Single-cell multiomics reveals exosome-mediated reprogramming and clonotypic remodeling of T cells in triple-negative breast cancer

Abstract

Triple-negative breast cancer (TNBC) is an aggressive and immunogenic subtype lacking targeted therapies. While tumor-derived exosomes are known to modulate immune function, their direct impact on human T cell plasticity and antigen specificity remains poorly defined. Here, we conducted a comprehensive single-cell multiomic analysis of primary human T cells exposed to exosomes derived from 17 genomically diverse TNBC cell lines and 35 patient samples. Integrating single-cell RNA-seq, V(D)J sequencing, non-coding RNA profiling, bulk and single-cell cytokine analyses, we uncovered conserved and subtype-specific immunomodulatory programs induced by TNBC exosomes. Exosome-treated T cells displayed skewing toward regulatory and dysfunctional phenotypes, including Th17-like, Treg, and PD-1⁺/PD-L1⁺ Tfh cells. Functional profiling revealed suppression of early activation markers and cytokine responses, alongside selective preservation of cytotoxic features in γδ T and NKT subsets. Transcriptomic and miRNA network analyses demonstrated widespread downregulation of immune effector genes (e.g., HBEGF and TNFSF9) mediated by exosome-delivered regulatory miRNAs (has-miR-98-5p). Notably, exosome-stimulated T cells displayed distinct clonotypic expansions, characterized by the emergence of five tumor-specific γδ TCR clonotypes and 30 unique αβ TCR CDR3 sequences that were absent in mock-treated controls, underscoring the role of exosomes in shaping TCR repertoire dynamics.

Introduction

Triple-negative breast cancer (TNBC) is an aggressive and molecularly heterogeneous subtype of breast cancer defined by the absence of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) expression.^1,2 Although TNBC accounts for only approximately 10–15% of all breast cancer cases, it contributes disproportionately to breast cancer-related mortality because of its high metastatic potential, early recurrence, and limited treatment options.^3–5 TNBC is more frequently diagnosed in younger women, in individuals of African ancestry, and in carriers of germline BRCA1 mutations.^3–5 In contrast to hormone receptor-positive or HER2-amplified breast cancers, TNBC lacks well-established actionable molecular targets, leaving cytotoxic chemotherapy as the principal systemic treatment for many patients. While some patients initially respond to chemotherapy, resistance is common, and relapse often occurs early in the clinical course. This aggressive biology is reflected in survival data showing that the 5-year survival rate falls sharply from more than 90% in patients with localized disease to approximately 12% in those with metastatic TNBC.⁶ These sobering outcomes underscore the urgent need to better understand the mechanisms that drive TNBC progression, dissemination, immune escape, and therapeutic resistance. A clearer mechanistic understanding of TNBC biology is essential for the development of more effective and durable treatment strategies tailored to this highly aggressive disease subtype.

A defining and somewhat paradoxical feature of TNBC is its relatively immunogenic phenotype compared with other breast cancer subtypes. Multiple studies have shown that TNBC tumors frequently harbor increased levels of tumor-infiltrating lymphocytes (TILs) and display elevated expression of immune-related and inflammatory gene signatures.^7,8 These features have fueled substantial interest in immunotherapy, particularly immune checkpoint blockade targeting the PD-1/PD-L1 axis.⁹ Indeed, the presence of TILs and an inflamed tumor microenvironment has supported the notion that TNBC may be particularly amenable to immune-based therapeutic strategies. However, despite encouraging results in selected patient subsets, the overall clinical benefit of checkpoint inhibitors in TNBC remains limited, and many patients either fail to respond or eventually develop resistance.⁹ These findings indicate that the immune microenvironment of TNBC is more complex than can be explained by checkpoint pathway activation alone. T cells within TNBC lesions often display impaired cytotoxicity, altered activation states, reduced proliferative capacity, dysfunctional cytokine responses, and features of exhaustion that collectively compromise anti-tumor immunity. Furthermore, given the substantial molecular heterogeneity of TNBC, it is likely that distinct tumor subtypes engage different mechanisms to shape immune behavior and suppress effective T cell responses. A deeper understanding of how TNBC tumors actively reprogram T cell function is, therefore, critical for overcoming resistance to immunotherapy and for identifying new therapeutic opportunities that can restore or enhance anti-tumor immune responses.

Among the factors that may mediate this tumor–immune crosstalk, exosomes have emerged as central regulators of intercellular communication within the tumor microenvironment.¹⁰ Exosomes are small extracellular vesicles, typically 30–150 nm in diameter, that are secreted by most cell types and carry a complex cargo of proteins, lipids, DNA, mRNAs, microRNAs, and long non-coding RNAs. Rather than representing passive cellular by-products, exosomes function as biologically active signaling vesicles capable of altering the phenotype and function of recipient cells through transcriptional, post-transcriptional, and epigenetic mechanisms.¹⁰ In cancer, tumor-derived exosomes have been implicated in multiple hallmarks of disease progression, including angiogenesis, metastatic niche formation, therapy resistance, and immune evasion.^11–14 They can modulate anti-tumor immunity by expanding regulatory T cells, suppressing cytotoxic lymphocyte activity, altering myeloid cell polarization, and interfering with antigen presentation pathways.^11–14 In TNBC specifically, exosomes have been linked to epithelial-to-mesenchymal transition, increased invasiveness, angiogenesis, and the establishment of an immunosuppressive microenvironment.^11–14 Despite these important observations, the full extent to which TNBC-derived exosomes shape the functional and transcriptional states of human T cells remains incompletely understood. This gap is particularly important because TNBC is not a single disease entity, but rather a collection of molecularly diverse subtypes with distinct biological behaviors. Understanding how exosomal cargo varies across TNBC models and how this diversity influences human T cell programming is, therefore, essential for elucidating mechanisms of immune suppression and identifying subtype-specific therapeutic vulnerabilities.

In this study, we present an integrative multi-omic analysis of the effects of exosomes derived from 17 molecularly diverse TNBC cell lines on human immune cells. These TNBC cell lines were selected to capture inter-tumoral heterogeneity in genomic alterations, aggressiveness, and molecular subtype, thereby providing a robust model system for studying the breadth of TNBC-associated immune modulation. Using single-cell RNA sequencing, we profiled transcriptional changes in human CD3⁺ T cells following exposure to TNBC-derived exosomes, allowing high-resolution characterization of exosome-induced immune remodeling at the single-cell level. To complement these analyses, we also incorporated non-coding RNA datasets, including microRNA and long non-coding RNA profiles, in order to define regulatory circuits that may drive altered T cell states. This integrated framework enabled us to identify both conserved and subtype-specific immunomodulatory programs induced by TNBC exosomes, including pathways linked to immune checkpoint regulation, cytokine signaling, T cell activation, and T cell exhaustion. Notably, our analysis revealed a panel of exosome-associated coding and non-coding RNAs connected to known mechanisms of immune suppression and resistance to immunotherapy. By integrating single-cell transcriptomics with exosomal non-coding RNA profiling across a diverse TNBC panel, this work provides a comprehensive view of how TNBC-derived exosomes reshape the transcriptional and regulatory landscape of human T cells. Collectively, these findings establish a broader framework for understanding exosome-mediated immune modulation in TNBC and highlight potential targets for therapeutic intervention aimed at restoring effective anti-tumor immunity in this highly aggressive breast cancer subtype.

Results

To elucidate the complex immunological responses elicited by triple-negative breast cancer (TNBC), we performed a comprehensive transcriptional analysis of T cells exposed to exosomes derived from 17 distinct TNBC cell lines, in comparison to mock-treated controls. This investigation yielded a high-resolution single-cell reference atlas, providing unprecedented insight into the transcriptional and phenotypic landscapes of human T cells in response to TNBC-derived exosomal stimuli. Exosomes were isolated by size exclusion chromatography and characterized via nanoparticle tracking analysis (NTA) and transmission electron microscopy (TEM).

Leveraging a multi-omic framework, we integrated single-cell RNA sequencing (scRNA-seq), miRNA-mRNA network analysis, high-throughput bulk and single-cell cytokine profiling, and T cell receptor (VDJ-seq) sequencing. This multifaceted approach enabled robust identification of recurrently dysregulated genes and pathways, offering mechanistic insights into the immunomodulatory effects of TNBC exosomes on distinct T cell subsets (Fig. 1).

Fig. 1.

Schematic overview of the multi-omic framework to assess TNBC exosome-mediated T cell modulation. Exosomes were isolated from 17 molecularly distinct TNBC cell lines and used to treat primary human T cells; untreated T cells served as mock controls. Following exposure, single-cell RNA and surface protein (ADT) profiling, as well as T cell receptor (TCR) V(D)J sequencing, were performed using the 10x Genomics platform. Transcriptomic data underwent normalization, clustering, differential expression, and pathway enrichment analysis. TCR clonotype data were analyzed for CDR3 sequence and V-J-C gene usage to identify tumor-associated T cell expansions. In parallel, exosomal miRNA and lncRNA expression profiles were used for regulatory network analysis. Functional cytokine responses were assessed using Luminex and IsoCode single-cell cytokine analysis chips, highlighting exosome-driven immune modulation

Subset-specific uptake of TNBC-derived exosomes by human T cells

Exosomes were isolated via size exclusion chromatography from the conditioned media of 17 molecularly and genetically distinct triple-negative breast cancer (TNBC) cell lines, obtained from the American Type Culture Collection (ATCC). This panel included both basal-like (HCC1599, HCC1937, HCC1143, MDA-MB-468, HCC38, HCC70, HCC1806, HCC1187, DU4475), mesenchymal (BT-549, MDA-MB-231, MDA-MB-436, MDA-MB-157, Hs 578T), luminal androgen receptor (LAR) (MDA-MB-453), and Unclassified subtypes (HCC1395, BT-20), thereby encompassing the clinical and molecular heterogeneity of TNBC.

Exosome purity was confirmed by nanoparticle tracking analysis (NTA) and transmission electron microscopy (TEM), with mean particle sizes ranging from ~80–130 nm, consistent with exosomal dimensions. Representative size distributions and morphological profiles are shown in Supplementary Fig. 1a, b. Additionally, exosomes were isolated from 35 human plasma samples, comprising 7 healthy donors and 28 TNBC patients. The average particle sizes of these plasma-derived exosomes ranged from approximately 60 to 140 nm, consistent with the expected size distribution of plasma extracellular vesicles (Supplementary Fig. 1c, d).

We next investigated differential uptake and intracellular localization of TNBC-derived exosomes across human T cell subpopulations. Single-cell imaging flow cytometry (ImageStream) revealed exosome uptake by CD4⁺, CD8⁺, and regulatory T cell (Treg; CD4⁺, CD127^low/−, CD25⁺) subsets (Supplementary Fig. 1e). High-resolution image analysis confirmed that exosomes were not only bound to the cell surface but were also internalized into the cytoplasm and perinuclear regions, consistent with active uptake. Quantitative analysis of internalization features demonstrated a significantly higher rate of intracellular exosome accumulation in Tregs compared to conventional CD4⁺ and CD8⁺ T cells (Supplementary Fig. 1h), suggesting preferential uptake by immunosuppressive subsets.

Single-cell multiomics reveals exosome-induced immunosuppressive T cell states

Following 48-h treatment with TNBC-derived exosomes, we profiled the transcriptomic and surface protein landscapes of T cell subpopulations using CITE-seq, integrating gene expression with a 137-plex panel of oligo-conjugated antibodies. This high-dimensional single-cell dataset includes 179,169 high-quality human T cells exposed to exosomes isolated from 17 genomically diverse TNBC cell lines. The combination of transcriptomic and phenotypic profiling enabled fine-grained delineation of T cell states and functional programs modulated by tumor-derived exosomal cues. To systematically classify human peripheral blood T cells, we analyzed a high-dimensional CITE-seq dataset using an approach analogous to that applied in tumor-infiltrating T cell profiling. H5 files generated by Cell Ranger from 17 TNBC samples were processed, resulting in 179,169 high-quality cells that passed stringent filtering criteria. Subsequent clustering and downstream analyses were conducted using established R packages. To improve cluster resolution, antibody-derived tags (ADTs) were integrated with gene expression data, enabling finer discrimination of T cell subsets.

The success of data integration was quantitatively assessed using kBET, which measures the degree of batch mixing within the integrated dataset. A low rejection rate indicates that cells from different experimental batches are well interspersed, thereby reflecting effective batch correction.

To evaluate the robustness of integration across the dataset, the data were randomly partitioned into 20 subsets, and the average kBET p value was computed for each. The average p values remained consistently close to 1.0 across all subsets, indicating that batch effects were effectively mitigated and that the integration was stable across different partitions of the data (Supplementary Fig. 2a).

Comparable results were obtained when the same analysis was conducted at the cell-type level, where average p values across clusters were reported (Supplementary Fig. 2b) and aggregated subset results were visualized with boxplots (Supplementary Fig. 2c). The Supplementary Fig. 2c presents a box plot comparing the expected rejection rates with the observed rejection rates across these clusters. The results provide compelling evidence of successful batch correction. The observed median rejection rate is significantly lower than the expected median, which is near 1.0. The low observed rejection rates demonstrate that within each identified cell type, cells from different batches are well-interspersed. The high expected rejection rate represents the baseline rejection rate when cells are not corrected for batch effects, which would be close to 1.0. The substantial difference between these two distributions confirms that our integration method effectively removed technical variability while preserving the biological integrity of each cell type.

Furthermore, UMAP projections colored by the original GEM well (to evaluate batch mixing) and by the identified cell types (to evaluate biological structure) provided complementary visual confirmation of the successful integration (Supplementary Fig. 2d). These findings confirm that the integration successfully removed batch effects while preserving biological variation and avoiding spurious batch-driven structure.

Marker-based annotation was performed using a curated panel of canonical T cell markers derived from the literature, incorporating both transcriptomic and protein-level features to ensure robust classification.

We systematically identified and profiled 11 major T cell subsets, including CD4⁺, CD8⁺, NKT, T helper, Treg, and γδ T cells. These subsets were distinguished based on their characteristic marker expression patterns to provide a comprehensive overview of the T cell landscape (Supplementary Fig. 2e). We first assessed IFNG expression, a key effector cytokine indicative of T cell activation (Fig. 2a), followed by a comprehensive evaluation of 43 additional markers associated with T cell identity and functional states across all clusters (Supplementary Fig. 3). This integrative analysis delineated 14 transcriptionally distinct clusters, each defined by a unique constellation of marker expression signatures (Fig. 2b). Analysis of the single-cell RNA sequencing data revealed that the NKT cell cluster exhibited basal IFNG expression but showed upregulated levels of TNF and PD-L1 relative to other clusters. This expression profile distinguishes the NKT cluster as a small, phenotypically discrete subset within the broader T cell population. Among the five T follicular helper (Tfh)-associated clusters, two were predominantly defined by canonical Tfh markers. In contrast, two other clusters demonstrated a co-expression of Tfh and Th1 markers, while the final cluster displayed a mixed Tfh and Treg phenotype. Within the two Tfh.Th1 clusters, a specific subgroup, was notable for their co-expression of the immune checkpoint markers TIGIT, PD-1, and PD-L1. This subgroup formed a distinct island on the UMAP projection, suggesting unique transcriptional and functional characteristics, potentially indicative of an exhausted or regulatory state (Fig. 2b).

Fig. 2.

Transcriptional and phenotypic reprogramming of human T cells in response to TNBC-derived exosomes. a UMAP plot of IFNG expression across the T cell landscape. b UMAP plot illustrating 14 transcriptionally distinct T cell clusters defined by combined gene and protein expression signatures. c Line plot depicting cell count distributions per cluster across exosome-treated samples. Incucyte live-cell imaging of TNBC cell killing by control and exosome-treated T cells. Activated primary human T cells were left untreated or pre-incubated with TNBC-derived exosomes (Exo-T cells) for 48 h and then co-cultured with d HCC1395 or e HCC70 target cells at an effector-to-target ratio of 1:5. Apoptotic tumor cells were quantified every 2 h for 48 h as caspase-3/7-positive objects using the Incucyte system. Data represent the mean ± SEM from three independent replicates. Statistical significance was determined by a two-tailed Mann–Whitney U test (****p < 0.0001; **p < 0.01). f–i Flow cytometry analysis of Treg (CD4⁺CD25⁺CD127^−/low), Tfh (CD4⁺CXCR5⁺), TEX (CD8⁺PD⁻1⁺CD127⁻), γδ T cells (CD4⁻CD8⁻CD56⁻TCR γδ⁺), NKT (CD4⁻CD8⁻TCR γ^δ-CD56⁺). Antibodies included: PE anti-human CD4, APC anti-human CD8α, PE/Dazzle™ 594 anti-human CD127, PerCP/Cyanine5.5 anti-human TCR γ/δ, Brilliant Violet 421™ anti-human CD56, APC/Cyanine7 anti-human CD25, FITC anti-human CXCR5, and FITC anti-human PD-1. Data represent the mean ± SEM from three independent biological replicates. Statistical significance was determined by one-way ANOVA followed by Dunnett’s multiple comparisons test (****p < 0.0001; ***p < 0.001; **p < 0.01; *p < 0.05)

The overall composition of T-cell subsets was largely conserved between treated and untreated samples (Supplementary Fig. 4a). Similarly, the distribution of tumor cell subtypes of origin within treated T cells mirrored that observed in untreated mock controls (Supplementary Fig. 4b). Analysis of T-cell subset abundances across TNBC-treated and untreated samples revealed that BT549 displayed the highest overall T-cell counts (Supplementary Fig. 4c), particularly within the Th17.Treg and Tfh(IFNG+) populations (Supplementary Fig. 4d). Furthermore, basal-like TNBC samples exhibited consistently higher T-cell numbers than other tumor cell–of-origin subtypes across multiple T-cell categories (Supplementary Fig. 4e). These findings indicate that exosome exposure, while capable of eliciting transcriptional alterations, does not markedly perturb the proportional representation of major immune subsets.

By contrast, the distribution of DEGs varied substantially across conditions (Supplementary Fig. 4f), underscoring that the principal impact of exosome exposure is the transcriptional reprogramming of pre-existing cell types rather than broad shifts in cellular composition. This emphasizes the necessity of single-cell transcriptomic resolution to reveal nuanced but functionally consequential changes.

Quantitative comparisons revealed that the Th17.Treg and Tfh (IFNG⁺, CCR4⁺) clusters comprised the highest number of cells, whereas the NKT cluster contained the fewest (Fig. 2c). Within the γδ T cell population, two subclusters were identified, γδ T and γδ T.CD8⁺, with the latter predominating. The relative proportions of CD4⁺ and CD8⁺ T cell subclusters were broadly consistent across samples. However, samples from HCC1395 and MB231 displayed unique distributions of T cell cluster abundance, highlighting inter-sample heterogeneity (Fig. 2c).

To determine whether TNBC-derived exosomes alter the cytotoxic capacity of T cells, we performed Incucyte live-cell imaging of tumor cell apoptosis in co-cultures of primary human T cells and HCC1395 or HCC70 cells, with or without prior exposure of T cells to TNBC exosomes (Exo–T). In both TNBC models, T cells alone induced a progressive increase in caspase-3/7-positive tumor cells over 48 h, whereas Exo-T cells showed a consistently blunted apoptotic response. For HCC1395 targets, caspase-3/7 object counts in the Exo-T condition remained markedly lower than in control T cells throughout the assay (Fig. 2d). A similar, though somewhat less pronounced, impairment of killing was observed against HCC70 cells, where Exo-T cells again generated significantly fewer apoptotic events than control T cells (Fig. 2e). Representative Incucyte images at 48 h illustrate the decreased density of caspase-3/7-positive (green) tumor cells in wells containing Exo-T cells compared with wells containing untreated T cells (Supplementary Fig. 4g).

Flow cytometric validation of CITE-seq predictions showed that TNBC-derived exosomes remodel the T cell compartment (Supplementary Fig. 5a). Across multiple cell-line-derived exosome conditions, we observed a significant expansion of CD4⁺ Treg (CD25⁺CD127^−/low) frequencies relative to untreated controls (Fig. 2f). Tfh cells also increased, albeit heterogeneously across lines, and were most pronounced outside the basal-like subtype (Fig. 2g). Within the CD8⁺ compartment, exosomes consistently elevated TEX (PD-1⁺CD127⁻) cells (Fig. 2h), aligning with the exhaustion signatures seen by CITE-seq. Concomitantly, CD56⁺ NKT-like frequencies declined under all exosome conditions (Supplementary Fig. 5b). In contrast, TCR γδ⁺ cells were selectively enriched, most notably in basal-like and mesenchymal contexts, mirroring the γδ T-cell cluster expansion in the UMAP projections (Fig. 2i). We also observed that the response of T cells, particularly within the Treg and TEX populations, was concentration dependent (5–160 µg), with higher exosome concentrations eliciting progressively stronger effects that eventually plateaued (Supplementary Fig. 5c). Collectively, these orthogonal datasets indicate that TNBC exosomes promote immunoregulatory skewing (Treg and Tfh gains), augment CD8⁺ TEX, suppress NKT-like representation, and expand γδ T cells, providing a mechanistic bridge between the single-cell transcriptional shifts and functional immunophenotypes.

T cell subset-specific transcriptomic responses to TNBC exosomes reveal functional heterogeneity

The pattern of DEGs exhibited substantial divergence from the distribution of cell numbers across clusters and samples. The highest DEG counts were observed in the MB453 sample within the Tfh (IFNG+, CCR4+) cluster and in the HCC1143 sample within the NKT cluster. Notably, no DEGs were detected in the Tfh.Th1 (IL4+, IL17+, IL7R−, AHR+) or γδ T cells (IFNG+, LAG3+, TBX21+) clusters (Fig. 3a).

Fig. 3.

Transcriptional reprogramming and pathway enrichment across T cell subsets following TNBC exosome exposure. a Number of differentially expressed genes (DEGs) per sample within each detailed cluster, relative to the Mock sample; b Heatmap displaying the log₂ fold change of genes dysregulated in at least four samples within the same T cell type, compared to the Mock sample; c Enriched biological pathways identified in CD4/CD8 transitional clusters, based on DEG profiles; d Heatmap of log₂ fold changes for DEGs contributing to enriched pathways in CD4/CD8 cluster group samples; e Dot plot illustrating significantly enriched pathways in NKT and γδ T cell populations; f Heatmap of functional DEGs associated with enriched pathways in NKT and γδ T cell populations

A detailed compilation of all DEGs identified for each sample within each cluster is presented in Table S1. Genes dysregulated in at least four samples per cluster are depicted in a heatmap (Fig. 3b), with cluster-specific annotations indicated by numeric suffixes. These numeric designations, ranging from 1 to 9, correspond to the following T cell subtypes: (1) CD4.central memory (IFNG−, BCL6+), (2) CD4.Naive (IFNG−), (3) CD8.Naive (IFNG−), (4) CD8.Naive (IFNG−, TRGV10+, CD127−), (5) NKT (TNF+, PD-L1+, SPI1+, CXCR5+), (6) Tfh (IFNG+, CCR4+), (7) Th17.Treg (IFNG+, TNF+, IL22+, IL4+, LAMP-1+), (8) Treg (IFNG−, CXCR5+), (9) γδ T cells.CD8+ (IFNG+, LAG3+, TBX21+). The heatmap clustering of the 17 T cell samples treated with TNBC-derived exosomes revealed two distinct gene expression profiles: eight samples grouped in the left columns, including BT20, HCC70, MB436, HCC1395, BT549, HCC1599, HCC1187, MB453, and nine in the right columns encompassing HCC1937, HCC1143, HCC38, HCC1806, DU4475, Hs578T, MB231, MB157, and MB468 (Fig. 3b).

Among all subsets, NKT cells displayed the strongest transcriptional activation, marked by coordinated upregulation of immune regulatory genes, signaling molecules, and histone-associated transcripts. This profile suggests enhanced chromatin accessibility, elevated immune activation, and increased vesicular and metabolic activity, collectively supporting the heightened responsiveness of this population. NKT cells from the HCC1143 sample contributed most prominently to these changes. In contrast, CD4⁺ central memory T cells exhibited broad downregulation of genes involved in metabolic homeostasis, receptor signaling, vesicular transport, and cell-cycle control. These alterations indicate reduced activation potential under exosome exposure. However, simultaneous suppression of inhibitory regulators such as CISH and FOS suggests partial release of negative feedback constraints, which may facilitate compensatory activation. A similar pattern was observed in CD8⁺ naïve T cells, where gene downregulation appears to lift inhibitory signaling and may enhance proliferative competence. In a distinct CD8⁺ naïve subset lacking IFNG and CD127 but expressing TRGV10, downregulation of MIR155HG indicates reduced miR-155-mediated activation, suggesting impaired recruitment and cytokine responsiveness (Fig. 3b).

Within the Tfh(IFNG⁺) compartment, exosome exposure altered key signaling and metabolic pathways. Reduced expression of CCL3, LDLR, and RSAD2 points to diminished chemotactic capacity, disrupted lipid metabolism, and weakened antiviral defense, whereas upregulation of RGL4 reflects compensatory proliferative or signaling changes. Th17.Treg cells underwent notable regulatory reprogramming. Upregulation of ABCA1 suggests enhanced cholesterol efflux, while coordinated downregulation of inflammatory mediators and prosurvival genes indicates attenuated effector capacity and a shift toward a more quiescent or regulatory state. Classical Tregs demonstrated increased CXCR5 expression, consistent with enhanced follicular homing, together with broad suppression of effector-associated transcripts, reinforcing a tolerogenic phenotype. Finally, γδ CD8⁺ T cells showed decreased expression of cytoskeletal and interferon-induced genes, suggesting reduced motility and antiviral competence (Fig. 3b).

Among these DEGs, HCC70 and MB157 exhibited the highest overlap, with 49 common DEGs each, indicating significant contributions to the observed profiles. Conversely, HCC1187 displayed the least impact, with only five shared DEGs.

In conclusion, exposure to tumor-derived exosomes elicits profound immunomodulatory effects relative to untreated mock controls. The downregulation of key mediators such as CCL3, FUT7, and OSM attenuates T-cell trafficking and cytokine signaling, thereby compromising anti-tumor inflammatory responses. Concomitant suppression of transcriptional feedback regulators, including DUSP1, EGR1, and FOS, is indicative of sustained yet dysregulated T-cell activation. In parallel, reduced expression of co-stimulatory and inhibitory molecules such as CTLA4, HBEGF, and TNF highlights a multifaceted perturbation of immune synapse integrity. The upregulation of RTN4 further implicates endoplasmic reticulum stress pathways in impairing T-cell homeostasis and survival. Collectively, these transcriptional alterations delineate an immune-evasive phenotype that facilitates tumor progression.

Immunoregulatory and hormone-linked pathways define T cell subset responses to TNBC exosomes

To investigate the functional significance of DEGs within T cell subpopulations, we conducted enrichment analyses. The identified DEGs were systematically mapped to the MSigDB Hallmark database, enabling the characterization of enriched biological pathways. This analysis delineated both shared functional networks and subset-specific specializations across T cell populations, providing deeper insights into their molecular and immunological roles. Notably, significant pathway enrichment (adjusted p value < 0.05) was observed across three major cluster groups: CD4/CD8 clusters, Th/Treg clusters, and NKT/γδ clusters, with functionally convergent pathways enriched within each group.

Within the CD4/CD8 cluster group, the most prominent enriched pathways included ‘Interferon Gamma Response’, ‘TNF-α Signaling via NF-κB’, ‘Estrogen Response Late’, and ‘Estrogen Response Early.’ Among these, ‘Cholesterol Homeostasis’ exhibited the strongest enrichment in the MB157 sample, demonstrating the lowest adjusted p value across CD4/CD8.Naive (IFNG−) clusters (Fig. 3c).

To identify the most functionally impactful genes within the enriched pathways across cell types, we selected DEGs present in all pathways enriched in more than 10 sample-cluster combinations (Fig. 3d). Among the 35 identified genes, FOS was the only gene dysregulated across all four clusters, exhibiting upregulation in the CD4.central memory (IFNG−, BCL6+) subset while being downregulated in the remaining clusters.

Another upregulated gene, DUSP1, is upregulated in CD4.central memory (IFNG−, BCL6+) while downregulated in CD4.Naive (IFNG−) and CD8.Naive (IFNG−). The remaining influential genes were predominantly downregulated. Additionally, OSM, SOCS3, CD69 (a well-established early T-cell activation marker), and CISH were among the most frequently dysregulated DEGs, exhibiting downregulation across three CD4/CD8 clusters (Fig. 3d).

In the NKT-γδ cluster group, a distinctly different functional profile was observed, characterized by the enrichment of unique biological pathways. In the γδ T cells.CD8+ (IFNG+, LAG3+, TBX21+) cluster, the most significantly enriched pathways included ‘Androgen Response,’ ‘TNF-α Signaling via NF-κB,’ ‘p53 Pathway,’ and ‘Apical Junction.’ Conversely, in the NKT (TNF+, PD-L1+, SPI1+, CXCR5+) cluster, ‘IL-2/STAT5 Signaling’ was the most frequently enriched pathway across three samples (Fig. 3e).

Interestingly, ‘Cholesterol Homeostasis’ emerged as the most statistically enriched pathway in the MB157 sample; however, this enrichment was exclusively observed within the γδ T cells.CD8+ (IFNG+, LAG3+, TBX21+) cluster. Additionally, ‘Interferon Gamma Response,’ ‘Interferon Alpha Response,’ and ‘Fatty Acid Metabolism’ were distinctively enriched across both clusters. The pathways ‘Estrogen Response Late’ and ‘Estrogen Response Early’ were not only consistently enriched, as observed in the previous cluster groups, but also contributed to the unique functional signature of the γδ T cells.CD8+ (IFNG+, LAG3+, TBX21+) cluster (Fig. 3e).

Moreover, a total of 24 DEGs were identified within pathways enriched in more than five sample-cluster combinations across both γδ T cells and NKT cells. The color scale in the heatmap (Fig. 3f) reflects the mean log₂(fold-change) values calculated across all samples within each cell type. The majority of these genes were downregulated; however, seven genes, ANXA4, CD83, CTSZ, GLIPR2, IL15, RIPK2, and SYNGR2, were upregulated within the NKT (TNF+, PD-L1+, SPI1+, CXCR5+) cluster. Notably, among the 24 identified DEGs, no genes were shared between the two clusters, underscoring their functionally distinct transcriptional landscapes.

Pathway enrichment analysis in the Th/Treg compartment revealed several consistently overrepresented pathways, including TNF-α signaling via NF-κB, p53 signaling, inflammatory response, and IL-2/STAT5 signaling. Although these pathways appeared broadly across samples, marked variability in sample contribution was observed. Notably, the MB157 sample showed strong enrichment of cholesterol homeostasis, and all 17 samples contributed to pathway activation within the Treg (IFNG−, CXCR5+) cluster. In contrast, only the HCC70 sample showed significant pathway involvement within the Tfh/Treg (IFNG−, CCR10+, GATA3+) subset (Fig. 4a).

Fig. 4.

Pathway enrichment analysis and exosome-mediated regulatory networks in Treg cell subsets. a Dot plot showing significantly enriched pathways in T helper (Th) and regulatory T cell (Treg) populations following exosomal exposure; b heatmap of functionally relevant differentially expressed genes (DEGs) contributing to enriched pathways within the Th-Treg T cell populations; c the alluvial plot visualizes the relationships between downregulated DEGs, specific T-cell populations (Treg, Th17.Treg, and Tfh.Treg), and the significantly enriched MSigDB Hallmark pathways across basal-like, mesenchymal, unclassified, and LAR TNBC subtypes

Across Th/Treg clusters, 120 genes were recurrently differentially expressed in pathways enriched in more than ten sample–cluster combinations. Most were downregulated, whereas a small subset, including ABCA1, CD79B, CLIC3, and MT1E, showed upregulation. FOSB demonstrated cluster-specific divergence, with upregulation in the Tfh/Th1 subset but significant suppression in multiple Treg and Th17/Treg clusters. Additional frequently downregulated genes included TNF, HBEGF, HMGCS1, IER3, IFIT3, and EGR1, indicating a predominant transcriptional repression signature across Th/Treg populations (Fig. 4b).

Pathway enrichment analysis underscored key biological processes across clusters, with pathways such as ‘Cholesterol Homeostasis’ and ‘TNF-α Signaling via NF-κB’ prominently featured. Furthermore, hormone-related pathways were enriched in most of the clusters, reflecting their potential role in modulating T cell function within the tumor microenvironment and highlighting the influence of endocrine signaling on immune responses in TNBC. These findings advance our understanding of T cell functionality, shedding light on the molecular underpinnings of their behavior and potential roles in immune modulation.

Suppressive role of Treg cells in the TNBC exosome-conditioned T-cell compartment

To test whether TNBC-derived exosomes skew T cells toward immunoregulatory states, we examined Treg-associated transcriptional programs and validated phenotypes by flow cytometry. TNBC-Exo exposure increased the frequency of CD4⁺CD25⁺CD127^−/low Treg cells (Fig. 2f) and enriched Treg/Th17.Treg clusters by CITE-seq, indicating subset expansion rather than broad compositional drift (Figs. 2a–e and S4).

At the transcript level, ITGA4 and TSPAN2 were upregulated in Treg cells, suggesting enhanced trafficking and tissue retention consistent with recruitment into tumor sites. ABCG1 was elevated in both Treg and Th17.T_reg clusters, pointing to shared cholesterol handling programs often linked to immunoregulatory function. Upregulation of LAIR1 and TOX/TOX2 genes associated with inhibitory signaling and chronic stimulation supports a shift toward suppressive/exhaustion-related states within the Treg axis.

We next assessed downregulated DEGs across TNBC subtypes (basal-like, mesenchymal, LAR and unclassified) in Treg, Th17.Treg and Tfh.Treg populations (Fig. 4c). Processes reduced across these cells included lymphocyte/T-cell proliferation, cytokine production, and cell–cell adhesion/migration pathways. Rather than implying loss of Treg identity, these coordinated decreases are most consistent with tonic dampening of activation and motility programs under exosomal conditioning, i.e., a transcriptional state compatible with sustained immune regulation in the tumor milieu, not effector activation. Concordantly, metabolic pathways important for T-cell activation (acetyl-CoA and cholesterol biosynthesis) were also reduced, suggesting reprogramming toward lipid efflux and lower anabolic tone, a known feature of regulatory and exhaustion-prone states. Decreased DNA-templated and miRNA transcription further indicates global transcriptional restraint.

Apoptosis-related annotations were also diminished (e.g., extrinsic apoptotic signaling). We interpret this cautiously: reduced apoptotic signatures may reflect enhanced Treg persistence or general signaling quiescence, both of which would favor accumulation of immunoregulatory cells in the tumor microenvironment. Taken together, these data indicate that TNBC-Exosomes promote Treg expansion and imprint an immunosuppressive, metabolically restrained transcriptional state that is well positioned to dampen anti-tumor immunity (Fig. 4c).

Non-coding RNA networks mediate coordinated downregulation of T cell immune genes in TNBC

To enhance our understanding of gene expression dynamics in response to distinct TNBC types in T cell samples, we expanded our analysis to integrate exosomal non-coding RNA (ncRNA) data extracted from 17 TNBC cell lines, with T cell coding RNA profiles.

This approach was motivated by the hypothesis that tumor-derived ncRNAs significantly influence T cell gene expression. To improve analytical precision, we incorporated data from 35 exosome samples extracted from human plasma, including 7 healthy controls and 28 patient samples.

Differentially expressed miRNAs (DEMs) were identified from annotated datasets in both the TNBC cell line exosome samples and the patient plasma exosome samples. We applied a stringent filtering strategy to ensure the biological relevance of our findings. Only miRNAs that were significantly up- or downregulated in both the cell line-derived exosomes (relative to primary breast cell line exosomes) and the patient plasma exosomes (relative to healthy controls) were retained for further analysis. This dual-validation approach increased our confidence that the observed in vitro expression patterns in the cell lines were consistent with the in vivo expression profiles in human patients. In parallel, we identified dysregulated mRNAs from T cell samples that had been previously exposed to tumor-derived exosomes. These DEGs were obtained by comparing gene expression in each cancer sample to a mock-treated control, independent of clustering. The integrated analysis of these dysregulated mRNAs with the validated DEMs allowed us to construct a comprehensive network of exosome-mediated regulation, providing deeper insight into the molecular mechanisms governing T cell-tumor interactions in TNBC.

The dysregulated mRNAs and miRNAs that exhibited concordant expression patterns within the same samples and had documented interactions in the miRTarBase database were subsequently utilized to construct an miRNA-mRNA interaction network.

A total of ten downregulated genes, CDC42EP1, FOS, HBEGF, HMGCS1, IER3, OLAH, PRSS23, PTGIS, SLC6A9, and TNFSF9, were consistently observed across multiple samples. Concurrently, 20 significantly upregulated miRNAs were identified within the same samples, with their regulatory interactions previously validated in established databases (Fig. 5a, d).

Fig. 5.

The miRNA expression and cytokine profiling reveal immunosuppressive remodeling of T cell function by TNBC-derived exosomes. a miRNA–mRNA interaction network depicting validated inhibitory relationships between upregulated miRNAs identified in TNBC cell line-derived exosomes and their corresponding downregulated mRNA targets in T cells exposed to TNBC exosomes, based on established interaction databases; flow cytometric analysis of CD4⁺CD25⁺CD127⁻ Tregs following transfection with hsa-miR-98-5p mimic or control showing b percentage of CD4⁺CD25⁺CD127⁻ Tregs among CD4⁺ T cells and c percentage of PD-1⁺ cells within the CD4⁺CD25⁺CD127⁻ Treg population. Data are presented as mean ± SEM from independent experiments. Statistical significance was determined using an unpaired two-tailed Student’s t test. *p < 0.05; **p < 0.01. d Heatmap showing significantly downregulated mRNA targets in T cells, associated with the regulatory miRNAs; e Dot plot summarizing the integrated Luminex-derived cytokine profiles across all cancer-exosome-treated samples; Bar plots showing differential expression levels of key cytokines; f TNFα, g IL-13, h IL-1β, i IL-17α, j IL-2R, and k IL-6 across T cell samples treated with TNBC-derived exosomes, as measured by Luminex, relative to the untreated Mock control. Data are presented as mean ± SEM from 3 replicates. Statistical significance was determined using an ordinary one-way ANOVA. **p < 0.01; ****p < 0.0001; l UMAP plots displaying single-cell cytokine secretion profiles for control and exosome-treated samples

Downregulation of multiple genes by miRNAs contained in TNBC-derived exosomes impairs T cell function by disrupting essential signaling pathways that govern activation, proliferation, and anti-tumor immune responses. These exosomes deliver an immunosuppressive cargo that reprograms recipient T cells within the tumor microenvironment, thereby facilitating immune escape.

For example, TNFSF9, a co-stimulatory ligand normally expressed by antigen-presenting cells, enhances T cell proliferation, survival, and IFN-γ production through receptor engagement. Its downregulation diminishes this critical co-stimulatory signal. Similarly, repression of FOS, a key component of the AP-1 transcription factor complex, interferes with TCR downstream signaling and disrupts NFAT and NF-κB activation, leading to impaired cytokine production.

Alterations in metabolic pathways further contribute to dysfunction. Suppression of HMGCS1 and PTGIS perturbs lipid metabolism, shifting T cells toward a metabolic profile associated with exhaustion. Downregulation of SLC6A9, a glycine transporter, compromises amino acid availability and energy production, while reduced IER3 expression impairs cytokine regulation and increases susceptibility to apoptosis.

Additional targets reinforce this immunosuppressive phenotype. HBEGF downregulation disrupts survival and proliferative signaling; repression of PRSS23 interferes with cytokine processing or T cell mobility; and inhibition of CDC42EP1, a regulator of CDC42-dependent cytoskeletal remodeling, compromises immune synapse formation and T cell migration, thereby limiting tumor cell engagement.

Network topology analysis identified hsa-miR-98-5p as a highly connected node, targeting several of the downregulated genes, including HBEGF, TNFSF9, and IER3 (Fig. 5a). Based on this central positioning, we further investigated the functional consequences of hsa-miR-98-5p modulation on T cell phenotype. Transfection of T cells with a hsa-miR-98-5p mimic resulted in a significant increase in the proportion of CD4⁺CD25⁺CD127⁻ Tregs compared to control-treated cells (Fig. 5b). In addition, a higher frequency of these Tregs expressed PD-1 (CD4⁺CD25⁺CD127⁻PD-1⁺) (Fig. 5c). While PD-1 expression has been associated with enhanced suppressive capacity in certain contexts, it can also reflect activation or exhaustion depending on the microenvironment. Therefore, our data indicate an expansion of a PD-1⁺ Treg subset, consistent with a potentially more suppressive phenotype, but functional suppression assays would be required to confirm this directly.

At the molecular level, overexpression of hsa-miR-98-5p was associated with reduced expression of HBEGF and TNFSF9, consistent with their predicted targeting (Fig. 5a). We also observed increased phosphorylation of STAT3 (Y705) without marked changes in total STAT3 levels (Supplementary Fig. 6a), suggesting activation of STAT3 signaling under these conditions. However, based on the current data, we cannot definitively conclude that STAT3 activation is directly mediated through suppression of HBEGF and/or TNFSF9. Rather, these findings demonstrate a correlation between hsa-miR-98-5p overexpression, modulation of its predicted targets, and enhanced STAT3 phosphorylation.

Collectively, these results support a model in which hsa-miR-98-5p functions as a regulatory hub associated with coordinated gene downregulation and altered T cell phenotype, including expansion of a PD-1⁺ Treg population and increased STAT3 activation. Further mechanistic studies will be required to establish direct causal links between specific target suppression and STAT3-driven Treg modulation.

TNBC exosomes harbor consistently upregulated lncRNAs with potential immunomodulatory roles

Similar to the microRNA analysis, the expression profiles of long non-coding RNAs (lncRNAs) were assessed to identify potential regulators in the tumor microenvironment. This was achieved by utilizing data from both TNBC cell line-derived exosomes and a cohort of 35 human plasma samples, which included seven healthy controls and 28 patient samples.

For further analysis, we applied a stringent filtering criterion to identify biologically relevant lncRNAs. Only those lncRNAs that demonstrated significant upregulation in both datasets were selected. Specifically, we required upregulation in the TNBC cell line-derived exosomes (relative to healthy primary breast cell line exosomes) and a corresponding upregulation in the patient plasma exosomes (relative to healthy controls). This dual-validation approach ensured that the identified lncRNAs were not only present in our in vitro models but also exhibited a consistent expression pattern in TNBC patients.

However, interrogation of the LncSEAv2 database did not reveal any documented interactions between the identified lncRNAs and DEGs, which were defined by comparing each TNBC exosome sample to the mock condition, independent of clustering.

The identified lncRNAs represent novel transcripts whose interactions with other RNA species warrant further investigation. Among them, ENST00000651540, ENST00000574245, and ENST00000674020 demonstrated the most pronounced upregulation, being overexpressed in eight, six, and five samples, respectively, suggesting their potential biological relevance in the lncRNA dataset (Supplementary Fig. 6b).

Bulk and single-cell cytokine profiling uncovers functional impairment of T cells by TNBC exosomes

To gain deeper insights into functional alterations, we assessed cytokine expression using the bulk and single-cell cytokine analysis platforms. In the bulk Luminex approach, cytokine levels were quantified from supernatants collected 48 h post-exposure to TNBC-derived exosomes. Sixteen cytokines were measured, including IFNγ, IL-10, IL-12p70, IL-13, IL-15, IL-17α, IL-1β, IL-2, IL-2R, IL-4, IL-5, IL-6, IL-8, IL-9, MIF, and TNFα (Fig. 5e). Each cytokine was assessed in duplicate per sample, and statistical analyses were performed using nonparametric methods with Dunnett’s multiple comparisons test.

From this panel of cytokines, twelve exhibited differential expression across cancer samples relative to mock, based on adjusted p value < 0.05. TNFα, IL-13, and IL-1β varied significantly across all samples (Fig. 5f–h), and IL-17α, IL-2R, and IL-6 were significantly altered in all but BT549 (Fig. 5i–k). In contrast, IL-12p70, IL-4, IL-15, and MIF remained unchanged across samples (Supplementary Fig. 6c–f). Other cytokines displayed variable expression patterns, with MDA-MB-231 and MDA-MB-453 showing the most pronounced changes (11 cytokines), whereas BT549 exhibited the least variation, with significant alterations in only five cytokines (Fig. 5e).

To assess the impact of TNBC-derived exosomes on cytokine secretion at the single-cell level, we used the IsoCode single-cell cytokine analysis chips. This enabled high-dimensional cytokine profiling, capturing the functional secretion profiles of individual T cells 48–72 h after exposure to TNBC exosomes (Fig. 5l). This method provided a more granular and sensitive assessment, quantifying 30 cytokines and chemokines, including CCL11, GM-CSF, Granzyme B, IFNγ, IL-10, IL-12, IL-13, IL-15, IL-17A, IL-17F, IL-1β, IL-2, IL-21, IL-22, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IP-10, MCP-1, MCP-4, MIP-1α, MIP-1β, Perforin, RANTES, sCD137, sCD40L, TGF-β1, TNFα, and TNFβ. Consistent with the Luminex findings, cytokine expression was analyzed across cancer and control samples.

For both CD4⁺ and CD8⁺ T cells, the polyfunctional strength index (PSI) was decomposed into effector, stimulatory, chemotactic, regulatory, and inflammatory modules. In the CD8⁺ compartment, basal-like exosomes segregated into two patterns—some lines reduced PSI, whereas others increased PSI relative to the matched control. By contrast, exosomes from mesenchymal, LAR, and unclassified lines generally lowered CD8⁺ PSI, with BT549 as the notable exception (Fig. 6a). In the CD4⁺ compartment, PSI was predominantly suppressed across subtypes, with HCC70, HCC38, and MB231 showing the only clear increases (Supplementary Fig. 6g). For CD4⁺ T cells, most basal-like TNBC exosomes suppressed cytokine secretion relative to control (Fig. 6b). Decreases were prominent across type-1 effector mediators (IFN-γ, TNF-α/β), myeloid-activating (GM-CSF), chemotactic (IL-8, MIP-1α/β), and cytotoxic enzyme (Granzyme B) outputs. Two lines—MB468 and HCC1806—were clear exceptions, showing broad upregulation spanning growth/survival (IL-2, IL-15), helper-skewing (IL-4, IL-9), type-1 effector (IFN-γ, TNF-α/β), myeloid-activating (GM-CSF), chemotactic (IL-8, MIP-1α/β), the costimulatory ligand (sCD40L), and cytotoxic enzyme (Granzyme B), indicating a potent CD4⁺ activation signature despite basal-like classification.

Fig. 6.

Single-cell cytokine profiling and Polyfunctional Strength Index analysis of T cells following exosomal treatment. a Stacked bar plot representing Polyfunctional Strength Index (PSI) for CD8⁺ T cells across five cytokine classes in four representative sample groups; b Heatmap displaying the log₂ fold changes of dysregulated cytokines in each experimental condition compared with Mock sample and TNBC subtypes detected by single cell cytokine profiling within CD4⁺ and CD8⁺ T cells

Mesenchymal exosomes enhanced CD4⁺ responses with coordinated increases across myeloid-activating (GM-CSF), cytotoxic enzyme (Granzyme B), type-1 effector (IFN-γ, TNF-α/β), growth/survival (IL-2), helper-skewing (IL-4, IL-9), Th17-associated (IL-17A), regulatory (IL-10), and chemotactic (IL-8, MIP-1α/β) categories. Unclassified exosomes (BT20, HCC1395) similarly induced broad upregulation—most strongly in BT20—across myeloid-activating, cytotoxic enzyme, type-1 effector, growth/survival, Th17-associated, and chemotactic mediators. The LAR line MB453 displayed a dominant effector profile with marked increases in Granzyme B, IFN-γ, IL-2, IL-9, MIP-1α/β, TGF-β1 (regulatory), and TNF-α, consistent with a highly activated helper-like CD4⁺ response.

For CD8⁺ T cells, most basal-like TNBC derived exosomes suppressed CD8⁺ cytokine/cytotoxic outputs relative to control (Fig. 6b). Decreases were evident across type-1 effector and chemotactic/cytotoxic mediators, including TNF-α, MIP-1α/β, and Granzyme B. Notably, MB468 and HCC1806 were clear exceptions, showing strong upregulation across multiple effector and stimulatory mediators, including GM-CSF, Granzyme B, IFN-γ, IL-2, IL-22, IL-5, IL-8, IL-9, MIP-1α/β, TNF-α/β indicating a potent CD8⁺ activation signature despite basal-like classification.

Mesenchymal exosomes largely enhanced CD8⁺ programs, characterized by coordinated increases in GM-CSF, Granzyme B, IFN-γ, IL-2, IL-22, IL-4, IL-8, IL-9, MIP-1α, MIP-1β, Perforin, TNF-α, and TNF-β.

Unclassified exosomes (BT20, HCC1395) induced a broad, consistent upregulation of CD8⁺ mediators, most prominently GM-CSF, Granzyme B, IFN-γ, IL-17A, IL-8, MIP-1α, and TNF-α, reflecting a pro-effector, recruitment-competent profile.

The LAR line MB453 exhibited a mixed effector signature, with increased Granzyme B, IL-22, and IL-9 alongside reduced IL-2 and Perforin.

Single-cell TCR sequencing identifies tumor-specific clonotypes following exosome-mediated T cell stimulation

All clonotype data, including VDJ chain compositions, were analyzed across both Gamma-Delta (γδ) and Alpha-Beta (αβ) T cell populations. The distribution of γδ TCR subtypes, classified based on Delta (δ) and Gamma (γ) chain associations, is depicted in Fig. 7a. This dataset encompasses V (variable), J (joining), and C (constant) gene segments, with the left panel representing all samples, including mock (control) samples, and the right panel displaying only tumor-derived samples for comparison. Each dot is color-coded according to its clonotype ID, illustrating the genetic diversity within distinct TCR repertoires. While we acknowledge that differences in cell counts between samples could visually influence the dot distribution, our focus remains on the identification of specific γδ T-cell receptor subtypes and their recurrence across conditions.

Fig. 7.

Clonotypic landscape of γδ T cell repertoires reveals TNBC exosome-induced skewing and tumor-associated TCR signatures. a Distribution of γδ T cell receptor (TCR) subtypes based on the combinatorial pairing of delta (δ) and gamma (γ) chain gene segments, highlighting subtype diversity across samples; b Dot plot showing complementarity-determining region 3 (CDR3) sequences uniquely detected in γδ T cells from tumor-exosome-treated samples but absent in Mock controls, indicating tumor-associated clonotype emergence; c Frequency distribution of V-J-C gene combinations corresponding to the three most dominant γδ TCR clonotype identifiers, illustrating usage patterns in response to exosomal stimulation

The complementarity-determining region 3 (CDR3), critical for antigen recognition, is the most informative segment for assessing T-cell receptor (TCR) functionality. In our γδ TCR clonotype dataset, we identified six CDR3 sequences present across all tumor-derived samples but absent in mock samples (Fig. 7b). All of these unique CDR3 sequences were presented by “γδ T cells.CD8⁺” cells. Notably, DU4475 was the only sample harboring all six identified sequences. These sequences were associated with distinct V-J-C gene combinations, suggesting their selective expansion within the tumor microenvironment. Furthermore, the distribution of V-J-C gene combinations was displayed through the three most dominant γδ clonotype IDs, which were detected across all 10 tumor samples (Fig. 7c).

A comprehensive analysis of clonotype IDs uniquely detected in TNBC exosome-treated samples revealed γδ T cells (green dots) expansion in 10 cancer samples, whereas αβ T cells (red dots) were expanded in nine. Among αβ T cells, 743 unique clonotype IDs were identified as specific to TNBC exosome samples, of which the 200 most frequently occurring clonotypes are reported (Supplementary Fig. 7).

We further analyzed clonotype distributions within αβ TCR repertoires, where V, J, and C gene subtypes were detected at a much higher frequency (Supplementary Fig. 8).

Additionally, within the αβ TCR dataset, we identified 32 unique CDR3 sequences that were undetectable in Mock samples but observed in TNBC exosome-treated samples (Table 1 and Supplementary Fig. 9). These sequences were distributed across distinct V-J-C gene combinations, further supporting the notion that TNBC exosome-specific TCR clonotypes undergo selective expansion.

Table 1.

List of 32 unique CDR3 sequences related to αβ TCR

Symbol	CDR3 sequence	# of Samples	Symbol	CDR3 sequence	# of Samples
C1	CASSQDLTVSGANVLTF	54	C17	CASSLGDSNQPQHF	6
C2	CSAISGAGGRGNEQFF	25	C18	CASSLKGTQYF	6
C3	CATSRWGGETQYF	19	C19	CASSPGDEQFF	6
C4	CASSLAGGNNEQFF	10	C20	CASSQMGPRSTGELFF	6
C5	CATDAIRDTGRRALTF	9	C21	CATDGNNARLMF	6
C6	CALKGSGNTGKLIF	8	C22	CATPTGANSKLTF	6
C7	CALSGNNAGNMLTF	8	C23	CAVSGGYQKVTF	6
C8	CASSPRQGSYGYTF	8	C24	CIVRVAGGNQGGKLIF	6
C9	CAVPWGSYIPTF	8	C25	CAENSGGSNYKLTF	5
C10	CAVRDRGSTLGRLYF	8	C26	CASSLGQGRLGYGYTF	5
C11	CAVRPNTGFQKLVF	8	C27	CASSPNLGGDTGELFF	5
C12	CASSPWGDFTDTQYF	7	C28	CAVGGSQGNLIF	5
C13	CATEGDFGNEKLTF	7	C29	CAVNRGSTLGRLYF	5
C14	CAVGSGNTGKLIF	7	C30	CAVNSGGYQKVTF	5
C15	CAGESKAAGNKLTF	6	C31	CAVQGGSEKLVF	5
C16	CASSHRDNLYEQYF	6	C32	CVVNPSTGFQKLVF	5

Through the integration of scRNA-seq, non-coding RNA regulation data, cytokine expression patterns, and VDJ-seq, we provide a comprehensive synthesis of our findings. This analysis unifies shared and distinct mechanisms underlying exosome-mediated immunomodulation, specifically in T cells, offering deeper insights into their regulatory networks and functional implications in immune responses.

Discussion

Our multi-omic analysis reveals that TNBC-derived exosomes act as potent modulators of human T-cell function, inducing transcriptional, phenotypic, and secretory reprogramming without altering major subset frequencies. Across 17 molecularly distinct TNBC lines, exosomal exposure consistently reshaped pre-existing T-cell states, producing convergent immunosuppressive signatures with subtype-specific variation. Integration of CITE-seq, VDJ sequencing, and cytokine profiling established that TNBC exosomes remodel both CD4⁺ and CD8⁺ T-cell compartments through coordinated transcriptional and post-transcriptional mechanisms. These findings align with prior reports showing that tumor-derived exosomes reprogram immune effector cells and promote T-cell dysfunction through the transfer of immunoregulatory lipids, proteins, and RNAs.^15,16

Transcriptional analysis further revealed widespread modulation of genes involved in immune regulation, proliferation, and cellular stress. In particular, PD-L1, SPI1, MDM2, and ZEB2 were consistently upregulated in NKT and Treg clusters, implicating p53 inhibition, epithelial–mesenchymal transition, and immune suppression as coordinated exosome-driven programs. In contrast, FOS, DUSP1, CD69, SOCS3, and IL-21 were broadly downregulated across effector subsets, consistent with reduced T-cell activation, cytokine responsiveness, and MAPK signaling. Interestingly, an estrogen-response signature was detected in CD8⁺ γδ cells despite TNBC’s ER-negative status, suggesting either non-genomic hormonal signaling or exosome-mediated transfer of estrogen-related regulators. These results collectively support the concept that TNBC exosomes rewire immune transcriptional networks toward a regulatory and metabolically repressed phenotype through multiple parallel axes, including checkpoint upregulation, transcriptional reprogramming, and non-genomic hormonal modulation.

Kinetic killing assays demonstrated that exposure to TNBC-derived exosomes functionally impairs T-cell-mediated cytotoxicity, as reflected by reduced induction of caspase-3/7-dependent apoptosis in two independent TNBC cell lines (Fig. 2). The consistent decrease in apoptotic tumor cells across both HCC1395 and HCC70 models argues that the effect is not restricted to a single tumor context and supports a generalized exosome-driven suppression of T-cell effector function. Together with our phenotypic and transcriptional data on exosome-treated T cells, these findings suggest that TNBC exosomes do not merely alter T-cell markers but translate into a measurable loss of tumor-killing capacity, providing a functional link between exosome-mediated reprogramming and impaired anti-tumor immunity.

At the transcriptional level, TNBC exosomes promoted expansion of Treg and Tfh-like clusters and increased CD8⁺ TEX (PD-1⁺CD127⁻) frequencies, consistent with a regulatory and exhaustion-prone immune milieu (Fig. 2). Exosome-treated T cells displayed skewing toward immunosuppressive subsets, including Th17-like, Tregs, and Tfh cells expressing IFNG and CCR4. The expansion of these populations suggests early signs of immune exhaustion and regulatory feedback, particularly in the presence of checkpoint molecules such as PD-1 and PD-L1. While Th17 cells may play dual roles in cancer immunity,^17,18 Tregs are known to facilitate tumor progression. The appearance of Tfh-like cells co-expressing PD-1 and PD-L1 suggests an atypical feedback circuit with potential relevance to resistance mechanisms against PD-1/PD-L1 blockade.¹⁹

Upregulation of inhibitory and trafficking-associated genes (ITGA4, TSPAN2, LAIR1, TOX, TOX2) within Treg populations, together with elevated ABCG1, supports enhanced tissue retention and lipid efflux programs characteristic of suppressive phenotypes.²⁰ Conversely, coordinated downregulation of proliferative and metabolic pathways, including acetyl-CoA and cholesterol biosynthesis, indicates reduced anabolic activity. These features collectively define a metabolically restrained Treg state optimized for long-term immune suppression rather than effector activation, consistent with prior evidence that lipid efflux and TOX-driven signaling underlie chronic T-cell dysfunction.²¹

Functional cytokine profiling aligned with these transcriptional trends. Bulk and single-cell cytokine analyses revealed that most TNBC exosomes dampened aggregate cytokine production and polyfunctionality while selectively modulating effector pathways (Fig. 6). In the CD4⁺ compartment, exosomes from the majority of basal-like lines suppressed cytokine secretion, particularly type-1 effector (IFN-γ, TNF-α/β), myeloid-activating (GM-CSF), chemotactic (IL-8, MIP-1α/β), and cytotoxic (Granzyme B) outputs. Notably, MDA-MB-468 and HCC1806 deviated from this pattern, exhibiting broad upregulation across growth/survival (IL-2, IL-15), helper-skewing (IL-4, IL-9), effector (IFN-γ, TNF-α/β), and chemotactic mediators (IL-8, MIP-1α/β), suggesting subtype-specific preservation of helper functionality within basal-like contexts. These findings reinforce that TNBC exosomes can both suppress and license T-cell activation depending on tumor lineage, paralleling reports that tumor-intrinsic genomic programs shape the immunomodulatory composition of exosomal cargo.²²

Exosomes derived from mesenchymal TNBC lines enhanced both CD4⁺ and CD8⁺ effector programs, with coordinated increases in Granzyme B, Perforin, IFN-γ, IL-2, IL-22, IL-8, MIP-1α/β, and TNF-α/β, consistent with partial cytotoxic reactivation. Unclassified lines (BT20, HCC1395) similarly induced broad cytokine upregulation, particularly in GM-CSF, Granzyme B, IFN-γ, IL-17A, and TNF-α, reflecting a pro-effector, recruitment-competent state. The LAR line MB453 displayed a distinct mixed effector profile, marked by increased Granzyme B, IL-22, and IL-9, with concurrent reductions in IL-2 and Perforin, suggesting selective remodeling rather than global activation of cytotoxic programs (Fig. 6b). These subtype-specific patterns reveal that exosome-encoded cues can impose distinct immunological outcomes, ranging from exhaustion and suppression to partial effector rescue depending on the molecular subtype of TNBC, echoing subtype-resolved immunotranscriptomic studies.

Single-cell transcriptomic analysis also uncovered two transcriptionally distinct γδ T-cell subpopulations, predominantly CD8⁺, exhibiting differential enrichment of interferon-signaling, androgen-response, and IL-2/STAT5 pathways. This finding, together with retained expression of IL15 and CD83 in NKT-like clusters, points to a partially preserved cytotoxic capacity within these unconventional T-cell subsets. Although γδ T cells have been implicated in both tumor-promoting and protective roles,²³ our data indicate that exosomes from specific TNBC subtypes prime them toward an anti-tumor phenotype. Single-cell VDJ-seq extended these findings by revealing tumor-specific clonotypes within both αβ and γδ T-cell compartments. Several γδ CDR3 sequences were detected exclusively in exosome-treated samples but were absent in mock controls, indicating exosome-mediated antigenic stimulation. This suggests that tumor exosomes can promote clonal expansion of specific TCR repertoires, consistent with prior demonstrations that exosomes can transfer tumor antigens and MHC–peptide complexes to antigen-presenting cells or directly to T cells.²⁴ Such clonotype-specific induction underscores the potential of exosome-primed T cells as a platform for neoantigen discovery and TCR-guided immunotherapies.

Integration of miRNA and mRNA datasets uncovered post-transcriptional mechanisms underlying these immune alterations (Fig. 5). TNBC-derived exosomes carried a shared set of upregulated miRNAs that converged on the downregulation of key immune effectors and signaling regulators, including FOS, TNFSF9, HBEGF, HMGCS1, IER3, and CDC42EP1. These interactions collectively attenuate TCR signaling, co-stimulation, and metabolic support pathways critical for T-cell activation. Network topology analysis identified hsa-miR-98-5p as a highly connected node predicted to target multiple downregulated immune regulators, including HBEGF, TNFSF9, and IER3. Consistent with this, hsa-miR-98-5p overexpression in T cells increased the proportion of CD4⁺CD25⁺CD127⁻ Tregs and enriched a PD-1⁺ subset (Fig. 5b, c). This phenotypic shift was accompanied by reduced HBEGF and TNFSF9 expression and increased STAT3 phosphorylation without major changes in total STAT3 (Supplementary Fig. 6a), linking a central exosomal miRNA hub to coordinated target repression and Treg-associated signaling programs, although direct causality and suppressive function remain to be established. In contrast, several lncRNAs were consistently upregulated across TNBC exosome samples and patient plasma, although their direct regulatory interactions with immune genes remain to be characterized (Supplementary Fig. 6b). The recurrent upregulation of ENST00000651540 and ENST00000574245 suggests potential novel regulatory nodes, consistent with emerging evidence that exosomal lncRNAs contribute to immune evasion and metastasis.²⁵

Together, these data delineate an exosome-driven immune remodeling program in TNBC that integrates transcriptional, metabolic, and secretory reprogramming across T-cell subsets. Most exosomes promote immunosuppressive and exhaustion-like states through coordinated suppression of activation and co-stimulatory signaling, while select TNBC subtypes retain or amplify effector potential. The identification of distinct exosome-induced TCR clonotypes further suggests that tumor exosomes can serve as antigenic intermediates, shaping adaptive immunity in a genotype-dependent manner. This notion aligns with recent insights into exosome-based immune education, where tumor exosomes condition systemic immunity in ways that influence checkpoint response and metastasis.²⁶

Collectively, our study defines the functional and molecular heterogeneity of TNBC-derived exosome–T cell interactions, highlighting both suppressive and stimulatory axes of immune regulation. These findings provide a mechanistic foundation for exploring exosome-based biomarkers and for developing immunotherapeutic strategies that account for tumor subtype, exosomal cargo, and immune subset plasticity in triple-negative breast cancer. Future in vivo studies using murine TNBC models to test the effects of systemic or localized overexposure to tumor-derived exosomes will further validate these findings and clarify the mechanistic basis of exosome-mediated T-cell modulation within the tumor microenvironment.

Materials and methods

Exosome isolation and characterization

Triple-Negative Breast Cancer Panel 3 cell lines (TCP-1003) were obtained from the American Type Culture Collection (ATCC) and cultured in RPMI-1640 media supplemented with 10% exosome-depleted fetal bovine serum (Gibco) and 1% penicillin–streptomycin. Cells were passaged using TrypLE Express enzyme (Gibco™) upon reaching 70-80% confluency. The conditioned media were collected for exosome isolation. Exosome isolation was performed as previously described.^27,28 The media was first centrifuged at 1500 × g for 10 min to remove cells and large particles. It was then concentrated using Pierce™ Protein Concentrators PES, 100 K MWCO (Thermo Scientific). Specifically, 1 L of media was concentrated down to 5 mL and applied to size exclusion chromatography columns (SEC) (IZON Science). An automatic fraction collector was used to gather the exosome-rich fractions, which were then combined. Finally, the exosomes were filtered through a 0.2 µm filter (MilliporeSigma) and characterized for size and concentration using the NanoSight NS300 Nanoparticle Tracking Analyzer (Malvern Panalytical). The exosomes were aliquoted and stored at −80 °C.

Healthy human peripheral blood plasma (10 mL) from multiple donors was obtained from Stem Cell Technologies and centrifuged at 1500 × g for 10 min to remove large particles. Plasma was not concentrated and was applied directly to SEC columns for exosome isolation. Peripheral blood plasma samples from patients with TNBC were obtained through the Indiana University Simon Comprehensive Cancer Center (IUSCCC) Biospecimen Collection and Banking Core (BC²), under an Institutional Review Board (IRB)-approved protocol (12515). All specimens were de-identified in compliance with ethical guidelines, and no patient identifiers were provided. Samples were processed as mentioned before.

Transmission electron microscopy (TEM)

The size and morphology of isolated exosomes were assessed by transmission electron microscopy (TEM) using an FEI Tecnai Spirit electron microscope. Exosome samples were first negatively stained on carbon-coated TEM grids. Grids were glow-discharged for 30 s using a PELCO easiGlow™ Glow Discharge System (Ted Pella Inc.) to enhance sample adherence. A 2 μL aliquot of exosome suspension was applied to the grid and allowed to adsorb for 30 s before being gently blotted. The grid was then washed and stained with 0.75% uranyl formate (Polysciences) for contrast. After staining, grids were air-dried overnight in a desiccator. Imaging was performed using the FEI Tecnai Spirit TEM, and exosome size and morphology were analyzed from representative micrographs captured under standard operating conditions.

T cell isolation and culture

Peripheral blood mononuclear cells (PBMCs) were isolated from at least 10 healthy donor leukopaks by Lymphoprep (STEMCELL Technologies) density gradient centrifugation. CD3⁺ T cells were then purified using the EasySep™ Human T Cell Negative Isolation Kit (STEMCELL Technologies) according to the manufacturer’s instructions. Cells were resuspended in RPMI 1640 supplemented with 10% FBS and 10 ng/mL recombinant human IL-2 (STEMCELL Technologies). T cell activation was performed using ImmunoCult™ Human CD3/CD28 T Cell Activator (STEMCELL Technologies) at the recommended concentration for 72 h. Following activation, T cells were washed and treated with tumor-derived exosomes isolated from 17 distinct TNBC cell lines. Exosome exposure was carried out for 48 h before cells were harvested for downstream analyses.

Single-cell imaging flow cytometry

Exosomes were fluorescently labeled using AZDye™ 488 Maleimide (Vector Laboratories). Briefly, exosomes were concentrated to a final volume of 50 µL (1 mg/mL total protein concentration) and incubated overnight at 4 °C with the dye at a 10× molar excess. Post-labeling, unbound dye was removed by three sequential washes using 40 K molecular weight cut-off (MWCO) Zeba™ spin desalting columns (Thermo Fisher Scientific).

Purified CD3⁺ T cells were incubated with Alexa Fluor 488-labeled exosomes (50 μg/mL final concentration) for 3 h at 37 °C, followed by washes with FACS buffer (PBS + 2% FBS). Surface staining was performed using the following antibodies: PE anti-human CD4 (clone OKT4), APC anti-human CD8α (clone RPA-T8), PE/Dazzle™ 594 anti-human CD127 (clone A019D5), and APC/Cyanine7 anti-human CD25 (clone BC96) (all from BioLegend). Nuclei were counterstained with Hoechst 33342 (Thermo Fisher Scientific).

Cells were analyzed on a Cytek® Amnis® ImageStream®X Mk II imaging flow cytometer (Cytek Biosciences) at ×40 magnification, enabling high-resolution acquisition of multi-channel single-cell images. Image analysis was conducted using IDEAS® software (Cytek Biosciences). The analysis pipeline included compensation, internalization quantification, and gating strategies to exclude doublets and debris, ensuring accurate single-cell measurements. For internalization, a cell mask was generated from the brightfield image to define cell boundaries, and an internal mask was created by eroding the cell mask to exclude the membrane region. The Internalization Score was calculated as the ratio of fluorescence intensity inside the internal mask to the total cellular fluorescence and was used to quantify uptake. A higher score indicates greater internalization. Thresholds were established using control samples to distinguish internalized versus surface-bound signal.

Flow cytometry analysis

Purified CD3⁺ T cells were incubated with Alexa Fluor 488-labeled exosomes (50 μg/mL final concentration) for 48 h at 37 °C, followed by washes with FACS buffer (PBS + 2% FBS). Surface staining was performed using the following antibodies: PE anti-human CD4 (clone OKT4), APC anti-human CD8α (clone RPA-T8), PE/Dazzle™ 594 anti-human CD127 (clone A019D5), PerCP/Cyanine5.5 anti-human TCR γ/δ (Clone B1), Brilliant Violet 421™ anti-human CD56 (Clone HCD56), APC/Cyanine7 anti-human CD25 (clone BC96), FITC anti-human CXCR5 (clone J252D4), FITC anti-human PD-1 (clone EH12.2H7), PE Mouse IgG2b κ Isotype Ctrl (clone MPC-11), and FITC Mouse IgG1, κ Isotype Ctrl (clone MOPC-21) (all from BioLegend). Stained cells were acquired on a BD LSRFortessa (BD Biosciences) and analyzed using FlowJo software version 10.10.

IncuCyte live-cell caspase-3/7 apoptosis assay

Tumor-cell killing by T cells was quantified using the Incucyte® Caspase-3/7 Apoptosis Assay (Sartorius) according to the manufacturer’s instructions. HCC1395 and HCC70 TNBC cells were seeded in 96-well flat-bottom tissue-culture plates at 1000 cells per well in 100 µL of RPMI media and allowed to adhere overnight. Primary human T cells were isolated from PBMCs, activated as described above, and either left untreated (T cells) or pre-incubated with TNBC-derived exosomes (Exo-T cells). On the day of the assay, culture medium on tumor cells was replaced with medium containing Incucyte® Caspase-3/7 Green Apoptosis Reagent (final concentration 5 µM), and T cells were added at an effector-to-target (E:T) ratio of 1:5 in a final volume of 150 µL per well. Plates were placed into the Incucyte® live-cell imaging system (37 °C, 5% CO₂), and phase-contrast and green fluorescence images were acquired every 2 h for up to 48 h using a 10× objective. Caspase-3/7-positive apoptotic cells were quantified using the Incucyte® analysis software with a predefined fluorescence intensity and object size threshold, and data were expressed as caspase-3/7-positive object counts over time.

Single-cell RNA sequencing

Human CD3⁺ T cells were isolated from the peripheral blood of a healthy donor and cultured for 72 h in the presence of ImmunoCult™ Human CD3/CD28 T Cell Activator, followed by treatment with tumor-derived exosomes from 17 TNBC cell lines for 48 h. After incubation, cells were harvested, washed, and stained with the TotalSeq™-C Human Universal Cocktail, v2.0 (BioLegend), which includes a panel of oligo-conjugated antibodies targeting key T cell surface markers. To enable multiplexing of biological replicates and experimental conditions, cells were also stained with TotalSeq™-C anti-human Hashtag antibodies (BioLegend), following the manufacturer’s recommended titration and staining protocol.

Following staining, cells were counted and loaded onto the Chromium X system (10x Genomics) using the Chromium Single Cell 5′ High Throughput (HT) Immune Profiling Kit. Single-cell Gel Bead-In-Emulsions (GEMs) were generated, and barcoded cDNA libraries, including gene expression and antibody-derived tag (ADT) libraries, were constructed according to the manufacturer’s instructions. The resulting libraries were sequenced on an Illumina NovaSeq 6000 platform using an S4 flow cell to a targeted depth of ~50,000 reads per cell.

Single-cell data analysis

We processed raw sequencing data using the Cell Ranger pipeline (v7.2.0, 10x Genomics), which included demultiplexing, alignment to the GRCh38 human reference genome, cell barcode assignment, and generation of gene expression matrices via unique molecular identifiers (UMIs). The Cell Ranger multi-pipeline (5′ immune profiling workflow) was used to simultaneously process gene expression, antibody-derived tag (ADT), hashtag antibody, and TCR V(D)J data.

To integrate the gene expression data with the ADT data, we utilized the IntegrateLayers function in R. We employed the RPCAIntegration method, a robust technique for integrating multi-modal single-cell data that accounts for technical noise and batch effects. For this process, we set the orig.reduction parameter to “pca” and the new.reduction parameter to “integrated.rpca”. The integrated data was subsequently combined into a single Seurat object using the JoinLayers function for downstream analysis.

To assess the effectiveness of batch effect removal and to demonstrate the quality of integration, we performed quantitative evaluations using metrics that assess both batch correction and biological conservation. To specifically evaluate the removal of technical variation, we applied the k-nearest neighbor batch effect test (kBET), which quantifies the degree of batch mixing within the integrated dataset. A low kBET rejection rate indicates that cells originating from different GEM wells are well interspersed, thereby reflecting successful batch correction.²⁹

Because kBET is computationally demanding and memory-intensive when applied to large-scale single-cell datasets, we adopted a subsampling strategy, as widely recommended for such applications. The dataset was randomly partitioned into 20 subsets, and average rejection rates were computed across these subsets, providing a robust estimate of batch correction performance.

We further performed the same analysis at the cell-type level, reporting average rejection rates across individual clusters and visualizing the aggregated subset results using boxplots. In addition, UMAP projections were generated and colored by the original GEM well (to assess batch mixing), thereby providing a complementary visual confirmation of successful integration.

To distinguish γδ and αβ TCR clonotypes, we utilized the reference dataset refdata-cellranger-vdj-GRCh38-alts-ensembl-7.0.0. Background subtraction for isotype control normalization of antibody signals was performed according to standard practice by subtracting each antibody’s background expression level from its raw signal.

The resulting gene expression matrix was analyzed using the Seurat R package (v5.1).³⁰ Cells were filtered based on the following criteria: (1) retention of the top 99% of cells by UMI counts, and (2) <20% mitochondrial gene expression. Additionally, genes were retained if expressed in at least three cells, and cells were included only if they expressed more than 200 genes. After quality control, UMI counts were normalized to total expression per cell and transformed using a natural logarithmic scale. Samples were integrated using Seurat’s merge function. Dimensionality reduction was performed via principal components analysis (PCA) on the normalized expression matrix. Clustering was performed using graph-based clustering followed by Uniform Manifold Approximation and Projection (UMAP) for visualization. T cell clusters were identified using Seurat’s FindClusters function with a resolution parameter of 0.3. Cluster annotation was guided by literature-curated marker genes, unsupervised learning algorithms, and differential expression analysis using the FindAllMarkers function in Seurat.

Differential gene expression analysis was performed using the MAST algorithm,³¹ comparing each exosome-treated sample to its corresponding mock-treated control within each defined cluster. Mock-treated group refers to cells that underwent the same experimental manipulations as the treatment group but without exposure to TNBC-derived exosomes. For pathway enrichment analysis, significantly differentially expressed genes (DEGs) were analyzed using the MSigDB Hallmark gene sets via gene set enrichment analysis.³² p values were adjusted for multiple comparisons using the Benjamini–Hochberg method, and enriched pathways were grouped by T cell cluster identity based on the annotation described above.

Non-coding RNA profiling

Non-coding RNAs were isolated from exosomes derived from 17 TNBC cell lines as well as from plasma samples of 28 TNBC patients and 7 healthy donors, using the MagMAX™ mirVana™ Total RNA Isolation Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. The isolated RNA, enriched for small and non-coding RNAs, was quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific), and RNA quality was assessed with an Agilent Bioanalyzer 2100 (Agilent Technologies). Library preparation was performed using the NEXTFLEX® Small RNA-Seq Kit v4 with Unique Dual Indices (UDIs) (Revvity), which includes 3′ and 5′ adapter ligation, reverse transcription, PCR amplification, and size selection steps optimized for non-coding RNA capture. Prepared libraries were quantified using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific) and validated for size distribution using the Agilent Bioanalyzer. Sequencing was conducted on an Illumina platform following standard procedures for small RNA libraries.

Sequencing data were analyzed using the Partek Flow cloud platform (Illumina). Adapter trimming was performed using CUTADAPT v4.2, applying a Phred quality cutoff of 20 and removing reads shorter than 16 nucleotides. Both 3′ and 5′ adapter sequences from paired reads were trimmed. miRNA reads were aligned to miRBase v22.1 using Bowtie2 v2.2.5 with a seed length of 10 and allowance for one seed mismatch. For lncRNA, reads were aligned to the GRCh38 (hg38) reference genome using STAR v2.7.8a, allowing a maximum of 3 mismatches, with all other parameters left at default settings. Transcript abundance was quantified using the Partek Expectation-Maximization (E/M) algorithm, with feature filtering set to a minimum of one read per feature. Annotated transcript counts were extracted using the “Total miRNA” and “Transcript Count” outputs for downstream analysis. Differential expression of non-coding RNAs in patient-derived plasma samples was determined using a threshold of log₂ fold change (log₂FC) > 2. In exosome-derived datasets, fold change (FC) values were calculated for each TNBC cell line relative to a control primary mammary epithelial cell line (PBreast) (HMEC, PCS-600-010, ATCC). When non-coding RNA expression was below detection thresholds (read count <10 in PBreast and <20 in TNBC), an FC of zero was assigned to prevent artificial inflation of relative expression. To investigate miRNA-mRNA regulatory networks, we queried the miRTarBase database (release containing 938,073 experimentally validated interactions).³³ Genes and miRNAs that were concurrently dysregulated within the same sample were mapped to validated interactions, revealing putative regulatory relationships relevant to TNBC pathogenesis and immune modulation.

Electroporation of miRNA mimics and western blot analysis

Following T cell activation using a CD3/CD28 stimulation cocktail, cells were electroporated with miRNA mimics (mirVana miRNA Mimics; Thermo Fisher Scientific) using BTX electroporation buffer. Control samples were electroporated with the mirVana miRNA Mimic Negative Control. After electroporation, cells were returned to culture, and protein lysates were collected 48 h post-electroporation for downstream immunoblotting.

For Western blotting, cells were lysed in ice-cold lysis buffer supplemented with protease and phosphatase inhibitors, and lysates were clarified by centrifugation. Protein concentration was determined using the BCA assay, and equal amounts of protein were denatured in SDS sample buffer and resolved by SDS–PAGE using Bio-Rad stain-free gels. Total protein in each lane was visualized using the stain-free imaging workflow (Bio-Rad) and used to verify equivalent loading/transfer and for normalization.

Proteins were transferred to nitrocellulose membranes, blocked, and incubated 1 h at RT with primary antibodies against anti-STAT3 (Clone 4G4B45), anti-phospho-STAT3 (Tyr705) (Clone 13A3-1), anti-human HB-EGF (Clone 3H4), anti-human CD137L/4-1BBL (TNFSF9) (Clone 5F4), and anti-GAPDH (Clone FF26A/F9) (all from BioLegend). After washing, membranes were incubated with IRDye® 800CW Goat Anti-Mouse IgG secondary antibody (LI-COR), washed extensively, and the fluorescent signal was detected on the Bio-Rad near-infrared imaging system (800 nm channel).

Bulk cytokine analysis

Multiplex cytokine analysis was performed using the Luminex xMAP platform (Luminex Corporation). Supernatants from cell cultures were collected after 48 h of exosome treatment. Briefly, 50 μL of supernatant was incubated with antibody-coupled magnetic beads for 2 h at room temperature, followed by detection with biotinylated secondary antibodies and streptavidin-phycoerythrin. Fluorescence signals were measured using a Luminex 200 instrument, and cytokine concentrations were calculated against a standard curve using a 5-parameter logistic regression model. All samples were analyzed in duplicate, and data were processed using xPONENT software (version 4.2, Luminex Corporation). Results were expressed as pg/mL, with a lower limit of detection ranging from 3.2 to 10 pg/mL depending on the cytokine.

Single-cell cytokine secretome analysis

Single-cell cytokine profiling was conducted using the IsoLight system (Bruker Cellular Analysis) with the IsoCode chip technology. Following 48 h treatment with exosomes, cells were loaded onto IsoCode chips containing approximately 12,000 microchambers, each designed to capture a single cell. The chips were pre-coated with a 32-plex antibody array targeting cytokines. Following cell capture, chips were incubated for 16 h at 37 °C to allow cytokine secretion. Secreted cytokines were detected using fluorescence-based sandwich immunoassays. The IsoLight system automatically imaged and analyzed microchambers to quantify cytokine secretion at the single-cell level. Data were processed using IsoSpeak software (version 2.1, Bruker Cellular Analysis), which generated polyfunctionality metrics, including the Polyfunctional Strength Index (PSI), to assess the diversity and intensity of cytokine secretion. PSI was calculated according to the following formula: PSI = (% of polyfunctional cells)×(average normalized cytokine intensity of polyfunctional cells). Single-cell data were gated to exclude debris and doublets, and results were reported as the frequency of cytokine-secreting cells and mean fluorescence intensity per cytokine.

Author contributions

R.S. conceived and designed the study and provided funding support. S.R.B., K.W., and R.S. performed the experiments. S.M., S.R.B., K.W., and R.S. analyzed data and generated figures. S.R.B., S.M., L.A., and R.S. reviewed and interpreted data. S.M. and R.S. wrote the manuscript. All authors reviewed and edited the final manuscript.

Data availability

All sequencing data, including single-cell RNA sequencing and noncoding RNA sequencing data generated in this study, are deposited in the NCBI GEO database (https://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE302399.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41392-026-02784-4.