A total of 32 patients with AD fulfilled the diagnostic criteria of Hanifin and Rajka (22) and 32 healthy age-matched controls (HC) without a family history of asthma or other atopic disease were enrolled in the study. Participants who had received systemic corticosteroids or immunosuppressant and had ongoing infections within one month before the study were excluded. SCORAD (scoring atopic dermatitis) index was used to evaluate AD disease severity which sub-divided the patients into mild (scores<25), moderate (25–50) and severe (>50). Based on total serum IgE levels, AD was classified as intrinsic (IgE ≤ 150 IU/mL) or extrinsic (IgE > 150 IU/mL). Demographic data are listed in Table 1 . The study was approved by the local institutional review boards, and all participants (or their legal guardians) provided informed consent.

Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood by density gradient centrifugation on Ficoll-Paque PLUS (Cytiva). The isolated PBMCs were then analyzed with an 8-color flow cytometer (FACSCanto™II, BD Biosciences, CA, USA) after labeling with the following anti-human fluorochrome-conjugated antibodies: FITC-conjugated CD38 (clone HIT2); APC or FITC-conjugated PD-1 (clone EH12.2H7); FITC-conjugated CD45RA (clone HI100); PE-conjugated CD24 (clone ML5); PE-conjugated ICOS (clone C398.4A); PE-conjugated CCR6 (clone G034E3); PerCP-conjugated CD3 (clone UCHT1); PE/cyanine7-conjugated CD27 (clone O323); PE/cyanine7-conjugated CD4 (clone OKT4); APC-conjugated CD19 (clone HIB19); APC-conjugated CD45RO (clone UCHL1); APC-conjugated CXCR3 (clone G025H7); and Brilliant Violet 421-conjugated CXCR5 (clone J252D4) (all antibodies were from BioLegend, CA, USA). Cell viability was assessed using a live/dead assay (Fixable Viability Dye eFluor 780; eBioscience). All data were collected and analyzed using FlowJo software. The gating strategies of the isotype-matched IgG control for specific cell populations are shown in the supplementary figure ( Figure S1 ).

PBMCs from HCs and patients with AD were resuspended in culture medium (at 2 × 10⁶ cells/mL) in 48-well flat-bottom plates and stimulated for 48 h at 37 C with CpG (TLR9 ligand, 10 μg/mL; In vivogen) and CD40 ligand (CD40L, 1 μg/mL; BioLegend). Phorbol 12-myristate 13-acetate (PMA, 20 ng/mL; Sigma-Aldrich), ionomycin (1 μg/mL, Sigma-Aldrich), and brefeldin A (BFA, 5 mg/mL, BioLegend) were added for the last 5 h. Next, the cells were harvested and washed twice with phosphate buffer solution (PBS). Single-cell suspensions were then stained for 20 min on ice with predetermined optimal concentrations of APC-conjugated CD19 (BioLegend). Stained cells were washed twice with PBS before fixation and permeabilization in Fixation and Permeabilization Buffer (BD Biosciences). Finally, cells were stained for 30 min with PerCP-conjugated IL-10 (clone JES3-9D7, BioLegend).

Human naïve CD45RO⁻CD4⁺ T cells were sorted by Magnetic-Activated Cell Sorting (19555; StemCell Technologies). CD19⁺CD24^hiCD38^hi transitional B cells obtained from HCs or patients with EAD were sorted by flow cytometry (FACSAria™ II, BD Biosciences) and co-cultured (0.25 × 10⁵ cells) for 72 h with naïve CD45RO⁻CD4⁺ T cells (2 × 10⁵ cells) obtained from HC; cells were co-cultured in plates precoated with anti-CD3 (317325; BioLegend) and anti-CD28 (302934; BioLegend) antibodies (1 μg/mL). To polarize Tfh cells (21), recombinant human IL-21 (10 ng/mL, PeproTech), IL-6 (25 ng/mL, PeproTech), 5 μg/mL anti-IFN-γ (506532; BioLegend), anti-IL-4 (500838; BioLegend), and anti-TGF-β (361204; BioLegend) neutralizing antibodies were added to the co-culture system. In some experiments, an anti-IL-10 neutralizing antibody (5 μg/mL, 501427; BioLegend) was added to cultures.

IL-10, IL-21 and TARC levels in serum samples obtained from patients with AD and from HC were measured using the Luminex assay (RnD, Item No. LXSAHM-15).

B cells isolated from human PBMCs by Magnetic-Activated Cell Sorting (17954, StemCell Technologies) were immobilized onto glass slides coated with poly-L-lysine (Sigma-Aldrich). The BCR was labeled with a human IgM+G antibody (Alexa Fluro 546-(Fab)2-anti-Ig(M+G)). This antibody was generated from the F(ab′)2 fragment (Jackson ImmunoResearch, West Grove, PA) using a published protocol (23). Alexa Fluro 546-(Fab)2-anti-Ig(M+G) was incubated with B cells for 30 min on ice before Fc receptor blocking (422301; BioLegend). Then, the B cells were stimulated for 0, 5, 15, or 30 min at 37°C before fixation with 4% paraformaldehyde. In some experiments, B cells were stimulated for 48 h at 37°C with CpG (10 μg/mL) and CD40L (1 μg/mL) to detect activation of phospho-STAT3. The following specific antibodies were used to stain the cells after permeabilization with 0.05% saponin (Sigma): anti-CD38 (ab235118; abcam), anti-CD24 (ab202073; abcam), anti-phospho-STAT3 (9145S; Cell Signaling Technology), anti-phosphotyrosine (05-1050X; Merck Millipore), anti-pBTK (ab52192; abcam), and anti-pCD19 (3571S; Cell Signaling Technology). Images were obtained under a confocal microscope (Nikon A1R) using 405, 488, 546, 647 nm lasers. Colocalization and MFI were determined by the NIS-Elements AR 3.2 software.

All statistical analyses were performed using GraphPad Prism version 6. The results are expressed as median and interquartile range unless otherwise stated. Spearman’s correlation was used to evaluate the correlations between the two variables. The significance of differences between two groups was evaluated using Mann-Whitney U-test, and multiple comparisons were made using Kruskal–Wallis test. A P-value <0.05 was considered statistically significant.

PBMCs from patients with AD or HC were incubated with CpG and CD40L for 48 h, respectively, and then PMA/ionomycin and brefeldin A were added for the last 5 h before the assessment of intracellular IL-10 expression. The results showed that the percentage of CD19⁺IL-10⁺B cells in patients with EAD was significantly lower than that in HC but not in patients with IAD ( Figures 1A, B and S2A ). AD was categorized as mild, moderate, or severe in accordance with SCORAD. Patients with moderate to severe AD had a lower percentage of Bregs than those with mild AD ( Figure 1C ), suggesting that Breg cells may be associated with disease severity. However, we found no significant correlation between the percentage of Breg cells and SCORAD in total patients with AD. It is interesting to note that when the classification of IAD and EAD is taken into account, the percentage of Breg cells correlated negatively with SOCRAD (p=0.0103, r=-0.68) in patients with EAD ( Figure 1D ), again suggesting that the relative reduction in percentage of Breg cells in EAD patients shows an inverse correlation with AD severity.

Because most EAD patients have high IgE levels, we examined the correlation between the percentage of Breg cells and total serum IgE levels. There was no significant correlation between these parameters ( Figure 1E ). Furthermore, when we examined serum levels of IL-10 in patients and controls, we found no difference between the AD and HC groups; however, there was a significant increase in serum thymus and activation-regulated chemokine (TARC) levels in the AD group ( Figure 1F ).

Next, we analyzed the percentages of two phenotypes of Breg cells in peripheral blood and found no significant difference in the percentage of CD19⁺CD24^hiCD38^hi transitional B cells and CD19⁺CD24^hiCD27⁺ memory B cells between patients and HCs ( Figures 2A, B ). To determine the ability of these two above-mentioned subsets to produce IL-10, we measured expression of CD24/CD38 and CD24/CD27 by IL-10-producing B cells. Expression of CD24/CD38 and CD24/CD27 by IL-10⁺ B cells was higher than that by IL-10^- B cells in patients with AD ( Figures 2C–E ); also, the Breg subset was dominated by CD19⁺CD24^hiCD38^hi B cells ( Figure 2E ). Hence, the CD19⁺CD24^hiCD38^hi phenotype was used to denote Breg cells in the current study.

The percentage of CD3⁺CD4⁺CD45RO⁺CXCR5⁺ in the circulating Tfh population was no different in patients with AD than in HC ( Figure 3A ). Tfh cells expressing the costimulatory molecules ICOS and inhibitory receptor PD-1 are considered to be activated. The percentage of ICOS⁺PD-1⁺ cells among CD4⁺CXCR5⁺ Tfh cells was higher in patients with AD than in HC ( Figure 3B ); however, the percentage of CD4⁺CXCR5⁺PD-1⁺ICOS⁺ cTfh cells was higher only in EAD patients (i.e., not in IAD patients) than in HC ( Figure 3C ). In addition, there was a negative correlation (p=0.005, r=-0.8297) between the percentage of CD4⁺CXCR5⁺PD-1⁺ICOS⁺ cTfh cells and IL-10⁺ B cells only in EAD patients ( Figure 3D ). However, there was no correlation between the frequency of circulating Tfh cells and serum levels of IgE or disease severity (SCORAD) ( Figure S3 ).

Circulating Tfh cells were classified into three major subsets, each showing differing expression of CXCR3 and CCR6: CXCR3⁺CCR6⁻ type 1 Tfh (Tfh1) cells, CXCR3⁻CCR6⁻ type 2 Tfh (Tfh2) cells, and CXCR3⁻CCR6⁺ type 17 Tfh (Tfh17) cells (24). There was no significant difference between AD patients and HC with respect to the proportion of Tfh1, Tfh2, and Tfh17 cells ( Figure S2B ).

To assess the ability of Breg cells to suppress Tfh cell differentiation, we analyzed differentiation of CXCR5⁺ICOS⁺PD-1⁺ Tfh cells from CD4⁺ naïve T cells in a co-culture system. We confirmed that CD19⁺CD24^hiCD38^hi B cells were the predominant population of Breg cells in humans ( Figures 2C ). Accordingly, CD19⁺CD24^hiCD38^hi B cells were isolated from HCs and patients with EAD by flow cytometry, and CD4⁺ naïve T cells were isolated by MACS. Cells were co-cultured for 72 h under Tfh-polarizing conditions (21). We found that the percentage of PD-1⁺ICOS⁺CXCR5⁺CD4⁺ Tfh cells was significantly reduced upon co-culture with Breg cells from HC compared with blank controls (CD4⁺ naïve T cells were cultured with medium alone), but not upon co-culture with Breg cells from EAD patients ( Figures 4A–C and S4 ). This suggests that Breg cells from EAD patients are less able to suppress differentiation of Tfh cells

Concurrently, we added anti-IL-10 neutralizing antibody (aIL-10) to the co-culture system to see whether Breg cells effectively suppress Tfh differentiation in an IL-10-dependent manner. The results showed that the suppressive effect of Bregs on Tfh cell differentiation decreased significantly after adding aIL-10, suggesting that these Breg cells regulate Tfh cell differentiation in an IL-10-dependent manner ( Figure 4D ).

Current research suggests that naïve B cells differentiate into IL-10-producing Breg precursor cells via B-cell receptor (BCR) signaling, while development of mature Breg cells from immature precursor cells is critically dependent on TLR-STAT3 signaling (25, 26). Here, we investigated the ability of Breg cells to respond to BCR and TLR stimulation in vitro by monitoring certain key downstream targets (i.e., pY, pCD19, pBTK, and pSTAT3). The results revealed that BCR signaling in CD19⁺CD24^hiCD38^hi B cells from patients with EAD is similar to that in HC ( Figures 5A, B and S5 ). However, there was a significant reduction in phosphorylation of STAT3 when Bregs from patients with EAD were stimulated for 48 h by CpG ( Figures 5C-F ). This observation indicates that the defective function of Breg cells in patients with EAD may be related to abnormal signal transduction through the TLR-STAT3 pathway, not the BCR pathway.