Patients visiting the Pediatric Allergy and Respiratory Center of Soonchunhyang University Hospital who had aggravated AD according to the Hanifin and Rajka AD diagnostic criteria15 were enrolled. Patients were excluded if they had other skin diseases, or had received topical, oral or injectable antibiotics within the previous 14 days. Wet dressing, topical corticosteroids and antibiotics were used to treat AD aggravation in all patients, and 66.7% and 18.5% of patients were treated with topical calcineurin inhibitors and systemic corticosteroids, respectively. Normal control children who were recruited had no skin diseases including AD and did not have any other allergic or significant underlying diseases. They were recruited from children who visited the Pediatric Department of Soonchunhyang University Hospital for vaccination and had not used probiotics and systemic antibiotics within 14 days. All participants were less than 17 years of age. Blood and urine were collected from the subjects. Participants and their parents provided informed consent. The study was approved by the Ethics Committee of Soonchunhyang University Hospital (approval number 2013-061).

White blood cells and eosinophil count were measured in AD patients. The levels of total IgE, specific IgE and eosinophil cationic protein (ECP) were measured using ImmunoCAP testing according to the manufacturer's instructions (Phadia AB, Uppsala, Sweden). Laboratory tests were not performed in normal controls. AD severity was measured by the SCORing of Atopic Dermatitis (SCORAD) index16 by the same pediatric allergist to avoid inter-observer bias.

Multiple ultracentrifugation and purification processes were applied to isolate EVs, as described in a previous study.17 DNA was extracted using a commercial kit (Bioneer, Daejeon, Korea), and the quality and quantity of DNA were determined using spectrophotometry (NanoDrop Instruments, Wilmington, DE, USA).

After each clone was polymerase chain reaction (PCR)-amplified, the V1 to V3 regions were amplified using a FastStart High Fidelity PCR System (Roche Diagnostics GmbH, Mannheim, Germany) with a 16s rRNA gene fusion primer (27F-GAGTTTGATCMTGGCTCAG and 518R-WTTACCGCGGCTGCTGG). A GS-FLX plus emPCR Kit (454 Life Sciences, Branford, CT, USA) and Tissue Lyser II (Qiagen, Germantown, MD, USA) were used to make micro-reactors containing amplified compounds and beads. After emPCR amplification, the amplicon was purified using an AMpure Bead kit (Beckman Coulter, Brea, CA, USA) and quantified using the Picogreen method (Invitrogen, Carlsbad, CA, USA). Subsequently, the amplicon was diluted and analyzed with a GS-FLX Titanium sequencer (Roche Diagnostics GmbH) at Macrogen (Seoul, Korea).

Sequencing reads were obtained in high quality as determined by the Quality score (mean Phred score > 20 and read length > 300 bp). The operational taxonomic unit (OTU) was analyzed using UCLUST and USEARCH, with phylogenetic classification performed using QIIME based on the 16sRNA sequence database of GreenGenes 8.15.13. All sequences were classified based on the similarity as follows: species, > 97% similarity; genus, > 94% similarity; family, > 90% similarity; order, > 85% similarity; class, > 80% similarity; and phylum, > 75% similarity. If there was a significant difference (more than 2-fold difference between the case and control groups) in the bacterial EV composition of the genus level, then the data were configured as a heatmap.

We selected Lactobacillus in the mouse study because many studies have already confirmed the beneficial effects for eczema of Lactobacillus. 181920212223242526 Among Lactobacillus, the L. plantarum (KCTC 11401BP, Gen-Bank accession No. GQ336971) strain was isolated from kimchi, which is the most popular fermented food in Korea and is easily obtained.27 This strain also has the potential to improve mouse and human AD.2829 Finally, we decided to use L. plantarum for the experiment, and L. plantarum culture medium was donated by Probiotics Research Center, CJ CheilJedang Corporation, Korea. L. plantarum-derived EVs (Lp EV) were isolated as EV preparation methods.

Purified EVs were diluted to 50 µg/mL with phosphate-buffered saline (PBS), and 10 µL of the EV-suspension were placed on 300-mesh copper grids (EMS, Hatfield, PA, USA) and stained with 2% uranyl acetate on the 300-mesh copper grids. The microscope images were recorded with a JEM1011 electron microscope (JEOL Ltd., Tokyo, Japan).

The protein sample (10 µg) from EVs and whole cell lysate (WCL) were loaded into each well and analyzed with SDS-PAGE (10% resolving gel). The SDS-PAGE gel was stained with CBB R-250 (Bio-Rad Laboratories, Inc., Hercules, CA, USA).

The purified EVs were diluted with PBS to 1 µg/mL and measured using a Zetasizer Nano ZS (Malvern Instruments, Worcestershire, UK) and Dynamic V6 software32.

Immortalized human epidermal keratinocytes (HaCaT cells) were donated by Jeung-Hoon Lee (Chungnam National University, Daejeon, Korea). The 2 × 10⁵ HaCaT cells were seeded on 24-well plate in DMEM (Hyclone Laboratories Inc., South Logan, UT, USA) containing 5% fetal bovine serum (FBS) (Gibco, Grand Island, NY, USA) and 1% antibiotics (Hyclone Laboratories Inc.). Cells were treated with 0.1, 1 and 10 µg/mL EVs for 12 hours to evaluate the immunogenicity of Lp EVs. After treatment, IL-6 levels in the cultured media were evaluated via enzyme-linked immunosorbent assay (ELISA) following manufacturer's instructions (R&D Systems, Minneapolis, MN, USA).

The cells were pretreated with Lp EV for 12 hours, followed by stimulation with S. aureus-induced EVs (Sa EV) for 12 hours. Co-treatment with Lp EV and Sa EV was performed to compare with the pretreatment group. After treatment, the cultured media was evaluated via ELISA by following manufacturer's instruction (R&D Systems). Some of the wells were treated for 24 hours and viability was measured using thiazolyl blue tetrazolium bromide (MTT; Sigma-Aldrich, St. Louis, MO, USA).

Sa EVs were administrated to mouse skin 3 times per week for 4 weeks to develop the mouse model for skin inflammation. Before administration of Sa EV, the skin of each mouse was stripped 5 times using Durapore surgical tape (3M Co., St. Paul, MN, USA). Gauze (1 × 1 cm) was placed on the stripped mouse skin, and Sa EV in 100 µL PBS were sprayed onto the gauze, which was fixed to the skin using Tegaderm bio-occlusive tape (3M Co.). The Lp EV was administered orally 12 hours before each Sa EV administration for the Lp EV-treated group.

Skin tissue from each mouse was sampled to evaluate the production of IL-4 and IL-5 in the mouse model. Skin tissue samples were homogenized in RIPA Buffer (ThermoFisher Scientific, Waltham, MA, USA) and were centrifuged at 10,000 ×g for 20 minutes at 4°C. The supernatant was collected, and IL-4 and IL-5 levels were determined using ELISA according to the manufacturer's instructions (R&D Systems).

Isolated skin tissues were fixed in paraffin, sectioned and stained with hematoxylin and eosin (H&E). The skin tissues were analyzed at 200× magnification.

The results of continuous variables are presented as the mean ± standard deviation. Continuous variables were analyzed using Student's t test or the Mann-Whitney U test, and categorical variables were analyzed using the chi-square test or Fisher's exact test. Statistical significance indicated by a P value of < 0.05. SPSS version 19.0 (SPSS Inc., Chicago, IL, USA) was used for analyses.

Twenty-seven AD patients and 6 normal controls were enrolled. The mean ages of the AD patients and controls were 91 and 83 months, respectively. In both AD and control participants, 66.7% were boys. Eosinophil count and ECP were 877.8 cells/µL and 66.4 µg/L, respectively. Specific IgE to any allergen was positive in 26 patients. Among them, 66.7% were positive for the specific IgE for house dust mites. All patients were treated with wet dressings, topical corticosteroids, antihistamines and antibiotics. Topical calcineurin inhibitors and systemic corticosteroids were used in 66.7% and 18.5% of the patients, respectively. The SCORAD index was significantly decreased from 75.5 to 41.4 after treatment (P < 0.001). Laboratory tests were not conducted in the normal control group (Supplementary Table S1).

Bacterial EV compositions in the urine and serum of the AD patients were compared each other to confirm the similarity of both sample types. The 2 sample types closely resembled each other in 18 AD patients whose serum and urine samples were collected simultaneously (Supplementary Fig. S1A and B). As shown in Supplementary Fig. S1C, the cluster of samples in blood (in blue) cannot be separated from the cluster of samples in urine (in orange), which implies that the bacterial compositions of the 2 groups are quite similar. This interpretation was confirmed further by the t test results with PC1 and PC2 (P value > 0.1). After confirming the similarity of bacterial EV composition in blood and urine by principal component analysis (PCA; Supplementary Fig. S1C), we decided to continue the analysis using urine samples, which were easier to collect.

Bacterial diversity as determined by the Chao 1 index was markedly decreased in AD patients compared to the control group (Fig. 1A). Definite separation of bacterial EV composition between AD patients and the control group was observed in PCA (Fig. 1B). At the phylum level, Firmicutes was dominant in the controls, and Proteobacteria was dominant in AD patients (Fig. 1C). A heatmap of the genus level revealed notable differences in bacterial EV composition in the urine of the control and AD patients (Figs. 1D and 2A). The proportions of Lactococcus, Leuconostoc, Lactobacillus and Lactobacillales(o) were significantly higher in the control group than in the AD patient group, and Alicyclobacillus, Propionibacterium and Streptophyta(o) were more increased in the AD patient group compared with the control group. Pseudomonas was common in both groups (Fig. 2A and B).

Lactobacillus comprised 20.9% and 4.6% of the total flora in the control and AD patient groups, respectively. Respective values were 18.2% and 0.02% for Lactobacillales(o), 14.6% and 0.1% for Leuconostoc, and 11.6% and 0.03% for Lactococcus. Alicyclobacillus comprised 0.3% and 8.5% of the control and AD patient groups, respectively. Other respective values were 2.9% and 7.9% for Streptophyta(o), and 2.8% and 7.6% for Propionibacterium (Fig. 2B, Supplementary Table S2).

Urine samples were collected before and after treatment from 10 AD patients. Then, we compared the changes in bacterial EVs composition based on the treatment in this group. Before the treatment, Alicyclobacillus and Comamonadaceae(f) were frequently found in AD patients. These proportions were decreased after treatment, while those of Acinetobacter and Oxalobacteraceae(f) were increased (Supplementary Fig. S2A and B). The proportions of Alicyclobacillus and Comamonadaceae(f) decreased from 8.7% to 1.9% and from 7.2% to 2.6%, respectively, after the treatment. The proportions of Acinetobacter and Oxalobacteraceae(f) were increased from 0.6% to 12.4% and from 3.0% to 9.2%, respectively, after the treatment (Supplementary Table S3). At the phylum level, Proteobacteria was dominant in both groups, and Firmicutes decreased after treatment. (Supplementary Fig. S2C).

The Lp EV were purified according to EV isolation methods used in the previous study.26 First, we analyzed the morphology of EVs using TEM. We observed spherical lipid bi-layer vesicles (Fig. 3A). The size distribution of EVs was measured using a dynamic light-scattering analyzer. The diameter of EVs ranged from 20 to 100 nm (Fig. 3B). Next, we analyzed the protein patterns of EVs and WCL with SDS-PAGE. The EVs showed distinct protein patterns compared with WCL (Fig. 3C). These results suggest that EVs contain sorted proteins from host cells and that there are sorting mechanisms when EVs are produced.

We assumed that EV from lactic acid bacteria would have some protective or beneficial roles against AD because we found that EVs of the lactobacilli group were dominant in the urine of normal controls compared to that of AD patients. As mentioned above, L. plantarum was selected as an experimental material among the lactic acid bacteria because it was easily obtained from kimchi, which is a familiar Korean food. First, we evaluated the therapeutic or preventive effect of EVs from L. plantarum on keratinocytes and macrophages. The concentration of the Lp EV used to treat the keratinocytes was determined at a level that did not cause skin irritation or inflammation compared to Sa EV, a positive control, which was already known to cause AD-like skin inflammation (Fig. 4A).3031 IL-6 secretion from keratinocyte was decreased with Lp EV treatment prior to Sa EV treatment in a dose-dependent manner, but not in the co-treatment model of Lp EV and Sa EV (Fig. 4A). IL-6 secretion from macrophages was also decreased with Lp EV pre-treatment, while no decrease was observed in the co-treatment group (Supplementary Fig. S3A). Cell viability was also restored with Lp EV pre-treatment on the keratinocytes (Fig. 4B). However, these trends failed to demonstrate statistical significance in macrophages (Supplementary Fig. S3B). These findings suggest that Lp EV could have a preventive effect on skin inflammation.

In the previous study, we demonstrated that Sa EV causes AD-like skin inflammation which was characterized by epidermal thickening with infiltration of the dermis by mast cells and eosinophils, and cutaneous production of IL-4, IL-5, interferon (IFN)-γ and IL-17.30 Epidermal thickening with the infiltration of inflammatory cells, especially eosinophils and increase in type 2 cytokines has been regarded as AD-like skin inflammation in mouse experiments.28303132 In the current study, we administered oral Lp EV to the Sa EV-induced AD mouse model to evaluate the in vivo effects of Lp EV on AD. Lp EV was administered orally for 25 days together with Sa EV application 3 times a week, and the effect was compared with that of dexamethasone treatment (Fig. 5A). Dexamethasone markedly decreased skin Sa EV-induced skin inflammation (Fig. 5B). Although low-dose Lp EVs (1 µg) were not effective in improving of skin inflammation, an increased dose reduced skin inflammation and showed an effect similar to the dexamethasone treatment (Fig. 5B). Histological analysis elucidated that eosinophilic inflammation and epidermal thickening were reduced after dexamethasone treatment. Lp EV administration also reduced epidermal thickening, but not eosinophilic infiltration (Fig. 5C). Elevated IL-4 levels in the positive control were significantly decreased with Lp EV administration (Fig. 5D). These results show that Lp EV may decrease skin inflammation by reducing IL-4 cytokine in a dose-dependent manner.