Integrated transcriptome study reveals a stress response state CD4 +T cells related to immune tolerance in breast cancer Authors: Jinyuan Gu, Qi Qi, Xinrui Mao, Rong Cong, Yue Sun, Xinyu Tang, Xiaoxiang Guan, Ji Wang, Muxin Yu, Shui Wang, Hong Pan, Kai Zhang, Wen Qiu, Wei Li, Cong Wang, Wenbin Zhou Published: 2025 ABSTRACT BackgroundLymph node status is of great clinical importance in tumor staging and prognosis where antitumoral immunity is generated and preserved. However, the immune landscape of lymph nodes before and after metastasis in breast cancer is not clear.MethodsWe analyze a total of 67 samples from breast cancers and lymph nodes to profile immune cells at single-cell resolution. Additionally, 13 patients with breast cancer were enrolled to confirm our findings using flow cytometry and immunofluorescence with primary breast tumors and axillary lymph nodes.ResultsThe distribution of immune cells is similar in uninvolved lymph nodes, but heterogeneity of immune cells is observed in metastatic lymph nodes (metLNs) across different subtypes of breast cancer. In metLNs, a cluster of CD4 +T cells (CD4 +Tstr cells) is predominantly found in luminal-like breast cancer with a state of anergy, probably due to immunosuppression caused by myeloid and tumor cells. By comparing scRNA and single-cell T cell receptor seq data, these un-expanded CD4+Tstr cells seldom show shared T cell receptor sequence between primary tumors and paired metLNs, therefore do not migrate from paired metLNs. Moreover, CD4+Tstr cells are associated with poor survival and predicted tolerance to immunotherapy in triple-negative breast cancer.ConclusionsIn summary, we identify a specific cluster of CD4 +T cells related to immune tolerance that serves as a new predictor for immunotherapy in breast cancer.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13058-025-02205-4. Background Regional lymph node status is of great clinical importance in tumor staging and prognosis in many tumor types. Patients with or without lymph node metastasis vary markedly in recurrence rate and overall survival. Lymph node state is even one of the most potent prognostic factors in breast cancer [1, 2]. More than that, lymph nodes are also important immune organs where antitumoral immunity is generated and preserved. After antigens are transported to regional lymph nodes by lymphatic fluid or antigen presenting cells, lymphocytes in the lymph nodes are primed and this immune activation can spread into systemic immune response through lymphocyte recirculation [3–5]. Recent studies have determined uninvolved lymph nodes (uiLNs) as the source and reservoir of tumor-responsive T cells during immunotherapy, and treatments that target regional lymph nodes could significantly enhance the efficacy of immunotherapy [4, 6]. In metastatic lymph nodes (metLNs), however, there has been preliminary evidence that both the antitumoral immunity and immunotherapy responsiveness might be disrupted [7]. Moreover, patients with lymph nodes metastasis have shown a relative better survival than patients with other metastases in breast cancer [8]. It remains to be uncovered what exact changes happen inside the lymph nodes along with the occurrence of metastasis. Breast cancer is a highly heterogeneous disease. Based on the expression of estrogen receptor (ER), progesterone receptor (PR), and human epithelial growth factor receptor 2 (HER2), breast cancer can be clinically stratified into three molecular subtypes: hormone receptor-positive/HER2-negative (luminal-like), HER2 gene amplified/receptor overexpressed (HER2+), and triple-negative breast cancer (TNBC) [9–11]. Different subtypes of breast cancer show great differences in prognosis and immune microenvironment [12, 13]. In the era of immunotherapy, different subtypes are further found to show divergent responses to immunotherapies [14]. In general, HER2 + and TNBC show better response to immunotherapy whether administered alone or in combination, while luminal-like breast cancer is considered “immune inert” [15–17]. Since lymph node is the primary site where antitumoral immune response is generated, it is highly possible that the divergent immune responses and immunotherapy responsiveness of different subtypes come from their draining lymph nodes. In this study, the features and changes of lymph nodes before and after metastasis in different subtypes of breast cancer were explored comprehensively for the first time. We found that uiLNs of different subtypes show similar immune features, while metLNs show heterogeneity of T cells across different subtypes, with an anergic stress response state CD4+ T (CD4 +Tstr) cell subgroup dominating in luminal-like breast cancer. This subgroup received antigen presentation but lacked co-stimulatory molecules, probably due to immunosuppression caused by myeloid and tumor cells. These un-expanded CD4 + Tstr cells did not migrate from paired metLNs, associated with local tumor microenvironment, related to poor prognosis and immunotherapy responsiveness in TNBC. Overall, our study might give new insights into the immune heterogeneity of breast cancer, and provide a predictor for immunotherapy. Methods Data Availability We used a total of 67 samples of scRNA-seq data. The gene expression data from our and additional sources included in this study were available from the NCBI GEO database (https://www.ncbi.nlm.nih.gov/geo/) under accession numbers: GSE180286, GSE158399, GSE169246, GSE161529, GSE190870. A total of 25 lymph nodes were available from 24 individuals. The T cell receptor (TCR) sequencing data were obtained from GSE190870, including 16 samples from 8 individuals with paired lymph nodes and primary tumors. The scRNA-seq data (GSE246613) with 34 samples of 33 individuals was utilized to validate our findings (Supplementary Figure S1). All patient information was summarized in Supplementary Table 1. Sample collection This study was conducted with the approval of the institutional ethics committee of The First Affiliated Hospital with Nanjing Medical University (approval no.2024-SR-701), and were in accordance with the Declaration of Helsinki. Patients provided written informed consent for the collection of samples used for research purposes, with no impact on their clinical treatment. This study included samples from 13 breast cancer patients, and all patients were treatment-naïve. Primary breast tumors were obtained from ten patients for flow cytometry. The remaining three patients had already metastasized to axillary lymph nodes confirmed by pathology. According to the immunohistochemical results of ER, PR and HER2, one patient was luminal-like, one patient was HER2+, and one patient was diagnosed with TNBC. Axillary lymph nodes were collected for immunofluorescence (Supplementary Figure S1). All patient information was summarized in Supplementary Table 1. scRNA-seq data quality control, data filtering and integration Seurat (v4.3.0) [18] was used for data quality control, normalization, dimensionality reduction, clustering and annotation. In total, 209,177 cells from different studies were obtained. Furthermore, quality control was applied to the cells based on several criteria. Briefly, cells with < 200 detected genes as well as those with > 20% mitochondrial content were removed. Cells having over 6000 detected genes were eliminated to further exclude the possible doublets. After filtering, 191,337 high-quality cells were preserved for subsequent analyses. The filtered gene expression matrix was normalized by the NormalizeData function. Then FindVariableFeatures function was applied to select the top 2000 highly variable genes, which were then used to construct principal components (PCs) with RunPCA function. RunHarmony function in Harmony package (v0.1.1) [19] was used to integrate samples in order to remove batch effects. Clustering with 20 PCs and resolution 0.05 were performed by graph-based clustering, visualized using Uniform Manifold Approximation and Projection (UMAP) with RunUMAP function. For subclustering analysis, a similar procedure was applied. For T&NK cells the top 17 PCs were selected with a resolution set to 0.4. For myeloid cells the top 20 PCs were selected with a resolution set to 0.2. And for B cells the top 15 PCs were selected with a resolution set to 0.1. Cell types were identified based on canonical markers from published references. Differential gene expression analysis and enrichment analysis Differentially expressed genes (DEGs) of the resulting clusters were identified using the FindAllMarkers function with the default parameters, using the default Wilcoxon Rank Sum test with Bonferroni correction. DEGs were filtered by log2FC (fold change) > 0.25 and p value < 0.05 and used for enrichment analysis. The Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses and gene set enrichment analysis (GSEA) [20] were performed by clusterProfiler (v4.8.0) [21]. Functional characterization of cells Gene signature scores were calculated for gene sets using Seurat’s AddModuleScore function. Signature gene sets of T&NK cells and macrophages were derived from previously published works [22–26]. The gene sets of antigen processing and presentation score also came from the previous study [27]. (Supplementary Table 2) Defining cell cluster signature scores In the validation of the findings, CD4 + Tstr signature (Supplementary Table 3) was defined as genes that were significantly upregulated in CD4 + Tstr cluster compared to the other T&NK cell clusters. Using the FindAllMarkers function, and after filtering for adjusted p-value < 0.05, absolute value of average log2FC > 1.4, the CD4 + Tstr signature included 12 genes: NR4A1, HSPA1B, FOSB, EGR1, HSPA1A, SLC2A3, HSPA6, BTG2, HSPH1, CD69, FOS, HSP90AA1. Trajectory analysis Monocle 3 (v.1.3.1) [28] was applied to reconstruct the cellular differentiation trajectory of CD4 + and CD8 + T cell subclusters. Clustering cells and calculating partitions using the cluster_cells function. And the trajectory analysis was performed using the learn_graph function. T01_CD4-SATB1, T05_CD4-CCR7 and T11_CD8-CCR7 were selected as the root for the trajectory for CD4 + and CD8+T cells. We performed the cluster_cells function, setting the resolution to 0.00001 and 0.000001 for CD4 + and CD8 + T cell, respectively. The cell trajectories were visualized by the plot_cells function. The graph_test function identified genes differentially expressed across single-cell trajectories and visualized the genes using the plot_genes_in_pseudotime function. Cell–Cell communication analysis The CellChat (v1.6.1) [29] was used to infer and analyze intercellular communication networks between clusters. The normalized expression matrix was imported to create a cellchat object with the createCellChat function. Followed by preprocessing data with dentifyOverExpressedGenes, identifyOverExpressedInteraction, and projectData functions, the computeCommunProb, filterCommunication (the minimum number of cells required for cell-cell communication analysis in each cell group was set to 10), and computeCommunProbPathway functions were used to compute the communication probability on signaling pathway level by summarizing all related ligands/receptors. Finally, using the aggregateNet function to calculate the aggregated cell–cell communication network. T-cell receptor analysis TCR sequencing data were obtained from GSE190870 [30]. Included TCR sequencing data for T&NK cluster from 8 paired primary tumors and metLNs. Annotation information of T&NK subclusters, TCR clonal expansion numbers, and whether matched TCR (TCR located both in the primary tumor and metLNs) were also obtained for analysis. Multiplex immunofluorescent staining Lymph node samples from patients were placed in 4% Paraformaldehyde for 24 h, and then dehydrated and embedded in paraffin. Paraffin blocks were cut into 4 μm sections and adhered to a glass slide heated at 62℃ for 1 h, deparaffinized in xylene, and then rehydrated in 100%, 95%, 70% alcohol in sequence. Antigen was retrieved by citric acid buffer (pH 6.0) in the 95℃ water bath for 20 min. Endogenous peroxidase was inactivated by incubation in 3% H2O2 for 10 min. After a pre-incubation with blocking buffer (5% BSA) at room temperature for 20 min, the sections were incubated with primary antibodies including CD4 (Abcam, Cat# ab133616, 1:400) and HSP70 (Proteintech, Cat# CL488-10995, 1:200), followed by incubation with horseradish peroxidase-conjugated secondary antibodies (Abcam, Cat# ab6721, 1:1000) and tyramide signal amplification. Finally, slides were enclosed using Fluoroshield Mounting Medium with DAPI, scanned using the Pannoramic SCAN (3D Histech), and analyzed with SlideViewer software. Flow cytometry Fresh surgical tissues were stored in tissue storage solution at 4℃ and used for analysis within 4 h. Tissues were digested with 20 mg/ml Dispase II, 20 mg/ml hyaluronidase and 20 mg/ml collagenase for 1 h at 37℃ with rotation to promote dissociation, and then filtered through a 70 μm cell strainer to obtain single-cell suspensions. Cells were treated with PMA (MultiScience Biotech, Cat# CS1001) and BFA (MultiScience Biotech, Cat# CS1002) for 6 h to augment cytokine secretion and block extracellular transportation. Zombie Aqua™ Fixable Viability Kit (BioLegend, Cat# 423101) was used to detect live cells. For surface staining of cells, cells were stained with fluorescence-conjugated antibodies in flow cytometry staining buffer (Biolegend, USA) for 30 min at 4 °C. After extracellular staining, cells were fixed and permeabilized (Biolegend, USA), followed by intracellular staining. The following fluorescence-conjugated antibodies were use: FITC anti-human CD45 Antibody (Biolegend, Cat# 304005), PerCP/Cyanine5.5 anti-human CD4 Antibody (Biolegend, Cat# 300529), PE/Cyanine7 anti-human CD25 Antibody (Biolegend, Cat# 356107), PE Anti-Hsp70 Antibody (Abcam, Cat# ab208878), Brilliant Violet 421™ anti-human IFN-γ Antibody (Biolegend, Cat# 502531), APC anti-human CD185 (CXCR5) Antibody (Biolegend, Cat# 356907). BD FACSCanto cytometer (BD Biosciences) was used for data acquisition, and data were analyzed with FlowJo 10.8.1 software. Survival analyses with gene signatures Public gene expression and survival data from The Cancer Genome Atlas Program (TCGA) dataset were accessed to validate the prognostic value of the CD4 + Tstr signature (https://www.cancer.gov/ccg/research/genome-sequencing/tcga), including 701 cases of breast cancer patients. The T10_CD8-CXCL13 signature was defined as genes significantly upregulated compared to all the other clusters (Supplementary Table 4). Multivariable Cox analyses were adjusted by age at diagnosis and cancer stage. All survival analyses and hazard ratio (HR) calculations were performed using the survival (v3.3-1) R package. And multivariate Cox analysis results were visualized as forest plots via the forestplot (v3.1.1) R package. Statistical analyses Continuous variables were described as mean and standard deviation (SD) for normal distribution or with medians and interquartile ranges for abnormal distribution. Paired or unpaired Student’s t-test and Wilcoxon rank-sum test were utilized in this study. Statistical significance was accepted for P < 0.05. All statistical analyses were performed in R (v4.2.0) (https://www.r-project.org/). Results Single-cell expression profile and cell types of lymph nodes in breast cancer To study the single-cell transcriptomic profile of lymph nodes in breast cancer, we compiled the scRNA-seq data of 25 lymph nodes from our 2 datasets and 3 additional published datasets (Fig. 1a) [30–34]. Our 2 datasets included 5 female patients with 6 uiLNs. 3 published datasets included 18 female patients and 1 male patient with a total of 19 metLNs. None of the patients had received any treatment. For molecular subtypes, uiLNs were from 1 patient with TNBC (including 2 uiLNs), 2 patients with HER2 + breast cancer, and 2 patients with luminal-like breast cancer. MetLNs were from 4 patients with TNBC, 3 patients with HER2 + breast cancer and 12 patients with luminal-like breast cancer (Fig. 1b). After quality filtering (Supplementary Figure S2a), the single-cell transcriptomic data from 191,337 high-quality cells were analyzed, including 51,817 (27.08%) cells from uiLNs and 139,520 (72.92%) cells from metLNs. We integrated the diverse data using the batch effect correction algorithm “Harmony” [19] and cells were well mixed in the integrated data. Fig. 1Single-cell expression profile and cell types of lymph nodes in breast cancer. (a) Schematic overview of the analytical workflow of this study. (b) Baseline clinical characteristics of enrolled patients. Subtypes included luminal-like (Lum), HER2 + and triple-negative breast cancer (TNBC). (c) Uniform Manifold Approximation and Projection (UMAP) plot, showing cell types of uninvolved lymph nodes (uiLNs) and metastatic lymph nodes (metLNs). (d) Histogram showing the proportions of cells in uiLNs and metLNs Single-cell expression profile and cell types of lymph nodes in breast cancer. (a) Schematic overview of the analytical workflow of this study. (b) Baseline clinical characteristics of enrolled patients. Subtypes included luminal-like (Lum), HER2 + and triple-negative breast cancer (TNBC). (c) Uniform Manifold Approximation and Projection (UMAP) plot, showing cell types of uninvolved lymph nodes (uiLNs) and metastatic lymph nodes (metLNs). (d) Histogram showing the proportions of cells in uiLNs and metLNs All single cells underwent unsupervised grouping based on uniform flow UMAP analysis (Supplementary Figure S2b, c). Six major cell groups were identified (Fig. 1c, Supplementary Figure S2d): T&NK cells (CD3D, CD3E, NKG7), B cells (CD19, CD79A, MS4A1), epithelial cells (EPCAM, KRT19), myeloid cells (LYZ, CD68), endothelial cells (VWF, PECAM1), and fibroblasts (COL1A1, COL1A2, DCN). The cells in uiLNs were almost exclusively immune cells, while the cell compartments in metLNs included immune cells, epithelial cells, endothelial cells and fibroblasts (Fig. 1d). Heterogeneity of T cells in MetLNs across different subtypes of breast cancer After re-clustering 99,268 T&NK cells, a total of 15 cell clusters were identified based on their signature genes [22, 35], including 8 clusters in the CD4+ T compartment, 6 clusters in the CD8+ T compartment and 1 cluster in the NK compartment (Fig. 2a, Supplementary Figure S3a). Fig. 2Heterogeneity of T cells in metLNs across different subtypes of breast cancer. (a) UMAP visualization of 15 T&NK cell clusters. (b) UAMP showing T&NK cells in metLNs across three subtypes of breast cancer. Pie charts displaying cell composition. (c) Histogram showing the proportions of T&NK cells in metLNs across three subtypes of breast cancer. (d) Heat map illustrating expression of gene signatures across CD4 + T cell clusters in metLNs. Heat map was generated based on scaled gene signature scores. T07 (18 cells) and T08 (9 cells) were excluded owing to small quantities. (e) Heat map illustrating expression of gene signatures across CD8 + T cell clusters in metLNs. Heat map was generated based on scaled gene signature scores. T14 (4 cells) was excluded owing to small quantities. (f) Monocle 3 trajectory analysis of CD4 + T cell differentiation in metLNs. Cells are color coded for their corresponding pseudotime. (g) Monocle 3 trajectory analysis of CD8 + T cell differentiation in metLNs. Cells are color coded for their corresponding pseudotime Heterogeneity of T cells in metLNs across different subtypes of breast cancer. (a) UMAP visualization of 15 T&NK cell clusters. (b) UAMP showing T&NK cells in metLNs across three subtypes of breast cancer. Pie charts displaying cell composition. (c) Histogram showing the proportions of T&NK cells in metLNs across three subtypes of breast cancer. (d) Heat map illustrating expression of gene signatures across CD4 + T cell clusters in metLNs. Heat map was generated based on scaled gene signature scores. T07 (18 cells) and T08 (9 cells) were excluded owing to small quantities. (e) Heat map illustrating expression of gene signatures across CD8 + T cell clusters in metLNs. Heat map was generated based on scaled gene signature scores. T14 (4 cells) was excluded owing to small quantities. (f) Monocle 3 trajectory analysis of CD4 + T cell differentiation in metLNs. Cells are color coded for their corresponding pseudotime. (g) Monocle 3 trajectory analysis of CD8 + T cell differentiation in metLNs. Cells are color coded for their corresponding pseudotime In uiLNs across different subtypes of breast cancer, the proportions of CD4 + T cells and CD8 + T cells were similar, consisting of numerous CD4 + T cells together with a small proportion of CD8 + T cells and NK cells (Supplementary Figure S3c). In uiLNs, T cells were mainly composed of T07_CD4-IL7R, T08_CD4-RPL10P9 and T14_CD8-CCL5 clusters (Supplementary Figure S3c). T07_CD4-IL7R cluster expressed IL7R (Supplementary Figure S3a), suggesting a memory phenotype. T08_CD4-RPL10P9 cluster was found mainly in uiLNs of HER2 + breast cancer and expressed STAT1, suggesting that this cluster might be quiescent cells [36]. T14_CD8-CCL5 cluster expressed CCL5, recognized as a cluster of effector memory T cells [37]. In metLNs, 6 CD4 + T cell clusters (T01 to T06) and 5 CD8 + T cell clusters (T09 to T13) were identified (Fig. 2b, c). T01_CD4-SATB1 and T05_CD4-CCR7 displayed a naive-like phenotype (Fig. 2d, Supplementary Figure S3b), with T05_CD4-CCR7 expressing the highest naive gene signature and low activation markers. T02_CD4-HSPA1A cluster was characterized by its unique expression of stress-related heat shock genes (such as HSPA1A and HSPA1B) (Supplementary Figure S3a) and stress response genes [38, 39], high expressions of HIF1A gene and “response to hypoxia” gene signatures, and the highest expression of activation/effector function genes (Fig. 2d, Supplementary Figure S3b). T03_CD4-CD74 cluster highly expressed HLA-DRA and HLA-DRB1, markers of T cell activation [40]. T04_CD4-ANXA1 cluster expressed “memory” marker genes, such as ANXA1 and IL7R. T06_CD4-FOXP3 cluster exhibited a classical Treg gene signature (such as FOXP3 and IL2RA), expressing exhaustion genes. Moreover, high expression of IFN response genes which might be related with T cell clonotype expansion [41], as well as oxidative phosphorylation gene signature and cytokine/cytokine receptors (such as IL2RA, IL2RG, IL32) were also observed (Fig. 2d). In addition, T06_CD4-FOXP3 clusters expressed ICOS, indicating that they might play a role in the antitumor effects mediated by anti-CTLA-4 therapy [42, 43]. Interestingly, we found that TCR signaling was highly expressed in T02_CD4-HSPA1A and T06_CD4-FOXP3, suggesting that they bound to MHC class II molecules [44], but only T06_CD4-FOXP3 cluster highly expressed costimulatory molecules (Fig. 2d), suggesting that possibly only T06_CD4-FOXP3 was activated while T02_CD4-HSPA1A might be in a state of anergy [45]. KEGG pathway analysis also found T06_CD4-FOXP3 was enriched with T cell receptor signaling pathway genes (Supplementary Figure S3e), and T02_CD4-HSPA1A was enriched in apoptotic genes, which might be related to the anergy caused by T02_CD4-HSPA1A costimulation-deficient [45]. T11_CD8-CCR7 clusters expressed high levels of naive markers (such as CCR7 and SELL) and highly expressed naive gene signature (Fig. 2e, Supplementary Figure S3b). The T10_CD8-CXCL13 cluster was characterized by the highest expression of exhaustion-related markers (such as LAG3, TIGIT, PDCD1 and CTLA4) and effector molecules (such as CST7, GZMK, GZMA, NKG7). In addition, T10_CD8-CXCL13 cells also highly expressed CXCL13 and anti-tumor signatures (Fig. 2e, Supplementary Figure S3b), indicating that they were tumor-reactive T cells [46] and likely to be related to response to immunotherapy [33]. T09_CD8-LMNA and T13_CD8-CD82 clusters showed high expression of early activation markers CD44 and CD69, and effector molecules (such as CST7, GZMK, GZMA, NKG7) although weaker compared to T10_CD8-CXCL13 (Supplementary Figure S3b). Consistently, high IFN response and TCR signaling were found in these three clusters (T09, T10 and T13), and enrichment of T cell receptor signaling pathway and cell adhesion molecules were also found in KEGG pathway analysis (Supplementary Figure S3f). T12_CD8-TRAC cluster highly expressed costimulatory molecules but without TCR signaling, so they might not function as anti-tumor cells. However, the distribution pattern of T cell clusters was heterogeneous across the metLNs of different molecular subtypes of breast cancer (Fig. 2b, c). MetLNs of luminal-like breast cancer had a higher proportion of CD4 + T cells than HER2 + and TNBC. T01_CD4-SATB1, T02_CD4-HSPA1A, T03_CD4-CD74 and T12_CD8-TRAC were predominantly found in metLNs of luminal-like breast cancers. Specifically, T02_CD4-HSPA1A highly expressed HIF1A, and GSEA also showed that CD4 + T cells in metLNs of luminal-like breast cancer upregulated the hypoxia pathway (Supplementary Figure S3d). Thus, hypoxia might partly explain the high expression of heat shock proteins in T02_CD4-HSPA1A. T04_CD4-ANXA1, T09_CD8-LMNA and T13_CD8-CD82 clusters were mainly found in metLNs of HER2 + breast cancer. For TNBC, T05_CD4-CCR7, T10_CD8-CXCL13 and T11_CD8-CCR7 clusters were mainly found in metLNs of TNBC, suggesting TNBC was suitable for immunotherapy. To understand the differentiation trajectories of T cells, pseudotime analyses were performed in metLNs [28]. For CD4 + T cells, we selected T01_CD4-SATB1 and T05_CD4-CCR7, which expressed high naive signature and low activation signature (Fig. 2d), as the root of the trajectory. In metLNs of luminal-like breast cancer, CD4 + T cells ended with T02_CD4-HSPA1A, T03_CD4-CD74, or T04_CD4-ANXA1, whereas in metLNs of HER2 + and TNBC, they ended with T04_CD4-ANXA1 or T05_CD4-CCR7 (Fig. 2f). Expression of heat shock genes increased along the inferred pseudotime axis (Supplementary Figure S3g). And for CD8 + T cells, T11_CD8-CCR7 was selected as the root of the trajectory path (Fig. 2g). Interestingly, the differentiation pattern of CD8 + T cells in metLNs of different subtypes of breast cancer varied. In metLNs of TNBC, T10_CD8-CXCL13 served as the main terminal state CD8 + T cell. Whereas in both metLNs of luminal-like and HER2 + breast cancer, CD8 + T cells differentiated to T09_CD8-LMNA, and also to T12_CD8-TRAC in metLNs of luminal-like breast cacner and further from T12 to T13 in metLNs of HER2 + breast cancer. Alongside this transition process, “naïve” marker gene expression levels (CCR7 and SELL) significantly decreased, while genes related to cytotoxicity (GZMA, GZMK, and GZMB) and exhaustion (CXCL13, CTLA4 and LAG3) exhibited an increase (Supplementary Figure S3h). In summary, the metLNs of luminal-like breast cancer had a larger proportion of CD4 + T cells compared to other molecular subtypes, including abundant naïve T01_CD4-SATB1 and activated but anergic T02_CD4-HSPA1A. And CD8 + T cells in metLNs of HER2 + and TNBC tended to differentiate into functional cells to exert anti-tumor effects. Overall, T cells in metLNs of luminal-like breast cancer were more likely to differentiate into low-functioning cells. NK cells consisted of distinct groups in metLNs of different subtypes of breast cancer, based on the expression levels of NCAM1 (CD56) and FCGR3A (CD16) (Supplementary Figure S3i). In metLNs of luminal-like breast cancer, CD56brightCD16lo NK cells were the predominant population and, similar to T02 cells, highly expressed HIF1A and stress gene signature, which might contribute to an NK cell dysfunctional phenotype [47]. The NK cells in metLNs of TNBC were CD56dimCD16hi and highly expressed PRF1 and GZMB, which agreed with the previous NK cell classification [24]. NK cells in metLNs of TNBC also highly expressed chemokine receptors (Supplementary Figure S3j), indicating that this group of NK cells might be recruited. NK cells in metLNs of HER2 + did not exhibit a genetic signature, which was probably due to the small number (267 cells) of NK cells. NK cells in metLNs also exhibited distinctive phenotypes of activating and inhibiting receptors. Distinct crosstalks in MetLNs contributing to heterogeneity of T cells To determine the potential mechanism driving distinct T cell infiltration patterns in metLNs of different molecular subtypes of breast cancer, we used CellChat to predict ligand-receptor interactions between other cells and T cells (Supplementary Figure S4a). Based on the strength of the interactions, we found that other cells showed strong interactions with CD4 + T cells in the metLNs of luminal-like breast cancer, whereas the interactions were mainly with CD8 + T cells in metLNs of HER2 + and TNBC (Fig. 3a). Fig. 3Distinct crosstalks in metLNs contributing to heterogeneity of T cells. (a) Chord diagrams of the inferred strength of signaling networks between other cells and T cells in three subtypes of metLNs. (b) Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between other cells and T cells in three subtypes of metLNs. (c) Bubble plots of ligand–receptor pairs of MHC-I and MHC-II signaling pathways between other cells and T cells with significant differentially expressed across three subtypes of metLNs. Dot color reflects communication probabilities, and dot size represents computed p-values. Empty space means the communication probability is zero. P-values are computed from a two-sided permutation test. (d) Violin plots of B cells and myeloid cells antigen processing and presentation scores compared in metLNs of three subtypes of breast cancer. P-values were computed from a two-sided wilcoxon test Distinct crosstalks in metLNs contributing to heterogeneity of T cells. (a) Chord diagrams of the inferred strength of signaling networks between other cells and T cells in three subtypes of metLNs. (b) Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between other cells and T cells in three subtypes of metLNs. (c) Bubble plots of ligand–receptor pairs of MHC-I and MHC-II signaling pathways between other cells and T cells with significant differentially expressed across three subtypes of metLNs. Dot color reflects communication probabilities, and dot size represents computed p-values. Empty space means the communication probability is zero. P-values are computed from a two-sided permutation test. (d) Violin plots of B cells and myeloid cells antigen processing and presentation scores compared in metLNs of three subtypes of breast cancer. P-values were computed from a two-sided wilcoxon test MHC-I signaling pathways were predominantly active in metLNs of TNBC, while MHC-II signaling pathways existed in metLNs of luminal-like and HER2 + breast cancer (Fig. 3b). Therefore, we focused on the communication probability of antigen presenting ligand-receptor pairs between antigen-presenting cells, epithelial cells and T cells (Fig. 3c). The MHC-II signaling pathway mainly targeted anergy T02_CD4-HSPA1A in metLNs of luminal-like breast cancer and targeted T06_CD4-FOXP3 in metLNs of HER2 + breast cancer. The MHC-I signaling pathway mainly targeted T10_CD8-CXCL13 in metLNs of TNBC, while in metLNs of luminal-like and HER2 + breast cancer T09_CD8-LMNA were targeted. The interactions between epithelial cells and CD8 + T cells were weak, suggesting the importance of cross-presentation by antigen-presenting cells in metLNs [48, 49]. We also evaluated the antigen processing and presentation scores of B cells and myeloid cells. Although the interactions between B cells and T cells were stronger, myeloid cells had higher antigen processing and presentation scores than B cells in metLNs of all three subtypes of breast cancers (Fig. 3d), which suggested that myeloid cells served as important antigen-presenting cells in metLNs. In addition, other signaling pathways also varied in metLNs of different subtypes of breast cancers. MIF signaling pathway was predominantly observed in metLNs of luminal-like breast cancer (Fig. 3b). MIF-(CD74 + CXCR4) and MIF-(CD74 + CD44) pairs mainly existed between epithelial cells and T01_CD4-SATB1, T02_CD4-HSPA1A, and MIF was also highest expressed in epithelial cells (Supplementary Figure S4a, b). Previous studies have shown that the MIF signaling pathway suppressed T cell function and caused the targeted cells to produce HIF1A [50, 51], which might be responsible for the differentiation of T01_CD4-SATB1 in metLNs of luminal-like breast cancer to the highly HIF1A-expressing T02_CD4-HSPA1A. The MDK-NCL pair was predominantly present in metLNs of luminal-like and HER2 + breast cancer, and MDK was expressed mainly by epithelial cells (Fig. 3b, Supplementary Figure S4c). This interaction may be associate with an immunosuppressive microenvironment [52]. LGALS9_CD44 could enhance the stability and function of Tregs [53]. This interaction existed in metLNs of HER2 + and TNBC (Supplementary Figure S4a), which might explain the relatively low proportion of T06_CD4-FOXP3 in metLNs of luminal-like breast cancer. In metLNs of HER2 + and TNBC, the MHC-I and MHC-II signaling pathways were presenting to functional cells, whereas in metLNs of luminal-like breast cancer, they were presenting mainly to low-functional cells. T10_CD8-CXCL13 were important cells for antitumor immunity in metLNs of TNBC, while T02_CD4-HSPA1A received antigenic presentation by MHC-II signaling pathway in metLNs of luminal-like breast cancer but were anergy with low co-stimulation scores. Overall, distinct crosstalks with other cells may drive distinct T cell infiltration patterns in metLNs of different molecular subtypes of breast cancer. Myeloid cells shape the immune tolerance microenvironment in MetLNs of luminal-like breast cancer As important antigen-presenting cells, myeloid cells were re-clustered, which were further partitioned into 13 clusters, including mast cells (M01), dendritic cells (DCs) (M02-M10), and macrophages (M11-M13) (Fig. 4a, b Supplementary Figure S5a) [26]. DCs included conventional DCs type 1 (cDC1) (M02), conventional DCs type 2 (cDC2) (M03-M05), plasmacytoid DCs (pDCs) (M06-M08), and mature conventional DCs (M09-M10), characterized by specific high expression of CLEC9A, CD1C, LILRA4, and LAMP3, respectively. We found that in uiLNs of different subtypes, myeloid cells were composed of a large proportion of DCs and a small proportion of mast cells and macrophages (Fig. 4c). While in metLNs, the proportion of DCs and macrophages varied widely, with metLNs of TNBC having the highest proportion of macrophages and metLNs of luminal-like breast cancer having the lowest proportion of macrophages but higher proportions of M06_DC-IRF4 and M09_DC-CD1B (Fig. 4c). All macrophages in the three clusters had an M2 phenotype [25] (Supplementary Figure S5b). M11_macro-C1QC exhibited significant higher phagocytosis scores [26] and were present in metLNs of all three subtypes of breast cancer. M13_macro-SPP1 expressed a highest pro-angiogenic signature, which might be related to the expression of SPP1 [37, 54], and were specifically present in metLNs of luminal-like breast cancer (Supplementary Figure S5d). Among all myeloid cells in metLNs, M02_DC-CLEC9A, M03_DC-TIMP1, and M11_macro-C1QC had higher antigen processing and presentation scores. KEGG pathway analysis also found these clusters enriched in antigen processing and presentation pathway (Fig. 4d, Supplementary Figure S5e, f). Fig. 4Myeloid cells shaping the immune tolerance microenvironment in metLNs of luminal-like breast cancer. (a) UMAP visualization of 13 myeloid cell clusters. (b) UAMP showing myeloid cell across three subtypes of breast cancer. (c) Histogram showing the proportion of myeloid cells in uiLNs and metLNs across three subtypes of breast cance. (d) Violin plots showing different expression patterns of antigen processing and presentation scores among dendritic cells and macrophage subclusters in metLNs. M04 (3 cells) excluded owing to small quantities. P-values are computed from a two-sided wilcoxon tests. (e) Chord diagrams of the inferred strength of signaling networks between myeloid cells and T cells in metLNs (left panel). Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between myeloid cells and T cells in three subtypes of metLNs (right panel). (f) Chord diagrams of the inferred MHC-II (left panel) and MHC-I (right panel) signaling pathway networks between myeloid cells and T cells in three subtypes of metLNs Myeloid cells shaping the immune tolerance microenvironment in metLNs of luminal-like breast cancer. (a) UMAP visualization of 13 myeloid cell clusters. (b) UAMP showing myeloid cell across three subtypes of breast cancer. (c) Histogram showing the proportion of myeloid cells in uiLNs and metLNs across three subtypes of breast cance. (d) Violin plots showing different expression patterns of antigen processing and presentation scores among dendritic cells and macrophage subclusters in metLNs. M04 (3 cells) excluded owing to small quantities. P-values are computed from a two-sided wilcoxon tests. (e) Chord diagrams of the inferred strength of signaling networks between myeloid cells and T cells in metLNs (left panel). Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between myeloid cells and T cells in three subtypes of metLNs (right panel). (f) Chord diagrams of the inferred MHC-II (left panel) and MHC-I (right panel) signaling pathway networks between myeloid cells and T cells in three subtypes of metLNs We further analyzed the crosstalk between myeloid and T cells in metLNs (Fig. 4e). The MHC-I signaling pathway was also predominant in metLNs of TNBC, while the MHC-II signaling pathway was mainly in metLNs of luminal-like and HER2 + breast cancer. Similarly, MHC-I signaling pathway primarily targeted T10_CD8-CXCL13 with antitumor function in metLNs of TNBC and T09_CD8-LMNA in metLNs of luminal-like and HER2 + breast cancer. DCs are very important cells for cross-presentation of antigens and are essential for the initiation of CD8 + T cell responses [48, 55]. In metLNs of breast cancer, M03_DC-TIMP1 might play an important anti-tumor role through cross-presentation of antigens. MHC-II signaling pathway mainly targeted low-functional T02_CD4-HSPA1A in metLNs of luminal-like breast cancer and T06_CD4-FOXP3 in metLNs of HER2 + breast cancer (Fig. 4f). Importantly, SPP1-CD44 axis participated in immune suppression and tumor promotion [54]. We found that the SPP1-CD44 pair was highly expressed in metLNs of luminal-like breast cancer (Fig. 4e). Outgoing and incoming communication probabilities also showed that this signaling pathway might be enriched between M13_macro-SPP1 and T01_CD4-SATB1, T02_CD4-HSPA1A (Supplementary Figure S5c), all of which were found to be in large numbers in metLNs of luminal-like cells. Myeloid cells were mainly presenting to functional CD4 + and CD8 + T cells in metLNs of HER2 + and TNBC, and to low-functional T cells such as T02_CD4-HSPA1A in metLNs of luminal-like breast cancer. Overall, myeloid cells in metLNs of luminal-like breast cancer had an immunosuppressive impact on the tumor immune microenvironment. B cells participate in the immune tolerance microenvironment in MetLNs of luminal-like breast cancer We identified 8 different B cell clusters (Fig. 5a, b, Supplementary Figure S6a) based on signature genes [56], including 2 naïve B cell clusters (B01-B02), 2 memory B cell clusters (B03-B04), 1 germinal center (GC) B cell cluster (B05), 1 plasma cell cluster (B06), and 2 proliferating B cell clusters (B07-B08). Fig. 5B cells participating in the immune tolerance microenvironment in metLNs of luminal-like breast cancer. (a) UMAP visualization of 8 B cell clusters. (b) UAMP showing B cells across three subtypes of breast cancer. (c) Histogram showing the proportion of B cells in uiLNs and metLNs across three subtypes of breast cancer. (d) Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between B cells and T cells in three subtypes of metLNs. (e) Heatmaps showing the relative signaling strength of incoming signaling patterns of T cell subclusters between B cells and T cells interaction network in metLNs across three subtypes of breast cancer B cells participating in the immune tolerance microenvironment in metLNs of luminal-like breast cancer. (a) UMAP visualization of 8 B cell clusters. (b) UAMP showing B cells across three subtypes of breast cancer. (c) Histogram showing the proportion of B cells in uiLNs and metLNs across three subtypes of breast cancer. (d) Bar plots of ranked signaling pathways based on differences in the overall information flow within the inferred networks between B cells and T cells in three subtypes of metLNs. (e) Heatmaps showing the relative signaling strength of incoming signaling patterns of T cell subclusters between B cells and T cells interaction network in metLNs across three subtypes of breast cancer In uiLNs of distinct subtypes of breast cancer (Fig. 5c), B cells had a similar composition, consisting of a high proportion of naïve B cells (B01_naive-IGHD), a small proportion of GC B cells (B05_GC-LRMP) and plasma cells (B06_plasma-MZB1). In metLNs (Fig. 5c), naïve B cells were still predominant but mainly B02_naive-FCER2, and the highest proportion of naïve B cells was found in metLNs of TNBC. Memory B cells varied across metLNs of three subtypes of breast cancer. MetLNs of luminal-like breast cancer contained the highest proportion of memory B cells with B03_memory-ANXA1 predominanting. In metLNs of HER2 + breast cancer, the majority was B04_memory-SERHL2, whereas there were few memory B cells in metLNs of TNBC. Proliferating B cells were predominantly present in metLNs of luminal-like breast cancer. We performed functional enrichment of B cell clusters in metLNs (Supplementary Figure S6b). Antigen processing and presentation gene sets were significantly enriched in B02_naive-FCER2, and B06_plasma-MZB1 were related to protein processing and export. Cell cycle and metabolism-related pathways were enriched in B05_GC-LRMP. We also found significant enrichment of CD4 + T cell differentiation, such as Th17 cell differentiation, Th1 and Th2 cell differentiation, in B02_naive-FCER2 and B03_memory-ANXA1. B cells also act as antigen-presenting cells. Although they are weak in nonspecific antigen uptake, B cells are sensitive to low concentrations of antigen [57]. So we also investigated the interaction network between B cells and T cells (Fig. 5d, e). Although the antigen processing and presentation scores of B cells were lower than those of myeloid cells, the antigen presentation signaling pathways between B and T cells were similar with myeloid and T cells. In metLNs of luminal-like breast cancer, the MHC-I signaling pathways were mainly targeting T09_CD8-LMNA and the MHC-II signaling pathways were mainly targeting T02_CD4-HSPA1A. The MHC-II signaling pathways were mainly targeting T06_CD4-FOXP3 in metLNs of HER2+, and the MHC-I signaling pathways were targeting T10_CD8-CXCL13 in metLNs of TNBC. Moreover, the MK and NECTIN signaling pathways were specifically active in metLNs of luminal-like breast cancer, which probably correlated with the immune tolerant microenvironment in metLNs of luminal-like breast cancer [58]. CD4+Tstr cells may derive not from draining lymph nodes but, are associated with local tumor microenvironment To further investigate the migration characteristics of stress response state CD4 + T cells (CD4 + Tstr cells, T02_CD4-HSPA1A), scRNA and single-cell TCR data of 8 paired primary tumors and metLNs were analyzed [30]. According to the clusters previously reported [30], 15 clusters were identified from 59,327 T&NK cells (Fig. 6a, b). Of CD4 + T cells, 7 clusters were found. The functions of these CD4 + T cell clusters were calculated (Fig. 6c), and CD4-C4-FOS cluster from these samples were recognized as CD4 + Tstr cells. To determine the distribution of CD4 + Tstr cells, the proportions were compared between primary tumors and metLNs (Supplementary Figure S7a, b). Less CD4 + Tstr cells were found in primary breast cancers than in metLNs (Supplementary Figure S7b), although no significant difference was observed. Fig. 6CD4 + Tstr cells may derive not from draining lymph nodes but, are associated with local tumor microenvironment. (a) Schematic overview of the analytical workflow of Liu et al. (GSE190870). (b) UMAP visualization of 15 T&NK cell clusters. (c) Heat map illustrating expression of gene signatures across 7 CD4 + T cell clusters. Heat map was generated based on the scaled gene signature scores. (d) Violin plots comparing the percentages of T cells with T cell receptor (TCR) clone size ≥ 3 of all T&NK cells in primary tumor and metLNs samples. Each point represented one sample. P-values were calculated by a two-sided paired t-test. (e) Violin plots comparing the percentages of T cells with matched TCR of all T&NK cells in primary tumor and metLNs samples. Each point represented one sample. P-values were calculated by a two-sided paired t-test CD4 + Tstr cells may derive not from draining lymph nodes but, are associated with local tumor microenvironment. (a) Schematic overview of the analytical workflow of Liu et al. (GSE190870). (b) UMAP visualization of 15 T&NK cell clusters. (c) Heat map illustrating expression of gene signatures across 7 CD4 + T cell clusters. Heat map was generated based on the scaled gene signature scores. (d) Violin plots comparing the percentages of T cells with T cell receptor (TCR) clone size ≥ 3 of all T&NK cells in primary tumor and metLNs samples. Each point represented one sample. P-values were calculated by a two-sided paired t-test. (e) Violin plots comparing the percentages of T cells with matched TCR of all T&NK cells in primary tumor and metLNs samples. Each point represented one sample. P-values were calculated by a two-sided paired t-test In this study, we defined T cells with at least one pair of identical paired α-β chains as one clone from the same ancestry, and the expanded clones were defined as those whose α and β TCR pairs were shared by at least three cells in a given cell population [59]. Of the 47,803 cells with TCR data, most were unique TCR. Interestingly, clonal (n ≥ 3) TCRs were mainly in CD8 + T cells, and the primary tumors showed more clonal (n ≥ 3) TCRs than the paired metLNs (Fig. 6d, Supplementary Figure S7c). To determine whether CD4 + Tstr cells in primary tumors migrated from lymph nodes, TCR sequences between primary tumors and metLNs were compared. CD4 + Tstr cells seldom shared the same TCR sequence between primary tumors and paired metLNs (Fig. 6e, Supplementary Figure S7d). Overall, un-expanded CD4 + Tstr cells did not migrate from paired metLNs, but were associated with local tumor microenvironment. CD4 + Tstr cells are associated with poor survival and unfavorable responses to immunotherapy in TNBC Although CD4 + Tstr cells were abundant in metLNs of luminal-like breast cancer, they were present in all three subtypes of breast cancer (Supplementary Figure S8a-c). We further performed immunofluorescence analysis (Fig. 7a) to demonstrate the existence of HSPA1A-expressing CD4 + T cells in metLNs. In addition, we compared function of T06_CD4-FOXP3 (CD4 + CD25+) and CD4 + Tstr (CD4 + HSP70+) using flow cytometry on primary breast tumors (Fig. 7b, c). IFNG, a crucial modulator of immunity that can be produced by CD4 + T cells [60], was expressed at a low level by CD4 + Tstr. The low expression of CXCR5 on CD4 + Tstr also indicated that it had low expression of “Chemokine/Chemokine receptor” gene signatures (Fig. 2d). This demonstrated CD4 + Tstr anergy. The previous study [61] has found that increased tumor-infiltrating lymphocytes predict response to neoadjuvant chemotherapy in all subtypes and are associated with a survival benefit in HER2-positive breast cancer and TNBC. By contrast, increased tumor-infiltrating lymphocytes is an adverse prognostic factor for survival in luminal-HER2-negative breast cancer. To further determine the prognostic value of CD4 + Tstr cells, CD4 + Tstr signature was evaluated in the TCGA dataset. A total of 701 patients with clearly defined subtypes and available prognostic information were included. After adjusting potential confounders including age and stage, T10_CD8-CXCL13 signature and CD4 + Tstr signature were not associated with survival in luminal-HER2-negative breast cancer (Supplementary Figure S8d, e). Importantly, CD4 + Tstr signature was significantly associated with poor progression-free interval (PFI) in TNBC patients and poor overall survival (OS) in HER2-positive breast cancer (Fig. 7d, Supplementary Figure S8d). Fig. 7CD4 + Tstr cells are associated with poor survival and unfavorable responses to immunotherapy in TNBC. (a) The immunofluorescence analysis for HSPA1A-expressing CD4 + T cells in metLNs sections of three subtypes of breast cancer. Arrows denote the area with higher magnification. (b) Scatter plot displaying the mean fluorescence intensity (MFI) of IFNG in T06_CD4-FOXP3 and stress response state CD4 + T cells (CD4 + Tstr cells) in primary breast cancer tumors (n = 10), with points connected by lines representing the same patient. Data are expressed as mean ± standard deviation (SD), and p-value was calculated from a paired t-test. Representative flow cytometric histograms showing the expression of IFNG. (c) Scatter plot displaying the MFI of CXCR5 in T06_CD4-FOXP3 and CD4 + Tstr cells in primary breast cancer tumors (n = 10), with points connected by lines representing the same patient. Data are expressed as mean ± SD, and p-value was calculated from a paired t-test. Representative flow cytometric histograms showing the expression of CXCR5. (d) Prognostic value of specific characteristics and cluster signatures for progression-free interval (PFI) in TNBC patients from The Cancer Genome Atlas Program (TCGA) dataset. Forest plot showing hazard ratio (HR) and 95% confidence interval (CI) derived from multivariate Cox analyses adjusted for age and stage. Within the forest plot, HR for each variable is depicted as a box, and 95% CIs are shown as horizontal lines. The vertical line crossing the value of 1 represents a non-statistically significant effect, and odds greater than one indicate worse effects. (e) Schematic overview of the analytical workflow of Shiao et al. (GSE246613). (f) UMAP visualization of 8 T cell clusters from Shiao et al. (GSE246613). (g) Heat map illustrating expression of anergy CD4 + Tstr gene signature across 8 T cell clusters. Heat map was generated based on the scaled gene signature scores. (h) Violin plots comparing Tcell_02 cluster proportion in non-responders (NR) and responders (R) of all T cells. P-value was calculated from a two-sided t-test CD4 + Tstr cells are associated with poor survival and unfavorable responses to immunotherapy in TNBC. (a) The immunofluorescence analysis for HSPA1A-expressing CD4 + T cells in metLNs sections of three subtypes of breast cancer. Arrows denote the area with higher magnification. (b) Scatter plot displaying the mean fluorescence intensity (MFI) of IFNG in T06_CD4-FOXP3 and stress response state CD4 + T cells (CD4 + Tstr cells) in primary breast cancer tumors (n = 10), with points connected by lines representing the same patient. Data are expressed as mean ± standard deviation (SD), and p-value was calculated from a paired t-test. Representative flow cytometric histograms showing the expression of IFNG. (c) Scatter plot displaying the MFI of CXCR5 in T06_CD4-FOXP3 and CD4 + Tstr cells in primary breast cancer tumors (n = 10), with points connected by lines representing the same patient. Data are expressed as mean ± SD, and p-value was calculated from a paired t-test. Representative flow cytometric histograms showing the expression of CXCR5. (d) Prognostic value of specific characteristics and cluster signatures for progression-free interval (PFI) in TNBC patients from The Cancer Genome Atlas Program (TCGA) dataset. Forest plot showing hazard ratio (HR) and 95% confidence interval (CI) derived from multivariate Cox analyses adjusted for age and stage. Within the forest plot, HR for each variable is depicted as a box, and 95% CIs are shown as horizontal lines. The vertical line crossing the value of 1 represents a non-statistically significant effect, and odds greater than one indicate worse effects. (e) Schematic overview of the analytical workflow of Shiao et al. (GSE246613). (f) UMAP visualization of 8 T cell clusters from Shiao et al. (GSE246613). (g) Heat map illustrating expression of anergy CD4 + Tstr gene signature across 8 T cell clusters. Heat map was generated based on the scaled gene signature scores. (h) Violin plots comparing Tcell_02 cluster proportion in non-responders (NR) and responders (R) of all T cells. P-value was calculated from a two-sided t-test Tumor-infiltrating lymphocytes in TNBC are associated with the response to immune checkpoint inhibitors [61, 62]. Considering that CD4 + Tstr cells may be related to immune tolerance, CD4 + Tstr cells may predict the response to immunotherapy in TNBC. Shiao et al. conducted a neoadjuvant trial [63] in patients diagnosed with TNBC. Single-cell RNA sequencing data were obtained from 34 tumors of 33 patients before receiving neoadjuvant regimens containing pembrolizumab. Among them, 23 patients responded with RCB scores of 0–1, while 10 patients were non-responders with RCB scores of 2–3 (Fig. 7e). According to the 8 T cell clusters provided by Shiao et al. (Fig. 7f), T cell_02 had the highest CD4 + Tstr signature score (Fig. 7g), recognized as CD4 + Tstr cells. Interestingly, non-responders showed a significantly higher proportion of CD4 + Tstr cells than responders (Fig. 7h). Overall, CD4 + Tstr cells may be associated with poor survival and predict tolerance to immunotherapy in TNBC patients. Discussion Axillary lymph nodes are the most common sites of metastasis in breast cancer. As important immune organs, regional lymph nodes are also recognized as the source and reservoir of tumor-responsive T cells. However, the immune landscape of lymph nodes across different subtypes of breast cancer has not been clearly reported. The metLNs showed heterogeneity of immune cells across different subtypes of breast cancer, with a stress response state CD4 + T cells mainly in luminal-like breast cancer. CD4 + Tstr cells received antigen presentation but were not expanded, with a state of anergy. Moreover, CD4 + Tstr cells in primary tumors were not elicited from paired metLNs, associated with local tumor microenvironment. Importantly, CD4 + Tstr cells were associated with poor prognosis and resistant to immunotherapy for TNBC (Figure S9). Although CD4 + Tstr cells have been reported previously, the characteristics of these cells and the association with immune tolerance in breast cancer are firstly reported in this study, to the best of our knowledge. HER2 is overexpressed in 15–20% of breast cancers. Neoadjuvant therapy and HER2-targeting treatment have clearly improved the prognosis of HER2-positive breast cancer [64]. However, many patients experience relapse, and deaths resulting from HER2-positive breast cancer remain persistently high. Our study indicated that CD4 + Tstr cells were associated with overall survival in HER2 + breast cancer. Tumor-infiltrating lymphocytes have also been reported to predict response in patients receiving neoadjuvant therapy and serve as a long-term outcome prognostic indicator for HER2-positive breast cancer [61, 65]. This all suggests the significant role of immune cells in HER2-positive breast cancer. There have been abundant studies investigating into the landscape of circulating and tumor-infiltrating immune cells. Our study first focused on the immune landscapes in uninvolved and involved lymph nodes of different subtypes of breast cancer. The uiLNs contained similar immune compartments across different subtypes of breast cancer, while metLNs of different subtypes showed heterogeneity of T cells, with an anergic CD4 + Tstr cell cluster mainly in luminal-like breast cancer. Since regional lymph nodes are the sites where tumor-specific immune responses are generated, our study might enhance the comprehension of tumor immunology. By comparing the TCR data from paired primary tumors and metLNs, we determined the important CD4 + Tstr cells were not derived from lymph node. The immune tolerance in lymph nodes of luminal-like breast cancer might explain the immunosuppressive systemic and intra-tumoral immune landscapes in this subtype. Moreover, as lymph nodes are origins of responder cells in immunotherapies, our comparison of the differences across different tumors could facilitate understanding the distinct immunotherapy response rates in different tumor types. In our study, we identified a unique CD4 + Tstr cell cluster which mainly existed in luminal-like breast cancer. The cluster shared similar phenotypic features with the previously reported Tstr cells [22]. In the previous pan-cancer analysis, Tstr cells were enriched in immunotherapy nonresponsive tumors including basal cell carcinoma, non-small cell lung cancer and renal cell carcinoma. The cell cluster existed in TNBC, while their implication in immunotherapies was unclear. The study indicated a potential role of Tstr cells in immunotherapy resistance, but the functioning mechanism and the origin of this cell cluster remained unexplained. Here in this study, Tstr cells were found to be in a state of stress, hypoxia, activation but anergy, which lacked proper costimulatory function. Immunofluorescence also validated the colocalization of CD4 and HSP70. Analysis of crosstalks of Tstr cells indicated that MIF signaling pathway sourced from epithelial cells might be responsible for the anergy of this cell type. Given the high expression of HIF1A on Tstr cells, this pathway probably played a crucial role in the formation and function of Tstr cells. As the upstream antigen presenting cells, myeloid cells could shape the immune tolerance through SPP1-CD44 interaction with Tstr cells, and B cells might also play a role through MK and NECTIN signaling pathways. Thus, we interpreted the functions of Tstr cells and how these cells might be induced in lymph nodes. Prognostic analysis suggested that in the TCGA cohort, CD4 + Tstr signature correlated with progression-free interval in TNBC patients. Additional analysis is needed to explore whether the association between Tstr cells and prognosis changes when adjusted for stress/exhaustion modules. By establishing the correlation of Tstr cells with unfavorable responses to immunotherapy, we suggested that Tstr cells might be associated with the immunosuppressive microenvironment. Further association analysis of the CD4 + Tstr signature with IFNG or antigen-presentation signatures as surrogates for immune activity could provide a more comprehensive demonstration of the immunosuppressive function of CD4 + Tstr cells. This suggested the feasibility of targeting CD4 + Tstr cells to block immune tolerance and immunotherapy resistance in breast cancer. Several limitations still existed in this study. First, data enrolled in this study were derived from two test platforms. So, the immune cells of uiLNs and metLNs were not directly compared. Second, because of the small sample size and lack of immunotherapy datasets, we only validated CD4 + Tstr cells resistance to immunotherapy in TNBC. Subsequent large cohort studies are necessary to validate the finding in all breast cancer subtypes. Final, the clinical value and the prediction of breast cancer prognosis of CD4 + Tstr cells were still at the hypothesis-generating stage. Mechanistic studies are required to extend clinical significance, and prospective research is needed to validate these findings. Conclusions In summary, we identified a specific cluster of CD4 + T cells related to immune tolerance that provided possible prognostic information and a new predictor for immunotherapy for breast cancer. Future studies are still needed to confirm our results. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6