Ampicillin, neomycin, metronidazole, vancomycin, 2,4-dinitrobenzene (DNCB), and sodium acetate anhydrous were purchased from Sigma Chemical Co. (St. Louis, MO, USA). Mouse enzyme-linked immunosorbent assay (ELISA) kits for immunoglobulin E (IgE) and thymic stromal lymphopoietin (TSLP) were all obtained from Abcam (Cambridge, MA, USA). TRIzol reagent, SYBR Green Master Mix, Microamp Fast Optical 96-Well Reaction Plate, and primary antibody against free fatty acid receptor 3 (FFAR3) as well as peroxidase-conjugated secondary antibodies were acquired from Thermo Fisher Scientific (Wilmington, DE, USA). Primary antibodies against GATA binding factor 3 (Gata3) and β-actin were obtained from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Primary antibody against free fatty acid receptor 2 (FFAR2) was obtained from Alomone Labs (Alomone Labs, Jerusalem, Israel).

The animals used were six-to-eight-week-old male BALB/c mice (20 ± 2 g), which were obtained from the Guangdong Medical Laboratory Animal Center (China) and subjected to a 7-day acclimation period in a controlled animal facility maintained at 22 ± 3°C, with 55 ± 5% humidity, and a 12-h light/dark cycle.

AD-like dermatitis was induced in mice by applying DNCB (Sigma-Aldrich) to the dorsal skin and ears (Figure 1A). After complete hair removal via depilation, 200 μL of 1% (w/v) DNCB solution was challenged for sensitization. Beginning on day 0 (4 days post-sensitization), the ears and dorsal skin were applied with 0.5% DNCB solution three times weekly for three consecutive weeks. The control group was treated with a vehicle. Anesthesia was performed by isoflurane induction. On day 21, the mice were sacrificed using pentobarbital sodium. Blood, dorsal skin, spleens, and colonic tissue samples, as well as intestinal contents, were collected. An electronic digital caliper was used to measure and record mouse ear thickness. Dorsal skin, spleens, and colonic tissue samples, as well as intestinal contents, were collected. Dermatitis scores were assessed at days−5, 0, 7, 14, and 21 by scoring skin lesions according to symptoms, including erythema/bleeding, dryness/scaling, edema, and erosion/scratching. The dermatitis severity was scored from 0–3 points, indicating no, mild, moderate, and severe symptoms, respectively. The final score was the sum of the four symptom scores. An investigator blinded to the grouping conducted the assessments.

We established a mouse model with gut microbiota depletion to investigate the role of the gut microbiota in AD. The specific pathogen-free mice were randomly divided into the following four groups: control (Ctrl), AD model (AD), antibiotic exposure control (ABX), and AD+antibiotic exposure (AD+ABX) groups. The ABX and AD+ABX mice were fed drinking water containing ampicillin (1 g/L), neomycin (1 g/L), metronidazole (1 g/L), and vancomycin (0.5 g/L) ad libitum 2 days before the establishment of the AD model, with antibiotic intervention for 28 days (Hong et al., 2019). Subsequently, we used 16S rRNA sequencing and metabolomics to screen for the differential microbiome components and metabolites in the feces of the AD mice.

After screening and identifying the key SCFA as acetate, the mice were categorized into the following four groups: control (Ctrl), AD model (AD), sodium acetate supplementation (Ac), and AD+acetate supplementation (AD+Ac) groups. We added sodium acetate (NaOAc) (200 mM, Sigma-Aldrich, catalog number: S2889) to the drinking water from the beginning of the experiments (day−5) to day 21 for Ctrl+Ac and AD+Ac mice, similar to the experiments performed with diet (Antunes et al., 2019).

Fresh fecal pellets (~150–200 mg/mouse) were collected and frozen in liquid nitrogen within 1 h after sampling until further analysis. Microbial community DNA were extracted and quantified. The quality of the DNA was verified and the V3-V4 hypervariable region of the bacterial 16S rRNA gene was amplified. Purified PCR amplicons were obtained using Agencourt AMPure XP beads (Beckman Coulter, USA) with an optimized elution buffer, followed by quality control on an Agilent 2100 Bioanalyzer (Agilent Technologies, USA). The DNA samples were sequenced on the DNB MGI-2000 platform (BGI, Shenzhen, China) to generate a 2 × 300 bp paired-end reads.

Raw sequencing data underwent quality control to remove low-quality bases, adapter sequences, and reads with >5% N-content, resulting in high-quality clean reads. The paired-end sequencing data were processed through VSEARCH (version 2.13.6) to generate consensus sequences, employing a stringent 97% similarity criterion, followed by clustering analysis to identify operational taxonomic units (OTUs). Representative OTU sequences were classified according to SILVA or Greengenes 16S rRNA reference database. Microbial alpha diversity was quantified using Faith's phylogenetic diversity metric, while LEfSe analysis (LDA effect size ≥3.0) was conducted to identify discriminative features at different taxonomic ranks.

Fresh fecal pellets from Ctrl and AD mice were analyzed using a Waters UPLC I-Class Plus system (Waters, USA) coupled with a QTRAP 6500 Plus triple quadrupole mass spectrometer (SCIEX, USA) (Liang et al., 2024). The Skyline software suite (MacCoss Lab, University of Washington) was used for targeted metabolite quantification in fecal extracts, followed by multivariate statistical analysis. Prior to principal component analysis (PCA), the data were log2-transformation and normalized using Pareto scaling. To evaluate metabolite-class associations, partial least squares (PLS) regression was implemented to assess the discriminatory power of metabolite expression profiles, enabling the evaluation of their impact on sample classification. Metabolites demonstrating significant differential abundance were determined based on a two-fold threshold criterion (≥1.2-fold increase or ≤ 0.83-fold decrease) with statistical significance set at P < 0.05.

Quantification of SCFA-containing ether fractions was performed using liquid chromatography-mass spectrometry (LC-MS). Following extraction, sample analysis was conducted on a Waters UPLC I-Class Plus system (Waters Technologies, MA, USA) integrated with a QTRAP 6500 Plus mass spectrometer. Concentration determination was achieved through an external standard calibration protocol covering the relevant concentration range.

The chromatographic separation was conducted using a Waters ACQUITY Premier BEH C18 column (2.1 × 50 mm, 1.7 μm) with a gradient elution program configured as detailed below: mobile phase B (containing 0.1% formic acid) was maintained at 15% for 1.0 min, linearly increased to 55% over 4.0 min, and then rapidly reversed to the initial concentration at a rate of 0.5%/min within 1.0 min (5.1–6.0 min). The system was operated under the following optimized parameters: column temperature set to 40°C (midpoint of the manufacturer's recommended range, 20–90°C), and flow rate maintained at 0.35 mL/min.

The QTRAP 6500 Plus mass spectrometer, equipped with a Turbo Ion Spray interface, was operated in negative ion mode with the following source parameters: ion source temperature of 550°C, ion spray voltage at −4,500 V, auxiliary gases (gas I and II) at 50 psi, and curtain gas at 30 psi. Data acquisition was performed using multiple reaction monitoring (MRM), with collision energy and declustering potential optimized for each analyte based on their physicochemical properties. MultiQuant (SCIEX, USA) was used to identify and quantify the metabolites. The resulting data matrix for obtaining information includes metabolite identification and quantitative results was processed for analysis.

IgE and TSLP were determined using ELISA kits with the supernatant obtained from mouse blood samples after centrifugation. Quantitative analysis involved duplicate measurements of absorbance at 450 nm using a standardized microplate reader system.

All dorsal skin tissue samples were fixed with 4% paraformaldehyde, underwent routine dehydration, and was subsequently embedded in paraffin. The prepared paraffin sections were stained with hematoxylin and eosin (H&E) and subjected to histopathological observation using an Olympus BX53 light microscope.

RNA was extracted from the colon, and cDNA synthesis was performed. Quantitative PCR was conducted using SYBR Green Master Mix on a ViiA 7 PCR system (Applied Biosystems). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was selected as the housekeeping reference gene, with the QPCR data quantified through the following formula: Relative mRNA expression = 2^{(Δ Ct target gene−ΔCt GAPDH)}. The primer sequences used were: FFAR3: Forward 5′-AGTCGCCTGGTGTGGATACTGAG-3′, Reverse 5′-GCCGAAGCAGACGAAGAAGATGAG-3′; FFAR2: Forward 5′-GCTGACAGGCTTCGGCTTCTAC-3′, Reverse 5′-CAGAGCAGCGATCACTCCATACAG-3′; GPR109a: Forward 5′-TGAGGCAGAGACAGATGGACAGAC-3′, Reverse 5′-GAGAAGCCAGAAGATGCGGATGC-3′; T-bet: Forward 5′-AGCCGTTTCTACCCCGAC-3′, Reverse 5′-GCTCACAGCTCGGAACTCC-3′; Gata3: Forward 5′-TCTGGAGGAGGAACGCTAATGGG-3′, Reverse 5′-CGGGTCTGGATGCCTTCTTTCTTC-3′; RORγt: Forward 5′-TGTCCCGAGATGCTGTCAAGTTTG-3′, Reverse 5′-TCCTGTTGCTGCTGCTGTTGC-3′; Foxp3: Forward 5′-AAGAATGCCATCCGCCACAACC-3′, Reverse 5′-TACGGTCCACACTGCTCCCTTC-3′; GAPDH: Forward 5′-TCCACTCACGGCAAATTCAAC-3′, Reverse 5′-GTAGACTCCACGACATACTCAGC-3′.

Mouse colon tissues were homogenized and processed for protein extraction, with concentrations subsequently quantified following the manufacturer's guidelines. Protein samples were fractionated by 8–12% SDS-PAGE and subsequently transferred onto polyvinylidene difluoride (PVDF) membranes pre-activated with methanol. Following blocking and washing, the membranes were incubated overnight with anti-Gata3, anti-FFAR3, anti-FFAR2, and anti-β-actin primary antibodies. After incubation with the secondary antibody at room temperature and three washes, protein signals were visualized using a Bio-Rad GelDoc imaging system.

Statistical analysis was performed using GraphPad Prism 8.0.2 (San Diego, CA, USA). The experimental results are presented as mean ± standard error of the mean for normally distributed data, while quartiles were used for non-normally distributed data. One-way analysis of variance, followed by Tukey's post-hoc multiple comparisons, was used. Statistical significance was determined at P < 0.05.

To evaluate whether the gut microbiome mediates the protective effects in AD, we administered oral antibiotics to mice for 28 days. The alpha diversity index reflects the abundance and diversity of microbial communities in samples through five indices, namely observed species, Chao, Ace, Shannon, and Simpson. The observed species, Chao, and Ace indices reflect the species richness index, while the Shannon and Simpson values indicate the species diversity of a community. The complexity of the sample is directly proportional to the first four values and negatively correlated with the Simpson value.

No significant difference was observed in the alpha diversity index between the Ctrl and AD groups, while antibiotic treatment significantly inhibited the gut microbiota of mice, affecting diversity and abundance (Table 1). The antibiotic-treated mice showed dehydration and weight loss in the early stages owing to their resistance to antibiotics in their drinking water. The ABX and AD+ABX mice gradually recovered on the 10th day, and their weights did not differ significantly from the other groups at the final endpoint (data not shown). Skin lesions, ear swelling, and increased serum IgE and TSLP expression are the main clinical symptoms of AD. The Ctrl and ABX groups did not receive DNCB, and their dermatitis scores were zero at all time points. Dermatitis (p > 0.05) and ear thickness (p < 0.05) were more severe in the AD+ABX mice than in the control group (Figures 1B, C). Hair regrowth in most mice after day 14 hindered accurate assessment of erythema and hemorrhage due to limitations in the dermatitis scoring system. The IgE and TSLP levels increased significantly after ABX treatment with AD mice (Figure 1D). Therefore, reduction in skin inflammation and allergies depends on the preserved intestinal microbiota.

The changes in the microbiome bacteria with proportions ≥9% at the six levels (phylum, class, order, family, genus, and species) are shown in Table 2 and Figure 2. 16S rRNA sequencing of fecal samples revealed that antibiotic intervention led to a significant reduction in microbial diversity by over 90% (Table 2). Phylum-level analysis showed a significant microbial compositional alteration in AD models, characterized by reduced Bacteroidetes and elevated Firmicutes, suggesting a potential shift in microbiota composition associated with AD pathogenesis. At the order level, we observed a similar change in the Bacteroidales/Clostridiales ratio (p < 0.01). In AD mice, Bacteroidia was significantly under-represented, and Clostridia was significantly over-represented at the class level. The abundance of Porphyromonadaceae did not change significantly in the AD group. In contrast, at the family, genus, and species levels, Prevotellaceae, Rikenellaceae, Alistipes, Alloprevotella, and Alloprevotella rava were significantly reduced, whereas Lachnospiraceae, and Clostridium XlVa were significantly over-represented. To further identify the specific commensal bacteria associated with AD, we compared the microbial compositions of the Ctrl and AD groups. LDA demonstrated significant over-representation of Bacteroidetes, Bacteroidia, Bacteroidales, Alistipes, Rikenellaceae, Prevotellaceae, Comamonadaceae, Anaerotruncus, Alphaproteobacteria, Anaerovorax, Clostridiales, Butyrivibrio, Paraprevotella, Deferribacteres, Deferribacteraceae, Mucispirillum, Deferribacteres, and Deferribacterals in the Ctrl group and significant over-representation of Firmicutes, Clostridiales, Clostridia, Lachnospiraceae, Clostridium XlVa, Acetatifactro, and Sporobacter in the AD group (Figure 3B). The LDA effect sizes (Figure 3C) were consistent with the microbiome composition analysis. The above data showed a significant change in the ratio of the Bacteroides/Firmicutes in AD mice.

Partial least squares discriminant analysis (PLS-DA) and PCA model of Ctrl and AD samples showed good stability and repeatability (Figures 3A, B). Differential metabolite screening revealed significant alterations between two groups, as shown in Figure 3C and Table 3. Specifically, five metabolites were markedly upregulated, while six were notably downregulated. Compared with the Ctrl group, the expression levels of 3,4-Dihydroxymandelic acid, 3-Mercaptolactic acid, Indole-3-propionic acid, Pentacosanoic acid, Amygdalin, and Acetate were significantly under-represented in AD mice, while 12-Ketolithocholic acid acetate, Etiadienic Acid, Stearoylcarnitine, Palmitoylcarnitine, and Arachidonic acid were significantly over-represented. Among the differential metabolite, the expression of acetate was the highest. We further examined the SCFA levels in AD mice using LC-MS analysis. Gut microbiota-derived SCFA profiles revealed a marked decrease in acetate concentrations along with total SCFA levels in AD model mice, whereas propionate, butyrate, valerate, caproate, isobutyrate, and isovalerate remained at comparable levels relative to healthy controls (Figure 3D). These observations imply that acetate, serving as a principal microbial metabolic product, may function as a critical modulator in the pathophysiological progression of AD.

We explored whether acetate is involved in the mechanism of AD. Currently, no animal model for AD fully replicates the complexity of human clinical conditions. However, DNCB-induced AD mice are widely used due to their skin lesion characteristics and Th2 polarization, which closely resemble those of human AD, making them suitable for the objectives of this study. After supplementing with acetate, the acetate content in the Ac group significantly increased, and no significant difference was observed between the AD+Ac and Ctrl group, indicating that the supplementation was effective (data not shown). Additionally, acetate supplementation did not cause any sensitivity reaction or redness in the skin of mice (Figures 4A, B).

Acetate supplementation alleviated the inflammation symptoms of AD mice, as evidenced by thinner skin, decreased lesion, and significantly reduced spleen weight (Figures 4A, B, D). As shown in Figure 4B, the AD group exhibited significant epidermal pathology compared to controls, including increased epidermal thickness, spinous layer hyperplasia, spongiotic edema with inflammatory infiltration, and impaired keratinization. In the AD+Ac group, these pathological changes were measurably ameliorated, as evidenced by reduced epidermal hyperplasia, partially restored keratinization, attenuated spongiosis, and diminished inflammatory cell infiltration. At the end of the intervention period, the AD+Ac mice exhibited significantly lower dermatitis scores than the AD group did (Figure 4C). Acetate supplementation in the AD group significantly decreased the IgE and TSLP levels (Figure 4E). Thus, acetate might be a critical modulator in DNCB-induced AD mice. To better understand the interaction between gut microbiota and acetate, we conducted 16S rRNA sequencing analysis to characterize microbial community alterations following acetate supplementation. We found that acetate supplementation did not affect gut microbiota composition or relative abundances (Figure 5A). The gut microbiota can affect the level of acetate, but an increase in acetate concentration within a certain concentration range does not affect the gut microbiota, indicating that the gut microbiota and acetate may have a unidirectional causal relationship.

SCFAs have been reported to activate FFAR3 (previously termed GPR41), FFAR2 (previously termed GPR43), and GPR109a, with FFAR2 agonists shown to ameliorate DNCB-induced AD in mice (Moniri and Farah, 2021; Kang and Im, 2020). We examined the relative expression of SCFAs receptors (ffar2, ffar3, gpr109a) and T-helper cell transcription factors t-bet (Th1, T helper type 1), gata3 (Th2), rorγt (Th17), and foxp3 (T regulatory) following acetate supplementation. The results revealed that ffar2 mRNA expression was reduced in the colon of AD mice (p < 0.01), whereas gata3 mRNA expression was significantly increased (Figure 5B). After acetate supplementation, ffar2 mRNA expression was significantly upregulated, and gata3 mRNA expression was significantly downregulated. No significant differences were detected in the mRNA expression of ffar3, gpr109a, t-bet, rorγt, and foxp3 in the colon across all experimental groups. At the protein level, GATA3 expression was markedly upregulated in AD mice relative to controls, whereas FFAR2 levels were significantly diminished, and FFAR3 expression remained unaltered (Figure 5C). Hence, the therapeutic mechanism of acetate in AD is related to the FFAR2 and Th2 (Gata3) pathways.