CD24^hiCD27⁺ Bregs within Metastatic Lymph Nodes Promote Multidrug Resistance in Breast Cancer

Axillary lymph nodes (LN) are recognized as the primary metastatic sites in breast cancer where the existence of LN metastasis strongly indicates a high risk of tumor recurrence and poor overall survival (1). Pooled analysis of 11,955 patients with breast cancer who had received neoadjuvant therapy (NAT) confirmed that total pathologic complete response (defined as absence of invasive tumor cells both in primary lesion and LNs) had stronger prognostic value than breast pathologic complete response (absence of invasive tumor cells in the primary lesion only; ref. 2). Results from ongoing clinical trials including Keynote-522, Geparneuvo, and Impassion-031, are strongly suggesting the profound benefit of targeting the PD-1/PD-L1 pathway in patients with breast cancer with metastatic lymph nodes (mLN; refs. 3–5). Other recent studies have continued to highlight the key role of tumor-draining lymph nodes (TDLN) in generating effective antitumor immunity (6, 7), indicating that metastatic cancer cells within mLNs are able to disrupt the antitumor immunity of LNs (8). Of note, only about 50% of patients with breast cancer with mLNs receiving NAT achieved a pathologic complete response (pCR) for mLNs; while increasing the intensity of chemotherapy could improve the pCR rate for breast lesions, it could not do so for mLNs (9). This suggests a differing drug resistance mechanism between the tumor cells in primary lesions and mLNs, and raises the important issue of optimizing the strategy to improve the therapeutic efficiency of patients with breast cancer with mLNs.

Numerous studies have illuminated the dominant role of the tumor microenvironment (TME) in regulating therapeutic response (10). Over recent years, B cells have attracted increasing attention where studies have suggested diverse effects of tumor-associated B cells within the TME (11). B cells and tertiary lymphoid structures (TLS) within tumor lesions, for example, are associated with favorable prognosis and improved response to chemotherapy and immune checkpoint blockade (12). Meanwhile, several studies have also revealed that tumor-infiltrating regulatory B cells (Breg) produce IL10, IL35, and TGFβ, to induce an immunosuppressive microenvironment by promoting the induction of regulatory T cells (Treg) or inhibiting the function of the CTLs (13–15). Compared with the primary tumor, B cells within mLNs are more likely to cause tumor metastasis (16, 17). However, most previous studies suggesting this have been based on mouse models and there is still a lack of corresponding reports relating to human mLNs, where any specific relationship between Bregs and the drug resistance of tumor cells has yet to be investigated in a human context.

Here, we clarify the crosstalk between human breast cancer cells and CD24^hiCD27⁺ Bregs within mLNs. Breast cancer cells were found to directly activate CD24^hiCD27⁺ Bregs in a CD40L/PD-L1–dependent manner, and CD24^hiCD27⁺ Bregs promoted the stemness and treatment resistance of breast cancer cells via producing IL6 and TNFα, thus forming a positive feedback loop. Meanwhile, anti-PD-L1 treatment greatly attenuated the protumoral function of CD24^hiCD27⁺ Bregs and increased tumor remission in mLNs. Our findings clarify the interaction between Bregs and breast cancer cells in supporting therapeutic resistance of tumor cells in mLNs and emphasize the importance of anti-PD-1/PD-L1 treatment in patients with breast cancer with mLNs.

The human breast cancer cell lines SKBR3, BT474, MCF7, T47D, and MDA-MB-231 were obtained from the ATCC. The identity of cell lines was confirmed by short tandem repeat (STR) analysis (Takara Bio CDM Center) and the STR profiles were matched to their original profiles in the cell line database at ATCC. The cells described in the experiments were used within 10 passages after thawing from the original frozen stocks. All cell lines have been tested for Mycoplasma contamination frequently to ensure that they were free of Mycoplasma contamination.

LN samples of patients with breast cancer were obtained from patients receiving axillary LN dissection at the Second Affiliated Hospital of Zhejiang University of Medicine (Zhejiang, P.R. China), with 10 mL of corresponding peripheral blood (PB) also collected. Concentrated leukocyte suspensions from healthy donors were obtained from the blood centers of Zhejiang Province. We also collected paraffin-embedded mLN samples from 26 patients with breast cancer who had received neoadjuvant therapy before surgery for multicolor immunofluorescence staining. The treatment responses of these cases were evaluated according to the RECIST (18): partial response (PR; reduction of target lesion volume by at least 30% from baseline), progressive disease (PD; increase in target lesion volume by at least 20%), and stable disease (SD; between PR and PD). The collection of all related clinical samples was approved by the human research ethics committee of the Second Affiliated Hospital of Zhejiang University of Medicine (Zhejiang, P.R. China). The written informed consent for sample acquisition was obtained from all patients, and the study was performed in full compliance with the principles of the Helsinki Declaration.

Fresh LN specimens were minced into small pieces (1 mm³) with sterile scissors in complete medium including 1 mg/mL type I, II, and IV collagenase (Sigma). The tissue was further dissociated with a GentleMACS tissue dissociator (Miltenyi Biotec) for 1 hour at 37°C. Thereafter, the cell suspension was filtered through an 80-μm cell strainer to remove tissue debris and then resuspended in PBS containing 2% FBS for further experiments.

Human peripheral blood mononuclear cells (PBMC) were isolated using human lymphocyte separating fluid (DKW) through density centrifugation (800 × g, 20 minutes) from PB. To isolate various B-cell subpopulations, CD19⁺ magnetic beads (Miltenyi Biotec) were first applied to enrich B cells, with CD24^hiCD27⁺ Bregs then sorted by FACS (BD Fortessa). The purity of all sorted cells was confirmed to be greater than 95% by flow cytometric analysis. B cells were cultured in RPMI 1640 with 10% FCS and stimulated with human sCD40L (PeproTech) at 1 μg/mL and class B CpG oligonucleotide (ODN2006; InvivoGen) at 10 μg/mL.

The cancer cells and CD24^hiCD27⁺ Bregs were cocultured either with direct contact or in a Transwell system. Sorted CD24^hiCD27⁺ Bregs were cocultured with adherent breast cancer cells at a 5:1 ratio for 48 hours. CD24^hiCD27⁺ Bregs were then washed off with PBS. Breast cancer cells were subjected to further analysis.

The indicated breast cancer cells were cocultured with sorted CD24^hiCD27⁺ Bregs for 24 hours, and the culture media were harvested as conditioned medium (CM). The obtained CM was filtered using Sartorius spin columns with a cutoff of 5 kDa (Sartorius) to remove proteins with molecular weights greater than 5 kDa. The volume of obtained CM (<5 kDa) was then supplemented to its original volume before filtration with 1640 complete medium.

For cell surface marker expression analysis, cells were resuspended in PBS with 2% FBS and incubated with the respective fluorescence-labeled antibodies at 4°C for 30 minutes. For intracellular staining analysis, B cells were fixed and permeabilized using a BD Cytofix/Cytoperm Solution Kit after surface staining, and then incubated with the respective fluorescence-labeled antibodies to detect cytokine production. All antibodies used in this study are listed in Supplementary Table S1.

An FITC Annexin V Apoptosis Detection Kit with propidium iodide (PI; BioLegend, 640914) was applied to evaluate the proportion of apoptotic breast cancer cells treated with lapatinib or pyrotinib or chemotherapeutic agents. Briefly, cells were incubated with 50 μL of binding buffer containing 2.5 μL of FITC annexin V antibody and 5 μL of PI solution for 15 minutes at room temperature. Then, 200 μL of annexin V binding buffer was added for flow cytometry analysis.

To detect the effect of B cells on the treatment sensitivity of breast cancer cells, breast cancer cells were cocultured with the indicated B subsets in 96-well plates, with subsequent drug treatments at the indicated concentrations. After 48 hours, the cells were replaced with 100 μL of fresh medium containing 10% Cell Counting Kit-8 (CCK-8) reagent (Invigentech, IV-1000) for 1 hour at 37°C. The survival rate of breast cancer cells was determined by measuring and analyzing the absorbance of cells at 450 nm.

Cancer cells were suspended in serum-free DMEM-F12 (Gibco) with 0.4% BSA (Sigma), 1× B27 (Invitrogen), 20 ng/mL EGF (Invitrogen), and 10 ng/mL bFGF (PeproTech) at 500 cells per well in ultralow adhesion 96-well plates (Corning). After culturing for 10 days, the formed tumorspheres were photographed and the number of tumorspheres with diameters larger than 50 μm was counted.

CD24^hiCD27⁺ Bregs isolated from PBMCs or LNs were cocultured 1:1 with carboxyfluorescein diacetate succinimidyl ester (CFSE)–labeled CD4⁺ T cells for 5 days, accompanied by stimulation with anti-human CD3/anti-human CD28 (BioLegend). The proliferation of CD4⁺ T cells under different culture conditions was determined by flow cytometry according to CFSE fluorescence intensity.

Breast cancer cells for immunofluorescence were seeded in 48-well culture plates with cell climbing slices. After coculture with CD24^hiCD27⁺ Bregs for 48 hours, slices were washed to remove B cells and then fixed with 0.4% paraformaldehyde for 15 minutes at room temperature. Fixed cells were washed with PBS, permeabilized with 0.2% Triton X-100 in PBS for 15 minutes, and then blocked with QuickBlock Blocking Buffer (Beyotime) for 15 minutes at room temperature. Afterward, slices were incubated with the indicated primary antibodies overnight at 4°C. Incubation with secondary antibodies was performed for 1 hour at room temperature. Finally, blocking solution containing DAPI for counterstaining the nuclei was added. Immunofluorescence images were acquired using a Leica DM microscope.

Ninety-two inflammation-related proteins in supernatants from CD24^hiCD27⁺ Bregs and SKBR3 cells under different culture conditions were measured and analyzed using the Olink Inflammation panel (list presented in Supplementary Table S2; Olink Proteomics AB) according to the manufacturer's instructions. The proximity extension assay technology used here has been well described previously (19). In brief, the corresponding 92 antibody probes bound to their specifically targeted proteins and then generated a unique PCR target sequence along with proximity-dependent DNA polymerization events. Such events were further detected and quantified using a microfluidic real-time PCR instrument (Biomark HD, Fluidigm). The final assay read-out is shown as Normalized Protein eXpression (NPX) values, an arbitrary unit on a log₂ scale, after the control and normalization of the resulting Ct data using a set of internal and external controls.

Supernatants from CD24^hiCD27⁺ Bregs were cocultured with breast cancer cells directly or cultured in a Transwell system, treated with various reagents (anti-PD-L1, BioLegend, 10 μg/mL; anti-CD154, BioLegend, 10 μg/mL), and then collected. Quantification of cytokine levels, including IL4, IL6, IL10, TNFα, and IFNγ, was performed using a human Th1/Th2 cytokine cytometric bead array (CBA) kit (BD Biosciences) according to the manufacturer's instructions. Briefly, samples (25 μL) were added to mixtures of cytokine beads (25 μL) and PE-conjugated detection reagent (25 μL). After incubation for 2.5 hours at room temperature, capture beads were washed with PBS and detected using flow cytometry, and the results were analyzed with CBA software (BD Biosciences).

Total RNA was isolated from breast cancer cells or CD24^hiCD27⁺ Bregs under various culture conditions at the indicated timepoints. cDNA was synthesized using HiScript II Q RT Supermix for qPCR (Vazyme). qPCR was performed using ChamQ Universal SYBR qPCR Master Mix (Vazyme) on a LightCycler 480 II (Roche Diagnostics). Gene expression levels were quantified by using the 2^−ΔΔCt method, and the level of GAPDH mRNA was used as an internal control. Primer sequences are listed in Supplementary Table S3.

Cells were fully lysed with ice-cold RIPA buffer (Beyotime) with protease inhibitors and phosphatase inhibitors. Thereafter, 20 μg protein samples along with loading buffer were separated by a 10% SDS-PAGE gel and transferred to polyvinylidene difluoride membranes (Bio-Rad). The indicated primary antibodies were incubated overnight at 4°C and then stained with the corresponding secondary antibody for 1 hour at room temperature. The protein bands were visualized using Bio-Imaging Systems and then photographed for recording.

RNA was extracted from breast cancer cells under distinct treatment conditions using a RNeasy kit (Qiagen) according to the manufacturer's instructions. After removing any raw data with low-quality, adapter pollution or high levels of N bases, clean data were obtained and then aligned against the NCBI human reference genome via HISAT. Differential gene expression was analyzed in R software using EdgeR and Limma. Gene set enrichment analysis (GSEA) was performed by utilizing GSEA software (http://www.broadinstitute.org/gsea), and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was conducted using the KEGG pathway database (https://www.genome.jp/kegg/pathway.html).

The paired primary breast tumors and mLNs from a patient with HER2⁺ breast cancer were obtained via diagnostic needles. The tissue samples were immediately processed for single-cell RNA sequencing (scRNA-seq). scRNA-seq data were first processed using the 10× Genomics CellRanger (v3.1.0) pipeline. After quality control and filtering, data were visualized using the Uniform Manifold Approximation and Projection (UMAP) package (https://github.com/ropenscilabs/umapr). Further iterations of reclustering were performed for B cells, and marker genes of each subset were identified by the FindAllMarkers function with a threshold of 0.25 for logFC.

Female BALB/C mice ages 6–8 weeks were obtained from Shanghai Slake Laboratory Animal Co., Ltd. For the mouse model of LN metastasis, we injected 2.5 × 10⁵ 4T1 cells into the right footpad of each mouse. Mice were randomly grouped and weighed before drug administration according to the experimental requirements (n = 5 per group). At 3 weeks, anti-TNFα antibody (adalimumab, i.p., 2 mg/kg, Vetter Pharma-Fertigung GmbH & Co. KG), anti-IL6 antibody (i.p., 200 μg/mouse, Biocell, MP5-20F3), anti-CD20 antibody (i.p., 250 μg/mouse, BioLegend, SA271G2), anti-PD-L1 antibody (atezoliumab, i.p., 5 mg/kg, MCE) or paclitaxel (PTX, i.p., 2.5 mg/kg, MCE) was administered every 3 days with a total of two doses, with control mice receiving an equal volume of PBS administration. In addition, the coadministration group with anti-CD20 antibody and paclitaxel was pretreated with anti-CD20 antibody by intraperitoneal injection 3 days prior to drug administration. Donor mice from each group were euthanized at day 28, and metastatic popliteal LNs and primary tumors were harvested separately for hematoxylin and eosin (H&E), IHC, multicolor immunofluorescence analysis, and flow cytometry. The animal protocol was reviewed and approved by the animal ethics committee of the Second Hospital Affiliated with the Zhejiang University School of Medicine (Hangzhou, P.R. China).

The metastatic popliteal LNs and primary tumor samples harvested from animals were formalin-fixed, paraffin-embedded, and then sectioned into 5-μm-thick sections. H&E was performed according to the standard protocol. Briefly, paraffin sections were first dewaxed and rehydrated and then stained with H&E for 5 minutes. After 5 dips in 1% acid ethanol (1% HCl in 75% ethanol) and rinsing in distilled water, tissue sections were stained with eosin solution again for 3 minutes, followed by dehydration with graded alcohol and clearing in xylene. Images were collected using a LEICA DM3000 LED [Leica DMshare (v3)]. As the definition of micrometastasis and macrometastasis in human mLNs was not suitable for a mouse model, here, a residual metastatic tumor diameter ≥1/4 of the length of the mLN was defined as a macroresidual tumor; a residual metastatic tumor diameter <1/4 of the length of the mLN was defined as a microresidual tumor; and the lack of residual metastatic tumor cells were defined as a pCR.

IHC was performed according to standard histologic protocols to evaluate the expression levels of CD44. The primary antibody against CD44 (Abcam, ab243894) was diluted 1:2,000 to incubate the slices at 4°C overnight. After three rinses with PBS, the tissue slides were incubated for 1 hour in secondary antibody followed by another three rinses with PBS. The sections were further colorized with 3,3′-diaminobenzidine (DAB) chromogenic agent, washed with tap water to terminate the reaction, and then redyed in hematoxylin. All images were captured using a LEICA DM3000 LED microscope [Leica DMshare (v3)], and areas of positively stained cells and staining intensity were quantified and analyzed using Image Pro Plus 6.0.

Multispectral immunofluorescence staining was performed as described previously (20). Paraffin sections of tumor and paired LNs from 7 different patients were incubated with the corresponding antibodies and then mounted with ProLong Diamond Antifade mounting medium containing DAPI (Invitrogen). Images of tissue specimens were captured using TissueFaxs imaging software (TissueGnostics). According to the fluorescence intensity and fluorescence area, the cell population was quantified using TissueQuest software (TissueGnostics). Furthermore, StrataQuest software was also used to quantify the number of defined cell populations distributed in the area of the tumor cells (0–25, 25–50, and 50–100 μm). Similarly, multicolor immunofluorescence analysis of paraffin-embedded mLN tissue sections of mice from different treatment groups was also performed as described above.

Clinical information and genomic profiling of The Cancer Genome Atlas (TCGA-BRCA: n = 1,065) were obtained through Sangerbox (a comprehensive, interaction-friendly clinical bioinformatics analysis platform). Among 1,042 female stage I–III breast cancer primary samples, 810 samples were HR positive, 124 samples were HER2 positive, and 79 were triple-negative breast cancer (TNBC). The xCell algorithm was used to infer the immune score and the relative immune cell abundance of the whole tumor by transcriptome data as described previously (21).

All data are presented as the means ± SDs of at least three independent experiments and were statistically analyzed using GraphPad Software (GraphPad Prism 7.0). Differences between two groups were compared using paired t tests (paired data) or unpaired t tests (unpaired data). One-way ANOVA followed by the Tukey post hoc test was performed for comparisons among multiple groups. A two-tailed test was performed, and a P value < 0.05 was considered statistically significant.

scRNA-seq data are available through National Human Genetic Resources Sharing Service Platform (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA004270). The codes used to generate the datasets that were analyzed are available upon request from the corresponding author. All other reasonable data including more detailed methods and materials generated in this study can also be made available upon request from the corresponding author.

First, we addressed whether B cells, as one of main cellular components of LNs, have a close association with residual tumor cells surviving standard NAT within mLNs. As IL10 production is well recognized as a dominant signature of Bregs (11), we performed a multicolor immunofluorescence assay to quantify IL10⁺ CD19⁺ B cells in the mLNs of patients with breast cancer who had completed standard neoadjuvant therapy (chemotherapy ± anti-HER2 agents). A significantly higher percentage of IL10⁺ B cells were evident in the mLNs of patients evaluated as having a poor response (PD or SD of mLNs, n = 9) than in patients with a substantial response (PR of mLNs, n = 17; Fig. 1A; Supplementary Table S4). We then collected paired primary tumors and mLNs from a patient with HER2⁺ breast cancer who had received multiple lines of treatment, including paclitaxel plus trastuzumab, followed by capcitabine plus pyrotinib. During such treatment, the primary tumor achieved significant remission (RCB I), but the mLNs had remained stable. Fresh specimens were then obtained from surgery and analyzed with scRNA-seq (Fig. 1B). Compared with the primary tumors, we found that a cluster of B cells with CD24 and CD27 expressions were enriched in mLNs (Fig. 1C and D), while the plasma cell markers CD38 and SCD1 (CD138) as well as the markers of antigen presentation, including CD80 and CD86, were almost absent in all B-cell subpopulations (Supplementary Fig. S1A). In addition, we analyzed scRNA-seq data of six mLNs of patients with treatment-naïve breast cancer from the Gene Expression Omnibus (GEO database (GSE195861_RAW). In line with these results, B cells with CD24 and CD27 expressions also represented one of the main subsets of immune cells within mLNs (Fig. 1D) and showed correspondingly negligible expressions of CD38, SCD1, CD80, and CD86 (Supplementary Fig. S1B).

To further understand the differences between B-cell subsets in metastatic and normal LNs, we assessed the proportion of B-cell subpopulations in paired non-metastatic LNs (non-mLNs) and mLNs of patients with breast cancer. Previous studies have confirmed that human CD24^hiCD27⁺ B cells were IL10-producing Bregs (22, 23). Here, CD24^hiCD27⁺ Bregs were found to represent the highest proportion of B-cell subsets in both non-mLNs and mLNs (Fig. 1E), supporting the natural dominant abundance of CD24^hiCD27⁺ Bregs in LNs. The proportion of CD24^hiCD27⁺ Bregs between non-mLNs and mLNs was also similar. To analyze the features of Bregs in mLNs, we conducted a bioinformatics analysis based on differentially expressed genes between paired primary tumors and mLNs of patients with breast cancer (GSE180186). In addition to CD27, a marker of human memory B cells, genes related to Breg differentiation (TLR9, CD40, and CD40L) were also found to be upregulated in mLNs compared with primary tumors (Supplementary Fig. S1C; ref. 24). Further functional enrichment analyses for upregulated genes in mLNs indicated significant enrichment in the biological process classes of “B-cell activation” and “B-cell differentiation” as well as the KEGG pathways of “NFκB signaling,” “JAK-STAT signaling,” and “PI3K-AKT signaling” (Supplementary Fig. S1D). In addition, genes associated with positive regulation of lymphocyte activation and cell-cell adhesion, protein synthesis and transport, and the TNF signaling pathway were also found to be significantly enriched in CD24^hiCD27⁺ Bregs within mLNs of breast cancer compared with those within non-mLNs based on the analysis of scRNA-seq data (GSE180286; Supplementary Fig. S1E). Taken together, we speculated that the activation status, but not the abundance of CD24^hiCD27⁺ Bregs in mLNs, determines their protumoral function.

As the activation status of immune cells can be determined by their spatial location within the TME (25), we then investigated the distribution of CD24^hiCD27⁺ Bregs in the mLNs of breast cancer. In mLN sections, the lymphoid follicles, usually rich in B cells, were disrupted, and B cells were more distributed around metastatic tumor nests (Supplementary Fig. S1F). Further quantitative analysis of multicolor immunofluorescence images showed that CD24^hiCD27⁺ Bregs were the predominant B-cell subset secreting IL10 in both mLNs and non-mLNs, and the proportion of activated CD24^hiCD27⁺ Bregs with IL10 expression in mLNs was significantly higher than that in non-mLNs (Fig. 1F). In addition, spatial location analysis using the StrataQuest software (TissueGnostics) showed that increased activated CD24^hiCD27⁺ Bregs were observed in close proximity to breast cancer cells within the mLNs (Supplementary Fig. S1G). In this, the proportion of IL10⁺CD24^hiCD27⁺ Bregs was significantly higher around tumor cells (radius < 25 μm) than that in farther from tumor cells, which may indicate a direct interaction mechanism supporting the activation of CD24^hiCD27⁺ Bregs (Fig. 1G).

Considering the limitation of LN specimen acquisition to perform in vitro experiments, the phenotypic and functional similarities of CD24^hiCD27⁺ Bregs from paired PB and LNs were investigated. We first confirmed that human CD24^hiCD27⁺ B cells, regardless of whether derived from LNs or PB, produced comparable higher levels of IL10 in response to CD40L and CpG than any other B-cell subpopulations, showing the IL10 secreting Breg cell phenotype (Supplementary Fig. S2A). Similar expression levels of CD5, CD38, CD39, IgD, and IgM on CD24^hiCD27⁺ Bregs from the two sources were also observed (Supplementary Fig. S2B). Comparative transcriptome analysis also supported the similarities within the B-cell marker gene set and functional gene set between sorted CD24^hiCD27⁺ Bregs from paired PB and LNs (Supplementary Fig. S2C and S2D). Correspondingly, PB-derived and LN-derived CD24^hiCD27⁺ Bregs with CD40L and CpG stimulation presented comparable cytokine secretion levels, including those of IL2, IL4, IL6, IL10, TNFα, and IFNγ (Fig. 2A). A CD4⁺ T-cell proliferation assay also demonstrated the comparable suppressive ability of CD24^hiCD27⁺ Bregs from paired PB and LNs (Supplementary Fig. S2E). Therefore, CD24^hiCD27⁺ Bregs from PB were sorted and then used for the further functional experiments of this study.

The sensitivity of breast cancer cells to anti-HER2 targeted agents (pyrotinib and lapatinib), as well as to chemotherapeutic drugs (paclitaxel), was first assessed by coculture of breast cancer cells with CD24^hiCD27⁺ Bregs. The CCK-8 assay showed that anti-HER2 targeted therapy and paclitaxel both effectively kill breast cancer cells, but that coculturing with CD24^hiCD27⁺ Bregs significantly improves the survivability of breast cancer cells (Fig. 2B). Consistently, CD24^hiCD27⁺ Bregs were capable of conferring breast cancer cell resistance to the apoptotic effect induced by either anti-HER2 therapy or paclitaxel (Fig. 2C). Given the high sensitivity of TNBC to chemotherapy, we also determined whether CD24^hiCD27⁺ Bregs influenced the response of MDA-MB-231 cells (a TNBC cell line) to paclitaxel. Upon coculturing with CD24^hiCD27⁺ Bregs, the survival of MDA-MB-231 cells was indeed significantly promoted with significant decreases in apoptotic cells upon paclitaxel exposure (Fig. 2D and E). Taken together, these data suggested that CD24^hiCD27⁺ Bregs were able to markedly promote the multidrug resistance of breast cancer cells.

In addition to increased drug resistance, the adherent breast cancer cells cocultured with CD24^hiCD27⁺ Bregs presented with elongated morphologic features (Fig. 3A) indicating epithelial–mesenchymal transition (EMT). The expression of EMT-related genes, Snail, Vimentin, Twist, N-Cadherin, E-Cadherin, ZEB1, and ZEB2, was also assessed and most of these genes associated with the mesenchymal transition were confirmed to be significantly upregulated in breast cancer cells cocultured with CD24^hiCD27⁺ Bregs (Fig. 3B). Immunofluorescence staining of Snail, ZEB1, and ZEB2 also supported the ability of CD24^hiCD27⁺ Bregs to induce EMT in breast cancer cells (Fig. 3C). EMT is closely related to cancer cell stemness and able to contribute to drug resistance in malignant diseases (26). Changes in breast cancer stemness-related genes, ALDH1A, CD44, Nanog, OCT4, and SOX2 (27), were therefore evaluated. Most of the expressions of these genes were also significantly higher in breast cancer cells cocultured with CD24^hiCD27⁺ Bregs compared with the control group (Fig. 3D). CD44 is well recognized as a classical marker of breast cancer stem cells (28); both transcriptional and protein expression levels of CD44 were significantly elevated upon coculture with CD24^hiCD27⁺ Bregs (Fig. 3D and E). In addition, the sphere formation of breast cancer cells was greatly inhibited by anti-HER2 therapy and paclitaxel, which could also be markedly rescued by coculturing with CD24^hiCD27⁺ Bregs (Fig. 3F).

We next examined whether CD24^hiCD27⁺ Bregs would have similar effects on luminal and TNBC cells. Using the same treatment conditions, the morphologic alterations and EMT-related gene expressions indicated that CD24^hiCD27⁺ Bregs had strongly induced EMT in breast cancer cells of both luminal and TNBC subtypes (Supplementary Fig. S3A and S3B). CD24^hiCD27⁺ Bregs also consistently promoted the expression of stemness-related genes and sphere formation in both luminal and TNBC cells (Supplementary Fig. S3C–S3F). Collectively, these findings suggest that CD24^hiCD27⁺ Bregs may be able to potently promote the EMT and stemness of breast cancer cells independent of molecular subtype.

Most previous studies have reported that Bregs exert protumor effects or support the immune suppressive microenvironment in a contact-independent manner, via cytokines such as IL10, IL35, and TGFβ (29). One recent study revealed that the neurotransmitter GABA could also be released by activated B cells and plasma cells to limit antitumor immunity by inducing IL10⁺ macrophages (14). To evaluate the role of secreted factors of various molecular weights, BT474 breast cancer cells were stimulated with conditioned media (CM) filtered with or without a 5-kDa ultrafiltration tube. CM without protein ingredients failed to promote pyrotinib and paclitaxel resistance in BT474 cells (Supplementary Fig. S4A and S4B). In this way, the possibility that such a mechanism could be simply mediated by small molecule substances (<5 kDa), including soluble metabolites from activated CD24^hiCD27⁺ Bregs, was excluded.

As our above results of multicolor immunofluorescence revealed that the activated CD24^hiCD27⁺ Bregs within the mLNs were located close to the tumor cells (Fig. 1G; Supplementary Fig. S1G), we next evaluated whether cell‒cell contact was critical for the effect of CD24^hiCD27⁺ Bregs on breast cancer cells. Although CD24^hiCD27⁺ Breg-induced upregulation of EMT- and stemness-related genes was also found in breast cancer cells cocultured with CD24^hiCD27⁺ Bregs in a 0.4 μm Transwell system, a significantly stronger effect was observed related to direct contact (Supplementary Fig. S5A and S5B). In addition, in terms of CD44 expression and sphere formation of breast cancer cells, direct contact cocultures showed markedly stronger effects than Transwell cocultures (Supplementary Fig. S5C and S5D). Similarly, a higher cell survival rate and lower apoptotic proportion of breast cancer cells were also observed in direct contact conditions than in the Transwell system upon treatment with pyrotinib or paclitaxel (Supplementary Fig. S5E and S5F). On the basis of these findings, we concluded that CD24^hiCD27⁺ Bregs could promote the EMT, stemness, and drug resistance of tumor cells by producing cytokines, but also more importantly, that this effect could be potentiated by direct contact with tumor cells.

The detailed mechanism of how Bregs exert direct protumoral function by soluble factors in mLNs is still poorly understood. We proceeded to identify cytokine production by CD24^hiCD27⁺ Bregs using an Olink inflammation panel comprised of 92 inflammatory proteins. Compared with the Transwell coculture and control groups, 21 cytokines were upregulated in the supernatants from the direct contact coculture group (Fig. 4A). Among them, IL6, IL8, IL10, TNFα, and TNFβ have been previously reported to be associated with cancer stemness and tumor drug resistance (30–32). Although the well-known EMT-promoting cytokine TGFβ was not included in this Olink panel, we found no substantial increase of TGFβ in the supernatant when CD24^hiCD27⁺ Bregs were activated with CpG and CD40L (Supplementary Fig. S6A). Through supplementation with neutralizing antibodies against various cytokines within the direct contact coculture system, we found that blocking IL6 or TNFα impaired the antiapoptotic effect of CD24^hiCD27⁺ Bregs on tumor cells in response to pyrotinib. However, antibodies against IL8, IL10, or TNFβ did not produce the same effect (Fig. 4B). The combination of anti-IL6 and anti-TNFα further restrained the drug resistance of breast cancer cells when cocultured with CD24^hiCD27⁺ Bregs (Supplementary Fig. S6B and S6C). The tumor cells were also capable of producing multiple cytokines, such as IL6, IL10, and TNFα, as reported by previous studies (33–35). Therefore, we further analyzed the transcriptomic levels of these key cytokines within CD24^hiCD27⁺ Bregs and SKBR3 cells from the direct contact coculture group. The mRNA levels of IL6, IL10, and TNFα in SKBR3 cells were significantly lower than those in CD24^hiCD27⁺ Bregs. This remained unaffected by the coculture conditions, suggesting CD24^hiCD27⁺ Bregs to be the major source of these cytokines (Supplementary Fig. S6D). Although IL10 unexpectedly did not seem to participate in the protumural function of CD24^hiCD27⁺ Bregs, flow cytometry analysis revealed that the activated IL10⁺CD24^hiCD27⁺ Bregs within the mLNs largely overlapped with IL6- and TNFα-producing B cells (Fig. 4C).

Thereafter, we performed GSEA and KEGG analysis of the transcriptional profiles in SKBR3 cells cocultured with CD24^hiCD27⁺ Bregs in a contact manner that indicated activation of both the NFκB and JAK-STAT signaling pathways (Supplementary Fig. S7A and S7B). It is well acknowledged that the NFκB pathway can be activated by TNFα and the JAK-STAT pathway can be initiated by IL6 (36). Moreover, higher levels of phosphorylated STAT3 (p-STAT3) and p65 (p-p65) were observed in breast cancer cells cocultured with CD24^hiCD27⁺ Bregs with direct contact, than in the Transwell system (Supplementary Fig. S7C). The addition of JSH-23 (NFκB inhibitor) and SH-454 (STAT3 inhibitor) significantly attenuated the antiapoptotic effect of CD24^hiCD27⁺ Bregs on breast cancer cells (Supplementary Fig. S7D and S7E). Taken together, these results implied that CD24^hiCD27⁺ Bregs promote drug resistance in breast cancer cells by simultaneously activating the TNFα-NFκB and IL6-STAT3 pathways.

We next employed a mouse model of LN metastasis with 4T1 cells to explore the effect of TNFα and/or IL6 blockade on the therapeutic efficacy of mLNs in vivo (Fig. 4D). The influence of B-cell depletion was also investigated (Fig. 4D; Supplementary Fig. S8A). Significantly, the addition of anti-IL6 improved the inhibition of tumor growth within mLNs, and the simultaneous blockade of TNFα and IL6 was comparable with the inhibitory effects of the anti-CD20 antibody on the tumor growth in mLNs (Fig. 4E). We further assessed the residual tumor burden across various treatment groups. As shown in Fig. 4F, the addition of either anti-TNFα or anti-IL6 antibody was capable of decreasing tumor cells in mLNs, compared with paclitaxel alone, while the pCR rate of mLNs was significantly elevated by up to 60% when anti-TNFα and anti-IL6 antibodies were both administered along with paclitaxel. In contrast, the pCR rate of mLNs in the anti-CD20 antibody– and paclitaxel-treated groups was only slightly improved (Fig. 4F). Further quantitative analysis of IL6, IL10, and TNFα positive cells within mLNs from the control group and anti-CD20 antibody–treated group by flow cytometry determined that B cells within mouse mLNs are the major source of IL6 and IL10, but have inferior production of TNFα compared with T cells (Supplementary Fig. S8B). T cells, dendritic cells, macrophages, and tumor cells were also able to secrete these cytokines when B cells are depleted (Supplementary Fig. S8C), which might explain the poorer outcomes of paclitaxel combined with anti-CD20 antibody compared with the therapeutic effect of paclitaxel combined with blockade of IL6 and TNFα on mLNs (Fig. 4F). We next evaluated whether CD24^hiCD27⁺ Bregs were the dominant sources of both IL6, IL10, and TNFα in human mLNs. Upon assessment by flow cytometry, our results confirmed that CD24^hiCD27⁺ Bregs were the main cellular sources for all three cytokines (IL6, IL10, and TNFα) among CD45⁺ immune cells within human mLNs (Supplementary Fig. S9A and S9B). Taken together, our findings suggest the importance of TNFα and IL6 in maintaining the drug resistance of metastatic tumor cells within mLNs.

Our above data strongly suggested that the protumoral effect of CD24^hiCD27⁺ Bregs was significantly potentiated by direct contact with tumor cells. We therefore further investigated the underlying mechanisms of this. Both the transcriptional levels and cytokine production of IL6, IL10, and TNFα were upregulated in CD24^hiCD27⁺ Bregs cocultured with breast cancer cells in contact conditions compared with the other groups (Fig. 5A and B). In addition to CD40L, PD-L1 was also found to be involved in the regulation of B-cell activation and differentiation. In previous studies, Xiao and colleagues and Guan and colleagues both reported that the induction of IL10⁺ Bregs was mediated by PD-L1, as expressed by tumor cells (37, 38). Here, we confirmed higher PD1 levels on CD24^hiCD27⁺ Bregs than the other B cells (Supplementary Fig. S10A). Moreover, expression levels of CD40L and PD-L1 on different molecular subtypes of breast cancer cells were also determined (Fig. 5C). Interestingly, breast cancer cells cocultured with CD24^hiCD27⁺ Bregs presented upregulated CD40L and PD-L1 levels (Fig. 5D; Supplementary Fig. S10B). In the mouse model of LN metastasis with 4T1 cells, elevated expression of PD-L1 on metastatic tumor cells within mLNs was far higher than that on tumor cells in primary lesions. This was consistent with our in vitro experiment, and the induction of PD-L1 expression on metastatic tumor cells could be weakened by B-cell depletion (Supplementary Fig. S10C). Next, we addressed whether CD40L-dependent and PD-L1–dependent signaling are involved in the activation of CD24^hiCD27⁺ Bregs when in direct contact with breast cancer cells. As expected, pretreatment with either anti-CD40L or anti-PD-L1 impaired the ability of breast cancer cells to induce increased expression of IL6, IL10, or TNFα by CD24^hiCD27⁺ Bregs in direct contact conditions, and inhibition of both ligands simultaneously presented an even stronger effect (Fig. 5E).

Compared with CD40L, PD-L1 is an actionable target in breast cancer. We therefore investigated the therapeutic effect of anti-PD-L1 antibody to impair the protumoral effect of CD24^hiCD27⁺ Bregs. We first confirmed that supplementation with an anti-PD-L1 antibody attenuated CD24^hiCD27⁺ Breg-induced drug resistance in BT474 and MDA-MB-231 cells with high PD-L1 expression (Fig. 6A and B). Thereafter, we further confirmed the contribution of anti-PD-L1 toward eliminating tumor cells within mLNs in vivo. Regarding the volume of mLNs in the different treatment groups, paclitaxel or anti-PD-L1 alone significantly suppressed the tumor growth in mLNs, while combined administration presented the highest tumor growth inhibition for mLNs (Fig. 6C). More importantly, by assessing the pathologic response rate of different treatments, the pCR rate of mLNs was 60% in the anti-PD-L1–treated group, representing a much higher level than that of the paclitaxel-treated group. The highest pCR rate (80%) was clearly presented in the combined treatment groups (Fig. 6D).

In addition, IHC revealed much stronger staining of CD44 on metastatic tumor cells within mLNs than in primary tumors, while CD44 expression on metastatic tumor cells was greatly inhibited by the administration of anti-PD-L1, but not by paclitaxel (Fig. 6E). Similar to the effect of B-cell depletion in vivo, increased PD-L1 levels on metastatic tumor cells within mLNs were also abrogated by anti-PD-L1 treatment (Supplementary Fig. S10D). This could be due to the inhibition of Bregs activation. By quantitative analysis of multicolor immunofluorescence in non-PCR mLNs, we further confirmed that the proportion of IL10-expressing B cells, with or without the simultaneous expression of IL6 and/or TNFα, was positively correlated with the residual tumor burden (Fig. 6F; Supplementary Fig. S11A). Moreover, a close spatial relationship between TNFα⁺IL6⁺IL10⁺ B cells and tumor cells within mLNs was also noted here, showing that the B cells closer to metastatic cancer cells tended to be activated in the secretion of those cytokines. However, this effect was not observed in the anti-PD-L1 group, also suggesting the importance of PD-L1–dependent activation of Bregs (Supplementary Fig. S11B). Further evaluation of the therapeutic effect of mLNs with different combination regimens also suggested that the addition of anti-PD-L1 has achieved significant improvement in eradicating tumor cells within mLNs, which was comparable with that of combined IL6 and TNFα blockade (Supplementary Fig. S11C and S11D). We did not observe significantly better efficiency for the elimination of tumor cells within mLNs of mice that received combo administration of paclitaxel plus blocking antibody of IL6, TNFα, and PD-L1 compared with other combination therapy groups, including paclitaxel combined with the blockade of IL6 and TNFα as well as paclitaxel combined with the blockade of PD-L1 (Supplementary Fig. S11C and S11D). Because of the low expression of PD-L1 in the primary lesion of breast cancer (39), anti-PD-L1 was still not included in the standard neoadjuvant therapeutic strategy. However, our results strongly indicated that anti-PD-L1 therapy has more prominent effects on cancer cells in mLNs.

Altogether, the results above elucidated that CD40L-dependent and PD-L1–dependent direct-contact interactions dominate the effect of CD24^hiCD27⁺ Bregs to induce EMT, stemness, and multidrug resistance in breast cancer cells of mLNs (as summarized in Fig. 6G), and that the addition of anti-PD-L1 agents may contribute to the prevention of drug resistance and improvements in the axillary pCR rate in patients with breast cancer.

Over recent years many reports have focused on the characteristics of the TME within metastatic lesions and their critical role in fostering tumor progression. In this study, we first reveal the spatial and biological features of activated CD24^hiCD27⁺ Bregs in mLNs that promoted the stemness and multidrug resistance of breast cancer cells mainly via TNFα and IL6. Meanwhile, increased CD40L and PD-L1 expressions on breast cancer cells are shown to further support the activation of CD24^hiCD27⁺ Bregs, forming a positive feedback loop. As our results identify the interaction between CD24^hiCD27⁺ Bregs and breast cancer cells as critical for inducing the therapeutic resistance of mLNs, these findings could provide a novel therapeutic opportunity for PD-L1 blockade for patients with breast cancer with mLNs.

In the event of LN metastasis, metastatic cancer cells invade lymphatic vessels to induce the formation of peritumor lymphatic vessels (40). Just as Dubey and colleagues previously confirmed the close contact of lymphatic vessels and B cells within mesenteric LNs (41), the accumulation of metastatic breast cancer cells in the peripheral region of B cells in mLNs was also confirmed by our study. Traditionally, peripheral B cells have been classified into naïve B cells, germinal center B cells, memory B cells, and plasma cells (42). Recent studies at the single-cell level have also suggested B-cell heterogeneity within the mLNs of breast cancer, where a very small proportion of B cells showed antitumor functions, but presented with exhausted status (43). By collecting mLNs with residual disease after various standard treatments, we performed scRNA-seq and multicolor immunofluorescence. This identified the large abundance of activated CD24^hiCD27⁺ Bregs in the mLNs of breast cancer, implicating their potential role in supporting the drug resistance of metastatic breast cancer cells within the mLNs.

Primary infiltrating B cells, which are generally associated with favorable prognosis in multiple cancers, are mainly distributed in TLS in the tumor stroma (44). Compared with the well-known antitumor effect of TLS-resident B cells, the significant accumulation of breast cancer–educated B cells within TDLNs has been found to promote the formation of a premetastatic niche (17). Ruddell and colleagues also described the function of B cells in remodeling the TDLNs and tumor metastasis using murine melanoma models (45), their team also further identifying the preferential accumulation of Bregs in the TDLNs to accelerate tumor growth (16). However, the role of Bregs within mLNs is comparatively understudied, and whether Bregs could be involved in the drug resistance of tumor cells in the mLNs of breast cancer, and if so by what underlying mechanism, has remained unclear.

To date, multiple B-cell subsets have been characterized as human Bregs. These primarily include CD24^hiCD27⁺ and CD24^hiCD38^hi B cells, the distribution of which has been confirmed in multiple tissues such as PB, spleen, and tonsils (23, 29). By flow cytometric analysis, we demonstrated that CD24^hiCD27⁺ B cells, but not CD24^hiCD38^hi B cells, occupied a large proportion of B cells in both non-mLNs and mLNs of the tumor draining region. Abundant evidence indicates that Bregs produce multiple cytokines simultaneously to exert protumoral and immune suppressive functions, including IL10, IL35, and TGFβ (13, 15). Here, we made a comprehensive evaluation of the effect of CD24^hiCD27⁺ Breg-derived cytokines and found both TNFα and IL6, but not IL10, were responsible for the stemness and multidrug resistance of breast cancer cells. This result was similar to the findings of Su and colleagues, who found that CD10⁺GPR77⁺ carcinoma-associated fibroblasts induced the stemness and chemoresistance of breast cancer cells and were mediated by IL6 and IL8, but not dependent on abundantly produced IL10 (46). IL6 has been reported to mediate tumor immune evasion by impeding the cytotoxic effect of T cells, decreasing CD8⁺ T cells’ infiltration, inducing FoxP3⁺ Treg and M2 macrophage polarization (47–49). In addition, the critical role of TNFα in the onset of the immune response was also well recognized; however, TNFα blockade could also improve antitumor immune response through inhibiting TNFα-mediated upregulation of PD-L1 and TIM-3 expression on CD8⁺ T cells (50). Therefore, CD24^hiCD27⁺ Bregs within mLNs may play pleiotropic roles to promote tumor progression, and combined blockade of IL6 and/or TNFα may be capable of regulating the numbers and function of immune cell subsets within mLNs to a certain extent.

Of note, we found a positive feedback loop between CD24^hiCD27⁺ Bregs and breast cancer cells, with the activation of CD24^hiCD27⁺ Bregs potentiated by direct contact with CD40L and PD-L1 expression on cancer cells. Several mechanisms have been reported to activate or polarize Bregs, including those involving IL1β, IL6, CD40L, and PD-L1 (24, 38, 51). With the notable exception of PD-L1, most of these are not currently actionable targets for breast cancer treatment. Nevertheless, a high expression of PD-L1 has been positively correlated with the efficiency of anti-PD-L1/PD-1 treatment, although most related studies seem to have only focused on the PD-L1 expression of cancer cells in primary tumor lesions and have overlooked metastatic cancer cells within mLNs (52). One recent meta-analysis by Boman and colleagues revealed the spatial heterogeneity of PD-L1 expression in breast cancer among various metastatic sites where the highest PD-L1 expression was found in mLNs (53). Yang and colleagues also found that PD-L1 expression was significantly higher in breast cancer cells within mLNs than in the primary tumor (54). Such observations could be partially explained by our finding that CD24^hiCD27⁺ Bregs were capable of promoting PD-L1 expression in tumor cells. In terms of mechanism, numerous cytokines within TME, including IL6, IL8, IL10, and TNFα, have been reported to promote the expression of PD-L1 (55, 56). These cytokines as mentioned above were found highly produced by activated CD24^hiCD27⁺ Bregs in our study.

Several randomized clinical trials (Keynote-522, Geparneuvo, and Impassion-031) have demonstrated the significant benefit of adding anti-PD-L1/PD1 to neoadjuvant chemotherapy in patients with breast cancer (3–5). Further subgroup analysis revealed that the additional pCR benefit of anti-PD-L1/PD1 therapy was much more pronounced in patients with axillary LN metastasis than those without (Supplementary Fig. S12A). Similarly, the randomized phase II study conducted by Ademuyiwa and colleagues included 67 patients with TNBC and demonstrated statistically significant improvements in the pCR rate of the addition of atezolizumab to neoadjuvant chemotherapy in patients with positive LNs at baseline, but not in those without LN metastasis (57). By comparing the TCGA dataset, we found the immune score of breast cancer primary lesions with different LN statuses was similar. This suggested that the benefit of adding anti-PD-L1/PD1 was independent of the TME of the primary lesion (Supplementary Fig. S12B–S12F). In addition, previous clinical studies of advanced urothelial bladder cancer and non–small cell lung cancer have already reported a similar interesting phenomenon where, compared with other metastatic lesions, metastatic LNs were presented as the best responding organ under treatment with PD-L1/PD1 immune checkpoint inhibitors (58, 59). However, the underlying mechanism for this was not uncovered in such studies. Here, with both in vitro and in vivo experiments, we confirmed the effectiveness of anti-PD-L1 treatment in reversing the drug resistance of breast cancer cells induced by CD24^hiCD27⁺ Bregs and improving the pCR rate of mLNs. The blockade of PD-L1/PD-1 has also been reported to support T-cell immunity in TDLNs in mouse models (7). Our results suggest that PD-L1 expression assessment should not be limited to primary breast tumors, but rather should also include a stronger focus upon mLNs. Clinical trials would be extremely helpful to evaluate the anti-PD-L1/PD1 treatment in patients with breast cancer with mLNs, but are likely to be more limited and therefore are less recommended for patients with TNBC simply with high PD-L1 expression in the primary lesion.

In conclusion, here we highlight the importance of activated CD24^hiCD27⁺ Bregs within the mLNs of breast cancer, which simultaneously promotes the stemness and multidrug resistance of cancer cells. Meanwhile, increased CD40L and PD-L1 expression on breast cancer cells further supports the activation of CD24^hiCD27⁺ Bregs in a contact-dependent manner, forming a positive feedback loop. Our results present the first clarification of the interaction between CD24^hiCD27⁺ Bregs and breast cancer cells within mLNs and provided novel evidence for anti-PD-L1 treatment in patients with breast cancer with mLNs. This may improve the chance of achieving pCR in patients with mLNs and may reduce the requirement for axillary LN dissection.

No disclosures were reported.

H. Huang: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. Y. Yao: Investigation, visualization, methodology. L. Shen: Formal analysis, investigation, visualization, methodology. J. Jiang: Software, visualization, methodology. T. Zhang: Data curation, supervision, methodology. J. Xiong: Formal analysis, visualization. J. Li: Data curation, investigation, visualization. S. Sun: Visualization, methodology. S. Zheng: Formal analysis, investigation. F. Jia: Visualization, methodology. J. Zhou: Resources, investigation. X. Yu: Resources, investigation. W. Chen: Resources, investigation. J. Shen: Resources, investigation. W. Xia: Resources, investigation. X. Shao: Resources, investigation. Q. Wang: Formal analysis, supervision, project administration, writing–review and editing. J. Huang: Formal analysis, supervision, project administration, writing–review and editing. C. Ni: Supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.

This work is supported by the Natural Science Foundation of Zhejiang Province (grants LR19H160001, LY21H100004), the Natural Science Foundation of China (grants 82073151, 82273275, 82173089, 32100725), the Leading Innovative and Entrepreneur Team Introduction Program of Zhejiang (grant 2019R01007), the Zhejiang Basic Public Welfare Research Project (grant LGF20H160026), and the National Key Research and Development Program of China (grant 2021YFC2501800). We would also like to express our thanks to Mozhuo Biotech (Zhejiang) for their help with analysis of the scRNA-seq data.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.