Multi-omics and functional validation reveal that Methanobrevibacter-derived L-3-aminoisobutyrate alleviates subclinical mastitis in dairy goats via the HSPA1B-p65 signaling pathway

Abstract

Background

Subclinical mastitis (SCM) is prevalent in dairy livestock and compromises milk quality and lactation performance. Although often attributed to bacterial infection, many cases lack identifiable pathogens, suggesting alternative mechanisms. While evidence supports a gut-mammary gland axis, the microbial drivers and microbiota-derived metabolites linking gut dysbiosis to SCM remain unclear. Here, we aimed to identify SCM-associated gut microbial markers, prioritize candidate therapeutic metabolites and define the underlying mechanism.

Results

Based on differences in somatic cell count (SCC) and inflammatory phenotypes across a cohort of 167 mid-lactation Saanen dairy goats, we selected 6 healthy and 6 SCM goats for downstream analyses. By integrating metagenomics, metabolomics, cross-species fecal microbiota transplantation (FMT) and functional validation in vitro and in vivo, we found that SCM was accompanied by reduced milk yield and heightened inflammatory signatures. Compared with the Healthy group, SCM goats exhibited marked remodelling of the gut microbiota, with enrichment of opportunistic taxa (Eubacterium and Blautia) and a pronounced depletion of archaeal Methanobrevibacter spp. Notably, FMT from SCM donors recapitulated mammary inflammatory phenotypes in mice, supporting a causal contribution of gut dysbiosis to mammary inflammation. Joint metagenomic functional profiling and metabolomics further identified the branched-chain amino-acid-derived metabolite L-3-aminoisobutyrate (BAIBA) as significantly enriched in the gut of healthy goats. Moreover, Methanobrevibacter spp. harboured key enzyme genes (vorA, vorB and vorD) implicated in BAIBA biosynthesis. In an LPS-challenged MAC-T model, BAIBA attenuated mammary epithelial inflammation by activating endoplasmic reticulum protein quality control programmes and restoring HSPA1B expression, thereby suppressing NF-κB activation and reducing pro-inflammatory cytokine production. Finally, in naturally infected goats, intramammary administration of BAIBA lowered SCC, highlighting translational potential.

Conclusions

This study identifies BAIBA as a microbiota-derived metabolite that protects against SCM by restraining mammary inflammation via the HSPA1B-NF-κB axis, establishing a mechanistic gut-mammary link and highlighting a potential non-antibiotic intervention strategy.

Download video file ^{(54.5MB, mp4)}

Video Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40168-026-02464-z.

Introduction

Mastitis, particularly subclinical mastitis (SCM), is one of the most challenging subclinical diseases faced by the global dairy livestock industry [1]. Its prevalence is significantly higher than that of clinical mastitis, with clinical symptoms being subtle, and diagnosis primarily relying on an elevated somatic cell count (SCC) in milk [2]. Subclinical mastitis often leads to adverse consequences including impaired milk quality [3], reduced lactation performance [4], and early culling of dairy livestock [5], ultimately causing significant economic losses. However, in some SCM cases, traditional microbiological diagnostic methods fail to identify the causative pathogens [6]. Mammary tissue damage and immune activation can still be observed even in the absence of typical pathogenic bacterial infections, suggesting atypical or endogenous pathogenic mechanisms [7, 8].

The gut microbiota has gained widespread attention as a key regulator of host immune homeostasis [9], nutrient metabolism [10], and epithelial barrier function [11]. The “gut-X” axis theory posits that gut microbes can influence distal organs through circulating microbially derived metabolites, thereby modulating inflammatory responses and tissue homeostasis [12]. In parallel, the gut microbiota may affect systemic immunity by altering intestinal barrier integrity and regulating immune cell activation and trafficking, ultimately shaping the functions of distant organs such as the lung [13], brain [14] and liver [15]. Building on this, some studies have proposed the concept of a “gut-mammary gland” axis, suggesting that gut dysbiosis may promote the development of mastitis through immune-mediated or metabolic pathways [16–18]. This concept is particularly attractive because it offers a route to intervention through modulating the gut ecosystem rather than targeting the mammary gland directly. However, three major gaps remain in existing studies. First, there is a lack of systematic identification of endogenous microbiota associated with SCM. Second, the functional profile of gut microbial-derived metabolites in healthy dairy livestock remains incompletely understood. Third, the molecular mechanisms by which microbiota-derived metabolites regulate local immune responses in the mammary gland are still unclear. Systematically identifying SCM-associated microbial markers could enable early risk stratification for atypical SCM, whereas defining the functional landscape of microbiota-derived metabolites in healthy goats may reveal mechanism-guided, non-antibiotic intervention avenues to support mammary health.

Dairy goats are known for their high-quality dairy products [19], lactation physiology similar to that of dairy cows [20], and enhanced sensitivity to immune and lactation-related changes during SCM [21]. As such, they serve as an ideal animal model for studying the atypical pathogenic mechanisms of SCM. Based on this, our study focused on a cohort of mid-lactation Saanen dairy goats with consistent parity and uniform feeding conditions. Healthy and SCM individuals were selected based on SCC in milk. By integrating metagenomics, metabolomics, and cross-species fecal microbiota transplantation (FMT) experiments, we systematically investigated the pathogenic potential of gut microbiota in SCM development. Furthermore, we identified functional microbial-derived metabolites and elucidated their molecular mechanisms in regulating mammary gland health.

Results

SCM dairy goats exhibit characteristic mastitis pathology and systemic inflammatory phenotype

To assess the impact of SCM on mammary gland secretory function and systemic inflammation in dairy goats, we selected 6 healthy individuals (Healthy) with normal SCC and 6 SCM individuals (Mastitis) with significantly elevated SCC from a cohort of 167 mid-lactation Saanen dairy goats (Fig. 1A and Additional file 1: Table S1). The results showed that the SCC in the Mastitis group was significantly higher than in the Healthy group (approximately 110-fold, P < 0.001; Fig. 1B), accompanied by a decrease in milk yield (17.2% reduction) and changes in milk composition (increased milk protein percentage and decreased lactose percentage) (P < 0.05; Additional file 2: Fig. S1A, B), indicating impaired mammary gland secretion function. Moreover, serum globulin (GLOB) levels were significantly higher in SCM individuals (P < 0.05; Additional file 3: Table S2), and elevated levels of inflammation markers, including lipopolysaccharide (LPS), interleukin-6 (IL-6), interleukin-1β (IL-1β), and myeloperoxidase (MPO), were observed in both serum and milk (P < 0.05; Fig. 1C–J). These findings indicate that SCM is associated with reduced lactation performance and heightened systemic immune activation, reflecting the multilayered immune response characteristics observed in SCM.

Fig. 1.

Subclinical Mastitis (SCM) triggers systemic immune responses in lactating dairy goats. A From the cohort of 167 mid-lactation Saanen dairy goats, we selected 6 goats with SCM (Mastitis, n = 6) and 6 Healthy goats (Healthy, n = 6) for downstream analyses. B SCM goats exhibited a significant increase in somatic cell count (SCC) in milk (n = 6/group). Statistical significance was determined using Student’s t-test. ***P < 0.001. Comparative analyses of milk levels of MPO (C), IL-6 (D), IL-1β (E), and TNF-α (F), as well as serum levels of LPS (G), IL-6 (H), IL-1β (I), and TNF-α (J), revealed a pronounced inflammatory profile in the Mastitis group (n = 6/group), indicating that SCM elicits systemic inflammatory responses beyond the mammary gland. Statistical significance was determined using Student’s t-test. *P < 0.05, **P < 0.01. Box plots show the median and interquartile range, and dots represent individual goats. Abbreviations: MPO, Myeloperoxidase; IL-6, Interleukin-6; IL-1β, Interleukin-1β; TNF-α, Tumor necrosis factor-α; LPS, Lipopolysaccharide

Identification of gut microbial signatures associated with mammary health status

Intestinal dysbiosis has been proposed to be associated with the onset of mastitis in dairy livestock [16, 22, 23]. To systematically identify microbial biomarkers closely associated with mammary gland status in healthy and SCM dairy goats, we performed metagenomic sequencing analysis on faecal samples from the Healthy and Mastitis groups. The results revealed that the α-diversity of gut microbiota in the Mastitis group was significantly higher (Simpson index, P < 0.05; Fig. 2A), and the community structure also showed significant differences in the Bray–Curtis distance matrix (PERMANOVA, R² = 0.227, P = 0.007; Fig. 2B).

Fig. 2.

Distinct gut microbial profiles in Mastitis and Healthy dairy goats. A α-diversity, as measured by the Simpson index, was significantly higher in the gut microbiota of the Mastitis group compared with the Healthy group (n = 6/group). Statistical significance was determined using the Wilcoxon rank-sum test. *P < 0.05. B PCoA visualized the compositional differences between groups, with Bray–Curtis dissimilarity used for distance calculation and statistical significance assessed by PERMANOVA (n = 6/group). Microbiota composition was compared between the Mastitis and Healthy groups at the domain (C), phylum (D), family (E), and genus (F) levels (n = 6/group). Statistical significance was determined using the Wilcoxon rank-sum test. *P < 0.05, **P < 0.01. G Species-level differences within the genus Methanobrevibacter between the Mastitis and Healthy groups were evaluated using the Wilcoxon rank-sum test (n = 6/group). *P < 0.05, **P < 0.01. H LEfSe analysis further identified signature species distinguishing the Mastitis and Healthy groups, with LDA scores indicating significant taxonomic enrichment (P < 0.05, |LDA|> 3), treating all samples as independent (n = 6/group). I Associations between microbial signatures and key subclinical mastitis phenotypes were assessed using Spearman correlation analysis (|R|> 0.5, P < 0.05) in the Mastitis and Healthy groups (n = 6/group). Box plots show the median and interquartile range, and dots represent individual goats. Bar plots show mean ± SEM. Abbreviations: M., Methanobrevibacter; SCC, somatic cell count; GLOB, Globulin; MPO, Myeloperoxidase; IL-6, Interleukin-6; IL-1β, Interleukin-1β; TNF-α, Tumor necrosis factor-α; LPS, Lipopolysaccharide

Microbial taxonomy analysis showed that the Mastitis group had a higher relative abundance of bacteria and a lower abundance of archaea than the Healthy group (P < 0.05; Fig. 2C). The abundance of archaea was also significantly negatively correlated with SCC (|R|> 0.5, P < 0.05; Additional file 2: Fig. S2). At the phylum level, Euryarchaeota (archaeal phylum), Verrucomicrobia, and Planctomycetota were notably more abundant in the Healthy group, whereas Thermotogae was significantly more prevalent in the Mastitis group (P < 0.05; Fig. 2D). At the family level, Methanobacteriaceae (archaeal family), Akkermansiaceae, and Thermoguttaceae were significantly enriched in the Healthy group, while the Mastitis group showed a marked increase in Lachnospiraceae, Eubacteriaceae, and Thermotogaceae (P < 0.05; Fig. 2E). At the genus level, the Healthy group exhibited notable enrichment in Methanobrevibacter (archaeal genus), Akkermansia, and Solobacterium, while the Mastitis group was predominantly enriched in Anaerotignum, Eubacterium, Blautia, and Lachnospira (P < 0.05; Fig. 2F). Based on differential analysis and linear discriminant analysis effect size (LEfSe), characteristic species of the Healthy group included Methanobrevibacter sp., M. cuticularis, M. oralis, and M. smithii (P < 0.05; Fig. 2G), as well as Akkermansia sp. and its CAG:344 subpopulation (linear discriminant analysis [LDA] > 3, P < 0.05; Fig. 2H). In contrast, the gut microbiota of SCM individuals was enriched with Eubacterium spp., Blautia spp., Ruminococcus spp., and Coprococcus spp. (LDA > 3, P < 0.05; Fig. 2H).

Based on microbiota correlation network analysis, species enriched in the Healthy group were found to be negatively correlated with high-abundance taxa in the Mastitis group (|R|> 0.5, P < 0.05; Additional file 2: Fig. S3), suggesting distinct co-occurrence patterns between the two groups. Further Spearman correlation analysis revealed that the abundance of Methanobrevibacter spp. and Akkermansia spp. was significantly negatively correlated with milk SCC and the concentration of pro-inflammatory cytokines (such as IL-1β, tumor necrosis factor-α [TNF-α], IL-6) (|R|> 0.5, P < 0.05; Fig. 2I), suggesting that these taxa were associated with a less inflammatory phenotype. In contrast, bacterial species enriched in the Mastitis group showed a significant positive correlation with these inflammatory markers (|R|> 0.5, P < 0.05; Fig. 2I), indicating their potential association with the inflammatory state of SCM. Overall, these results identify Methanobrevibacter spp. and Akkermansia spp. as health-associated gut taxa in dairy goats, whereas Eubacterium spp. and several other bacterial taxa were enriched in SCM individuals. Together, these taxa may serve as candidate microbial biomarkers of the Healthy or SCM state.

Cross-species fecal microbiota transplantation supports a causal contribution of SCM-associated gut microbiota to mammary inflammation

To validate the pathogenic potential of gut microbiota from Healthy and SCM dairy goats in the context of mammary inflammation, we established an antibiotic-treated pseudo-germ-free mouse model using a composite antibiotic regimen (ABX) to deplete the host native gut microbiota. ABX-treated mice receiving PBS served as the negative control (NC). We then transplanted fecal microbiota from Healthy dairy goats (H-FMT) or SCM dairy goats (M-FMT) into recipient mice, respectively (Fig. 3A). The results indicated that the mammary gland tissue in the H-FMT group maintained its structural integrity, with well-organized acini and no obvious inflammatory cell infiltration, closely resembling the morphology observed in the NC group. In contrast, the mammary gland tissue of the M-FMT group exhibited pronounced inflammatory features, including disorganized acinar architecture, neutrophil infiltration, and interstitial fibrosis, which are hallmark pathological alterations typically associated with mammary gland inflammation (Fig. 3B). Further analysis of immune markers revealed that the activity of the neutrophil-specific marker MPO was significantly increased in the mammary tissue of the M-FMT group, and pro-inflammatory cytokines, including IL-1β and TNF-α, were also significantly upregulated (P < 0.01; Fig. 3C–E). Additionally, compared to the H-FMT and NC groups, the M-FMT group showed significantly higher levels of IL-6, IL-1β, and TNF-α in serum (P < 0.01; Fig. 3F–H), suggesting that SCM-associated microbiota not only induced typical local inflammatory responses in the mammary gland but also drove systemic immune activation. Notably, the cross-species fecal microbiota transplantation model successfully induced mammary gland inflammation that closely resembled the SCM phenotype, strengthening the evidence supporting a causal contribution of gut microbiota to SCM-associated mammary inflammation.

Fig. 3.

Fecal microbiota transplantation from goats to mice links SCM-associated gut microbiota to mammary gland inflammation. A Schematic diagram illustrating the goat-to-mouse FMT protocol. Female BALB/c mice were treated with a 20-day composite antibiotic regimen (ABX) to deplete resident gut microbiota and then orally gavaged with PBS (NC), fecal microbiota from SCM dairy goats (M-FMT), or fecal microbiota from Healthy dairy goats (H-FMT) for 7 days (n=6/group). B Representative H&E-stained images of mouse mammary gland tissue after FMT. Scale bars, 200 μm. Concentrations of MPO (C), IL-1β (D), and TNF-α (E) in mammary gland tissue after FMT (n=6/group). Serum concentrations of IL-6 (F), IL-1β (G), and TNF-α (H) after FMT (n=6/group). I Absolute quantification of fecal bacterial load at the end of acclimation, after ABX treatment, and after FMT (n=6/group). Different lowercase letters indicate significant differences among groups. J PCoA based on Bray–Curtis distances showing differences in fecal microbiota composition between M-FMT and H-FMT mice after FMT (n=6/group). Statistical significance was assessed using PERMANOVA. Differentially abundant taxa between M-FMT and H-FMT groups at the phylum (K), family (L), and genus (M) levels after FMT (n=6/group). N Absolute abundance of Methanobrevibacter in fecal samples after FMT (n=6/group). O Spearman correlations between selected microbial taxa and inflammatory markers in mammary gland tissue and serum after FMT (n=6/group). Statistical significance was determined using one-way ANOVA followed by Tukey’s test for panels C–I and N, the Wilcoxon rank-sum test for panels K–M, and Spearman correlation analysis for panel O. *P < 0.05, **P < 0.01, ***P < 0.001; n.s., not significant. Box plots show the median and interquartile range, and dots represent individual mice. Bar plots show mean ± SEM. Abbreviations: MPO, Myeloperoxidase; IL-6, Interleukin-6; IL-1β, Interleukin-1β; TNF-α, tumor necrosis factor-α

To validate the establishment of the FMT model and assess donor-derived microbial reconstitution in recipient mice, we further profiled the fecal microbiota after FMT. Absolute quantification showed that ABX treatment markedly reduced the total fecal bacterial load, confirming successful generation of pseudo-germ-free mice (P < 0.05; Fig. 3I). After FMT, the PBS-treated group showed partial spontaneous recovery of the gut microbiota; however, its total bacterial load remained significantly lower than those of the M-FMT and H-FMT groups, reaching only approximately 1/25.0 and 1/25.7 of their levels, respectively (P < 0.05; Fig. 3I). By contrast, both M-FMT and H-FMT markedly restored fecal bacterial loads, indicating that goat-derived FMT effectively promoted gut microbial reconstitution in recipient mice (Fig. 3I). Further 16S rRNA gene sequencing showed that FMT significantly increased the Shannon diversity of the gut microbiota after ABX treatment (P < 0.05), with the M-FMT group exhibiting significantly higher diversity than the H-FMT group (P < 0.05; Additional file 2: Fig. S4). This suggests that the extent of microbial reconstitution differed according to donor source. Principal coordinates analysis (PCoA) based on Bray–Curtis distances showed that FMT significantly altered the gut microbial structure of pseudo-germ-free mice (PERMANOVA, R² = 0.394, P = 0.002; Additional file 2: Fig. S5), with a clear separation between the M-FMT and H-FMT groups (PERMANOVA, R² = 0.153, P = 0.042; Fig. 3J). These findings indicate that FMT reshaped the recipient gut microbiota in a donor-dependent manner. At the phylum level, the NC group was dominated by Pseudomonadota, consistent with the rapid expansion of this phylum in low-competition ecological niches after ABX treatment [24] (Additional file 2: Fig. S6A). In contrast, the M-FMT and H-FMT groups were dominated by Bacteroidota and Bacillota, resembling the phylum-level composition observed in the dairy goat donor cohort (Additional file 2: Fig. S6A, B). Collectively, these results demonstrate that goat-derived FMT restored the total bacterial load and microbial diversity of ABX-treated mice and partially induced donor-dependent remodeling of the recipient gut ecosystem.

Subsequent differential taxonomic analysis showed that the microbial signatures observed in recipient mice were partly consistent with those identified in the dairy goat donor cohort. Taxa enriched in Healthy donors, including Verrucomicrobiota, Akkermansiaceae and its representative genus Akkermansia, were also significantly increased in the H-FMT group (P < 0.05; Figs. 2D–F; Fig. 3K–M). In contrast, Lachnospiraceae and its associated genera Blautia and Enterocloster, which were enriched in SCM donors, were significantly increased in the M-FMT group (P < 0.05; Figs. 2E, F; Fig. 3L, M). Notably, Methanobrevibacter, the most prominent archaeal genus enriched in Healthy goats, was negatively correlated with SCC and pro-inflammatory cytokine levels in the donor cohort (Fig. 2F, I). Genus-specific qPCR further showed that the absolute abundance of Methanobrevibacter in mouse feces was significantly lower in the M-FMT group than in the H-FMT group after FMT (P < 0.05; Fig. 3N), suggesting donor-dependent differential retention of this archaeal feature in recipient mice. Spearman correlation analysis further showed that Akkermansia and Methanobrevibacter were negatively correlated with pro-inflammatory cytokine levels in mammary tissue and serum, whereas Blautia and Enterocloster were positively correlated with these inflammatory markers (|R|> 0.5, P < 0.05; Fig. 3O). These results indicate that goat-derived FMT did not merely induce nonspecific microbial recovery, but partially reconstructed donor-enriched bacterial and archaeal features associated with distinct inflammatory states in recipient mice.

Integrated metabolic network analysis reveals BAIBA as a critical mediator alleviating subclinical mastitis

To systematically characterize the differences in intestinal microbiota metabolic functions between healthy and SCM states, and to identify potential anti-inflammatory metabolites, we integrated metagenomic functional annotation with untargeted fecal metabolomics data. Comparative analysis based on the KEGG database revealed significant differences in the microbiota functional profiles at both the KO level (R² = 0.200, P = 0.027) and the Module level (R² = 0.180, P = 0.048) between the Healthy and Mastitis groups (Fig. 4A, B). Z-score functional enrichment analysis indicated significant enrichment of pathways involved in the Valine, leucine and isoleucine degradation (ko00280), Lysine degradation (ko00310), D-Amino acid metabolism (ko00470), and Butanoate metabolism (ko00650) in the Healthy group. Conversely, the Mastitis group showed significant enrichment of pathways related to carbohydrate metabolism and biofilm formation, including Starch and sucrose metabolism (ko00500), Galactose metabolism (ko00052), Fructose and mannose metabolism (ko00051), Amino sugar and nucleotide sugar metabolism (ko00520), and biofilm formation-Escherichia coli (ko02026) (Z-score > 1.65; Fig. 4C). In addition, a total of 1,643 metabolites were identified through metabolomics, with Benzene and substituted derivatives (16.92%), Organic acid and Its derivatives (14.18%), and Amino acid and Its metabolites (13.63%) being the predominant categories (Fig. 4D). PCoA analysis confirmed a clear separation of the metabolic profiles between the Healthy and Mastitis groups (Bray–Curtis; PERMANOVA, R² = 0.112, P = 0.032; Fig. 4E). Focused KEGG pathway enrichment analysis also identified the Valine, leucine and isoleucine degradation (ko00280) as significantly enriched (P < 0.05; Fig. 4F). Notably, the key end product of this pathway, L-3-aminoisobutyrate (BAIBA), was highly enriched in the Healthy group and was one of the metabolites with the largest differential fold change (FC) (P < 0.01; Fig. 4G and Additional file 4: Table S3).

Fig. 4.

Comparative functional profiling of gut microbiota reveals L-3-aminoisobutyrate (BAIBA) as a characteristic microbial metabolite in healthy dairy goats. Principal coordinates analysis (PCoA) based on KO (A) and KEGG Module (B) levels revealed significant differences in the functional profiles of gut microbiota between Mastitis and Healthy groups (n = 6/group). Bray–Curtis dissimilarity was used for distance calculation, and statistical significance was assessed by PERMANOVA. C KEGG enrichment analysis (Z-score > 1.65) identified distinct microbial pathways between Mastitis and Healthy groups (n = 6/group). Red and green dots represent pathways enriched in the Mastitis and Healthy groups, respectively. D Class I distribution of fecal metabolites in the Mastitis and Healthy groups (n = 6/group). E PCoA analysis of metabolite profiles further confirmed significant differences in gut metabolic functions between Mastitis and Healthy groups (n = 6/group), evaluated using Bray–Curtis dissimilarity and PERMANOVA. F KEGG Level 3 enrichment analysis based on differential abundance (DA) scores revealed significant alterations in microbial metabolic pathways. The x-axis denotes DA scores, dot size reflects the number of differentially enriched metabolites, and P-values for each pathway were calculated using the hypergeometric test (P < 0.05). G Scatter plots showing the top 10 significantly upregulated (P < 0.05) fecal metabolites in the Mastitis group (orange) and Healthy group (green), respectively. Statistical significance was determined using Student’s t-test. H Overview of the BAIBA biosynthesis pathway. Orange nodes represent metabolites with no significant difference in relative density, while green nodes indicate significantly elevated levels in the Healthy group. The colour gradient above the arrows reflects increasing expression levels of microbial enzyme genes from orange to green. At left, the stacked bar chart shows the taxonomic assignment of the rate-limiting enzyme genes in the L-Valine-derived BAIBA pathway (vorA, vorB and vorD). I The bar plots compare L-Valine and BAIBA concentrations in the Mastitis and Healthy groups. Statistical significance was determined using Student’s t-test. *P < 0.05, **P < 0.01. Data involving error bars are presented as mean ± SEM. Abbreviations: Val, Valine; leu, leucine; isoleu, isoleucine; BAIBA, L-3-aminoisobutyrate; M., Methanobrevibacter

To trace the origin of BAIBA synthesis, we examined the expression of relevant microbial enzyme genes. The results revealed that the gut microbiome of dairy goats possesses the complete enzymatic machinery to synthesize BAIBA from L-Valine. Moreover, key rate-limiting enzyme genes involved in this synthesis, including vorA (FC = 1.62), vorB (FC = 1.78), and vorD (FC = 1.53), were highly expressed in the Healthy group (Fig. 4H; Additional file 5: Table S4). Phylogenetic analysis of KOs indicated that these enzymes are predominantly encoded by Methanobrevibacter sp., a species that exhibited significantly higher abundance in the Healthy group (P < 0.05; Fig. 2G), ultimately leading to the accumulation of BAIBA (P < 0.01; Fig. 4I). Subsequent Spearman correlation analysis further confirmed that the relative abundance of BAIBA was significantly negatively correlated with SCC and pro-inflammatory cytokine concentrations (e.g., IL-1β, TNF-α) (|R|> 0.5, P < 0.05; Additional file 2: Fig. S7), suggesting its potential anti-inflammatory effects through modulation of the host immune response. Importantly, the integrative multi-omics analyses identified Methanobrevibacter sp. as the principal microbial source of BAIBA and indicated that this metabolite exerts a negative immunoregulatory role within the “gut-mammary gland” axis, thereby nominating BAIBA as a microbiota-derived bioactive with translational potential.

BAIBA negatively regulates NF-κB signaling to suppress mammary epithelial inflammatory response in an HSPA1B-dependent manner

To further investigate the molecular mechanisms underlying the regulation of mammary gland inflammation by BAIBA, we employed an LPS-induced bovine mammary epithelial cell (MAC-T) inflammation model to systematically evaluate its effects on cell proliferation and immune modulation potential. Cell viability assays indicated that exposure to BAIBA at 5–20 µM for 6–24 h did not impair MAC-T proliferative capacity (Additional file 2: Fig. S8), suggesting its favourable biocompatibility. The inflammation model was established by treating MAC-T with 50 μg/mL LPS for 12 h (P < 0.05; Additional file 2: Fig. S9). BAIBA intervention at 10 μM significantly downregulated the levels of pro-inflammatory cytokines (IL-1β, IL-6, TNF-α) while simultaneously upregulating the anti-inflammatory cytokine IL-10 (P < 0.001; Additional file 2: Fig. S10), indicating its potent anti-inflammatory activity.

To identify the key signalling pathways and regulatory proteins involved in the action of BAIBA, we performed DIA quantitative proteomics on MAC-T under normal (MAC-T), LPS-stimulated (LPS), and BAIBA-treated (BAIBA) conditions. A total of 9,307 proteins were identified (Fig. 5A). GO functional annotation revealed that the identified proteins were primarily involved in cellular processes, cellular anatomical entities, and binding (Additional file 2: Fig. S11). PCoA in conjunction with subcellular localization analyses indicated that LPS stimulation and BAIBA treatment significantly altered the cytoplasmic and nuclear proteomes (R² = 0.463, P = 0.001; Figs. 5B and Additional file 2: Fig. S12). Differential comparison further revealed that LPS treatment induced the upregulation of pro-inflammatory proteins such as BOLA-DQA5 [25], while significantly downregulating the expression of negative regulators, including ZNF423 [26], LF [27], and HSPA1B [28] (P < 0.05). Following BAIBA intervention, anti-inflammatory proteins such as LF [27], HSPA1B [28], and the deubiquitinase OTUD4 [29] were significantly upregulated, while ID1 [30], which is associated with tissue fibrosis, was significantly downregulated (P < 0.05; Fig. 5C). KEGG pathway enrichment analysis indicated that proteins with differential expression in the LPS group were significantly enriched in several inflammation-related and disease-related pathways, including the p53 signalling pathway (ko04115), Antigen processing and presentation (ko04612), and MicroRNAs in cancer (ko05206) (P < 0.05; Additional file 2: Fig. S13). In contrast, the proteins whose expression was significantly restored by BAIBA were mainly enriched in the Protein processing in endoplasmic reticulum (ko04141), involving molecular chaperones (HSPA1B, HSPA1L, HYOU1), misfolded protein recognition factors (EDEM1, MAN1B), and the ubiquitination execution system (UBE2J1). These results suggest that BAIBA may alleviate mammary inflammation by enhancing the clearance of misfolded and unassembled proteins under inflammatory conditions (P < 0.05; Figs. 5D and Additional file 2: Fig. S14).

Fig. 5.

L-3-aminoisobutyrate (BAIBA) attenuates LPS-induced inflammatory phenotypes in bovine mammary epithelial cells (MAC-T) via the HSPA1B-p65 signalling axis. A DIA-based proteomic profiling identified shared and unique proteins across control MAC-T, LPS-induced inflammatory MAC-T (LPS), and BAIBA-treated inflammatory MAC-T (BAIBA) (n = 4/group). B PCoA analysis visualised the protein expression profiles of the three groups, with Bray–Curtis dissimilarity used for distance calculation and statistical significance assessed by PERMANOVA (n = 4/group). C Volcano plots illustrate differentially expressed proteins (DEPs; FC > 1.2; P < 0.05) from pairwise comparisons among the MAC-T, LPS, and BAIBA groups. Green dots indicate significantly upregulated proteins in the former group relative to the latter, while red dots represent significantly downregulated proteins. Dot size reflects fold change, and only the top 10 DEPs by P-value are labelled (n = 4/group). Statistical significance was determined using Student’s t-test. D Functional enrichment analysis of proteins downregulated following LPS stimulation but restored by BAIBA treatment. Dot size reflects the number of enriched proteins per pathway, and colour indicates statistical significance based on the hypergeometric test (P < 0.05) (n = 4/group). E Western blot analysis shows the effects of HSPA1B knockdown on HSPA1B and p-p65 protein levels in LPS-induced or BAIBA-treated inflammatory MAC-T, with GAPDH used as a loading control. Effects of BAIBA treatment on the concentrations of IL-6 (F), IL-1β (G), TNF-α (H), and IL-10 (I) in the culture supernatant of LPS-induced inflammatory MAC-T following HSPA1B knockdown (n = 3/group). J Schematic overview of the BAIBA treatment experiment in dairy goats with clinical or subclinical Mastitis (n = 5/group). Differences in milk somatic cell count before and after BAIBA treatment in goats with subclinical (K) and clinical (L) Mastitis. Statistical significance was determined using one-way ANOVA for multi-group comparisons and paired two-tailed Student’s t-test for before–after comparisons in (K) and (L) (n = 5/group). Statistical significance was determined using one-way ANOVA. **P < 0.01, ***P < 0.001; n.s., not significant. Box plots show the median and interquartile range, and dots represent individual biological replicates. Bar plots show mean ± SEM. Abbreviations: SCC, somatic cell count; IL-6, Interleukin-6; IL-1β, Interleukin-1β; TNF-α, tumour necrosis factor-α; IL-10, Interleukin-10

Notably, the expression of HSPA1B, a member of the 70 kDa heat shock protein family, was significantly downregulated following LPS stimulation, while BAIBA treatment effectively restored its expression levels (P < 0.05; Figs. 5C and Additional file 2: Fig. S14), suggesting that HSPA1B may serve as a key downstream target of the anti-inflammatory activity of BAIBA. To further validate this hypothesis, we focused on the activation status of the inflammatory transcription factor NF-κB p65 across different treatment groups. The results demonstrated that BAIBA significantly inhibited the LPS-induced increase in p-p65 levels (Fig. 5E). By contrast, total p65 protein abundance remained unchanged (Additional file 2: Fig. S15), indicating that BAIBA primarily affected the phosphorylation status rather than the overall abundance of p65. However, in the context of HSPA1B knockdown via siRNA, the suppressive effect of BAIBA on p-p65 expression was completely abrogated (Fig. 5E). Furthermore, HSPA1B knockdown abrogated BAIBA-mediated repression of the pro-inflammatory cytokines IL-1β, IL-6 and TNF-α, as well as its induction of the anti-inflammatory cytokine IL-10 (Fig. 5F–I), indicating that BAIBA’s cytokine remodelling is HSPA1B-dependent. Collectively, our study establishes the “BAIBA-HSPA1B-NF-κB” signalling pathway, revealing that BAIBA alleviates the inflammatory response in MAC-T by activating the endoplasmic reticulum protein quality control system and promoting HSPA1B expression, which in turn suppresses NF-κB signalling activation and downstream pro-inflammatory cytokine expression, providing a solid mechanistic basis for the targeted application of BAIBA.

To evaluate the clinical translational potential of BAIBA, we conducted an in vivo local treatment experiment in dairy goats naturally infected with SCM and clinical Mastitis. Specifically, 10 μM BAIBA was administered via intramammary infusion at a dose of 5 mL daily for 7 days (Fig. 5J). The results revealed a significant decrease in SCC in milk following treatment, with some individuals showing recovery of SCC to below clinical reference ranges (P < 0.05; Fig. 5K, L), suggesting a substantial improvement in the inflammatory state. Taken together, these findings demonstrate that BAIBA, as a functional microbial metabolite, has significant efficacy in alleviating local mammary inflammation and reducing milk SCC, validating its potential for non-antibiotic treatment of Mastitis in dairy livestock.

Discussion

SCM is a prevalent yet frequently under-recognized mammary disorder in dairy livestock and has been associated with sustained decrements in milk yield, deterioration of milk quality, and compromised animal welfare [31]. Owing to the paucity of overt clinical manifestations and the limited diagnostic yield of conventional microbiological assays in a subset of cases, research focus has increasingly shifted toward atypical and systemic dimensions of its pathogenesis [32]. Within this context, we adopted an integrative microbiome-metabolite-immunity framework to delineate the multiscale regulatory architecture underpinning SCM pathobiology, with particular emphasis on the putative contribution of the intestinal microbiome to local mammary inflammation.

We observed pronounced restructuring of the intestinal microbiota in individuals with SCM. Compared with the Healthy group, archaeal Methanobrevibacter spp. were markedly depleted, whereas several putative opportunistic pathobionts (such as Eubacterium spp., Blautia spp., and Lachnospira sp.) were enriched. This pattern is highly consistent with prior studies implicating gut microbial composition in mammary gland health [7, 8, 16, 22]. For example, in a dairy cow mastitis cohort, the abundance of Blautia and Lachnospira was significantly associated with elevated serum levels of pro-inflammatory cytokines, suggesting that shifts in microbiota composition may elicit systemic immune responses and potentially modulate the mammary microenvironment [8]. Moreover, the “gut-mammary gland” axis proposes that gut microbiota influence mammary epithelial function via metabolite-dependent or immune-mediated routes [16]. By integrating correlations between microbiota composition and inflammatory readouts, our study provides additional empirical support for this axis and indicates that, in the context of SCM, intestinal dysbiosis may constitute an important driver of the inflammatory response (Fig. 6).

Fig. 6.

Differential metabolic functions in the gut microbiota of Mastitis and Healthy dairy goats regulate mammary gland inflammation via the HSPA1B-p65 signalling axis. This schematic illustrates key differences in the microbial composition and metabolic functions of the gut microbiota between the Mastitis and Healthy groups and their potential impact on mammary gland inflammation. In the Mastitis group, representative taxa such as Blautia spp. and Eubacterium spp. were significantly associated with circulating levels of IL-6, IL-1β, TNF-α, and LPS. These microbes are hypothesized to be involved in the activation of mammary immune responses. In contrast, the Methanobrevibacter sp., a dominant lineage in the gut of Healthy dairy goats, harbors vorA, vorB, and vorD, which encode enzymes catalyzing the rate-limiting steps of the L-Valine-derived biosynthetic pathway to L-3-aminoisobutyrate (BAIBA), culminating in BAIBA accumulation in the hindgut. BAIBA is proposed to act along the “gut-mammary gland” axis to activate the endoplasmic reticulum quality control pathway in mammary epithelium, induce HSPA1B expression, and thereby attenuate NF-κB signalling and lower pro-inflammatory cytokine levels, mitigating mammary inflammation. Abbreviations: LPS, Lipopolysaccharide; IL-6, Interleukin-6; IL-1β, Interleukin-1β; TNF-α, tumor necrosis factor-α

Building on these findings, we integrated metagenomic functional profiles with the metabolomics dataset and identified BAIBA as significantly elevated in the Healthy group, with inverse associations to SCC and multiple inflammatory readouts. Notably, BAIBA is not novel in this context. Previous studies have established its derivation from the branched-chain amino-acid-derived degradation pathway and its roles in modulating glucose and lipid metabolism and attenuating low-grade systemic inflammation [33, 34]. Particularly in high-fat diet-induced mice models of metabolic syndrome and skeletal muscle inflammation, BAIBA exerted anti-inflammatory effects through the AMPK-PPARγ signaling axis [35]. However, whether BAIBA modulates inflammation within the mammary gland remained largely unexplored. Using MAC-T, we provide experimental evidence that BAIBA attenuated NF-κB activation by upregulating HSPA1B, thereby suppressing local pro-inflammatory responses and expanding the tissue-specific repertoire of its immunoregulatory actions. Furthermore, functional proteomics showed that BAIBA markedly activated key components of the endoplasmic reticulum protein quality control pathway (such as HSPA1B and HYOU1), consistent with its role in maintaining stress homeostasis in other tissues [36, 37]. This pathway has been underappreciated in the context of mammary inflammatory resolution, suggesting that BAIBA may act as a mediator within an inflammation-proteostasis-immune homeostasis triad.

It should be noted that, although BAIBA was supported by the most consistent evidence in the present study and was therefore prioritized for in-depth characterization as a candidate functional metabolite, this does not exclude the possibility that other differential metabolites also contribute to the regulation of mammary health. Notably, Akkermansia spp. was another bacterial taxon significantly enriched in the gut of the Healthy group. Previous studies have suggested that Akkermansia is closely linked to tryptophan metabolism and is capable of converting tryptophan into indole derivatives, including indole-3-acetic acid and indole-3-lactic acid [38]. Consistent with this, we further found that both metabolites were significantly elevated in the gut of the Healthy group. Indole-3-acetic acid, acting as an AhR ligand, has been shown to promote intestinal epithelial barrier function, induce anti-inflammatory mediators such as IL-10 and IL-22, and suppress the production of pro-inflammatory cytokines including TNF-α and IL-6 [39, 40]. By contrast, indole-3-lactic acid has been reported to suppress inflammatory cell activation and inflammatory mediator release through activation of the AhR/Nrf2 signaling axis, while also promoting epithelial homeostasis [41, 42]. Taken together with our finding that Methanobrevibacter has the potential to contribute to BAIBA biosynthesis, these results suggest that a metabolite network jointly sustained by specific gut archaeal and bacterial communities, rather than any single microbial taxon or metabolite alone, may be important for maintaining mammary health and constraining the progression of mammary inflammation.

Beyond the central role of HSPA1B highlighted in this study in the resolution of mammary inflammation, we also observed significant increases in the abundance of several other proteins following BAIBA intervention, including LF, OTUD4 and ID1. This suggests that the protective effects of BAIBA may not be restricted to a single HSPA1B-centered axis, but instead may involve a broader restoration of protein homeostasis and inflammatory regulatory networks. LF is a classical host-defense molecule in the mammary secretory system, with established antimicrobial, immunomodulatory and inflammation-buffering properties. Previous studies have shown that LF exerts broad immunoregulatory and anti-inflammatory effects, and that mammary epithelial cells can upregulate and secrete LF in response to innate immune stimulation, thereby attenuating inflammatory responses such as TNF-α production [43, 44]. By contrast, OTUD4 and ID1 have been much less extensively studied in the context of mastitis, although their biological properties suggest that they may likewise contribute to the protective program induced by BAIBA. OTUD4 is a member of the deubiquitinase family, and has been reported to regulate TLR-dependent NF-κB signaling through its action on MyD88, thereby influencing the magnitude of innate immune responses. The upregulation of OTUD4 following BAIBA treatment in the present study therefore suggests that BAIBA may, at least in part, alleviate mammary epithelial inflammation through reshaping ubiquitination-dependent inflammatory signaling [29]. ID1 is a classical helix-loop-helix transcriptional regulator that has been implicated in the control of mammary epithelial cell proliferation, differentiation and apoptosis. Studies in other epithelial tissues have further linked ID1 to epithelial injury responses and the maintenance of tissue homeostasis [45, 46]. In light of the marked recovery of ID1 abundance after BAIBA treatment observed here, we speculate that its upregulation may reflect restoration of mammary epithelial homeostatic capacity by BAIBA, rather than merely a passive change in inflammatory markers. Taken together, the coordinated increase in these proteins following BAIBA treatment supports the notion that BAIBA may alleviate mammary inflammation and promote epithelial homeostasis through a multilayered molecular network. In addition to HSPA1B, LF, OTUD4, ID1 and other simultaneously upregulated candidate proteins may jointly contribute to the mammary-protective effects of BAIBA. Although the precise roles of these proteins in maintaining mammary health and regulating mammary inflammation remain to be fully defined, their consistent changes provide important leads for future mechanistic studies. Further gain- and loss-of-function investigations, at both the cellular and vector-based levels, will be required to systematically dissect their functional roles, interaction networks and potential translational value for mastitis intervention.

To further substantiate the linkage between gut microbiota, microbial metabolites, and mammary inflammation, we conducted a cross-species FMT experiment and observed that donor communities from SCM individuals elicited inflammatory changes in mice mammary tissue. Although interpretation is constrained by interspecies differences and other experimental limitations, the overall pattern supports the hypothesis that SCM-associated gut microbiota possesses mammary inflammation-inducing potential, consistent with reports that FMT can modulate immune responses in distal organs [47, 48]. Importantly, we did not establish strain-level sufficiency, as the present data do not demonstrate that any single strain alone is causative for mammary inflammation. Accordingly, this study should be regarded as delineating an association network and plausible mechanistic routes rather than providing definitive evidence from single-factor intervention studies.

Building on the “gut-X” axis framework [12], our study provides a multi-omics evidentiary chain linking mammary inflammation with intestinal dysbiosis in the context of SCM, suggesting that gut-derived metabolites may modulate mammary immune function. This perspective complements pathogen-centric models of Mastitis and informs the development of non-antibiotic, microbiota-based intervention strategies. In contrast to prior work that emphasized canonical metabolites such as butyrate [8] and inulin [49, 50], we focused on branched-chain amino-acid-derived catabolites, particularly BAIBA that is predominantly synthesized by Methanobrevibacter, thereby adding conceptual novelty and extending mechanistic scope. Several limitations merit consideration. First, functional annotation of the goat gut metagenome remains incomplete, and the microbial provenance of BAIBA requires verification using isolates or defined synthetic consortia. In addition, although our study implicates BAIBA as a potential modulator of SCM, whether BAIBA can serve as an endogenous biomarker for SCM risk stratification will need to be established in larger, independent cohorts encompassing both diseased and healthy lactating animals. Second, the absence of feed intake and body weight represents an important limitation, because diet and host physiological status are major determinants of gut microbiome composition and function and may have contributed, at least in part, to the observed fecal microbiome shifts [51]. Future studies should therefore incorporate controlled feeding, longitudinal sampling, and detailed host phenotyping to better distinguish whether gut microbiome alterations precede SCM or arise secondarily to mammary inflammation. Moreover, although regulatory effects of BAIBA were demonstrated in vitro, its consistency and robustness in mammalian mammary tissue will require additional in vivo evidence. Third, strain-level reconstitution was not performed; therefore, shifts in the gut community cannot be deemed sufficient to cause SCM. Taken together, within the current “gut-organ” axis paradigm [12], our study outlines a putative conduit connecting microbiome, microbial metabolites, and local mammary immune status in SCM, and it identifies a BAIBA-HSPA1B-NF-κB signaling axis as a candidate regulator of mammary inflammation. These insights motivate construction of metabolism-immunity regulatory networks and the design of microbiota-targeted interventions, while underscoring the need to more tightly integrate gut microbiome functional research with strategies for the prevention and control of mammary disease.

Conclusions

Taken together, this study delineates a metabolic-signaling coupling between intestinal microbiome and mammary immunity in SCM, identifies the gut-derived metabolite BAIBA as a key determinant of mammary immune homeostasis, and outlines a feasible strategy for non-antibiotic interventions that target microbial metabolites in mastitis. These findings extend the “gut-organ” interaction paradigm to mammary disease and provide a mechanistic rationale together with a portfolio of candidate targets for advancing microbiome-based therapeutics in livestock health.

Materials and methods

Husbandry and management of dairy goats

The discovery cohort comprised 167 mid-lactation Saanen dairy goats (2–4 years old; parity, 2.17 ± 0.40; days in milk, 121.47 ± 7.95) from a commercial farm in Huan County, Qingyang, Gansu Province, China (36°57′N, 107°31′E) (Additional file 1: Table S1). All animals were housed in the same temperature-controlled, well-ventilated barn. Goats were milked daily at 05:00 and 17:00 and were fed immediately after each milking (06:00 and 18:00). After feed delivery, goats were allowed to consume feed ad libitum and had free access to water throughout the day. The concentrate-to-forage ratio was 45:55, and diet composition and nutrient formulation are provided in Additional file 6: Table S5.

A total of 10 Guanzhong dairy goats with mastitis (5 subclinical and 5 clinical; 2–3 years old; parity, 2) were enrolled from a commercial farm in Fuping County, Shaanxi Province, China (34° 45′ 06″ N, 109° 10′ 48″ E). Affected goats were housed in isolation in the same temperature-controlled, well-ventilated barn, which was cleaned daily by farm staff. Farm records indicated that all cases occurred during the current lactation and that no animals received antibiotics during the disease period. The diet formulation before BAIBA treatment is provided in Additional file7: Table S6. After enrollment, goats underwent a 7-day acclimation period and then received intramammary BAIBA infusion via the teat canal for 7 consecutive days. Throughout the experimental period, goats were maintained on the same diet formulation as the discovery cohort (concentrate-to-forage ratio, 45:55) and followed the same feeding and ad libitum water-access regimen (Additional file 6: Table S5).

Experimental design

The experimental design is depicted in Fig. 7. We first measured milk SCC in 167 mid-lactation Saanen dairy goats and, on this basis, selected 6 goats for the Mastitis group and 6 for the Healthy group. To delineate the potential contribution of the gut microbiota to SCM pathogenesis, fecal microbiota from Mastitis (M-FMT) and Healthy (H-FMT) donors were transplanted into pseudo-germ-free mice, respectively. Guided by the marked between-group differences in microbial composition and metabolic function, and integrating untargeted metabolomics, we prioritized BAIBA as a candidate effector metabolite and, using an LPS-induced inflammatory model in MAC-T, investigated the BAIBA-mediated anti-inflammatory mechanism and its putative molecular targets.

Fig. 7.

Experimental design. Our study began by quantifying the SCC in a cohort of 167 Saanen dairy goats, from which 6 goats from both the Mastitis and Healthy groups were selected for further analysis. Fecal microbiota from these two groups were then transplanted into mice to investigate the relationship between gut microbiota and mammary gland inflammation. The microbial composition and functional profiles of the gut in both groups were analyzed, followed by LC–MS/MS-based metabolomic analysis to identify functional metabolites in the gut of healthy goats. Finally, the therapeutic potential of the metabolite was explored in an LPS-induced bovine mammary epithelial cell inflammation model to investigate its role in alleviating mammary inflammation. Abbreviations: SCC, somatic cell count; BAIBA, L-3-aminoisobutyrate; MAC-T, bovine mammary epithelial cell

Collection of dairy goat samples

To assess lactation phenotypes in the 167 Saanen dairy goats, daily milk yield was recorded throughout the first and fourth weeks of September 2022, and the mean of these values was used as the average daily milk yield. On the final day of milk-yield recording, composite milk samples were prepared by mixing morning and evening milk at a 3:2 ratio in duplicate. These samples were then analyzed for milk components and SCC (leukocytes per mL in fresh milk) using a milk composition analyzer (Combi 300, BENTLEY, MN, USA). Goats with the highest SCC (Mastitis, n = 6) and the lowest SCC (Healthy, n = 6) were then selected, and a post hoc power analysis was performed in G*Power [52] (v3.1.9.7). The effect size was computed from group means and standard deviations (SD); with α = 0.05 under an independent-samples t-test, the achieved power > 99%. After group assignment, fecal and blood samples were collected from all 12 goats during the same sampling session. Fresh feces were obtained by rectal abdominal massage, snap-frozen in liquid nitrogen and stored at −80 °C for metagenomic sequencing and untargeted metabolomics. Blood was then collected by jugular venipuncture into serum collection tubes, allowed to clot at room temperature for 15–30 min, and centrifuged at 3,000 rpm for 10 min at 4 °C to obtain serum [53]. Serum was transferred to sterile tubes and stored at −20 °C for biochemistry assays and quantification of inflammatory cytokines.

To evaluate the therapeutic efficacy of BAIBA in goats with naturally occurring subclinical or clinical mastitis, goats underwent a 7-day acclimation period before treatment, during which the feeding schedule was standardised and the animals were familiarised with the intramammary infusion and milk sampling procedures. Following acclimation, each goat received an intramammary infusion of BAIBA (10 μM; 5 mL per day) through the teat canal for 7 consecutive days, administered each morning after feeding. Milk samples were collected before feeding at 06:00 and 18:00 on day 7 of acclimation and again on the morning and evening of the day after completion of BAIBA treatment. Milk from the morning and evening milkings was pooled at a 3:2 ratio in duplicate and analysed for SCC using a milk composition analyser (Combi 300, BENTLEY, MN, USA). Samples obtained prior to BAIBA administration constituted the CON group, and those collected after treatment constituted the BAIBA group. After acquiring pre- and post-treatment SCC data for the subclinical and clinical mastitis cohorts, we conducted a post hoc power analysis in G*Power [52] (v3.1.9.7). Effect sizes were calculated from group means and SD of SCC; with α = 0.05 under a paired two-tailed Student’s t-test and a sample size of five per cohort, the achieved power > 90%, indicating adequate statistical power.

Animals and FMT treatment

The goat–to–mouse FMT protocol was adapted from previously published procedures with appropriate modifications [7, 54]. Briefly, fecal samples from Healthy and Mastitis donor goats were collected by abdominal massage and immediately transferred into cryovials containing sterile aqueous glycerol solution (50%, v/v), then stored at −80 °C to preserve bacterial viability as much as possible [55]. To prepare the fecal suspension for FMT, donor fecal samples were thawed on ice and subsequently transferred to an anaerobic workstation (YY-XXL PLUS, MAWORDE, Qiqihar, China) for all downstream procedures. For each donor group, 1 g of feces from each donor animal was pooled and resuspended in sterile PBS containing 50% glycerol at a ratio of 1 mL per gram of feces. After thorough mixing, the suspension was centrifuged at 4 °C and 3,000 rpm for 6 min. The supernatant was collected, aliquoted into sterile 2-mL cryovials, and stored at −80 °C for subsequent FMT gavage.

To establish a goat-derived FMT recipient mouse model, 30 specific-pathogen-free (SPF) BALB/c mice weighing 20–22 g were used as recipients and were purchased from Teng Xin Biotechnology Co., Ltd. (Chongqing, China). Mice were housed in the SPF barrier facility of Northwest A&F University under controlled conditions (22 ± 2 °C, 50–60% relative humidity, and a 12-h light/12-h dark cycle) [56]. Throughout the experiment, mice had ad libitum access to sterilized feed and sterile water and were maintained in individually ventilated cages. Bedding, water bottles, and all related husbandry supplies were autoclaved and replaced regularly under barrier conditions to minimise exposure to exogenous microorganisms [57].

After arrival, mice were acclimated for 1 week. Each female then received an intraperitoneal injection of 10 IU pregnant mare serum gonadotropin, followed 48 h later by 8 IU human chorionic gonadotropin. Females were subsequently housed with males at a female-to-male ratio of 2:1 for 3 days, and Eighteen plug-positive females were selected for the subsequent experiments. To generate a pseudo-germ-free mouse model, drinking water was supplemented with ampicillin (1 g/L), gentamicin (1 g/L), streptomycin (1 g/L), and vancomycin (0.5 g/L) (Macklin, Shanghai, China) for 20 days [54]; antibiotic-containing water was freshly prepared and replaced every 3 days under sterile conditions. The 18 mice were then randomly assigned to three groups: negative control (NC, n = 6), mastitis-donor FMT (M-FMT, n = 6), and healthy-donor FMT (H-FMT, n = 6). Mice in the M-FMT and H-FMT groups received 200 μL of the corresponding fecal suspension by oral gavage once daily for 7 consecutive days, whereas NC mice received 200 μL PBS on the same schedule. At the end of FMT, mice were anesthetized in an induction chamber with 5% isoflurane until the loss of the righting reflex and pedal withdrawal reflex [58], and were then euthanized by gradual CO₂ inhalation. Immediately after euthanasia, blood was collected following enucleation, allowed to clot at room temperature for 30 min, and centrifuged at 3,000 rpm for 10 min at 4 °C to isolate serum, which was stored at −80 °C until further analysis [59]. The skin over the inguinal region was opened to expose the fourth pair of mammary glands, which were then carefully dissected in their entirety based on their anatomical location [60]. During tissue collection, the surrounding skin, fascia, non-mammary adipose tissue, and connective tissue were carefully removed to ensure the integrity and consistency of the isolated samples. One portion was fixed in 4% paraformaldehyde for histological analysis, and the other was snap-frozen in liquid nitrogen and stored at −80 °C for subsequent ELISA-based quantification of inflammatory factors in mammary tissue. To monitor microbial depletion and reconstitution during the FMT experiment, fresh fecal pellets were collected from each mouse at three time points: before antibiotic treatment, after completion of antibiotic treatment, and after completion of FMT. For fecal collection, mice were individually placed in sterile clean cages, and freshly voided fecal pellets were immediately collected using sterile forceps to avoid contamination from bedding or cage surfaces. Samples were transferred into sterile cryovials, snap-frozen in liquid nitrogen, and stored at −80 °C until DNA extraction for 16S rRNA gene sequencing and qPCR analysis.

Serum physio-biochemical analysis in dairy goats

Serum clinical chemistry parameters were measured using a fully automated analyser (Celercare V5, MNCHIP, Tianjin, China) in accordance with the manufacturer’s instructions. Briefly, serum samples from Mastitis and Healthy dairy goats and from NC, M-FMT, and H-FMT mice were retrieved from storage at −20 °C and thawed on ice. For each sample, One hundred microlitres of serum was loaded into a 19-parameter health-check cartridge (14 + 5 panel; MNCHIP, Tianjin, China). The cartridge was inserted into the analyser, and the resulting clinical chemistry readouts were recorded and archived.

Histological analysis

Mouse mammary tissues were fixed in 4% paraformaldehyde for more than 48 h, embedded in paraffin, and sectioned at 5 μm. Sections were stained with hematoxylin and eosin (H&E) and examined under a light microscope (Olympus, Tokyo, Japan). Inflammatory phenotypes were evaluated, including disruption of tissue architecture, inflammatory cell infiltration, and any additional histopathological alterations.

Comprehensive analysis of inflammatory cytokine in milk, mammary tissue, serum, and bovine mammary epithelial cells

Inflammatory cytokines were quantified in goat milk and serum, BALB/c mouse mammary tissue and serum, and MAC-T suspensions; mammary tissue and MAC-T samples required pre-processing prior to assay. For mammary tissue, 1 g of tissue was added to 9 mL of PBS homogenization buffer (pH 7.2–7.4; 0.1 mg/L) and homogenized in a refrigerated tissue homogenizer. The resulting homogenate was centrifuged at 12,000 × g for 10 min at 4 °C, and the supernatant was collected for analysis. For MAC-T, cultures were gently washed with PBS and dissociated with trypsin, then centrifuged at 1,000 × g for 5 min to collect the cell pellet. The pellet was washed three times with pre-chilled PBS, resuspended at 1 × 10⁶ cells in 200 µL PBS, and lysed by repeated freeze–thaw cycles. Lysates were centrifuged at 1,500 × g for 10 min and the supernatant was collected for downstream analysis. Concentrations of MPO, LPS, IL-6, IL-1β, TNF-α, and IL-10 in milk, serum, and the pre-processed tissue and MAC-T samples were measured using enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturer’s instructions (catalog numbers F3875-A, F72200-A, F3885-A, F6400-A, F3849-A, and F4050-A; Kexing Trading, Shanghai, China).

Absolute qPCR quantification of microbial loads in mouse fecal samples

To assess microbial depletion and reconstitution during antibiotic treatment and FMT, absolute qPCR was performed to quantify total bacterial load and Methanobrevibacter spp. copy numbers in mouse fecal samples. After genomic DNA extraction, bacterial 16S rRNA genes were amplified using the universal primer pair [61] 27 F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′-TACGGTTACCTTGTTACGACTT-3′), whereas Methanobrevibacter spp. were amplified using the specific primer pair [62] F (5′-CTCCGCAATGTGAGAAATCG-3′) and R (5′-GCGGTGTGTGCAAGGAGC-3′). Primer specificity was verified using Primer-BLAST. qPCR was performed on a LightCycler® 96 Real-Time PCR System (Roche, Pleasanton, CA, USA) using qPCR SuperMix (TransGen, Beijing, China). Target amplicons were cloned into plasmid vectors and sequence-verified as standards, followed by tenfold serial dilution to generate standard curves. Copy numbers of total bacteria and Methanobrevibacter spp. were calculated by interpolating sample Ct values against the corresponding standard curves and expressed as gene copies per gram of feces after log10 transformation for downstream statistical analysis.

DNA extraction, shotgun metagenomic sequencing, and bioinformatic analysis

Genomic DNA was extracted from faecal samples of the Mastitis and Healthy groups using the mini-bead-beating method described by Yu and Morrison with minor adaptations [63], with the following modifications: Two hundred milligrams of faecal material was used as input, and mechanical disruption was performed using 0.1-mm zirconia/silica beads. DNA concentration and purity were assessed with a NanoDrop 2000 UV–Vis spectrophotometer (Thermo Fisher Scientific, Wilmington, USA), and DNA integrity was evaluated by 1% agarose gel electrophoresis. Paired-end (PE) libraries were prepared from fragmented DNA using the NEXTFLEX Rapid DNA-Seq kit (Bioo Scientific, Austin, USA). Metagenomic sequencing was performed on an Illumina NovaSeq 6000 platform following the manufacturer’s protocol to generate 150 bp PE reads. Across 12 dairy goat faecal samples, sequencing yielded 142.11 Gb of raw data, averaging 11.84 Gb per sample. Adapter trimming and quality control of raw reads were conducted with fastp [64] (v0.20.0; parameters: –cut_by_quality3 -W 4 -M 20 -n 5 -c -l 50 -w 3). To eliminate host DNA contamination, reads were aligned to the goat reference genome using BWA-MEM [65] (v0.7.17), and high-quality reads were retained, resulting in 104.14 Gb of high-quality data (8.68 Gb per sample on average) with a mean host contamination rate of 26.76%. Assembly of high-quality reads was performed with MEGAHIT [66] (v1.2.9; parameters: –min-contig-len 500 -t 40), yielding on average 777,409.50 contigs per sample with a mean N50 of 639.69 bp. Open reading frames (ORFs) were predicted from assembled contigs using Prodigal [67] (v2.6.3; parameters: -p meta), producing 15.78 million ORFs with a mean length of 444.96 bp. After removing ORFs shorter than 100 bp, clustering was carried out with CD-HIT [68] (v4.8.1; parameters: -n 9 -g 1 -G 0 -m 0 -d 0 -as 0.9 -c 0.95) to generate a non-redundant (NR) goat gut microbial gene catalog comprising 4,399,515 predicted genes with a mean length of 518.74 bp. For taxonomic and functional annotation, NR nucleotide sequences were translated into predicted protein sequences using EMBOSS Transeq [69] (v6.6.0.0; parameters: -frame 6). Gene abundances were then estimated by mapping each sample’s high-quality reads to the NR gene catalog with SOAP2 [70] (v2.21; parameters: -r 1 -l 35 -M 4 -p 6 -v 20 -c 0.95 -m 400—× 600).

To obtain taxonomic annotations for the NR gene catalogue, the associated predicted protein sequences were aligned against the NCBI-nr protein database (version: 20,230,830) using DIAMOND [71] (v2.1.8.162) with an e-value threshold of 1e-5, and assignments were retrieved across standard taxonomic ranks.

KOs annotation for the NR predicted genes was performed with KOfamScan [72] (v1.1.0; parameters-cpu 20 -E 1e-5 -f mapper), which uses HMMER to search the KOfam database (version: 202,209), a profile hidden Markov model collection tailored to KOs. For KEGG enrichment analysis, a Z-score framework was applied to evaluate pathway enrichment across groups [73]. Briefly, significantly different KOs between two groups were identified using the Wilcoxon rank-sum test, and P-values were then halved so that the transformed values lay within the interval (0, 0.5], expressed as $P_{\mathit{koi}}=P_{\mathit{koi}}/2$, where $P_{\mathit{koi}}$ denotes the P-value for ${\mathit{KO}}_i$.

Subsequently, P-values were transformed to Z-values using the inverse normal transform, $Z_{\mathit{koi}}={\theta }^{-1}\left(1,-,P_{\mathit{koi}}\right)$. We then computed the abundance difference $\mathrm{\Delta }K{O}_i=\overline{K{O}_{i_{g1}}}-\overline{K{O}_{i_{g2}}}$ and assigned the sign of $Z_{\mathit{KOi}}$ according to directionality, such that $Z_{\mathit{KOi}}$ was positive if $\mathrm{\Delta }K{O}_i$ > 0 and negative if $\mathrm{\Delta }K{O}_i$ < 0. Pathway-level scores were obtained by aggregating KO-level statistics as $Z_{\mathit{pathway}}=\frac{1}{\sqrt{k}}\sum Z_{\mathit{koi}}$, where k denotes the number of KOs mapped to the pathway. We next performed 1,000 permutations to derive the null distribution of $Z_{\mathit{pathway}}$ and calculated an adjusted score $Z_{\mathit{adjustedpathway}}=\left(Z_{\mathit{pathway}},-,{\mu }_k\right)/{\sigma }_k$, where ${\mu }_k$ and ${\sigma }_k$ are the permutation-based mean and standard deviation, respectively. The adjusted statistic was taken as the final Z-score, also referred to as the reporter score, and pathways with |Z-score|> 1.65 were considered significantly different between groups.

To trace the microbial provenance of enzyme genes involved in the rate-limiting steps of BAIBA biosynthesis, we first used a custom Python script to identify ORFs encoding the relevant key enzymes from the microbial NR gene catalogue and extracted their corresponding taxonomic annotations. We then retrieved the abundance information for these ORFs from the ORF abundance matrix. Finally, taxonomic annotations were integrated with abundance data to visualize the source composition of enzyme genes involved in the rate-limiting steps of BAIBA biosynthesis at the species level.

16S rRNA gene sequencing and analysis of mouse fecal microbiota after FMT

Microbial DNA was extracted from mouse fecal samples using the same mini-bead-beating protocol described above for goat fecal samples, with minor modifications. The extracted DNA was used for 16S rRNA gene sequencing to assess the gut microbial composition of recipient mice after FMT. The V3-V4 region of the bacterial 16S rRNA gene was amplified using the primer pair 338 F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). Purified amplicons were quantified, pooled in equimolar amounts, and subjected to paired-end sequencing on an Illumina X Plus platform. Raw reads were demultiplexed according to sample-specific barcodes and quality-filtered using fastp (v0.23.4; parameters: –cut_by_quality3 -W 4 -M 20 -n 5 -c -l 50 -w 3). PE reads were then merged using FLASH (v1.2.11; parameters: -m 10 -M 250—× 0.20 -p 33 -t 8), and the resulting merged sequences were denoised, filtered and chimera-removed using the DADA2 plugin in QIIME2 in single-end mode (v2024; parameters: –p-trunc-len 0 –p-max-ee 2 –p-chimera-method consensus). Amplicon sequence variants (ASVs) were taxonomically assigned using a naive Bayes classifier trained against the SILVA 138 16S rRNA gene database. To minimize bias arising from uneven sequencing depth, all samples were rarefied to an equal sequencing depth before downstream diversity analyses.

α-diversity indices were calculated to evaluate microbial diversity in mouse fecal samples after FMT. PCoA based on Bray–Curtis distances was performed to assess microbial community structure, including comparisons between NC and FMT-treated mice, as well as between the M-FMT and H-FMT groups. Differences in community composition were assessed using PERMANOVA. Taxonomic differences between the H-FMT and M-FMT groups at the phylum, family and genus levels were analyzed using the Wilcoxon rank-sum test. Bacterial taxa detected in mouse feces after FMT were further compared with donor-enriched microbial signatures in goats to assess whether donor-associated taxa were detectable in the recipient mouse gut after FMT.

Untargeted metabolomics sequencing

Fecal samples were thawed on ice. For each sample, 50 mg (± 1 mg) was homogenized with 500 µL ice-cold methanol/water (70%, v/v). The mixture was vortexed for 3 min, sonicated in an ice-water bath for 10 min, and vortexed again for 1 min. The samples were then centrifuged at 12,000 rpm for 10 min. A 200 µL aliquot of the supernatant was incubated at −20 °C for 30 min and subsequently centrifuged at 12,000 rpm for 30 min at 4 °C. A 200 µL portion of the clarified supernatant was transferred to a vacuum concentrator and evaporated to dryness. The residue was reconstituted in 150 µL of 70% methanol, vortexed for 3 min, and sonicated in an ice-water bath for 10 min. A final centrifugation was performed at 12,000 rpm for 3 min at 4 °C, and 120 µL of the supernatant was used for UHPLC-ESI–MS/MS analysis (UHPLC, ExionLC AD; ESI; QTRAP system). Chromatographic separation employed a Waters ACQUITY UPLC HSS T3 C18 column (1.8 µm, 2.1 mm × 100 mm). The mobile phases were ultrapure water with 0.1% formic acid and acetonitrile with 0.1% formic acid. The column temperature was maintained at 40 °C, the flow rate was 0.4 mL/min, and the injection volume was 2 µL. Mass spectrometric data were acquired in both positive and negative ion modes.

Raw LC–MS/MS data were processed with Analyst software (v1.6.3). To broaden metabolite identification coverage, annotation was carried out against an in-house database (MWDB, METWARE) and public repositories including METLIN, HMDB, KEGG, and MoNA. Differential metabolites were defined by a variable importance in projection (VIP) > 1 together with a Wilcoxon rank-sum test P-value < 0.05 (Additional file 4: Table S3). Identified metabolites were mapped to KEGG pathways, and pathway enrichment was evaluated using a hypergeometric test.

Cell culture and inflammatory stimulation of MAC-T

MAC-T were cultured at 37 °C in a humidified incubator with 5% CO₂ in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum. To establish an LPS-induced inflammatory model, passaged cells were treated with 25, 50, 75, or 100 µg/mL LPS for 12, 24, or 48 h to determine the optimal induction conditions. To evaluate the effect of BAIBA on MAC-T proliferation, cells were seeded into 96-well plates and treated with 5, 10, 15, or 20 µM BAIBA for 6, 12, 18, or 24 h. CCK-8 working solution (C0042; Beyotime, Shanghai, China) was then added and incubation continued for 1 h, after which OD450 was measured according to the manufacturer’s instructions. Relative cell viability was calculated by normalizing to the untreated control, which was set to 100%. To evaluate the therapeutic effect of BAIBA on inflammatory MAC-T, the inflammatory model was first established using the optimal LPS concentration and incubation time, after which BAIBA was administered at graded concentrations (5, 10, 15, 20 µM) and time points (6, 12, 18, 24 h). Following treatment, levels of IL-1β, IL-6, TNF-α, and IL-10 in culture supernatants were quantified by ELISA to determine the optimal BAIBA treatment conditions. For HSPA1B knockdown, nonspecific control siRNA and HSPA1B-specific siRNA (GenePharma, Shanghai, China) were mixed with Lipofectamine 3000 at a 1:1 volume ratio, diluted in serum-free medium, and incubated at room temperature for 30 min. The siRNA-Lipofectamine complexes were then added to MAC-T in 6-well plates for a 24 h transfection. After transfection, cells were transferred to 12-well plates, treated with 50 µg/mL LPS for 12 h, and subsequently treated with 10 µM BAIBA for 12 h. The HSPA1B siRNA sequence was 5'-CGUCGAUGGUGCUGACCAAGA-3'.

Protein extraction and peptide separation

To evaluate the therapeutic effect of BAIBA in an LPS-induced inflammatory MAC-T model, MAC-T were revived and passaged into 12-well plates. Three conditions were established: untreated MAC-T control (MAC-T), LPS stimulation (LPS), and BAIBA treatment (BAIBA). For the MAC-T group, passaged cells were incubated in PBS for 24 h. For the LPS and BAIBA groups, cells were first exposed to 50 µg/mL LPS for 12 h and were then treated with PBS or 10 µM BAIBA, respectively, for an additional 12 h. After treatment, cells were washed three times with PBS, detached with 0.25% trypsin, collected into centrifuge tubes, and centrifuged at 1,500 × g for 5 min at 4 °C; the supernatant was discarded. RIPA lysis buffer containing 0.1% protease inhibitor was added in proportion to cell number, and lysates were incubated on ice for 30 min. The lysates were centrifuged at 12,000 × g for 15 min at 4 °C, the supernatants were collected, and protein concentrations were determined by the BCA method using the Omni-Easy™ Instant BCA Protein Assay Kit (Epizyme, Shanghai, China). Protein samples were stored at −80 °C. For protein digestion and peptide preparation, 100 µg of protein was equilibrated in 100 mM triethylammonium bicarbonate (TEAB), reduced with 10 mM tris(2-carboxyethyl)phosphine at 37 °C for 60 min, alkylated with 40 mM iodoacetamide at room temperature in the dark for 40 min, and precipitated with acetone at −20 °C for 4 h followed by centrifugation at 10,000 × g. The pellet was resuspended in 100 µL of 100 mM TEAB and digested overnight at 37 °C with trypsin at an enzyme-to-protein ratio of 1:50. Digests were acidified with 0.1% trifluoroacetic acid, desalted using HLB solid-phase extraction cartridges, vacuum-concentrated, and quantified by UV spectrophotometry on a NanoDrop One (Thermo Fisher Scientific, Wilmington, USA).

DIA proteomic and data analysis

After trypsinization and desalting on HLB cartridges, equal amounts of peptides from the MAC-T, LPS, and BAIBA groups were dissolved in MS buffer and analyzed on a Vanquish Neo UHPLC system (Thermo Fisher Scientific, Wilmington, USA) coupled to a µPAC high-throughput column (75 µm × 5.5 cm). Chromatographic separation used solvent A consisting of 2% acetonitrile in water with 0.1% formic acid and solvent B consisting of 80% acetonitrile in water with 0.1% formic acid, with an 8 min gradient. Peptides were acquired on an Orbitrap Astral mass spectrometer (Thermo Fisher Scientific, Wilmington, USA) in DIA mode under positive ion conditions with a spray voltage of 1.5 kV, an MS1 scan range of 380–980 m/z, and an MS2 scan range of 150–2000 m/z. Raw files were searched and quantified in Spectronaut version 18 using stringent criteria that included a protein and peptide FDR not exceeding 1% and a peptide confidence of at least 99%, with shared and modified peptides excluded from quantification. Peak area integration used up to six peptides per protein and three fragment ions per peptide. In total, 9,307 proteins were identified. Statistical testing employed the aov() function in R to obtain P-values, and FC thresholds were applied so that proteins with P < 0.05 and FC > 1.2 were considered upregulated, whereas proteins with P < 0.05 and FC < 0.83 were considered downregulated. Functional annotation of differentially expressed proteins was performed by aligning protein sequences against relevant protein databases using DIAMOND [71] (v2.1.8.162) in order to infer biological functions and pathway involvement.

Western blotting

Protein samples were boiled at 100 °C for 15 min and separated by polyacrylamide gel electrophoresis. Proteins were transferred to 0.45 µm nitrocellulose membranes using a sandwich transfer method. Membranes were blocked with 5% nonfat dry milk at room temperature for 1 h and then incubated with primary antibodies overnight at 4 °C. The primary antibodies were GAPDH (10,494–1-AP, Proteintech, Chicago, USA), p65 (4764 T, Cell Signaling Technology, Boston, USA), and HSPA1B (YM8139, Immunoway, Plano, USA). After primary incubation, membranes were washed three times with 1 × TBST (T1081, Solarbio, Beijing, China) for 5 min each and then incubated with horseradish peroxidase-conjugated secondary antibodies for 1 h at room temperature. Signals were developed with the Omni-ECL™ ultra-sensitive chemiluminescence kit (Epizyme, Shanghai, China), and GAPDH served as the loading control. All blots and gels were derived from the same experiment and were processed in parallel.

Statistical analysis

All figures were generated in RStudio (v4.3.1) and refined in Adobe Illustrator (v2021). Unless otherwise specified, two-group comparisons were performed using two-tailed unpaired Student’s t-test, whereas comparisons among three or more groups were performed using one-way ANOVA followed by Tukey’s HSD test in SPSS (v20.0, SPSS, Chicago, IL, USA). This approach was also applied to absolute qPCR data, including total bacterial load and Methanobrevibacter copy numbers. Statistical significance was set at P < 0.05. For microbiome and multi-omics datasets, Bray–Curtis dissimilarities were calculated and visualized by PCoA, with group separation assessed by PERMANOVA with 999 permutations using the vegan package in R [74]. This included comparisons between Mastitis and Healthy goats across species, KO, KEGG module, metabolite and protein profiles, as well as comparisons between NC and FMT-treated mice and between M-FMT and H-FMT mice in the mouse 16S rRNA gene sequencing dataset. Differences in microbial relative abundance across taxonomic ranks were assessed using Wilcoxon rank-sum tests. LEfSe was used to identify discriminative taxa, with taxa meeting LDA score (log10) > 3.0 and P < 0.05 considered significantly enriched. Spearman correlation analysis was performed using the Hmisc package in R to evaluate associations between microbial abundances and host phenotypes. In metabolomic analyses, metabolites with VIP ≥ 1 and P < 0.05 were considered differentially abundant.

Authors’ contributions

Conceptualization, Y.L., K.Z., and Y.L.C.; methodology, Y.L., Y.B.X., Y.T.Y., J.R.Z., T.Y.Z., J.J.H., Y.L.H., J.Y.Z., X.L.W., K.Z., and Y.L.C.; investigation, Y.L., Y.B.X., Y.T.Y., J.R.Z., T.Y.Z., J.J.H., K.Z., and Y.L.C.; writing - original draft, Y.L., K.Z., and Y.L.C.; writing - review and editing, Y.L., Y.B.X., Y.T.Y., K.Z., and Y.L.C.; funding acquisition, X.L.W., K.Z., and Y.L.C.; resources, Y.L., and Y.L.C.; supervision, K.Z., and Y.L.C. All authors read and approved the final version of the paper.

Funding

This research was supported by the National Key Research and Development Program of China (2022YFD1300203), the National Natural Science Foundation of China (32161143010 and 32402785), the Key Research and Development Program of Qinghai Province (2025-NK-130), the Key Research and Development Program of Ningxia Hui Autonomous Region (2025BBF02007), the local grants from Shaanxi (2025RS-CXTD-059 and Z2024-ZYFS-0044), and the China Agricultural Research System (CARS-39–03).

Data availability

The raw metagenomic sequencing data generated in this study are accessible at NCBI under BioProject accession PRJNA1311753. The 16S rRNA gene sequencing data of fecal samples from BALB/c mice have been deposited in the Genome Sequence Archive (GSA), hosted by the China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number CRA043057. The raw DIA proteomics data are available in the iProX repository under Project ID IPX0013451000. The raw metabolomics data can be found in the NGDC OMIX repository under project accession PRJCA046493.

Declarations

Ethics approval and consent to participate

All animal care procedures and experimental protocols described in this study received approval from the Animal Ethics Committee at Northwest A&F University (Xianyang, Shaanxi, China). The research was carried out strictly following the university’s guidelines for animal experimentation, authorized under License Number 20220923008.

Competing interests

The authors declare no competing interests.