CD25^high Effector Regulatory T Cells Hamper Responses to PD-1 Blockade in Triple-Negative Breast Cancer

An imbalance between effector CD8⁺ T cells and CD25^high effector Tregs marks immunosuppressive microenvironments in αPD-1–resistant TNBC and can be reversed through effector Treg depletion to increase αPD-1 efficacy.

Abstract

Regulatory T cells (Treg) impede effective antitumor immunity. However, the role of Tregs in the clinical outcomes of patients with triple-negative breast cancer (TNBC) remains controversial. Here, we found that an immunosuppressive TNBC microenvironment is marked by an imbalance between effector αβCD8⁺ T cells and Tregs harboring hallmarks of highly suppressive effector Tregs (eTreg). Intratumoral eTregs strongly expressed PD-1 and persisted in patients with TNBC resistant to PD-1 blockade. Importantly, CD25 was the most selective surface marker of eTregs in primary TNBC and metastases compared with other candidate targets for eTreg depletion currently being evaluated in trials for patients with advanced TNBC. In a syngeneic TNBC model, the use of Fc-optimized, IL2 sparing, anti-CD25 antibodies synergized with PD-1 blockade to promote systemic antitumor immunity and durable tumor growth control by increasing effector αβCD8⁺ T-cell/Treg ratios in tumors and in the periphery. Together, this study provides the rationale for the clinical translation of anti-CD25 therapy to improve PD-1 blockade responses in patients with TNBC.

Significance:

An imbalance between effector CD8⁺ T cells and CD25^high effector Tregs marks immunosuppressive microenvironments in αPD-1–resistant TNBC and can be reversed through effector Treg depletion to increase αPD-1 efficacy.

Graphical Abstract

Introduction

Triple-negative breast cancer (TNBC) is the deadliest subtype of breast cancer. Standard-of-care options for patients with TNBC include in stage II–III diseases the combination of neoadjuvant chemotherapy (NAC) with pembrolizumab, an immune-checkpoint inhibitor targeting the programmed cell death-1 (PD-1) receptor, while stage IV (metastatic or locally advanced unresectable) programmed death-ligand 1 (PD-L1)-positive diseases also benefit from the combination of chemotherapy and pembrolizumab (1). Treatment with anti–PD-1 antibodies (αPD-1) reinvigorates PD-1⁺ antigen-experienced αβCD8 T cells in patients with TNBC (2). However, about 40% and 80% of patients with early-stage and advanced TNBC do not respond to chemoimmunotherapy, respectively (3, 4).

Regulatory T cells (Treg; CD4⁺ FOXP3⁺ IL2Rα [CD25]⁺) exert immunosuppressive functions that are crucial for maintaining self-tolerance in physiologic settings. However, immune-evasive tumor microenvironments favor the accumulation of Tregs harboring highly suppressive phenotypes corresponding to effector Tregs (eTreg) originally described by Sakagushi and colleagues (CD45RA⁻ FOXP3^high CD25^high; refs. 5, 6). The prognostic value of the intratumoral quantity of FOXP3⁺ Tregs is controversial in TNBC, as studies indicate associations with favorable (7) or worse clinical outcomes (8). In patients with other types of solid tumor [e.g., non–small cell lung cancer (NSCLC) and melanoma], an imbalance between effector αβCD8 T cells and Tregs harboring canonical markers of eTregs has been associated with decreased survival and poor efficacy of αPD-1, providing the rationale for combination therapy based on eTreg depletion (9). To our knowledge, no study has yet defined subsets of Tregs in TNBC.

Current strategies aiming at depleting Tregs include the use of monoclonal antibodies (mAb) harboring fragment crystallizable (Fc) domains optimized for higher binding affinity to activating Fc gamma receptors (FcγRs; ref. 10). These mAbs target several eTreg-related proteins in combination with αPD-1 and are currently being tested in clinical trials for patients with various types of solid tumors, including TNBC (10). However, tailoring candidate targets to the context of TNBC is crucial to maximizing the therapeutic effects of combination strategies based on Treg depletion. Indeed, Tregs acquire transcriptional programs reflecting tissue adaptation (11, 12), and thereby express distinct phenotypes according to tissue location and tumor type (13). Moreover, a systemic increase in Treg abundance is associated with poor outcomes in patients with breast cancer (14, 15). These data prompt the targeting of proteins selectively expressed by Tregs at multiple sites, namely, primary tumors, invaded tumor-draining lymph nodes (iTDLN) and distant metastases. CD25 was the first marker used to identify Tregs (16), whose development and survival depend on IL2 signaling (17). In solid tumors, CD25^high eTregs exert local IL2 deprivation, thus inhibiting antitumor immunity (17). The combination of Fc-optimized anti-CD25 mAbs with αPD-1 has shown preclinical antitumor efficacy (18), making CD25 a prime target for Treg depletion in patients with cancer resistant to αPD-1. However, on-target off-Treg effects should be considered in the context of combination therapy, as recent studies have shown that IL2 signaling is required for reinvigoration of antigen-experienced αβCD8 T cells by αPD-1 (19).

Here, we first sought to dissect the ontogenesis and functional state of breast tumor-infiltrating eTregs using single-cell RNA sequencing (scRNA-seq) or single-cell TCR sequencing (scTCR-seq) data from patients with early, advanced, or metastatic TNBC. We report an imbalance between cytotoxic effector αβCD8 T cells and bona fide CD25^high 4-1BB⁺ PD-1^high eTregs in aggressive primary TNBC. This imbalance was a multifaced hallmark of immune alterations related to TNBC immunosuppression, including the accumulation of PD-1^high exhausted αβCD8 T cells and M2-like macrophages. CD25^high eTregs were further detected in iTDLNs and distant metastases. Mass cytometry validation revealed CD25 as the most selective surface target for intratumoral eTreg depletion at multiple tumor sites. Importantly, CD25^high eTregs persisted in TNBC resistant to αPD-1 therapy. We therefore evaluated the activity of an Fc-optimized anti-CD25 mAb that does not block IL2 signaling (αCD25NIB; ref. 20) in a syngeneic TNBC model. Combination of αCD25NIB and αPD-1 promoted durable antitumor immunity by increasing ratios of cytotoxic effector αβCD8 T cells to Tregs in primary TNBC and in the periphery.

Materials and Methods

Human subjects

After providing written informed consent, patients over 18 years of age with a diagnosis of breast cancer were prospectively enrolled onto an Institutional Review Board [Comité d'Orientation Stratégique (COS), Marseille, France]–approved protocol from the Paoli-Calmettes Institute (BC-BIO-IPC protocol, NCT01521676, 2009 approval). The study protocol complied with the Good Clinical Practice guidelines and the Declaration of Helsinki. Fresh EDTA-anticoagulated blood samples (N = 7) and primary tumor samples (N = 8) were collected prior to any treatment at diagnosis and at debulking surgery, respectively. Fresh normal breast tissues (prophylactic mastectomies, N = 4) were collected at surgery. Fresh EDTA-anticoagulated blood samples (N = 4) from healthy donors were recruited by the Etablissement Français du Sang (EFS Alpes-Méditerranée). Clinical characteristics of patients with breast cancer enrolled in the BC-BIO-IPC protocol, as well as clinical characteristics of patients with breast cancer enrolled onto public scRNA-seq studies, are summarized in Supplementary Table S1.

Mice

Female Balb/c mice were purchased from Shanghai Lingchang Biotechnology Co., Ltd. All mice were housed and maintained at CrownBioscience Inc. The protocol and any amendment(s) or procedures involving the care and use of animals in this study were reviewed and approved by the Institutional Animal Care and Use Committee of CrownBioscience Inc. prior to execution. During the study, the care and use of animals were conducted in accordance with the regulations of the Association for Assessment and Accreditation of Laboratory Animal Care.

Cell lines

EMT-6 cells were obtained from ATCC. Mycoplasma-free EMT-6 cells were cultured in vitro as a monolayer in DMEM supplemented with 10% heat-inactivated fetal calf serum at 37°C in an atmosphere of 5% CO₂. EMT-6 cells were subcultured twice weekly, harvested in an exponential growth phase, and counted before tumor inoculation.

Syngeneic tumor studies

EMT-6 tumor studies were performed by inoculating 6- to 7-week-old female BALB/c mice with a subcutaneous injection of 5 × 10⁵ EMT-6 cells in 100 μL PBS in the right rear flank region. Mice were randomized to a treatment arm once tumors reached 80 to 120 mm³ (6 days following tumor inoculation, day 0). Tumor-bearing mice were intraperitoneally injected with either vehicle (PBS, day 0), 200 μg/mouse non-IL2-blocking anti-CD25 antibodies (αCD25NIB, clone 7D4 ALD2510 murine surrogate, day 0), 10 mg/kg*4 anti–PD-1 antibodies (αPD-1, mIgG2aκ, clone RMP-1-14, days 1, 4, 8, and 11), or a combination of αCD25NIB plus αPD-1. Seven days following treatment administration, 4 animals per group of mice were sacrificed for flow cytometry analysis of isolated tumor-infiltrating lymphocytes and peripheral blood mononuclear cells. On day 39, naïve control and surviving mice from αCD25NIB and αCD25NIB plus αPD-1 groups were rechallenged in the left rear opposite flank region with a subcutaneous injection of 5 × 10⁵ EMT-6 cells in 100 μL PBS. Mice bearing tumors that exceeded 2,500 mm³ were euthanized following approved protocols. Tumor volumes were measured twice a week inside a laminar flow cabinet after randomization in two dimensions using a caliper. The volumes were expressed in mm³ using the formula: V = (L × W × W)/2, where V is tumor volume, L is tumor length (the longest tumor dimension), and W is tumor width (the longest tumor dimension perpendicular to L). No signs of adverse effects were observed during treatments.

scRNA-seq

Data collection, filtering, and normalization

In this study, we reanalyzed 8 publicly available scRNA-seq data sets using BBrowser V3.5.26 (BioTuring). Data sets were queried through the BioTuring database or the Gene Expression Omnibus (GEO) database [managed by the National Center for Biotechnology Information (NCBI)] and imported into BBrowser. Data sets filtering for exclusion of dropouts, doublets, and apoptotic cells were performed according to the quality control criteria described in the source publication. For public data sets without a detailed filtering process in the source publication, we excluded genes having at least 1 UMI count in less than three cells, cells with less than 200 genes having at least 1 UMI count, and cells with a mitochondria genes ratio > 5. Raw expression matrices were then subjected to log normalization.

Batch-effect removal, dimensionality reduction, cell clustering, and annotation

To avoid donor-driven and study-driven cell clustering, we used the MNN batch-effect removal algorithm when computing dimensionality reduction by Uniform Manifold Approximation and Projection (UMAP). The top 2000 highly variable genes and the first 30 components of the principal component analysis were used to compute UMAP using the uwot package available in BBrowser. Perplexity for UMAP was set automatically based on the count of cells in a data set, or the count of cells in clusters selected for subclustering. Unsupervised graph-based clustering (Louvain clustering) was performed with a number of nearest neighbors ≤ 8 using the igraph package available in BBrowser. Clusters were annotated based on the expression of markers identified by differential gene-expression analysis using the Venice test.

Pseudotime analysis

To infer the pseudotime trajectory on UMAP, user-selected cell embeddings were fed to Monocle 3′s algorithm available in BBrowser.

Gene set enrichment analysis

Log-normalized expression matrices were exported from BBrowser and imported into the gene set enrichment analysis (GSEA) software developed by the Broad Institute. Fold-change expression values and adjusted P values were combined to rank the genes as input for GSEA.

Between-group analysis

The percentage of 35 unique intratumoral immune cell types relative to the total count of immune cells was exported from BBrowser and imported into R for between-group analysis (BGA) using the made4 package. R scripts performing BGA are provided in Supplementary File S1.

Ligand–receptor interaction analysis

Log-normalized expression matrices were exported from BBrowser, and imported into R for annotation of ligand–receptor pairs and identification of significant cellular interactions using the iTALK package. The top 2,000 highly expressed genes were selected to perform the analysis.

Single-cell TCR sequencing

Already curated scTCR-seq data from Zhang and colleagues (21) were queried through the BioTuring database and analyzed using the BBrowserX software (Talk2Data, BioTuring). The BBrowserX software only considered clonotypes that were full-length (valid V and J annotations) and productive (transcripts translate to a protein with a CRD3 region).

Bulk RNA-seq and microarray

Clinical characteristics of patients with TNBC and log-normalized gene-expression values from two public bulk RNA-seq data sets were used in this study. Data sets were queried through cBioPortal (TCGA–BRCA, Firehose Legacy) and the GEO (NCBI) database (GSE177043). Clinical characteristics of patients with TNBC and log-normalized/batch-corrected gene-expression values from one public microarray data set were used in this study. The data set was queried through the GEO database (GSE194040). Patients with TNBC treated with paclitaxel as monotherapy or combined with pembrolizumab were selected. Tumors were sampled prior to any treatment.

Mass cytometry

Tissue dissociation, cell staining, data acquisition, and analysis were performed as described in (22). Additional details are provided in the Supplementary Materials and Methods and Supplementary Table.

Statistical analysis

The statistical methods used for each analysis are described in the figure legends. A P value < 0.05 was considered significant. Statistical significance between two unpaired groups was calculated using the nonparametric Wilcoxon–Mann–Whitney test. Statistical significance between two paired groups was calculated using the nonparametric Wilcoxon matched-pairs signed-rank test. In cases of multiple comparisons, P values were adjusted by false discovery rate (FDR) using the Benjamini–Hochberg process. For GSEA, only FDR-adjusted P values < 0.05 were considered. All dot plots are expressed as mean ± SEM. All violin plots report the kernel density estimates as the width, with box plots expressed as mean ± SEM. For survival analyses, the overall survival was defined as the time from diagnosis until death from any cause. Patients were censored at a cutoff date of 108 months. Survival times were estimated by the Kaplan–Meier method and compared using the log-rank test. Statistical analyses were generated using BBrowser V3.5.26 (BioTuring) and Prism V8.0.2 (GraphPad).

Data availability

The data analyzed in this study were obtained from GEO at GSE164898, GSE169246, GSE179994, GSE140228, GSE164690, GSE108989, GSE177043, and GSE194040, from the European Genome-phenome Archive (EGA) at EGAS00001004809, from The Tabula Sapiens Consortium at https://tabula-sapiens-portal.ds.czbiohub.org/, from TCGA Program at https://www.cbioportal.org/study/summary?id=brca_tcga, and from the BioTuring database at https://bioturing.com/. All other raw data are available upon request from the corresponding author.

Additional materials and methods

Additional details and product catalog numbers are described in the Supplementary Materials and Methods and Supplementary Table.

Results

A low ratio of cytotoxic effector αβCD8 T cell to terminally differentiated eTreg marks untreated primary TNBC

We first sought to define subsets of Tregs in primary tumors of patients with breast cancer. To this end, we collected public scRNA-seq data from breast tissues of 6 healthy donors and 24 patients with untreated early breast cancer (Supplementary Table S1; refs. 2, 23). After rigorous quality control filtering, we excluded stromal and epithelial cells to obtain the expression profiles of 42,943 immune cells (Supplementary Fig. S1A–S1B). Graph-based clustering identified 25 major immune cell populations (Supplementary Tables S2 and S3). A single cluster of Tregs (FOXP3⁺, IKZF2⁺ [HELIOS], IL2RA⁺ [CD25]) was first identified. Subclustering of 15,430 αβCD4⁺ T cells enabled improved resolution of the Treg clusters and identification of eTregs (FOXP3^high, IL2RA^high; Fig. 1A; Supplementary Fig. S1C).

Figure 1.

A low ratio of cytotoxic effector αβCD8 T cell to terminally differentiated eTreg marks untreated primary TNBC. A, UMAP visualization of 13,463 αβCD4⁺ T cells from primary breast tumor (PBT, N = 24; Supplementary Table S1). Louvain clustering was used for the identification of cell types (Supplementary Fig. S1; Supplementary Tables S2 and S3). Top left, UMAP colored by an inferred pseudotemporal reconstruction of Treg activation/differentiation processes. Top right, spline plots showing, along the inferred trajectory (x-axis), expression levels of genes (y-axis). The dashed line indicates the pseudotime (≃ 0.6) at which Tregs acquire an eTreg transcriptional program. Bottom, heat map of Tregs ordered by the inferred trajectory (y-axis). The color gradient indicates expression levels of 50 genes whose expression changes significantly along the trajectory (Supplementary Table S4); Moran's I test for spatial autocorrelation of genes; genes annotated in red encode transcription factors or candidate targets for intratumoral eTreg depletion. Tn:cm, naive and central memory T cell; T non-act, nonactivated T cell; Tearly act, early activated T cell; Teff, cytotoxic effector T cell; Tfh, follicular helper T cell. B, Top, UMAP colored by eTreg populations. Conventional αβCD4⁺ T cells, light gray; Tregs, gray; TNFRSF9⁻ eTregs, yellow; TNFRSF9⁺IL2RA^high eTregs, red; TNFRSF9⁺IL2RA^high eTregs, TNFRSF9 log₂ expression ≥ 0.5. Bottom, percentage of TNFRSF9⁻ eTregs (yellow) and TNFRSF9⁺IL2RA^high eTregs (red) expressing differentially expressed genes (Supplementary Fig. S2A; Supplementary Table S5). Each single dot represents a patient. Wilcoxon matched-pairs signed-rank test. C, Left, representative overlaid histograms comparing CD25 and PD-1 protein expression levels by Tregs from normal breast tissue (NBT; green) and by intratumoral 4–1BB⁻ Tregs (yellow) or 4-1BB⁺ eTregs (red) from primary TNBC (Supplementary Fig. S2B–S2D). Right, percentage of Tregs from normal breast tissue and percentage of intratumoral 4-1BB⁻ Tregs or 4-1BB⁺ eTregs expressing high levels of CD25 (Supplementary Fig. S2B) and PD-1. Each single dot represents a donor. Wilcoxon matched-pairs signed-rank test; mean ± SEM. D, Left, expression levels of two nonoverlapping signature scores related to eTreg activation and IL2/STAT5 signaling pathway, computed per intratumoral αβCD4⁺ T cell. Right, Hallmark/Reactome/ImmuneSigDB pathways enriched in TNFRSF9⁺IL2RA^high eTregs versus TNFRSF9⁻ eTregs in tumors (Table 1). Adj P, adjusted P value; NES, normalized enrichment score. E, Jaccard indices computed for sharing of identical TCR clonotypes between intratumoral TNFRSF9⁺IL2RA^high eTregs and indicated cell types from paired peripheral blood (blood) and primary TNBC (PBT) samples (N = 4; Supplementary Fig. S3A–S3B). Each single dot represents a patient. Wilcoxon–Mann–Whitney test; mean ± SEM. Tem, effector memory T cell. F, Schematic overview of key molecular mediators involved in CD25^high eTreg differentiation in human primary TNBC. G, Top, percentage of TNFRSF9⁺IL2RA^high eTregs and percentage of TNFRSF9⁺IL2RA^high eTregs displaying IL2RA log₂ expression ≥ 2 (Supplementary Fig. S2C), relative to the total count of Tregs. Bottom, ratio of cytotoxic effector αβCD8 T-cell counts (Supplementary Fig. S4A) to TNFRSF9⁺IL2RA^high eTreg counts. Each single dot represents a donor. αβCD8 Teff, cytotoxic effector αβCD8 T cell; HR⁺, hormone receptor–positive breast cancer (N = 12); HER2⁺, human epidermal growth factor receptor 2-positive breast cancer (N = 2); TNBC (N = 10). Wilcoxon–Mann–Whitney test; mean ± SEM. H, Left, heat map summarizing CD8A and TNFRSF9 expression levels per immune cell type in untreated primary TNBC. Color gradient indicates gene-expression levels. Circle size indicates the percentage of the cell type that expresses the gene. pDC, plasmacytoid dendritic cell; cDC, conventional dendritic cell; NK, natural killer cell; Tmp, memory precursor T cell; Tex, exhausted T cell; Tprlf, proliferating T cell. Right, Kaplan–Meier estimates of overall survival according to the ratio of CD8A expression to FOXP3 or TNFRSF9 expression in primary tumors from two independent cohorts of patients with TNBC (Supplementary Fig. S4B–S4C). Group “low” = 1st quartile of ratios; Log-rank test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. (F, Created with BioRender.com.)

Defining the ontogeny of intratumoral eTregs could guide the choice of therapeutic strategy to reverse immunosuppression in breast cancer. Indeed, previous studies have suggested that conventional αβCD4⁺ T cells recruited from peripheral blood or residing in breast tissue are the main source of breast tumor-infiltrating Tregs (24). Here, to investigate the ontogenesis of intratumoral eTregs, we first ordered total breast tumor-infiltrating Tregs on an inferred pseudotemporal reconstruction of Treg activation/differentiation processes using Monocle 3 (Fig. 1A). Of note, cell embeddings were uniquely selected within the clusters of Tregs, as there was no pattern of transition between Tregs and conventional αβCD4 T cells on the UMAP plot. This indicates that intratumoral Tregs are less likely to derive from the local conversion of peripherally recruited or tissue-resident conventional αβCD4 T cells. To gain insight into key regulators, we ranked the top 100 genes whose expression changed significantly over the inferred trajectory using the Moran's I test (adjusted P < 0.0001; Fig. 1A; Supplementary Table S4). Consistent with recent work identifying BATF and JUNB as early drivers of Treg activation in melanoma, lung cancer, and kidney cancer (25), BATF and JUNB expression levels were, respectively, gradually increased and decreased at early stages of the Treg trajectory, suggesting conserved Treg activation programs across solid tumors. Several genes encoding other transcription factors were upregulated at latter stages of the Treg trajectory, namely, Nur77 (NR4A1), NF-κB (RELB, NFKB2, and REL), and SOX4. SOX4 is a key downstream target of pathways supporting eTregs’ suppressive functions, such as TGFβ signaling. Nur77 and NF-κB have been shown to transactivate FOXP3 by acting downstream of TCR signaling. Consistently, we observed increased FOXP3 levels along the Treg trajectory. Moreover, the expression of genes encoding candidate targets for eTreg depletion was increased along the Treg trajectory, including IL2RA (CD25), CCR8, and TNFRSF9 (4-1BB). Several studies have used 4-1BB expression to define highly suppressive eTregs in human tumors (5). For uniform nomenclature, intratumoral TNFRSF9⁺ eTregs were manually gated and compared with TNFRSF9⁻ eTregs (Fig. 1B; Supplementary Fig. S2A; Supplementary Table S5). TNFRSF9⁺ eTregs were located at the very end of the eTreg cluster. Consistent with the trajectory analysis, the latter had increased expression levels of IL2RA (mean % IL2RA^high = 29% of eTregs, 44% of TNFRSF9⁺IL2RA^high eTregs; P < 0.01). TNFRSF9⁺IL2RA^high eTregs strongly expressed regulators of Treg differentiation mentioned above, and genes encoding molecules involved in Treg-suppressive functions, including a decoy receptor neutralizing proinflammatory IL1 (IL1R2), a membrane protein required for the release of active TGFβ1 [LRRC32 (GARP)] and tumor necrosis factor receptor (TNFR)-induced antiapoptotic proteins (TRAF1, BCL2L1). PD-1 is a negative regulator of Treg activation (26). PDCD1 was preferentially expressed by TNFRSF9⁺IL2RA^high eTregs in most patients (mean % PDCD1⁺ = 11% of Tregs, 21% of eTregs, 32% of TNFRSF9⁺IL2RA^high eTregs; P < 0.05). We confirmed these observations at the protein level using mass cytometry (Supplementary Table S6), as evidenced by the high expression of CD25 and PD-1 on breast tumor-infiltrating 4-1BB⁺ eTregs (mean % of PD-1⁺ = 27% of intratumoral Tregs, 50% of intratumoral 4-1BB⁺ CD25^high eTregs; P < 0.01; Fig. 1C; Supplementary Fig. S2B–S2D), warranting monitoring of this population in patients treated with PD-1 blockade. Noteworthy, we found similar results in the peripheral blood of breast cancer patients (Supplementary Fig. S2E).

Trajectory analysis evidenced the involvement of key transcription factors acting downstream of inflammatory pathways in the accumulation of intratumoral eTregs (e.g., NR4A1 and TCR signaling, SOX4 and TGFβ signaling, REL/RELB/NFKB2 and NF-κB signaling). GSEA further revealed that components of signatures related to Treg activation/differentiation are indeed enriched in TNFRSF9⁺IL2RA^high eTregs compared with TNFRSF9⁻ eTregs, including TGFβ signaling (KLF10, UBE2D3, SKIL, TGFB1, SPTBN1; adjusted P < 0.01), signatures of metabolic adaptations to low-glucose tumor microenvironments (GAPDH, MDH1/2, LDHA, ATP5s; adjusted P < 0.0001) and IL2 signaling (CCND2, SLC1A5, MYC, BATF3, IRF4; adjusted P < 0.0001; Fig. 1D; Table 1), indicating that CD25 expression on breast tumor-infiltrating eTregs is indeed functional. To further confirm our trajectory analysis, we investigated the ontogeny of TNFRSF9⁺IL2RA^high eTregs using scTCR-seq data from untreated primary TNBC (N = 4; ref. 21). A total of 603 full-length/productive TCR clonotypes were detected for 1,049 intratumoral Tregs. We first computed Jaccard indices per patient for the identification of shared clonotypes between intratumoral Tregs, TNFRSF9⁺IL2RA^high eTregs, and subsets of conventional αβ T cells from the peripheral blood and tumors (Fig. 1E; Supplementary Fig. S3A–S3B). Intratumoral Tregs shared very few clonotypes with peripheral and intratumoral αβCD4 T cells (mean % of shared clonotypes: 3.5% of Treg clonotypes, 1.5% of TNFRSF9⁺IL2RA^high eTreg clonotypes), indicating that most intratumoral Tregs do not originate from the recruitment of their analogues from the periphery or the conversion of conventional αβCD4 T cells. In contrast, intratumoral Tregs and TNFRSF9⁺IL2RA^high eTregs shared a significant proportion of their clonotypes (mean [min–max]% of shared clonotypes: 8.3 [2–20]%; P < 0.05). These data are consistent with trajectory analysis suggesting intratumoral IL2RA⁺ Treg differentiation into TNFRSF9⁺IL2RA^high eTreg (Fig. 1F). In rarefaction and extrapolation analysis, the high diversity of the TCR repertoire further indicated that intratumoral Tregs are more likely to be self-reactive bona fide Tregs (Supplementary Fig. S3C). Indeed, despite the activated phenotype of these Tregs, they exhibited a TCR diversity close to that of naïve/central memory αβ T cells in tumors. Conversely, exhausted αβCD8 T cells had lower TCR diversity, reflecting engagement of antigen-experienced αβCD8 T cells in primary TNBC.

Table 1.

GSEA outputs based on the differentially expressed genes between TNFRSF9⁺IL2RA^high eTregs and TNFRSF9⁻ eTregs, related to Fig. 1D.

Pathways enriched in TNFRSF9+ IL2RAhigh eTregs vs. TNFRSF9-eTregs	Adjusted P value	NES	Gene set leading edge genes
GSE15659_RESTING_vs._ACTIVATED_TREG_DN	0	2.12	TNFAIP3, RAN, SEC61B, TPI1, TOP1, TBC1D4, WDR92, TMEM208, WARS, PTPN23, ROCK1, TYK2, SLC25A3, SEC24D, SCO2, ZMIZ2, TOMM22, RNF10, TRIP12, RANGRF, TMBIM1, PRDX3, RIT1, PRDX4, SUSD1, TIPRL, TMEM126A, PTK2B, SARS2, ZBED6, TPMT, SLC35F3, SNX3, ZC3H13, SAP30BP, TTC7A, SPATC1, RSPH3, SLC9A8, VPS28, UBA3, TSTA3, PPP2CA, PPP2R5A, RAP2B, SLC43A3, S1PR2, POLR2F, ZNF324B, RNF214, STK25, UBFD1, TSNAX, PPHLN1, UBE2D1, STK36, ZNF596, SEPHS2, TMEM70, PSRC1, RPL26L1, ZNRD1, ZFAND5
REACTOME_COSTIMULATION_BY_THE_CD28_FAMILY	0	2.10	PTPN6, PDCD1, HLA-DQA1, CD80, FYN, ICOSLG, RAC1, HLA-DQA2, CD3G, HLA-DRB1, CTLA4, ICOS, BTLA, HLA-DQB1, PTPN11, HLA- DRB5, CD4, MAPKAP1, HLA-DPA1, CD3D, MAP3K8, CD274, AKT2, AKT3, PPP2R1B, VAV1, SRC, YES1, PAK2, PAK1, PPP2CA, PPP2R5A, AKT1, CDC42
HALLMARK_PI3K_AKT_MTOR_SIGNALING	0	2.09	MAP2K3, HSP90B1, TIAM1, UBE2D3, CALR, ACTR3, RAC1, RPS6KA3, CDKN1A, ARHGDIA, UBE2N, MKNK1, CLTC, RPS6KA1, EIF4E, PTPN11, SQSTM1, PFN1, VAV3, MAPKAP1, RIT1, AKT1S1, RAF1, AP2M1, GSK3B, CDK4, PIKFYVE, CAB39, ECSIT, PPP2R1B, CAB39L, PPP1CA, ARF1, CAMK4, PRKAG1, DDIT3, AKT1, PRKAR2A
HALLMARK_PROTEIN_SECRETION	0	1.99	SOD1, ANP32E, GNAS, STAM, YIPF6, RPS6KA3, PAM, CLTC, SCAMP3, IGF2R, SEC24D, ATP1A1, LMAN1, TMED2, M6PR, CLN5, VAMP7, TMED10, CLTA, ARFGAP3, AP2M1, COPB2, AP1G1, AP3B1, GLA, USO1, AP3S1, GOLGA4, ARF1, GALC, CLCN3, RER1, YKT6, VPS4B, RAB2A, COPB1, ERGIC3, KIF1B, TSG101, ATP6V1H, SEC22B, ZW10
REACTOME_PD_1_SIGNALING	0.001	1.91	PTPN6, PDCD1, HLA-DQA1, HLA-DQA2, CD3G, HLA-DRB1, HLA-DQB1, PTPN11, HLA-DRB5, CD4, HLA-DPA1, CD3D, CD274
HALLMARK_INFLAMMATORY_RESPONSE	0.001	1.91	CCL22, EBI3, TNFRSF1B, NFKBIA, CD82, NAMPT, NFKB1, EMP3, MYC, RHOG, IL18R1, ICOSLG, SELENOS, TNFAIP6, SEMA4D, CDKN1A, CCL20, RGS16, IL1R1, GCH1, CXCR6, LTA, IL4R, SLC7A1, CYBB, ADRM1, AHR, RAF1, GNA15, IL2RB, RELA, IL18RAP
REACTOME_TCR_SIGNALING	0.001	1.83	ID2, ID3, KLF10, UBE2D3, SKIL, PMEPA1, CTNNB1, SLC20A1, NCOR2, JUNB, TGFB1, PPP1CA, BCAR3, HIPK2, IFNGR2, CDH1, SPTBN1, FKBP1A, WWTR1, XIAP
HALLMARK_TNFA_SIGNALING_VIA_NFKB	0	2.51	REL, DUSP4, NINJ1, NR4A1, TRAF1, MAP2K3, NFKBIA, ID2, NFKB2, RELB, TNFAIP3, NAMPT, GADD45B, BIRC3, CD83, TNIP2, SDC4, FOSL2, NFKBIE, KDM6B, CD80, GADD45A, ZC3H12A, JUN, NFKB1, STAT5A, MYC, KLF10, GEM, LITAF, DUSP5, BTG3, NR4A3, ICOSLG, TRIB1, DUSP2, ZFP36, BCL2A1, TNFAIP6, CDKN1A, CCL20, PHLDA2, PMEPA1, PHLDA1, PANX1, GCH1, CCND1, MAFF, SOD2, CFLAR, PLAU, SQSTM1, HES1, TNIP1, G0S2, PER1, IL23A, NFAT5, TUBB2A, NFE2L2, NR4A2, CLCF1, EIF1, KYNU
HALLMARK_IL2_STAT5_SIGNALING	0	2.20	TNFRSF18, TNFRSF4, BCL2L1, IL1R2, BATF, ODC1, TRAF1, SYNGR2, TNFRSF1B, IL2RA, SNX9, CCND2, MYO1E, GADD45B, CD83, SLC1A5, TNFRSF8, RHOH, TIAM1, MYC, UCK2, IL18R1, BATF3, GPX4, PUS1, IRF4, DCPS, ARL4A, RGS16, P4HA1, PRKCH, CST7, RNH1, PLAGL1, PHLDA1, PRNP, IGF2R, PLPP1, MAFF, CTLA4, ICOS, IL1RL1, IL4R, GSTO1, UMPS, ITGAE, IKZF2, AHR, NOP2, CTSZ, ANXA4, MAP3K8, IL2RB
REACTOME_INTERLEUKIN_1_SIGNALING	0	2.11	IL1R2, NFKBIA, NFKB2, PSMA7, TNIP2, PSMA6, NFKB1, PSME2, IL1RN, PSMD8, PSMB5, UBE2N, IKBIP, IL1R1, PSMB1, PSMA5, PSME3, PSMD13, UBB, PSMC4, SQSTM1, PSMC3, PSMD7, PSMD11, PSMB7, PSMA4, PSMC5, UBA52, RBX1, MAP3K8, RELA, MAP2K1, PSMB2, PSMA1, PSMD4, PSMB3, PSMB6, PSMF1, PSMD1, TAB3, TP53, SEM1, PSMC2, PSMD14, PSMD2, IRAK2, NFKBIB, IKBKB, IL1RAP, PSMC1, PSME1, CUL1, FBXW11, TIFA, PSMA3, PSMD12, PSMB4
REACTOME_INTERLEUKIN_10_SIGNALING	0	2.11	CCL22, IL1R2, TNFRSF1B, STAT3, CD80, IL1RN, CCL20, IL1R1, JAK1, TYK2
REACTOME_INTERLEUKIN_12_SIGNALING	0	2.10	SOD1, PPIA, MIF, PSME2, STAT4, ANXA2, JAK1, TYK2, SOD2, RPLP0, LCP1, SNRPA1, VAMP7, MTAP, GSTO1, RAP1B, HNRNPDL, HNRNPA2B1, ARF1, PAK2, BOLA2B, CDC42, HSPA9, AIP, TALDO1, IL12RB1, P4HB, HNRNPF
HALLMARK_IL6_JAK_STAT3_SIGNALING	0	2.00	IL1R2, EBI3, TNFRSF1B, STAT3, HAX1, IL2RA, PTPN1, JUN, IL18R1, IL1R1, TYK2, CSF2RB, PTPN11, IL4R, TNFRSF12A, LEPR
HALLMARK_TGF_BETA_SIGNALING	0.002	1.78	ID2, ID3, KLF10, UBE2D3, SKIL, PMEPA1, CTNNB1, SLC20A1, NCOR2, JUNB, TGFB1, PPP1CA, BCAR3, HIPK2, IFNGR2, CDH1, SPTBN1, FKBP1A, WWTR1, XIAP
HALLMARK_MYC_TARGETS_V1	0	2.39	HSP90AB1, ODC1, RANBP1, NOLC1, EIF1AX, ILF2, PPIA, HNRNPC, PSMA7, LDHA, RAN, SERBP1, PSMA6, GNL3, NOP16, STARD7, VDAC1, NDUFAB1, MYC, EIF2S1, SET, EIF3D, PSMD8, SNRPG, SNRPD3, HSPE1, SSBP1, PA2G4, C1QBP, RPS6, CCT2, DDX21, HSPD1, HNRNPU, CCT7, ETF1, GOT2, APEX1, CNBP, PGK1, COX5A, SRSF3, CCT5, EIF4G2, RPS5, SLC25A3, RSL1D1, NHP2, TRIM28, EIF4E, PSMC4, PTGES3, RPLP0, EIF4H, SRSF7, YWHAQ, NME1, SYNCRIP, PSMD7, DDX18, EEF1B2, PHB2, AIMP2, HNRNPA1, RPL6, TUFM, SNRPA1, KPNA2, PRDX3, PSMA4, PABPC4, SRPK1, PRDX4, HNRNPA3, COPS5, RNPS1, PCBP1, SNRPD2, IFRD1, PPM1G, SNRPD1, RPS2, SNRPB2, SRM, PRPF31, SF3A1, CSTF2, ERH, CDK4, LSM2, NAP1L1, HNRNPA2B1, SRSF2, TXNL4A, CCT3, PSMB2, NPM1, PSMA1, AP3S1, UBA2, PSMB3, YWHAE, ACP1, DUT, UBE2L3, CTPS1, KPNB1, PSMD1, TOMM70, XRCC6, GSPT1, CAD, CANX, XPO1, PABPC1, MCM4, LSM7, TFDP1, MRPS18B, MRPL23, BUB3, RUVBL2, NOP56, RFC4, PSMD14, EIF4A1, SRSF1, SMARCC1, GLO1
HALLMARK_OXIDATIVE_PHOSPHORYLATION	0	2.32	NDUFB2, LDHA, ATP5MC3, VDAC1, UQCRQ, NDUFAB1, NDUFB7, ATP5MC1, COX17, CYCS, COX6C, ATP5PF, ATP5F1B, TIMM50, GPX4, NDUFB6, CS, ATP5MF, COX7C, NDUFA6, TIMM8B, NDUFV2, DLD, HTRA2, NDUFA4, GOT2, ATP5F1A, COX5A, COX4I1, COX7A2, TIMM17A, PHYH, NDUFV1, SLC25A3, ACADSB, COX6A1, ATP6V0B, ATP6V1F, COX5B, ATP5PD, NDUFS6, ISCU, NDUFS8, NDUFC1, HCCS, OPA1, TOMM22, NDUFA3, NDUFA8, PHB2, MDH2, HADHA, SDHB, PRDX3, COX7B, TIMM10, TIMM13, HADHB, ETFA, UQCRFS1, DECR1, ATP5F1C, DLAT, MRPL15, MRPS12, SLC25A5, ATP5PB, SLC25A11, ATP6V0C, MDH1, CYB5A, VDAC2, OXA1L, NDUFB3, MRPL35, HSD17B10, ALAS1, FXN, NDUFS3, ATP5F1D, CASP7, SLC25A6, TOMM70, ATP1B1, NDUFB1, NDUFS1, MRPS11, AFG3L2, COX6B1, NDUFA1, ATP6V1E1, MRPL11, MRPS15, ATP5PO, HSPA9, POR, MRPL34, POLR2F, PDHB, ACADM, ECI1, TIMM9, PDHX, IDH1, OAT, NDUFB4, BAX, ATP5F1E, SURF1, COX11, RETSAT, ACAA1, ATP6V1H, UQCRC1, MTRR, MRPS22, GLUD1, AIFM1, MTRF1
REACTOME_MITOCHONDRIAL_TRANSLATION	0	2.25	MRPS5, AURKAIP1, MRPL32, MRPL51, MRPL16, MRPS25, MRPL14, MRPS16, GADD45GIP1, MRPL41, MRPL44, MRPL58, MRPL20, MRPL4, MRPL21, MRPS27, MRPL10, MRPS2, MRPL52, TUFM, MRPS10, MRPL54, TSFM, MRPL50, MRPL17, MRPL3, CHCHD1, MRPL15, MRPS12, MRPS35, MRPL33, MRPL46, MRPL28, MRPS18C, OXA1L, MRPS7, MRPL48, GFM1, MRPL35, MRPL19, MRPL37, MRPL2, MRPS18A, MRPL55, MRPS28, MRPS11, MTRF1L, MRPL57, MRPS31, MRPL11, MRPS15, MRPL36, MRPL34, MRPS18B, MRPL23
HALLMARK_MYC_TARGETS_V2	0	2.12	NOLC1, GNL3, NOP16, MYC, DUSP2, HSPE1, PUS1, PA2G4, MYBBP1A, HSPD1, NIP7, WDR74, DCTPP1, PES1, PPAN, MRTO4, PPRC1, FARSA, PRMT3, DDX18, NDUFAF4, AIMP2, EXOSC5, UTP20, RRP12, NOP2, WDR43, SRM, IMP4, CDK4, TMEM97, NPM1, GRWD1, BYSL
REACTOME_GLUCONEOGENESIS	0	2.02	GAPDH, PGAM1, ENO1, TPI1, GOT2, SLC37A1, PGK1, MDH2, SLC25A11, ALDOA, MDH1, PCK2, PC, SLC25A10, SLC25A1
REACTOME_GLYCOLYSIS	0	1.99	PKM, GAPDH, PGAM1, ENO1, TPI1, SEC13, NUP50, PGK1, RANBP2, NDC1, PGM2L1, NUP62, NUP188, ALDOA, NUP98, NUP43, NUP35, PPP2R1B, POM121C, PPP2CA, GNPDA1, SEH1L, AAAS, NUP155, NUP85, NUP214, NUP58, HK1
HALLMARK_FATTY_ACID_METABOLISM	0	1.92	ODC1, SMS, ACSL4, LGALS1, LDHA, UROS, MIF, HADH, IL4I1, ADIPOR2, DLD, ADSL, APEX1, HSPH1, HSP90AA1, OSTC, HCCS, HMGCS1, CRYZ, MDH2, G0S2, HADHB, DECR1, ALDOA, RAP1GDS1, MDH1, ACSL1, NCAPH2, UBE2L6, HSD17B10, ALAD, ACSM3, NTHL1, EPHX1, ECI2, PDHB, SLC22A5, ACADM, ECI1, IDH1, GLUL, RDH11, ENO2, RETSAT, ACAA1, GCDH, UROD, ACADS, BLVRA, PSME1, METAP1, NBN
REACTOME_METABOLISM_OF_AMINO_ACIDS_AND_DERIVATIVES	0.001	1.88	SLC7A5, ODC1, SMS, RPL27, PSMA7, SLC3A2, PSMA6, NDUFAB1, AIMP1, RPL27A, PSME2, RPL3, PSMD8, RPL35, RPL37A, IL4I1, PSMB5, GCSH, RPS6, RPS17, PSMB1, PSMA5, DLD, GOT2, RPS25, SECISBP2, RPL10A, PSME3, RPS16, PSMD13, RPS5, EEF1E1, RPSA, ACADSB, PYCR1, CARNS1, CHDH, GSR, RPL23, PSMC4, PSMC3, RPLP0, PSMD7, NAGS, PSMD11, AIMP2, RPS9, RPL6, PSMB7, RPS26, RPL36AL, RPL4, PSMA4, MTAP, SERINC3, RPL8, HIBADH, PSMC5, RPL38, RPL22L1, UBA52, DLAT, KYNU, IDO1, MCCC2, PXMP2, RPS2, GLS, SRM, SRR, SLC25A44, HSD17B10, RPL21, PSMB2, PSMA1, RPS18, TST, PSMD4, CARNMT1, PSMB3, PSMB6, RPS19, SDSL, RPL18A, OAZ2, BCKDK, PSMF1, PSMD1, RPL9, ALDH18A1, RPLP2, PSPH, GLUD2, SEM1, ENOPH1, CGA, SLC25A10, BCAT2, PDHB, ASS1, ARG2, PDHX, OAT, TXN2, GLUL, PSMC2, PSMD14, RPL15, PSMD2, SEPHS2, RPS11, GCDH, RPL26L1, MTRR, PSMC1, RPL36, RPL7, GLUD1, PSME1, FAH, ALDH6A1, PSMA3, APIP, PSMD12, PSMB4
HALLMARK_G2M_CHECKPOINT	0.014	1.51	SLC7A5, ODC1, NOLC1, CKS2, MYC, SFPQ, TOP1, UCK2, DR1, EWSR1, ILF3, PRPF4B, HSPA8, NUP50, HNRNPU, SQLE, ATF5, JPT1, E2F4, MNAT1, CCND1, NCL, PRMT5, SLC7A1, SYNCRIP, TMPO, KPNA2, RBM14, DTYMK, SLC38A1, RPA2, NUP98, TNPO2, SNRPD1, CDK4, LIG3, MCM3, TGFB1, FOXN3, CUL5, SRSF2, PBK, TRAIP, RAD21, HIF1A, KPNB1, KIF5B, BCL3, CUL3, GSPT1, PLK1, TPX2, TLE3, UBE2C, XPO1, SAP30, CENPF, PTTG1, E2F3, CCNB2, TFDP1, CKS1B, TOP2A, BUB3, BIRC5, CDKN3, SRSF1, SMARCC1, HMMR, CHMP1A, ARID4A, SMC4, RAD23B, EZH2, CBX1, CUL1
HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION	0.046	1.35	ID2, TNFAIP3, GADD45B, LGALS1, SDC4, GADD45A, JUN, EMP3, GEM, SERPINE2, PMEPA1, TPM4, PDLIM4, TGFBR3, TNFRSF12A, VIM, BASP1, FLNA, EFEMP2, SFRP4, COL1A2, PPIB, POSTN, COL6A3, TGFB1, CALU, FBLN5, CAPG, FN1, FBLN1, HTRA1, APLP1, COL4A1, COL12A1, COL5A1, RHOB, THBS2, COL8A2, ADAM12, FAS, ENO2, CD59, COL3A1, LAMC1, COL6A2, SGCB, CD44, TPM2, PMP22

Abbreviations: eTregs, effector regulatory T cells; GSEA, gene set enrichment analysis; NES, normalized enrichment score.

TNFRSF9 ⁺ IL2RA ^high eTregs were increased in primary TNBC (mean [min–max]% = 21 [3–38]% of Tregs), compared with normal tissues (P < 0.01) and hormone receptor–positive primary breast cancers (HR⁺; P < 0.05; Fig. 1G). Moreover, TNFRSF9⁺IL2RA^high eTregs displayed higher IL2RA levels in primary TNBC (P < 0.01). Importantly, ratios of cytotoxic effector αβCD8 T cells (Supplementary Fig. S4A) to TNFRSF9⁺IL2RA^high eTregs were decreased in primary TNBC compared with HR⁺ tumors (P < 0.05). Given the clinical issues for patients with TNBC, we decided to conduct this study on the latter indication. We evaluated whether the ratio of CD8A to TNFRSF9 expression is associated with TNBC outcome using bulk RNA-seq data from two independent cohorts (TCGA and GSE177043). Primary TNBC samples were stratified into two groups based on the CD8A/TNFRSF9 ratio (Supplementary Fig. S4B). A low CD8A/TNFRSF9 ratio was associated with decreased survival in TNBC (P < 0.05; Fig. 1H), independently of the TNM stage (data not shown). CD8A/FOXP3 ratio as well as levels of TNFRSF9, FOXP3, and CD8A were not associated with survival in TNBC (Supplementary Fig. S4C).

Altogether, these data suggest that reversing the ratio of cytotoxic effector αβCD8 T cells to 4-1BB⁺ CD25^high eTregs by depletion of intratumoral eTregs, rather than blocking their recruitment from the periphery or induction from conventional αβCD4 T cells, would represent an attractive therapeutic strategy in patients with TNBC.

Accumulation of TNFRSF9⁺IL2RA^high eTregs is associated with an immunosuppressive TNBC microenvironment

Because a low CD8 T-cell to eTreg ratio was associated with decreased survival in patients with untreated TNBC, we sought to explore the immune microenvironment associated with elevated tumor infiltration by TNFRSF9⁺IL2RA^high eTreg. To this end, two groups of TNBC samples were defined based on the ratio of αβCD8 T cells to TNFRSF9⁺IL2RA^high eTregs [below and above geometric mean = “low-ratio TNBC” (N = 6) and “high-ratio TNBC” (N = 4), respectively; Fig. 2A). Groups of TNBC exhibited similar overall infiltration by immune cells or T cells (Supplementary Fig. S5A). The relative abundance of 35 immune populations reflecting the tumor immune microenvironment was used to discriminate groups of TNBC using BGA (Supplementary Table S7). Low-ratio TNBC displayed a homogeneous immune composition, as these were grouped together on the left axis of the BGA (Fig. 2A). In contrast, BGA showed that high-ratio TNBC differed markedly in their immune composition compared with low-ratio TNBC.

Figure 2.

Accumulation of TNFRSF9⁺IL2RA^high eTregs is associated with an immunosuppressive TNBC microenvironment. A, Left, two groups of untreated primary TNBC samples defined by the geometric mean of αβCD8 T-cell counts/TNFRSF9⁺IL2RA^high eTreg count ratio. Each single blue or red dot indicates primary TNBC with a high or low ratio, respectively. The immune composition of primary TNBC with high or low ratio was compared by BGA (Supplementary Fig. S5A). The BGA axis shows projection of samples. Each sample origin is linked to its own group. The distance between each sample origin indicates how much the samples differ in their immune composition (Supplementary Table S7). Right, pie charts illustrating the immune composition of primary TNBC with high or low ratio. Cell types annotated in blue or red and enriched in primary TNBC with high or low ratio, respectively. cDC, conventional dendritic cell; NK, natural killer cell; Tn:cm, naive and central memory T cell; Tearly act, early activated T cell; Tex, exhausted T cell; Tprlf; proliferating T cell; Teff, cytotoxic effector T cell; Tfh, follicular helper T cell. B, Top, UMAP visualizations of 5,463 myeloid cells and 29,108 T cells from normal breast tissues (N = 6) and primary TNBC (N = 10; Supplementary Figs. S1C, S4A, and S5B). Bottom, violin plots showing expression of genes or signature scores by indicated cell types (Supplementary Tables S2 and S3). Cells colored in blue or red are enriched in TNBC with a high or low ratio, respectively; naïve and exhausted T signature scores were queried from Andreatta and colleagues (30); effector αβCD8 T expression score: KRLD1⁺NKG7⁺PRF1⁺GZMB⁺GNLY⁺FGFBP2⁺; box plots are expressed as mean ± SEM. C, Percentage of indicated cell types. Each single green, blue, or red dot indicates normal breast tissue (NBT) and TNBC with high or low ratio, respectively. Wilcoxon–Mann–Whitney test; mean ± min/max. D, Circos plots showing ligand–receptor interactions predicted between indicated cell types. Width of arrow lines and tips indicates ligand and receptor relative expression, respectively; genes annotated in red are involved in immunosuppressive interactions. *, P < 0.05; **, P < 0.01.

Tregs have been shown to inhibit type-2 conventional dendritic cells (cDC2), which prime αβCD4 T-cell antitumor responses and are associated with good prognosis in breast cancer (27, 28). Consistently, we found that cDC2 were enriched in high-ratio TNBC compared with normal tissues and low-ratio TNBC (P < 0.01; Fig. 2B–C; Supplementary Fig. S5B). Oppositely, APOE⁺ were enriched in low-ratio TNBC compared with normal tissues (P < 0.01) and high-ratio TNBC (P < 0.05). These macrophages coexpressed genes related to M2-like macrophages (CD163⁺, CD204⁺, and TREM2⁺) and lipid metabolism (LIPA⁺, APOE⁺, APOC1⁺, and PSAP⁺), thus exhibiting hallmarks of immunosuppressive tumor-associated macrophages (TAM) previously associated with poor prognosis in TNBC (Supplementary Fig. S5B; ref. 29). Tregs and TAMs are well-known mediators of T-cell exhaustion in cancer. Consistently, αβCD4 T and αβCD8 T cells expressing hallmarks of T-cell exhaustion (e.g., TOX⁺, PDCD1^high, and HAVCR2⁺; ref. 30) were markedly enriched in low-ratio TNBC compared with normal tissues and high-ratio TNBC (P < 0.01; Fig. 2B and C). These terminally differentiated T-cell populations comprised antigen-experienced exhausted/proliferating αβCD8 T cells and CXCL13⁺CXCR5⁻PDCD1^high follicular helper αβCD4 T cells (αβCD4 Tfh; Supplementary Fig. S1C). This is consistent with previous studies showing that intratumoral CXCL13⁺ CXCR5⁻ αβCD4 Tfh differentiate in response to Treg accumulation and subsequent exposure to TGFβ1 and IL2 deprivation (31). Besides, αβCD4 Tfh promote B-cell differentiation into long-lived plasma B cell. In line, plasma B cells were enriched in low-ratio TNBC, but not in high-ratio TNBC, compared with normal tissues (P < 0.05).

To extract additional information from the immune alterations identified by BGA, we defined potential cell–cell interactions in low-ratio TNBC using iTALK. We focused on enriched ligand–receptor interactions between TNFRSF9⁺IL2RA^high eTregs, APOE⁺ M2 macrophages, exhausted/proliferating αβCD8 T cells, and monocytes (Fig. 2D). Several predicted ligand–receptor interactions were found to be antagonists of the antitumor functions of monocytes and antigen-experienced αβCD8 T cells, namely, (i) PDCD1⁺ antigen-experienced αβCD8 T cells interacting with CD274 (PD-L1)⁺/CD273 (PD-L2)⁺APOE⁺ M2 macrophages, (ii) HAVCR2 (TIM3)⁺ antigen-experienced αβCD8 T cells and HAVCR2⁺ monocytes interacting with LGALS9 (galectin-9)⁺APOE⁺ M2 macrophages, (iii) CD80⁺/CD86⁺ monocytes interacting with CTLA4⁺TNFRSF9⁺IL2RA^high eTregs, and (iv) IL1β⁺ monocytes interacting with IL1R2⁺TNFRSF9⁺IL2RA^high eTregs. Other predicted ligand–receptor interactions could promote eTreg-suppressive functions, such as ICOSLG (ICOS-ligand)⁺APOE⁺ M2 macrophages interacting with ICOS⁺TNFRSF9⁺IL2RA^high eTregs.

Altogether, these data indicate that the accumulation of bona fide TNFRSF9⁺IL2RA^high eTregs is a multifaceted hallmark of immune alterations in the primary TNBC microenvironment.

CD25 is the most selective surface target for intratumoral eTreg depletion at multiple tumor sites

We next aimed to determine the most suitable target for selective intratumoral eTreg depletion in patients with TNBC at various stages of the disease. Among the potential therapeutic targets for intratumoral eTreg depletion whose expression was increased during differentiation (Fig. 1A), only IL2RA, TNFRSF9, and CCR8 were restricted to eTregs in primary early breast tumors, including TNBC (Supplementary Fig. S6). We evaluated whether TNFRSF9⁺IL2RA^high eTregs are detected in primary advanced tumors (N = 5) and metastases (N = 8) of patients with untreated TNBC (Fig. 2A; Supplementary Table S1; ref. 21). TNFRSF9⁺IL2RA^high eTregs were detected at multiple tumor sites, with similar abundances of TNFRSF9⁺IL2RA^high eTregs in primary advanced TNBC and associated distant metastasis (mean = 36% of Tregs; Fig. 3A). Importantly, we observed selective CD25 mRNA and protein expression by 4-1BB⁺ CD25^high PD-1⁺ eTregs at multiple tumor sites, namely, advanced primary TNBC (P < 0.01) as well as associated iTDLNs and distant metastasis (P < 0.01; Fig. 3A and B). In contrast, CCR8 marked fewer eTregs than IL2RA (P < 0.0001) in primary TNBC (Fig. 3C; Supplementary Fig. S6A). Moreover, CCR8 was weakly expressed by eTregs in iTDNLs from patients with advanced TNBC (Fig. 3A). CCR8 appeared to be an attractive target for eTreg depletion in patients with primary colorectal cancer and some patients with NSCLC. However, eTregs-infiltrating primary head and neck cancer as well as hepatocellular cancer did not display sufficient CCR8 expression to consider the use of an anti-CCR8 mAb for eTreg depletion in these cancer indications (Fig. 3C; Supplementary Fig. S7). Additionally, TNFRSF9, but not IL2RA, was expressed by intratumoral CXCL13⁺PDCD1⁺ exhausted/proliferating antigen-specific αβCD8 T cells, which have been shown to be reinvigorated by PD-(L)1 blockade (2, 21) in all types of primary solid tumors (Fig. 3C; Supplementary Fig. S7). Taken together, these arguments guided the choice of CD25 for intratumoral eTreg depletion in patients with early, advanced or metastatic TNBC, with better expression by eTregs, and lower risk of antigen-experienced T-cell depletion at multiple tumor sites.

Figure 3.

CD25 is the most selective surface target for intratumoral eTreg depletion at multiple tumor sites. A, Left, heat map summarizing eTreg-related gene-expression levels in untreated advanced primary TNBC and associated metastasis. Color gradient indicates gene-expression levels. Circle size indicates the percentage of the cell type that expresses the gene in the indicated tumor tissue. Right, percentage of TNFRSF9⁺IL2RA^high eTregs relative to the total count of Tregs. Each single dot represents a donor. Green dots indicate samples from healthy donors. LNs, lymph nodes; TDLNs, tumor-draining lymph nodes. Wilcoxon–Mann–Whitney test; mean ± SEM. B, CD25 gene and protein expression by lymphoid cell subsets in untreated breast tumors (N = 32), including primary TNBC (N = 13) and associated metastasis (N = 8). Each single dot represents a patient. Wilcoxon–Mann–Whitney test or Wilcoxon matched-pairs signed-rank test; mean ± SEM. C, CD25, 4-1BB, CCR8, and PD-1 gene expression by eTregs and tumor antigen–specific αβCD8⁺ T cells in different types of untreated primary solid tumors (Supplementary Fig. S7). Each single dot represents a patient. Wilcoxon matched-pairs signed-rank test; mean ± SEM. CRC, colorectal cancer; HNC, head and neck cancer; HCC, hepatocellular cancer. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

TNFRSF9 ⁺ IL2RA ^high eTregs persist following anti–PD-(L)1 blockade in primary tumors and metastases of patients with TNBC

We collected public scRNA-seq data from primary TNBC sampled before and after PD-1 blockade by pembrolizumab monotherapy (αPD-1; N = 10) or after NAC plus pembrolizumab (N = 4; Supplementary Table S1; ref. 2). Batch-corrected visualization and clustering of 36,868 T cells/NK cells identified 23 cell populations (Fig. 4A). TNFRSF9⁺IL2RA^high eTregs were manually gated. We found no difference in the relative abundance of TNFRSF9⁺IL2RA^high eTregs before and after treatment (Fig. 4B). However, we observed interpatient variability in changes between sampling time points: 6/10 patients with TNBC had stable relative abundance of TNFRSF9⁺IL2RA^high eTregs after αPD-1 alone, and 3/10 patients with TNBC had increased relative abundance of TNFRSF9⁺IL2RA^high eTregs after αPD-1 alone (mean [min–max]% increase = 9.5 [4.1–14.2]% in Tregs). Additionally, 7/10 patients with TNBC had decreased αβCD8 T cell/TNFRSF9⁺IL2RA^high eTreg ratios after αPD-1 alone. We also found no significant difference in PDCD1⁺ antigen-experienced αβCD8 T cell/TNFRSF9⁺IL2RA^high eTreg ratios, indicating that αPD-1 alone does not preferentially activate αβCD8 T cells over eTregs. Indeed, TNFRSF9⁺IL2RA^high eTregs maintained eTreg hallmarks (FOXP3^high, BATF^high, REL/B^high; ref. 32), such as IL2RA levels (Fig. 4C and D). Interestingly, these transcripts were downregulated in Tregs from primary NSCLC responding to PD-1 blockade, but were upregulated in nonresponders (Supplementary Fig. S8; ref. 33). Treg functional reprogramming into IFNγ-producing FOXP3⁻ exTregs has been shown to be required for an effective response to PD-1 blockade (34). Blimp-1 has been shown to prevent the methylation of the FOXP3 locus in Tregs at sites of inflammation (35). Consistently, loss of Blimp-1 expression has been shown to destabilize Tregs, thus restoring responses to PD-1 blockade (36). Interestingly, eTregs from TNBC displayed increased expression of genes related to a stable immunosuppressive phenotype after αPD-1 alone, including the transcription factor PRDM1 (Blimp-1) along with the suppressive cytokine TGFβ1 (Fig. 4D; Supplementary Table S8).

Figure 4.

TNFRSF9 ⁺ IL2RA ^high eTregs persist following αPD-(L)1 in primary tumors and metastases of patients with TNBC. A, Batch-corrected UMAP visualization of 36,868 T cells and NK cells from primary TNBC sampled before (pre–αPD-1) and after (post–αPD-1) pembrolizumab monotherapy (N = 10), or after neoadjuvant chemotherapy plus pembrolizumab (post-NAC:αPD-1; N = 4). Left, UMAP colored by sampling time points. Right, UMAP colored by Louvain clustering or a signature score characterizing human Tregs. TNFRSF9⁺IL2RA^high eTregs, TNFRSF9 log₂ expression ≥ 0.5. B, Percentage of TNFRSF9⁺IL2RA^high eTregs relative to the total count of Tregs. Ratio of total αβCD8 T-cell counts, or subsets of αβCD8 T-cell counts, to TNFRSF9⁺IL2RA^high eTreg counts are also presented. Each single dot represents a patient. Wilcoxon matched-pairs signed-rank test. C, Heat map summarizing eTreg-related gene-expression levels in primary TNBC. D, Top, expression levels of two nonoverlapping signature scores related to eTreg-suppressive functions in Tregs (#3), eTregs (#2), and TNFRSF9⁺IL2RA^high eTregs (#1) computed per patient according to sampling time points. Signature scores were queried from the top 100 genes previously identified by trajectory analysis (Treg activation program; Supplementary Table S4) or the top 100 genes from Mijnheer and colleagues (ref. 32; inflammation-derived Treg). Mean ± SEM. Bottom, percentage of Tregs (#3) or eTregs (#2/1) expressing differentially expressed genes (Supplementary Table S8). Each single dot represents a patient; red and blue dots indicate pre–αPD-1 and post–αPD-1, respectively. Wilcoxon matched-pairs signed-rank test; *, P < 0.05; **, P < 0.01; ***, P < 0.001. E, Metastases from patients with TNBC (N = 3) sampled before (pre-NAC:αPD-L1) and after (post-NAC:αPD-L1) neoadjuvant chemotherapy plus atezolizumab. Percentages of TNFRSF9⁺IL2RA^high eTregs relative to the total count of Tregs are shown. Ratio of αβCD8 T-cell counts to TNFRSF9⁺IL2RA^high eTreg counts is presented as well. Each single dot represents a patient. F, Heat map summarizing eTreg-related gene-expression levels in metastasis. G, Jaccard indices computed for sharing of identical TCR clonotypes between intratumoral post-NAC:αPD-L1 Tregs (left), intratumoral post-NAC:αPD-L1 TNFRSF9⁺IL2RA^high eTregs (right), and cell types from paired peripheral blood (Blood) and metastases sampled at indicated time points. Each single dot represents a patient; mean ± SEM. H, Rarefaction and extrapolation analysis of the TCR repertoire of indicated cell types in metastases sampled at indicated time points. Lighter shades, 95% confidence intervals. P #, patient ID (Supplementary Table S1).

TNFRSF9 ⁺ IL2RA ^high eTregs were detected in metastases of patients with TNBC after NAC plus PD-L1 blockade by atezolizumab (N = 3; Fig. 4E and F; ref. 21). scTCR-seq analysis further revealed that posttreatment Tregs shared only a few clonotypes with other lineages of αβ T cells from the peripheral blood and tumors (mean % of shared clonotypes: 5% of Treg clonotypes, 4% of TNFRSF9⁺IL2RA^high eTreg clonotypes; Fig. 4G). We obtained similar results in pretreatment biopsies (data not shown). Only exhausted αβCD8 T cells had decreased TCR diversity after treatment, reflecting the engagement of tumor antigen–specific αβCD8 T cells (Fig. 4H).

Intratumoral 4-1BB⁺ CD25^high eTregs hamper responses to PD-1 blockade in TNBC

We identified markers restricted to eTregs or proinflammatory/cytotoxic lymphoid cells in scRNA-seq data from primary TNBC (Supplementary Fig. S9A). In the TCGA data set, low ratios of expression of proinflammatory cytotoxic markers to eTreg markers were associated with decreased survival in patients with TNBC (Supplementary Fig. S9B). We evaluated whether these ratios are associated with responses to αPD-1 in patients with TNBC using microarray data from the I-SPY-2 trial (37). Primary tumors were sampled before patients received a combination of NAC (paclitaxel) and αPD-1 (pembrolizumab). We observed lower proinflammatory cytotoxic/eTreg marker ratios in patients who did not achieve pathologic complete response (no pCR, N = 10) than in patients who did (pCR, N = 19; Supplementary Fig. S9C–S9D). The most significant differences were observed in ratios of GZMB, GNLY, and IFNG expression to IL2RA expression (P < 0.01). Importantly, we did not observe such differences in patients treated with paclitaxel alone. Altogether, these data suggest that accumulation of CD25^high eTregs is associated with human TNBC resistance to αPD-1.

We aimed to determine whether Treg depletion could improve the clinical outcome of patients with TNBC resistant to αPD-1. ALD2510 is a non-IL2 blocking, Fc-optimized, anti-CD25 mAb developed for selective depletion of Tregs in patients with cancer (38). In vitro, human Fc-optimized, non-IL2 blocking, ALD2510 showed increased capacity to promote NK cell–mediated depletion of CD25⁺ cells compared with human non-Fc–optimized, non-IL2 blocking, anti-CD25 mAbs (Supplementary Fig. S10A). In addition, the administration of human ALD2510 alone increased the proportion of ADCC-competent NK cells (CD16⁺ or CD16–2⁺) in primary tumors from immunocompetent, human CD25 knock-in, syngeneic colorectal cancer model (MC38, N = 4; Supplementary Fig. S10B and S10C). We evaluated the activity of a murine surrogate of ALD2510 (αCD25NIB) in combination with αPD-1 (mIgG2aκ, clone RMP1-14) in a syngeneic TNBC model (Fig. 5A). Six days after subcutaneous implantation of EMT-6 cells, groups of mice (N = 14/group) received either vehicle (PBS), αCD25NIB, αPD-1, or αCD25NIB plus αPD-1. On day 7 after treatment, αCD25NIB monotherapy or combined with αPD-1 had effectively depleted intratumoral FOXP3⁺ Tregs and FOXP3⁺ 4–1BB⁺ eTregs (P = 0.05; Fig. 5B and C). Importantly, with respect to the maintenance of self-tolerance, peripheral blood Tregs were less sensitive to αCD25NIB-mediated depletion (mean % of Tregs depleted by αCD25NIB alone: 58% in tumors versus 37% in the periphery blood). Consistently, in immunocompetent, human CD25 knock-in, syngeneic colorectal cancer model (MC38, N = 4), human αCD25NIB alone preferentially depleted intratumoral Tregs compared with splenic Tregs (mean % of Tregs depleted by αCD25NIB alone: 89% in tumors versus 37% in the spleen; Supplementary Fig. S10D). The activation status of splenic conventional T cells was weakly affected by human αCD25NIB alone, as shown by a slight increase in the percentage of granzyme B⁺ Ki-67⁺ αβCD4 and αβCD8 T cells in the spleen on day 7 posttreatment (Supplementary Fig. S10E). Importantly, in the syngeneic TNBC model, tumors from 2 out of 4 mice treated with αPD-1 alone displayed increased proportions of Tregs. In the peripheral blood, proportions of Tregs were increased in mice treated with αPD-1 alone versus mice treated with vehicle (P < 0.05). These data indicate that αPD-1 activates subsets of Tregs in vivo. Importantly, only the combination of αPD-1 and αCD25NIB increased the ratio of total αβCD8 T cells, or cytotoxic effector αβCD8 T cells (granzyme B⁺ Ki-67⁺), to Tregs in tumors and in the peripheral blood (Fig. 5C). In human primary TNBC, scTCR-seq analysis revealed that a significant proportion of intratumoral cytotoxic effector αβCD8 T cells originates from the recruitment of their counterparts from the periphery, as opposed to exhausted αβCD8 T cells expressing hallmarks of tissue residency (Fig. 5D; Supplementary Fig. S4A). In the syngeneic TNBC model, the combination of αPD-1 and αCD25NIB, but not αPD-1 alone nor αCD25NIB alone, increased the proportion of cytotoxic effector αβCD8 T cells in primary tumors and in the peripheral blood (Fig. 5E). Interestingly, we found a positive correlation (R > 0.7) between the proportion of peripheral blood and tumor cytotoxic effector αβCD8 T cells in mice treated with αCD25NIB alone or combined with αPD-1 (Fig. 5E), but not in mice receiving vehicle or αPD-1 alone (data not shown). Altogether, these data indicate that αCD25NIB-mediated Treg depletion in combination with αPD1 favors the recruitment of cytotoxic effector αβCD8 T cells from the periphery to the tumor bed.

Figure 5.

Intratumoral 4-1BB⁺ CD25^high eTregs hamper responses to PD-1 blockade in TNBC. A, Schematic of experimental tumor models. On day −6, BALB/c mice received subcutaneous (s.c.) flank injection of tumor cells (5 × 10⁵ EMT6). On day 0, mice were randomized to a treatment arm once tumors reached 80–120 mm³. Tumor-bearing mice were intraperitoneally injected with either vehicle (PBS; day 0), murine non-IL2-blocking anti-CD25 antibodies (αCD25NIB; day 0), murine anti–PD-1 antibodies (αPD-1; days 1, 4, 8, and 11), or a combination of αCD25NIB + αPD-1, at indicated doses. On day 7, four animals per group of mice were sacrificed for flow cytometry analysis of isolated tumor-infiltrating lymphocytes and peripheral blood mononuclear cells. On day 39, naïve control and surviving mice from αCD25NIB and αCD25NIB + αPD-1 groups were rechallenged with tumor cells in the opposite flank (5 × 10⁵ EMT6 s.c.). On day 56, mice were sacrificed once tumors reached 2,500 mm³. Tumor growth and survival were monitored until the experimental endpoints. B, Representative overlaid histograms comparing CD25 and ICOS expression levels by intratumoral 4-1BB⁻ (light red) and 4-1BB⁺ (red) Tregs in mice that received vehicle. C, Percentage of Tregs and 4-1BB⁺ eTregs relative to the overall number of immune cells, according to treatment arms. Ratio of αβCD8 T-cell counts, or granzyme B⁺ Ki-67⁺ αβCD8 T-cell counts, to Treg counts is presented as well. Each single dot represents a mouse. Wilcoxon–Mann–Whitney test; mean ± SEM. D, Jaccard indices computed for sharing of identical TCR clonotypes between intratumoral effector (left) or exhausted αβCD8 T cells and indicated cell types from paired human peripheral blood (Blood) and primary TNBC (PBT) samples (N = 4; Supplementary Fig. S3A; right). Each single dot represents a patient. Mean ± SEM. Tn:cm, naive and central memory T cell; Tem, effector memory T cell; Teff, cytotoxic effector T cell; Tex:prlf, exhausted/proliferating T cell. E, Left, percentage of granzyme B⁺ Ki-67⁺ αβCD8 T cells relative to the overall number αβCD8 T cells in the tumor and peripheral blood of mice, according to treatment arms. Each single dot represents a mouse. Wilcoxon–Mann–Whitney test; mean ± SEM. Right, correlation between the proportion of granzyme B⁺ Ki-67⁺ αβCD8 T cells in paired peripheral blood and tumor samples from mice treated with αCD25NIB alone or combined with αPD-1. Correlation was analyzed by linear regression R (R = 0.74). F, Tumor growth curves of mice according to treatment arms (Supplementary Fig. S11A). Wilcoxon–Mann–Whitney test; mean ± SEM. G, Kaplan–Meier estimates of overall survival according to treatment arms. Log-rank test. H, Tumor growth curves of naïve control and surviving mice rechallenged with tumor cells in the opposite flank (Supplementary Fig. S11B). Wilcoxon–Mann–Whitney test; mean ± SEM. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

αPD-1 alone eradicated established tumors in 30% of the mice and lacked significant therapeutic activity (Fig. 5F; Supplementary Fig. S11A). In contrast, Treg depletion by αCD25NIB alone eradicated 70% of established tumors (P < 0.0001). Increasing effector αβCD8 T cell/Treg ratios by αPD-1 plus αCD25NIB further eradicated 90% of established tumors (P < 0.01). αCD25NIB alone and combined with αPD-1 extended survival of mice compared with vehicle (P < 0.001) and αPD-1 alone (P < 0.05; Fig. 5G). On day 39, naïve control and surviving mice from αCD25NIB and αCD25NIB plus αPD-1 groups were rechallenged with tumor cells in the opposite left flank (Fig. 5H; Supplementary Fig. S11B and S11C). αCD25NIB alone and combined with αPD-1 prevented tumor engraftment at opposite flanks in 12% and 22% of mice, respectively (P < 0.05). Importantly, αCD25NIB alone and combined with αPD-1 promoted durable tumor growth control, without apparent adverse effects (P < 0.001). Altogether, these data show that combination of αCD25NIB plus PD-1 blockade promotes systemic antitumor immunity by increasing the ratio of cytotoxic effector αβCD8 T cells to eTregs in a syngeneic TNBC model.

Discussion

Recent single-cell studies have provided meaningful insight into the biological behavior of scarce, yet crucial, subsets of antigen-experienced αβCD8 T cells in TNBC treated with αPD-1 (2, 21). However, Tregs are either disregarded or analyzed as a single population. Therefore, Treg subsets remained to be analyzed in depth in TNBC to (i) clarify the counterintuitive prognostic value of Tregs, and (ii) explore a potential role of Tregs in TNBC resistance to PD-1 blockade. We hypothesized that single-cell technologies could provide such information, and reanalyzed public scRNA-seq data sets. We found that TNBC is associated with an accumulation of CD25^high Tregs harboring canonical markers of eTregs (FOXP3^high BATF⁺ 4-1BB⁺) conserved across solid tumors. This is consistent with pioneering work reporting an accumulation of highly suppressive CD25^high Tregs in TNBC compared with other breast cancer subtypes (39). However, the accumulation of CD25^high eTregs may reflect T cell-permissive tumors, as TNBC exhibits higher immunogenicity than other breast cancer subtypes (40), and Tregs—as sensors of inflammation—coinfiltrate tumors with αβCD8 T cells (41). Here, we evidenced that an imbalance between cytotoxic effector αβCD8 T cells and CD25^high eTregs is associated with decreased survival in TNBC, indicating that accumulation of eTregs indeed reflects an immunosuppressive microenvironment in TNBC.

Krummel and colleagues (42) recently introduced the concept of “reactive immunity archetypes,” suggesting that several immune alterations compose a unique functional network shaping tumor responses to immune-checkpoint inhibitors. Here, we showed that the immune microenvironment of primary TNBC enriched in CD25^high eTregs displays features previously associated with resistance to αPD-1, namely, increased abundance of APOE⁺ TAMs and decreased abundance of cDC2 (29, 43). In-depth characterization of eTregs and cell–cell interaction analysis further suggested that the immune composition of TNBC enriched in CD25^high eTregs is shaped by suppressive cross-talk detrimental to αPD-1 efficacy, including the release of active TGFβ1 via GARP, the potentiation of CD25^high eTregs by APOE⁺ TAMs (e.g., ICOS–ICOSL axis) and the deprivation of proinflammatory IL1β and IL2 via IL1R2 and CD25, respectively. Importantly, PD-1⁺ T cells known to expand following αPD-1 in cancer patients, namely, differentiated αβCD4 Tfh and antigen-experienced αβCD8 T cells (2, 44) were more abundant in primary TNBC enriched in CD25^high eTregs, suggesting that combination therapy based on eTreg depletion may further promote PD-1⁺ T cell reinvigoration. Importantly, we have extended previous work conducted in solid tumors for which PD-1 blockade was approved much earlier than for TNBC (melanoma, NSCLC), by demonstrating that (i) PD-1 is predominantly expressed by CD25^high eTregs in primary TNBC, as evidenced by mass cytometry; (ii) αPD-1 does not preferentially activate antigen-experienced αβCD8 T cells over eTregs in primary TNBC, as demonstrated by scRNA-seq and experimental models; and (iii) TNBC resistant to αPD-1 display a further imbalance between cytotoxic effector lymphoid cells and CD25^high eTregs in primary TNBC, as supported by scRNA-seq and microarray. Thus, we expect that depletion of intratumoral CD25^high eTregs may reprogram various facets of resistance to PD-1 blockade in patients with TNBC, notably the balance between cytotoxic effector αβCD8 T cells and CD25^high eTregs. The mechanisms of action of αCD25NIB alone or in combination with PD-1 blockade identified in mice will be subject to confirmation by scRNA-seq monitoring in early-phase trials.

Exogeneous Ag-specific/autoreactive Tregs residing in peripheral tissues, such as the skin, exert a pleiotropic role in the maintenance of tissue homeostasis by promoting immune tolerance to commensal bacteria, as well as tissue repair and regeneration (45). The elevated abundance of TNFRSF9⁺IL2RA^high eTregs in normal skin tissue thus warned of potential immune-related skin toxicity of our αCD25NIB. As in invaded skin tissues from TNBC patients, normal skin-resident TNFRSF9⁺ eTregs displayed increased IL2RA expression compared with TNFRSF9⁻ Tregs (Supplementary Fig. S11D). Yet, no drug-induced skin toxicity was observed in mice treated with αCD25NIB as monotherapy or combined with αPD-1. This suggests that preferential depletion of CD25^high eTregs might not be sufficient to break immune tolerance in peripheral tissues in vivo. In line, no rash or cutaneous toxicity was observed upon αCD25NIB administration at doses up to 100 mg/kg in cynomolgus monkeys (data not shown). As in (46), further characterization of eTregs in peripheral tissues from mice treated with αCD25NIB therapy would be required before considering clinical translation.

Current Fc-optimized mAbs aim at depleting intratumoral Tregs and target several eTreg-related molecules (10). However, there is no consensus on the most selective target for intratumoral Treg depletion. We showed that CD25, 4-1BB, and CCR8 are the most selective surface markers of eTregs in primary TNBC. In contrast, we raise a note of caution about the risk of potential on-target off-Treg effects on conventional lymphoid cells when targeting other eTreg-related molecules, including TIGIT and ICOS. PD-(L)1 blockade restores the antitumor function of exhausted antigen-experienced CXCL13⁺ PD-1⁺ αβCD8 T cells in patients with TNBC (2, 21). Consistent with preclinical studies showing that 4-1BB agonists synergize with αPD-1 (47), we found that a subset of activated PD-1⁺ antigen-experienced αβCD8 T cells expresses 4-1BB in solid tumors, including TNBC, making it difficult to deplete 4-1BB⁺ eTregs while sparing antigen-experienced T cells. In addition, eTregs from primary colorectal cancer, but not primary TNBC or metastasis, highly expressed CCR8. CD25 was expressed by all Treg subsets, while only CD25^high eTregs expressed 4-1BB and CCR8. The selective deletion of BATF⁺ eTregs, as well as the use of Fc-optimized anti-CCR8/-4-1BB mAbs, has shown preclinical antitumor efficacy (46, 48, 49). Here, we showed that CD25^high eTregs originate in part from the differentiation of activated bona fide CD25⁺ CCR8⁻ 4-1BB⁻ Tregs in primary TNBC and metastasis, as demonstrated by trajectory analysis and TCR clonotype analysis. Therefore, to tailor Treg-depleting strategies to the tumor context, we identified CD25 as the most selective surface marker for complete eTreg depletion in human primary TNBC and metastasis.

Pioneering work from Vargas and colleagues (18) showed that Fc-engineering of anti-CD25 mAbs for higher binding affinity to activating FcγRs improves antibody-dependent cell-mediated cytotoxicity/phagocytosis of intratumoral Tregs, and synergizes with PD-1 blockade (18). However, Fc-optimized anti-CD25 mAbs still hamper IL2 signaling on effector lymphoid cells upregulating CD25 upon activation. IL2 signaling has been shown to be required for the effective reinvigoration of antigen-experienced αβCD8 T cells by αPD-1, and therapeutic strategies are being developed to improve responses to PD-1 blockade by selectively delivering IL2 signaling to antigen-experienced αβCD8 T cells in tumors (19). Therefore, we and others have developed next-generation Fc-optimized anti-CD25 mAbs that do not block IL2 signaling on conventional lymphoid cells (αCD25NIB; refs. 38, 50). We showed that a combination of αCD25NIB and αPD-1 promotes systemic antitumor immunity and durable tumor growth control in a syngeneic TNBC model by increasing the ratio of cytotoxic effector αβCD8 T cells to CD25^high eTregs, with limited effect on peripheral Tregs and good tolerance in vivo. This increased ratio involves the recruitment of proliferating, cytotoxic effector CD8⁺ T cells from the peripheral blood to the tumor bed. Outside the present work, ALD2510 showed improved preclinical antitumor efficacy in animal models of cancer compared with IL2-blocking anti-CD25 mAbs (38). Overall, this study supports the involvement of CD25^high eTregs in TNBC resistance to PD-1 blockade and prompts clinical evaluation of αCD25NIB combination therapy to improve responses in patients with TNBC.

Supplementary Material

Supplementary_Materials_and_Methods_Figures_and_Figure_legends

Supplementary_Materials_and_Methods_Figures_and_Figure_legends

File_S1

BGA R script

Table S1

Clinical characteristics of patients with breast cancer

Table S2

T cell and NK cell subpopulations gene features, related to Figure 1

Table S3

cDC_monocyte_macrophage subpopulations gene features, related to Figure 1

Table S4

Inferred pseudotime gene features, related to Figure 1

Table S5

Differentially expressed genes between intratumoral TNFRSF9+ eTregs versus TNFRSF9- eTregs, related to Figure 1

Table S6

Mass cytometry panels, related to Figure 1

Table S7

Percentage of immune cell types in primary TNBC, related to Figure 2

Table S8

Differentially expressed genes between pre-αPD-1 Tregs versus post-αPD-1 Tregs, related to Figure 4

Authors' Disclosures

S. Fattori reports grants and nonfinancial support from the French Ministry of Research, Institut National du Cancer, Site de Recherche Intégrée en Cancérologie de Marseille, Cancéropôle Provence-Alpes-Côte d'Azur, Fondation de France, and Fondation pour la Recherche Médicale during the conduct of the study; in addition, S. Fattori reports nonfinancial support from Alderaan Biotechnology outside the submitted work. J. Houacine reports being an employee of Alderaan Biotechnology. P. Rochigneux reports grants from Novartis during the conduct of the study and grants from MSD, AstraZeneca, BMS, SANOFI, and nonfinancial support from Pfizer and Viatris outside the submitted work. A. Foussat reports a patent for WO 2022106665 pending to Alderaan Biotechnology, a patent for WO2022106663 pending to Alderaan Biotechnology, and a patent for WO 2020 234399 pending to Alderaan Biotechnology. A. Chrétien reports grants from Alderaan during the conduct of the study. D. Olive reports grants and personal fees from Alderaan Biotechnology during the conduct of the study; grants and personal fees from Imcheck Therapeutics outside the submitted work; in addition, D. Olive has a patent for CD25 Mabs pending. No disclosures were reported by the other authors.

Authors' Contributions

S. Fattori: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A. Le Roy: Data curation, formal analysis, validation, investigation, visualization, methodology. J. Houacine: Validation, methodology. L. Robert: Investigation. R. Abes: Validation, methodology. L. Gorvel: Validation, methodology. S. Granjeaud: Software, methodology. M. Rouvière: Resources. A. Ben Amara: Resources. N. Boucherit: Resources. C. Tarpin: Resources. J. Pakradouni: Resources. E. Charafe-Jauffret: Resources. G. Houvenaeghel: Resources. E. Lambaudie: Resources. F. Bertucci: Resources. P. Rochigneux: Resources. A. Gonçalves: Resources. A. Foussat: Conceptualization, resources, supervision, funding acquisition, validation, methodology, writing–original draft, project administration, writing–review and editing. A.-S. Chrétien: Conceptualization, resources, software, formal analysis, supervision, funding acquisition, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing. D. Olive: Conceptualization, resources, supervision, funding acquisition, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing.