Introduction

Breast cancer (BC) is a heterogeneous disease in which prognosis and treatment response are influenced by both the intrinsic properties of tumor cells and their surrounding microenvironment. The molecular subtype of the tumor currently plays a fundamental role in guiding treatment decisions. Among the various subtypes, luminal and triple-negative BC are the most diverse in terms of clinical behavior and biological characteristics.

Luminal BC is hormone receptor-positive (HR⁺) and is further classified into two distinct subtypes: luminal A and luminal B. Luminal A tumors are characterized by estrogen receptor (ER) and progesterone receptor (PR) positivity with low Ki-67 expression (< 20%), reflecting low tumor proliferation. In contrast, luminal B tumors also express ER and PR but present with higher Ki-67 levels and can be either HER2-positive or HER2-negative (human epidermal growth factor receptor 2). Luminal BC is the most frequently diagnosed subtype and is typically the least clinically aggressive [1]. At the molecular level, luminal A tumors show increased expression of genes linked to ER function, including BCL2, ESR1, PGR, and FOXA1. Luminal B tumors share elevated expression of BCL2, FOXA1, CCND1, and GATA3 but differ by expressing higher levels of proliferation-associated genes such as CCNB1, CCND1, CCNE1, MYBL2, and MKI67 [2]. Notably, tumors of the luminal subtype are generally immunologically “cold,” characterized by low infiltration of immune cells, particularly tumor-infiltrating lymphocytes (TILs). However, when high TIL levels are present, they have been associated with adverse prognostic features, including high histological grade and lymphovascular invasion [3].

Triple-negative BC is a clinically aggressive subtype accounting for approximately 10% of BC cases. It is defined by the absence of ER, PR, and HER2. Triple-negative BC typically exhibits a high Ki-67 proliferation index (> 30%) and is associated with the poorest prognosis among BC subtypes [1]. Basal-like BC, which is often over-lapping with triple-negative BC, is characterized by increased expression of basal cytokeratins 5, 14, and 17, along with dysregulation of key signaling pathways including PI3K/AKT, JAK/STAT, and ERK/MAP. These tumors also show high expression of proliferation-associated genes such as FOXM1, c-MYC, CCNE1, BIRC5, and CCND1. Additionally, there is elevated expression of genes involved in cell cycle regulation (CDC20, CDC6) and the epidermal growth factor receptor (EGFR) pathway, coupled with decreased expression of estrogen-related genes [2]. Notably, high levels of TILs are more frequently observed in triple-negative BC compared to luminal subtypes. The presence of abundant TILs in triple-negative BC is associated with a more favorable prognosis and improved response to neoadjuvant chemotherapy [3].

The list of approaches for BC research that hold potential for translation into clinical practice is expanding [4], making it possible to move beyond the study of morphological tumor characteristics and toward an emphasis on the functional state of the tumor and its microenvironment, including intercellular contacts. Recent advances in single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics have enabled the development of detailed molecular atlases that capture the heterogeneity of both tumor and stromal cell populations in BC [5, 6]. While the characteristic features of tumors and their microenvironments across different BC subtypes are increasingly well understood, data on tumor-stromal interactions specific to these subtypes remain scarce. This gap is significant given that interactions between tumor and microenvironmental cells are recognized as key factors influencing chemotherapy response, recurrence, and metastasis [7, 8]. Given that luminal and triple-negative BCs exhibit fundamentally different clinical behaviors, treatment responses, and microenvironmental compositions, the transcriptomic features of tumor cells and their adjacent stromal partners may be subtype-specific. Defining these distinct features is a prerequisite for developing subtype-informed spatial biomarkers that could refine diagnostic classification or predict differential therapeutic sensitivity.

The aim of our study was to investigate the transcriptomic features of tumor and stromal cells in close proximity within tumor nodules of luminal and triple-negative BC. We employed Visium 10x spatial transcriptomics profiling, which enables the integration of transcriptome data with tumor tissue morphology, allowing detailed molecular characterization of cell populations in specific morphological regions. Special emphasis was placed on the precise manual annotation of these tumor tissue regions. We identified three key zones: regions containing isolated tumor cells (tumor-only cluster), regions with isolated stromal cells (stromal cluster), and regions where tumor and stromal cells are in direct contact (mixed cluster). For patients with luminal and triple-negative BC, we generated gene expression profiles specific to the contact regions, identified the most active biological processes, and pinpointed characteristic ligand-receptor pairs that may mediate intercellular communication.

Materials and methods

Results

Manual spots annotation

Manual morphological annotation of spots was performed directly within the Loupe Browser software v8.0.0 on hematoxylin and eosin-stained tissue by three independent researchers sequentially, ensuring high annotation accuracy. The interobserver discordance rate was below 5%. Discrepancies were resolved by consensus discussion, involving an experienced pathologist when necessary. Spots were classified into three categories: tumor spots (containing tumor cells without contact with stromal cells), stromal spots (containing tumor microenvironment cells not in contact with tumor cells), and mixed spots (containing both tumor cells and stromal cells in contact) (Figure 1).

Display full size

Figure 1. Classification of spots in the Loupe Browser v8.0.0. A and C: t-SNE; B and D: manual spots annotation.

Classification of spots based on their spatial localization in the tissue was conducted according to the following criteria: tumor spots included only tumor cells and no stromal cells; stromal spots contained only stromal cells of the tumor microenvironment (such as immune cells and fibroblasts) without any tumor cells; mixed spots included both tumor and stromal cells in varying proportions, with each mixed spot comprising at least three tumor cells and at least three stromal cells.

All analyzed tumor tissue regions were localized within the primary tumor mass; transcriptomic features of the invasive tumor margin were outside the scope of this study. Additionally, spots containing no cells, areas with normal mammary gland structures, necrotic regions, vascular structures, erythrocytes, tertiary lymphoid structures, spots situated more than 300 μm from tumor cells, edge spots, as well as spots with artefactual staining or unclear imaging were excluded from analysis.

DEGs and biological processes

To identify transcriptomic features of tumor and stromal cells in direct contact, we performed the following comparisons: mixed spots versus tumor spots and mixed spots versus stromal spots in luminal BC, as well as mixed spots versus tumor spots and mixed spots versus stromal spots in triple-negative BC. Since mixed spots contain both tumor and stromal cells, both corresponding genes and biological processes are functionally active simultaneously within these regions. In the comparison of mixed spots versus tumor spots, DEGs and biological processes related to tumor growth that are enriched in mixed spots were evaluated. Conversely, in the comparison of mixed spots versus stromal spots, DEGs and processes characteristic of stromal cells—including immune cells, fibroblasts, and endothelial cells—were analyzed.

Luminal breast cancer

In luminal BC, the mixed cluster (tumor cells colocalized with stromal cells) showed higher expression of genes encoding S100A family Ca²⁺-binding proteins (S100A4 (log2FC = 0.61, p_adj = 0.0000), S100A8 (log2FC = 1.99, p_adj = 0.0000), S100A9 (log2FC = 1.88, p_adj = 0.0000)), matrix metalloproteinases (MMP2 (log2FC = 0.64, p_adj = 0.0000), MMP7 (log2FC = 0.79, p_adj = 0.0000), MMP14 (log2FC = 0.59, p_adj = 0.0000)), epithelial markers and cytokeratins (KRT5 (log2FC = 0.70, p_adj = 0.0000), KRT7 (log2FC = 0.68, p_adj = 0.0000), KRT15 (log2FC = 0.93, p_adj = 0.0000), KRT23 (log2FC = 0.67, p_adj = 0.0000), KRT81 (log2FC = 0.75, p_adj = 0.0000)), the mesenchymal marker VIM (log2FC = 0.65, p_adj = 0.0000), and genes associated with epithelial-mesenchymal transition (EMT) (ICAM1 (log2FC = 0.79, p_adj = 0.0000), PRRX1 (log2FC = 0.82, p_adj = 0.0000)) compared to tumor-only cluster.

Gene Ontology Biological Process analysis for luminal BC revealed that genes upregulated in mixed cluster were enriched for processes related to regulation of the ERK/MAPK cascade (e.g., Regulation of ERK1 and ERK2 Cascade, GO:0070372, p_adj = 0.0001; Positive Regulation of ERK1 and ERK2 Cascade, GO:0070374, p_adj = 0.0007; Negative Regulation of ERK1 and ERK2 Cascade, GO:0070373, p_adj = 0.0161), apoptosis regulation (e.g., Regulation of Apoptotic Process, GO:0042981, p_adj = 0.0016; Positive Regulation of Programmed Cell Death, GO:0043068, p_adj = 0.0106; Negative Regulation of Programmed Cell Death, GO:0043069, p_adj = 0.0328), and regulation of cell adhesion and migration (Positive Regulation of Cell Junction Assembly, GO:1901890, p_adj = 0.0071; Negative Regulation of Cell Motility, GO:2000146, p_adj = 0.0019; Positive Regulation of Cell Adhesion, GO:0045785, p_adj = 0.0194) compared to tumor-only cluster (Figure 2, Table S2).

Display full size

Figure 2. Biological processes enriched in tumor cells in contact with stromal cells in luminal breast cancer. FDR: false discovery rate (adjusted p-value, Benjamini–Hochberg correction).

Stromal cells adjacent to tumor cells (mixed cluster) within the tumor mass in luminal BC exhibited hyperexpression of only two immune response-associated genes, FDCSP (log2FC = 1.44, p_adj = 0.0000) and CXCL13 (log2FC = 1.11, p_adj = 0.0000), compared to stromal cells neighboring exclusively other stromal cells (stromal cluster), without upregulation of other stromal-related processes. Consistently, enrichment analysis of biological processes did not reveal dominance of immune response, stromal remodeling, angiogenesis, or growth factor production processes in the contact region compared to the stromal areas. All these processes were more active in microenvironmental cells not in direct contact with tumor cells.

Triple-negative breast cancer

In triple-negative BC, the mixed cluster showed hyperexpression of S100A family genes S100A2 (log2FC = 0.97, p_adj = 0.0000), S100A8 (log2FC = 0.60, p_adj = 0.0000), and S100A9 (log2FC = 0.71, p_adj = 0.0000), the epithelial gene KRT6B (log2FC = 0.73, p_adj = 0.0000), the classical cancer stem cell marker CD44 (log2FC = 0.64, p_adj = 0.0000), and the gene NOTCH2 (log2FC = 0.72, p_adj = 0.0000), which is associated with negative regulation of EMT, compared to tumor-only cluster.

Similar to luminal BC, mixed cluster (tumor cells colocalized with stromal cells) in triple-negative BC were characterized by increased activity of processes regulating the ERK/MAPK cascade, including Positive Regulation of ERK1 and ERK2 Cascade (GO:0070374, p_adj = 0.00002), Regulation of ERK1 and ERK2 Cascade (GO:0070372, p_adj = 0.0001), and Positive Regulation of MAPK Cascade (GO:0043410, p_adj = 0.0005). Increased regulation was also observed in apoptosis-related pathways, such as Negative Regulation of Intrinsic Apoptotic Signaling Pathway in Response to DNA Damage by p53 Class Mediator (GO:1902166, p_adj = 0.0005), Negative Regulation of Intrinsic Apoptotic Signaling Pathway by p53 Class Mediator (GO:1902254, p_adj = 0.0010), and Negative Regulation of Intrinsic Apoptotic Signaling Pathway in Response to DNA Damage (GO:1902230, p_adj = 0.0028). Additionally, processes involved in cell adhesion and migration, including Cell Junction Disassembly (GO:0150146, p_adj = 0.0031), were more active compared to the tumor center (Figure 3, Table S2).

Display full size

Figure 3. Biological processes enriched in tumor cells in contact with stromal cells in triple-negative BC. FDR: false discovery rate (adjusted p-value, Benjamini–Hochberg correction).

In regions of colocalization between tumor and stromal cells, genes associated with angiogenesis and immune processes, including VEGFA (log2FC = 0.86, p_adj = 0.0000), CXCL17 (log2FC = 1.74, p_adj = 0.0000), CCL28 (log2FC = 1.45, p_adj = 0.0000), and HLA-A (log2FC = 1.04, p_adj = 0.0000), were more highly expressed compared to stromal areas distant from the tumor, despite their expression also being detected in tumor cells. Biological process enrichment analysis revealed activation of Regulation of Blood Vessel Branching (GO:1905553, p_adj = 0.0233) and Positive Regulation of Leukocyte Chemotaxis (GO:0002690, p_adj = 0.0252) in mixed spots; however, similar processes were also activated in stromal spots.

Luminal versus triple-negative breast cancer

We also compared transcriptomic features between luminal and triple-negative BС across corresponding spot clusters: mixed spots (luminal BC versus triple-negative BC), stromal spots (luminal BC versus triple-negative BC), and tumor spots (luminal BC versus triple-negative BC). Enriched biological processes were identified for each comparison (Table S2). Consistently, processes related to immune response, inflammation, and cytokine signaling were more highly enriched in all triple-negative BC regions compared to corresponding luminal BC regions. Figure 4 presents the top 10 enriched processes in each region for both studied subtypes.

Display full size

Figure 4. Top 10 enriched biological processes in each cluster in luminal and triple-negative BC (breast cancer).

Discussion

Active interaction between tumor cells and stromal components (fibroblasts, immune cells, endothelium) is a driving force in tumor progression. In the study by J. Kaufmann et al. [43], a prognostic parameter called the Tumor-Stroma Contact Ratio was proposed for patients with oropharyngeal cancer. This parameter evaluates the relationship between tumor cells in contact with the stromal surface and the total number of epithelial tumor cells and predicts poor response to (chemo-)radiotherapy, as well as shorter overall and progression-free survival in locally advanced oropharyngeal cancer [43]. This work highlights the particular importance of investigating the characteristics of the direct contact zone between tumor nests and stroma, where the most intense intercellular communication occurs.

In our study, we employed manual morphological annotation of Visium 10x tumor tissue spots, selectively highlighting regions of interest and excluding non-representative areas. This approach provided preliminary evidence that tumor cells in contact with stromal cells in luminal BC show increased expression of genes including S100A4, S100A8, S100A9, MMP2, MMP7, MMP14, KRT5, KRT7, KRT15, KRT23, KRT81, VIM, ICAM1, and PRRX1. In triple-negative BC, the contacting tumor cells showed increased expression of S100A2, S100A8, S100A9, KRT6B, CD44, and NOTCH2.

The functional versatility of S100 proteins derives from their ability to interact with numerous targets, enabling their involvement in critical processes such as Ca²⁺ signaling, cell growth, differentiation, apoptosis, inflammation, and motility [44]. Data regarding their prognostic role are contradictory [44], but their detection at the tumor boundary may indicate increased intra- and intercellular interaction activity.

MMPs play a critical role in cancer metastasis by degrading the extracellular matrix and blood vessel walls, enabling tumor cells to invade and spread to distant organs [45]. In luminal BC, higher expression of MMP genes was observed in tumor cells at the tumor-stroma edge, whereas in triple-negative BC, no differences in MMP gene expression were found between mixed and tumor clusters. This may indicate that both edge and central tumor cells are equally adapted to interact with and degrade the extracellular matrix, potentially contributing to the aggressiveness of this subtype. Comparison of tumor-stromal co-localization regions between luminal and triple-negative BC showed a relatively higher enrichment of immune response‑associated GO terms in triple‑negative BC, including those related to innate immunity, inflammation, immune cell recruitment, secretion, and cytokine signaling. Notably, these processes were even more highly enriched in triple-negative BC stromal regions not in contact with tumor cells and in tumor regions not in contact with stroma compared to luminal BC. Therefore, the enrichment of immune-related processes in triple-negative BC characterizes not only the tumor-immune cell interface but also the entire tumor mass. This indicates that heightened immune activity results not solely from direct tumor-immune cell contact but also from active paracrine interactions between spatially separated cell populations.

Interestingly, we detected hyperexpression of the CD44 gene in tumor cells at the tumor-stroma interface in triple-negative BC. CD44 is strongly linked to cancer stem cell hallmarks, notably self-renewal, EMT, therapy resistance, and enhanced tumorigenicity. In the study by O. Mansour et al. [46], CD44 expression showed significant positive correlations with BRCA1, p38 MAPK, E-cadherin, DSG1, HOTAIR, and BC200, suggesting its involvement in both stemness and tumor progression pathways in BC [46]. Expression of stemness markers in the CD44^hi/CD24^low and ALDH^hi combination in tumors of BC patients correlates with shorter disease-free survival and overall survival in those receiving neoadjuvant chemotherapy [47]. The elevated expression of CD44 at the tumor-stroma interface observed in our study underscores its pro-tumorigenic functional significance in BС. However, it has also been shown that ALDH1⁺ cancer stem cells demonstrate superior self-renewal and tumorigenic abilities compared to CD44^hi/CD24^low/– cancer stem cells across different molecular subtypes of BC [48]. Differences in the expression patterns of cytokeratin genes and EMT-associated genes might reflect fundamentally distinct strategies of invasion and cellular plasticity between BC subtypes.

In both BC subtypes, the set of genes upregulated in tumor cells colocalized with stromal cells was enriched for GO terms related to regulation of the ERK/MAPK cascade, apoptosis, cell adhesion, and migration. The ERK/MAPK signaling pathway is closely associated with cell proliferation and differentiation and plays a pivotal role in the cellular signal transduction network. It also participates in extracellular matrix degradation by upregulating MMPs expression, thereby promoting tumor invasion and metastasis [49]. It is also known that ERK1/2 can exert pro-apoptotic effects [50], whereas the JNK and p38 MAPK cascades mediate pro-apoptotic processes [51].

Apoptosis is recognized as a process with “double-edged” consequences: on one hand, it is tumor-suppressive by eliminating malignant or pre-malignant cells; on the other hand, it promotes tumor progression by stimulating reparative and regenerative responses in the tumor microenvironment [52]. The enrichment of these processes in co-localized regions reflects not only the high invasive potential of tumor cells in these areas but also the intensified cellular response to growth-restrictive conditions in the surrounding microenvironment. This phenomenon is characteristic of both luminal and triple-negative BC. On the other hand, activation of these pathways in tumor cells may be induced by factors secreted by stromal cells. For example, the tumor microenvironment can activate the JAK/STAT pathway through factors such as IL-6, which synergizes with ERK signaling to further diminish the efficacy of targeted therapies [53]. Additionally, induction of the extrinsic apoptosis pathway in tumor cells requires interaction between death receptors and their corresponding ligands (TNF-α, FasL, and TRAIL), which are predominantly expressed in immune cells, including granulocytes, monocytes, T cells, B cells, dendritic cells, and NK cells [54].

In regions of tumor-stromal cell contact in luminal BС, despite its characterization as an immunologically cold tumor, a broader spectrum of genes encoding various HLA class II molecules co-expressed with the CD4 gene was observed compared to triple-negative BC. Our data corroborate findings obtained using a different approach by the research group led by Y. Cui et al. [55], which identified a distinct tumor subpopulation that upregulates MHC-II genes and actively interacts with immune cells in triple-negative BC [55]. According to recent studies, effective antitumor immunity in BC may be linked to MHC-II-restricted presentation of optimal tumor antigens and requires the cooperation of CD4⁺ T cells, B cells, and antigen-specific antibodies [56]. In experimental models, MHC-II-restricted neoantigen vaccination enhances inflammatory signaling within the tumor microenvironment of “cold” tumors, increases infiltration of CD4⁺ and CD8⁺ T cells and IFN-γ production, and exhibits synergy with immune checkpoint inhibitors [57]. Given that luminal BC subtypes are classified as immunologically “cold” tumors, this strategy holds promise for their treatment. In this context, our observed enrichment of HLA-II/CD4 pairs in areas of direct contact between tumor and stromal cells suggests that these spatial niches may serve as potential targets for MHC-II-directed immunotherapy, particularly in patients with luminal BC.

Other notable immune cell interactions detected include the APP–CD74 axis. APP expressed on tumor cells binds to CD74 on TAMs, delivering an inhibitory signal that suppresses TAMs phagocytic activity [24]. This mechanism enables tumor cells to evade immune-mediated destruction, thereby promoting tumor progression. In a study by H. Zeng et al. [58], a ligand‑receptor pair‑based signature score derived from on‑treatment tumor specimens was developed to predict immune checkpoint blockade response in metastatic melanoma. This score included seven ligand‑receptor pairs, among them APP–CD74 [58]. Another study by O. Chen et al. [59] demonstrated that modulating tumor‑associated macrophages through APP–CD74 blockade using IL4R‑exosomes synergizes with PD‑1 inhibition in gastric cancer mouse models [59]. In BC, the prognostic or therapeutic significance of this axis has not been established and requires validation.

In our study, tumor-stroma co-localization regions in luminal BC exhibited a greater number of ligand-receptor pairs involved in tight junction formation between tumor cells and immune cell recruitment to the tumor compared to triple-negative BC. Such interactions may restrict tumor growth and enhance anti-tumor immune responses. Notably, a broad range of these ligand-receptor pairs included integrins composed of various α- and β-subunits. Integrin expression and activity are closely associated with multiple stages of tumor development, such as initiation, angiogenesis, motility, invasion, and metastasis. While some integrins promote tumor formation or potentiate oncogenic signaling through receptor interactions, others have minimal or even inhibitory effects [60]. Previous studies have investigated the spectrum of integrin interactions with other molecules, demonstrating their high heterogeneity in BC [12]. The participation of integrins in a wide range of molecular partnerships likely underlies their functional duality. Anti‑integrin therapy is currently used in the treatment of ulcerative colitis and Crohn’s disease [61]. Accumulating evidence also supports the potential of anti‑integrin approaches in cancer. For instance, inhibition of integrin αVβ3 has been shown to induce cytotoxicity and suppress migration in ovarian cancer cells [62]. Moreover, integrin‑targeting molecules may be employed to improve the efficiency of targeted drug delivery to tumors [63].

Also, we identified an expanded repertoire of JAM family interactions in luminal BC, including homotypic (JAM2-JAM2, JAM3-JAM3) and heterotypic (JAM2-JAM3) contacts as well as integrin-mediated pairs (JAM2-ITGAV_ITGB1, JAM2-ITGA3_ITGB1, JAM3-ITGAM_ITGB2). Given that JAM3 acts as a tumor suppressor in BC brain metastases [37] and JAM2 is involved in leukocyte trafficking and angiogenesis [35], the enrichment of these contacts specifically at the tumor-stroma interface suggests they may contribute to the less aggressive phenotype of luminal BC.

Our study aligns with the rapidly expanding field of spatial transcriptomics in BC, an approach that holds promise for refining the molecular classification of this disease and identifying spatial tumor microenvironment structures relevant to the development of personalized therapy [64, 65]. A key advantage of spatial approaches is that they preserve tissue architecture, which is critically important for investigating interactions between tumor and stromal cells. Currently, few studies comprehensively investigate the characteristics of tumor and stromal cells in direct contact. For example, Wu et al. [8] demonstrated that the tumor-stroma edge is characterized by extensive extracellular matrix remodeling, immunomodulatory regulation, and EMT. They also identified significant interactions between cancer-associated fibroblasts (CAFs) and M2-like TAMs that contribute to immune exclusion and drug resistance [8]. Additionally, the spatial organization and immune status of the tumor-stroma edge were identified as distinctive features differentiating mismatch repair-deficient (dMMR) and proficient (pMMR) colorectal cancers, correlating with responses to immune checkpoint blockade therapy [66]. Importantly, these studies investigated the invasive front—the tumor edge with adjacent normal tissue. In contrast, our study focused on the characteristics of tumor and stromal cells located within the tumor mass itself. Our approach is consistent with existing clinical guidelines for the assessment of TILs in BC, which restrict analysis to the stroma within the invasive margin of the tumor rather than the invasive edge with adjacent normal tissue [67]. It is within these internal tumor regions that therapy-resistant tumor cell populations and progression-driving clones emerge, alongside high concentrations of immunosuppressive cells. Thus, studies of tumor and stromal cells in direct contact within the tumor nodules may provide refined insights into mechanisms of tumor progression and identify novel therapeutic targets for malignant neoplasms.

The main limitation of our study is the cohort size—19 patients (10 with luminal BC and 9 with TNBC). On the one hand, this number is comparable to many pilot spatial transcriptomics studies, and the large number of spatial spots analyzed per sample provides adequate statistical power. On the other hand, the high interpatient heterogeneity of BC, particularly in the triple-negative BC group, raises the possibility that some variants of gene expression or ligand-receptor pairs may have remained undetected. Therefore, the presented results should be considered hypothesis‑generating; independent validation cohorts using additional analytical methods (multiplex immunohistochemistry, spatial sequencing on other platforms) are required to confirm these findings.