Dissection of mammary cell landscape in ruminal dysbiosis-induced mastitis by single-cell RNA sequencing

Abstract

Growing evidence has underscored the contribution of gastrointestinal dysbiosis to the onset of mastitis, however, the local cellular changes responsible for the pathological processes of ruminal dysbiosis-induced mastitis (RDIM) are still unclear. Here, we profiled mammary single-cell transcriptomes in goats with RDIM, complemented by ruminal metagenomic and untargeted metabolomic analyses of rumen fluid and serum. Our results indicated that compromised lactation and barrier integrity in LumSec were linked to RDIM. Increased inflammatory macrophages and DCs, γδT and CD4⁺ T_H cell populations, along with reduced T_ex/T_reg and B cells were implicated in RDIM. Fibroblasts exhibited increased gene expression related to the extracellular matrix, while lymphatic endothelial cells and Vas-venous structures displayed elevated inflammatory gene expression. Tight junction integrity and apelin signaling pathways were compromised in Vas-capillary during RDIM. Notably, metagenomic analysis indicated that RDIM correlated with reduced ruminal microbial diversity and shifts in microbial community composition. Key metabolic pathways including microbial tryptophan metabolism, secondary bile acid biosynthesis, and vitamin metabolism were significantly diminished during RDIM. Furthermore, tryptophan-induced AHR signaling and secondary bile acid receptor GPBAR1, primarily expressed in vascular endothelial cells and macrophages, respectively, which were reduced during RDIM. Collectively, our study provides a comprehensive atlas of mammary cell landscapes in RDIM, which may enhance the understanding of mastitis pathogenesis.

Introduction

Mastitis is among the most severe diseases affecting the dairy industry, leading to substantial economic losses and posing a potential risk to public health. It has long been believed that infection of the mammary gland by pathogens, including Staphylococcus aureus (S. aureus), Escherichia coli, and Klebsiella, is the primary driver of mastitis^1,2. However, recent studies have shown that gut dysbiosis can independently induce mastitis development^3,4. Specifically, cows with mastitis exhibit altered ruminal and gut microbial profiles, characterized by increased abundances of opportunistic pathogenic Proteobacteria and reduced levels of commensal bacteria capable of metabolizing tryptophan and secondary bile acid production^1,3–5. Moreover, ruminal and fecal microbiota transplantation has been shown to trigger mastitis in recipient animals in the absence of pathogen invasion, supporting the contribution of gastrointestinal dysbiosis to mastitis pathogenesis^3,4. Multiple mechanisms are implicated in the onset of ruminal dysbiosis–induced mastitis. Specifically, ruminal dysbiosis leads to an increase in harmful bacterial metabolites or components such as lipopolysaccharide (LPS) and histamine, which trigger host immune activation and compromise barrier function, thereby promoting inflammation⁶. Disruption of microbiota-mediated metabolism further exacerbates this process. For example, ruminal dysbiosis increases the levels of sialic acid and succinate, which can serve as carbon sources to support the proliferation and virulence of opportunistic pathogens, thereby further worsening dysbiosis^6–8. Moreover, the breakdown of the gastrointestinal barrier enables opportunistic pathogens to migrate directly to distal organs, including the mammary gland, leading to mastitis^4,9. Additionally, decreased production of beneficial metabolites, such as tryptophan-derived ligands for the aryl hydrocarbon receptor (AHR) and secondary bile acids, compromises host defenses and heightens susceptibility to inflammation and pathogenic infection^1,5,10. Despite evidence supporting the role of rumen microbial and metabolic disorders in mastitis development, the specific responses of mammary cell subpopulations and genes to these changes remain unclear.

Based on single-cell RNA sequencing (scRNA-seq), earlier studies have shown that the mammary gland contains a highly heterogeneous cellular landscape, with abundant luminal epithelial cells as well as perivascular cells, vascular endothelial cells (VCs), fibroblasts, immune cells, and lymphatic vessels^11,12. Together, these populations constitute a complex architecture of epithelial ducts and lobules that are situated within connective and adipose tissues^12,13. The extensively branched lobular units and ductal system provide the structural basis for efficient lactation by enabling material exchange with the body via rich vascular and lymphatic networks, which are supported by extracellular matrix (ECM)–enriched connective tissue and adipose-dominant niches. Ducts and lobules display a bilayer organization, comprising an inner luminal epithelial layer and an outer basal myoepithelial layer^14,15. Notably, ductal areas consist primarily of branched epithelium characterized by luminal secretory cells (LumSec), while the lobular regions exhibit a bilayered epithelium predominantly composed of luminal hormone-responsive cells (LumHR)¹³. In patients with plasma cell mastitis, aberrant T-cell and neutrophil proliferation is observed, alongside impaired B-cell trafficking and differentiation mediated by the CXCL12/CXCR4 axis¹⁶. Similarly, elevated macrophage and neutrophil infiltration have been documented in individuals with granulomatous lobular mastitis¹⁷. Furthermore, increased granulocyte accumulation is detected in the mammary gland tissue of cows with S. aureus-induced chronic mastitis¹⁸.

In this study, we constructed a comprehensive single-cell atlas of mammary cells in ruminal dysbiosis-induced mastitis (RDIM). Our analysis identified 11 major cell types. Notably, the RDIM group showed higher proportions of fibroblasts, T cells, myeloid cells, basal cells, LumHR, lymphatic cells, and progenitor cells, while VCs, perivascular, LumSec, and B cells were reduced. We observed that RDIM impaired lactation function in LumSec, weakened the blood-milk barrier in VCs and epithelial cells, and increased gene expression related to inflammatory responses in myeloid cells and T cells. In addition, RDIM reshaped ECM and lymphatic vessel functions. Importantly, changes induced by ruminal microbiota included reductions in microbial tryptophan metabolism, secondary bile acid production, and vitamin metabolism. Microbial tryptophan metabolism-induced AHR signaling was predominantly expressed in VCs. The secondary bile acid receptor G protein-coupled bile acid receptor 1 (GPBAR1, formerly TGR5) was mainly expressed in macrophages and was downregulated in the RDIM group. Vitamin D receptor (VDR) was primarily expressed in LumSec, while folate receptor 2 (FOLR2) was predominantly expressed in myeloid cells; both were reduced during RDIM. Overall, this work delineates the cellular landscape of RDIM and may deepen our understanding of mastitis pathogenesis.

Results

Single-cell analysis and cell-type identification in RDIM

To assess how RDIM affects the mammary gland, goats were maintained on a high-grain diet (HGD, approximately 70% concentrate) for two months to establish an RDIM model ⁶. Subsequently, single-cell analysis, metagenomics, and untargeted metabolomics were performed to characterize the mammary cellular landscape and the accompanying microbiota-driven metabolic changes (Fig. 1a). As anticipated, goats fed a HGD exhibited significant mastitis symptoms, characterized by compromised mammary structure and elevated inflammatory markers (Suppl. Fig. 1A–C). To characterize the initiation and progression of RDIM, we performed scRNA-seq on a 10×Genomics platform using samples from the RDIM and control (Ctr) groups, yielding 48,864 cells for downstream analyses. In total, 17 clusters (C0-C16) were identified based on marker gene expression (Suppl. Fig. 1D). These cells were subsequently classified into 11 distinct cell types according to canonical lineage markers. Four clusters (C1, C4, C12, C13) were identified as fibroblasts based on the elevated expression levels of marker genes such as KERA, SFRP4, COL1A1, DCN, COL1A2, and PDGFRA (Fig. 1b–d, Suppl. Fig. 1E, F). Cluster C0 was classified as VCs due to the higher expression of genes including CCL14, MECOM, CDH5, FABP4, VWF, and PTPRB. Three clusters (C2, C11, C14) were designated as perivascular cells (PVCs) because of the increased expression of RGS5, RERGL, CLDN10, ABCC9, DES, and RASL11A (Fig. 1b–d, Suppl. Fig. 1E, F). Three clusters (C5, C8, C9) were categorized as epithelial cells: basal-myoepithelial cells (Basal, C8), characterized by KRT17, KRT14, COL17A1, MIA, IL17B, IRX4; LumHR (C9), characterized by the expression of PGR, ESR1, AREG, PCLO, NCMAP, ERICH3); and LumSec (C5), distinguished by ELF5, GLYCAM1, LALBA, SNORC, FABP3, BTN1A1) (Fig. 1b–d, Suppl. Fig. 1E, F). Two clusters (C3 and C7) were designated as T cells (TCs) owing to high expression of CTSW, IFNG, CD52, CD96, TRDC, and IL2RB (Fig. 1b–d, Suppl. Fig. 1E, F). Cluster C15 was assigned as B cells (BCs) based on enriched expression of MZB1, CD27, TNFRSF17, EAF2, DERL3, and JCHAIN. Cluster C6 was classified as myeloid cells (MCs) due to the increased expression of CXCL8, IL1B, C5AR1, CD163, C1QC, and AIF1 (Fig. 1b–d, Suppl. Fig. 1E, F). In addition, lymphatic endothelial cells (C10; CCL21, LYVE1, MMRN1, RELN, CLDN11, PKHD1L1) and progenitor cells (PCs, C16; TOP2A, BUB1B, NDC80, ASPM, CENPF, DLGAP) were also annotated (Fig. 1b–d, Suppl. Fig. 1E, F). Importantly, relative to the Ctr group, the RDIM group showed decreased proportions of VCs, PVCs, LumSec, and BCs, while fibroblasts, TCs, MCs, Basal, LumHR, and lymphatic endothelial cells were increased (Fig. 1E). These findings imply that these cell populations may be involved in lactation-related functions and the progression of mastitis.

Fig. 1. Single-cell analysis and cell-type identification in RDIM.

a Overview of the experimental workflow for single-cell RNA sequencing (scRNA-seq). Goats were maintained on a high-grain diet to induce RDIM, and mammary glands were subsequently harvested for scRNA-seq (n = 3). In parallel, ruminal and blood samples were profiled by untargeted metabolomics (n = 10), and ruminal samples were further assessed by metagenomic analysis (n = 10). b UMAP visualization showing clusters of major cell types. c Dot plot summarizing marker-gene expression across cell-type clusters. d Heatmap presenting scaled expression patterns of discriminative gene sets for major cell types. e Bar plot showing the relative abundance of major cell types in the Ctr versus RDIM groups. ***p < 0.001 by two-tailed unpaired Student’s t-test. LumHR luminal hormone-responsive epithelial cell, LumSec luminal secretory epithelial cell.

Impaired epithelial lactation function and tight junction integrity are associated with RDIM

Epithelial cells are crucial for milk production ¹³. Throughout RDIM, the proportion of LumSec cells declined, whereas LumHR became more abundant (Fig. 1e). Both LumSec and LumHR are reported to originate from the same progenitor cell ¹³. To investigate the relationships and potential transitions among luminal epithelial cells during RDIM, we conducted pseudotime analysis. The results revealed that PCs predominantly constituted branch 1, representing the initial states of luminal epithelial cells (Fig. 2a). LumSec cells were primarily found in branches 4 and 5, indicative of the cellular states associated with the secretory phenotype. Notably, the RDIM-dominant trajectory showed a marked loss of the LumSec phenotype (Fig. 2a). Milk-associated genes, including GLYCAM, LALBA, LTF, and SPP1 ¹³, showed increased expression in branches 4 and 5 (Fig. 2b), whereas ESR and PGR were upregulated in branches 2 and 3 (Suppl. Fig. 2A). Epithelial cells were subsequently partitioned into nine discrete clusters according to their transcriptomic profiles. Three clusters (C0, C2, and C7) were assigned to LumSec based on high expression of marker genes such as ELF5, XDH, MUC15, and LTF. Specifically, C0 and C2 exhibited higher expression of milk protein genes including LALBA and SPP1, whereas C7 showed increased expression of KRT15, KIT, APOE, PDCD4, and PTGDS, indicating distinct secretory characteristics (Fig. 2c, d). Three additional clusters (C3, C6, and C8) were classified as LumHR, supported by enriched expression of canonical markers including KRT7, KRT8, KRT19, ESR1, PGR, and GATA3¹². Notably, C8 showed increased expression of IFN-production–related genes, such as MX1, MX2, IFI44, and ISG15¹², while C3 showed higher expression of ATF3, FTH1, and S100A2. Cluster C4 was designated as Basal cells owing to elevated expression of KRT5, KRT14, KRT17, MYLK, and ACTA2 (Fig. 2c, d). Additionally, a hybrid Basal-LumHR population (C5) was identified, characterized by concurrent expression of luminal and basal markers together with progenitor markers including STMN1 and TOP2A^12,19. Finally, cluster C1 was annotated as an epithelial progenitor cell population due to the higher expression of genes including HSPB1, TUBA1B, ADAMTS1, ADAMTS4, NOTCH3, and RGS16 (Fig. 2c, d).

Fig. 2. RDIM is associated with impaired epithelial lactation capacity and disrupted tight-junction integrity.

a Trajectory inference of luminal epithelial cells. b Pseudotime-resolved single-cell expression dynamics of GLYCAM, LALBA, LTF, and SPP1. c Dot plot summarizing marker-gene expression across epithelial subsets. d UMAP visualization of epithelial-cell clusters based on transcriptomic profiles. e Bar plot showing the relative abundance of each epithelial subset. f Heatmap depicting scaled expression of discriminative gene sets in epithelial cells. g Volcano plot of DEGs in LumSec-1 and LumHR-1 subsets from the RDIM versus Ctr comparison. h Representative KEGG pathways enriched among DEGs in LumSec-1 and LumHR-1 subsets in the RDIM/Ctr comparison. Epithelial prol epithelial proliferating cells.

The proportions of LumSec-1 (C0), LumSec-3 (C7), and LumHR-3 (C8) decreased during the RDIM process, whereas those of Basal (C4) and most LumHR subtypes increased (Fig. 2e). Differentially expressed genes (DEGs), Gene Ontology (GO), and KEGG analyses revealed that upregulated DEGs in the LumSec-1 subtype were associated with biosynthetic processes (e.g., peptide, amide, lactose synthesis) and metabolic pathways, including vitamin B6 and fatty acid metabolism (Fig. 2f, Suppl. Fig. 2C). In the LumHR-1 subtype, upregulated DEGs exhibited regulatory effects on biosynthetic and metabolic processes, while DEGs related to tight junctions, bacterial invasion of epithelial cells, glycolysis, necroptosis, and breast cancer were also enriched (Fig. 2f, Suppl. Fig. 2D). Additionally, DEGs in the Basal subtype were linked to cell differentiation, anatomical structure morphogenesis, and organelle organization (Fig. 2f, Suppl. Fig. 2E). Notably, DEGs of LumSec-1 between RDIM and control samples were associated with the TGF-β signaling pathway (TGFB2, MAPK3 and NEO1), tight junctions (TJP1, TJP3 and CLDN8), cellular metabolic processes, and milk component–synthesis–related pathways, including fatty acid metabolism (ACSL1 and ACSL3) and steroid biosynthesis (Fig. 2g, h, Suppl. Fig. 2F). RDIM-induced DEGs in the LumHR-1 subtype were associated with oxidative phosphorylation, tight junctions (TJP1, TJP3 and CLDN8), and inflammatory response pathways (HIF-1 and MAPK signaling pathway) (Fig. 2g, h, Suppl. Fig. 2G). EGR1, a key regulator of inflammation in macrophages^20,21, was upregulated in the LumHR-1 subtype in the RDIM group, indicating enhanced inflammation and potential tissue damage. Additionally, genes related to milk production, such as GLYCAM1 and LTF, were downregulated in the LumHR-1 subtype in the RDIM group (Fig. 2g). In the LumHR-3 subtype, RDIM-induced DEGs were linked to inflammation (CXCR4, CSF1 and TNFAIP3) and the apelin signaling pathway (SPP1, SLC8A1, NRAS, and MRAS) (Suppl. Fig. 2H, I). In Basal cells, RDIM-induced DEGs were significantly enriched in TGF-β signaling, ECM–receptor interaction, and cancer-related pathways (Suppl. Fig. 2J, K). Collectively, these findings suggest that RDIM impairs mammary lactation function and disrupts tight junction integrity in epithelial cells.

Inflammatory MCs and dendritic cells represent pathogenic cell types implicated in the development of RDIM

MCs showed an increased proportion in the RDIM group and have been reported to be involved in inflammation²² (Fig. 1e). MCs were further classified into 11 subsets, including dendritic cells (DCs), macrophages, monocytes (Mono), and proliferating myeloid cells (Mye-prol) (Fig. 3a). Four distinct DC subsets were identified: DC-1 and DC-2 highly expressed CD1E, TXN, LMNA, and GPR183; DC-3 expressed CST3, CLEC9A, BATF3 and IRF8, indicating a conventional DC state¹². DC-4 highly expressed RGS16 and APOE. Macrophages were defined by the expression of C3, C1QA, C1QB, and C1QC, and were divided into five distinct subsets. Cluster C0 displayed an M2-like phenotype (Macro M2), supported by high expression of ACP5, CD163, F13A1, MRC1, and LYVE1. The Macro-IFN subset (C4) was enriched for genes associated with the M1-like phenotype (CXCL10) and interferon response genes (IFIT1, IFIT2, IFIT3, MX1, and MX2). Additionally, three other macrophage clusters were identified: Macro-1 (ATF3, GADD45B, RGS1, and SGK), Mcaro-2 (PAEP, PYCARD, GNAS, and NDUFA13) and Macro-3 (COTL1, CXCL12, ITGA4, OLR1, PLAC8, PTGS2, and TNFSF10). Mono cells exhibited high expression of inflammation-associated genes, including CD14, S100A8, S100A12, CXCL8, and IL1B. A Mye-prol cluster, marked by STMN1, TUBA1B, and TOP2A, was also identified (Fig. 3b).

Fig. 3. Inflammatory MCs and DCs represent pathogenic cell types implicated in the development of RDIM.

a UMAP visualization of myeloid-cell clusters. b Dot plot summarizing marker-gene expression across myeloid subsets. c Heatmap depicting the scaled expression patterns of discriminative gene sets in myeloid cells. d UMAP plot showing M2 and M1 gene signature scores for distinct myeloid cell states. e, f DEGs and representative KEGG pathways enriched by DEGs of Macro-M2 subset from the RDIM versus Ctr comparison. g, h DEGs and representative KEGG pathways enriched by DEGs in the Macro-1 subset in RDIM/Ctr comparison. i, j Volcano plot showing DEGs of Macro-IFN and DC-1 subsets in RDIM/Ctr comparison. k, l DEGs and representative KEGG pathways enriched by DEGs of DC-1 subset in RDIM/Ctr comparison. m, n DEGs and representative KEGG pathways enriched by DEGs of DC-4 subset in RDIM/Ctr comparison. o, p DEGs and representative KEGG pathways enriched by DEGs of Mono subset in RDIM/Ctr comparison. Macro macrophage, Mono monocyte, DC dendritic cell.

Analysis of upregulated DEGs together with KEGG enrichment indicated that inflammation-related programs—encompassing the Toll-like receptor signaling pathway, NOD-like receptor signaling pathway, and Th1/Th2/Th17 differentiation—were mainly enriched in the Macro-1, Macro-3, Macro-IFN, DC-1, DC-3, and Mono subsets (Fig. 3c, Suppl. Fig. 3A). M1 signature scores were also significantly elevated in the Macro-1, Macro-3, and Macro-IFN subsets (Fig. 3d). Notably, myeloid cell populations were largely elevated in the RDIM group (Suppl. Fig. 3B). Interestingly, RDIM reduced the expression of genes associated with the M2-like phenotype, such as ACP5, IL10, F13A1, MRC1, and LYVE1, while upregulating genes related to inflammatory response (IL-1B, NFKBIA, TNF, and CXCL8) and lipid metabolism (APOE)²³ (Fig. 3e, f). In the RDIM group, genes involved in the Toll-like receptor signaling pathway, NOD-like receptor signaling pathway, and Th cell differentiation were upregulated in the Macro-1, Macro-2, and Macro-IFN subsets (Fig. 3g–i, Suppl. Fig. 3C–E). Specifically, RDIM was associated with dysregulated expression of genes linked to lipid metabolism, atherosclerosis, and neutrophil extracellular trap formation in the Macro-1 subset (Fig. 3g, h), and genes related to natural killer cell-mediated cytotoxicity (IFNAR1/2, IFI6) in the Macro-IFN subset (Fig. 3I, Suppl. Fig. 3E). In the DC-1 subset, RDIM increased the expression of genes associated with inflammatory responses (IL-1B, IL-1A, TNF, ICAM1, NFKBIA, and CXCL8) (Fig. 3j, k). In parallel, genes involved in inflammation, antigen processing and presentation, and lipid metabolism were modulated in the DC-2 and DC-4 subsets (Fig. 3l–n, Suppl. Fig. 3F). In the DC-4 subset, DEGs related to cell death (TNF, CXCL9, and SPP1) were enriched in the RDIM group (Fig. 3m, n). DEGs in the DC-3 subset were primarily associated with ECM-receptor interaction (Suppl. Fig. 3G, H). DEGs in the Mono subset were associated with NF-κB and PI3K-Akt signaling pathways (Fig. 3o, p). Taken together, these observations indicate that an increased abundance of proinflammatory MCs and DCs may contribute substantially to RDIM pathogenesis.

Increased proinflammatory TCs and reduced BCs were involved in RDIM

Altered cells associated with adaptive immunity were also observed in RDIM, characterized by increased TCs and reduced BCs counts (Fig. 1e). TCs were further classified into eight subtypes: two CD8⁺ T cell subsets (CD8⁺ T-1/2), defined by markers such as CD96, IL2RB, GZMA, CTSW, BEND5, and IBTK; CD8⁺ effector memory TCs (CD8⁺ T_em), marked by IFNG, FASLG, ITGAL, GZMK, and CRTAM; CD4⁺ helper T cells (CD4⁺ T_H), identified by CD4, IL7R, and CD69; CD4⁺ regulatory T cells (CD4⁺T_reg), characterized by CD4, IL7R, CTLA4, NDFIP2, GATA3, DUSP4, TIGIT, and IKZF2; γδT cells, distinguished by TRDC, ICOS, S100A4, SAMSN1, and IL17A; exhausted T cell (T_ex)/T_reg-like cells, noted for MX1, IFI6, ISG15, OAS1, IFIT3, and GBP2; and NK/NKT, recognized by NCR1, KLRF2, KLRF1, and CSF (Fig. 4a, b). Notably, the RDIM group showed significantly higher proportions of CD8⁺T-1, γδT, and CD4⁺T_H compared with the control group, whereas CD8⁺T_em, T_ex/T_reg, and NK/NKT were reduced (Fig. 4c). In contrast, no significant differences were observed in the proportions of CD4⁺T_reg and CD8⁺T-2 between the RDIM and control groups (Suppl. Fig. 3I).

Fig. 4. Increased proinflammatory TCs and reduced BCs were involved in RDIM.

a UMAP visualization showing the clustering of NK/T cell populations. d Dot plot summarizing marker-gene expression across NK/T cell subsets. C Box plots comparing the proportions of the indicated NK/T cell subsets. d Heatmap depicting the scaled expression patterns of discriminative gene sets in NK/T cells. e Volcano plot showing DEGs of CD8⁺ T-1, γδT and CD4⁺T_H subsets in RDIM/Ctr comparison. f Representative KEGG pathways enriched by DEGs of CD4⁺ T_H subset in RDIM/Ctr comparison. g, h DEGs and representative KEGG pathways enriched by DEGs of NK/NKT and T_ex/T_reg subsets in RDIM/Ctr comparison. i UMAP plot showing clusters of B cells. j The numbers of B cells of indicated subsets. k Dot plot showing the expression of marker genes in B cell subsets. l Heatmap showing scaled expression of discriminative gene sets for B cells. m Volcano plot showing DEGs of B plasma-1/2 and B naive subsets in RDIM/Ctr comparison. *p < 0.05 and ***p < 0.001 by two-tailed unpaired Student’s t-test (C). T_ex, exhausted T cells; T_em, effector memory T cells; T_reg, T regulatory cells; T_H, T helper cells.

Analysis of upregulated DEGs together with KEGG enrichment showed that DEGs involved in Toll-like receptor signaling, NOD-like receptor signaling, and Th cell differentiation were significantly overrepresented in the CD8⁺T_em, CD4⁺T_H, CD4⁺T_reg, γδT, and NK/NKT cell subtypes. Specifically, DEGs related to NK cell-mediated cytotoxicity were enriched in the CD8⁺T-1/2 and CD8⁺T_em subtypes (Fig. 4d, Suppl. Fig. 3J). Further analysis showed that in the CD8⁺T-1 subtype, DEGs were predominantly involved in antigen processing and presentation (Fig. 4e, Suppl. Fig. 3K). In the γδT subtype, DEGs associated with Th17 cell differentiation, including AHR, HSP90AB1, and HSP90AA1, were identified (Fig. 4e, Suppl. Fig. 3L). Additionally, in the CD4⁺Th subtype, DEGs related to both Th1/2 and Th17 cell differentiation, such as AHR, HSP90AB1, and HSP90AA1, were observed (Fig. 4e, f). Moreover, RDIM altered genes associated with apoptosis (PRF1, TNF, ATF4, MAP2K2, TUBA1B, PARP1, and PNPT11) and NK cell-mediated cytotoxicity (PRF1, TNF, MAP2K2, and ACTG1) in the NK/NKT subtype (Fig. 4g, h). Similarly, DEGs related to apoptosis, including PRF1, CTSW, GADD45B, ACTB, FASLG, and LMNA, were also found in the T_ex/T_reg subtype (Fig. 4g, h). In the CD8⁺T_em subtype, DEGs were enriched for pathways involved in antigen processing and presentation as well as NF-κB signaling (Suppl. Fig. 3M, N). Subsequently, BCs were classified into three subtypes: B plasma-1 (characterized by EAF2, JCHAIN, TNFRSF17, LMAN1, SPCS2, and FKBP11), B plasma-2 (marked by NT5E, SAP30, IRF4, MANF, and OSTM1) and B naïve (distinguished by YBX3, KLF2, APLP2, CD9, and CXCL12). Notably, the number of BCs in all three subtypes was significantly reduced in RDIM (Fig. 4i–l). The DEGs in BCs were predominantly involved in apoptotic processes and proteolysis (Fig. 4m, Suppl. Fig. 3O), indicating impaired antibody production and processing in RDIM. Collectively, these findings demonstrate that RDIM is characterized by an increase in proinflammatory TCs and a reduction in B cell populations.

Alterations in fibroblasts and lymphatic endothelial cells were observed in RDIM

Fibroblasts represent one of the major cell populations in the mammary gland, and both their abundance and transcriptional programs can be remodeled, thereby promoting inflammatory responses. Indeed, fibroblast counts were elevated in RDIM (Fig. 1e), and these cells were classified into five subtypes based on classical marker genes such as COL1A1, OGN, and MFAP5 (Fig. 5a, b). Specifically, Fibro-2 showed elevated expression of IGFBP2, POSTN, and IGFBP3, whereas Fibro-5 displayed increased levels of inflammation-related genes, including ISG15, MX1, IFIT3, IFIT1, and EPSTI1 (Fig. 5a, b). Notably, the proportions of Fibro-1 and Fibro-2 were increased, whereas those of Fibro-3, Fibro-4, and Fibro-5 were decreased in the RDIM group (Fig. 5c). Analysis of upregulated DEGs together with GO enrichment indicated that DEGs in Fibro-1/2 were mainly linked to the ECM, implying that ECM remodeling may be crucial for RDIM development (Fig. 5d, e). In Fibro-3/4, upregulated DEGs were linked to vasculature development (Fig. 5d, Suppl. Fig. 4A). Notably, DEGs in Fibro-5 were associated with immune system processes and response to stress (Suppl. Fig. 4A). Further analysis showed that RDIM-induced changes in DEGs in Fibro-1 were involved in ECM-receptor interaction (AGRN, COL6A5, LAMA2, HSPG2, COL1A1, FN1, COL4A1, and COL4A2), TGF-β signaling pathway (TGFBR2, THBS1, FBN1, and TGFBR1), and cellular senescence (TGFBR2, MYC, and TGFBR1) (Fig. 5f). Similarly, in Fibro-2/3, RDIM-induced DEGs were enriched in pathways related to ECM–receptor interaction and TGF-β signaling (Fig. 5g, Suppl. Fig. 4B). Specifically, DEGs associated with cellular senescence (TGFB3, VDAC3, and TGFB2) and proteoglycans in cancer were identified in the Fibro-2 subtype (Fig. 5g), while DEGs associated with cancer-related processes were observed in the Fibro-4 subtype (Suppl. Fig. 4C). Additionally, RDIM-induced DEGs in the Fibro-5 subtype were linked to Th17 cell differentiation (NFKBIA, IFNGR2, AHR, IL1B, and GATA3) and infection-related inflammatory responses (Suppl. Fig. 4D). These results suggest that alterations in the ECM and inflammatory responses within fibroblasts are integral to the development of RDIM.

Fig. 5. Alterations in fibroblasts and lymphatic endothelial cells were observed in RDIM.

a UMAP visualization of fibroblast clusters. b Dot plot summarizing marker-gene expression across fibroblast subsets. c Box plots comparing the proportions of the indicated fibroblast subsets. d Heatmap depicting the scaled expression patterns of discriminative gene sets for fibroblasts. e Representative GO pathway enriched by DEGs of Fibro-1/2 subsets. f, g DEGs and representative KEGG pathways enriched by DEGs of Fibro-1/2 subsets in RDIM/Ctr comparison. h UMAP plot and marker genes of lymphatic endothelial cells. i Heatmap showing scaled expression of discriminative gene sets for lymphatic endothelial cells. j Representative GO pathways enriched by DEGs of Lym-1/5 subsets. k Representative KEGG pathways enriched by DEGs of Lym-1/5 subsets in RDIM/Ctr comparison. ***p < 0.001 by two-tailed unpaired Student’s t-test (c). Fibro fibroblasts, Lym lymphatic endothelial cells.

Lymphatic endothelial cells were next divided into six subtypes, designated as Lym-1 through Lym-6 (Fig. 5h). Lym-1 and Lym-2 exhibited expressed canonical LC marker genes, including CCL21, LYVE1, and PROX1¹², and their proportions did not differ significantly between the Ctr and RDIM groups (Fig. 5h, i, Suppl. Fig. 4E). Lym-3 showed high expression of MEOX1, a transcription factor that promotes profibrotic responses in human cardiac fibroblasts²⁴, whereas Lym-6 exhibited elevated expression of genes related to adipogenesis or lipid accumulation, including APOE (Fig. 5h, i). The proportions of Lym-3 and Lym-6 were notably reduced in the RDIM group (Suppl. Fig. 4E). Lym-4 and Lym-5 expressed genes associated with fibroblasts (DCN, COL1A1, and FAP) and immune function (CTSW, TRDC, and FASLG), respectively, and both subsets were significantly expanded in the RDIM group (Fig. 5h, i, Supplementary Fig. 4E). GO analysis revealed that upregulated genes in Lym-1 and Lym-2 were primarily involved in tube development, while those in Lym-3 and Lym-5 were associated with immune responses. Additionally, genes related to the ECM were upregulated in the Lym-4 subtype (Fig. 5j, Suppl. Fig. 4F). RDIM-induced DEGs in the Lym-1 subtype were linked to focal adhesion and Hippo signaling pathways, while changes in the apelin signaling pathway were observed in the Lym-5 subtype during RDIM, suggesting its role in regulating tissue development and inflammation (Fig. 5k, Suppl. Fig. 4G, H).

Enhanced inflammatory activity together with impaired barrier integrity in VCs may play a crucial role in RDIM pathogenesis

VCs represent the principal cellular components of the blood–milk barrier and regulate molecular exchange between the blood and milk. We subsequently categorized VCs into three subtypes: Vas-venous (characterized by ACKR1, SELP, and SELE), Vas-capillary (marked by CA4, RGCC, and ESM1), and Vas-1 (identified by NOTCH3, ABCC9, and RGS5) (Fig. 6a–c). Notably, during RDIM, the proportion of Vas-venous increased, whereas that of Vas-capillary and Vas-1 decreased (Fig. 6d). Upregulated genes in the Vas-venous subtype were involved in inflammatory responses, including Th17 cell differentiation and TNF signaling (Fig. 6e), which is consistent with previous findings that ACKR1⁺ vascular cells are associated with inflammation development²³. Upregulated DEGs in the Vas-capillary subtype were associated with tight junctions and the apelin signaling pathway, implying that compromised barrier integrity and disrupted apelin signaling may contribute to RDIM pathogenesis (Fig. 6e). We further confirmed that genes related to the apelin signaling pathway were predominantly expressed in the Vas-capillary and Vas-1 subtypes, but not in the Vas-venous subtype (Fig. 6f, Suppl. Fig. 5A). DEGs associated with apoptosis and NF-κB signaling, such as NFKBIB, IL6, and TLR4, were altered in the Vas-venous subtype during RDIM (Fig. 6g). In the Vas-capillary and Vas-1 subtypes, genes associated with apelin signaling pathway (PIK3CG, SPP1, PIK3C3, and MAPK3) and tight junctions (TJP2) were also changed during RDIM (Fig. 6h, Suppl. Fig. 5B).

Fig. 6. Enhanced inflammatory responses and compromised barrier function of VCs contribute to RDIM pathogenesis.

a UMAP visualization of vascular endothelial cell clusters. b UMAP feature plots showing the expression patterns of selected marker genes across vascular endothelial cell subsets. c Heatmap depicting the scaled expression profiles of discriminative gene sets in vascular endothelial cells. d Box plots comparing the proportions of the indicated vascular endothelial cell subsets. e Representative KEGG pathways enriched by DEGs of Vas-venous and Vas-capillary subsets. f Dot plot showing the expression of genes associated with apelin signaling pathway in vascular endothelial cell subsets. g, h DEGs and representative KEGG pathways enriched by DEGs of Vas-venous and Vas-capillary subsets in RDIM/Ctr comparison. i UMAP plot and marker genes of perivascular cells. j UMAP feature plots showing the expression patterns of selected marker genes across perivascular cell subsets. k Heatmap depicting the scaled expression profiles of discriminative gene sets for perivascular cells. l Box plots comparing the proportions of the indicated perivascular cell subsets. m Representative KEGG pathways enriched among DEGs in the pericyte and VSMC-2 subsets from the RDIM versus Ctr comparison. ***p < 0.001 by two-tailed unpaired Student’s t-test (d). VSMC, vascular smooth muscle cell.

Perivascular cells were further classified into three subtypes based on marker expression: vascular smooth muscle cell (VSMC)-1, VSMC-2, and pericytes (Fig. 6i–k). In the RDIM group, the proportion of pericytes was increased, whereas VSMC-2 showed a reduced proportion; by contrast, VSMC-1 displayed no significant change (Fig. 6l, Suppl. Fig. 5C). Upregulated genes in these subtypes were associated with tight junctions and cellular processes such as oxidative phosphorylation (Suppl. Fig. 5D). RDIM-induced DEGs in pericytes were related to ECM-receptor interaction and inflammatory responses, including CCL2, NFKBIA, and HIF1A (Fig. 6m, Suppl. Fig. 5E). In the VSMC-2 subtype, RDIM-responsive DEGs were significantly enriched in tight-junction–related pathways (e.g., JUN, ROCK2, and MYH10) and Wnt signaling pathway (e.g., TLE3, SFRP2, and DKK2) (Fig. 6m, Suppl. Fig. 5F). Additionally, DEGs in the VSMC-1 subtype were associated with tight junctions and ECM-receptor interactions (Suppl. Fig. 5G).

Alterations in microbial metabolism influence the single-cell transcriptional landscapes of mammary cells

We subsequently examined the profiles of ruminal microbiota and their metabolic functions, as well as their role in altering mammary single-cell transcriptional profiles. The results showed that the ruminal microbiota in the RDIM group had lower alpha diversity than that in the control group (Fig. 7a, Suppl. Fig. 5H). In addition, the overall rumen microbial community structure differed significantly between the Ctr and RDIM groups (Fig. 7b). At the phylum level, RDIM was characterized by decreased abundances of Bacteroidota, Euryarchaeota, and Uroviricota, along with increased abundances of Firmicutes and Actinobacteria (Fig. 7c). LEfSe analysis further revealed that the RDIM group had lower abundances of Prevotella_sp, Clostridia_bacterium, and Bacteroidaceae_bacterium abundances, and higher abundances of Ruminococcus_sp__M6_2020, Ruminococcus_sp, and Butyrivibrio_sp (Fig. 7d). Furthermore, the ruminal microbiota in the RDIM group displayed functional profiles that were clearly distinct from those of the Ctr group (Suppl. Fig. 5I). KEGG enrichment analysis indicated that pathways including methane metabolism, folate biosynthesis, riboflavin metabolism, tryptophan metabolism, and secondary bile acid biosynthesis were downregulated in RDIM, whereas peptidoglycan biosynthesis, bacterial secretion systems, histidine metabolism, and vancomycin resistance were upregulated (Fig. 7e). We further confirmed the DEGs associated with tryptophan metabolism (Suppl. Fig. 5J, K). Consistent with the KEGG analysis of ruminal microbiota, our results demonstrated that the RDIM group exhibited decreased levels of metabolites associated with tryptophan metabolism (e.g., 5-hydroxyindoleacetic acid [5-HIAA]), secondary bile acids (e.g., deoxycholic acid [DCA] and lithocholic acid [LCA]), and vitamins metabolism (e.g., riboflavin, pantothenic acid, niacinamide), along with increased levels of harmful metabolites linked to inflammation (e.g., histamine and cholesterol) in the rumen (Fig. 7f, g). Additionally, serum analysis showed reduced levels of metabolites associated with tryptophan metabolism (e.g., indole acetaldehyde [IAld], indole lactate [ILA], and tryptamine) in the RDIM group (Fig. 7h, i).

Fig. 7. Changes in microbial metabolism reshape the single-cell transcriptional landscape of mammary cells.

a Sobs index of the ruminal microbiota in the Ctr and RDIM groups (n = 10). b Beta-diversity analysis of the ruminal microbiota comparing the Ctr and RDIM groups (n = 10). c Phylum-level composition of ruminal microbiota (n = 10). d LEfSe analysis revealing differentially enriched species in indicated groups (n = 10). e Representative KEGG pathways for microbial functions enriched in ruminal microbiota of indicated groups. f, g Differentially abundant metabolites and their enriched pathways in ruminal samples from indicated groups (n = 10). h, i Differentially abundant metabolites and their enriched pathways in serum samples from indicated groups (n = 10). j Dot plot illustrating expression levels of receptor genes for microbial metabolites in specified cell subsets. k UMAP feature plot showing GBPAR1 expression across myeloid cell subsets, accompanied by violin plots comparing GBPAR1 levels between the RDIM and Ctr groups. L Violin plots comparing AHR expression in the indicated cell subsets between RDIM and Ctr groups. m UMAP plot illustrating CYP1A1 expression in vascular endothelial cell subsets. n, o Violin plots showing VDR and FPLR2 levels in specified cell subsets between RDIM and Ctr groups. ***p < 0.001 by two-tailed unpaired Student’s t-test (a, k, l, n, o).

We further investigated the potential effects of microbial metabolites on mammary transcriptional profiles by examining the expression of receptors for bile acids, microbial tryptophan metabolism, and vitamins. The results demonstrated that the secondary bile acid receptor GBPAR1 (also known as TGR5) was predominantly expressed in myeloid cells (Fig. 7j, k). Consistent with the reduced levels of DCA and lithocholic acid in the rumen, GBPAR1 expression was decreased in the Macro-IFN, Macro-2, and Macro-M2 subtypes within the RDIM group (Fig. 7k). The AHR, which recognizes tryptophan-derived indole derivatives^5,25, was expressed in most cell types, including basal cells, TCs, and myeloid cells, with increased AHR expression observed in the RDIM group (Fig. 7j, l). However, CYP1A1, an enzyme that prevents ligand-induced overactivation of AHR²⁵, was primarily expressed in vascular subtypes, particularly in the Vas-capillary subtype, suggesting that microbial ligand-induced activation of AHR mainly occurs in VCs (Fig. 7j, m). Additionally, VDR expression was predominantly found in LumSec cells and was reduced in the RDIM group, specifically in the LumSec-1/2 and LumHR subtypes (Fig. 7j, n), Meanwhile, FOLR2 expression was mainly detected in myeloid cells and was also reduced in the RDIM group, indicating its role in regulating immunity (Fig. 7j, o). These results imply that shifts in microbial metabolic activity can reshape the single-cell transcriptional landscape of mammary cells.

Discussion

Emerging evidence has demonstrated that gastrointestinal dysbiosis can induce mastitis in the mammary gland without direct pathogen infection, and regulate host health and physiological homeostasis^3,4,26. However, the local cellular landscape of the mammary gland during RDIM remains insufficiently characterized. Here, we constructed a comprehensive atlas of cellular and molecular alterations during RDIM at single-cell resolution, elucidating the potential relationship between microbial metabolism and mammary transcriptional profiles. Our findings reveal impaired lactation function in LumSec, compromised barrier function and apelin signaling in VCs, and heightened inflammatory responses in macrophages, DCs, γδT cells, and Th cells. Additionally, we observed alterations in the ECM and lymphatic vessels. We also identified impaired metabolism of tryptophan, secondary bile acids, and vitamins, which may influence cell signaling in myeloid cells, VCs, and LumSec (Fig. 8). The data generated in this study constitute a valuable resource for elucidating RDIM pathogenesis and will guide future efforts to identify potential therapeutic strategies for mastitis and breast diseases.

Fig. 8. Proposed mechanistic framework linking rumen microbial metabolic perturbations to cell-type-specific transcriptional reprogramming in the mammary gland during the RDIM.

Chronic HCD feeding triggers ruminal dysbiosis and systemic metabolic disruption, coinciding with significant shifts in the mammary single-cell transcriptome: expansion of T cells, myeloid cells, LumHR, basal cells, fibroblasts, and lymphatic endothelial cells, alongside contraction of B cells, Lumsec, vascular endothelial, and perivascular cell populations. Impaired microbial production of tryptophan catabolites (e.g., 5-HIAA, IAld, ILA), secondary bile acids (e.g., DCA, LCA), and essential vitamins—coupled with accumulation of potentially detrimental metabolites such as histidine—may collectively drive RDIM-associated mammary remodeling by engaging distinct receptor-mediated signaling pathways in a cell-type-restricted manner. Specifically, tryptophan-derived AHR ligands preferentially activate AHR in vascular endothelial cells; secondary bile acids signal predominantly through GPBAR1 in myeloid cells to modulate inflammatory tone; and microbial-derived vitamins—likely engage the VDR in Lumsec cells and FOLR2 in monocytes to support epithelial differentiation and immune regulation.

The proportions of LumSec were decreased while those of LumHR were increased during RDIM, accompanied by reduced expression of lactation-associated genes, including GLYCAM, LALBA, and LTF¹³. LumSec and LumHR can be differentiated from luminal progenitor cells¹³. Pseudotime analysis indicated that the differentiation process from progenitor cells into LumSec may be inhibited during RDIM, suggesting that impaired lactation function is associated with RDIM. Notably, these differentiation processes can also be influenced by developmental stages¹³. An increased proportion of luminal progenitor cells has been linked to poor prognosis and a dysfunctional immune microenvironment in breast cancer²⁷, implying that diet-induced gut dysbiosis and alterations in the mammary epithelial cell landscape may influence breast cancer development. Impaired tight junction-related genes, such as TJP1, TJP3 and CLDN8, which have been reported to decrease during mastitis^3,6, were also altered in LumSec and LumHR subtypes. Additionally, many metabolic genes and signaling pathways, including oxidative phosphorylation and mitophagy, which have been shown to modulate breast cancer metastasis²⁸, were altered in epithelial cells during RDIM.

MCs, particularly macrophages and DCs associated with inflammatory responses, were significantly increased during RDIM. related to inflammatory responses, such as Th1/2/17 cell differentiation and Toll- and NOD-like receptor signaling pathways, were predominantly observed in DC-1/2 (GPR183⁺), RGS⁺ macrophage (M1-like), Mcro-IFN and Mono. Previous studies have reported that RGS⁺ macrophages are mainly distributed in the ductal and lobular regions^12,19, and their role in regulating epithelial functions and inflammation has been demonstrated during S. aureus-induced mastitis²⁹. Additionally, we identified an M2-like macrophage subtype characterized by increased gene expression associated with lipid metabolism and atherosclerosis (APOE) and proinflammatory responses (TNF, IL1B, and CXCL8). This subtype also exhibited altered signaling pathways related to cell death, including apoptosis and ferroptosis, suggesting its involvement in RDIM pathogenesis. In general, M2-like macrophages are found in both the epithelial and stromal compartments of the mammary gland and are frequently enriched in the breast microenvironment^12,19,30,31. DC-3 expressed markers of conventional DCs (CLEC9A) with lower levels of inflammation-related gene expression, indicating a tissue-resident phenotype¹². CD14⁺ monocytes were also identified, exhibiting highly proinflammatory gene expression profiles (S100A8, CXCL8, and IL1B)²³, further implicating their role in RDIM development. In addition to MCs, increased inflammatory gene expression was also observed in CD8⁺ T_em, CD4⁺ T_H and γδT cells, consistent with previous reports that gut dysbiosis drives the expansion of inflammatory T cells in the mammary gland⁴. Many oxidative stress-associated genes, such as FOS, JUN, and DUSP4, were altered in CD8⁺ T and CD4⁺ T_reg cells, indicating an enhanced response to oxidative stress during RDIM³². The proportions of T_ex/T_reg and NK/NKT cells decreased during RDIM, correlating with changes in gene expression related to apoptosis. Additionally, B-cell abundance was markedly reduced during RDIM, which may increase vulnerability to pathogen-induced mastitis under conditions of gastrointestinal dysbiosis^5,33.

Increased fibroblast activity was observed during RDIM, particularly with upregulated genes associated with the ECM and ECM-receptor interactions. Studies have demonstrated that stress-induced inflammatory responses, especially those mediated by IL-1β, can activate RELA-dependent expression of the transcription factor MEOX1 in fibroblasts, leading to tissue fibrosis^24,34,35. Elevated reactive oxygen species (ROS) levels and mitochondrial damage have also been linked to mastitis-associated fibrosis in dairy cows³⁶. Future research should investigate how gut dysbiosis mediates inflammation to promote fibrosis. Notably, MEOX1 was highly expressed in Lym-3, and DEGs related to ECM-receptor interaction were altered during RDIM; however, it remains unclear whether lymphatic vessels exhibit a profibrotic response during this process. Additionally, reductions in genes associated with the apelin signaling pathway were identified in lymphatic vessels and vasculature, along with decreased expression of genes involved in tight junctions, implying that apelin signaling contributes to regulating barrier function in the mammary gland. In the lung, apelin signaling has also been reported to participate in host defense against pathogen invasion and tissue inflammation, both of which can be modulated by microbiota-mediated activation of the AHR³⁷. Moreover, an increase in Vas-venous cells (ACKR1⁺) during RDIM was accompanied by elevated proinflammatory gene expression, consistent with previous findings that ACKR1⁺ endothelial cells contribute to inflammation development²³. Notably, aberrant proliferation of T cells and neutrophils is observed in patients with plasma cell mastitis, accompanied by impaired CXCL12/CXCR4 axis–mediated B-cell trafficking and differentiation¹⁶. Similarly, elevated macrophage and neutrophil infiltration were detected in individuals with granulomatous lobular mastitis¹⁷. Moreover, in bovine models of S. aureus-induced chronic mastitis, granulocyte accumulation is markedly elevated within mammary gland tissue¹⁸. These results indicated that the increase in immune cell infiltration may be conserved in different types of mastitis. However, neither increased fibroblast abundance nor impaired Lumsec and VC function has been reported in prior studies, suggesting that vascular barrier leakage and fibrotic remodeling are pathological hallmarks specific to RDIM.

We further characterized the alterations in ruminal microbiota and metabolism during RDIM. The results demonstrated that ruminal diversity and structure were significantly altered in the RDIM group, particularly with a notable decrease in Prevotella_sp. As one of the primary species responsible for propionate production, which is essential for lactose synthesis³⁸, this reduction suggests Prevotella_sp’s critical role in regulating lactation function. This dynamic interaction between immune cells and their metabolic environment orchestrates the balance between anti-inflammatory and pro-inflammatory responses³⁹. Metagenomics and metabolomics analyses revealed that individuals from the RDIM group exhibited impaired tryptophan metabolism, secondary bile acid biosynthesis, and vitamin metabolism, while histidine and cholesterol metabolism were enhanced. These observations align with prior reports that cows with mastitis exhibit reduced ruminal 5-HIAA and lower mammary DCA levels^1,6. Administration of 5-HIAA and DCA alleviated mastitis in mice by activating AHR and GBPAR1, respectively^6,10. We observed reduced GBPAR1 expression in macrophages during RDIM, consistent with previous reports of decreased GBPAR1 levels in transition cows experiencing excessive lipolysis and associated immunosuppression in MCs⁴⁰. This suggests that secondary bile acid-mediated immune regulation in macrophages may influence the development of mastitis. Notably, ligand-induced AHR activation was predominantly expressed in endothelial cells, potentially modulating the apelin signaling pathway to enhance mammary homeostasis and protect against RDIM^37,41. Interestingly, gut microbial bile acid metabolism has been linked to activation of the apelin signaling pathway⁴²; however, its functional role in RDIM remains to be elucidated. Additionally, we noted reduced levels of vitamin-associated receptors VDR and FOLR2 in LumSec and MCs, respectively. VDR has been recognized as an essential regulator of mammary development and the pregnancy–lactation cycle⁴³. Importantly, bile acids can promote vitamin metabolism, thereby mitigating metabolic syndrome induced by an excessive-energy diet⁴⁴. FOLR2⁺ macrophages located in a perivascular niche have been reported to be associated with CD8⁺ T cell infiltration and improved breast cancer prognosis⁴⁵. This implies that microbial dysbiosis-induced reduction of FOLR2 in macrophages may affect breast cancer outcomes.

In conclusion, we generated a comprehensive single-cell atlas of the cellular landscape and molecular features of RDIM, underscoring the potential involvement of ruminal microbiota and metabolism in shaping mammary transcriptional programs. These results not only deepen our understanding of mastitis pathogenesis but also provide a valuable resource for future investigations into mammary physiology and disease. Notably, while our study provides a comprehensive cellular landscape for RDIM and highlights the potential link between ruminal microbial metabolism and mammary cell responses, future research should address several limitations. First, the limited sample size necessitates validation in larger, independent cohorts to strengthen the robustness and generalizability of our conclusions. Second, future studies should investigate whether the single-cell transcriptional and functional characteristics observed in the mammary gland of this model are conserved across other mammalian species such as humans and cows. Third, spatially resolved information for RDIM is needed to map the distribution of cell clusters and clarify their interactions. Finally, additional studies are required to verify the contribution of microbial-metabolism–driven cell receptor signaling to mastitis pathogenesis.

Methods

Ethics statement

All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of Jilin University (DY202605004).

Animal and treatments

The ruminal dysbiosis–induced mastitis model was established as previously described in ref. ⁶. Twenty dairy goats (2–3 years old; mean body weight, ~45 kg) were randomly assigned to two groups: control (Ctr) and RDIM (n = 10 goats per group). Group allocation was performed using a computer-generated randomisation sequence. None of the goats had received antibiotics or other medications within the preceding six months, and all animals were free of pre-existing diseases. Goats in the Ctr group were provided a standard dietline based on grass–legume hay (40% concentrate, Suppl. Table 1), whereas those in the RDIM group were fed a high-grain diet (approximately 70% grain and 30% forage, Suppl. Table 1). Treatment and measurement order were randomised, and animals/cages were evenly distributed with regular position rotation to control potential confounding factors. Investigators were blinded to group allocation during data collection and analysis. Blinding was not feasible during animal treatment, as the administering personnel necessarily required knowledge of each goat’s assigned intervention to ensure accurate delivery of the RDIM or control procedure. After 8 weeks of treatment, milk samples were collected from both groups by aseptic milk collection, and somatic cell count (SCC) more than 2.0 × 10⁶ cells/mL was used for preliminary mastitis assessment^4,6. Rumen fluid was collected by rumenocentesis using nonpyrogenic needles (2.0 × 120 mm) with 50 mL syringes, and serum samples were obtained from the tail vein. All samples were stored at −80 °C until analysis. At the end of the trial, mammary tissue samples were collected from three randomly selected goats per group for single-cell RNA sequencing.

Single-cell suspension preparation

Single-cell suspension of the mammary gland was prepared as previously described in refs. ^23,46. Briefly, mammary tissues were collected and processed immediately. Each sample was finely minced into fragments (<1 mm) and digested in DMEM (Thermo Fisher Scientific) containing 2% FBS (Thermo Fisher Scientific), 1 mg/mL collagenase II (Sigma-Aldrich), and 100 μL DNase I (Sigma-Aldrich) for 1 h at 37 °C with shaking. The resulting suspension was then diluted by adding 4 mL DMEM and passed through a 70 μm cell strainer. Cells were pelleted by centrifugation at 300×g for 10 min at 4 °C, after which the supernatant was removed and the pellet was washed twice with PBS. The cells were subsequently resuspended in 10 mL ice-cold red blood cell lysis buffer and incubated at 4 °C for 15 min. Next, 10 mL ice-cold PBS was added, followed by centrifugation at 250 × g for 10 min at 4 °C. After discarding the supernatant, the pellet was resuspended in 5 mL calcium- and magnesium-free PBS supplemented with 0.04% (w/v) BSA. The final single-cell suspension was adjusted to 700–1200 cells/μL with > 85% viability, as assessed using a Countess II Automated Cell Counter.

GEM generation and barcoding

Cell suspensions were processed on the 10× Genomics GemCode single-cell system to generate single-cell Gel Bead-in-Emulsion (GEM) droplets. cDNA libraries were prepared and sequenced using the Chromium Next GEM Single Cell 3’Reagent Kits v3.1. Following gel-bead dissolution within each GEM, primers containing (i) the Illumina® R1 sequence (read 1 sequencing primer), (ii) a 16-nt 10× barcode, (iii) a 10-nt unique molecular identifier (UMI), and (iv) a poly(dT) sequence were released and combined with the cell lysate and master mix. Subsequently, barcoded full-length cDNAs were synthesized by reverse transcription from polyadenylated mRNAs.

GEM-RT clean up and cDNA amplification

Silane magnetic beads were employed to remove residual biochemical reagents and primers from the post-GEM reaction mixture. The resulting full-length, barcoded cDNAs were subsequently PCR-amplified to obtain sufficient material for library construction.

Single-cell library preparation and sequencing

During GEM incubation, the R1 (read 1 primer) sequence was incorporated into the molecules. During library construction, P5, P7, the sample index, and the R2 (read 2 primer) sequence were introduced through end repair, A-tailing, adaptor ligation, and PCR. The resulting libraries contained P5 and P7 sequences required for Illumina bridge amplification, and the Single Cell 3′ protocol generated Illumina-ready sequencing libraries. Each Single Cell 3’ library followed a standard Illumina paired-end architecture flanked by P5 and P7. The 16-bp 10× barcode and 10-bp UMI were read in Read 1, whereas Read 2 was used to sequence the cDNA insert. Sample index sequences were captured as the i7 index read. Read 1 and Read 2 corresponded to the standard Illumina® primer sites for paired-end sequencing.

Preprocessing of scRNA-seq data

10× Genomics Cell Ranger software (version 3.1.0) was utilized to process raw BCL files into FASTQ files, perform alignment, and quantify counts. Low-quality reads with unreliable barcode or UMI sequences were filtered out prior to alignment to the reference genome. Only reads that uniquely mapped to the transcriptome and overlapped an exon by ≥ 50% were retained for UMI counting. Before quantification, sequencing errors in UMI sequences were corrected, and valid barcodes were determined using the EmptyDrops approach⁴⁷. Cell-by-gene matrices were subsequently produced through UMI counting and cell barcode calling.

Determination of cell type

Cell-by-gene matrices from each sample were imported separately into Seurat (v3.1.1) for downstream analyses. Cells with excessively high UMI counts (≥ 8000) or a high mitochondrial gene fraction (≥10%) were removed. Cells with fewer than 500 or more than 4000 detected genes were also excluded. Doublet GEMs were identified and filtered using DoubletFinder (v2.0.3), which simulates artificial doublets and computes the proportion of artificial k-nearest neighbors (pANN) based on PC distance, ranking cells according to the expected doublet rate. Seurat clustering was performed using a graph-based strategy: cell–cell distances were calculated in the space of selected principal components, and cells were embedded into a shared nearest neighbor (SNN) graph by connecting transcriptionally similar cells. The SNN graph was constructed using Euclidean distances in PCA space, and edge weights were refined using Jaccard similarity to quantify overlap in local neighborhoods. Communities were then identified with the Louvain algorithm to maximize modularity. For cell-type annotation, log-normalized expression matrices were analyzed with SingleR, which correlates reference cell-type expression profiles with single-cell data. Spearman correlations were first computed between each cell and each reference sample, and correlations were then aggregated to obtain a per-cell score for each candidate cell type⁴⁸.

Trajectory analysis by Monocle 2 analysis

DEGs among epithelial cell subclusters were identified and analyzed using Monocle. Dimensionality reduction and trajectory visualization were then performed with the “DDRTree” and “plot_cell_trajectory” functions in Monocle 2²³.

Metagenomic sequencing

Ruminal fluid was collected, and total genomic DNA was extracted using the FastPure Stool DNA Isolation Kit (Magnetic Bead) according to the manufacturer’s protocol. DNA concentration and purity were measured with the Synergy HTX and NanoDrop 2000, respectively, and DNA integrity was evaluated on a 1% agarose gel. Genomic DNA was then sheared to an average fragment length of ~350 bp using a Covaris M220 instrument (Gene Company Limited, China) for paired-end library preparation. Paired-end sequencing was performed on an Illumina NovaSeq™ X Plus system (Illumina Inc., San Diego, CA, USA) at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China) following the NovaSeq X Series 25B Reagent Kit instructions. Raw reads were quality-controlled with fastp (v0.20.0) by trimming adapters and discarding low-quality reads (length < 50 bp or mean quality score < 20). Contigs ≥ 300 bp were retained for final assembly. ORFs were predicted from assembled contigs using Prodigal (v2.6.3), and ORFs ≥ 100 bp were kept for downstream analyses. Taxonomic assignment of non-redundant genes was determined by best-hit alignment against the NCBI NR database using DIAMOND (v2.0.13) with an e-value threshold of 1e − 5. Functional annotation of non-redundant genes was performed against GO, KEGG, eggNOG, CAZy, CARD, and PHI databases. Based on taxonomic/functional annotations and the abundance profiles of non-redundant genes, principal coordinate analysis (PCoA) and LEfSe were applied to characterize ruminal microbial community structure and identify differentially abundant taxa.

Untargeted metabolomics

Rumen fluid and serum (100 μL) were extracted with 400 μL of prechilled methanol via vigorous vortex mixing, followed by 5 min incubation on ice. Samples were centrifuged at 15,000 × g for 5 min at 4 °C, and the supernatants were diluted with LC–MS-grade water to achieve a final methanol concentration of 53% (v/v). Aliquots were transferred to fresh low-binding Eppendorf tubes and centrifuged again at 15,000 × g for 10 min at 4 °C to remove residual particulates; the clarified supernatants were then injected into the UHPLC–MS/MS system. Raw data were processed using Compound Discoverer 3.1 (Thermo Fisher Scientific) for peak alignment, detection, and relative quantification. To monitor analytical reproducibility, pooled quality control (QC) samples were interspersed throughout the batch (every 10 injections); retention time stability and signal intensity consistency across QCs were assessed. Metabolite features exhibiting a relative standard deviation (RSD) > 30% in QC samples were excluded from downstream analysis. Further data curation included retention time correction, peak identification (against mzCloud and MassList databases), missing value imputation (using the minimum observed value per feature), and probabilistic quotient normalization (PQN). Principal component analysis (PCA) was performed to confirm tight clustering of QC samples and to identify and exclude technical outliers. Metabolite annotation was conducted by matching accurate mass (± 5 ppm), retention time (± 0.2 min), and MS/MS spectra against the KEGG, HMDB, and METLIN databases. Statistical significance was evaluated using two-tailed Student’s t-tests (P < 0.05); metabolites meeting all three criteria—variable importance in projection (VIP) score > 1, P < 0.05, and absolute log2-fold change ≥ 1 (i.e., FC ≥ 2 or ≤ 0.5)—were designated as differentially abundant. KEGG-based pathway enrichment and functional overrepresentation analyses were subsequently performed on this metabolite subset.

Detection of somatic cell counts

SCC was quantified using the LACTOSCAN MCC COMBO, a fully automated fluorescence-based milk cell counter, following the manufacturer’s instructions. Briefly, 100 μL of fresh milk was vortexed with 100 μL of SOFIA staining reagent (provided in the kit) under complete darkness; 8 μL of the resulting mixture was then loaded onto the sample well of the detection card, incubated for 1 min at room temperature, and immediately analyzed on the instrument.

Histological analysis

Mammary tissues were collected for histological evaluation and immediately fixed in 4% paraformaldehyde. After dehydration and paraffin embedding, 5-μm sections were prepared and stained with hematoxylin and eosin (H&E). Histopathological changes were examined using an optical microscope (Olympus, Tokyo, Japan)^3,4. The histological score of the mammary gland was assessed using a semi-quantitative scoring system adapted from ref. 5, evaluating two key parameters: (1) acinar structural integrity—scored as 0 (intact), 1 (mild disruption), 2 (moderate disruption), or 3 (severe disruption); and (2) inflammatory cell infiltration—scored as 0 (absent), 1 (mild), 2 (moderate), or 3 (severe).

RNA extraction and qPCR

Mammary tissues were harvested from goats in the Ctr and RDIM groups, and total RNA was isolated with TRIzol reagent (Invitrogen, Carlsbad, CA, USA) following the manufacturer’s instructions⁴. Genomic DNA contamination was eliminated by RNase-free DNase I treatment during RNA isolation. First-strand cDNA was synthesized from the extracted RNA using TransStart Tip Green qPCR SuperMix (TransGen Biotech, Beijing, China). Quantitative PCR was performed with gene-specific primers and FastStart Universal SYBR Green Master Mix (ROX) (Roche, Basel, Switzerland) on a Step One Plus Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). Primer sequences were as follows¹: TNF-α (sense: 5′- CCCTCACACTCAGATCATCTTCT-3′; antisense: 5′- GCTACGACGTGGGCTACAG-3′), IL-1β (sense: 5′- GCAACTGTTCCTGAACTCAACT-3′; antisense: 5′-ATCTTTTGGGGTCCGTCAACT-3′), and GAPDH (sense: 5′-AACTTTGGCATTGTGGAAGG-3′; antisense: 5′-ACACATTGGGGGTAGGAACA-3′). The expression stability of GAPDH was validated prior to its use as an endogenous reference gene for RT-qPCR normalization. Relative transcript levels were calculated using the 2^-ΔΔCt method, with the control group as the calibrator.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 8.0 (San Diego, CA, USA). No statistical methods were used to predetermine sample size for the animal experiments; however, the sample sizes employed in this study are consistent with those reported in prior works^4,40. No animals were excluded during the experiment. Data are presented as a violin plot or as boxplots, where the center line represents the median, the box spans the interquartile range (25th to 75th percentiles), and the whiskers extend to the minimum and maximum values. Data normality was checked via Shapiro-Wilk test. If parametric assumptions were not satisfied, corresponding non-parametric tests were used instead. Differences between two groups were assessed using the Mann–Whitney U test (nonparametric) or a two-tailed unpaired Student’s t-test (parametric), as appropriate. P < 0.05 was considered statistically significant. All experiments were performed at least three independent experiments, with similar results across replicates.

Author contributions

Caijun Zhao, Yun Zhang, Fanglei Han, and Yunhe Fu designed the study. Caijun Zhao, Zhuohan Li, and Miao Liu performed animal experiments and all statistical analysis. Lijuan Bao, Chongshan Yuan, Yihong Zhao, Min Qiu, Yuhong He, Naisheng Zhang, Xiaoyu Hu, and Keyi Wu assisted with histological analysis and experimental parameter determinations. Caijun Zhao, Yunhe Fu, and Xiaoyu Hu obtained fundings. Caijun Zhao wrote the manuscript, and all authors revised and approved the manuscript.

Data availability

All data used in this study are present in the article and supplementary materials. The raw sequencing data generated in this study have been deposited in the Genome Sequence Archive (GSA; Genomics, Proteomics & Bioinformatics 2025) at the National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession GSA: CRA034123. The dataset is publicly accessible at https://ngdc.cncb.ac.cn/gsa.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41522-026-01076-7.