Patients

Patients in cohorts A and B were eligible for enrollment if they were at least 18 years of age and had stage I–III (clinical tumor stage T1c-3 and nodal stage N0–3, according to the primary tumor regional lymph node staging criteria of the American Joint Committee on Cancer, 7th edition) TNBC with confirmation of estrogen receptor (ER) and HER2 negativity (ER < 10% and HER2 0, 1 or 2 in the absence of amplification as determined by in situ hybridization) on a biopsy from the primary tumor in the breast; newly diagnosed, previously untreated disease; a WHO PS score62 of 0 or 1 and adequate organ functions. The TIL percentage is needed to be 5% or more. To ensure balanced enrollment based on TIL levels, each cohort included five patients with low (5–10%), five patients with intermediate (11–49%) and five patients with high (≥50%) TIL levels. Patients with concurrent ipsilateral, bilateral or multifocal primary tumors were also eligible for enrollment. For cohort C, patients had to meet the same criteria, but the nodal stage had to be N0, tumor stage T1c–T2 and TILs had to be 50% or more. The intention for cohort C was to explore the potential feasibility of chemotherapy de-escalation in patients with high TILs. As withholding adjuvant capecitabine for high-risk patients and/or escalating locoregional treatment for patients with more extensive disease was undesired, cohort C included only patients who were lymph node-negative.

Exclusion criteria included history of immunodeficiency, autoimmune disease or conditions requiring immunosuppression (>10 mg d−1 prednisone or equivalent); other immunosuppressive medications intake within 28 days of study drug administration; chronic or recurring infections; occult breast cancer; fertility preservation due to breast cancer diagnosis; active hepatitis B virus or hepatitis C virus infection; clinically overt cardiovascular disease; or previous systemic anticancer treatment.

Trial design and treatments

The BELLINI trial (Preoperative Trial for Breast Cancer With Nivolumab in Combination With Novel IO; ClinicalTrials.gov registration: NCT03815890) is a single center, nonblinded, nonrandomized, noncomparative phase 2 study designed to evaluate the feasibility and efficacy of checkpoint inhibition before regular neoadjuvant therapy or surgery in patients with primary breast cancer. Cohorts for prespecified breast cancer subgroups are opened in a sequential manner. Here we report the first three TNBC cohorts for patients who were treated with nivolumab (cohort A) or nivolumab + ipilimumab for 4 (cohort B) or 6 (cohort C) weeks. Cohort A had nivolumab monotherapy, 240 mg on day 1 (D1) and D15. Cohort B had nivolumab + ipilimumab 1 mg kg−1 on D1 and nivolumab 240 mg on D15. Cohort C had nivolumab + ipilimumab 1 mg kg−1 on D1 and D21. Regular therapy, consisting of neoadjuvant chemotherapy or primary surgery, started on D29 and onwards. Given the poor prognosis of patients with low TIL levels and the hypothesis that these women will probably not be the super-responders to ICI, patients were only eligible with TILs ≥ 5%. A threshold of 5% TILs was selected to exclude true immune-deserted tumors. Equal distribution of patients with different levels of tumor of infiltrating lymphocytes over the cohorts was ensured by inclusion of five patients with low TIL (5–10%), five patients with intermediate TIL (11–49%) and five patients with high TIL (≥50%) scores per cohort.

After cohorts A (in the protocol defined as cohort 1B) and B (in the protocol defined as cohort 2B) the protocol was amended to open cohort C (in the protocol defined as cohort 3B). Cohort C had the same inclusion criteria as cohort A and B, except that only inclusion of patients with clinically node-negative disease and with TIL levels of 50% or higher was allowed. With the amendment to open cohort C, the WOO design was changed into a true neoadjuvant design with all patients proceeding to surgery after the immunotherapy. After completing the interim analysis of cohorts A and B, an amendment was approved to use pCR as a primary end point instead of immune activation for cohort C and subsequent cohorts (see details on end points below).

Ethics statement

All patients provided written informed consent before enrollment. This investigator-initiated trial was designed by the Netherlands Cancer Institute (NKI).

The trial was conducted in accordance with the protocol, Good Clinical Practice standards and the Declaration of Helsinki. The full protocol, amendments and the informed consent form were approved by the medical ethical committee of the NKI.

End pointsCohorts A and B

The primary end point for cohorts A and B is immune activation following two cycles of neoadjuvant ICI, defined as a twofold increase in CD8+ T cells assessed via immunohistochemistry and/or an increase in IFNG gene expression. High-quality paired biopsies are necessary for the evaluability of this primary end point.

Clinical response

As a secondary end point for cohorts A and B, we evaluated the clinical response. Clinical response is defined as having a radiological and/or pathological response.

Radiological signs of response

At least a 30% decrease on MRI (PR according to RECIST v.1.1, not confirmed). The target (or index) lesion is defined as the largest enhancing lesion. In case of multifocality or multicentricity the largest mass and/or nonmass enhancement was measured in the axial–sagittal or coronal plane and defined as target/index lesion. In these cases, the total area occupied by the tumor (including all masses and nonmass enhancement) was also measured. The total tumor area was used for the RECIST measurements.

Pathological signs of response

Pathological response could be studied in biopsies from 28 patients due to the WOO design. The absence of viable tumor after 4 weeks of therapy in the post-treatment biopsy was classified as a clinical response. For patients proceeding to surgery this was defined as partial or pCR, according to the EUSOMA criteria.

Cohort C

The primary end point for cohort C is pCR, defined as no viable tumor remaining in the breast and lymph nodes (ypT0N0)63. MPR (the secondary end point) is a frequently used surrogate end point for efficacy in neoadjuvant trials evaluating immune checkpoint blockade across cancer types8,11,26. MPR was defined as ≤10% of residual viable tumor in the surgical specimen17,64,65 or no viable tumor in the breast but residual tumor cells in the lymph nodes.

All cohorts (A, B and C)

Secondary end points included feasibility, safety and radiological response. Feasibility was determined based on any treatment-related complications that led to a delay in chemotherapy or primary surgery beyond 6 weeks from the start of therapy. All patients were closely monitored for AEs for 100 days after the administration of the last study treatment, following the Common Terminology Criteria for Adverse Events (CTCAE) v.5 (ref. 66). In addition, we reported all immune-related AEs in the first year of follow-up. Radiological response was assessed according to the RECIST v.1.1 guidelines, but not confirmed.

Statistical analysis

For this exploratory, hypothesis-generating study, no formal sample size calculation was performed for efficacy because there were no data on the efficacy of neoadjuvant immunotherapy in breast cancer at the time of the design of this study. For cohorts A and B, the null hypothesis of a true immune activation in ≤30% of patients was tested against a one-sided alternative. For cohort C, design was identical with the exception of null hypothesis being pCR in ≤30% of patients tested against a one-sided alternative. For 80% power, at a one-sided significance level of 0.05, 15 patients were accrued per cohort to be evaluated in the first stage. If there were 5 or fewer responses among these 15 patients, the cohort was closed for futility. Otherwise, the cohort could be expanded with 31 additional patients, reaching a total of 46. We decided to publish after stage I, which was allowed by protocol, due to the observation that very early responses to ICI without chemotherapy are possible in TNBC, which warrants efforts to de-escalate therapy for a subset of patients, in contrast to the current therapy escalation for all patients with TNBC. The median follow-up time was obtained using a reverse Kaplan–Meier method. Analyses were performed using R67 v.4.2.1.

Pathology assessments and IHC analyses

All patients underwent baseline tumor staging, consisting of ultrasound of the breast, axilla and periclavicular region and MRI imaging of the breast. Positron emission tomography and computed tomography imaging was performed in all participants to confirm the clinical stage. Pretreatment tumor histological biopsies (four core biopsies, 14G needle) were taken for all patients and post-treatment tissue was either obtained through a biopsy (three core biopsies, 14G needle) for patients continuing neoadjuvant chemotherapy (n = 28) and the surgical specimen was used for those undergoing surgery right after the ICI study treatment (n = 3). Histopathological examination of biopsies and resection specimens was carried out by five experienced breast cancer pathologists (H.M.H., R.S., K.v.d.V., J.v.d.B. and N.K.). Resected tumors were examined in their entirety and regression of resected tumors was assessed by estimating the percentage of residual viable tumor of the macroscopically identifiable tumor bed, as identified on routine hematoxylin and eosin (H&E) staining. Formalin-fixed paraffin-embedded (FFPE) tissue sections were used for H&E staining and for immunohistochemical analysis of CD8 (C8/144B, DAKO), PD-L1 (22C3, DAKO) and PD-1 (NAT105, Roche Diagnostics). The percentage of tumor cells and TILs was assessed by pathologists trained for TIL assessment on H&E-stained slides according to the international standard from the International Immuno-Oncology Biomarker Working Group22 (see www.tilsinbreastcancer.org for all guidelines on TIL assessment in solid tumors). After a pathologist provided an initial TIL score, an ‘expert TIL score’ was generated as a consensus score from at least two out of four trained pathologists using slidescore.com for online scoring (www.slidescore.com). TIL scores for inclusion were scored on the diagnostic biopsy of the patient to allow for stratification of patients (low ≥ 5–10%, intermediate = 11–49% and high ≥ 50%).

Immunohistochemistry

IHC of the FFPE tumor samples was performed on a BenchMark Ultra autostainer (Ventana Medical Systems). The double stain was performed on a Discovery Ultra autostainer. In brief, paraffin sections were cut at 3 μm, heated at 75 °C for 28 min and deparaffinized in the instrument with EZ prep solution (Ventana Medical Systems). Heat-induced antigen retrieval was carried out using Cell Conditioning 1 (CC1, Ventana Medical Systems) for 48 min at 95 °C (PD-L1) or 64 min at 95 °C (PD-1/CD8 double). PD-L1 was detected using clone 22C3 (1:40 dilution, 1 h at room temperature, Agilent/DAKO, lot 11654144). Bound antibody was detected using the OptiView DAB Detection Kit (Ventana Medical Systems). Slides were counterstained with Hematoxylin and Bluing Reagent (Ventana Medical Systems).

For the double-staining PD-1 (Yellow) followed by CD8 (Purple), PD-1 was detected in the first sequence using clone NAT5 (Ready-to-Use, 32 min at 37 °C, Roche Diagnostics, lot 11654144). The PD-1-bound antibody was visualized using anti-mouse NP (Ventana Medical Systems, Ready-to-Use dispenser, lot K09956) for 12 min at 37 °C followed by anti-NP AP (Ventana Medical Systems, Ready-to-Use dispenser, lot J23971) for 12 min at 37 °C, followed by the Discovery Yellow detection kit (Ventana Medical Systems). In the second sequence of the double-staining procedure, CD8 was detected using clone C8/144B (1:200 dilution, 32 min at 37 °C, Agilent, lot 41527763). CD8 was visualized using anti-mouse HQ (Ventana Medical systems, Ready-to-Use dispenser, lot K20711) for 12 min at 370 °C followed by anti-HQ HRP (Ventana Medical Systems, Ready-to-Use dispenser, lot K22062) for 12 min at 37 °C, followed by the Discovery Purple Detection kit (Ventana Medical Systems). Slides were counterstained with Hematoxylin and Bluing Reagent (Ventana Medical Systems). A PANNORAMIC 1000 scanner from 3DHISTECH was used to scan the slides at a ×40 magnification.

Distance analysis between tumor and CD8+ T cells

Spatial analysis was performed on the pretreatment biopsies of all included patients. The stained slides were scanned and image analysis was performed with the HALO image analysis software from Indica Labs, v.3.4.2986.185 (cohorts A and B) and v.3.6.4134 (cohort C). Within HALO, the multiplex IHC module was used to phenotype and quantify CD8+ cells. Cell segmentation was performed by the detection of hematoxylin (detection weight of 1) and PD-1 (detection weights 0.045 for cohorts A and B; and 0.5 for cohort C) and CD8 for cohort C (detection weight of 0.5) staining, utilizing a nuclear segmentation aggressiveness of 0.045. Minimal intensity thresholds to consider a cell positive for a marker were set for hematoxylin (0), PD-1 (0.25 for cohorts A and B and 0.1 for cohort C) and CD8 (0.1) separately. Biopsies were analyzed in total, while for resection specimens the analysis was restricted to representative tumor beds as annotated by a breast cancer pathologist. The quantified levels of CD8+ and PD-1+CD8+ cells were corrected for the analyzed tissue area (cells per µm2).

Artificial intelligence tumor classifiers (Object Phenotyper, HALO AI) were developed to discriminate between tumor and nontumor cells in cohorts A and B and in cohort C. Individual cells were segmented (nuclei seg BF v.1.0.0), and the classifiers were trained by annotating single cells as tumor or nontumor. The annotations were guided by marked tumor regions on H&E-stained slides by a trained breast cancer pathologist. The classifiers were finalized with 20,000 iterations and a cross-entropy of 0.009 (cohort A and B) and >10,000 iterations and cross-entropy of 0.021 (cohort C).

Merging the results of the multiplex IHC and tumor classifier enabled the visualization of the spatial distribution of tumor and CD8+ cells (Extended Data Fig. 1b–f). Using the nearest neighborhood analysis, the average distance between the tumor and immune cells was quantified by taking the mean of the distances between every tumor cell and its nearest cell of the above-mentioned immune phenotypes in the pretreatment biopsies (Extended Data Fig. 1f). Distances from tumor cells to the nearest CD8+ T cells were taken as a measure of proximity of CD8+ T cells to the tumor.

DNA and RNA isolation

DNA and RNA were extracted from fresh-frozen, pre- and post-treatment tumor material using the AllPrep DNA/RNA kit (QIAGEN) for frozen material, following the manufacturer’s protocol, in a QIAcube (QIAGEN). Germline DNA was isolated from patient peripheral blood mononuclear cells using the DNeasy Blood & Tissue kit (QIAGEN).

Bulk RNA sequencingTotal RNA quality control

Quality and quantity of the total RNA was assessed by the 2100 BioAnalyzer using a Nano chip (Agilent). Total RNA samples having a RIN > 8 were subjected to library generation.

TruSeq stranded mRNA library generation

Strand-specific libraries were generated using the TruSeq stranded mRNA sample preparation kit (Illumina, RS-122-2101/2) according to the manufacturer's instructions (Illumina, document no. 1000000040498 v00). In brief, polyadenylated RNA from intact total RNA was purified using oligo-dT beads. Following purification, the RNA was fragmented, random primed and reverse transcribed using SuperScript II Reverse Transcriptase (Invitrogen, part no. 18064-014) with the addition of Actinomycin D. Second-strand synthesis was performed using Polymerase I and RNaseH with replacement of dTTP for dUTP. The generated cDNA fragments were 3' end adenylated and ligated to Integrated DNA Technologies (IDT) xGen UDI(10 bp)-UMI(9 bp) paired-end sequencing adaptors (Integrated DNA Technologies) and subsequently amplified by 12 cycles of PCR. The libraries were analyzed on a 2100 BioAnalyzer using a 7500 chip (Agilent), diluted and pooled equimolar into a multiplex sequencing pool.

Sequencing

The libraries were sequenced with 54 paired-end reads on a NovaSeq 6000 using S1 Reagent kit v.1.5 (100 cycles) (Illumina).

Data analysis

RNA-seq data were aligned to GRCh38 with STAR68 v.2.7.1a, with the twopassMode = ‘Basic’. FPKM were obtained with RSeQC69 v.4.0.0 FPKM_count.py and subsequently normalized to transcripts per million. Data quality was assessed with FastQC70 v.0.11.5, FastQ Screen71 v.0.14.0, the Picard CollectRnaSeqMetrics72,73 and RSeQC69 v.4.0.0 read_distribution.py and read_duplication.py and were found to be suitable for the downstream analysis. TNBCtype74 was used for the Lehmann subtype classification75. The Gseapy76 v.1.0.3 ssgsea tool with the sample_norm_method = ‘rank’ was used for gene set signature scoring. For the signature analysis, P values were significant after FDR correction (Benjamini–Hochberg) at a 10% significance level. Data were analyzed with Python77 v.3.10.5. Pandas78,79 v.2.0.0 and numpy80 v.1.22.4 were used for data handling. Matplotlib72 v.3.5.2, seaborn81 v.0.12.2 and statannotations82 v.0.5.0 were used for plotting.

Whole-exome sequencing

For each sample the amount of double-stranded DNA was quantified by using the Qubit dsDNA HS Assay kit (Invitrogen, cat. no. Q32851). A maximum amount of 2 μg double-stranded genomic DNA was fragmented by covaris AFA technology to obtain fragment sizes of 200–300 bp. Samples were purified using Agencourt AMPure XP Reagent (Beckman Coulter, cat. no. A63881) in a 2× reaction volume settings according to the manufacturer’s instructions. The fragmented DNA was quantified and qualified on a BioAnalyzer system using the DNA7500 assay kit (Agilent Technologies cat no. 5067- 1506). With a maximum input amount of 1 μg fragmented DNA, next-generation sequencing library preparation for Illumina sequencing was performed using the KAPA HTP Prep kit (KAPA Biosystems, KK8234) in combination with xGen UDI-UMI adaptors (IDT). During the library amplification step, four cycles of PCR were performed to obtain enough yield for the exome enrichment assay. All DNA libraries were quantified on a BioAnalyzer system using the DNA7500 assay kit. Exome enrichment was performed on library pools of six unique dual indexed libraries, 500 ng each, using the xGen Exome Hyb Panel v.2 (IDT, cat. no. 10005152) and xGen Hybridization Capture Core Reagents according to manufacturer’s protocol, with hybridization time adjusted to 16 h and ten cycles of PCR performed during post-capture PCR. All exome enriched library pools were quantified on a BioAnalyzer system using the DNA7500 assay kit, pooled equimolar to a final concentration of 10 nM and subjected to paired-end 100-bp sequencing on an Illumina Novaseq 6000 instrument using a NovaSeq 6000 S4 Reagent Kit v.1.5 (Illumina, 20028313), according to the manufacturer’s instructions.

Data analysis

Sequencing reads were aligned to the human reference GRCh38 (Ensemble, v.105) using BWA83 v.0.7.17. Duplicated reads were marked using Picard73 MarkDuplicates v.2.25.0, after which quality scores were recalibrated using GATK4 (ref. 84) BaseRecalibrator v.4.2.2.0. Single-nucleotide variants and short insertions and deletions (indels), were called using GATK4 (ref. 84) Mutect2 v.4.2.2.0 on the tumor samples matched with germline samples. Subsequently, variants were filtered by the PASS filter, and annotated using Ensembl Variant Effect Predictor 105. The maftools85 v.2.10.5 package was used for the analysis. Tumor mutational burden was calculated by summarizing the total number of nonsynonymous somatic mutations with a minimal variant allele frequency of 20%. Data were analyzed with Python77 v.3.10.5 and R67 v.4.1.3. Pandas78,79 v.2.0.0 was used for data handling. maftools85 v.2.10.5, Matplotlib72 v.3.5.2, seaborn81 v.0.12.2 and statannotations82 v.0.5.0 were used for plotting.

scRNA-seq and TCR sequencingPreparation of the single-cell suspension

Following biopsy or obtaining resection specimens, samples were rapidly processed for scRNA-seq. Samples from cohort A were minced on ice and frozen in 10% dimethylsulfoxide FCS at −80 °C. Within 4 weeks after freezing, samples were defrosted in 37 °C medium. Samples from cohort B were minced on ice and immediately processed for single-cell sequencing (not frozen), which did not result in a batch effect.

Samples were transferred to a tube containing 1 ml digestion medium containing collagenase P (2 mg ml−1, Thermo Fisher Scientific) and DNase 1 (10 U µl−1, Sigma) in RPMI (Thermo Fisher Scientific). Samples were incubated for 20 min at 37 °C and were pipetted up and down every 5 min for 30 s. Next, samples were filtered on a 40-µm nylon mesh (Thermo Fisher Scientific) and directly after the same volume of ice cold PBS containing 0.04% BSA was added. Following centrifugation at 300g and 4 °C for 5 min, the supernatant was removed and discarded, and the cell pellet was resuspended in red cell blood lysis buffer for 5 min at room temperature and then centrifuged again at 300g at 4 °C for 5 min. The supernatant was removed and discarded and the pellet was resuspended in PBS containing 0.04% BSA. Next, 10 μl of this cell suspension was counted using an automated cell counter (ChemoMetec NucleoCounter NC-200) to determine the concentration of live cells. The entire procedure was usually completed within 1 h and 15 min.

scRNA-seq data acquisition and preprocessing

Libraries for scRNA-seq were generated using the Chromium Single Cell 5′ library and Gel Bead & Multiplex kit from 10x Genomics. We aimed to profile 10,000 cells per library if a sufficient number of cells was retained during dissociation. All libraries were sequenced on a HiSeq4000 or NovaSeq 6000 until sufficient saturation was reached.

Data analysis

After quality control, raw sequencing reads were aligned to the human reference genome GRCh38 and processed to a matrix representing the unique molecular identifiers’ per-cell barcode per gene using Cell Ranger (10x Genomics, v.2.0). The data were analyzed with scanpy86 v.1.9.3 and Seurat87 v.3. Cellbender88 v.0.3.0 was used for eliminating technical artifacts and cells above the quality cutoff of 0.5 were filtered out. Cells with mitochondrial RNA content >0.25, the number of genes <200 or >6,000 and <400 counts were filtered out. After normalization, regression for the number of unique molecular identifiers, percentage mtRNA, sample ID, cell cycle, hypoxia, interferon content and cell stress was performed on the 2,000 most variable genes followed by principal-component analysis. Next, a Uniform Manifold Approximation and Projection (UMAP) was generated and clustering was performed at resolution of 0.2 using the 30 most informative components. Major cell types were identified based on canonical marker genes.

For T cell subclustering, the T cells were selected from the full Seurat object and the analysis described above was repeated with ten principal components based on the elbow plot and clusters were identified at a resolution of 0.6 and were annotated based on breast cancer tissue-specific marker genes89. Cells expressing markers of other cell types (immunoglobulins and hemoglobin) were filtered out. Principal-component analysis was calculated on highly variable genes with k = 30. Clustering was performed with Phenograph90 with k = 30. Cluster identification was performed based on canonical marker genes. Signature scores were calculated with sc.tl.score_genes. Groups were compared to sc.tl.rank_genes_groups, with method = ‘wilcoxon’ and use_raw = True. EnrichR91,92 was used for the pathway enrichment analysis. Activated Treg cells were defined based on the level of CD137 gene expression >0.5 in the Treg cell population. PD-1+Ki-67+CD4+ cells were defined based on the level of MKI67 gene expression >0 in the TFH cell population. Scirpy93 v.0.11.2 was used for the TCR analysis. Clonotypes were defined based on the amino acid structure. Clonality was calculated as (1 − normalized Shannon entropy). Data were analyzed with Python77 v.3.10.5. Pandas78,79 v.2.0.0 and numpy80 v.1.22.4 were used for data handling. Matplotlib72 v.3.5.2, seaborn81 v.0.12.2, sc-toolbox94 v.0.12.3 and statannotations82 v.0.5.0 were used for plotting.

ctDNA analysis

A proprietary bioinformatics tissue variant calling pipeline was used to select a set of 16 high-ranked, patient-specific, somatic, clonal single-nucleotide variants from whole-exome sequencing. The Signatera amplicon design pipeline was used to generate multiplex PCR (mPCR) primer pairs for the given set of 16 variants. For cfDNA library preparation, up to 20,000 genome equivalents of cfDNA from each plasma sample were used. The cfDNA was end-repaired, A-tailed and ligated with custom adaptors, followed by amplification (20 cycles) and purified using Ampure XP beads (Agencourt/Beckman Coulter). A proprietary mPCR methodology was used to run patient-specific assays. Sequencing was performed on these mPCR products on an Illumina HiSeq 2500 Rapid Run (50 cycles) using the Illumina Paired End v.2 kit with an average read depth of >100,000× per amplicon. All paired-end reads were merged using Pear v.0.9.8 software and mapped to the hg19 reference genome with Novoalign v.2.3.4 (http://www.novocraft.com/). Plasma samples with at least two variants with a confidence score above a predefined algorithm threshold were defined as ctDNA-positive.

Flow cytometry of fresh blood

Flow cytometry was performed as previously described95. In brief, fresh blood samples were processed and analyzed within 24 h after blood draw. Peripheral blood was collected in EDTA vacutainers (BD) and subjected to red blood cell lysis (lysis buffer, dH2O, NH4Cl, NaHCCO3 and EDTA). Cells were suspended in PBS containing 0.5% BSA and 2 mM EDTA and counted using the NucleoCounter NC-200 (Chemometec) automated cell counter. To obtain absolute white blood cell counts per ml human blood, the total amount of post-lysis cells was divided by the volume (ml) of blood obtained from the patient. For surface antigen staining, cells were first incubated with human FcR Blocking Reagent (1:100 dilution, Miltenyi) for 15 min at 4 °C and then incubated with fluorochrome-conjugated antibodies for 30 min at 4 °C. For intracellular antigen staining, cells were fixed with Fixation/Permeabilization solution 1× (Foxp3/Transcription Factor Staining Buffer Set, eBioscience) for 30 min at 4 °C and stained with fluorochrome-conjugated antibodies in Permeabilization buffer 1× (eBioscience) for 30 min at room temperature. Viability was assessed by staining with either 7AAD staining solution (1:10 dilution; eBioscience) or Zombie Red Fixable Viability kit (1:800 dilution, BioLegend). Data acquisition was performed on an LSRII SORP flow cytometer (BD Biosciences) using Diva software and data analysis was performed using FlowJo v.10.6.2. The gating strategy is displayed in Extended Data Fig. 5a.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.