Multi‐Omics Integration Reveals Metabolic Reprogramming and EMT‐Driven Metastasis in Triple‐Negative Breast Cancer

ABSTRACT

Triple‐negative breast cancer (TNBC) is an aggressive subtype with high metastatic potential and poor prognosis, yet the molecular mechanisms driving metastasis remain incompletely understood. Here, we performed integrated transcriptomic, metabolomic, and imaging mass cytometry (IMC) analyses on tumor tissues from 52 TNBC patients stratified into NonMeta (long‐term survival without metastasis, n = 39) and MetaPoor (metastasis‐related death within 36 months after metastatic diagnosis, n = 13) groups. Transcriptomic analysis revealed significant enrichment of epithelial‐mesenchymal transition (EMT), WNT signaling pathway, and angiogenesis pathways in MetaPoor tumors, with EMT identified as the top enriched hallmark by GSEA. Metabolomic profiling showed decreased branched‐chain amino acids (BCAAs) and arginine derivatives, alongside elevated glycerophosphorylcholine (GPC), indicating enhanced metabolic demand and membrane remodeling. IMC‐based spatial proteomics identified distinct tumor cell subpopulations, with MetaPoor tumors harboring higher proportions of stem‐like and EMT‐like cells characterized by upregulated HADHA (BCAA oxidation) and P38, and downregulated ATP5A (oxidative phosphorylation). Multi‐omics integration revealed strong correlations between GPC and EMT markers, and between BCAA depletion and HADHA expression. Collectively, our multi‐omics study demonstrates that metastatic TNBC is characterized by EMT activation, metabolic reprogramming toward altered BCAA metabolism and oxidative phosphorylation‐related protein expression, and remodeling of the tumor immune microenvironment, suggesting candidate molecules that warrant independent validation and further mechanistic investigation.

Integrated transcriptomic, metabolomic, and spatial proteomic profiling of TNBC tumors identifies a metastatic‐associated molecular landscape characterized by EMT activation, metabolic reprogramming, and tumor microenvironment (TME) remodeling. Multi‐omics integration reveals coordinated alterations in EMT signatures, amino acid metabolism, immune infiltration, and spatial cellular organization, defining potential biomarkers associated with metastatic progression in triple‐negative breast cancer.

Abbreviations

BCAA

branched‐chain amino acid

EMT

epithelial‐mesenchymal transition

FC

fold change

FDR

false discovery rate

GO

gene Ontology

GPC

glycerophosphorylcholine

GSEA

gene set enrichment analysis

HADHA

hydroxyacyl‐CoA dehydrogenase trifunctional multienzyme complex subunit alpha

IMC

imaging mass cytometry

KEGG

Kyoto encyclopedia of genes and genomes

LC–MS/MS

liquid chromatography–tandem mass spectrometry

MetaPoor

metastasis‐related death within 36 months

NonMeta

long term survival without metastasis

PLS‐DA

partial least squares discriminant analysis

SDMA

symmetric dimethylarginine

TNBC

triple‐negative breast cancer

UMAP

uniform manifold approximation and projection

VIP

variable importance in projection

1. Introduction

Triple‐negative breast cancer (TNBC) accounts for approximately 15%–20% of all breast cancers and is defined by the absence of estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 expression [1]. Molecularly, TNBC is characterized by marked heterogeneity and a predominance of basal‐like transcriptional profiles, contributing to its aggressive clinical behavior [2]. Unlike other breast cancer subtypes, TNBC lacks well‐established targeted therapies and is primarily treated with chemotherapy, which often results in suboptimal and short‐lived responses [3]. Consequently, patients with TNBC frequently experience poor clinical outcomes. Metastasis represents the leading cause of mortality in TNBC, and patients who develop metastatic disease typically exhibit dramatically reduced survival [4]. Notably, most metastatic events occur within the first 3 years following diagnosis, suggesting that tumors with high metastatic potential may possess distinct biological features at the time of initial presentation [5]. Therefore, identifying the molecular characteristics of metastasis‐prone tumors and elucidating the mechanisms driving metastatic progression are critical for improving patient outcomes in TNBC [6].

Metastasis is a complex, multi‐step process that involves coordinated changes in tumor cells and their surrounding microenvironment [7]. Previous studies have implicated several biological processes in the metastatic progression of TNBC [8]. Epithelial–mesenchymal transition (EMT) enables tumor cells to acquire migratory and invasive capabilities, facilitating dissemination from the primary tumor [9]. Immune evasion mechanisms, including suppression of cytotoxic T‐cell activity and recruitment of immunosuppressive cells, allow disseminated tumor cells to survive and establish metastatic niches [10]. Angiogenesis and vascular remodeling further support metastatic spread by promoting tumor cell intravasation and systemic circulation [11]. In parallel, metabolic reprogramming provides the energetic and biosynthetic demands required for rapid tumor growth and metastatic colonization [12]. The tumor microenvironment (TME), composed of stromal cells, immune cells, and extracellular matrix components, also plays a critical role in regulating tumor progression and metastatic behavior [13]. Although these processes have each been linked to metastasis, they are often studied in isolation, and how they interact to drive metastatic progression in TNBC remains incompletely understood.

To better understand the complex biology underlying metastatic progression, integrative multi‐omics approaches have emerged as powerful tools for characterizing tumors from complementary perspectives [14]. Transcriptomic profiling provides insights into gene regulatory programs and pathway‐level alterations associated with tumor progression, including processes such as EMT, WNT signaling pathway, and angiogenesis [15]. In contrast, metabolomic analysis captures functional metabolic states by quantifying small‐molecule metabolites, thereby reflecting alterations in amino acid utilization, energy metabolism, and lipid remodeling that support tumor growth and dissemination [16]. However, these molecular measurements alone cannot fully explain how different cell populations interact within the TME. Imaging mass cytometry (IMC), a spatially resolved proteomic technology, enables highly multiplexed protein detection while preserving tissue architecture, allowing the identification of cellular phenotypes and their spatial organization at single‐cell resolution [17]. Integrating these complementary platforms therefore provides a comprehensive framework for understanding the molecular, metabolic, and spatial determinants of metastatic progression.

In this study, we performed an integrated multi‐omics analysis to systematically characterize the molecular and spatial features associated with metastatic progression in triple‐negative breast cancer. By comparing tumors from patients with poor metastatic outcomes (MetaPoor) and those without metastatic progression (NonMeta), we combined transcriptomic profiling, metabolomic analysis, and IMC to investigate key biological processes involved in metastasis, including epithelial–mesenchymal transition, metabolic reprogramming, and remodeling of the tumor immune microenvironment. Through integrative analysis across these platforms, we aimed to identify coordinated molecular networks underlying metastatic progression and to discover exploratory biomarker candidates associated with clinical outcomes. These findings provide new insights into the mechanisms driving TNBC metastasis and may offer potential clues for risk stratification and therapeutic intervention.

2. Materials and Methods

2.1. Study Cohort and Sample Collection

A total of 52 patients with histologically confirmed TNBC who underwent surgical resection between November 2016 and August 2017 at Harbin Medical University Cancer Hospital were enrolled in this study. Inclusion criteria were: (1) histologically confirmed primary TNBC; (2) complete clinicopathological records; and (3) availability of fresh frozen tumor tissues with sufficient RNA quality for sequencing. Exclusion criteria included: (1) previous or concurrent malignancies; (2) receipt of neoadjuvant chemotherapy prior to surgery; and (3) incomplete follow‐up data. All patients underwent standard surgical resection and received adjuvant chemotherapy according to institutional guidelines, with treatment regimens recorded from medical charts. Patients were retrospectively divided into two groups based on metastasis status and prognosis: the NonMeta group (n = 39) comprised patients who remained metastasis‐free throughout follow‐up, while the MetaPoor group (n = 13) consisted of patients who developed metastasis and subsequently died of metastasis‐related causes within 36 months of metastasis diagnosis. Follow‐up was conducted every 3 months for the first 2 years, every 6 months for the next 3 years, and annually thereafter, with a median follow‐up duration exceeding 10 years. Clinical and pathological characteristics, including age, tumor size, and lymph node status, were collected from medical records. Fresh frozen tumor tissues were obtained at the time of surgery and immediately stored at −80°C until further processing. Detailed clinical information for all patients is provided in Table S1.

2.2. Transcriptomic Profiling and Analysis

Total RNA was extracted from tumor tissues and mRNA was enriched using Oligo(dT) magnetic beads. Following fragmentation, first‐strand cDNA was synthesized using random hexamer primers, followed by second‐strand synthesis. After end repair, A‐tailing, adapter ligation, size selection, and PCR amplification, libraries were sequenced on an Illumina platform with paired‐end reads. Raw reads were processed using fastp to remove adapter‐contaminated and low‐quality reads, yielding clean reads which were aligned to the GRCh38 reference genome using HISAT2. Gene expression levels were quantified using featureCounts, and differential expression analysis between NonMeta and MetaPoor groups was performed using the DESeq2 R package. Genes with adjusted p value (padj) < 0.05 and |log₂FC| ≥ 1 were considered significantly differentially expressed. A total of 1847 differentially expressed genes (DEGs) were identified. To focus downstream analyses on genes with established roles in cancer biology, the DEGs were further filtered using the Cancer Gene Census (CGC), a manually curated database of genes with well‐established causal roles in human cancer [18]. This filtering step identified 54 cancer‐related DEGs, which were subsequently used for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses. Functional enrichment analyses were performed using the clusterProfiler R package, with an adjusted p < 0.05 considered statistically significant. Gene Set Enrichment Analysis (GSEA) was performed independently using the ranked list of all expressed genes and the Hallmark gene set collection. Gene sets with an FDR < 0.05 were considered significantly enriched. Immune cell fractions were estimated using xCell based on transcriptomic data, and differences between groups were assessed using the Wilcoxon rank‐sum test.

2.3. Metabolomic Profiling and Analysis

Metabolite extraction was performed from tumor tissues using a methanol: water (7:3) solution containing internal standards. Samples were vortexed, sonicated in an ice bath, and incubated at −20°C to precipitate proteins. After centrifugation, the supernatant was collected for LC–MS/MS analysis. Chromatographic separation was carried out on a Waters ACQUITY HSS T3 column (1.8 μm, 2.1 × 100 mm) using a gradient of 0.1% formic acid in water and 0.1% formic acid in acetonitrile. Mass spectrometry was performed on a Q Exactive HF‐X instrument in both positive and negative ion modes, with full scan range of m/z 75–1000 at 35,000 resolution. Raw data were processed using ProteoWizard and XCMS for peak picking, alignment, and retention time correction. Metabolites were identified by matching m/z and MS/MS spectra against an in‐house database as well as public databases (HMDB, KEGG). After quality control filtering (CV < 0.3 in QC samples), missing values were imputed using K‐nearest neighbors or 1/5 minimum value. For statistical analysis, data were log₂‐transformed and Pareto‐scaled. Orthogonal partial least‐squares discriminant analysis (PLS‐DA) was performed using the MetaboAnalystR package, and variable importance in projection (VIP) scores were extracted. Model robustness was evaluated by permutation testing with 200 random permutations. Differential metabolites were defined by VIP > 1 and p < 0.05.

2.4. Agents and Antibody Labeling Protocol

Two serial sections from each tumor tissue block were used for IMC with two different antibody panels, designed to address complementary biological questions. Details of the antibodies used for IMC are provided in Tables S2 and S3. Antibodies were custom‐labeled with metals using the Maxpar X8 Multi‐Metal Labeling Kit (Standard Biotools). Other reagents included TCEP (Thermo Scientific), Amicon Ultra centrifugal filters (3 kDa and 50 kDa), DPBS (Corning), Cell‐ID Intercalator‐Ir (Standard Biotools), and BSA (Solarbio).

After preparing the antibodies and metals to be labeled, first add 95 μL of L‐Buffer to an X8‐polymer tube and mix until the polymer is completely dissolved, then add 5 μL of 50 mM lanthanide metal solution, mix well, and incubate in a 37°C water bath for 30 min. Add 100 μg of antibody to a 50 kDa centrifugal filter unit and replenish with R‐Buffer to a total volume of 400 μL, then centrifuge at 12,000 × g for 10 min at room temperature. Dilute TCEP with R‐Buffer to a final concentration of 4 mM (1:100 dilution). Add 200 μL of L‐Buffer to a 3 kDa centrifugal filter unit and transfer the metal‐polymer mixture into this 3 kDa filter unit, then centrifuge at 12,000 × g for 30 min at room temperature. Add 100 μL of 4 mM TCEP solution to the 50 kDa filter unit containing the antibody, mix well, and immediately transfer to a 37°C water bath for 30 min incubation. Discard the waste liquid from the collection tube of the 3 kDa filter unit, add 400 μL of C‐Buffer, and centrifuge at 12,000 × g for 30 min at room temperature. Add 300 μL of C‐Buffer to the 50 kDa filter unit and pipette along the wall to collect residual antibody, then centrifuge at 12,000 × g for 10 min at room temperature. After centrifugation, discard the waste liquid from the collection tube of the 50 kDa filter unit, add another 400 μL of C‐Buffer for a second wash, and centrifuge again at 12,000 × g for 10 min at room temperature. After centrifugation, remove the 3 kDa filter unit, discard the waste liquid from the collection tube, and confirm the correspondence between each antibody and metal. Add 60 μL of C‐Buffer and transfer the entire liquid to the 50 kDa filter unit, gently pipette to mix, and incubate in a 37°C water bath for 90 min. After incubation, remove the 50 kDa filter unit from the water bath, add 300 μL of W‐Buffer to the filter unit, and centrifuge at 12,000 × g for 10 min at room temperature. Discard the waste liquid from the collection tube and repeat this washing step with W‐Buffer three times. Finally, add 100 μL of antibody storage buffer, invert the filter unit into a new collection tube, and centrifuge at 1,000 × g for 2 min at room temperature to collect the labeled antibody. Store the labeled antibody at 4°C for future use.

2.5. Imaging Mass Cytometry and Analysis

The IMC antibody panels were designed around predefined biological questions concerning tumor‐cell phenotypes, metabolic reprogramming, signaling activity, hypoxia, mitochondrial function, and the tumor immune microenvironment. This targeted strategy was informed by previous studies demonstrating the utility of mass cytometry for resolving cellular metabolic states and functional programs at single‐cell resolution, as well as by established applications of IMC for characterizing spatially organized tumor and immune phenotypes in breast cancer and TNBC [19, 20, 21].

Formalin‐fixed, paraffin‐embedded (FFPE) slides were deparaffinized in fresh xylene for 20 min (fume hood) and rehydrated through graded ethanol (100%, 95%, 80%, 70%; 5 min each). After a 5 min wash in Maxpar Water (horizontal shaker), slides were subjected to heat‐induced epitope retrieval in preheated EDTA solution at 96°C for 30 min, followed by cooling to 70°C at room temperature. Slides were then washed sequentially in Maxpar Water and Maxpar PBS for 10 min each (horizontal shaker). Following sample encircling with a PAP pen, non‐specific binding was blocked with 3% BSA in Maxpar PBS for 45 min at room temperature in a hydration chamber. Antibody cocktail, prepared in PBS with 0.5% BSA, was applied to slides and incubated overnight at 4°C in a humidified chamber. Subsequently, slides were washed twice with 0.2% Triton X‐100 in Maxpar PBS and twice with Maxpar PBS, each for 8 min with slow agitation (horizontal shaker). Tissue was stained with Intercalator‐Ir (1:400 in Maxpar PBS) for 30 min at room temperature in a humidified chamber, followed by a 5 min wash in Maxpar Water with gentle agitation. After air‐drying for at least 20 min at room temperature, protein detection was performed using a Hyperion imaging system (Fluidigm, SF, USA) at 1 μm resolution and 200 Hz. For each sample, a 500 × 500 μm region was selected for analysis.

2.6. Statistical Analysis

All statistical analyses were performed using R software (version 4.3.3). Data are presented as mean ± SEM unless otherwise specified. Comparisons between two groups were performed using the Wilcoxon rank‐sum test. Correlations were assessed using Spearman's rank correlation coefficient. For transcriptomic differential expression analysis, the DESeq2 package was used with an independent filtering step. Genes with |log₂FC| > 1 and adjusted p value (padj) < 0.05 (Benjamini–Hochberg correction) were considered significantly differentially expressed. Metabolomic analyses were performed as an exploratory profiling approach. Differential metabolite features were evaluated using effect size and nominal p values, and metabolites with |log₂FC| > 0.5 and nominal p < 0.05 were considered candidate features for exploratory visualization and biological interpretation. Because these analyses were exploratory rather than confirmatory, nominal p values were used without multiple‐testing correction, and the resulting associations were interpreted cautiously. For IMC analyses, predefined antibody panels were used to investigate specific biological programs and cellular phenotypes. Group comparisons of IMC‐derived protein expression were likewise considered exploratory, with nominal p values reported for descriptive purposes rather than as confirmatory statistical evidence. Functional enrichment analyses (GO, KEGG, GSEA) were performed using the clusterProfiler package, with adjusted p values (padj) < 0.05 considered statistically significant after Benjamini–Hochberg correction.

2.7. Immunohistochemistry (IHC)

Formalin‐fixed, paraffin‐embedded (FFPE) tumor tissue sections were cut at 4 μm thickness. Sections were deparaffinized in xylene and rehydrated through graded ethanol. Antigen retrieval was performed by heating sections in EDTA buffer (pH 9.0) at 95°C for 20 min, followed by cooling at room temperature. Endogenous peroxidase activity was blocked with 3% H₂O₂ for 10 min. Sections were incubated with primary antibodies overnight at 4°C. After washing, sections were incubated with HRP‐conjugated secondary antibody for 30 min at room temperature. Immunoreactivity was visualized using 3,3′‐diaminobenzidine (DAB) as chromogen. Stained sections were scanned using a digital pathology scanner and viewed with SlideViewer software.

3. Results

3.1. Transcriptomic Profiling Reveals Distinct Molecular Features of Metastatic Tumors

A total of 52 patients with TNBC were included in this study, with a median follow‐up of over 10 years. All patients who developed metastasis experienced metastatic events within 36 months and subsequent metastasis‐related death. Kaplan–Meier analysis revealed a marked survival difference between groups, with patients in the metastatic group exhibiting dramatically shortened overall survival compared to their non‐metastatic counterparts (Figure 1A). Quality assessment of the transcriptomic data, including PLS‐DA analysis, confirmed the overall data structure and consistency among samples (Figure S1).

FIGURE 1.

Transcriptomic Landscape of Metastatic and Non‐metastatic TNBC Tumors. (A) Kaplan–Meier survival curves comparing overall survival between NonMeta (n = 39) and MetaPoor (n = 13) groups. (B) Gene Ontology (GO) enrichment analysis of biological processes enriched in cancer‐related genes differentially expressed between MetaPoor and NonMeta tumors. Bubble size represents gene count, color indicates‐log10(p. adjust). (C) KEGG pathway enrichment analysis of differentially expressed genes. (D) (Figure 1D). GSEA plot showing enrichment of the Hallmark_Epithelial_Mesenchymal_Transition signature in MetaPoor tumors. (E) xCell deconvolution analysis comparing immune cell fractions between NonMeta and MetaPoor tumors. Box plots show significantly lower B cell and CD8+ T cell infiltration, and higher M2 macrophage, fibroblast, and endothelial cell proportions in MetaPoor tumors. p values are indicated in the figure.

To elucidate the molecular underpinnings of metastatic progression, we performed transcriptomic profiling on tumor tissues from both groups. Differential expression analysis revealed substantial gene expression alterations in metastatic tumors compared to non‐metastatic tissues. To focus on cancer‐relevant alterations, we restricted our analysis to genes from the CGC and performed functional enrichment analysis. GO analysis of biological processes revealed significant enrichment of pathways critically involved in metastatic dissemination, including mesenchymal development, mesenchymal cell differentiation, epithelial‐mesenchymal transition (EMT), and regulation of the Wnt signaling pathway (Figure 1B). Notably, angiogenesis‐related processes such as vasculogenesis, positive regulation of angiogenesis, and vasculature development were also prominently enriched, highlighting the importance of vascular remodeling in metastatic progression. Additionally, metabolic reprogramming was evident through the enrichment of the tricarboxylic acid cycle, fatty acid oxidation, and glucose transmembrane transport, indicating that metastatic tumors undergo substantial metabolic adaptation to support their aggressive behavior.

KEGG pathway analysis further corroborated these findings, revealing enrichment of multiple cancer‐related signaling cascades (Figure 1C). The most prominently enriched pathways included PI3K‐Akt, MAPK, mTOR, Ras, Rap1, and Wnt signaling pathway, all of which are well‐established drivers of tumor progression and metastasis. Notably, breast cancer‐specific pathways and drug resistance pathways, including EGFR tyrosine kinase inhibitor resistance and platinum drug resistance, were also significantly enriched, suggesting that metastatic tumors may possess intrinsic resistance mechanisms. To complement these findings with an unbiased approach, we performed GSEA using the Hallmark gene set collection. This analysis identified EMT as the most significantly enriched hallmark pathway in metastatic tumors (Figure 1D), consistent with the GO results and reinforcing the central role of EMT in driving metastatic dissemination in TNBC.

Given the importance of TME in metastasis, we next characterized the immune landscape of these tumors using xCell deconvolution analysis. Compared to non‐metastatic tumors, metastatic tumors exhibited significantly lower infiltration of B cells and CD8+ T cells, accompanied by higher proportions of M2 macrophages, cancer‐associated fibroblasts, and endothelial cells (Figure 1E). This immunosuppressive and pro‐angiogenic microenvironment is consistent with the EMT and angiogenesis signatures observed in the transcriptomic analysis, suggesting that metastatic tumors actively remodel their microenvironment to evade immune surveillance and facilitate vascular invasion.

Collectively, these findings demonstrate that metastatic TNBC tumors are characterized by activation of EMT and Wnt signaling pathway, enhanced angiogenesis, metabolic reprogramming, and an immunosuppressive microenvironment, representing a multi‐faceted adaptation program that distinguishes metastatic from non‐metastatic tumors.

3.2. Metabolomic Profiling Reveals Distinct Metabolic Reprogramming in Metastatic Tumors

To investigate the metabolic alterations associated with metastatic progression, we performed LC–MS metabolomic profiling, quantifying 160 metabolites involved in amino acid, nucleotide, energy, and lipid metabolism. Partial least squares discriminant analysis (PLS‐DA) revealed a clear separation between the two groups (Figure 2A), indicating that metastatic tumors possess a distinct metabolic landscape compared to their non‐metastatic counterparts. Permutation testing (200 iterations) confirmed that the observed separation was not due to random chance, as the classification error rate of the original model was significantly lower than that of all permuted models (Figure S2). A heatmap of the top 30 discriminatory metabolites further illustrated distinct metabolic patterns between the MetaPoor and NonMeta groups (Figure 2B). The MetaPoor group generally exhibited lower levels of multiple metabolites, whereas the NonMeta group showed relatively higher metabolite abundance, suggesting increased metabolic consumption in metastatic tumors.

FIGURE 2.

Metabolomic Reprogramming in Metastatic TNBC Tumors. (A) Partial least squares discriminant analysis (PLS‐DA) score plot showing distinct metabolic profiles between NonMeta and MetaPoor tumor tissues. (B) Heatmap of the top 30 discriminatory metabolites identified by VIP scores. Colors represent scaled metabolite levels, with red indicating higher and blue indicating lower abundance. (C) Variable importance in projection (VIP) scores of the top 30 metabolites contributing to group discrimination. Higher scores indicate greater contribution to the PLS‐DA model. (D) Correlation heatmap showing co‐expression patterns among candidate differential metabolites (VIP > 1), revealing coordinated metabolic modules associated with the MetaPoor phenotype. (E) Box plots comparing relative abundance of key metabolites between NonMeta and MetaPoor groups. GPC and adenine were elevated, while branched‐chain amino acids (BCAAs) (valine, isoleucine, leucine), arginine derivatives (N‐methyl‐L‐arginine, SDMA‐ADMA), and taurine were decreased in MetaPoor tumors.

To characterize metabolic features associated with the MetaPoor phenotype, we first evaluated the contribution of individual metabolites to group discrimination using VIP scores from the PLS‐DA model. Metabolites with VIP > 1 were considered influential features for exploratory profiling (Figure 2C). Putrescine (VIP = 2.15), pantothenic acid (VIP = 1.86), and glycine (VIP = 1.82) ranked among the most influential features, followed by valine‐norvaline (VIP = 1.76), GPC (VIP = 1.73), isoleucine (VIP = 1.73), leucine (VIP = 1.71), N‐methyl‐L‐arginine (VIP = 1.67), and taurine (VIP = 1.56). These results identified a set of metabolites that contributed substantially to the separation between the NonMeta and MetaPoor groups.

We next examined the relationships among differential metabolites and observed coordinated variation across several amino acid‐ and nitrogen‐related metabolites (Figure 2D). Methionine‐sulfoxide, N‐acetyl‐L‐alanine, N‐methyl‐L‐arginine, SDMA‐ADMA, phenylalanine, histidine, valine‐norvaline, tyrosine, asparagine, arginine, methionine, isoleucine, and leucine showed coordinated abundance patterns. These metabolites are involved in interconnected processes related to amino acid metabolism, nitrogen metabolism, protein synthesis, and cellular signaling, suggesting broader metabolic remodeling associated with the MetaPoor phenotype.

Based on the discriminatory features identified in Figure 2C and the coordinated metabolic patterns observed in Figure 2D, we further examined representative metabolites showing consistent group‐level alterations for focused visualization and biological interpretation (Figure 2E). GPC and adenine were significantly elevated in MetaPoor tumors. GPC is a glycerophospholipid involved in choline phospholipid metabolism and has been associated with breast cancer progression and EMT‐related metabolic remodeling [22]. Adenine, a purine metabolite, is linked to altered nucleotide metabolism and proliferative activity in cancer [23]. In contrast, branched‐chain amino acids (BCAAs) (valine, isoleucine, and leucine), arginine‐related metabolites (N‐methyl‐L‐arginine and SDMA‐ADMA), and taurine were markedly decreased in MetaPoor tumors. Alterations in BCAA metabolism have been implicated in breast cancer progression and TNBC malignancy [24], while arginine metabolism is closely linked to tumor progression, angiogenesis, and immune regulation [25]. Taurine contributes to redox homeostasis and has also been implicated in breast cancer progression and metastasis [26]. Collectively, these metabolic alterations highlight coordinated changes in phospholipid, nucleotide, amino acid, and redox metabolism associated with the MetaPoor phenotype.

The coordinated downregulation of BCAAs and arginine metabolites aligns with our transcriptomic findings of mTOR and PI3K‐Akt pathway activation, suggesting that metastatic tumors actively consume these amino acids to fuel protein synthesis and support rapid proliferation. The elevation of GPC directly parallels the EMT signature identified by GSEA, providing metabolic evidence of membrane remodeling during the acquisition of migratory capacity. Furthermore, the altered arginine/ADMA balance points toward enhanced nitric oxide production and angiogenesis, consistent with the vasculogenesis pathways enriched in our GO and KEGG analyses. Together, these metabolomic findings demonstrate that metastatic TNBC tumors undergo profound metabolic reprogramming characterized by enhanced utilization of amino acids for protein synthesis, membrane remodeling associated with EMT, and altered energy metabolism to support increased proliferative demand. The convergence of transcriptomic and metabolomic findings provides a comprehensive view of the metastatic phenotype in TNBC and highlights candidate metabolic pathways that may merit further investigation.

3.3. Imaging Mass Cytometry Reveals Spatial Organization of the Tumor Microenvironment in Metastatic Tumors

To further investigate the cellular composition and spatial architecture of the TME at single‐cell resolution, we performed IMC on tumor tissue sections from both groups. A panel of 35 protein markers was used to characterize immune and stromal populations, with iridium (Ir) staining serving as a nuclear counterstain to identify individual cells (Figure 3A). Dimensionality reduction and unsupervised clustering using UMAP enabled the identification of 11 distinct cellular populations within the TME (Figure 3B). These populations were annotated based on their protein expression profiles, revealing the presence of malignant cells, immune subsets, and stromal components (Figure 3C). Comparative analysis of cellular abundance revealed that metastatic tumors exhibited lower proportions of CD8+ T cells and B cells, accompanied by higher frequencies of cancer‐associated fibroblasts and M2 macrophages compared to non‐metastatic tumors (Figure 3D), confirming the immunosuppressive nature of the metastatic microenvironment.

FIGURE 3.

Single‐Cell Profiling of the Tumor Microenvironment by Imaging Mass Cytometry (IMC). (A) Representative IMC image showing nuclear staining with iridium (Ir) to identify individual cells. Scale bar = 100 μm. (B) UMAP dimensionality reduction and unsupervised clustering of single cells from tumor tissues, revealing 11 distinct cellular populations. (C) Heatmap showing the expression levels of lineage‐specific markers across the 11 identified cell clusters, enabling annotation of malignant cells, immune subsets, and stromal components. (D) Comparison of cellular abundance between NonMeta and MetaPoor tumors. MetaPoor tumors exhibited significantly lower proportions of CD8+ T cells and B cells, and higher frequencies of cancer‐associated fibroblasts and M2 macrophages. p values are indicated in the figure.

Cell type annotation was performed based on expression of lineage‐specific markers visualized by IMC. Individual marker channels and merged images are presented to illustrate staining specificity and cellular resolution (Figure 4A,B). Nuclear staining (DNA1/Ir) was used to generate single‐cell masks (Figure S3A,B), allowing for subsequent quantitative analysis of cellular spatial relationships.

FIGURE 4.

Spatial Organization of the Tumor Microenvironment in Metastatic Tumors. (A–B) Representative multiplex IMC images showing the antibody panels used for cell type identification. Panel A (Pan‐CK, DNA1, CD8, CD20, α‐SMA, CD68) and Panel B (Pan‐CK, DNA1, CD31, VIM, CD68, CD15) display individual marker staining patterns along with merged images. Scale bars = 100 μm. (C–D) Distance analysis showing that malignant cells in MetaPoor tumors are positioned significantly farther from CD8+ T cells and B cells compared to NonMeta tumors. (E–F) Analysis of vascular components revealing that endothelial cells in MetaPoor tumors are located closer to stromal cells but farther from neutrophils.

Neighborhood analysis revealed distinct spatial patterns associated with metastatic progression. In metastatic tumors, malignant cells were positioned at significantly greater distances from CD8+ T cells and B cells (Figure 4C,D), suggesting exclusion of cytotoxic immune effectors from tumor cell proximity. Conversely, cancer‐associated fibroblasts were located in closer proximity to CD8+ T cells and B cells in metastatic tumors (Figure S3C,D), potentially creating a physical barrier that impedes immune cell infiltration. Similarly, M2 macrophages exhibited reduced distances to both CD8+ T cells and B cells (Figure S3E,F), consistent with their role in fostering an immunosuppressive niche. Analysis of vascular components revealed that endothelial cells in metastatic tumors were positioned closer to stromal cells but farther from neutrophils (Figure 4E,F), while their distances to malignant cells and M2 macrophages were reduced (Figure S3G). This spatial configuration suggests that metastatic tumors organize a pro‐angiogenic and immunosuppressive niche that facilitates vascular invasion.

Collectively, these IMC findings provide spatial evidence of immune exclusion and stromal remodeling in metastatic TNBC tumors, demonstrating that the molecular programs identified earlier manifest as distinct cellular organizations within the TME.

3.4. Spatial Proteomic Analysis Reveals Distinct Tumor Cell Subpopulations and Metabolic Reprogramming in Metastatic Tumors

To further investigate the spatial organization and metabolic heterogeneity of tumor cells at single‐cell resolution, we applied a separate IMC panel on a serial section, focusing on tumor cell phenotypes. Dimensionality reduction and unsupervised clustering using UMAP identified seven distinct cellular populations (Figure 5A), with feature plots illustrating the expression patterns of key markers across these clusters (Figure 5B). The expression patterns of additional markers across UMAP clusters are presented in the (Figure S4).

FIGURE 5.

Subpopulations and Metabolic Reprogramming in Metastatic Tumors. (A) UMAP dimensionality reduction and unsupervised clustering of single cells from tumor tissues, revealing seven distinct cellular populations. (B) Feature plots showing the expression patterns of key markers across the UMAP dimensions. (C) Representative imaging mass cytometry(IMC) images illustrating tumor cell subpopulation identification based on Pan‐CK, Vimentin, CD44, CD133, and CD24 expression. Individual marker channels and merged images are shown. Scale bar = 100 μm. (D) Ridgeline plots depicting the spatial distribution of distances between different tumor cell subpopulations and other cell types in NonMeta and MetaPoor groups. (E) Comparison of the proportions of stem‐like and EMT‐like epithelial cells between NonMeta and MetaPoor tumors. (F) Box plots showing relative protein expression levels in EMT‐like epithelial cells from NonMeta and MetaPoor tumors. (G) Box plots showing relative protein expression levels in stem‐like epithelial cells from NonMeta and MetaPoor tumors. p values are indicated in the figure.

Based on the expression patterns of Pan‐CK, CD44, CD133, CD24, and Vimentin, tumor cells were further categorized into three subpopulations: epithelial cells (Pan‐CK⁺), stem‐like epithelial cells (Pan‐CK⁺/CD44⁺/CD133⁺), and EMT‐like epithelial cells (Pan‐CK⁺/Vimentin⁺/CD44⁺). (Figure 5C). Spatial analysis of these subpopulations revealed distinct organizational patterns within tumor tissues. Stem‐like epithelial cells were positioned farthest from other cell types, suggesting their localization at the tumor core, while EMT‐like epithelial cells exhibited intermediate distances, potentially forming a transitional zone between the tumor core and periphery. This spatial arrangement suggests a radial organization of tumor cell subpopulations, with stem‐like cells at the center, EMT‐like cells in the middle, and differentiated epithelial cells at the invasive front (Figure 5D).

Comparative analysis between groups revealed that MetaPoor tumors harbored significantly higher proportions of both stem‐like and EMT‐like epithelial cells compared to NonMeta tumors (Figure 5E), indicating an enrichment of aggressive tumor cell phenotypes in metastatic lesions. Protein expression analysis within these subpopulations further uncovered distinct metabolic reprogramming associated with metastasis. In EMT‐like epithelial cells from MetaPoor tumors, we observed elevated levels of Vimentin, CD44, HADHA, ACADM, CD98, p4EBP1, PS6, HIF1α, CA9, and GLUT1, accompanied by decreased expression of ATP5A, CYTC, CS, and IKB (Figure 5F). This expression pattern indicates enhanced BCAA oxidation, mTOR activation, glycolytic shift, and hypoxia adaptation, coupled with suppression of oxidative phosphorylation and NF‐κB signaling. In stem‐like epithelial cells, MetaPoor tumors showed increased expression of CD44, CD98, p4EBP1, PS6, HIF1α, CA9, GLUT1, CYTC, and P38 (Figure 5G), suggesting that these cells retain some oxidative capacity while also adapting to hypoxic conditions and activating stress responses.

Collectively, these spatial proteomic findings demonstrate that metastatic TNBC tumors are characterized by an enrichment of stem‐like and EMT‐like tumor cell subpopulations with distinct metabolic programs. The radial organization of these subpopulations, coupled with their metabolic reprogramming toward enhanced BCAA utilization, glycolysis, and hypoxia adaptation, provides a spatial framework for understanding how tumor cells coordinate their metabolic strategies to support metastatic progression.

3.5. Integrated Multi‐Omics Data Were Used to Identify Coordinated Molecular Networks Associated With Metastasis

Validation of immune cell fractions between transcriptomic deconvolution (xCell) and IMC showed significant positive correlations across major immune cell types, including CD8⁺ T cells (R = 0.422, p = 0.002), B cells (R = 0.506, p < 0.001), M2 macrophages (R = 0.395, p = 0.004), fibroblasts (R = 0.567, p < 0.001), and endothelial cells (R = 0.436, p = 0.001), confirming the reliability of both platforms (Figure 6A). Correlation analysis between metabolites and clinical parameters revealed distinct patterns (Figure 6B). Glycerophosphorylcholine (GPC), a metabolite associated with EMT, showed negative correlations with overall survival and positive correlations with metastatic status. Conversely, BCAAs (isoleucine, valine‐norvaline) and arginine derivatives (N‐methyl‐L‐arginine) exhibited positive correlations with survival time and negative correlations with metastasis, indicating that their depletion in aggressive tumors reflects enhanced metabolic demand. Scatter plots of selected metabolites against OS_month further illustrated these relationships (Figure 6C), with isoleucine (R = 0.337, p = 0.015), valine‐norvaline (R = 0.329, p = 0.017), and N‐methyl‐L‐arginine (R = 0.392, p = 0.004) showing protective effects, while GPC emerged as a risk factor (R = −0.34, p = 0.014). Additional correlations between metabolites and overall survival are shown in the (Figure S5).

FIGURE 6.

Integrated Molecular Networks Associated with Metastatic Progression. (A) Correlation of immune cell fractions between transcriptomic deconvolution and imaging mass cytometry (IMC). (B) Heatmap of correlations between key metabolites and clinical parameters. *p < 0.05, **p < 0.01, ***p < 0.001. (C) Scatter plots showing correlations between selected metabolites and overall survival. (D) Heatmaps of correlations between protein expression in stem‐like cells and EMT‐like cells and clinical parameters. (E) Scatter plots showing correlations between selected proteins and overall survival. (F) Immunohistochemical validation of HADHA, ATP5A, and CD98 protein expression in NonMeta and MetaPoor tumor tissues. Scale bars = 50 μm. H‐Score quantification of protein expression. *p < 0.05.

Protein expression in stem‐like and EMT‐like cells showed distinct correlation patterns with clinical parameters (Figure 6D). In stem‐like cells, CD98 and CD44 exhibited strong negative correlations with survival and positive correlations with metastasis, suggesting that stemness‐associated amino acid transport and adhesion molecules contribute to poor prognosis. In EMT‐like cells, HADHA and P38 showed strong negative correlations with survival, while ATP5A demonstrated a positive correlation, indicating that preserved oxidative phosphorylation may confer protective effects. Scatter plots of selected proteins against OS_month confirmed these findings (Figure 6E), with HADHA showing the strongest negative correlation among all proteins analyzed (R = −0.608, p < 0.001), followed by CD98 (R = −0.511, p < 0.001) and CD44 (R = −0.394, p = 0.004) in stem‐like cells, while ATP5A in EMT‐like cells showed a positive correlation (R = 0.415, p = 0.002). Additional correlations between proteins and overall survival are presented in the (Figures S6 and S7). To assess the generalizability of our findings, we performed external survival analysis using the KM Plotter breast cancer mRNA database restricted to TNBC cases. High HADHA expression was significantly associated with shorter overall survival (HR = 1.40, log‐rank p = 0.030). High CD98 expression showed a consistent but non‐significant trend toward poorer survival (HR = 1.27, log‐rank p = 0.12) (Figure S8A). Sensitivity analyses confirmed that the HADHA–OS and CD98–OS correlations were robust to leave‐one‐out cross‐validation and extreme‐value exclusion (Figure S8B,C).

To further validate the protein expression differences identified by IMC, we performed immunohistochemistry (IHC) on an independent set of tumor sections from the same cohort. Consistent with the IMC findings, HADHA and CD98 expression were significantly elevated in MetaPoor tumors compared to NonMeta tumors, whereas ATP5A expression was markedly decreased in the MetaPoor group (Figure 6F). H‐Score quantification confirmed these differences, further supporting the reliability of the spatial proteomic observations and reinforcing the association of these proteins with metastatic progression.

4. Discussion

Metastasis represents a highly coordinated biological process requiring tumor cells to acquire invasive capacity, adapt to metabolic stress, and interact dynamically with the surrounding microenvironment [27]. In the present study, the integration of transcriptomic, metabolomic, and spatial proteomic analyses provides a systems‐level view of the biological programs associated with metastatic progression in TNBC [28]. Rather than representing independent alterations, our findings suggest that transcriptional reprogramming, metabolic adaptation, and spatial remodeling of the TME are tightly interconnected processes that collectively facilitate tumor dissemination [29].

One of the central features emerging from our analysis is the activation of EMT programs in metastatic tumors. EMT is widely recognized as a key driver of tumor invasion and dissemination, enabling epithelial tumor cells to acquire migratory capacity and phenotypic plasticity [30]. Beyond promoting cell motility, EMT has also been linked to the acquisition of stem‐like characteristics and increased resistance to cellular stress [31]. In this context, the enrichment of Wnt signaling pathway observed in metastatic tumors may further reinforce this plastic state, as Wnt signaling pathways are known to regulate stemness and cellular dedifferentiation [32]. Importantly, the spatial proteomic analysis revealed an increased abundance of stem‐like and EMT‐like epithelial cell populations in metastatic tumors, suggesting that transcriptional activation of EMT is accompanied by the expansion of specific tumor cell phenotypes [33]. This observation highlights the possibility that EMT‐associated plasticity may generate cellular states that are particularly suited for metastatic dissemination [34].

Metabolic reprogramming appears to represent a complementary adaptation that supports these phenotypic transitions [35]. Immunohistochemical validation further confirmed the differential expression of HADHA, CD98, and ATP5A between NonMeta and MetaPoor tumors, supporting the reliability of the spatial proteomic findings. The coordinated depletion of BCAAs and arginine‐related metabolites observed in metastatic tumors suggests enhanced metabolic consumption rather than reduced metabolic activity. Such metabolic shifts may provide both biosynthetic substrates and signaling inputs that sustain aggressive tumor phenotypes. BCAAs are known regulators of mTOR signaling, a pathway closely linked to protein synthesis, cell growth, and tumor progression [36]. Their increased utilization may therefore reflect elevated anabolic demand in rapidly proliferating metastatic cells [37]. At the same time, the accumulation of GPC may indicate enhanced phospholipid turnover and membrane remodeling, processes that are essential for dynamic cytoskeletal rearrangement during cell migration [38]. Taken together, these metabolic alterations suggest that metastatic tumor cells undergo a metabolic shift that supports both energetic demands and structural adaptations required for invasion [39].

Importantly, the metabolic and transcriptional changes identified in tumor cells appear to be closely coupled with remodeling of the TME [40]. The reduced presence of cytotoxic immune populations together with increased stromal and immunosuppressive cell types suggests the establishment of an immune‐permissive niche that may allow disseminating tumor cells to evade immune surveillance [41]. Spatial analyses further revealed that immune cells were positioned at greater distances from tumor cells in metastatic tissues, indicating a pattern of immune exclusion. Such spatial segregation has recently been recognized as an important mechanism of immune escape in multiple tumor types [42]. In this context, stromal components such as fibroblasts and macrophages may function not only as supportive cells but also as organizers of the tumor ecosystem, shaping spatial barriers that limit immune cell access to malignant cells [43]. These observations suggest that metastatic tumors may actively reorganize their microenvironment to create conditions that favor tumor cell survival and dissemination.

The integration of these observations supports a model in which metastatic progression in TNBC arises from the convergence of cellular plasticity, metabolic adaptation, and microenvironmental remodeling [44]. EMT‐associated transcriptional programs may generate highly plastic tumor cell states capable of invasion, while metabolic rewiring provides the biochemical resources required to sustain these phenotypes. Simultaneously, spatial restructuring of the TME may promote immune evasion and create supportive niches that facilitate metastatic spread. This coordinated reprogramming highlights the importance of studying tumor biology across multiple molecular layers and spatial scales [45], as single‐dimensional analyses may fail to capture the complex interactions that ultimately drive disease progression.

The relatively small size of the MetaPoor subgroup (n = 13) limits the statistical power and precision of our survival analyses. Based on the observed effect sizes, the estimated power was approximately 64% for HADHA (r = −0.608) and 44% for CD98 (r = −0.511) at a two‐sided α of 0.05. Therefore, these associations should be considered exploratory and warrant validation in larger, independent TNBC cohorts.

Beyond providing mechanistic insights into metastatic progression, our findings may also suggest possible directions for future research. The coordinated molecular and metabolic features identified in this study suggest that primary tumors with a high metastatic potential may already harbor detectable biological signatures prior to the occurrence of clinically evident metastasis. Profiling these features within tumor tissues could therefore contribute to improved risk stratification by identifying patients who are more likely to develop early metastatic relapse. In addition, several molecules highlighted in our analyses may represent molecular features that could be explored in future studies. Metabolic pathways associated with amino acid utilization and mTOR signaling, together with proteins involved in cellular plasticity and nutrient transport, may provide opportunities for further investigation. Importantly, the convergence of metabolic rewiring and immune microenvironment remodeling observed in metastatic tumors also raises the possibility that combinatorial therapeutic strategies targeting both tumor metabolism and the TME may enhance treatment efficacy. Future studies integrating molecular profiling with clinical intervention strategies may therefore help translate these findings into more effective therapeutic approaches for TNBC (Figure 7).

FIGURE 7.

Graphical abstract. Schematic overview of the study design and integrated multi‐omics findings in triple‐negative breast cancer (TNBC). Tumor tissues from 52 TNBC patients were stratified into NonMeta and MetaPoor groups. Multi‐omics profiling, including transcriptomics, metabolomics, and imaging mass cytometry (IMC)‐based spatial proteomics, was performed to characterize the molecular and spatial features associated with metastatic progression. Integration of these platforms revealed coordinated alterations in immune infiltration, differential metabolites, and spatial neighborhood organization, collectively defining an EMT activation signature within the tumor microenvironment (TME) of metastatic TNBC.

Author Contributions

Yuheng Pang: writing – original draft, data curation. Yuefeng Shang: investigation, resources. Wenjing Wang: project administration, methodology. Tong Liu: writing – review and editing, supervision. Xiaolong Zou: writing – review and editing, writing – original draft, conceptualization.

Funding

This work was supported by grants from Beijing Natural Science Foundation (Z230008 to W.W), National Natural Science Foundation of China (82272872 to T.L), Heilongjiang Provincial Department of Science and Technology, China (JD24D001) (Heilongjiang Province Science and Technology Innovation Base Reward Project, to T.L).

Ethics Statement

This study was approved by the Ethics Committee of Harbin Medical University Cancer Hospital (KY2016‐34), and written informed consent was obtained from all participants.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Clinical and pathological characteristics of 52 TNBC patients.

Table S2: Mass cytometry antibodies panel 1 design.

Table S3: Mass cytometry antibodies panel 2 design.

Figure S1: PLS‐DA analysis of transcriptomic profiles. Partial least squares discriminant analysis (PLS‐DA) score plot showing the separation between NonMeta (n = 39, green) and MetaPoor (n = 13, red) groups based on log₂ − transformed TPM expression data. Ellipses represent 95% confidence intervals for each group.

Figure S2: Permutation test for the OPLS‐DA model. The histogram shows the distribution of classification error rates from 200 random permutations of group labels. The red vertical line indicates the error rate of the original model, confirming significantly better performance than expected by chance.

Figure S3: Cell Segmentation and Spatial Analyses. (A‐B) Representative images showing cell segmentation masks generated based on nuclear (DNA1/Ir) staining, Scale bars = 100 μm. (C‐D) Distance analysis revealing that cancer‐associated fibroblasts in MetaPoor tumors are positioned closer to CD8+ T cells and B cells. (E‐F) M2 macrophages in MetaPoor tumors exhibit reduced distances to CD8+ T cells and B cells. (G) Additional spatial analysis showing that endothelial cells in MetaPoor tumors are located closer to malignant cells and M2 macrophages.

Figure S4: Protein expression patterns across UMAP clusters (A) Feature plots showing the expression patterns of key protein markers across the seven identified UMAP clusters.

Figure S5: correlations between metabolites and overall survival (A) Scatter plots showing correlations between additional metabolites and overall survival (OS_month). Each plot displays individual patient samples color‐coded by group (NonMeta in green, MetaPoor in red).

Figure S6: Correlations between EMT‐like cell proteins and overall survival (A) Scatter plots showing correlations between additional proteins expressed in EMT‐like cells and overall survival.

Figure S7: Correlations between stem‐like cell proteins and overall survival (A) Scatter plots showing correlations between additional proteins expressed in stem‐like cells and overall survival.

Figure S8: External validation and sensitivity analysis. (A) Kaplan–Meier curves for HADHA and CD98 in a public TNBC cohort (n = 534) using the Kaplan‐Meier Plotter database (ER‐/PR‐/HER2‐negative; OS endpoint; auto‐select best cutoff). (B) Leave‐one‐out cross‐validation showing stable Spearman R values for HADHA and CD98. (C) Correlation R values remained stable after excluding extreme expression values.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.