Multi-faceted characterization of the gut microbiome and metabolome in patients with primary Sjögren syndrome

Summary

The gut microbiome and its metabolomic potential in primary Sjögren syndrome (pSS) remain largely unexplored. Here, we perform whole-metagenome shotgun sequencing of fecal samples from 206 pSS patients and 355 non-pSS controls, integrating compositional and functional profiling with serum and fecal metabolomes. pSS is associated with extensive multi-kingdom alterations, including 49 bacterial (e.g., Streptococcus parasanguinis, Ligilactobacillus salivarius, and Veillonella parvula), 19 fungal (notably Candida albicans), and 1,323 viral species. These signatures form robust inter-kingdom correlations and achieve high diagnostic accuracy in an independent validation cohort. Functional and metabolomic analyses reveal enrichment of toxin-related and aromatic pathways and depletion of protective metabolites in patients. pSS-enriched bacteria harbor abundant immunogenic epitopes, virulence factors, and antimicrobial resistance genes, and induce proinflammatory responses ex vivo. Together, these findings outline a multi-faceted microbial framework for pSS and suggest mechanistic links between gut dysbiosis and immune dysregulation.

Graphical abstract

Highlights

• •Primary Sjögren syndrome shows coordinated bacterial, fungal, and viral alterations
• •Integrated multi-kingdom signatures achieve diagnostic performance across cohorts
• •pSS gut dysbiosis is associated with harmful metabolic remodeling and SCFA depletion
• •pSS-enriched bacteria encode virulence traits and trigger inflammatory PBMC responses

Chen et al. map the multi-kingdom gut microbiome in primary Sjögren syndrome and identify diagnostic microbial signatures, metabolic disruption, and proinflammatory species with potential therapeutic relevance.

Introduction

Primary Sjögren syndrome (pSS) is a chronic systemic autoimmune disorder characterized by inflammation and destruction of exocrine glands, clinically manifesting as dry eyes and dry mouth (keratoconjunctivitis sicca and xerostomia).¹ The pathogenesis of pSS involves aberrant immune responses against self-antigens, resulting in epithelial cell damage and impaired secretion from salivary and lacrimal glands. Beyond glandular involvement, pSS patients may develop a broad spectrum of systemic manifestations, including arthritis, interstitial lung disease (ILD), renal involvement, neurological complications, and an increased risk of lymphoma. Although the concept of autoimmune epitheliitis has been proposed to explain the pathogenesis of pSS, the precise underlying mechanisms remain incomplete. Both genetic predisposition and environmental factors are implicated in disease onset and progression.^2,3

In recent years, the skin and mucosal microbiota have emerged as critical contributors to autoimmune pathogenesis.⁴ The gut microbiome, comprising bacteria, viruses, and fungi, harbors a gene repertoire vastly exceeding that of the host and plays a pivotal role in modulating the digestive, immune, and nervous system functions.^5,6 Accumulating evidence from metagenomic shotgun sequencing has linked gut microbial dysbiosis to multiple immune-related disorders, including inflammatory bowel disease (IBD),⁷ rheumatoid arthritis (RA),⁸ and systemic lupus erythematosus (SLE).⁹ Emerging studies also implicate gut dysbiosis in pSS, with changes in microbial diversity correlating to disease severity.¹⁰ Three principal mechanisms have been proposed to explain microbiota-driven autoimmunity: breakdown of epithelial barrier integrity leading to bacterial translocation and immune activation¹¹; immunomodulation by microbial metabolites such as short-chain fatty acids (SCFAs)¹²; and molecular mimicry, whereby microbial antigens trigger cross-reactive immune responses against host tissues.¹³ However, most studies focus predominantly on bacteria, overlooking the potential contributions of viruses and fungi and the complex inter-kingdom interactions within the gut ecosystem. We hypothesize that a multi-faceted approach incorporating bacteria, viruses, and fungi will provide deeper insights into the gut microbiota’s role in pSS and elucidate its multifaceted interactions with host immunity.

To date, research on the gut microbiota in pSS has largely relied on 16S rRNA gene sequencing,^14,15,16 with only one small metagenomic study (78 samples) reported.¹⁵ In the present study, we analyzed fecal samples from 206 pSS patients and 355 non-pSS controls, comprehensively characterizing the gut bacteriome, mycobiome, and virome, along with their functional potential and serum and fecal metabolomic profiles. We further evaluated the impact of clinical characteristics, including medication use and the presence of ILD, on the gut microbiome composition and function. In vitro peripheral blood mononuclear cell (PBMC) stimulation assays were conducted to assess the immunostimulatory effects of key microbial taxa. By integrating multi-kingdom microbiome data with metabolic profiling and functional validation, this study aims to provide a holistic view of the gut ecosystem in pSS, identify robust disease-specific microbial signatures, and uncover potential links between gut dysbiosis, host immunity, and clinical phenotypes.

Results

Study population and basic characteristics

To investigate the gut microbiome and metabolome in pSS, we collected fecal samples from 206 pSS patients and 355 non-pSS controls. Analysis of the study population revealed that pSS patients were slightly but significantly older than non-pSS controls (mean age 57.5 vs. 53.0 years, p < 0.001) and exhibited a markedly higher proportion of females (89.4% vs. 44.3%, p < 0.001; Table S1). Among pSS patients, 68.9% had a history of medication use, including drugs such as prednisone, leflunomide, hydroxychloroquine, and mycophenolate mofetil. Additionally, 16.0% of pSS patients were diagnosed with ILD, further underscoring the high prevalence of this comorbidity in the pSS population.

Diversity and composition of gut multi-kingdom microbiome in pSS patients

We performed whole-metagenome shotgun sequencing on fecal DNA from 206 pSS patients and 355 non-pSS subjects. After quality control and host read removal, high-quality sequencing data were profiled using MetaPhlAn4,¹⁷ a cultivated gut fungal genome catalog, and the cnGVC database to characterize gut microbial composition across multiple kingdoms.

Diversity and composition of the gut bacteriome

Rarefaction analysis revealed slightly higher bacterial richness in non-pSS subjects compared to pSS patients with increasing sample size (Figure 1A). However, neither observed species richness nor Shannon diversity differed significantly between groups (Figures 1B and 1C), suggesting that pSS has limited impact on gut bacterial alpha diversity. After adjusting for medication use, principal coordinates analysis (PCoA) based on Bray-Curtis distances still revealed significant separation between pSS patients and non-pSS subjects (PERMANOVA p = 0.001; Figure 1D), indicating that compositional shifts in the gut bacteriome persist independently of medication and are associated with pSS.

Figure 1.

Gut bacteriome diversity and compositional differences between pSS patients and non-pSS controls

(A) Rarefaction curves depicting the cumulative number of observed bacterial species as a function of sequencing depth in pSS (orange) and non-pSS (green) groups.

(B and C) Boxplot comparing the number of observed bacterial species (B) and the Shannon diversity index (C) between pSS and non-pSS groups.

(D) Principal coordinates analysis (PCoA) based on Bray-Curtis distances at the species level, after adjusting for medication effects. Ellipsoids represent 95% confidence intervals for each group, with upper and right boxplots showing sample score distribution along PCoA1 and PCoA2, respectively. Statistical significance was determined using the Wilcoxon rank-sum test with Benjamini and Hochberg adjustment. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

(E) Relative abundance of the top eight bacterial phyla across all samples.

(F) Boxplots illustrating significantly altered genera between groups (fold change >1.2, LDA score >2, relative abundance >0.01).

At the phylum level, Firmicutes, Bacteroidetes, Actinobacteria, and Proteobacteria dominated both groups (Figure 1E). Proteobacteria abundance was significantly higher in pSS patients (10.6% vs. 3.4%, Wilcoxon rank-sum test, q < 0.05), whereas Bacteroidetes was lower (14.5% vs. 27.3%, q < 0.05; Table S2). At the genus level, pSS patients exhibited elevated abundances of Escherichia, Streptococcus, Veillonella, Ligilactobacillus, and Limosilactobacillus, along with 17 additional genera, while Faecalibacterium, Phocaeicola, Bacteroidota, Roseburia, and Anaerobutyricum, together with 11 other genera, were enriched in non-pSS controls (Figure 1F; Table S3).

At the species level, 49 taxa differed significantly between groups (Wilcoxon rank-sum test, q < 0.05, linear discriminant analysis score [LDA] score >2), with 19 species enriched in pSS patients and 30 enriched in controls (Figure 1H; Table S4). Species from the orders Veillonellales (e.g., Veillonella parvula), Lactobacillales (e.g., Lactobacillus mucosae, Ligilactobacillus salivarius, and Streptococcus parasanguinis), and Enterobacterales (e.g., Escherichia coli, Klebsiella pneumoniae, and Citrobacter freundii) were significantly increased in pSS patients. Conversely, some Bacteroidetes members such as Prevotella copri clade A, Phocaeicola vulgatus, Bacteroidota uniformis, Anaerostipes hadrus, and Phocaeicola plebeius were notably reduced in pSS patients. After adjusting for gender, age, body mass index (BMI), and medication use, the number of significant species modestly decreased, but most differential taxa, including S. parasanguinis, L. salivarius, and V. parvula, remained robustly associated with pSS.

Diversity and composition of the gut mycobiome

Rarefaction analysis revealed significantly higher pooled fungal richness in non-pSS subjects compared to pSS patients (Figure 2A). Conversely, between-sample dispersion of observed species was lower in non-pSS individuals, indicating a more homogeneous fungal community structure than in pSS (Wilcoxon rank-sum test, p = 0.001; Figure 2B). Shannon diversity index also differed significantly between groups (Figure 2C). After adjusting for medication use, Bray-Curtis-based PCoA demonstrated distinct clustering of gut mycobiota between pSS patients and non-pSS controls (PERMANOVA, p = 0.001; Figure 2D), mirroring the dysbiosis observed in bacterial communities.

Figure 2.

Gut mycobiome diversity and compositional differences between pSS patients and non-pSS controls

(A) Rarefaction curves depicting the cumulative number of observed fungal species as a function of sequencing depth in pSS (orange) and non-pSS (green) groups.

(B and C) Boxplot comparing the number of observed fungal species (B) and the Shannon diversity index (C) between pSS and non-pSS groups. Statistical significance was determined using the Wilcoxon rank-sum test with Benjamini and Hochberg adjustment. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

(D) Principal coordinates analysis (PCoA) based on Bray-Curtis distances at the fungal species level, after adjusting for medication effects. Ellipsoids represent 95% confidence intervals for each group, with upper and right boxplots showing sample score distributions along PCoA1 and PCoA2, respectively.

(E) Pie plot showing the phylum-level mycobiome composition for pSS patients and non-pSS controls.

(F) Relative abundance of the top 10 fungal genera across all samples.

(G) Boxplots showing significantly altered fungal genera between groups (fold change >1.2, relative abundance >0.01).

Compositionally, the gut mycobiome was predominantly composed of Saccharomycotina (yeasts), followed by Pezizomycotina (filamentous fungi), Basidiomycota, and Mucoromycota across all subjects (Figure 2E). At the genus level, Candida, Malassezia, and Orbilia were significantly enriched in pSS patients, whereas Pichia, Aspergillus, and Kodamaea were more abundant in controls (Figure 2F; Table S5). A total of 19 fungal species met criteria for differential abundance (fold change >1.2, relative abundance ≥0.1% in at least one group, occurrence frequency ≥10%, q < 0.05), with the majority enriched in pSS patients (Table S6). Specifically, species from Penicillium (e.g., Pichia Sumatraense and Pichia glabrum), Candida (e.g., C. albicans), and Malassezia (e.g., Malassezia arunalokei and Malassezia restricta) were more abundant in pSS patients. In contrast, only three species, including Kodamaea ohmeri, Rhizopus stolonifer, and Aspergillus sp. c38, were significantly depleted in patients, suggesting their potential as fungal biomarkers for the disease. After adjusting for potential confounders, most of these fungal taxa remained significantly associated with pSS. These results suggest that alterations in fungal composition, particularly enrichment of Candida and Malassezia species, may represent microbial signatures linked to pSS.

Diversity and composition of the gut virome

A total of 62,816 viral operational taxonomic units (vOTUs) were quantified across 561 metagenomic samples, of which 62.7% (39,388/62,816) were assigned to 33 known viral families for downstream analysis. Rarefaction analysis revealed slightly lower viral richness in pSS patients compared to non-pSS subjects (Figure 3A), a trend corroborated by observed species counts (Figure 3B). Intragroup Shannon diversity was also significantly reduced in pSS patients (Figure 3C). After medication adjustment, Bray-Curtis-based PCoA showed clear separation between pSS and non-pSS groups (PERMANOVA, p = 0.006; Figure 3D), underscoring persistent alterations in gut viral community structure independent of medication effects.

Figure 3.

Gut virome diversity and compositional differences between pSS patients and non-pSS controls

(A) Rarefaction curves depicting the cumulative number of observed viral species as a function of sequencing depth in pSS (orange) and non-pSS (green) groups.

(B and C) Boxplot comparing the number of observed vOTUs (B) and the Shannon diversity index (C) between pSS and non-pSS groups.

(D) PCoA based on Bray-Curtis distances at the vOTU level, after adjusting for medication effects. Ellipsoids represent 95% confidence intervals for each group, with upper and right boxplots displaying sample score distributions along PCoA1 and PCoA2, respectively. For (B–D), statistical significance was assessed using the Wilcoxon rank-sum test with Benjamini and Hochberg adjustment. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

(E) Relative abundance of the top 10 viral families across all samples.

(F and G) Stacked bar plots showing the distribution of detected viral species by their predicted bacterial hosts. Each bar is color-coded according to the host bacterial taxon.

We next examined the composition of the gut virome at the family level and identified Siphoviridae, Myoviridae, Podoviridae_crAss-like, and Quimbyviridae as dominant families (Figure 3E). Notably, Myoviridae and Podoviridae were enriched in pSS patients, whereas Quimbyviridae was significantly higher in non-pSS subjects (q < 0.05; Table S7). At the vOTU level, 1,323 vOTUs exhibited significant changes in pSS patients, with 327 enriched and 951 depleted (q < 0.05; Table S8). After adjusting for potential confounding variables, the number of significant vOTUs decreased markedly, indicating that some viral associations were partially influenced by individual characteristics. Host prediction of vOTUs revealed that viruses enriched in pSS patients were primarily associated with bacterial genera including Escherichia, Streptococcus, Akkermansia, Klebsiella, and Megamonas (Figures 3F and 3G; Table S9). In contrast, depleted vOTUs were predominantly linked to Bacteroidota, Faecalibacterium, Prevotella, Parabacteroides, and Roseburia. These pronounced differences suggest that viruses may play important roles in the pathogenesis or progression of pSS.

Multi-kingdom microbial interactions and their diagnostic potential

Gut multi-kingdom microbial co-occurrence networks

Given the coexistence and potential interactions among bacteria, fungi, and viruses within the gut ecosystem, as well as the marked differences observed in each of these microbial components between patients and controls, we next investigated inter-kingdom interaction networks. The phylogenetic distribution and enrichment patterns of the differential abundant multi-kingdom signatures (i.e., 49 bacterial species, 19 fungal species, and 1,323 vOTUs) are summarized in a taxonomic tree (Figure 4A). To construct comprehensive interaction networks, we conducted correlation analysis on these signatures, retaining only robust associations (|ρ| > 0.6). This yielded networks comprising 1,160 and 1,101 edges for pSS patients and non-pSS controls, respectively (Figures 4B and 4C). In both networks, Faecalibacterium prausnitzii, B. uniformis, P. copri clade A, E. coli, and P. vulgatus exhibited high connectivity, suggesting their potential keystone roles in gut homeostasis (Figures 4D and 4E). We further identified 116 species whose interaction patterns differed markedly between the two networks (Table S10). Notably, Rothia mucilaginosa and Rhizopus stolonifer c238 displayed higher connectivity in the pSS network, potentially implicating them in disease pathogenesis.

Figure 4.

Multi-kingdom differential taxa and co-occurrence networks of the gut microbiome

(A) Phylogenetic tree illustrating significantly altered multi-kingdom taxa between pSS and non-pSS groups. Outer colored blocks represent taxonomic orders; red and blue triangles indicate taxa enriched or depleted in the pSS group, respectively.

(B and C) Co-occurrence networks of bacterial, fungal, and viral taxa in the non-pSS (B) and pSS (C) groups. Green circulars, orange squares, and blue triangulars represent bacteria, fungi, and viruses, respectively. Edges indicate significant correlations (|ρ| > 0.6).

(D and E) Bar plots showing the top 20 taxa with the highest connectivity (degree) in non-pSS (D) and pSS (E) groups.

Classification of disease status based on multi-kingdom signatures

To assess whether gut microbial profiles could discriminate pSS patients from controls, we trained random forest classifiers using bacterial, fungal, and viral features separately. Models based on 29 bacterial, 10 fungal, and 64 viral features achieved optimal areas under the curve (AUCs) of 0.841, 0.849, and 0.943, respectively (Figure 5A). Interestingly, a combined model integrating all three kingdoms outperformed individual classifiers, attaining a cross-validated AUC of 0.950 (95% confidence interval [CI]: 0.945–0.955). Top-ranking features included pSS-enriched bacteria such as S. parasanguinis, Clostridium innocuum, Ruthenibacterium lactatiformans, L. salivarius, and Streptococcus anginosus, as well as several vOTUs (Figures 5B–5E). To validate the robustness and generalizability of these signatures, we recruited an independent cohort from a hospital in western China (56 pSS patients and 58 healthy controls), geographically distinct from the northeastern discovery cohort. Consistent classification performance was observed, with AUCs of 0.711 (bacteria), 0.659 (fungi), 0.730 (viruses), and 0.742 (combined model) (Figure 5F). Several taxa, including Veillonella, Limosilactobacillus, Ligilactobacillus, Faecalibacterium, Escherichia, and Candida, exhibited reproducible enrichment trends in pSS patients across both cohorts (Figures 5G–5J). Notably, pSS-enriched species such as V. parvula, S. parasanguinis, and L. salivarius showed consistent direction changes in the validation cohort.

Figure 5.

Classification of pSS status by the compositions of gut multi-kingdom signatures

(A) Receiver operating characteristic (ROC) curves for pSS classification using bacterial, fungal, and viral signatures individually and combined multi-kingdom signatures.

(B) External validation in an independent cohort from different regions, showing ROC curves for bacterial, fungal, viral, and combined multi-kingdom models.

(C–F) Top-ranked microbial features contributing to classification models for bacteria (C), fungi (D), viruses (E), and multi-kingdom combined (F). Bar lengths indicate feature importance, with orange and green bars representing features enriched in pSS and non-pSS groups, respectively.

(G) PCoA of the validation cohort, demonstrating distinct clustering between pSS patients and non-pSS controls.

(H–J) Relative abundance of key bacterial genera (H), bacterial species (I), and fungal genera (J) previously identified as significant in the discovery cohort, validated in the independent cohort. Error bars represent mean ± standard error.

Gut functional profiling and serum and fecal metabolomes in pSS patients

Functional characteristics of gut microbiota

To characterize the functional attributes of the gut microbiota in pSS, we applied the UHGP pipeline to generate functional profiles from all fecal metagenomes, yielding 9,041 KOs and 765 KEGG modules for subsequent analysis. Functional diversity analysis revealed that pSS patients exhibited lower module richness but higher Shannon index compared to controls (Figures 6A and 6B), indicating a more even distribution of function capacities across samples. Consistent with taxonomic shifts, PCoA of module composition showed clean separation between groups (PERMANOVA p = 0.001; Figure 6C), underscoring substantial functional remodeling of the gut microbiome in pSS. Among the 765 modules, 120 differed significantly between groups, with 115 enriched in pSS and only 5 enriched in non-pSS subjects (q < 0.001; Figures 6D and 6E; Table S11). pSS-enriched modules were predominantly involved in aromatic compound degradation (e.g., benzoate, toluene, xylene, anthranilate, and tyrosine), polyamine biosynthesis, bacterial and fungal secondary metabolite biosynthesis, pathogenicity, antimicrobial resistance, and two-component regulatory systems. Representative examples include the SaeS-SaeR staphylococcal virulence regulation system, manganese/zinc/iron transport system, and multiple drug resistance modules. These findings indicate that the pSS gut microbiome is functionally reprogrammed toward stress response, virulence potential, and xenobiotic metabolism.

Figure 6.

Functional and metabolomic alterations in pSS patients

(A and B) Boxplots comparing the richness (A) and Shannon diversity index (B) of functional modules between pSS and non-pSS groups.

(C) PCoA based on Bray-Curtis distances showing overall differences in functional module composition between groups. For (A–C), statistical significance was assessed using the Wilcoxon rank-sum test with Benjamini and Hochberg adjustment. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

(D) Volcano plot highlighting differentially abundant functional modules between pSS and non-pSS groups.

(E) Bar chart illustrating significantly altered functional modules, annotated with higher-order functional classifications and enrichment trends.

(F and G) Network diagrams depicting associations between differentially abundance bacterial/fungal taxa and functional modules.

(H and I) Bar plots showing differentially abundant metabolites in fecal (H) and serum (I) identified by untargeted metabolomic analysis.

(J) Correlation network linking differentially abundant bacterial and fungal taxa with altered fecal metabolites.

Serum and fecal metabolomic alterations in pSS patients

To experimentally assess the functional consequences of gut microbial dysbiosis, we conducted untargeted metabolomic profiling of paired serum and fecal samples from 75 pSS patients and 66 non-pSS controls. A total of 18 serum metabolites and 7 fecal metabolites showed significant differences between groups (Figures 6H and 6I; Tables S12 and S13). In serum, pSS patients exhibited elevated levels of metabolites related to oxidative stress and immune activation, including creatinine, indoxyl sulfate, trimethylamine N-oxide (TMAO), hippuric acid, kynurenine, tryptophan, and uric acid—consistent with previously reported metabolic alterations in autoimmune conditions.¹⁸ In feces, concentrations of the SCFAs butyrate and propionate were markedly reduced in pSS patients, mirroring trends observed in other inflammatory diseases. Additionally, indole—a microbial tryptophan metabolite and aryl hydrocarbon receptor (AhR) ligand with immunoregulatory properties—was significantly depleted, suggesting impaired AhR-medicated mucosal immune modulation in the pSS gut environment. Collectively, these metabolomic alterations point to a systemic shift toward pro-inflammatory and oxidative metabolic states, alongside loss of protective microbial metabolites, likely reflecting functional consequences of gut dysbiosis and its impact on host immune homeostasis.

Functional network analysis linking microbial taxa

To explore functional connections between altered microbial taxa and their functional metabolic potential, we constructed correlation networks integrating differential bacterial species and KEGG modules for pSS and non-pSS groups. Given that modules were inferred from the same metagenomic data, associations with bacterial taxa were expected to be strong; we therefore applied a stringent threshold (Spearman’s ρ > 0.6, q < 0.05) to retain only the most robust interactions (Figures 6F and 6G; Table S14). In the pSS network, E. coli, K. pneumoniae, S. parasanguinis, and Streptococcus salivarius exhibited the highest connectivity, suggesting central roles in the dysbiotic ecosystem. Notably, these species were strongly associated with pSS-enriched modules involved in virulence regulation (e.g., SaeS-SaeR two-component system), metal iron transport, and stress adaption—functions that may enhance bacterial persistence under inflammatory conditions. In contrast, the non-pSS network featured tight associations between commensal taxa such as F. prausnitzii, Alistipes putredinis, and Odoribacter splanchnicus and modules related to SCFA biosynthesis and amino acid metabolism, consistent with a homeostasis metabolic profile.

Extending this analysis to cross-kingdom associations among bacteria, fungi, and serum/fecal metabolites, we applied a more permissive threshold (q < 0.05) to capture biologically meaningful connections given the inherently weaker inter-domain relationships (Figure 6J; Table S15). The resulting integrative network revealed distinct pSS-associated patterns. Several pSS-enriched bacteria, including S. anginosus, L. salivarius, Bifidobacterium dentium, S. parasanguinis, E. coli, and V. parvula, showed significant negative correlations with SCFAs and positive correlations with pro-inflammatory or toxic metabolites (e.g., TMAO, indoxyl sulfate, uric acid, and kynurenine). Conversely, putatively beneficial taxa such as Bifidobacterium animalis and Faecalibacillus intestinalis displayed opposite trends, positively correlating with SCFAs and negatively with proinflammatory metabolites. Most fungal taxa, particularly the three pSS-enriched Candida albicans subspecies, exhibited strong positive correlations with deleterious metabolites, suggesting that fungal dysbiosis may further exacerbate systemic inflammation through metabolic crosstalk. Collectively, these results reveal complex, multi-kingdom metabolic reprogramming in pSS, characterized by cooperative pathogenic networks linking dysbiotic bacteria, fungi, and pro-inflammatory metabolites that may contribute to disease.

Functional characterization of gut species based on MAGs

To further explore the functional potential of pSS-enriched microbes, we reconstructed high-quality metagenome-assembled genomes (MAGs) for three key species (S. parasanguinis, L. salivarius, and V. parvula), yielding 13, 15, and 10 MAGs for each species. Protein-coding genes were annotated against multiple databases, including B cell and T cell epitopes, VFDB, DRG (drug resistance gene database), and antimicrobial peptide prediction tools. Our analysis revealed several notably findings. First, both B cell and T cell epitope prediction showed that all species harbor numerous immunogenic proteins (Figure 7; Tables S16 and S17). Conserved stress-related and metabolic housekeeping proteins such as BiP/GRP78, HSP60/HSP70, alpha/beta-enolase, phosphoglycerate mutase, and pyruvate kinase were consistently identified as antigenic sources across species. Each species also displayed distinct antigenic profiles. S. parasanguinis encoded epitopes derived from CTP synthase, aminopeptidase N, and various transcription-related proteins. V. parvula carried epitopes resembling collagen alpha-chains and thyroid-stimulating hormone receptor. L. salivarius contained multiple epitopes homologous to host mitochondrial and metabolic proteins. Some epitopes shared high sequence similarity to human proteins such as collagen, PKM, IMPDH, and CTPS, raising the possibility that these microbial antigens could trigger autoimmune responses through molecular mimicry.

Figure 7.

Genomic features of three key pSS-associated species

(A–C) Functional annotation of MAGs for Streptococcus parasanguinis (A), Ligilactobacillus salivarius (B), and Veillonella parvula (C). Blue bar charts represent the total number of genes in each species. Colored circles indicate counts of specific genomic features: red, predicted B cell epitopes; blue, predicted T cell epitopes; green, antimicrobial resistance genes; purple, virulence factor genes; orange, antimicrobial peptides.

Second, comprehensive virulence factor annotation revealed distinct pathogenic trait profiles among the three species (Figure 7; Table S18). S. parasanguinis and L. salivarius possessed the highest diversity of virulence-associated genes, L. salivarius encoded a broad range of elements linked to mucosal colonization and immune evasion, including sortase (srtC),¹⁹ capsular biosynthesis enzymes (cap8P, cpsA-I), and bile salt hydrolase (bsh). Chaperone proteins (DnaK, GroEL, and ClpC/P/E), glycosyltransferases, and two-component regulator systems (RegX3 and LisR) were also detected, suggesting enhanced adaptive and stress tolerance capacity. S. parasanguinis contained adhesion and immune-interactive factors, such as fibronectin-binding protein (fbp54), laminin-binding protein (lmb), and collagen-binding adhesin (cnm). Additionally, secretion-associated virulence determinants including type VII secretion protein EsxA, ZmpB metalloprotease, and Lap adhesion protein were identified, likely facilitating epithelial adherence and promoting inflammatory responses. Although V. parvula encoded fewer canonical virulence factors, it harbored stress and redox-related proteins (catalase katA and thioredoxin/methionine sulfoxide reductase msrA/B), adhesion proteins (Lap and LirB), and capsular polysaccharide biosynthesis clusters (cps4I, ugd, galU, and galE), which may support persistence and immune modulation under inflammatory conditions.²⁰

Third, metagenome analysis revealed widespread antimicrobial resistance determinants across all three species, encompassing both intrinsic and acquired resistance types (Figure 7A; Table S19). S. parasanguinis carried multidrug resistance genes conferring resistance to macrolides, lincosamides, rifampicin, tetracyclines, and quinolones. L. salivarius contained genes associated with resistance to aminoglycosides, macrolides, and tetracyclines, while V. parvula exhibited mainly tetracycline and natural resistance markers. In contrast, antimicrobial-peptide-related sequences were limited across all three species, indicating a relatively weak intrinsic capacity for producing antimicrobial compounds (Table S20).

Immune activation induced by pSS-associated gut microbes

Multi-kingdom gut microbiome analysis revealed that both bacterial (notably S. parasanguinis, L. salivarius, and V. parvula) and fungal (particularly C. albicans) members of the pSS-associated microbiota possess molecular features that may disrupt immune homeostasis. Based on these observations, we conducted in vitro experiments to assess their immunostimulatory potential using PBMCs. Exposure of PBMCs to fecal microbiota supernatant (FMS) from pSS patients led to marked induction of inflammatory cytokines, including tumor necrosis factor α (TNF-α), interleukin (IL)-6, IL-8, IL-17, and interferon gamma (IFN-γ), whereas FMS from non-pSS controls elicited no such response (Figure 8A). Untargeted metabolomic profiling of these FMS samples revealed significant enrichment of phenylacetaldehyde and hippuric acid—two metabolites that were also elevated in the feces and serum of pSS patients, respectively, compared to non-pSS controls (Figure 8B). We next tested four representative species: S. parasanguinis, V. parvula, L. salivarius, and Candida albicans. Metabolic profiling of culture supernatants from these species showed that all four could utilize aromatic amino acids and related substrates and each produced multiple pSS-enriched metabolites. For example, L. salivarius and S. parasanguinis generated indoxyl sulfate; C. albicans and L. salivarius produced biliverdin; and V. parvula generated kynurenine. PMBC stimulation assays revealed species-specific immunostimulatory profiles (Figure 8C). L. salivarius significantly upregulated TNF-α, IL-6, IL-8, and IL-17; V. parvula increased TNF-α, IL-6, and IL-8 expression; C. albicans strongly induced TNF-α, IL-6, and IFN-γ; and S. parasanguinis predominantly elevated TNF-α and IL-6. These findings demonstrate that both the dysbiotic gut microbiota as a whole and specific pSS-enriched taxa can trigger proinflammatory responses in human immune cells, providing early mechanistic evidence linking microbiome alterations to immune dysregulation in pSS.

Figure 8.

In vitro immune activation of PBMCs by gut microbiota and pSS-associated representative microbial species

(A) Expression levels of proinflammatory cytokines (TNF-α, IL-6, IL-8, IL-17, and IFN-γ) in PBMCs stimulated with fecal microbiota supernatants (FMSs) derived from pSS patients and non-pSS controls.

(B) Untargeted metabolomic profiling of FMSs and culture supernatants from four representative pSS-associated species. Left panel: bar plot showing the fold change in metabolite concentrations in FMS from pSS patients relative to non-pSS controls. Right four panels: comparison of metabolite levels in culture supernatants of the indicated microbial species versus sterile culture medium. Metabolites shown include those significantly elevated in pSS patient feces and/or serum, including phenylacetaldehyde, hippuric acid, indoxyl sulfate, biliverdin, and kynurenine.

(C) Cytokine responses of PBMCs exposed to microbial supernatants or particle fractions from Streptococcus parasanguinis, Veillonella parvula, Ligilactobacillus salivarius, and Candida albicans, as quantified by qPCR. NC, negative control; LPS, positive control by lipopolysaccharide; GAM, negative control by Gifu Anaerobic Medium. For (A) and (C), statistical significance was assessed using the Wilcoxon rank-sum test with Benjamini and Hochberg adjustment. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Impact of patient characteristics on the gut microbiota

Effects of medications on the gut microbiota of pSS patients

Immunosuppressants, antibiotics, and biologics have been shown to modulate the gut microbiota through distinct mechanisms, potentially affecting both its composition and function.^21,22 Having established significant compositional differences in the gut bacteriome between pSS patients and non-pSS controls, we next examined whether these differences are influenced by medication treatments. Our analysis revealed that medicated pSS patients (MP) exhibited significantly reduced gut bacterial diversity compared to non-medicated pSS patients (NMPs). Although observed species richness did not differ between groups, the Shannon index was notably lower in MP patients (Figures 9A and 9B), suggesting that medications may selectively deplete certain antibiotic-sensitive taxa, a pattern consistent with previous reports. However, PCoA showed no significant separation between MP and NMP groups (PERMANOVA, p = 0.291; Figure 9C), indicating that overall community composition remains relatively stable despite medication-associated diversity loss. Differential abundance analysis identified specific taxa altered in MP patients (Figure 9D). Ruminococcus sp. NSJ-71, B. dentium, Dysosmobacter welbionis, and Lacrimispora celerecrescens were significantly enriched in the MP group, whereas Ruminococcus bromii, Phocaeicola coprophilus, Eubacterium siraeum, Adlercreutzia equolifaciens, and Parasutterella excrementihominis were significantly depleted. These findings indicate that while medications do not broadly reshape the gut microbial community structure, they may exert selective pressure on specific bacterial populations, leading to targeted compositional shifts in medicated pSS patients.

Figure 9.

Impact of patient characteristics on the gut microbiota diversity and composition

(A and B) Boxplots comparing observed bacterial species richness (A) and Shannon diversity index (B) between medicated (MP) and non-medicated pSS patients (MNP).

(C) PCoA based on Bray-Curtis distances at the bacterial species level, showing distinct clustering between MP and MNP groups. Ellipsoids represent 95% confidence intervals; upper and right boxplots show sample score distributions along PCoA1 and PCoA2, respectively.

(D) Linear discriminant analysis (LDA) effect size (LEfSe) bar plot showing bacterial species with significant differential abundance between MP and NMP groups.

(E and F) Boxplots comparing observed bacterial species richness (E) and Shannon diversity index (F) between pSS patients with ILD (pSS-ILD) and those without (pSS-NO-ILD).

(G) PCoA based on Bray-Curtis distances at the bacterial species level, showing distinct clustering between pSS-ILD and pSS-NO-ILD groups.

(H) LEfSe bar plot showing bacterial species with significant differential abundance between pSS-ILD and pSS-NO-ILD patients.

Effects of ILD on the gut microbiota of pSS patients

Patients with pSS are at increased risk of developing ILD. To determine whether the presence of ILD is associated with distinct gut microbial features, we compare the gut microbiota of patients with ILD (pSS-ILD) to those without ILD (pSS-NO-ILD). Rarefaction curves suggested a slightly lower microbial richness in the pSS-ILD group with increasing sample size, but neither observed species counts nor Shannon diversity differed significantly between groups (Figures 9E and 9F). PCoA likewise revealed no separation in overall community composition (Figure 9G). Despite the absence of global compositional differences, species-level analysis identified several taxa differentially abundant between the two groups. Blautia wexlerae, Enterococcus casseliflavus, Bifidobacterium longum, Fusobacteriaceae bacterium, and Staphylococcus epidermidis were significantly enriched in pSS-ILD patients, whereas Lachnospira pectinoschiza, Bacteroidota eggerthii, and Raoultella ornithinolytica were notably reduced (Figure 9H). These microbial shifts may reflect or contribute to the pathophysiology of ILD in the context of pSS, offering potential leads for future mechanistic studies and biomarker development.

Discussion

In this study, we conducted a comprehensive multi-kingdom analysis of the gut microbiome in 206 pSS patients and 355 non-pSS controls, integrating compositional, functional, and metabolomic profiling to explore its role in pSS. Our investigation revealed significant alterations in bacterial, fungal, and viral communities, along with corresponding shifts in microbial metabolic functions. We further characterized microbial virulence genes, potential mimicry peptides, drug resistance, and AMPs, providing further insights into host-microbe interactions in pSS. Importantly, we delineated complex inter-kingdom microbial networks and identified disease-specific signatures that were validated in an independent geographic cohort, demonstrating robust classification performance. In vitro experiments confirmed the immunomodulatory effects of key microbial metabolites and taxa, supporting the mechanistic relevance of our multi-omic findings. Additionally, we assessed the impact of medication use and ILD on microbial composition, offering a more nuanced understanding of how clinical factors shape the gut microbiome in pSS.

Several key discrepancies emerged when comparing our finding with previous pSS studies. While earlier reports described reduced bacterial α-diversity in pSS,¹⁵ we observed no significant decrease. Similarly, the Firmicutes/Bacteroidetes (F/B) ratio, a hallmark of gut dysbiosis in autoimmune diseases such as SLE and ankylosing spondylitis (AS),^23,24,25 was unexpectedly elevated in our pSS cohort, aligning instead with observations in RA.²⁶ At the taxon level, Ruminococcus gnavus was markedly enriched in our pSS cohort, contradicting earlier studies reporting its depletion¹⁵ despite its well-documented association with autoimmune conditions including SLE,²⁷ Sjögren syndrome (SS),¹⁵ and RA.²⁸ Methodological differences, including updated reference databases and revised bioinformatic pipelines, may partially account for these discrepancies.

We identified several pSS-related species with potential relevance to immune modulation and autoimmune pathogenesis, including S. parasanguinis, L. salivarius, V. parvula, L. mucosae, C. freundii, and B. dentium. C. freundii has been shown to activate the NLRP3 inflammasome in macrophages through its type VI secretion system (T6SS), leading to IL-1β secretion and pyroptosis²⁹—mechanisms central to autoimmune inflammation. L. mucosae modulates cytokines such as IFN-β, TNF-α, IL-1β, and IL-10, which are closely linked to autoimmune disease progression.³⁰ The enrichment of S. parasanguinis, V. parvula, and L. salivarius in pSS patients aligns with previous reports.^15,31,32 These bacteria are well-recognized oral commensals frequently enriched in inflammatory oral conditions such as periodontitis and oral lichen planus.^33,34,35 Increasing evidence supported the “oral-gut axis,” whereby oral taxa translocate or ectopically colonize the gut, disseminating immune-stimulatory molecules or perpetuating systemic inflammation.³⁶ L. salivarius has also been associated with elevated systemic inflammatory markers (erythrocyte sedimentation rate [ESR], immunoglobulin G [IgG], and IgM),¹⁵ further supporting oral-gut immunological crosstalk. B. dentium can induce both Th1 and Th2 cytokines, enhancing IFN-γ and TNF-α secretion, promoting Th1-oriented responses³⁷ and boosting natural killer (NK) cell cytotoxicity and T lymphocyte proliferation.³⁸ Notably, B. dentium remained abundant in pSS patients even after treatment, highlighting the need for monitoring in therapeutic interventions.

ILD, the most common and severe pulmonary complication of pSS (prevalence ∼20%), may be influenced by the gut microbiota. Our analysis identified Blautia wexlerae, Enterococcus casseliflavus, Fusobacteriaceae bacteria, and S. epidermidis as potential contributors to ILD. Intestinal colonization by S. epidermidis may trigger systemic inflammatory responses affecting multiple organs, potentially leading to severe pathology,³⁹ and its strong association with lung diseases suggest a role in ILD pathogenesis through systemic inflammation and immune dysregulation.⁴⁰

Fungi, despite low abundance, play an important role in pSS. Previous studies reported candidiasis in 87.5% of saliva samples from pSS patients, with Candida albicans accounting for the majority,⁴¹ suggesting oral C. albicans overgrowth may reflect oral-gut microbial exchange and contribute to chronic mucosal inflammation.³⁴ C. albicans activates the IL-17/Th17 pathway: hyphal transition exposes β-glucans, activating Dectin-1/CARD9 signaling in antigen-presenting cells and promoting Th17 differentiation—consistent with our cytokine findings.⁴² M. restricta, associated with Crohn disease,⁴³ may also influence pSS through immunomodulation, though further research is needed. The convergence of intestinal fungal communities in pSS may represent both a consequence and driver of disease progression: immune dysfunction and altered gut conditions may select for stress-tolerant fungi, while dominance by proinflammatory species may amplify mucosal and systemic immune activation. Similar bidirectional relationships have been reported in Crohn disease and RA.^44,45

Viruses play a key role in autoimmune pathogenesis.^46,47 Our findings align with previous studies showing reduced Myoviridae in healthy individuals and enrichment of Quimbyviridae in non-pSS controls. Host analysis reveals that pSS-enriched vOTUs were primarily associated with pathogenic bacteria such as Streptococcus, Escherichia, and Klebsiella—genera linked to rheumatic diseases,⁴⁸ SLE,⁴⁹ and IBD.^50,51 In contrast, vOTUs enriched in non-pSS subjects often correlated with beneficial taxa including Prevotella (SCFA producers),⁵² Bacteroidota (vitamin synthesis),⁵³ and butyrate-producing anti-inflammatory genera such as Faecalibacterium and Roseburia.^54,55 These virus-host interactions were consistent with bacteriome alterations, underscoring their potential impact on disease.

Functional analysis revealed enrichment of aromatic compound degradation (e.g., tyrosine, benzoate, toluene, and xylene) in pSS patients. Bacterial metabolites such as phenol and phenyl sulfates are recognized as uremic toxins linked to kidney disease, which affects ∼5% of pSS patients.⁵⁶ Metabolomic validation confirmed accumulation of nephrotoxins and depletion of beneficial metabolites (e.g., propionate), suggesting phenol-related metabolites may contribute to renal damage in pSS. Phenol has also been implicated in autoimmune diseases such as pemphigus through induction of IL-1α and TNF-α from keratinocytes.⁵⁷ Notably, we observed enrichment of the SaeS-SaeR two-component regulatory system, originally characterized in Staphylococcus aureus, in pSS metagenome, despite no increase in S. aureus abundance. This module, which governs virulence gene expression,⁵⁸ was strongly associated with S. parasanguinis and S. salivarius, suggesting commensal streptococci may harbor homologous signaling pathways that enhance persistence under chronic inflammatory stress.⁵⁹ This aligns with our MAG-based findings in S. parasanguinis, which revealed multiple virulence-associated genes potentially regulated by such systems.⁵⁹ Additionally, we observed increased abundance of aflatoxin and pyocyanin biosynthesis modules. Aflatoxins have immunosuppressive effects and been linked to autoimmune diseases including IBD, RA, and multiple sclerosis.^60,61 Pyocyanin promotes neutrophil extracellular trap (NET) formation, which, when excessive, contributes to autoimmune conditions such as SLE and vasculitis.^62,63

MAG-based analysis revealed that S. parasanguinis, L. salivarius, and V. parvula harbor abundant immunogenic, virulence-associated, and antimicrobial resistance traits. Conserved antigens including HSP60, GRP78, enolase, and pyruvate kinase—known moonlighting proteins capable of triggering autoimmune responses through molecular mimicry—were consistently identified across species,^64,65,66 suggesting shared mechanisms of immune activation.⁶⁷ Virulence features complemented this antigenic potential: S. parasanguinis displayed strong adhesive and invasive capabilities; L. salivarius, despite probiotic associations, contained capsular and stress-adaptive genes that may promote inflammation during dysbiosis; and V. parvula possessed oxidative stress defenses enabling persistence under inflammatory conditions.⁶⁸ Widespread antibiotic resistance, particularly in S. parasanguinis, raises clinical concerns about its survival and spread under selective pressure.⁶⁹ The limited presence of antimicrobial peptide sequences suggests weak intrinsic ability to maintain microbial balance. Collectively, these traits support a model in which pSS-associated microbes contribute to chronic inflammation through persistence, immune activation, and microbial imbalance.

In vitro PBMC stimulation assays provided functional validation of these genomic predictions. Exposure to pSS-associated microbial components was associated with robust proinflammatory cytokine responses, suggesting their potential immunostimulatory effects. The strong induction of IL-17 by L. salivarius and TNF-α/IFN-γ by C. albicans suggests activation of both Th17 and Th1 pathways, consistent with pSS immunopathogenesis. Concurrent activation across bacterial and fungal taxa implies that cross-kingdom dysbiosis may amplify inflammation through cooperative immune signaling,⁷⁰ supporting an emerging concept in autoimmune disease.

In conclusion, this multi-kingdom, multi-omic study provides a comprehensive resource for understanding gut microbiome alterations in pSS, identifies robust diagnostic signatures, and offers mechanistic insights into host-microbe interactions that may inform future therapeutic strategies.

Limitations of the study

Several limitations should be acknowledged. First, the observational, cross-sectional design precludes causal inference; although multivariable adjustment was performed, residual confounding may persist. Second, both discovery and validation cohorts were Chinese, limiting generalizability to other populations with different genetic backgrounds, diets, and lifestyles. Third, functional modules and resistance determinants were inferred from gene content rather than directly measured; thus, pathway expression and phenotypic resistance remain unconfirmed. vOTU host assignments were computationally predicted, and co-occurrence networks describe associations without mechanistic directionality. Fourth, absence of longitudinal data precludes distinguishing disease effects from treatment or temporal variation. Finally, despite strong classification performance and independent validation, prospective clinical applicability and cross-laboratory reproducibility require further evaluation.

Resource availability

Lead contact

Further information and resource requests should be directed to and will be fulfilled by the lead contact, Wukai Ma (walker55@163.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Raw whole-metagenome shotgun sequencing data generated in this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under BioProject:PRJEB85773. (https://www.ebi.ac.uk/ena/browser/view/PRJEB85773).
• •The data processing and statistical analysis codes used in this study are publicly available at https://github.com/Bioinformatics-xgr/pSS_Multi-kingdom.
• •All other data and materials supporting the findings of the study are available in the paper or from the corresponding authors upon request.

Acknowledgments

This work was supported by National Natural Science Foundation of China (82474497 and 82370563), the China Postdoctoral Science Foundation (2024MD754009), Guizhou Provincial Basic Research Program (General Project) (Qian Ke He Basic MS [2026] 699), Young Elite Scientist Sponsorship Program by GAST (GASTYESS202532), Dalian Outstanding Young Scientific and Technological Talents Program (2024RJ018), Young Top Talents Program of the Liaoning Revitalization Talents Initiative (XLYC2403207), Basic Research Project of Liaoning Provincial Department of Education for Universities (LJ212410161043), Dalian Municipal Guidance Program for the Life and Health Sector (2024ZDJH01PT130), and the Key Laboratory of Guizhou Provincial Education Department (Guizhou Education Technology [2023] No.017).

Author contributions

W.M., X.K., Xueming Yao, and Q.Y. designed and directed the study. C.C., Y.X., F.Z., S.S., L.Z., Y.L., Xiaoling Yao, C.L., T.M., and W.S. have participated in clinical protocol, sample collection, and result interpretation. G.X., N.Z., Yue Zhang, Yidi Zhang, and R.G. performed data analyses, investigation, and visualization. J.K., L.C., and S.F. assisted with data curation. C.C., G.X., and S.L. wrote the manuscript. All authors revised and approved the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Raw sequencing reads for the metagenomic data	This paper	Database: European Nucleotide Archive (ENA), Accession No. PRJEB85773: https://www.ebi.ac.uk/ena/browser/view/PRJEB85773
The pipelines and scripts used in omics data analyses	This paper	https://github.com/Bioinformatics-xgr/pSS_Multi-kingdom
Software and algorithms
R	https://www.r-project.org	4.0.1
fastp	Chen et al.71	0.23.4
Bowtie2	Langmead and Salzberg72	2.5.1
MetaPhlAn4	Blanco-Míguez et al.17	4.0.6
MS-DIAL	Tsugawa et al.	5.5 (20250820)
MEGAHIT	Li et al.73	1.1.3
MetaBAT	Kang et al.74	2.12.1
CheckM2	Chklovski et al.75	0.1.3
GTDB-Tk	Chaumeil et al.76	2.0.0
Prodigal	Hyatt et al.77	2.6.3
DIAMOND	Buchfink et al.78	0.9.21.122
PhyloPhlAn	Segata et al.79	3
USEARCH	Edgar et al.37	5.2.32
MUSCLE	Edgar et al.80	3.8.1551
FastTree	Price et al.81	2.1
FigTree	Rambaut et al.82	1.4.32
iTOL	https://itol.embl.de/	6
Cytoscape	https://cytoscape.org	3.8.2
vegan (R package)	Dixon et al.83	2.6–6
randomForest (R package)	R package	4.7–1.1

Experimental model and study participant details

Participants and study design

A clinical cohort comprising 206 pSS patients and 355 non-pSS controls was recruited at the Second Affiliated Hospital of Dalian Medical University. None of the pSS patients had received steroids, immunosuppressive agents, or antibiotics within the three months preceding sample collection. The diagnosis of pSS strictly followed the 2016 American-European Consensus Group (AECG) classification criteria.⁸⁴ Patients were included if they presented with at least one symptom of ocular or oral dryness, including: persistent daily sensation of dry eyes lasting for more than three months; recurrent sensation of sand or gravel in the eyes; use of artificial tears three or more times daily; persistent daily sensation of dry mouth for more than three months; or frequent deed to drink water to assist in swallowing dry food. Exclusion criteria included conditions that could confound diagnosis or mimic pSS manifestations, such as history of head or neck radiation therapy, active hepatitis C virus infection, acquired immunodeficiency syndrome (AIDS), sarcoidosis, amyloidosis, graft-versus-host disease, or IgG4-related disease. A diagnosis of pSS was confirmed when the total score reached or exceeded four points based on the following weighted items: focal lymphocytic sialadenitis in labial salivary gland biopsy with a focus score ≥1 focus/4 mm² (3 points); positive serum anti-SSA antibody (1 point); ocular staining score (OSS) ≥ 5 or Van Bijsterveld score ≥4 in at least one eye (1 point); Schirmer’s test ≤5 mm/5 min in at least one eye (1 point); and unstimulated whole salivary flow rate ≤0.1 mL/min measured according to Navazesh and Kumar (1 point). Patients regularly taking cholinergic medications were instructed to discontinue them prior to ocular and oral dryness assessments. Disease activity was quantified using the EULAR Sjögren’s Syndrome Disease Activity Index (ESSDAI).⁸⁵ Among the 206 pSS patients, 142 had a prior history of antibiotics or immunosuppressive drug use, while the remaining 64 were treatment-naïve. Additionally, 33 of the 206 pSS patients had a concurrent diagnosis of ILD.

Non-pSS subjects were carefully screened to exclude those with arthralgia, heart failure, renal failure, autoimmune diseases, or inflammatory disorders. Additional exclusion criteria included severe systemic diseases, malignancies, septicemia, cardiovascular or metabolic disorders, and diarrheal symptoms. Participants with a history of excessive alcohol consumption or those who had ingested sour milk within the week prior to sampling were also excluded to minimize potential confounding factors. All participants were of Chinese ethnicity, ensuring a homogenous study population.

The study was approved by the Ethics Committee of the Second Affiliated Hospital of Dalian Medical University (approval number: 2022071) and conducted in compliance with the principles outlined in the Declaration of Helsinki and the International Council for Harmonization Guidelines for Good Clinical Practice. Informed consent was obtained from all participants prior to inclusion.

To validate the generalizability of our findings, an independent external validation cohort comprising 56 pSS patients and 58 healthy controls was recruited from a hospital in western China. All samples from this cohort were processed and analyzed using the same metagenomic and bioinformatic pipelines as the discovery cohort.

Method details

Sample collection, DNA extraction, and whole-metagenome shotgun sequencing

Fecal samples were self-collected by participants following defecation at the hospital and placed on dry ice. Samples were subsequently transported to the laboratory, divided into two aliquots, and stored at −80°C until further processing. Total DNA was extracted from 170 mg of each fecal samples using the Tiangen fecal DNA extraction kit (Tiangen, China) following the manufacturer’s instructions. DNA concentration and purity were assessed using a NanoDrop2000 spectrophotometer and Qubit 4.0 fluorometer. Extracted DNA was fragmented using a Covaris M220 ultrasonicator (Gene Company Limited, China), and a paired-end library with a read length of 150bp and an insert size of approximately 350 bp was constructed for each sample. All libraries were barcoded, pooled, and subjected to whole-metagenome shotgun sequencing on the Illumina NovaSeq platform. Initial base calling was conducted using the platform’s default parameters. Raw sequencing reads were processed independently for quality control using fastp v0.23.4.⁷¹ Low-quality bases (Q < 30) were trimmed from the ends of the reads, and reads containing ambiguous bases (N), adapter contamination, or those shorter than 90 bp were filtered out to obtain high-quality reads. Human reads were removed by aligning high-quality reads to the human reference genome (GRCh38) using Bowtie2 v2.5.1.⁷²

Gut bacteriome, mycobiome, and virome profiling

Gut bacteriome profiling

The prokaryotic composition of the gut microbiome in fecal metagenomes from all samples was profiled using the MetaPhlAn4 v4.0.6.¹⁷ Relative abundances of microbial species were calculated by normalizing the abundance of each species to the total number of reads in the corresponding sample and multiplied by 100 to express the values as percentages. Species-level abundances were subsequently aggregated to obtain relative abundances at higher taxonomic levels (such as phylum and genus) by summing the relative abundances of all species within each taxonomic group.

Reference database construction and gut mycobiome profiling

We integrated fungal genomes reported in a recently study with all publicly available genomes from the NCBI genome database, applying the same processing and integration methods as described therein.⁸⁶ From a total of 16,634 genomes, 1,384 were excluded due to extremely low assembly quality (N50 length <2,000bp or number of scaffolds >10,000) or evidence of mixed genome assemblies, Resulting in 15,250 high-quality genomes retained for downstream analyses. To minimize non-specific read mapping to fungal genomes, filtered reads were first aligned against three databases: the GRCh38 human genome, the Unified Human Gastrointestinal Genome (UHGG) collection,⁸⁷ and the SILVA rRNA database,⁸⁸ thereby excluding reads originating from human or prokaryotic sources. For each sample, the remaining reads were aligned to our customized gut fungal genome catalog using Bowtie2 v2.5.1,⁷² and read counts per genome were calculated. To generate mycobiome composition profiles, read counts for each genome were normalized by genome size. Normalized counts were then summed within each sample to preserve the relative abundance of each fungal population. For higher taxonomic levels, relative abundances were computed as the sum of relative abundances of all constituent populations assigned to that taxon.

Gut virome profiling

The Chinese Gut Viral Catalog (cnGVC),⁸⁹ a comprehensive gut virus catalog constructed from over 10,000 publicly available fecal metagenomes comprising more than 93,000 nonredundant viral operational taxonomic units (vOTUs), was used for virome analysis. High-quality reads from all samples were aligned to the cnGVC database using Bowtie2 v2.5.1 with a nucleotide similarity threshold of 95% to define viral “species-level” units.⁹⁰ To generate vOTU abundance profiles, reads mapped to each vOTU were aggregated and normalized to the total number of mapped reads per sample. Relative abundances at the viral family level were subsequently obtained by summing the relative abundances of all vOTUs assigned to the same family.

Gut microbiome functional profiling and host metabolome analysis

Microbial functional profiling

Functional profiling of the metagenomes was performed by mapping clean reads to the integrated gene catalog of the human gut microbiome,⁸⁷ followed by aggregating read counts for Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthologs (Kos).⁹¹ Abundances of KEGG functional modules were subsequently calculated by summing read counts across all KOs constituting each module.

Serum and fecal metabolome analysis

Untargeted metabolomic profiling was performed on paired serum and fecal samples from 75 pSS patients and 66 non-pSS controls. For serum, 100 μL of each sample was mixed with 400 μL of pre-chilled methanol to precipitate proteins, vortexed for 1 min, incubated at −20°C for 30 min, and centrifuged at 13,000 × g for 15 min at 4°C. The supernatant was collected for LC-MS/MS analysis. For fecal samples, approximately 80 mg of each specimen was homogenized in 800 μL of 80% methanol, vortexed, sonicated for 10 min, and centrifuged under identical conditions; the resulting supernatant was used for LC-MS/MS detection. Metabolomic data were acquired using an ultra-high-performance liquid chromatography coupled with high-resolution mass spectrometry (UHPLC–MS/MS, Thermo Fisher Scientific) in both positive and negative ionization modes. Raw data were processed using MS-DIAL (version 5.5 20250820) for peak detection, deconvolution, alignment, and normalization. Metabolite annotation was achieved by matching accurate mass and MS/MS spectra against public databases including HMDB, METLIN, and MassBank. The resulting abundance matrix was exported for downstream statistical and correlation analyses.

Analysis of metagenome-assembled genomes (MAGs)

Metagenomic assembly and binning, and taxonomic classification

Metagenomic assembly was performed using MEGAHIT v1.1.3.⁷³ Contigs shorter than 2,000bp were discarded, and the remaining contigs were aligned against the original raw reads using Bowtie2 v2.5.1 to calculate coverage information. Binning was conducted with MetaBAT v2.12.1,⁷⁴ and genome quality was assessed using CheckM2 v0.1.3.⁷⁵ MAGs meeting quality standards (completeness ≥50%, contamination <5%; QS > 50, where QS = completeness - 5 x contamination) were retained for downstream analysis. Taxonomical classification of each MAG was performed using GTDB-Tk v2.0.0 with database release 207,⁷⁶ employing the ‘classify_wf’ function, which utilizes 120 ubiquitous single-copy proteins for bacterial taxonomy assignment.

Functional annotation of MAGs

Open reading frames (ORFs) were predicted using Prodigal v.2.6.3.⁷⁷ Virulence genes were identified by aligning ORFs against the Virulence Factor Database (VFDB)⁹² using DIAMOND v0.9.21.122⁷⁸ (≥50% identity, ≥50% coverage). To identify potential mimicry epitopes, T cell (n = 8,487) and B-cell (n = 34,166) epitope sequences from the Immune Epitope Database (IEDB) were aligned against ORFs using BLAST (>75% identity, 75% coverage). Antimicrobial resistance genes were identified by searching ORFs against the comprehensive drug resistance Gene (DRG) database using DIAMOND BLASTp (e-value≤1 × 10⁻⁵, ≥80% identity, ≥70% query coverage). Antimicrobial peptides (AMPs) were predicted from small ORFs using an ensemble of neural network models (Long Short-Term Memory, Attention mechanisms, and Bidirectional Encoder Representations from Transformers) as previously described, with model combinations optimized to enhance precision and recall.

Phylogenetic analysis of MAGs

Phylogenomic analysis of MAGs was performed using PhyloPhlAn v3.0,⁷⁹ a pipeline designed for large-scale phylogenetic reconstructions of microbial genomes. Briefly, homologous sequences corresponding to the 400 most ubiquitous proteins in prokaryotic genomes were identified within each MAGs by searching against the PhyloPhlAn 3.0 non-redundant database using USEARCH v5.2.32.⁹³ Selected protein markers were individually aligned using MUSCLE v3.8.1551,⁸⁰ and the resulting alignments were concatenated to generate a super matrix for whole-genome phylogenetic inference. A maximum-likelihood phylogenetic tree was constructed using FastTree v2.1,⁸¹ and subsequently midpoint-rooted with FigTree v1.4.32.⁸² The final phylogenetic tree, incorporating taxonomic annotations, was visualized and annotated using iTOL v6.⁹⁴

Peripheral blood mononuclear cell (PBMC) stimulation assay

PBMCs were isolated from healthy donors using Ficoll–Paque density gradient centrifugation and cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum. To assess immune activation, PBMCs were stimulated with fecal microbiota supernatant (FMS) derived from pSS patients or healthy controls. Fecal samples were suspended in sterile PBS (1 g/10 mL), vortexed, centrifuged (12,000 × g, 10 min, 4°C), and filtered through 0.22 μm filters to obtain sterile FMS. PBMCs were treated with 1:200 diluted FMS for 4 h at 37°C in 5% CO₂ atmosphere.

For microbial stimulation experiments, four representative species identified in metagenomic analysis (i.e., Streptococcus parasanguinis, Veillonella parvula, Ligilactobacillus salivarius, and Candida albicans) were cultured under appropriate aerobic or anaerobic conditions. Bacterial species were grown in Gifu Anaerobic Medium (GAM, Nissui Pharmaceutical Co., Tokyo, Japan), while C. albicans was maintained in Modified Martin Medium (Hopebio, Qingdao, China). Cultures were harvested at mid-log phase, washed twice with Hank’s balanced salt solution, and resuspended in RPMI-1640. Both culture supernatants and heat-killed particle fractions were collected and used for PBMC stimulation. PBMCs were treated with these preparations at a 1:200 dilution for 4 h. Lipopolysaccharide (100 ng/mL) were used as positive control.

Following stimulation, total RNA was extracted using TRIzol reagent, and cDNA was synthesized using a reverse-transcription kit (Takara). Relative mRNA expression levels of TNF-α, IL-6, IL-8, IL-17, and IFN-γ were quantified by qPCR with SYBR Green chemistry (Applied Biosystems), normalized to GAPDH, and calculated using the 2⁻ΔΔCt method. Each experiment was performed in triplicate, and statistical significance was assessed using the Wilcoxon rank-sum test where appropriate.

Construction of multi-kingdom co-occurrence network

Correlation networks were constructed to explore associations among bacterial species, fungi species, and functional modules using Spearman’s rank correlation coefficient. Only correlations with |ρ| > 0.6 were considered robust and retained for network visualization. Networks were generated and visualized using Cytoscape v.3.8.2.⁹⁵ The same analytical approach was applied to construct interaction networks among bacterial species, fungal species, and vOTUs. For the associations between metabolites and microbial taxa (bacteria and fungi), Spearman correlations were calculated, and only those with |ρ| > 0.2 and q < 0.05 were deemed statistically significant. For all network analyses, p-values were adjusted for multiple testing using the Benjamini–Hochberg (BH) procedure, and only correlations with an adjusted q-value <0.05 were retained for visualization and downstream interpretation.

Quantification and statistical analysis

All statistical analyses were conducted using the R v4.0.1 platform. For each sample, gut microbial richness was evaluated by the observed number of species-level taxa (i.e., bacterial species, fungal species, and vOTUs), and microbiome diversity was estimated using the Shannon index. These indices were calculated with the vegan v2.6-6 package⁸³ after rarefying all samples to a uniform sequencing depth of 10 million reads. Principal coordinates analysis (PCoA) based on Bray-Curtis dissimilarities was performed using the vegan package. Permutational multivariate analysis of variance (PERMANOVA) was conducted with the adonis function, and statistical significance was assessed using 1,000 permutations. Differences in microbial diversity, taxonomic composition, and clinical characteristics between groups were evaluated using Student’s t test or Wilcoxon rank-sum test, as appropriate. p-values were adjusted for multiple testing using the BH procedure, with adjusted p-value (q) < 0.05 considered statistically significant. Random forest models were trained using the randomForest v4.7–1.1 package (1,000 trees) to discriminate between pSS patients and non-pSS controls based on the abundance profiles of the differentially abundant bacteria, fungi, and vOTUs. Linear discriminant analysis effect size (LEfSe, Galaxy version 1.1.2), an advanced algorithm designed for the elucidation of high-dimensional biomarkers,⁸³ was employed to identify high-dimensional biomarkers characterizing compositional differences between groups.

Published: May 11, 2026