Genome-resolved and culture-based atlas of the feline gut microbiome enables host-adapted probiotic development

Abstract

Domestic cats (Felis catus) depend on their gut microbiome for metabolism, immunity, and pathogen defense, yet its genomic characterization remains limited. We combined large-scale metagenomics and culturomics to define the feline gut microbiome and identify indigenous probiotic candidates. Analysis of 412 feline fecal metagenomes produced 2852 strain-resolved metagenome-assembled genomes (MAGs) grouped into 514 species-level genome bins, including 106 putative novel taxa. This catalog revealed 24 core species and two enterotypes: ET-P, deaminated by Prevotella, and ET-CB, enriched for Collinsella, Blautia, Bifidobacterium, Ligilactobacillus, MAG-based screening prioritized 113 candidate probiotic species. Culturomics recovered 2904 isolates representing 110 species-level taxa, including 75 putative novel species and a candidate novel genus. Six feline-derived isolates were selected for downstream testing, and five exhibited favorable probiotic traits in vitro, including acid and bile tolerance, anti-Escherichia coli activity, and favorable cytokine responses. In a pathogenic Escherichia coli-induced dirrhea model in cats, a five-strain indigenous consortium improved fecal scores and reduced IL-2, IL-1β, and IL-6, with TNF-α suppression superior to antibiotics or a commercial probiotic. These results establish FelMGDB as a resource for feline microbiome research and highlights indigenous probiotics as promising interventions for feline gut health.

Introduction

Domestic cats (Felis catus) are globally prevalent companion animals, and their health has become an increasing concern for owners and veterinarians. Beyond welfare considerations, the close human–animal interface has public-health implications, as companion animals and owners can exchange microbiota, including antibiotic resistance determinants^1,2. These factors motivate a deeper, clinically relevant understanding of the feline gastrointestinal microbiome.

The feline gut microbiome is a complex community that contributes to nutrient acquisition, digestion, immune modulation, and colonization resistance³. Dysbiosis has been linked to common feline disorders, including enteritis and kidney disease^3,4, as well as obesity⁵. Despite this relevance, systematic, genome-resolved characterization of the feline gut ecosystem remains limited, constraining ecological inference and translational applications. Shotgun metagenomics facilitates culture-independent profiling and the reconstruction of metagenome-assembled genomes (MAGs)^6,7, which has expanded microbial catalogs across multiple host species, including chickens⁸, swine⁹, cattle^10,11, and giant pandas^12,13. However, comparable resources for cats have, to date, been restricted in cohort breadth and taxonomic coverage.

Probiotic interventions have shown promise for prevention and adjunctive treatment of gastrointestinal and other conditions, with growing interest in companion animals^14–16. In this context, autochthonous (host-adapted) microorganisms, long-term colonizers of the native gastrointestinal niche, are of particular interest because they may exhibit enhanced ecological fitness and functional compatibility within the host^17–19. Developing such host-adapted probiotics requires strain-level resolution to nominate candidates and culture access to enable mechanistic evaluation.

Here, we address these gaps by integrating large-scale, genome-resolved metagenomics with complementary culturomics and targeted functional assessments in cats. Specifically, we (i) construct an expanded catalog of feline gut MAGs and delineate community features including a conservative core and two robust enterotypes; (ii) prioritize candidate autochthonous probiotics using a MAG-based framework and augment these via large-scale isolation; and (iii) evaluate a mixed indigenous formulation in a pathogenic Escherichia coli–induced diarrhea model in domestic cats. Together, these components provide scalable resources, testable strain candidates, and an experimental path from community structure to translational potential in the feline gut ecosystem.

Results

Comprehensive assembly and refinement of feline gut metagenomes

Fecal samples (n = 234) were collected from 210 domestic cats spanning nine breeds, with stratification by age, sex, and diet. Sources comprised accredited breeding facilities and private households (Supplementary Table 1). After quality control, sequencing yielded 2.44 Tb of reads (mean 10.41 Gb per sample), providing depth sufficient to resolve community structure. We additionally retrieved publicly available datasets (n = 178) from five studies (PRJEB4391, PRJNA758898, PRJNA908260, PRJNA944553, PRJEB52014) and reprocessed them through the same pipeline, yielding 1.28 Tb of clean data (mean 7.21 Gb per sample; Fig. 1a). Metadata (geolocation, sex, health status, diet, etc.) are summarized in Supplementary Data 1.

Fig. 1. Geographic distribution and phylogeny of feline gut MAGs.

a Global map of metagenomes analyzed; circle size reflects sample count per site. Cities represented by a single sample were omitted for clarity. b Phylogenomic tree of species-level genome bins (SGBs; 95% ANI), annotated with GTDB-Tk v2.4.0 and visualized in iTOL (v6.5.2). Concentric rings from inner to outer indicate clade colors, genome size, completeness (purple), MAG quality category (high-quality MAGs shown in white and ultra-high-quality MAGs highlighted in purple), contamination (orange), probiotic potential, and novelty (putative novel species defined as genomes with FastANI ANI < 95% to GTDB reference genomes with alignment fraction >0.5).

A dual-mode assembly strategy (per-sample and co-assembly) across the 234 in-house metagenomes produced 8942 MAGs meeting minimal quality (≥70% completeness, <5% contamination). To expand representation, we integrated 6,606 publicly available MAGs from Plaza Oñate et al.²⁰ that met identical criteria. After dereplication with dRep v3.4.0 at 99% ANI, the non-redundant catalog comprised 2852 MAGs (Supplementary Data 2).

Assembly quality was high: 1143 MAGs (40.1%) were high-quality (≥90% completeness, <5% contamination), and 982 (34.4%) achieved near-complete status (≥95% completeness, <3% contamination; Supplementary Fig. 1). Prior to dereplication, 1379 strain-level representatives (48.4%) were supported by at least two independent MAGs; of these multi-supported clusters, 1066 (77.3%) were high-quality, indicating robust biological reproducibility.

Across 16 phyla (Supplementary Data 2), the catalog was dominated by Bacillota (n = 1549; 54.31%), Actinobacteriota (n = 722; 25.32%), Bacteroidota (n = 307; 10.76%), Pseudomonadota/Proteobacteria (n = 128; 4.48%), and Campylobacterota (n = 60; 2.10%). Eleven MAGs lacked genus-level assignment and formed six de novo clusters at genus-scale thresholds (≥85% ANI), suggesting six previously uncharted genera in the feline gut. In addition, 214 MAGs could not be assigned to known species and grouped into 106 species-level clusters (95% similarity).

Clustering 2852 strain-level MAGs at 95% ANI yielded 514 SGBs (Fig. 1b) including 444 SGBs represented by ≥2 MAGs, 70 by a single MAG, and 247 by ≥10 MAGs (Supplementary Data 3). The largest SGB was SGB091, annotated as Prevotella copri in GTDB v2.4.0 (corresponding to Segatella copri in the NCBI taxonomy), with 450 MAGs. Because large MAG resources often nominate putative new taxa without assessing reliability, we evaluated candidate novel genera using representative completeness, contamination, and SGB membership size. Two SGBs (SGB240 and SGB310) exhibited near-complete genomes (>95% completeness, <1% contamination) and included 15 and 37 MAGs, respectively, supporting higher confidence for these two novel genera relative to the other four.

To prioritize probiotic candidates, we applied MetaProbiotics to all SGBs, identifying 113 candidate SGBs: 54 named species, 33 GTDB-defined species, and 26 novel species from this study (Fig. 1b; Supplementary Data 3). Classical probiotic taxa, including Bifidobacterium longum, B. adolescentis, B. pullorum, Lactococcus lactis, and Lactiplantibacillus plantarum scored highly. Several novel lineages (e.g., SGB503, CALAMD01 SCATMAG00644; SGB438, GCATMAG05267 SCATMAG05267, a putative novel genus) also ranked highly, indicating previously unrecognized probiotic potential within the feline gut.

To enable community exploration, we built FelMGDB (https://FelMGDB.bacbase.com), integrating the 2,852 MAGs with host/sample metadata from 412 feline gut metagenomes. The portal supports bulk downloads, keyword search, and interactive views of taxonomic assignment, genome quality, functional annotation, and sample context (sex, geography, diet, and relative abundance).

Core gut microbiota and enterotype structure in domestic cats

Core taxa were defined as species present in ≥80% of samples (n = 412) at ≥0.05% relative abundance. Twenty-four species met these criteria including 17 Bacillota and 4 Actinomycetota. Of these, 11 corresponded to formally named species, 11 were assigned to GTDB-defined placeholder species, and two lacked species-level assignment in GTDB, indicating potentially novel taxa recovered in this study (Fig. 2a). MAGs representing core species were generally high-quality; 15 exceeded stringent “extra-high” thresholds (≥97% completeness, <3% contamination), and only four fell below standard high-quality criteria (Supplementary Fig. 2a).

Fig. 2. Feline gut microbiome enterotypes and their taxonomic signatures.

a Heatmap displays the proportion of abundance level for 24 core microbiota (presence in ≥80% samples, ≥0.05% abundance) across feline gut microbiome samples. Abundance levels were categorized as: Low (0–0.05%), Medium (0.05–0.1%), High (0.1–1%), and Very High (>1%). b Principal Coordinates Analysis (PCoA) of Bray-Curtis dissimilarity between enterotypes (ET-P: orange; ET-CB: blue). c Boxplots comparing top five genera (mean relative abundance) between enterotypes. Shared taxa (Collinsella, Blautia_A) and enterotype-specific taxa are annotated. Significance levels: ***p < 0.001 (wilcox.test). d Volcano plot of species-level differential abundance (|log₂FC | > 1, FDR < 0.05).

Using species-level profiles from the 514 SGBs, we resolved two robust enterotypes (Fig. 2b): ET-P, dominated by Prevotella (mean 20.60%) with secondary Collinsella (6.56%) and Phocaeicola (5.28%); and ET-CB, enriched for Collinsella (13.62%), Blautia_A (12.07%), Bifidobacterium (10.01%), and Ligilactobacillus (6.70%) (PERMANOVA p < 0.001; Fig. 2c; Supplementary Fig. 2b). Notably, ET-CB concentrates two putatively probiotic genera (Bifidobacterium, Ligilactobacillus). Species-level contrasts indicated that ET-P’s signature was driven by Prevotella copri, whereas ET-CB was characterized by Collinsella sp902362275, Blautia_A caecimuris, Blautia_A wexlerae, and Bifidobacterium pullorum_B (Fig. 2d). For accessibility, we implemented the enterotype classifier as a Python pipeline and a web app (https://FelMGDB.bacbase.com/enterotype_predict).

Culturomics of the feline gut

From 2904 colonies identified by 16S rRNA gene sequencing, pre-clustering at 99.5% similarity yielded 657 non-redundant clusters. Of these, 219 colonies could not be assigned to any RefSeq/NCBI species at ≥98.7% similarity. These unassigned colonies at species level were subsequently clustered at 98.7% similarity, yielding 75 clusters, suggesting the presence of 75 novel species distributed across 9 known genera: Vagococcus, Enterococcus, Escherichia, Shigella, Proteus, Companilactobacillus, Clostridium, Bacteroides, and Staphylococcus. One cluster showed 94.50% similarity to Clostridium perfringens, consistent with a candidate novel genus.

The remaining 438 clusters mapped to 35 known species (Supplementary Table 2). Ten species overlapped with MAG-derived assemblies, whereas 25 species were not recovered from metagenomics—highlighting complementary yield of culturomics. Only two species (Bifidobacterium longum and Enterococcus faecalis) overlapped with in silico probiotic predictions.

Identification of candidate indigenous probiotics for cats

To generate an initial feline-derived probiotic screening panel, we integrated MAG-based prioritization with literature-guided selection of culturomics-derived isolates. Bifidobacterium longum and Enterococcus faecalis were retained because they were the only two species identified by both culturomics and in silico probiotic prediction. The candidate set was then broadened to include additional feline isolates representing taxa with reported relevance to companion-animal gut health, particularly Enterococcus faecium, Ligilactobacillus animalis, and Bacillus spp. (represented here by B. subtilis), together with two exploratory isolates, Bacillus zanthoxyli and Companilactobacillus nuruki, for first-pass in vitro evaluation^21–25. To reduce redundancy among closely related isolates, we reconstructed a phylogeny based on 16S rRNA gene sequences and selected one representative strain per retained species cluster for downstream experiments (Supplementary Fig. 3). The final set comprised Bifidobacterium longum A-1-12, Enterococcus faecalis A-A-2, Enterococcus faecium M-3, Ligilactobacillus animalis A-11-15, Bacillus subtilis O2-5-1-11, and Companilactobacillus nuruki A-148; although Bacillus zanthoxyli A-3 was included in the initial panel, it was not retained after phylogeny-guided down-selection. Growth curves showed five strains entered logarithmic growth at 4–7 h; B. longum A-1-12 required ~9 h (Supplementary Fig. 4a). During B. subtilis O2-5-1-11 growth, pH decreased then rose after OD ≈ 0.8, consistent with onset of apoptosis/lysis (Supplementary Fig. 4b). Bile-salt tolerance was observed for five of six strains (Supplementary Fig. 5a–f). For acid tolerance, all strains except B. subtilis O2-5-1-11 remained viable (Supplementary Fig. 5g–l). Against pathogenic E. coli, all strains except B. subtilis inhibited growth (Fig. 3b).

Fig. 3. In vitro probiotic traits.

a Cytotoxicity test. The “+“ symbol indicates the simultaneous supplementation of LPS and probiotics. Values are presented as Mean ± SE, with different letters denoting significant differences between treatment groups (p < 0.05). b Bacteriostatic ring experiment for candidate probiotics. c–e Comparative effects of different probiotic treatments on gene expression of inflammatory cytokines (IL-1β, IL-8, and TNF-α). The “+“ symbol indicates the simultaneous supplementation of LPS and probiotics.

In HT29 assays, LPS reduced viability and increased LDH release; probiotics alone were non-toxic (Fig. 3a). Co-treatment with probiotics mitigated LPS-induced cytotoxicity for all strains except B. subtilis. Similarly, LPS upregulated pro-inflammatory cytokines (TNF-α, IL-1β, IL-17, IL-6, IL-8), whereas probiotics alone did not alter basal expression (Supplementary Fig. 6). Probiotic co-treatment attenuated LPS-induced cytokine expression (Fig. 3c–e), again with B. subtilis as the exception.

Impact of indigenous Probiotic Supplementation on Pathogenic E. coli-induced diarrhea of cats

Based on in vitro performance, we formulated a mixed indigenous probiotic (MProT) comprising B. longum A-1-12, L. animalis A-11-15, C. nuruki A-148, E. faecalis A-A-2, and E. faecium M-3 (excluding B. subtilis). We compared five groups: Blank Control (BC), Infection Control (InfC), Antibiotic Treatment (AntT), Commercial Probiotic Treatment (CProT), and Mixed Probiotic Treatment (MProT) (Fig. 4a).

Fig. 4. In vivo efficacy against E. coli–induced diarrhea in cats.

a Experimental design. b Representative fecal scores for Post-Infection (Post-Inf) and Post-Treatment (Post-Tre) phases. Fecal scores across Pre-Inf, Post-Inf, and Post-Tre for BC (c), InfC (d), AntT (e), CProT (f), and MProT (g). h–j Serum IgA, IgG, IgM at endpoint. k–p Blood cytokine mRNA (IL-2, IL-1β, IL-6, IL-10, IFN-α, TNF-α) at endpoint. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001.

Body weight and feed intake did not differ among groups across phases (Supplementary Fig. 7). As expected, fecal consistency worsened after E. coli challenge (higher fecal scores; Fig. 4b–g). Post-treatment, only MProT showed a significant reduction in fecal scores relative to the infected phase (Fig. 4g). In contrast, AntT and CProT exhibited higher post-treatment fecal scores than pre-infection baselines (Fig. 4c–g).

Cytokine/immunoglobulin baselines did not differ prior to challenge (Supplementary Fig. 8). E. coli exposure elevated IgG/IgM and cytokines (IL-2, IL-1β, IL-6, TNF-α), confirming model establishment (Supplementary Fig. 9). At endpoint, IgA/IgG/IgM levels in AntT, MProT, and CProT were significantly lower than InfC and were indistinguishable from BC (Fig. 4h–j). For cytokines, AntT remained comparable to InfC, whereas MProT and CProT reduced IL-2, IL-1β, and IL-6 to near-baseline (BC) levels (Fig. 4k–m). Notably, only MProT significantly reduced TNF-α compared with both AntT and InfC (Fig. 4p); CProT did not (Fig. 4p). IL-10 and IFN-α did not differ among groups (Fig. 4n–o).

Discussion

As companion animals whose popularity is rising with urbanization, cats face a considerable health burden that is of increasing concern to owners. The gut microbiota is closely linked to host immunity and disease susceptibility²⁶, including common feline conditions such as kidney disease⁴, enteritis³, and obesity⁵. Moreover, microbial exchange between companion animals and their owners may have implications for human health². These considerations underscore the need for a deeper understanding of the feline gastrointestinal microbiome. Previous work by Yang et al². assembled 161 MAGs to profile antibiotic resistance genes (ARGs) and mobile genetic elements (MGEs) in cats; however, the limited cohort (30 fecal samples) and modest MAG yield constrain representativeness. By contrast, the present study substantially expands genome-resolved resources through analysis of a new metagenomic cohort (n = 234) that spans three dietary regimens, four geographic regions, nine breeds, and multiple age groups, and by integrating 178 publicly available samples from five studies. In aggregate, we constructed a large-scale, non-redundant catalog of 2852 strain-level MAGs (99% similarity), which provides a foundation for future investigations into feline gut microbial diversity and its roles in health and disease.

Using 412 fecal metagenomes, we delineated the core feline gut microbiota as taxa present in ≥80% of individuals at ≥0.05% relative abundance²⁷. Twenty-four species fulfilled these criteria, with a substantial proportion affiliated with Firmicutes, a phylum commonly associated with protein digestion, fiber fermentation, and immune modulation in the mammalian gut. Consistent with prior observations that shifts between Firmicutes and Bacteroidetes are linked to feline obesity⁵, these findings highlight the putative contribution of Firmicutes to gut homeostasis. Among core Firmicutes, several taxa, such as Collinsella spp., contribute short-chain fatty acids (SCFAs), metabolites with energy-providing and anti-inflammatory properties; decreases in SCFA-producing bacteria have been associated with feline disease onset and progression²⁸. While functional inference from composition warrants cautious interpretation, the persistence of these taxa across diverse hosts and contexts supports their ecological importance in the feline gut.

Inter-individual heterogeneity in community structure and function is a hallmark of the gut ecosystem. Enterotyping offers a framework to summarize such variation across physiological and pathological states^29,30. Despite limited prior work in cats, our integration of reference genomes from FelMGDB with species-level composition profiles identified two robust enterotypes: ET-P, typified by Prevotella predominance, and ET-CB, enriched for Collinsella, Blautia_A, Bifidobacterium, and Ligilactobacillus. At the species level, ET-P was driven by Prevotella copri, a taxon previously implicated in feline obesity^5,31 and linked to adipose deposition in swine studies³². Although the current data do not permit causal inference regarding ET-P and feline metabolic status, this pattern nominates ET-P as a candidate community configuration for future investigation into metabolic dysregulation. ET-CB exhibited a more balanced taxonomic distribution, with Bifidobacterium among its abundant genera; notably, Bifidobacterium longum has been associated with maintenance of gut homeostasis via immunomodulatory and metabolic effects³³. These observations, together with the high prevalence of B. longum in our dataset, support its prioritization as a feline-adapted probiotic candidate, while emphasizing the need for targeted validation.

Probiotics are increasingly recognized as viable strategies for disease prevention and adjunctive therapy in both humans and animals³⁴. In cats, strains or products have been investigated for gastrointestinal disorders³⁵ and for improving digestion^36,37, with additional applications reported for oral, hepatic, and renal conditions. Genome-resolved prediction of probiotic potential from MAGs can guide discovery and isolation of novel candidates^38,39. Applying metaProbiotics³⁸ to our catalog, we identified 113 species-level candidates, including 54 well-documented taxa such as Bifidobacterium longum, B. adolescentis, B. pullorum, Lactococcus lactis, and Lactiplantibacillus plantarum. Importantly, several novel species, including SGB503 (CALAMD01 SCATMAG00644) and SGB438 (GCATMAG05267 SCATMAG05267; putative novel genus), received elevated scores, suggesting the presence of unexplored probiotic resources within the feline gut. These predictions provide a tractable shortlist for targeted culture and functional assessment.

Because MAG-based inference is necessarily provisional, isolation remains essential for mechanistic interrogation and translational use of probiotics. We therefore performed large-scale culturomics, yielding 2904 colonies from healthy feline feces. These isolates spanned 35 known species and 75 putative novel species. To our knowledge, this represents the first large-scale culturomic survey of the feline gut and substantially broadens the culture collection available for developing cat-specific probiotics. Concordant with observations in other hosts^40,41, overlap between culture-derived species and MAGs was limited (10 shared species), and 25 cultured species were not recovered by metagenomic assembly. These findings reinforce the complementarity of cultivation and genome-resolved metagenomics in recovering microbial diversity⁴² and argue for integrated strategies when building host-adapted probiotic libraries.

Diarrhea is common among cats in household and breeding settings and arises from diverse etiologies, including parasites, viruses, and bacteria⁴³. Dysbiosis is frequently observed irrespective of the immediate cause¹⁶. While antibiotics can be beneficial, misuse risks resistance and gastrointestinal adverse effects, motivating increasing interest in probiotics as alternatives for prevention and management. Prior studies have reported benefits of Enterococcus faecium⁴³, Bacillus licheniformis⁴⁴, Bacillus alleviates²³, Bacillus coagulans⁴⁵, and mixed formulations⁴⁶ in pet cats. Building on our in silico prioritization and literature, we selected six indigenous isolates for in vitro evaluation. Across assays of acid and bile tolerance, anti-Escherichia coli activity, cytotoxicity (lactate dehydrogenase), and cytokine gene expression, five strains—excluding Bacillus subtilis O2-5-1-11—exhibited probiotic-like properties, supporting their candidacy as feline indigenous probiotics.

We further established a pathogenic E. coli–induced diarrhea model in cats⁴⁷. Post-challenge increases in fecal scores and in circulating immunoglobulins (IgG, IgM) and cytokines (IL-2, IL-1β, IL-6, TNF-α) indicated an active immune response⁴⁸ and validated the model. Therapeutically, fecal scores improved significantly only in the mixed indigenous probiotic group (MProT), whereas the antibiotic-treated and commercial probiotic groups did not show comparable normalization. Consistent with these clinical readouts, endpoint cytokine profiles revealed that the antibiotic group remained elevated relative to controls, while both MProT and the commercial probiotic (CProT) reduced IL-2, IL-1β, and IL-6 toward baseline; notably, only MProT significantly lowered TNF-α compared with both antibiotic and infection controls. IL-10 and IFN-α did not differ across groups, suggesting selective immunomodulation. These findings accord with prior reports that probiotics attenuate diarrhea caused by enterotoxigenic E. coli⁴⁹, and further indicate that an indigenous mixed formulation may outperform both antibiotics and a commercial product in reducing specific cytokines and alleviating clinical signs in domestic cats.

This study delivers an expanded genome-resolved catalog of the feline gut microbiome, delineates a conservative core microbiota, resolves two robust enterotypes, and couples culture-independent prediction with culture-dependent validation to nominate indigenous probiotic candidates. While mechanistic dissection and controlled trials remain necessary for translation, the combined metagenomic-culturomic framework presented here provides a scalable route toward feline-adapted probiotic development and a resource for future studies of gut ecology and feline health.

In conclusion, we present a genome-resolved baseline for the feline gut microbiome by assembling 2,852 non-redundant MAGs across 514 SGBs, defining a 24-species core, and resolving two robust enterotypes (ET-P and ET-CB). Using this catalog, metaProbiotics prioritized 113 candidate probiotic species, including previously uncharacterized lineages. Complementary culturomics recovered 110 species (75 putative novel), and in vitro screening identified five indigenous isolates with favorable probiotic traits. In vivo, a mixed indigenous formulation (MProT) mitigated Escherichia coli-induced diarrhea in cats, improving fecal scores and selectively lowering pro-inflammatory cytokines relative to an antibiotic regimen and a commercial probiotic. Together with FelMGDB and the enterotype classifier, these results provide actionable strain targets and tools for advancing feline, host-adapted probiotics.

Methods

Sample collection

Animal experiments conform to the ethical norms of animal experiments and the relevant provisions of animal welfare. This study was approved by the Animal Care and Use Committee of Foshan University (FOSU2019029909).

The central portion of recently freshly voided domestic cat feces was collected using a cotton swab immediately following defecation. The feces were then placed in 2 mL tubes. Fecal samples were stored on dry ice and then transferred to laboratory to store at −80 °C until DNA extraction was performed. Fecal samples utilized for culturomics were collected in Cary-Blair enteric transport medium and promptly transferred to the laboratory for culturing.

DNA extraction, library preparation, and sequencing

The initial process involved the extraction of DNA from fecal samples employing the QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany) as per the prescribed manufacturer protocols. DNA concentration was quantified with NanoDrop (Thermo Fisher Scientific, Waltham, MA, USA).

The DNA’s quality and integrity were assessed by electrophoresis on 1% agarose gels. The Qubit® DNA Assay Kit and a Qubit® 3.0 Fluorometer (Invitrogen, China) were utilized to quantify the DNA concentration. We prepared the library using the NEB Next® Ultra™ DNA Library Prep Kit for Illumina (NEB, USA) following the manufacturer’s instructions. The prepared DNA libraries, meeting the required quality standards, were sequenced on the Illumina NovaSeq 6000 (PE150) platform.

Quality control and MAGs reconstruction

The Kneaddata pipeline v0.7.2 was employed for the purpose of filtering out low-quality reads and eliminating host-contamination. The parameters utilized in Kneaddata consisted of a 4-bp window for assessing low-quality reads, an average Q score cutoff of 30, and a minimum read length of 60 bp. Additionally, the Bowtie2 software⁵⁰ was utilized to detect and remove host contamination specifically originating from reads associated with the domestic cat genome (RefSeq: GCF_018350175.1).

After achieving a set of clean reads, we grouped these samples by their breeds and housing method to perform multiple co-assembly using the megahit v1.2.9⁵¹. We selected megahit for the co-assembly due to its remarkable memory efficiency, which accommodates larger sample sizes in the co-assembly process. Separate assemblies for each group were executed using the SPAdes 3.15.5⁵². Subsequently, the assembled scaffolds longer than 2000 bp were subjected to genome binning using MetaBAT2 v2.14⁵³. Bins that were longer than 1 M bp were chosen for further quality check. Redundant bins were eliminated using dRep v3.3.0⁵⁴ with the parameters at “-sa 0.99 -nc 0.3.” To generate species and genus level clusters, we grouped the MAGs using dRep v3.3.0, applying 95% ANI for species and 85% ANI for genus clustering.

We then utilized CheckM (lineage_wf)⁵⁵ to assess the final quality of the bins, taking into account completeness and contamination, retaining only those MAGs with a completeness level greater than 70% and a contamination level below 5%. The tRNA and rRNA genes of MAGs were annotated using Barrnap and tRNAscan-SE⁵⁶, respectively. Finally, we inferred the taxonomic assignment of these MAGs using GTDB-TK v2.4.0⁵⁷, with reference to the GTDB (R09-RS220, July 7, 2024).

Prediction of probiotics from MAGs

The unique MAGs with medium or high quality were conducted probiotics prediction using the pipeline of metaProbiotics³⁸, which was developed for mining of probiotic candidates from assembled MAGs using a language model. The MAGs assigned a score higher than 0.5 by metaProbiotics would be considered as potential probiotics.

Enterotyping analysis

Gut microbiome enterotypes were identified through genus-level relative abundance data underwent quality control by replacing missing values with zeros and removing taxa/samples with zero total abundance. Taxa were retained if present in ≥10% of samples (prevalence filter) and collectively accounting for >0.5% of total sequencing reads. Hellinger-transformed data were subjected to β-diversity analysis using Bray-Curtis dissimilarity matrices (vegdist, vegan v2.6-4). Optimal enterotype clusters (k = 2–5) were determined via partitioning around medoids (PAM) algorithm, with cluster stability assessed by silhouette width maximization. The final model used k = 2 to assign samples to enterotypes.

Culture media and bacterial isolation

A total of 43 cultural media were employed in the study. Modifications were made to the culture media, such as the addition of extra reagents and the utilization of both aerobic and anaerobic methods for culturing. Further details regarding the culture media can be found in Supplementary Table 3.

Fecal samples were obtained from 10 healthy cats (5 males and 5 females), all of whom had not received any antibiotic or probiotic treatments within the preceding six months. Following collection, each sample underwent vortexing in sterile phosphate-buffered saline (PBS) to achieve bacterial suspensions. The resulting supernatant was then diluted to varying working concentrations for subsequent culturing procedures. In each culturing instance, 100 µl of the bacterial suspension was spread onto an agar plate and incubated under proper conditions at 37 °C (Supplementary Table 3). Anaerobic cultures were incubated in an anaerobic chamber under an atmosphere composed of 85% N₂/10% H₂/5% CO₂.

Colonies were picked and streaked for isolation on fresh agar plates to obtain pure cultures. DNA of each picked colony was extracted using the TIANcombi DNALyse&Det PCR Kit (TIANGEN BIOTECH, Beijing, China) with default instruction. Polymerase chain reaction (PCR) amplification was subsequently performed, in accordance with the manufacturer’s instructions. The PCR primers used for the amplification of the 16S rRNA genes were 27F: 5′- AGAGTTTGATCCTGGCTCAG-3′ and 1492 R: 5′-GGTTACCTTGTTACGACTT-3′. The amplified 16S rRNA gene sequences were subsequently sequenced using the Sanger sequencing by the Tsingke Biotech Co., Ltd (Beijing, China).

The 16S rRNA gene sequences obtained from the Sanger sequencer were processed using the DNASTAR Seqman software program to trim the ends with quality stringency of high. Additionally, we conducted a manual verification to confirm the accuracy of base calling. Sequences shorter than 600 bp were excluded from further analysis. To eliminate duplicate colonies, we initially pre-clustered them using a 99.5% similarity threshold with USEARCH software (version 11.0.667)⁵⁸. Subsequently, we aligned the 16S rRNA genes to the NCBI/RefSeq database (Access 08 May 2024) for taxonomy assignment. Colonies that aligned with an existing species with a similarity of greater than 98.7% were categorized as known species. In contrast, those with less than 95% similarity to any reference species were considered indicative of a potential new genus. Further, 16S rRNA gene sequences identified as candidate novel species were re-clustered using a 98.7% similarity threshold with USEARCH software to ascertain the number of novel species.

In vitro biological assays

Growth curves were determined for selected bacterial strains using 40 mL of culture medium supplemented with 200 µL of bacterial suspension. Anaerobic bacteria were sealed with liquid paraffin. Optical density (OD) measurements were taken at selected time points until a stable growth phase was observed. OD values between 0.2 and 0.8 at 600 nm are directly proportional to cell concentration. OD values greater than 1 at 600 nm may indicate cell death, and thus, may not accurately reflect viable cell counts.

Concurrent with the growth curve analysis, the pH of the bacterial suspension was measured using a pH meter, and a pH curve was plotted.

The pH levels of the MRS and LB media were adjusted to 2.0, 3.0, and 4.0 using a 1 M HCl solution. Following this adjustment, cultures were incubated at 37 °C for a period of 12 h. Optical density measurements were taken at both the beginning and end of the incubation period and compared to untreated control samples.

Bile salt tolerance was assessed by adding 0.1%, 0.3%, and 0.5% (w/v) bovine bile salts to MRS and LB media. After thorough agitation to ensure complete dissolution, cultures were incubated at 37 °C for 12 h. OD measurements at 0 and 12 h were conducted, like the acid tolerance test, to determine bile salt tolerance.

E. coli suspension was inoculated onto LB agar cooled to about 40 °C at a ratio of 1:10000. Sterile Oxford cups (6 mm inner diameter, 8 mm outer diameter) were evenly placed on the surface of the agar according to markings. After slowly inverting to mix the medium, it was poured into Petri dishes and allowed to solidify and dry. After drying, the Oxford cups were removed, and the plates were marked. Then, 100 µL of the fermentation liquid from the test strains was added. Plates were incubated at 37 °C.

The presence and size of inhibition zones were observed. Larger clear zones indicate higher sensitivity to the antibiotic; smaller zones indicate lower sensitivity. The diameters of the inhibition zones were measured with a millimeter ruler and classified as sensitive (diameter ≥15 mm, S), intermediate (15 mm > diameter ≥ 10 mm, I), and resistant (diameter <10 mm, R).

Toxicity assay lactate dehydrogenase (LDH)

The cellular cytotoxicity was assessed by the LDH in-vitro cytotoxicity assay (TOX7, Sigma-Aldrich, St. Louis, MO, USA). The study included 5 treatment groups: (1) Empty wells with only culture medium for absorbance baseline. (2) Untreated cells in wells at 37 °C for baseline cellular activity. (3) Cells treated with probiotics to assess impact on cellular integrity. (4) cells exposed to LPS to induce inflammation and toxicity. (5) cells treated with both LPS and probiotics to study protective effects.

After treatments, LDH detection reagent was added to all wells and mixed thoroughly. Plates were then incubated in the dark at room temperature for 30 min before measuring absorbance at 490 nm with a spectrophotometer to calculate cell cytotoxicity or LDH enzyme activity.

$$LDHActivity(\%)=\left(\frac{O{D}_{\mathrm{treated}}-O{D}_{\mathrm{control}}}{O{D}_{\mathrm{maxenzyme}\mathit{activity}}-O{D}_{\mathrm{control}}}\right)\times 100$$

Where $O{D}_{\mathrm{treated}}$ is the optical density (OD) of the treated sample, $O{D}_{\mathrm{control}}$ is the OD of the control cell well, and $O{D}_{\mathrm{maxenzyme}\mathrm{activity}}$ represents the maximum enzyme activity observed under lysed cell conditions. This calculation provides a normalized measure of LDH release, indicative of cell membrane integrity and cytotoxicity.

Cytokine gene expression by qRT-PCR

RNA was extracted from the cells using the Takara RNAiso Plus Kit (Japan), according to the manufacturer’s protocol. Reverse transcription of total RNA was performed using M-MLV Reverse Transcriptase. The primers were synthesized by Shanghai Sangon Co., Ltd. qRT-PCR analysis was done on the ABI 7500 system with SYBR green detection. Samples were measured in duplicate and gene expression levels were compared to β-actin using the 2^−ΔΔCt method. See Supplementary Table 4 for primer sequences.

Animal experiments in cats

A total number of 30 adult British Short-hair blue cats, aged 1–2 years and without probiotic or antibiotic treatment in the past 6 months, were pre-fed for 14 days in a room maintained at 25 °C. Each domestic cat was housed in a separate cage. All cats were fed the same commercial cat diet and provided with water ad libitum.

We formulated a mixed probiotic based on results from in vitro experiments. Subsequently, we conducted an animal study to investigate the characteristics of this probiotic. All cats were randomly assigned to the five groups: Blank Control (BC), Infection Control (InfC), Antibiotic Treatment (AntT), Commercial Probiotic Treatment (CProT), and Mixed Probiotic Treatment (MProT). The cats of InfC, AntT, CProT, and MProT groups were exposed to Escherichia coli for 8 days to establish a diarrhea model. The E. coli bacterial powder was encapsulated and orally administered at a dosage of 1 × 10⁹ CFU/kg, twice daily. For the Antibiotic Treatment Group (AntT), antibiotics were administered after E. coli exposure. The amoxicillin-clavulanate potassium (50 mg per tablet) was administered once daily. For the probiotic treatment groups of Commercial Probiotic Treatment Group (CProT) and Mixed Probiotic Treatment Group (MProT), probiotics were administered at a dose of 1 × 10⁹ CFU/kg, once daily.

The cats were weighed on days 0 and 30 of the experiment. Food intake was recorded daily at 19:00, and fecal scores were assessed daily following the Nestle Purina fecal scoring system (Supplementary Table 5). Blood samples (>3 mL) were collected from each cat using the disposable vacuum blood collection tubes and blood collection needles at day 0, day 9, and day 30. The blood was centrifuged, and serum (>500 μL) was extracted for cytokine and immunoglobulin analyses.

The levels of cytokines (IL-2, IL-1β, IL-6, IL-10, IFN-α, and TNF-α) and immunoglobulins (IgA, IgG, and IgM) in cat serum were quantified utilizing Mlbio ELISA kits in accordance with the manufacturer’s instructions (China, Shanghai).

Author contributions

F.D., Y.F., J.Y. and X.Z. contributed to methodology, investigation, and drafting of the original manuscript. F.D. also performed formal analysis and data curation, and prepared the visualizations. J.Y. contributed to validation and formal analysis. Y.G. and X.Z. also contributed to visualization. M.L. contributed to visualization. Y.P. and H.J. contributed to the investigation. L.Z. and J.Z. contributed to writing, review, and editing. F.L., Y.Z., B.D. and S.C. contributed resources. J.D. and J.C. contributed to validation. Y.L. conceived and supervised the study, administered the project, acquired funding, and contributed to writing, review, and editing. All authors read and approved the final manuscript.

Data availability

All sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1237286. Comprehensive metadata supporting these findings are documented in Supplementary Tables S4 and S5. The complete catalog of MAGs is available for download from Figshare (https://doi.org/10.6084/m9.figshare.31841092). The feline Metagenome Database (FelMGDB; https://FelMGDB.bacbase.com) provides integrated access to the MAG collection, associated annotations, sample metadata, and related dataset information.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41522-026-01038-z.