Single-cell and spatial transcriptome profiling identifies cellular heterogeneity and immunosuppressive tumor microenvironment in inflammatory breast cancer.

Introduction
Inflammatory breast cancer (IBC) is a rare yet highly aggressive and lethal subtype of breast cancer, accounting for approximately 2–4 % of locally diagnosed cases but contributing to 7–10 % of breast cancer-related mortality [1]. It is distinguished by the absence of a primary tumor and the swift buildup of cancerous epithelial cells within dermal lymphatic vessels. IBC presents with rapid-onset inflammatory symptoms such as widespread redness and swelling (peau d'orange) in the breast, yet without detectable palpable mass, making it initially challenging to distinguish from inflammatory breast disease, such as acute mastitis [[2], [3], [4]]. The enigmatic origin of IBC cells is still not fully understood, although studies have shown that approximately 11.1 % of IBC cases arise from non-IBC in the same breast [5]. The current treatment strategies for IBC involve surgery, adjuvant locoregional radiotherapy, and neoadjuvant chemotherapy [6,7]. Despite these multimodal therapies, the prognosis for IBC patients remains considerably poorer compared to non-IBC cases [[8], [9], [10]]. Due to the limited understanding of its underlying mechanisms and the absence of a distinct genomic/transcriptomic signature, targeted therapies tailored specifically to manage IBC have yet to be developed. Therefore, a comprehensive understanding of the pathological and biological basis specific to IBC is crucial to pave the way for more precise therapeutic strategies for this aggressive disease.
IBC is not only phenotypically different, but also exhibits unique molecular features that distinguish it from other forms of breast cancer. Efforts have been directed towards elucidating the molecular signature of IBC through direct comparisons with non-IBC cases, even though facing challenges in reproducibility and specificity. A previous study employed a nonparametric random forest (RF) machine learning approach and identified a robust IBC-specific gene signature, termed G59, composed of 59 genes [11]. This signature was found to be independent of ER/HER2 status and molecular subtypes, demonstrating particular efficacy in pre-treatment samples. Functional analysis revealed enrichment for plasma membrane proteins and interleukin signaling pathways, suggesting a potential role in tumor-microenvironment interactions. In addition to molecular profiling, immune alterations in IBC have also been investigated. A study analyzing peripheral blood from 14 stage IV metastatic IBC patients and 11 age-matched healthy women using flow cytometry revealed significant lymphopenia in IBC patients [12]. Notably, reductions were observed across T, B, and NK cell populations, with CD4+ T cell subsets being particularly affected. While IBCs exhibit lower basal frequencies of Th1 and Th2 groups [13], differences in immune cells between IBCs and non-IBCs were believed to arise from variations in subpopulations rather than overall immune cell quantities. Unfortunately, such nuances cannot be captured through solely bulk RNA-seq analysis of samples. Single-cell RNA-sequencing (scRNA-seq) provides an opportunity to systematically describe the cellular landscape of tumors, uncover intratumor heterogeneity [14], characterize tumor evolutionary lineages [15], identify rare subpopulations [16], and gain insights into cell phenotypes specific to a disease. Generating a comprehensive transcriptional atlas of IBC and elucidating cellular components and their interactions in the tumor microenvironment is crucial for understanding the underlying molecular mechanisms of this disease and developing more precise therapeutic strategies to combat IBC.
In this study, we applied scRNA-seq to comprehensively characterize IBC tumor microenvironments at the subpopulation level. The samples included in this study represented a diverse range of breast diseases, including different molecular subtypes of IBC and non-IBC, as well as normal breast tissue controls. Comparative analyses of these samples revealed a compromised immune system in IBC patients, with decreased CXCL13 expression in T cells and down-regulated inflammatory genes in immune and tumor cells. Overexpression of CXCL13 in tumor cells, under co-culture with immune cells, significantly promoted tumor cell death and CXCL13 can also enhance the efficacy of anti-PD-1 therapy. Cell-cell interactions between immune and tumor cells were significantly weaker in IBCs compared to non-IBCs. These findings highlight the intrinsic heterogeneity within the tumor microenvironment of IBCs and the potential contribution of impaired immune responses to the unfavorable clinical outcomes. To identify potential therapeutic compounds for IBC, natural products screening was performed, and sanguinarine and α-mangostin were identified to have significant inhibitory effects on IBC cells and upregulate inflammatory gene expression. These findings provide valuable insights for precision medicine strategies in IBC treatment, offering opportunities to target the unique characteristics of this challenging disease.

Materials and methods

Clinical samples collection
Human tissue samples were provided by the First Affiliated Hospital of Zhejiang Chinese Medical University under an approved protocol by the local medical ethics committee (2021-KL-148-02). All patients were required to provide written informed consent. Immunohistochemistry was performed according to the manufacturer’s instructions. The monoclonal antibodies were used for Ki67, ER, PR, and HER-2 staining. The diagnosis of IBC was confirmed independently by a pathologist using AJCC-accepted clinical and pathological criteria. According with the results from ER, PR, HER2, and Ki67, the tumor samples included 3 Luminal A, 2 Luminal B, 3 HER2+, 2 triple negative, and 3 IBC. Among them, IBC3 patient received 4 cycles of neoadjuvant chemotherapy (docetaxel + carboplatin + trastuzumab) before sample collection. In addition, 2 normal breast tissues were obtained.

Single-cell RNA-sequencing
Tissue samples were immediately subjected to dissociation on ice after collection. A Countess II Automated Cell Counter was used to count single cells in the suspensions, and these single cell suspensions were loaded onto 10 × Chromium, following the manufacturer's instructions. The subsequent steps, including cDNA amplification and library construction, were carried out according to the standard protocol. The libraries were sequenced on an Illumina sequencing system. Detailed information of sample processing for single-cell RNA-sequencing is provided in Supplementary methods.

scRNA-seq data analysis
The processing of raw reads was executed utilizing the Cell Ranger Single-Cell Software Suite (version 6.0.1, 10 × Genomics Inc., CA, USA). The primary analyses of the data involved various steps, including alignment, filtering, barcode counting, and UMIs quantification to ascertain gene transcript counts per cell. Quality control, clustering, and statistical analysis were subsequently performed using the CellRanger count command. Genes were annotated based on Ensembl build 93. The gene expression matrices, produced per sample through CellRanger, were brought into R (version 4.0.0) and transformed into a Seurat object using the Seurat R package (version 4.1.0). Subsequently, ambient RNA was cleared using DecountX (R package celda, version 1.4.7). To enhance data quality, dead cells and doublets were systematically eliminated. The remaining cells were integrated and batch effects were corrected using rPCA in Seurat. To pinpoint marker genes associated with specific cell types, a comparison was made between the gene expression values of cells from the cluster of interest and those from the remaining clusters using the Seurat FindMarkers function with the default 'MAST' test [17]. Trajectory and pseudotime analysis were conducted using monocle2 (version 1.0.0) [18]. CellChat was utilized for the inference and analysis of cell–cell communication. Detailed information of scRNA-seq data analysis is provided in Supplementary methods.

Gene set enrichment analysis
For the calculation of differential expression genes between two groups of cells, the Seurat FindMarkers function with the 'MAST' method was applied. Parameters included ‘min.pct = 0.01′ and ‘logfc.threshold = 0.01′. Differentially expressed genes with adjusted p-values below 0.05 were discerned among various disease conditions of cell types using the 'FindMarkers' function. Following this, 'clusterProfiler' was utilized to identify GO and KEGG pathways enriched by these differentially expressed genes. Terms with a false discovery rate (FDR) less than 0.05 were considered as significantly enriched.

Cell culture
The MCF-10A cell line was obtained from the Institute of Shanghai Biochemistry and Cell Biology (Shanghai, China). MCF7, BT-474, MDA-MB-231, and MDA-MB-453 cell lines were purchased from the American Type Culture Collection (ATCC, Rockville, MD). Inflammatory breast cancer cell lines SUM149PT (IBC-TNBC) and KPL-4 (IBC-ER+, PR+, HER2+) were obtained from the Institute of Shanghai Biochemistry and Cell Biology (Shanghai, China). The cell lines MCF7, MDA-MB-231, MDA-MB-453, SUM149PT, and KPL-4 were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) with 10 % fetal bovine serum (FBS) and double antibiotic. BT-474 was maintained in RPMI 1640 medium supplemented with 10 % FBS and double antibiotic. All cells were cultured in a humidified incubator at 37 °C with 5 % CO2.

Co-culture experiments
Mouse T cells were isolated from the spleens of 6–8 week-old BALB/c mice using the EasySep™ T Cell Isolation Kit (STEMCELL, #19851) and activated with Mouse CD3/CD28 T Cell Activation Beads (Elabscience, MIM001A). T cell proliferation was promoted by supplementing IL-2 (MCE, HY-P7077) at a concentration of 8 ng/mL. 4T1 cells were transfected to overexpress CXCL13, with normal 4T1 cells serving as controls. When 4T1 cells reached appropriate confluence, activated T cells were added for co-culture. After 24 h, the culture supernatants were collected, centrifuged to remove cells, and analyzed for TNF-α (CZKEWEI, SU-BN20852), Granzyme B (CZKEWEI, SU-BN20140), and IFN-γ (CZKEWEI, SU-BN20455) levels by ELISA. Co-cultured cells were subsequently harvested for the assessment of 4T1 cell apoptosis by flow cytometry. In parallel, for macrophage-related assays, normal or CXCL13-overexpressing 4T1 cells were seeded in the lower chambers of six-well plates, and RAW264.7 macrophages were placed in the upper inserts for indirect co-culture. After 24 h, RAW264.7 cells were collected, and macrophage polarization was evaluated. M1-type macrophages were labeled with anti-CD86 antibodies and M2-type macrophages with anti-CD206 antibodies, followed by flow cytometric analysis. Additionally, the expression levels of iNOS, TNF-α, and IL-6, markers of M1 polarization, were quantified by qPCR. To assess the impact on tumor cell survival, 4T1 cells were collected after 24 h of co-culture with macrophages, and apoptosis was detected by flow cytometry. For phagocytosis assays, RAW264.7 macrophages and 4T1 cells were pre-labeled with Dil (YEASEN, 40726ES10) and Cellhunt Green CMFDA (Aladdin, 136832-63-8), respectively, mixed at a 2:1 ratio, and co-cultured for 24 h. Macrophage phagocytosis of tumor cells was then observed under a fluorescence microscope.

In vivo experiments
The 4T1-luc cells were constructed and 5 × 105 of these cells were inoculated into the fourth mammary fat pads of 6–8 week-old female BALB/C mice. Tumor volume was measured every second day and calculated as (length × width2)/2. The mice were divided into two groups: one group treated with anti-mouse PD-1 alone (200 μg) and the other group treated with the combination of anti-PD-1 and recombinant mouse CXCL13. Anti-PD-1 administration (200 μg) was carried out every three days starting from the 7th day after tumor implantation. Recombinant mouse CXCL13 (rmCXCL13, 5 μg per mouse) was intraperitoneally injected starting from the 7th day after tumor implantation and then administered every other day until the end of the experiment.

Results

Single-cell transcriptome atlas of normal breast and IBC/non-IBC cancerous tissues
To gain insights into the cellular architecture of IBCs and to differentiate them from non-IBCs and normal breast tissue, a comprehensive analysis of scRNA-seq was performed. The study protocol received approval from the local medical ethics committee and included 18 biopsies: three IBCs (IBC1: Luminal B, IBC2: Luminal A, and IBC3: HER2-enriched), 13 non-IBCs (comprising 4 LABC, 2 LBBC, 5 HER2-enriched, and 2 TNBC), as well as 2 normal human breast tissue samples (Fig. 1A). The molecular subtypes of breast cancer were classified according to clinical criteria based on the combined expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2), yielding four primary molecular subtypes: Luminal A (LABC: ER+, PR > 20 %, HER2−, and Ki-67 < 15 %), Luminal B (LBBC: ER+, PR < 20 %, HER2−, and Ki-67 ≥ 15 %), HER2-enriched (HPBC: ER−, PR−, and HER2+), and triple-negative (TNBC: ER−, PR−, and HER2−). The molecular subtypes of IBCs were defined using the same criteria as those applied to non-IBCs. Detailed information regarding the expression levels of ER, PR, HER2, and Ki-67 for each patient included in this study is available in Supplementary Table S1, while pathohistological information is presented in Supplementary Fig. S1. Following the application of cell filtering and normalization, a total of 170,865 cells exhibiting high-quality transcriptomic data were obtained. These cells were categorized into eight distinct clusters (Fig. 1B), including epithelial cells, endothelial cells, plasma cells, fibroblasts, myeloid cells, mast cells, T cells, and B cells, utilizing established cell type markers (Fig. 1C). Cell fraction analysis for each sample revealed a diverse array of cell types across all analyzed groups (Fig. 1D). Consistent with previous studies on breast cancer [19], substantial variations in the proportions of the identified cell populations were observed among different samples, reflecting significant inter-patient heterogeneity. Notably, a higher overall frequency of immune cells, including T cells, myeloid cells, and B cells, was detected within tumor tissues, both IBC and non-IBC, compared to healthy controls, suggesting a redirected immune response in cancer-affected individuals (Fig. 1E). Furthermore, when comparing IBC to non-IBC samples, a reduced overall frequency of T cells was observed in IBC cases (Fig. 1F). These results suggest that the tumor microenvironment of IBC may impede the effective recruitment of T cells to the tumor site.

The fraction of exhausted CXCL13+ CD4+ T cells was reduced in IBC tumors
Building upon the aforementioned results, a comprehensive analysis of T cell subpopulations was performed to further investigate the intricate immune landscape of IBCs at a high-resolution level. This analysis revealed 14 distinct T cell subgroups across the entire cohort (Fig. 2A), identified based on the expression of specific cell subtype markers (Fig. 2B and Supplementary Fig. S2A). Among the CD4+ subgroups, identifiable populations included CD4 naïve cells, CD4 TCM, CD4 Treg, and CD4 exhausted cells, each exhibiting a unique expression profile characterized by marker genes. Similarly, the CD8+ subgroups comprised CD8 naïve cells, CD8 TEM, CD8 TCM, CD8 TRM, CD8 TEMRA, and CD8 exhausted cells. In addition to these established T cell subsets, two additional populations were identified: mucosal-associated invariant T cells (MAIT) and MKI67-expressing T proliferating cells (CD4+/CD8+ T cells). Furthermore, NK cells expressing marker genes such as NKG7 and GNLY, as well as NK proliferating cells characterized by the expression of ANXA2, STMN1, and MKI67, were detected based on their distinct marker gene expression patterns.
Comparative analysis of the cell composition of T cell subtypes revealed significant differences between IBCs and non-IBCs (Fig. 2C). Using pathological immunohistochemistry to analyze ER, PR, HER2, and Ki-67, IBC1 was classified as resembling a luminal B subtype (LBBC), while IBC2 and IBC3 were categorized as luminal A (LABC) and HER2-enriched subtypes (HPBC), respectively. Notably, when comparing matched molecular subtypes (IBC1 vs. LBBC and IBC2 vs. LABC), a consistent pattern emerged, characterized by a reduced fraction of CD4+ exhausted T cells (CXCL13+ CD4+ T cells) in IBCs compared to their non-IBC counterparts (Fig. 2D). However, this difference was not evident in the IBC3 patient, likely due to clinical interventions administered prior to sample collection. This reduction in CXCL13+ CD4+ T cells was especially pronounced within the total T cell population of the IBC1 patient, who exhibited the poorest clinical prognosis, suggesting that this cell cluster may play a critical role in the progression of IBC tumors. To investigate the biological functions of the specific cluster of CXCL13+ CD4+ T cells, pathway enrichment analysis was conducted using marker genes identified in this cluster via FindMarkers. The results indicated that the marker genes of CXCL13+ CD4+ T cells were significantly enriched in KEGG pathways associated with Th17 cell differentiation, Th1 and Th2 cell differentiation, and the PD-L1 expression and PD-1 checkpoint pathway in cancer (Fig. 2E), underscoring the crucial role of this cell group in mediating immune responses against tumor cells.
CXCL13+ CD4+ T cells are a specific subgroup of helper T cells that play an important role in the tumor microenvironment, and their presence is associated with improved clinical response [20]. This cell cluster identified in this study was characterized by high expression of the chemokine CXCL13 (Supplementary Fig. S2A), which is involved in the recruitment of B cells and other immune cells. The previous study has established a positive correlation between CXCL13+ T cells and favorable responses to anti-PD-L1 therapy in breast cancer [21]. To further explore the clinical implications of CXCL13 expression in breast cancer, we assessed its correlation with cancer stages and patient survival outcomes. The analysis consistently showed that CXCL13 expression is relatively high in early-stage breast cancer (Fig. 2F) and that elevated CXCL13 levels significantly predicted improved patient outcomes (Fig. 2G), aligning with previous reports [22]. This promoted us to speculate that the unfavorable clinical outcome of IBC1 patient may be related to the significant reduction of CXCL13+ CD4+ T cells.

CXCL13 is significantly downregulated in IBC and its expression positively correlates with anti-tumor immune factor levels in breast cancer
T cell activity in the tumor microenvironment is one of the keys to determining the efficacy of cancer therapy. Activated T cells can recognize and attack tumor cells, contributing to anti-tumor immunity. To investigate the differences in T cell activity between IBC and non-IBC, we performed a comparative analysis of differentially expressed genes (DEGs) in T cells obtained from patients with IBC and those with non-IBC. This analysis revealed a significant and widespread downregulation of multiple genes, including GZMB, GNLY, and IFITM1, in T cells derived from IBC patients compared to those from non-IBC patients (Fig. 3A). The expression of GZMB, GNLY, and IFITM1 is closely linked to T cell activity, particularly in the context of immune responses [[23], [24], [25]]. Increased expression of GZMB, GNLY, and IFITM1 in T cells indicates enhanced activation and cytotoxicity, while reduced expression of these genes in T cells in the context of cancer indicates a diminished cytotoxic response and impaired immune surveillance. Specifically, reduced GZMB and GNLY levels may lead to decreased apoptosis / pyroptosis of cancer cells [26,27], while reduced IFITM1 expression may hinder effective antitumor activity [28]. IFITM1 expression is closely associated with interferon (IFN) signaling, especially in the context of immune responses. IFNs is a class of cytokines produced primarily by activated T cells in the tumor microenvironment and plays a key role in regulating the immune response. Their reduction may lead to T cell dysfunction and an aggressive phenotype of IBC. Supporting this, gene set enrichment analysis of DEGs in T cells reveals that interferon alpha and gamma response signaling pathways were significantly down-regulated in T cells of IBC patients (Fig. 3B). This finding highlights the reduced activity of T cells in the IBC immune microenvironment.
It is important to note that CXCL13 emerges as the most significantly downregulated gene in the majority of T cells, as illustrated in Fig. 3A. This chemokine plays a key role in immune cell recruitment and inflammation, and interacts with various immune factors to form a complex immune response network. As a result, CXCL13 influences both anti-tumor immunity and tumor progression. In the context of breast cancer, our gene expression correlation analysis, using data from The Cancer Genome Atlas (TCGA) via GEPIA online tool, identified a strong positive correlation between CXCL13 mRNA levels and genes linked to T cell activity, such as IFNG, GZMB, and GZMK (Fig. 3C). Furthermore, a significant positive correlation was also observed between CXCL13 and immune checkpoint-related genes, including PDCD1, CTLA4, and LAG3 (Fig. 3D). These findings suggest that CXCL13 may be associated with increased cytotoxic activity of T cells and may favor tumor response to checkpoint blockade therapy. The significant downregulation of CXCL13 in IBCs may indicate suppression of T cell activity in the tumor microenvironment, possibly contributing to immune escape and the construction of a “cold” tumor microenvironment.

Downregulation of CXCL13 is associated with reduced immune infiltration in breast cancer
CXCL13-expressing cells are crucial in coordinating the immune response, especially in lymphoid tissue and tertiary lymphoid structures (TLS) [29,30]. Elevated levels of CXCL13 in the tumor microenvironment can attract immune cells, such as T and B cells, potentially enhancing anti-tumor immunity. This immune cell recruitment may lead to the formation of TLS, which supports local anti-tumor immune responses and is often associated with better clinical outcomes. Given the significant downregulation of CXCL13 observed in IBC tumors, we subsequently explored the correlation between CXCL13 expression and the immune microenvironment in breast cancer. Initial UMAP analysis of CXCL13 expression revealed that this chemokine is predominantly expressed by T cells in breast cancer (Fig. 4A). Notably, T follicular helper (Tfh) cells were shown to express high levels of CXCL13 in breast cancer (Fig. 4B), based on data from TCGA accessed through the GEPIA database. Furthermore, we identified a strong correlation between the proportion of Tfh cells and the frequencies of M1 macrophages and CD8+ T cells (Fig. 4C). This correlation further underscores the potential role of CXCL13 in modulating the immune landscape and enhancing anti-tumor immune activity.
To further investigate the relationship between CXCL13 expression and immune infiltration in breast cancer, we employed the Sangerbox 3.0 immune infiltration analysis. Invasive breast carcinoma (BRCA) datasets were extracted from the UCSC database, and expression data for the gene ENSG00000156234 (CXCL13) were obtained from each sample. Gene expression profiles for each tumor were extracted, and expression profiles were mapped to GeneSymbol to calculate the immune scores for each patient based on gene expression. Spearman correlation between gene expression and immunoinfiltration scores for each tumor was calculated. This analysis revealed a significant positive correlation between CXCL13 expression and immune infiltration in BRCA, with a p-value of 2.2e-111 and a correlation coefficient (r) of 0.61 (Fig. 4D). Additionally, we re-evaluated the infiltration scores of various immune cell types based on gene expression data, which included T cells, CD8+ T cells, cytotoxic lymphocytes, B lineage cells, NK cells, monocytic lineage cells, myeloid dendritic cells, neutrophils, endothelial cells, and fibroblasts for each tumor using Sangerbox 3.0 online tool. As a result, we observed a significant positive correlation between CXCL13 and most immune cell types, particularly T cells and cytotoxic lymphocytes (Fig. 4E). Further analysis revealed that CXCL13 expression exhibited a significant positive correlation with M1 macrophages and a negative correlation with M2 macrophages (Fig. 4F). Additionally, CXCL13 was found to be closely correlated with various immune factors, including chemokines such as CXCL9, CXCL10, and CXCL11 (Supplementary Fig. S2B). These chemokines play a crucial role in the immune response, especially in the context of cancer, primarily by attracting T cells and other immune cells to inflammatory and tumor sites [31]. This recruitment helps regulate immune surveillance, tumor suppression, and response to therapeutic interventions. Overall, these findings support a strong association between CXCL13 and the anti-cancer immune response in breast cancer.
Inspired by the strong correlation between CXCL13 expression and immune infiltration in breast cancer, we conducted spatial transcriptome analysis to investigate the differential infiltration of lymphocytes in IBC versus non-IBC using the NanoString GeoMx Digital Spatial Profiler (GeoMx DSP) spatial multi-omics platform, focusing on multiple spatial locations. Visual markers for tumor and immune cells included nuclear staining with SYTO13 (blue), PanCK (green), and CD45 (red). The presence of CD45-positive cells in tumor tissue is particularly significant, as it typically indicates immune cell infiltration within the tumor microenvironment. The DSP immunofluorescence results revealed a significant reduction in CD45-positive cells within the tumor tissues of IBC patients compared to those in non-IBC patients (Fig. 4G), suggesting that IBC tumors exhibit overall low immune infiltration, potentially indicative of “cold” tumor characteristics.

The interaction between exhausted T cells and other immune cells via the CXCL13-CXCR3 axis is significantly reduced in IBC tumors
Cell-cell communication plays a key role in shaping the intricate tumor microenvironment. The aforementioned findings prompted us to investigate the potential interactions between CXCL13+ T cells and other cell populations. To achieve this, we employed CellChat to quantitatively characterize and compare the predicted intercellular communication networks in both IBC and non-IBC tumors. The ligand-receptor interactions that serve as these complex communication mediators are critical to elucidating the interactions and their contribution to the tumor microenvironment. The analysis revealed a greater number of predicted receptor-ligand interactions in IBC tumors; however, there was a notable decrease in the strength of signaling interactions compared to non-IBC tumors (Fig. 5A). The heatmap of differential interaction numbers indicated an overall increase in communication among T cells, macrophages, and DCs in IBC, particularly among specific macrophage subgroups (macro_IL1B, macro_MRC1, macro_CXCL10, macro_CCL18, macro_SPP1, and macro_CCL5), although this communication occurred with significantly weaker signaling strength (Fig. 5B). This difference was particularly pronounced in the reduced signaling received by T cells in IBC compared to their non-IBC counterparts.
To further investigate how cells interact and communicate differently in IBC tumors, we compared cellular signaling between the IBC and non-IBC groups and calculated the information flow within each signaling pathway. Our analysis revealed a specific enrichment of RESISTIN, PARs, and PERIOSTIN within IBCs, while FLT3, EGF, and CSF exhibited exclusive enrichment in non-IBC tumors (Fig. 5C). Notably, several pathways, including TGF-β, MK, IL16, COMPLEMENT, TNF, GALECTIN, MIF, XCR, VISFATIN, BTLA, CD40, CCL, GRN, SEMA3, IGF, and CXCL, showed significant downregulation in IBC tumors and increased activity in non-IBC tumors. This decreased activity in these signaling pathways may play a critical role in shaping the “cold” tumor characteristics associated with IBC by reducing interactions with tumor-infiltrating lymphocytes.
To further elucidate the critical signaling mechanisms and potential mediators of immune and tumor cell interactions within IBCs, we focused on the top ligand-receptor interactions enriched in epithelial cells, T cells, macrophages, and DCs. Comparative analyses of these interactions between IBC and non-IBC tumors revealed notable differences. Specifically, the ligand-receptor pairs MIF-(CD74 + CXCR4) and MIF-(CD74 + CD44) emerged as predominant signaling pathways through which tumor epithelial cells communicate with immune cells in both IBCs and non-IBCs (Supplementary Fig. S2C and Table S2). Importantly, we identified significantly enhanced communication between tumor epithelial cells and various macrophage subgroups mediated by ANXA1-FPR1 signaling within IBCs (Supplementary Fig. S2D). This signaling pathway has previously been implicated in fostering a pro-tumor immune microenvironment in breast cancer by promoting the polarization of M2 macrophages [32], warranting further investigation. Conversely, the MDK-NCL signaling pathway, which originates from tumor epithelial cells and influences specific subgroups of T cells, macrophages, and DCs, was exclusively identified in non-IBCs (Supplementary Fig. S2D). This finding suggests distinct differences in cell communication-dependent signaling pathways between IBC and non-IBC tumors.
Additionally, our analysis revealed that CXCL signaling pathways were more enriched in non-IBC tumors compared to IBC tumors (Fig. 5C). We subsequently explored the CXCL signaling pathway network, with the results depicted in Fig. 5D as circular plots. Within this pathway, macrophages were identified as the primary signal senders, regulating other cells within the immune microenvironment. Notably, in IBC tumors, the predominant macrophage types acting as signal senders were MRC1+ macrophage (associated with M2 macrophage) and IL1B+ macrophage populations, both of which are strongly associated with the inhibition of anti-tumor immune responses. These populations also exhibited evident autocrine signaling within IBC tumors. In contrast, non-IBC tumors displayed CXCL signals originating from diverse cell types, with particularly robust contributions from CXCL10+ macrophages and CXCL13+ CD4+ T cells (exhausted CD4+ T cells). In healthy controls, CXCL signals were predominantly derived from CCL5+ macrophage populations. Interestingly, CXCL signaling in IBC was predominantly driven by the CXCL12 ligand and its receptor CXCR4 among all ligand-receptor pairs. In contrast, non-IBC tumors exhibited prominent interactions involving the CXCL10-CXCR3, CXCL13-CXCR3, and CXCL9-CXCR3 pairs (Fig. 5D). The CXCL13-CXCR3 interaction is particularly associated with extensive crosstalk between exhausted T cells and cytotoxic T cells (Fig. 5E). Our analysis revealed a significant reduction in communication between exhausted CD4+ and CD8+ T cells and other T cell populations, such as CD4 TCM, CD8 TRM, and CD8 TEM, through the CXCL13-CXCR3 pathway in patients with IBC compared to those with non-IBC (Fig. 5E). Given that CXCR3 is highly expressed on activated T cells, further analysis of the CXCL signaling pathway networks indicated that CXCR3 expression was significantly downregulated in the IBC group compared to non-IBCs (Supplementary Fig. S3A and Table S3). These findings suggest that the CXCL13-CXCR3 pathway plays a significant role in mediating immune cell communication and may be critical for maintaining an active immune microenvironment in breast cancer. The downregulation of this signaling axis may therefore contribute to the “cold” characteristics observed in IBC tumors. Moreover, we observed significantly diminished signaling of the CCL5-CCR1 axis in CD8+ T cells (including CD8 TEMRA, CD8 TCM, CD8 TRM, and CD8 TEM cells) within IBCs compared to non-IBCs (Supplementary Fig. S3). This specific ligand-receptor pair primarily facilitates communication between CD8+ T cells and macrophages, thereby elucidating the weaker interactions noted between these cellular subsets in IBCs. The observed absence or reduction of these signaling pathways in T cells within IBC tumors likely contributes to the attenuated intercellular interactions in the tumor microenvironment. Collectively, these data underscore the rewired cellular crosstalk network associated with IBC and highlight the central role of exhausted T cells in driving intercellular communication.

Tumor cells in IBCs exhibit high malignant potential coupled with a diminished immune response
To investigate the heterogeneity of epithelial cells, we conducted a comprehensive analysis by re-clustering 48,616 individual epithelial cells, resulting in the identification of 13 distinct clusters (Fig. 6A). These clusters represented various epithelial cell subtypes, including proliferative, mesenchymal, luminal progenitor, and several epithelial subsets (epithelial1-10). Marker genes associated with each cluster were identified, providing insights into their functional characteristics (Supplementary Fig. S4A). The analysis revealed that normal breast tissue samples predominantly consisted of luminal progenitor cells, while tumor tissues exhibited a diverse distribution of epithelial cell subsets (Fig. 6B). Notably, the IBC1 patient, who experienced a poor clinical outcome, demonstrated a higher abundance of cells from the epithelial5 cluster. Comparative analysis of DEGs within the epithelial5 cluster between the IBC1 and LBBC samples revealed a downregulation of interferon response pathways in the IBC1 patient (Fig. 6C). The epithelial1 cluster represented the second most abundant cell cluster present in the IBC1 tumor. Similarly, the epithelial1 cluster in the IBC1 patient exhibited downregulation of immune response-related pathways, further supporting the hypothesis of compromised immune responses in this individual. Moreover, the downregulation of estrogen response-related pathways in the epithelial1 cluster of the IBC1 patient was particularly noteworthy, as these pathways are closely linked to the prognosis of estrogen receptor-positive breast cancer.
Given that genomic alterations are a hallmark of tumor cells, we next performed a comprehensive copy number variation (CNV) analysis on the epithelial cells, resulting in the identification of nine subgroups: luminal progenitor, epithelial, and tumor1-7 clusters (Supplementary Fig. S4B). As expected, the tumor clusters exhibited a higher frequency of CNVs compared to luminal progenitor and other epithelial cells, with the tumor1 cluster displaying the highest frequency, indicative of its elevated malignant potential (Supplementary Fig. S4C). Furthermore, we analyzed the distribution of tumor cells across HC, IBC, and non-IBC tumors. Specifically, normal breast tissue samples were predominantly composed of luminal progenitor and epithelial cells, while both IBC and non-IBC samples exhibited a higher abundance of tumor cells, albeit with distinct proportions and disease subtype-specific characteristics (Fig. 6D). Remarkably, the tumor1 cluster was exclusively observed in the IBC1 patient, whereas the epithelial cells of the IBC2 patient predominantly comprised tumor4 cluster cells. The exclusive presence of the tumor1 cluster in the IBC1 patient, characterized by its highest frequency of CNVs, aligns with the patient’s poor clinical outcome (Fig. 6D). Marker genes for the tumor1 and tumor4 clusters were predominantly enriched in metabolic pathways such as oxidative phosphorylation, glycolysis, and xenobiotic metabolism, as well as signaling pathways associated with tumor growth, including mTORC1 signaling and MYC targets (Fig. 6E), implying the aggressive characteristics of these cells.
Given the above cumulative evidence suggesting an attenuated immune response in IBC patients, we then conducted a separate analysis to examine the expression patterns of immune-related genes among HC, IBC, and non-IBC samples. Consistently, we observed a profound decrease in the expression of immune-related genes, such as HLA-A and FLNA, in IBC patients (Fig. 6F). To validate these findings, we performed qRT-PCR analysis using normal breast epithelial cells (MCF-10A) and various breast cancer cell lines (MCF-7, MDA-MB-231, BT-474, MDA-MB-453, SUM149PT, and KPL-4). The results demonstrated a consistent gene expression pattern across both IBCs and non-IBCs, with IBC cells showing relatively lower expression of HLA-A and FLNA compared to most non-IBC cells, as illustrated in Fig. 6G. Previous studies have elucidated the advantages of HLA-negative tumor cells in cancer progression and metastasis [33]. The expression of HLA-A/B serves as an indicator of activated cytotoxic T lymphocytes and is associated with favorable clinical outcomes in immune-activated breast cancers [34]. To validate our findings at the protein level, we conducted fluorescence imaging analysis using tissue samples and the NanoZoomer S60 system. Fluorescence staining was performed for key proteins, including HLA-A, HLA-E, HLA-DMB, FLNA, FN1, and CXCL13. As illustrated in Fig. 7 and Supplementary Fig. S5, the staining results corroborated our earlier observations, demonstrating that immune-related proteins such as CXCL13 and HLA-A are expressed at lower levels in breast tissues derived from patients with IBC compared to non-IBC cases. This finding reinforces the finding of attenuated protein expression and provides visual confirmation of downregulation of these proteins in breast tissue from patients with IBC.

CXCL13 overexpression in tumor cells enhances tumor cell death in the presence of immune cells
We have previously demonstrated a strong correlation between the proportion of Tfh cells, M1 macrophages, and CD8+ T cells, suggesting a potential role for CXCL13 in regulating anti-tumor immunity. To further validate the immunomodulatory effects of CXCL13 in vitro, we conducted co-culture experiments using 4T1 tumor cells, either with or without CXCL13 overexpression, together with T cells or macrophages (RAW264.7). As shown in Fig. 8A and B, when 4T1 cells were co-cultured with T cells, CXCL13 overexpression significantly increased the rate of tumor cell apoptosis compared to 4T1 cells without overexpressing CXCL13. Moreover, ELISA analysis of the culture supernatants revealed elevated levels of TNF-α, Granzyme B, and IFN-γ (Fig. 8C), indicating enhanced T cell activation and cytotoxicity. Similarly, when 4T1 cells were co-cultured with RAW264.7 macrophages for 24 h, flow cytometry analysis showed a higher tumor cell death rate in the CXCL13-overexpressing group (Fig. 8D and E). In addition, flow cytometric profiling of macrophages revealed an increased proportion of M1-type macrophages (CD86+) and a reduced proportion of M2-type macrophages (CD206+) following CXCL13 overexpression (Fig. 8F-H). Consistent with this, qRT-PCR analysis confirmed upregulation of M1-associated markers, including iNOS, TNF-α, and IL-6 (Fig. 8I). Furthermore, microscopic observation demonstrated enhanced phagocytosis of tumor cells by macrophages in the CXCL13-overexpressing group (Fig. 8J and K). Collectively, these findings indicate that CXCL13 remodels the tumor microenvironment by promoting T cell-mediated cytotoxicity and macrophage M1 polarization, thereby enhancing anti-tumor immunity and promoting tumor cell death.

CXCL13 enhances the therapeutic efficacy of anti-PD-1 therapy in vivo
To further explore the prognostic significance of CXCL13 expression in breast cancer, we utilized the Kaplan-Meier Plotter online tool to assess its association with OS across different molecular subtypes (Supplementary Fig. S6A) and clinical grades (Supplementary Fig. S6B). Our analysis revealed that high CXCL13 expression was significantly associated with improved OS in several subtypes, including Luminal B (logrank P = 0.0023), HER2-positive (logrank P = 0.018), triple-negative (logrank P = 0.001), and basal breast cancer (logrank P = 1.9 × 10−8). In contrast, no significant correlation was observed in Luminal A patients (logrank P = 0.76), suggesting a subtype-specific prognostic role for CXCL13. When focusing specifically on ER-positive patients as a group, CXCL13 expression did not show a significant association with OS (logrank P = 0.24). This finding implies that the prognostic value of CXCL13 may be more relevant in ER-negative or more aggressive breast cancer subtypes. Further stratification by tumor grade revealed that high CXCL13 expression was significantly associated with better OS exclusively in Grade 3 tumors (logrank P = 0.003), while no significant associations were found in Grade 1 (logrank P = 0.88) or Grade 2 (logrank P = 0.061) tumors. Given that Grade 3 tumors are generally more aggressive, these findings suggest that CXCL13 may serve as a protective factor particularly in high-grade breast cancers, potentially reflecting an active anti-tumor immune microenvironment.
To further elucidate the potential mechanism by which CXCL13 improves outcomes in breast cancer, particularly through immune regulation, we performed in vivo experiments using a 4T1-luc orthotopic tumor model in female BALB/c mice. 4T1-luc cells were injected into the fourth pair of mammary fat pads to establish tumors. When tumor volumes reached approximately 50 mm3, mice (n = 5 per group) were evenly divided based on in vivo fluorescence imaging intensity and assigned to either an anti-PD-1 monotherapy group or a combination treatment group receiving anti-PD-1 and recombinant murine CXCL13. Mice were treated with anti-PD-1 (200 μg, every 3 days) alone or in combination with CXCL13 (5 μg, every 2 days) (Fig. 9A). Tumor volume and body weight were measured every 2 days, and fluorescence imaging was performed every 3 days to monitor tumor progression. As shown in Fig. 9B-C, tumors in the anti-PD-1 monotherapy group exhibited a more rapid increase in fluorescence intensity compared to the combination treatment group. Consistently, tumor volume assessments and tissue images (Fig. 9D-E) revealed that tumor growth was significantly inhibited in mice receiving the combination therapy relative to those treated with anti-PD-1 alone. Furthermore, no significant body weight loss was observed in the combination group (Fig. 9F), suggesting that the combined treatment was well tolerated. These results suggest that CXCL13 significantly enhances the therapeutic efficacy of anti-PD-1 treatment in vivo, providing further evidence for its potential as an immunomodulatory adjuvant in breast cancer therapy.

Sanguinarine and α-mangostin were identified as potential therapeutic agents for IBC through a screening of natural products
To identify potential therapeutic agents, we evaluated the inhibitory effects of 910 natural products across seven distinct cell lines: MCF-10A (normal breast epithelial cells), MCF-7 (luminal A), BT474 (luminal B), MDA-MB-231 (triple-negative), MDA-MB-453 (HER2-positive), KPL-4 (IBC cells), and SUM149PT (IBC cells), utilizing the CCK-8 assay (Supplementary Fig. S7). To visualize the efficacy of these natural products, we ranked them based on their inhibitory effects on breast cancer cell viability, with the top 50 compounds presented in a heatmap (Fig. 10A). From these compounds, we selected nine candidates that exhibited notable inhibitory effects on IBC cells and further investigated their impact on the expression of inflammatory genes, including HLA-A, FLNA, and CXCL13, in IBC cell lines (KPL-4 and SUM149PT) through qRT-PCR analysis (Fig. 10B and Supplementary Fig. S8). Notably, sanguinarine and α-mangostin were found to effectively upregulate the expression of several inflammatory genes in IBC cells, highlighting their potential as therapeutic candidates for IBC treatment by modulating immune-related factors. We further explored the effect of α-mangostin on tumor cell viability and its interaction with immune cells. In a co-culture model of 4T1 cells and macrophages, we observed that the presence of RAW264.7 cells significantly enhanced the cytotoxic effect of α-mangostin on tumor cells (Fig. 10C). This suggests that α-mangostin may exert its antitumor activity, at least in part, by modulating immune cell behavior within the tumor microenvironment. These findings provide preliminary evidence supporting α-mangostin’s potential to enhance antitumor immunity in IBC.

Discussion
IBC remains a significant clinical challenge due to its poor prognosis, diagnostic complexities, and limited treatment options. It usually progresses rapidly and does not always present as a palpable lump, which often leads to delays or misdiagnoses. Compared to other types of breast cancer, our understanding of the pathological and biological basis specific to IBC lags behind. The emerging body of research on IBC underscores its complex and aggressive nature, highlighting several critical factors that contribute to its pathophysiology, clinical behavior, and treatment response. Despite the lack of unique genomic drivers distinguishing IBC from non-IBC, numerous studies point to the TME, immune interactions, and specific molecular pathways as pivotal in dictating the disease's aggressiveness and response to therapies. Several studies investigating the genomic landscape of IBC have found that its tumor mutational burden (TMB), mutational spectra, and copy number alterations (CNAs) largely mirror those of non-IBC tumors. This challenges the assumption that IBC’s aggressive clinical behavior is driven by distinct genetic mutations. For instance, a whole-exome sequencing (WES) analysis comparing untreated IBC tumors with stage-matched non-IBC tumors revealed no significant differences in genomic features [35]. Non-genomic factors, including the immune microenvironment, epigenetic modifications, and tumor-stromal interactions, may be more influential in IBC’s progression and aggressiveness. This highlights the need for multi-omics approaches, such as scRNA-seq and spatial transcriptomics, to further unravel the contributions of the TME and other microenvironmental factors to IBC's unique biology. A previous study provides a single-cell analysis of IBC, revealing a unique subpopulation of luminal progenitor (LP) cells that highly express pleiotrophin (PTN) as a critical driver of IBC's aggressive phenotype [36]. Unlike non-IBC tumors, IBC tumors were found to exhibit significant enrichment of PTN + LP cells, which secrete PTN to promote angiogenesis by activating neuropilin-1 (NRP1) on endothelial tip cells, underscoring the role of non-malignant cells in shaping IBC's aggressive behavior. In this study, employing scRNA-seq to analyze molecular subtype-matched IBC and non-IBC samples at the single cell level, we determined that a significant reduction in exhausted CD4+ T cells (CXCL13+ CD4+ T cells) in IBC tumors compared to their corresponding molecular subtype matched non-IBCs. This cluster of cells was characterized by high expression of CXCL13, a gene associated with an anti-tumor immune microenvironment [21,37]. The reduction of CXCL13 in IBC tumors may contribute to the establishment of a “cold” immune microenvironment in IBC and is strongly associated with the poor prognosis of IBC.
In this study, differential expression analysis demonstrated that CXCL13 was the most significantly downregulated gene in IBC-derived T cells. CXCL13 plays a crucial role in cancer biology, particularly in regulating the tumor microenvironment and recruiting immune cells [38]. It is recognized primarily for its function in B cell homing and lymphoid organogenesis, binding to its receptor CXCR5, which is expressed on B cells and certain T cell subpopulations. In our study, we observed a significant correlation between CXCL13 expression and immune cell infiltration in breast cancer, particularly with cytotoxic lymphocytes and M1-type macrophages. This finding links CXCL13 expression to an enhanced anti-tumor immune microenvironment in breast cancer. In IBCs, infiltrated stromal immune cells are associated with improved outcomes [39]. Our analyses consistently show that CXCL13 is significantly down-regulated in IBC and its expression is positively associated with the prognosis of breast cancer patients. Most importantly, overexpression of CXCL13 in 4T1 cells, followed by co-culture with immune cells, significantly promoted tumor cell death, likely through modulation of immune cell behavior. Further analyses indicated that CXCL13 overexpression enhanced T cell activation, altered macrophage polarization, and influenced the secretion of key immune mediators, collectively contributing to an improved anti-tumor immune response. These findings underscore the important role of CXCL13 in regulating the tumor microenvironment in IBC, which may consequently influence the outcomes for IBC patients.
CXCL chemokines play a key role in the formation and maintenance of tumor microenvironment and are important mediators for recruitment of various immune cells to tumor sites. Among these chemokines, CXCL9, CXCL10, and CXCL11 are mainly associated with Th1-type immune responses, promoting the attraction of cytotoxic T lymphocytes and NK cells, thereby promoting the construction of a “hot” tumor microenvironment and anti-tumor immunity [[40], [41], [42]]. Correlation analysis conducted using data obtained from the UCSC database via the Sangerbox online tool revealed a significant association between CXCL13 and the aforementioned chemokines (CXCL9, CXCL10, and CXCL11). This suggests that there may be a feedback loop in which CXCL13-mediated recruitment of immune cells may stimulate increased production of CXCL9, CXCL10, and CXCL11, leading to further recruitment of immune cells into the tumor microenvironment. Moreover, CXCL13 has been shown to be involved in the activation of macrophages in cancer [43], which are critical sources of CXCL9 and CXCL10 within the tumor microenvironment. Macrophages can respond to CXCL13 by upregulating the expression of these chemokines, thereby enhancing the recruitment of T cells and other immune cells. Interactions between CXCL13 and other CXC chemokines significantly increased the infiltration of effector T cells into the tumor microenvironment. Since these immune cells rely on chemokines such as CXCL9, CXCL10, and CXCL11 to migrate to the tumor site, this loop is critical for establishing an effective anti-tumor immune response. Besides, previous report has demonstrated that IBCs promote the development of M2-type macrophages through a complex network of chemokines [44]. This was consistent with the observation that CXCL13 negatively correlated with M2-type macrophage frequency in breast cancer and significantly down-regulated in IBC tumors. However, it is important to note that although CXCL13-producing cells are often reported to be positively correlated with immunotherapy efficacy in several cancers [21,37,38], intratumoral CXCL13+ CD8+ T cell infiltration was demonstrated to be associated with poor clinical prognosis in clear cell renal cell carcinoma [45]. Therefore, CXCL13 may have cancer-type specific effects, which should be paid attention to in future studies. Additionally, analysis of ligand-receptor interactions demonstrated the downregulated CXCL13-CXCR3 axis in cell–cell communication, which is involved in communication between T cells. The CXCR3 axis facilitates recruitment of immune cells into the tumor microenvironment [46,47]. Thus, the significant downregulation of CXCL13 and CXCR3 expression across multiple T cell subpopulations may play a critical role in establishing a “cold” microenvironment in IBC tumors. This alteration could contribute to the aggressive phenotype and poor prognosis commonly associated with IBC patients. Therefore, further investigations are warranted to elucidate the precise role of CXCL13 and the CXCL13-CXCR3 axis and to assess its potential as a therapeutic target in patients with IBC.
Moreover, patients with IBC exhibited a significant downregulation of immune-related genes, including HLA-A, FLNA, and CXCL13. HLA class molecules are an important part of the immune system, and changes in their expression can adversely affect the immune surveillance of tumors and may affect cancer progression and metastasis [48,49]. Besides, a notable downregulation of the interferon signaling pathway was also observed in tumor cells derived from IBC patients compared to those from non-IBC patients. The specific role of the interferon signaling pathway, particularly the IFN-γ signaling axis, in tumorigenesis has engendered controversy [50]. Nonetheless, it is widely accepted that IFN-γ is crucial for promoting antitumor immune responses through direct and indirect effects on both tumor and immune cells. Indeed, the attenuation of intra-tumoral IFN signaling has been linked to immunosuppression in hypoxic tumors [51] and primary resistance to anti-CTLA-4 therapy in melanoma patients [52]. Conversely, the activation of IFN signaling has shown promise in enhancing the efficacy of immune checkpoint blockade in preclinical breast cancer models [53]. Thus, this warrants further investigation into the implications of the interferon signaling pathway in the immunosuppressive tumor microenvironment associated with IBC.
Recent advances in understanding the TME of IBC have highlighted its complex and dynamic immune landscape. Immune cell infiltration and immunosuppressive features in IBC have examined from different angles in previous studies, revealing both shared observations and important divergences. Bertucci et al. provided a comprehensive immune profiling of pre-treatment primary IBC, revealing a heterogeneous but overall immunologically active TME [54]. Their study demonstrated enrichment of memory-B cells, M1 macrophages, and γδ T cells, as well as elevated expression of immune checkpoint molecules such as PD-L1, TIM-3, and LAG3. This suggested a pro-inflammatory and potentially immune-responsive TME, supports the potential of immune checkpoint inhibitors (ICIs) in early-stage IBC. In contrast, De Schepper et al. [55] found that stromal tumor-infiltrating lymphocytes (sTILs) were generally low across IBC cases, but still retained predictive value for treatment response. Meanwhile, Fernandez et al. [12], focusing on metastatic IBC patients, observed profound systemic immunosuppression, including lymphopenia and depletion of peripheral T, B, and NK cells, reflecting immune suppression at advanced stages. However, PD-1/PD-L1 expression in tumor tissues still correlated with peripheral immune activation in some cases, suggesting residual immune responsiveness. On the other hand, methodological differences likely contribute to these discrepancies. Bertucci et al. utilized transcriptomic analyses to infer immune cell composition within tumor tissues, whereas Schepper et al. evaluated sTILs through histopathological assessment. In contrast, Fernandez et al. focused on immune profiling of peripheral blood using flow cytometry. Regardless of methodology, patient cohort, or disease stage, all these studies highlight the critical interplay between immune cell infiltration and immunosuppressive mechanisms in IBC. No matter what stage the disease is in, these findings together suggest that immune dynamics in IBC are complex and evolve over time. While the IBC TME can be immunologically active in earlier stages, disease progression may lead to systemic immune exhaustion or suppression. This may explain why, despite preclinical evidence suggesting that IBC tumors are immunologically vulnerable, clinical trials have shown limited efficacy of immune checkpoint inhibitors combined with chemotherapy in metastatic triple-negative IBC (mTN-IBC), with lower-than-expected progression-free survival (PFS) rates even among PD-L1-positive patients [56]. Thus, understanding the temporal and spatial evolution of the immune landscape is essential for designing stage-specific immunotherapeutic strategies and for identifying which patients may benefit most from immune modulation. Besides, our findings also indicate that the prognostic significance of CXCL13 is especially pronounced in high-grade breast tumors. Consistently, in vivo studies revealed that CXCL13 markedly enhanced the efficacy of anti-PD-1 therapy. These results offer important insights for developing novel strategies to treat highly aggressive forms of IBC.
Furthermore, in this study, we preliminarily explored the potential of natural products in treating IBCs. Natural products exhibit specific advantages in preventing tumor occurrence and reducing the risk of tumor recurrence and metastasis [57,58]. Through screening the inhibitory effects of 910 natural products on various types of breast cancer cells, sanguinarine and α-mangostin were identified as potential therapeutic compounds for IBC treatment. These compounds showed the ability to upregulate the expression of inflammatory genes in IBC cells, highlighting their potential as novel drugs for combating this aggressive form of breast cancer. Through a co-culture model of tumor cells and macrophages, we found that macrophages markedly enhanced the cytotoxic effect of α-mangostin on tumor cells. These findings suggest that α-mangostin’s antitumor activity may, at least partially, depend on its ability to modulate immune cell dynamics within the tumor microenvironment.
In summary, our scRNA-seq study provided a comprehensive transcriptional characterization of IBCs, revealing distinct cellular subsets and alterations in immune-related gene expression. The findings support the presence of a “cold” tumor microenvironment in IBCs and shed light on potential therapeutic options, such as sanguinarine and α-mangostin. However, given the relatively small number of IBC samples analyzed in this study, further investigations involving larger patient cohorts will be essential to comprehensively capture the heterogeneity of IBC and to validate the robustness of our findings.

Data availability

The The data that support the findings of this study have been deposited into CNGB Sequence Archive (CNSA) [59] of China National GeneBank DataBase (CNGBdb) [60] with accession number CNP0004647.

Compliance with ethics requirements

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2008 (5). Informed consent was obtained from all patients for being included in the study.

CRediT authorship contribution statement
Xueni Sun: Methodology, Investigation, Visualization, Writing – original draft. Jing Xia: Methodology, Investigation, Visualization. Haiyang Jiang: Investigation, Visualization. Ting Duan: Methodology, Investigation. Chunli Zhang: Methodology. Qinyi Li: Investigation, Visualization. Zuyi Yang: Methodology, Investigation. Ruonan Zhang: Investigation, Visualization. Xia Ding: Conceptualization, Writing – review & editing. Xidong Gu: Conceptualization, Writing – review & editing. Xiaohong Xie: Conceptualization, Writing – review & editing. Tian Xie: Conceptualization, Writing – review & editing. Xinbing Sui: Conceptualization, Supervision, Writing – review & editing.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.