Microbial single-cell transcriptomics links gut microbiota functional states to metabolic changes in male mice

Ziye Xu; Xin Long; Mengdi Song; Sanbao Zhang; Xiaoyue Li; Feifei Lv; Tianyu Zhang; Ting Cao; Yongcheng Wang

doi:10.1038/s41467-026-75546-z

. 2026 Jul 15;17:8682. doi: 10.1038/s41467-026-75546-z

Microbial single-cell transcriptomics links gut microbiota functional states to metabolic changes in male mice

Ziye Xu ^1,^2,^✉,^#, Xin Long ^1,^#, Mengdi Song ¹, Sanbao Zhang ¹, Xiaoyue Li ¹, Feifei Lv ¹, Tianyu Zhang ³, Ting Cao ^1,², Yongcheng Wang ^1,^✉

PMCID: PMC13490489 PMID: 42457693

Abstract

Increasing recognition that microorganisms within the same community can differ markedly in activity has motivated approaches that measure microbial function at single-cell resolution. However, microbial single-cell transcriptional profiling in mouse models remains limited. Here we show that the microbial single-cell RNA-seq platform smRandom-seq can be adapted to intestinal contents from male diabetic (db/db) and male control mice to profile microbial single-cell transcriptomes across the cecum, colon, and rectum. Using the species-identification workflow smClassify, together with an analysis strategy that integrates microbial transcriptomes with metabolomic profiles, we obtain functionally annotated single-microbe transcriptomes and characterize region- and phenotype-associated metabolic alterations. We also observe cross-species functional patterns that are associated with diabetes-related metabolic changes. Within-species analysis shows region-dependent transcriptional changes in carbohydrate and nitrogen pathways in Muribaculum gordoncarteri. This framework offers a practical approach for resolving microbial functional heterogeneity in the mouse gut and provides a basis for linking such heterogeneity to host metabolic changes, enabling the investigation of how single-microbe transcriptional states interface with host metabolism under diverse physiological and metabolic perturbations.

Subject terms: RNA sequencing, Microbiome

In this study, the authors adapt microbial single-cell RNA sequencing to profile individual gut microbes in male diabetic and control mice, showing how microbial activity varies across gut regions and with diabetes-linked metabolic changes.

Introduction

The gut hosts trillions of microorganisms collectively known as the gut microbiota¹, which play key roles in metabolism, immunity, and overall health^2–6. Previous meta-omics, imaging, and spatially resolved studies have shown that gut microbial communities exhibit pronounced spatial organization along the gastrointestinal tract and across ecological niches such as luminal and mucus-associated compartments^7,8. As recognition of microbial heterogeneity and its disease relevance has grown⁹, the limitations of bulk-level profiling, often limited in practice in resolving intra-species functional diversity, have become increasingly apparent^10,11. To help address this gap, single-cell approaches offer the potential to resolve microbial functional heterogeneity and ecological interactions at higher resolution.

Microbial single-cell RNA sequencing (scRNA-seq) enables transcriptional profiling of thousands of bacteria at single-cell resolution. Several methods, including PETRI-seq¹², microSPLiT¹³, ProBac-seq¹⁴, BacDrop¹⁵, and M3-seq¹⁶, have been developed based on distinct strategies for prokaryotic RNA capture and single-cell barcoding. These methods have demonstrated feasibility in bacterial isolates and synthetic communities (e.g., Escherichia coli and Staphylococcus aureus mixtures¹²) and have revealed intra-species heterogeneity, such as bet-hedging subpopulations in E. coli¹⁶. Our group previously developed smRandom-seq, a random-primer-based, high-throughput microbial scRNA-seq approach, which showed comparable performance in isolates and defined communities^17,18, and was subsequently extended to uncultured bacteria from human feces^19,20 and rumen fluid²¹. However, microbial communities vary substantially across environments in taxonomic composition and physicochemical properties, and the applicability of microbial single-cell transcriptomic methods to other complex ecosystems remains incompletely characterized. In particular, gut luminal communities present additional challenges, such as oxygen sensitivity, requiring further methodological evaluation and adaptation.

Extending microbial scRNA-seq from fecal profiling to region-resolved mouse gut ecosystems is timely and impactful. Human gut microbiome studies have primarily relied on fecal samples, which are convenient but do not capture location-dependent differences in microbial communities and metabolites along the gut^22–24. Mouse intestinal luminal contents, therefore, provide a practical stand-in for region-resolved analyses^25,26. Moreover, the db/db mouse model of type 2 diabetes (T2D) is widely used to investigate host-microbiome interactions in metabolic disease^27,28 and to validate therapeutic strategies^29,30. T2D is associated with well-characterized alterations in gut microbial metabolism, including loss of short-chain fatty acid producers, expansion of facultative anaerobes, and disturbances in amino-acid and bile-acid metabolism. While bulk multi-omics can capture these community-level changes, they do not resolve whether such alterations arise from uniform shifts across microbes or from changes in distinct functional cell states^10,11. These features make diabetic mouse models a relevant system for applying microbial single-cell RNA sequencing to investigate functional heterogeneity of the intestinal microbiota across gut regions.

Recent advances in mouse gut reference genomes and analysis tools have expanded available resources. New genome compendia for the mouse gut microbiome, including the Mouse Gastrointestinal Bacteria Catalogue (MGBC)³¹, the Comprehensive Mouse Microbiota Genome (CMMG) catalogue³², the Mouse Reference Gut Microbiome (MRGM)³³, and MGnify Genomes mouse gut catalogue v1.0³⁴, improve both taxonomic and functional resolution. However, species assignment in single-cell transcriptomic data still depends on how complete and well-annotated these references are, and downstream functional interpretation requires reliable functional annotations. Compared with eukaryotic transcriptomes, current mouse gut microbial reference catalogs and annotations are still incomplete^35,36. Moreover, resources developed for meta-omics are not yet directly suited to mouse microbial single-cell transcriptomics.

In this work, we apply an experimental and analytical workflow to profile microbial single-cell RNA-seq data from the mouse gut microbiome using currently available references. We optimize smRandom-seq for murine luminal contents and develop smClassify to generate species-level, functionally annotated single-microbe transcriptomes. At single-cell resolution, we show that microbial transcriptional programs are organized into functional states rather than species-level averages, and that community-level differences across host phenotypes and gut regions arise from shifts in these states. By integrating single-cell transcriptomics with untargeted metabolomics, we further link metabolite-associated pathway alterations to microbial transcriptional organization. Across the community, microbial functions exhibit coordinated yet complementary activity across species, alongside heterogeneity within species. In particular, we resolve spatially structured transcriptional heterogeneity within Muribaculum gordoncarteri, illustrating how distinct subpopulations contribute to region-specific functional programs. Together, these results provide a framework for understanding microbial functional heterogeneity and how community-level metabolic changes arise from coordinated transcriptional states across and within species.

Results

Study design and integrated microbial single-cell transcriptomic-metabolomic analysis framework

To investigate region-specific microbial activities in the distal gut, we established an integrated framework combining microbial single-cell RNA sequencing from db/db and WT mice (n = 2 per group) with untargeted metabolomics from a separate set of db/db and WT mice within the same experimental model (n = 6 per group) (Fig. 1, Suppl. Tables 1, 2).

To adapt smRandom-seq, which has been extensively validated for performance^17,18, to murine luminal contents, we implemented an optimized sample collection and preprocessing procedure. As shown in Fig. 1a, anatomically defined intestinal segments were dissected²⁶ and immediately fixed by full submersion in 4% paraformaldehyde (PFA), followed by in-tube longitudinal opening and gentle scraping of luminal contents while the tissue was kept fully submerged. This procedure minimized handling time and oxygen exposure, helping to preserve native transcriptional states. In smRandom-seq workflow, RNA capture and barcoding are performed in situ within fixed microbial cells, without requiring cell lysis prior to transcript capture. This design is compatible with rapid PFA fixation and supports preservation of transcriptional states in complex luminal samples.

Single-microbe barcoded libraries were generated using the smRandom-seq workflow^17,18 (Fig. 1b). Microbial cells were enriched by two-step centrifugation (sequential low-g and high-g), washed, and filtered to remove residual debris and aggregates. Cells were then permeabilized and subjected to combined lysozyme-lysostaphin digestion to improve cell-wall permeability across both Gram-positive and Gram-negative bacteria. In situ reverse transcription was performed using random primers to capture total microbial transcripts, which provides an efficient and practical approach for profiling bacterial RNA that lacks poly(A) tails. Following first-strand synthesis, poly(dA) tailing was applied to append adapters to cDNA termini, enabling second-strand synthesis and library construction. Single microbial cells were subsequently co-encapsulated with barcoded hydrogel beads in a droplet-based microfluidic system, enabling automated and high-throughput transcript barcoding at single-cell resolution. The resulting cDNA libraries were recovered, amplified, and sequenced.

Sequencing reads were processed using the smClassify pipeline to generate species-level single-microbe expression matrices for downstream analyses (Fig. 1c). To determine which reference resources best support taxonomic resolution in our dataset, three reference sets, Prokaryotic RefSeq (p-RefSeq)³⁷, MGBC³¹, and MGnify Mouse Gut v1.0 (m-MGnify)³⁴ were benchmarked. Based on this evaluation, we adopted a unified reference strategy, using m-MGnify for genome alignment with p-RefSeq-based taxonomy refinement, which was implemented in the smClassify workflow. The smClassify workflow, guided by reference-based annotation, performs preprocessing (including host RNA removal and exclusion of rRNA-derived reads), followed by alignment, gene annotation, and barcode-level species assignment using threshold-based scoring, providing both taxonomic annotations and functional expression profiles for downstream analyses.

In parallel, serum and luminal metabolites from a separate set of db/db and WT mice within the same experimental model were profiled by LC-MS/MS and annotated using the Kyoto Encyclopedia of Genes and Genomes (KEGG)³⁸. Microbial single-cell transcriptomes were clustered into functional clusters spanning multiple species and experimental conditions, and pathway activities were quantified at the level of these clusters and integrated with metabolite abundances through shared biochemical pathways, enabling systematic cross-omics alignment (Fig. 1d). This framework enables the analysis of microbial transcriptional activity at single-cell resolution and facilitates the integration of region-specific microbial activity with host metabolic changes.

Reference genome choice determines the taxonomic resolution achievable in microbial scRNA-seq

Accurate taxonomic classification in microbial single-cell RNA-seq critically depends on the choice of reference genomes. Three reference sets p-RefSeq³⁷, MGBC³¹, and m-MGnify³⁴ were evaluated (Fig. 1c). These databases differ in size, content, and host specificity (Suppl. Table 3). p-RefSeq was the largest (> 300,000 genomes across approximately 4000 genera and 20,000 species) but contained many incomplete genomes (Fig. 2a). MGBC comprised near-complete genomes (median completeness ~100%) but lacked functional annotation. m-MGnify provided the broadest functional coverage (82.6% annotated), with extensive assignments across COG, KEGG, GO, and CAZy databases (Fig. 2b). KEGG annotations further showed that most coding sequences in m-MGnify were associated with core metabolic pathways, predominantly contributed by Bacillota_A and Bacteroidota (Suppl. Fig. 1a, b). Using a previously described Kraken2-based classifier¹⁹, m-MGnify yielded the highest read-assignment rates of total reads at both the genus and species levels (Fig. 2c). As expected, p-RefSeq captured the widest taxonomic breadth (Fig. 2d). Overlap at the genus and species levels among the three resources was limited, and p-RefSeq classifications included taxa that are uncommon in the mouse gut (Suppl. Fig. 2a–c). Relative-abundance structures also differed, which may partly reflect that many m-MGnify MAGs lack rRNA features, potentially biasing assignments for rRNA-rich taxa. To quantify concordance between genome catalogs, we computed average nucleotide identity (ANI) between p-RefSeq genomes and m-MGnify MAGs using FastANI³⁹ (Suppl. Fig. 3a). Despite differences in species naming between the two databases (Suppl. Fig. 3b), 89.2% of RefSeq-based assignments of our samples corresponded to genomes with a high-ANI match (> 95%) to m-MGnify MAGs (Suppl. Fig. 3c), indicating substantial concordance between the two references. These results suggest that m-MGnify provides strong ecological relevance and functional annotation, whereas p-RefSeq offers more standardized species nomenclature, highlighting the challenge of achieving both ecological relevance and consistent taxonomy using a single reference resource. Therefore, we developed smClassify, a reference-guided workflow in which species assignment is primarily based on alignments of non-rRNA reads to the m-MGnify genome catalog, with species names subsequently refined using p-RefSeq to ensure consistent taxonomy (Fig. 1c).

Fig. 2 — a Taxonomic breadth, genome number and genome completeness of Prokaryotic RefSeq (p-RefSeq), Mouse Gastrointestinal Bacterial Catalogue (MGBC) and MGnify Mouse Gut v1.0 (m-MGnify). Genome completeness was calculated per genome; p-RefSeq, n = 21,247 genomes; MGBC, n = 26,640 genomes; m-MGnify, n = 112,951 genomes. b Functional database mapping. Vertical bars and dot plots indicate overlapping functional annotations, and horizontal bars show total annotated genes per database. c, d Classification performance across references, including overall assignment rate, genus/species-level assignment (c), and recovered genus/species numbers (d). Each point represents one gut-region sample; n = 12 samples per reference, derived from 4 biologically independent mice. e Community composition at family, genus, and species levels across cecum, colon, and rectum from WT and *db/db* mice. f Overlap of detected taxa between smClassify and Kraken2 applied to the same microbial single-cell RNA-seq dataset. g Genus-level abundance correlations between smClassify and Kraken2 across WT gut regions; dashed lines indicate y = x and solid lines indicate linear fits. Pearson’s r and Spearman’s ρ were calculated using log10-transformed abundances. h Bray-Curtis distances between biological replicates based on genus- and species-level compositions derived from Kraken2 or smClassify. n = 6 pairwise comparisons per method for each taxonomic rank. P-values were calculated using two-sided paired Wilcoxon signed-rank tests without multiple-comparison adjustment. i Correlations between microbial single-cell RNA-seq and metagenomic profiles for WT cecum samples using the same MGnify reference. Rows show family, genus and species levels; columns show Kraken2 and smClassify profiles versus metagenome. Abundances were log10-transformed, and correlation coefficients are shown in each panel. j Non-rRNA UMI counts and detected non-rRNA genes per cell after species assignment. Exact cell numbers for each sample are provided in Suppl. Table 7. For box plots in (a, c, d, h, j), center lines indicate medians, box bounds indicate the 25th and 75th percentiles, and whiskers extend to the most extreme values within 1.5× the interquartile range. Points beyond whiskers are plotted as outliers only in (a, j); points in (c, d, h) show all data points.

The smClassify workflow was applied to microbial single-cell transcriptomes from the mouse gut generated by smRandom-seq (Fig. 1a, b). During preprocessing, evaluation of barcode selection thresholds showed a tradeoff between cell recovery and transcriptome quality, leading to the selection of a Top15,000 cutoff (Suppl. Table 4). Sequencing depth and alignment quality were broadly consistent across samples, with rRNA-derived reads comprising a substantial fraction of total unique molecular identifiers (UMIs) (median ~67%), and all downstream analyses performed using coding sequence (CDS)-restricted alignments (Suppl. Fig. 4a; Suppl. Tables 5, 6). Sensitivity analyses of species-calling parameters indicated that species detection and community composition were stable across a range of thresholds (Suppl. Fig. 4b–d). The dataset showed low putative doublet rates, a strongly right-skewed confidence margin distribution, and a high proportion of cells passing species-calling thresholds across samples (Suppl. Fig. 5a–c, Suppl. Table 7), supporting reliable species assignments for downstream analyses. Comparisons of taxonomic assignments showed that conflicts were moderate at the genus level and more frequent at the species level across different methods and reference databases (Suppl. Fig. 5d, e). The harmonized taxonomy generated by smClassify was used for downstream analyses, with 111 genera and 266 species detected at ≥ 1 barcode, decreasing to 49 genera and 77 species at > 100 barcodes (Suppl. Fig. 5f). At the family level, Lachnospiraceae, Bacteroidaceae, and Muribaculaceae dominated, with the largest compositional shifts observed in the colon (Fig. 2e). At the genus level, COE1 (a Lachnospiraceae genus), Bacteroides, and Muribaculum were consistently detected, with abundance patterns that varied between groups. Species-level differences were most apparent for Bacteroides acidifaciens, Bacteroides muris, and Muribaculum gordoncarteri.

We next benchmarked these smClassify results against widely used taxonomic profiling tools, Kraken2 and MetaPhlAn, which were not designed specifically for microbial single-cell analyses (Suppl. Table 8). MetaPhlAn showed substantially lower mapping rates (~7–18%), whereas Kraken2 (~90–94%) and smClassify (~82–88%) showed higher and broadly comparable assignment rates. Comparing Kraken2 results generated using a Kraken2/Bracken-formatted m-MGnify catalog with smClassify results incorporating the p-RefSeq taxonomy refinement, detected taxa overlapped substantially at the genus level (110 shared genera), while differences were more pronounced at the species level (Fig. 2f, Suppl. Fig. 6a, b). Consistently, relative abundance profiles showed good concordance between smClassify and Kraken2 at the genus level (Fig. 2g). Reproducibility analysis based on Bray-Curtis dissimilarity between biological replicates showed that smClassify yielded lower dissimilarity relative to Kraken2 (Fig. 2h). In a defined four-species mock community (Escherichia coli, Staphylococcus aureus, Anaerobutyricum hallii, and Muribaculum gordoncarteri), smClassify recovered all expected species and consistently showed higher recovery rates across taxa compared to Kraken2 (Suppl. Fig. 6c–g). This difference may partly reflect the influence of abundant rRNA-derived reads, as Kraken2 includes conserved rRNA regions with limited species-level resolution, whereas smClassify focuses on CDS-derived features. These results indicate that smClassify provides comparable taxonomic coverage to existing tools, with improved recovery of expected species in the mock community.

We next compared single-cell taxonomic profiles with metagenomic data analyzed using Kraken2 with the MGnify reference database (Suppl. Fig. 7a–c). Species-level support showed partial concordance between the two datasets, with both shared and method-specific abundance patterns observed across taxa (Suppl. Fig. 7d). To account for the influence of highly abundant taxa, concordance was evaluated using both relative and log-transformed abundance, together with Spearman correlation (Suppl. Table 9). Under these metrics, concordance based on log-transformed abundance was higher for smClassify (Pearson r = 0.628, p = 4.54 × 10⁻⁴) than for Kraken2 (Pearson r = 0.541, p = 5.26 × 10⁻³) (Fig. 2i). A similar pattern was observed at the genus level (smClassify: Pearson r = 0.518, p = 2.85 × 10⁻⁷; Kraken2: Pearson r = 0.448, p = 2.20 × 10⁻⁵), and at the species level (smClassify: Pearson r = 0.444, p = 5.74 × 10⁻⁹; Kraken2: Pearson r = 0.237, p = 7.03 × 10⁻⁵). Metagenomic data detected a larger number of taxa than smClassify, particularly at the species level (Suppl. Fig. 7e). For example, Sporofaciens (a Gram-positive genus) was among the top 10 most abundant taxa in metagenomic data but was only minimally detected by smClassify (Suppl. Fig. 7f). These differences likely reflect inherent methodological distinctions between DNA-based metagenomic profiling and transcriptomic measurements.

Consistent with the sensitivity and coverage previously reported for smRandom-seq^17,18, the current dataset showed expected levels of transcript recovery in murine gut microbiota. Following species assignment, we detected a median of ~1000 genes per species across samples after aggregating across cells, representing a substantial fraction of typical bacterial gene repertoires (Suppl. Fig. 5g). smClassify yielded consistent expression metrics across samples, with median values of 150–300 non-rRNA UMIs and 100–200 detected non-rRNA genes per cell (Fig. 2j), comparable to previously reported microbial single-cell transcriptomic datasets^{12–16,19,20,40,41} (Suppl. Table 10).

In this study, we adopted a unified reference strategy within smClassify, using m-MGnify for genome alignment with p-RefSeq-based taxonomy refinement, thereby providing a harmonized framework for species-level transcriptional profiling of microbial communities, facilitating downstream analyses of functional heterogeneity at the single-cell level.

Functional annotation characterizes microbial transcriptional heterogeneity in the mouse gut microbiome

To characterize transcriptional heterogeneity across individual microbes, we excluded structural-RNA-associated genes and hypothetical proteins lacking functional annotation (Suppl. Table 11), retaining 53.5% annotated genes (46.5% hypothetical) and 32.9% of transcripts after all filtering steps (Suppl. Fig. 5h, i). After filtering and cell-level quality control, each sample retained ~6000–7000 cells (~80,000 in total). Cells from all 12 samples formed a continuous manifold in UMAP space, with minimal segregation by sample (mean ≈ −0.13; Suppl. Fig. 8a, b). When colored by host phenotype, cells appeared partially overlapping, with weak separation (mean ≈ 0.03) (Fig. 3a, Suppl. Fig. 8c, d), suggesting that phenotype-associated differences are distributed across transcriptional programs rather than forming discrete clusters. In contrast, cells showed substantial overlap across gut regions (mean ≈ −0.01) (Suppl. Fig. 8e, f). Cells assigned to the top 20 taxa were broadly intermixed in UMAP. Consistently, silhouette analysis indicated minimal segregation by species identity (mean ≈ −0.12) (Suppl. Fig. 8g, h), suggesting that individual species span multiple transcriptional states rather than a single averaged functional profile. Nevertheless, several taxa showed non-random enrichment in specific clusters. For instance, Muribaculum gordoncarteri was preferentially localized to cluster 6.

Unsupervised graph-based clustering identified eight major clusters, with modest separation among clusters (mean silhouette ≈ 0.22) (Suppl. Fig. 9a, b). These clusters represent distinct functional transcriptional states rather than taxonomic groupings. Cluster-specific marker genes further delineated characteristic transcriptional programs (Suppl. Fig. 9c), and pathway enrichment of these markers revealed distinct metabolic programs across clusters (Fig. 3b). Based on these profiles, clusters were annotated with dominant programs, including motile chemotactic and stress-adaptive (cluster 0), butyrate metabolism (cluster 1), polysaccharide uptake (cluster 2), organic acid metabolism (cluster 3), anaerobic energy metabolism (cluster 4), carbohydrate degradation (cluster 5), polysaccharide degradation (cluster 6), and oxidative stress responses (cluster 7) (Fig. 3c). These annotations were stable across downsampling thresholds, with comparable cluster structures and high retention of functional assignments (Suppl. Fig. 10a–e). All clusters were detected across region-phenotype groups, but their relative abundances varied across groups (Fig. 3d). db/db samples tended to contribute a larger proportion of cells to the motile chemotactic and stress-adaptive cluster (cluster 0), whereas WT samples were relatively enriched for the organic acid metabolism (cluster 3) and polysaccharide degradation (cluster 6) programs. Across regions, oxidative stress response cells (cluster 7) were most frequent in rectal samples, whereas butyrate metabolism cells (cluster 1) were more common in the cecum. Consistently, weighted KEGG scores, calculated for each group by combining their cluster composition with cluster-specific pathway enrichments, supported phenotype- and region-associated transcriptional shifts within this dataset (Fig. 3e). For example, pyruvate metabolism, a butyrate-metabolism-linked pathway, scored highest in cecum-WT and cecum-DB, whereas galactose metabolism, associated with organic acid metabolism, was more strongly represented in colon-WT and rectum-WT. These pathway-level differences reflect shifts in the relative abundance of functional cell states rather than uniform transcriptional changes across microbes.

Species-level profiles showed distinct community compositions across functional clusters in WT versus db/db mice (Fig. 3f). WT samples were dominated by Muribaculum (notably M. intestinale and M. gordoncarteri) and Bacteroides (for example, B. acidifaciens) in polysaccharide and carbohydrate/organic acid metabolism, whereas db/db samples contained higher proportions of Parabacteroides distasonis and taxa linked to stress- or motility-related clusters. Uncultured Lachnospiraceae/COE1-like taxa that could not be resolved to species level were detected in several clusters, but were most concentrated in cluster 0. Given the relative enrichment of cluster 0 in db/db mice, we further subdivided cluster 0 cells into five functional subgroups based on subcluster-specific marker genes (Fig. 3g, h, Suppl. Fig. 11a–c). Gene-set analysis using curated modules related to protective functions^42,43, risk-associated traits^44–46, and SCFA context/flux^47,48 further differentiated the five subgroups and highlighted distinct functional specializations (Fig. 3i–k, Suppl. Fig. 11d–f). Based on these signatures, Sub-0 was the subgroup most strongly associated with the db/db-enriched pattern within cluster 0, with Sub-1 showing a weaker association.

Together, single-cell-resolved microbial transcriptomes support the identification of cluster-level and subcluster-level functional states of individual microbial cells, rather than species-level averages. These analyses show that community-level functional differences across host phenotypes and gut regions are driven by shifts in the abundance of these states, which can be traced to their dominant contributing species and further resolved into phenotype-associated subpopulations within each state.

Cross-omics integration identifies metabolic changes in the mouse gut microbiome

To link microbial transcriptional changes with metabolic shifts, we profiled luminal (cecum, colon, rectum) and serum metabolites from a separate set of WT and db/db mice within the same experimental model. Intermediates of carbohydrate metabolism, including sugar phosphates (glucose-6-phosphate, mannose-6-phosphate, and fucose-1-phosphate) and oligosaccharides (stachyose, raffinose, maltotriose, and D-maltose), were elevated across intestinal segments (Suppl. Fig. 12a, b). Although glucose itself was not detected, these related metabolites showed consistent differences between genotypes, supporting altered carbohydrate-associated metabolic states under diabetic conditions. Consistently, unsupervised PCA and supervised PLS-DA showed that samples primarily separated by genotype, whereas separation by intestinal region was less pronounced (Suppl. Fig. 13a–f), indicating that disease status is a major contributor to variation in the gut luminal metabolome.

We next examined these differences at the level of individual metabolites. Volcano plots identified numerous differential metabolites (Fig. 4a), including increased succinic acid and taurodeoxycholic acid in db/db mice, as well as higher serum pyruvate, cholic acid, and branched-chain amino acids (leucine and isoleucine), consistent with prior studies^49–52. Hierarchical clustering of differential metabolites revealed both shared and region-specific metabolic changes across the cecum, colon, and rectum when samples were organized by intestinal segment and genotype (Suppl. Fig. 13g), and unsupervised clustering showed that samples grouped more strongly by genotype than by tissue type (Suppl. Fig. 13h). UpSet analysis showed substantial overlap of differential metabolites between rectum-cecum and rectum-colon, with more distinct profiles relative to serum (Suppl. Fig. 13i). At the class level, coordinated alterations were observed across multiple metabolite categories, including fatty acyls, prenol lipids, organooxygen compounds, and carboxylic acid derivatives (Suppl. Fig. 13j), consistent with metabolic changes associated with insulin resistance and diabetes^53–56.

KEGG pathway enrichment analysis of differential metabolites revealed coordinated alterations across amino acid, carbohydrate, energy, lipid, and bile secretion pathways (Fig. 4b), supporting pathway-level metabolic changes in db/db mice. To directly compare transcriptional and metabolic layers, we then mapped the annotated differentially expressed microbial transcripts and differential metabolites onto shared pathways. This cross-layer network analysis revealed substantial overlap between metabolite-derived pathways and microbial transcriptional programs, with serum profiles preferentially linked to polysaccharide uptake and oxidative-stress modules, and luminal profiles linked to carbohydrate and organic-acid metabolism (Fig. 4c).

To systematically evaluate metabolite classes implicated in T2D, we examined representative metabolites across amino acids, organic acids, fatty acids, and bile acids (Suppl. Fig. 12c–m), providing an overview across intestinal regions and serum. Consistent with pathway-level signals and the role of amino acid metabolism in diabetes^57,58, amino-acid-related metabolites showed coordinated and tissue-specific alterations across the cecum, colon, rectum, and serum (Fig. 4d), and unsupervised clustering further showed that tissue type defines the divergent profiles of the cecum and serum, whereas genotype-driven shifts predominate in the colon and rectum (Suppl. Fig. 14a). The glutamine/glutamic acid (Gln/Glu) ratio was significantly elevated in both the cecum and rectum of db/db mice, whereas the N-acetylglutamate/glutamate (NAG/Glu) ratio increased in the cecum and colon (P < 0.05, Fig. 4e). Upstream intermediates in nitrogen flow^59–61, including glutamine and citrulline, accumulated specifically in the cecum (Suppl. Fig. 12c, d), consistent with region-specific disruption of nitrogen handling. Luminal Homo-L-arginine declined, whereas serum DL-arginine rose (Suppl. Fig. 12e), pointing to a mismatch between local and systemic arginine-nitrogen handling. Elevated serum pyruvate and succinate (Fig. 4a), together with shifted levels of aromatic amino acids and related organic acids (Suppl. Fig. 12f, g), further support a metabolic bottleneck at the carbon-nitrogen entry point in db/db mice. At the transcriptional level, colonic microbes of db/db mice displayed lower expression of the glutamate-glutamine assimilation axis (gltD, gltB, gdhA, and glnA) (Suppl. Fig. 14b), with these genes mainly expressed by M. gordoncarteri and Bacteroides muris (Fig. 4f). By contrast, argF and argH were only mildly induced in the cecum (Suppl. Fig. 14b), mainly in Eubacterium_J and Parabacteroides (Fig. 4f), suggesting limited compensation through the arginine-ornithine branch. Organic acid-linked genes encoding pyruvate: ferredoxin oxidoreductase (PFOR, MGYG000321881-01184/-01489), glutaminase (glsA1), and aspartate aminotransferase/β-decarboxylase (asD), predominantly assigned to Duncaniella muricolitica and Duncaniella muris (Fig. 4f), showed reduced expression in the colon of db/db mice, with highest expression in the rectum and minimal expression in the cecum (Suppl. Fig. 14c). This regional pattern suggests reduced microbial conversion of pyruvate to acetyl-CoA and limited Gln/Glu flux in the colon, consistent with pyruvate/succinate accumulation and a distal shift of carbon-nitrogen coupling. Consistent with this microbial pattern, publicly available host single-cell data from WT mice⁶² showed proximal enrichment of nitrogen-shunting genes (Got1/2, Gpt, and Arg1/2) and distal Arg-NO metabolism genes (Gls, Glul, and Nos2) (Suppl. Fig. 15a–c), supporting spatial carbon-nitrogen uncoupling along the gut.

Given that both our metabolomic data and prior work in diabetes implicate long-chain fatty acids (LCFAs)^57,63 and bile acids (BAs)⁶⁴ as key metabolic nodes, we next examined LCFA and BA metabolism. Even-chain fatty acids (C16:0, C16:1, and C18:1) increased, whereas odd-chain fatty acids (C15:0 and C17:0) were decreased in db/db mice (Fig. 4g, Suppl. Fig. 12h), a pattern consistent with enhanced acetyl-CoA-driven lipid elongation and reduced propionyl-CoA-initiated synthesis under diabetic conditions^65–67. Consistent with altered microbial lipid metabolism, colonic fadH and fadF were primarily expressed by M. gordoncarteri (Fig. 4f) and displayed lower expression in db/db colonic samples (Suppl. Fig. 14d). In the host⁶², β-oxidation genes (Acsl1/3/5 and Cpt1a) were enriched proximally in the cecum, whereas fatty-acid signaling genes (Ptgs2 and Pla2g4a) predominated in the colon (Suppl. Fig. 15d, e). Turning to bile acid metabolism, conjugated bile acids (taurodeoxycholic acid and taurochenodeoxycholic acid) increased in both colon and serum (Fig. 4h, Suppl. Fig. 12j–m), accompanied by region-specific microbial gene changes. The bile salt hydrolase cbh⁶⁸ showed lower expression in colonic and rectal microbiomes (Fig. 4i) and was largely contributed by Cryptobacteroides sp. 910585445 and M. gordoncarteri (Fig. 4f). ompR exhibited reduced expression distally, whereas efflux pumps (acrA and mdtK) displayed higher expression proximally. In the host⁶², bile-acid transport and signaling machinery also displayed clear regional segregation (Fig. 4j). Colonic enterocytes showed higher expression of bile-acid transport components (Fabp6, Slc10a2, Slc51a/b), as well as FXR-FGF15 axis component Nr1h4 and TGR5-GLP-1 pathway genes (Pcsk1 and Gcg), whereas cecal enterocytes showed relatively higher expression of the FXR-FGF15 co-receptor pair (Fgfr4 and Klb) and TGR5-GLP-1 pathway components (Glp1r, Ins2).

Taken together, although the metabolic differences between groups and pathway-level alterations observed here are broadly consistent with prior studies of diabetes^49–56, our cross-omics analyses further suggest that these changes are associated with distinct microbial transcriptional states at single-cell resolution. By mapping metabolite-enriched pathways to functional clusters of gut microbes, and further to specific species and microbial gene expression programs, we relate these metabolic alterations to microbial transcriptional organization at the level of individual microbial cells.

Community-embedded functional states highlight cross-species functional complementation within the gut microbiota

Given the amino-acid-centered metabolic changes and associated lipid and bile-acid shifts identified in Fig. 4, we next examined how these functions are distributed across microbial species. We aggregated scRNA-seq cells into per-sample species abundance profiles and constructed a species-by-sample correlation matrix, which revealed coherent blocks of positively correlated species together with anticorrelated groups (Fig. 5a). Guided by the metabolic pathways highlighted in Fig. 4 and functional modules implicated in diabetes-associated microbiome alterations^69–71, gene-set scoring across the top 20 species showed uneven yet complementary activity patterns (Fig. 5b, Suppl. Fig. 16a). For example, Bacteroidales taxa, including D. muris, M. gordoncarteri, B. muris, Alloprevotella sp002933955, and Cryptobacteroides sp910585445, were enriched for carbohydrate metabolism modules plus bile-acid transport and stress responses.

Fig. 5 — a Spearman correlation heatmap of the top 20 species based on per-sample cell counts. Rows and columns represent species; blue, negative correlation; white, approximately zero; red, positive correlation. b Dot plot of metabolic and stress-response pathway activity across the top 20 species, including SCFA production, LCFA activation and β-oxidation, PUL, mucin degradation, amino acid and nitrogen metabolism, bile acid resistance, and stress/efflux modules. Dot size indicates the percentage of cells expressing each module, and color indicates average module activity. c, d UMAP embeddings of *D. muris*, *M. gordoncarteri* and *B. muris* cells colored by species (c) or five transcriptionally defined subpopulations (d). e Dot plot of eight T2D/IR-related functional modules across subpopulations. Dot size indicates the percentage of cells expressing each module, and color indicates average module activity. f UMAP embedding of *Duncaniella* cells colored by five transcriptionally defined clusters. g Faceted bar charts showing the fraction of total cells contributed by each *Duncaniella* subpopulation within each tissue-condition group. For each group, the denominator is all cells in that group, and the numerator is cells from the indicated subpopulation; bar height represents the corresponding proportion. Colors denote groups. h, i UMAP embeddings of *Parabacteroides* cells colored by species (h) or four transcriptionally defined clusters (i). j Heatmap of culture-related levers predicted to influence *Parabacteroides* isolation and growth. Columns represent culture levers, and rows represent *Parabacteroides* species. Tile color indicates evidence strength from pathway module scores and gene expression. Symbols indicate predicted relevance: +, favorable; −, unfavorable; ·, neutral. SCFA short-chain fatty acid, LCFA long-chain fatty acid, PUL polysaccharide utilization locus, T2D type 2 diabetes, IR insulin resistance.

To examine how complementary functions are distributed within lineages, we performed joint clustering of the three dominant Bacteroidales taxa (D. muris, M. gordoncarteri, and B. muris), identifying both species-specific clusters and a mixed functional cluster containing cells from all three species (Fig. 5c, Suppl. Fig. 17a). The mixed subcluster was enriched for facultative carbon-utilizer genes (Fig. 5d, Suppl. Fig. 17b), suggesting convergent strategies for nutrient acquisition despite phylogenetic distance. Functional module analysis showed distinct yet overlapping activity patterns (Fig. 5e, Suppl. Fig. 17c). B. muris showed stronger glycan utilization and carbohydrate uptake signatures, M. gordoncarteri favored efflux and acid resistance, and D. muris specialized in stress defense, motility/chemotaxis, and fermentative metabolism. Notably, these modules were active across multiple species, highlighting cross-species functional convergence.

Within Duncaniella, species showed substantial functional divergence (Suppl. Fig. 18a). UMAP clustering resolved four major subpopulations, including broad-spectrum degraders (D. muricolitica), substrate specialists, SCFA producers (D. muris), and mucin/nitrogen utilizers (D. freteri) (Fig. 5f, Suppl. Fig. 18b, c). Tissue- and condition-specific abundance analysis showed reduced representation of most subpopulations in db/db mice, particularly in the colon, whereas substrate specialists persisted (Fig. 5g, Suppl. Fig. 18d), suggesting that streamlined catabolic programs may confer a selective advantage under diabetic conditions.

Parabacteroides displayed a similar functional partitioning. While P. distasonis and Parabacteroides sp910577325 dominated overall abundance, P. goldsteinii specialized in redox and vitamin-related niches (Fig. 5b). Single-cell clustering identified four functional subtypes ranging from polysaccharide degraders to central carbon/redox specialists (Fig. 5h, i, Suppl. Fig. 18e, f). To infer culture conditions for isolating Parabacteroides sp910577325, we compared transcriptome-derived metabolic-requirement gene sets across three Parabacteroides species (Fig. 5j, Suppl. Fig. 18g). For P. sp910577325, transcriptomic profiles were consistent with mucin/GlcNAc and fucose utilization and a B12 requirement, alongside low bile tolerance, poor fumarate use, and largely neutral signals for other sugars and ethanolamine. In comparison, P. goldsteinii favored fumarate-based respiration, while P. distasonis showed broadly negative sugar-use signals and similarly low bile tolerance.

Together, our results suggest that microbial functions are coordinated across species at single-cell resolution. We observe functional convergence across microbes from different lineages alongside divergence within closely related lineages, patterns not captured by species-level abundance analyses. Transcriptome-derived signatures further delineate functional niches and provide insights into culture requirements. These findings provide a framework for resolving community-level functional organization beyond conventional abundance-based profiles.

Subpopulation heterogeneity of M. gordoncarteri links disrupted nitrogen and stress programs to host nitrogen imbalance

Building on the alterations of the glutamate-glutamine metabolic interface (Fig. 4) and the community-level complementation described above (Fig. 5), we next focused on M. gordoncarteri, a dominant Bacteroidetes taxon⁷². M. gordoncarteri mapped primarily to cluster 6 (polysaccharide degradation) and expressed high levels of two glycoside hydrolases, BoGH43B (non-reducing-end α-L-arabinofuranosidase, MGYG000413837-01743) and BoGH31A (α-xylosidase, MGYG000413837-01569) (Fig. 6a). The efflux pumps bepF and bepG also showed strong, species-specific expression, consistent with tolerance to remodeled bile acids and long-chain fatty acids. Subclustering of all cells assigned to M. gordoncarteri resolved three transcriptionally distinct subpopulations within this single species (Fig. 6b, Suppl. Fig. 19a, b). Cells belonging to Subtype 0, enriched for nitrogen and glutamate metabolism, predominated in the rectum and, to a lesser extent, the colon (Fig. 6c). Cells belonging to Subtype 1, characterized by carbohydrate and fatty acid utilization, were most abundant in the cecum, followed by the colon. Cells belonging to Subtype 2, defined by stress-adaptation signatures, were enriched in the WT rectum but depleted in db/db mice. These regional and phenotype-associated patterns reflect shifts in the relative abundance of distinct transcriptional subpopulations. Consistent with these observations, pseudobulk differential expression (DE) analysis further confirmed that several key nitrogen- and amino-acid-related genes showed lower expression in the db/db colon compared to WT, including glutamate/glutamine assimilation (gdhA), asparagine synthesis (asnA), nitrogen transport (nrgA), tryptophan biosynthesis (trpC, trpD, trpE), and histidine biosynthesis (hisG) (Suppl. Fig. 19c). In contrast, no genes reached statistical significance in the cecum or rectum, likely reflecting signal attenuation due to subpopulation averaging.

Pseudotime analysis ordered M. gordoncarteri cells along a spatial trajectory from cecum (early) to colon (intermediate) to rectum (late) (Fig. 6d, Suppl. Fig. 19a). WT cells were more concentrated in intermediate states, whereas db/db cells were more frequently observed toward trajectory ends, consistent with altered transcriptional-state distributions (Fig. 6e), in line with loss of the stress-adapted Subtype 2. Gene dynamics analysis identified four transcriptional modules (Fig. 6f). Glutamate-pathway genes gdhA and glnA in Module 3 defined WT-colon states with active nitrogen assimilation, whereas GABA cycle genes gadB and gadC in Module 2 characterized late db/db branches and were linked to the GABA shunt and malate metabolism (Fig. 6g). Stress defense genes (rbr and sodB) in Module 4 were rectum-WT specific, consistent with oxidative and nitrosative stress defenses. Fatty acid biosynthesis (fabF) was upregulated in db/db. Urease gene (ureG) showed rectal WT enrichment, suggesting altered urea utilization in db/db mice.

Urease profiling showed that M. gordoncarteri uniquely carries the urease operon (ureB-ureG) (Fig. 6h), with the highest expression in Subtype 0 (Suppl. Fig. 19d). Within this species, urease genes showed lower expression in the colon and higher expression in the rectum (Fig. 6i). Pseudobulk DE analysis also showed that ureB-ureG transcripts were reduced in the colon and slightly elevated in the rectum of db/db mice relative to WT (Suppl. Fig. 19e). The colonic reduction coincided with lower N-acetyl-L-glutamic acid and glutamate/D-glutamate, alongside upstream accumulation of glutamine and citrulline (Fig. 4e). Increased rectal urease expression may reflect a distal compensatory response to community-level urease loss in db/db mice. Consistent with coordinated changes in polysaccharide-utilisation and stress-adaptation, polysaccharide utilization locus (PUL) components (susC, susD, and pulA) showed higher expression in the colon of db/db mice, matching region-wide elevations of sugar phosphates and oligosaccharides (Fig. 6i). The glycoside hydrolases BoGH43B (MGYG000413837-01743) and BoGH31A (MGYG000413837-01569) displayed contrasting expression patterns, with BoGH43B higher in the cecum and BoGH31A lower in the colon. The efflux pumps bepF and bepG showed reduced expression in the colon, almost undetectable in the cecum, and bepF exhibited higher expression in the rectum. Metabolomics further showed increased luminal N-acetylputrescine and N-acetylcadaverine together with decreased luminal sinapoylspermine and serum N8-acetylspermidine (Fig. 6j), consistent with accumulation of acetylated early diamines and reduced downstream polyamine elaboration.

Taken together, species-level analyses show that M. gordoncarteri reorganizes into region-specific transcriptional states along the gut axis, shifting from carbohydrate utilization in the cecum to nitrogen-buffering and stress-response programs in the rectum. These intra-species transitions mirror the metabolic alterations identified in Fig. 4 and complement the cross-species patterns in Fig. 5. Importantly, rather than reflecting uniform changes at the species level, these patterns arise from shifts in distinct transcriptional subpopulations within the same species, a feature that is not resolved by bulk approaches. This provides evidence linking intra-species heterogeneity to metabolic remodeling, which defines a hierarchical framework from cell-level functional states to cross-species complementation and intra-species heterogeneity, integrated with region- and phenotype-associated metabolic alterations (Fig. 6k).

Discussion

In this study, we present an integrated framework combining microbial single-cell transcriptomics with luminal and systemic metabolomics to characterize gut microbial functional organization in diabetes. This framework integrates optimized smRandom-seq for gut contents, smClassify for species assignment, and cross-omics mapping of microbial transcripts and metabolites onto shared functional pathways. Using this approach, we show that microbial transcriptional programs are organized into cell-level functional states and that community-level differences across host phenotypes and gut regions arise from shifts in these states. Cross-omics integration further links these states to pathway-level metabolic alterations. At the intra-species level, M. gordoncarteri exhibits transcriptional heterogeneity, with distinct subpopulations underlying functional programs consistent with metabolomic changes.

We generated microbial single-cell transcriptomes across multiple regions of the murine gut, capturing major taxonomic groups. While these results showed concordance with metagenomics at higher taxonomic levels, it must be acknowledged that single-cell transcriptomic profiles should not be interpreted as an exact replicate of the underlying community composition derived from bulk metagenomics. Differences in analytical frameworks between metagenomic and single-cell transcriptomic data, including compositional effects, gene copy number variation and zero inflation, may introduce systematic biases in cross-platform comparisons⁷³. While gene copy number variation is not explicitly modeled, the single-microbe resolution partially reduces species-level confounding compared to bulk metatranscriptomic analyses. In addition, technical aspects of microbial single-cell RNA-seq, such as intact cell isolation, host-material removal, fixation and permeabilization, may introduce taxon-specific capture biases. For example, despite using an optimized cell-wall digestion strategy, Gram-positive taxa such as Sporofaciens, which are abundant in metagenomic profiles, were less efficiently recovered by smRandom-seq. Commensal fungi, increasingly recognized for their roles in gut health and metabolic disease^74,75, often possess highly resistant cell walls that hinder lysis. Similarly, biofilm-forming microorganisms embedded within extracellular matrices and microbes tightly associated with mucus layers or host tissues may be more difficult to dissociate or capture, potentially leading to underrepresentation of host-associated niches. The PFA fixation strategy may introduce bias in microbial recovery, as differences in cell-wall structure and permeability can affect RNA preservation and lysis efficiency. Nevertheless, validation in smRandom-seq^17,18 and related platforms, such as BacDrop¹⁵ and M3-seq¹⁶, demonstrated transcript recovery from both Gram-positive and Gram-negative bacteria, suggesting that PFA-fixed luminal samples can reasonably approximate microbial transcriptomes. Although transcript recovery has been validated using defined controls in our previous smRandom-seq study¹⁷, spike-in experiments in native microbiome backgrounds will be important for future benchmarking of detection sensitivity. Furthermore, current reference genome databases and widely used taxonomic profiling tools were not designed specifically for microbial single-cell analyses. While Kraken2⁷⁶ and MetaPhlAn⁷⁷ provide useful benchmarks, they do not model cell-level barcode structure or sparse transcript recovery, highlighting the need for tailored approaches such as smClassify. Nevertheless, the smClassify workflow developed here is not yet fully optimized. Beyond methodological considerations, the spatial coverage of this study is limited. We focused on the cecum, colon and rectum, whereas the small intestine (duodenum, jejunum and ileum), where metabolite absorption is substantial but microbial biomass is relatively low and host debris is more abundant^26,78,79, was not profiled. Together, these findings highlight that while smRandom-seq provides high-resolution insights into microbial gene expression, careful interpretation is needed at the species level, particularly for biofilm-protected, host-associated, or low-abundance taxa, and absolute abundances may differ from those observed in bulk metagenomic sequencing. Moving forward, microbial single-cell transcriptomics should be viewed not as a substitute for metagenomics, but rather as a high-resolution complement for in-depth transcriptional analysis. Future methodological optimizations should prioritize expanding taxonomic capture through diverse enzymatic cocktails, optimized microbial scRNA-seq chemistry, higher-depth microbial transcript recovery, and more complete strain-resolved references. In addition, future benchmarking and experimental quality-control strategies should incorporate defined spike-in species and synthetic mock communities to better monitor taxon-specific recovery, detection sensitivity, quantitative accuracy, and cross-platform consistency.

While metagenomic sequencing defines microbial composition and functional potential, it cannot resolve transcriptional activity of individual microbes. Prior studies have established the biogeography of gut microbial communities across intestinal regions and niches⁷, and more recent spatially resolved approaches have further mapped host-microbiome organization at high resolution⁸. Microbial single-cell transcriptomics directly measures gene expression in individual cells, enabling the identification of distinct functional states within microbial populations. Although overall community composition was broadly similar, functional annotation of microbial single-cell transcriptomes showed that db/db microbiota displayed enhanced motility, chemotaxis, and stress adaptation, whereas WT communities favored organic acid metabolism and polysaccharide degradation. These patterns are broadly consistent with features reported in human cohorts, including moderate dysbiosis, reduced butyrate producers, enrichment of opportunistic taxa, and increased oxidative-stress and sulphate-reduction pathways in T2D^10,80, but should be interpreted in the context of species and ecological differences between mouse and human gut microbiota, and will require validation in human cohorts and cross-species comparative studies to assess their broader translational relevance. A cross-cohort analysis has identified strain-level T2D signatures, suggesting that phylogenetic diversity and gene content shape inter-individual variation in disease risk¹¹. Here, single-cell transcriptomics further resolved subpopulations and mapped within-species programs, exemplified by M. gordoncarteri, offering a mechanistic basis for heterogeneous associations across studies.

Previous studies have linked high plasma glutamate and a low glutamine-to-glutamate ratio with T2D and with lower BMI and HOMA-IR in men^81,82. In our study, serum amino acids were unchanged, whereas gut contents showed decreased glutamate and increased glutamine. Single-cell analyses associate these shifts with M. gordoncarteri, which exhibited loss of a stress-adapted rectal subpopulation, colonic urease repression, and rectal urease induction, suggesting a spatially uncoupled nitrogen flux. This reorganization coincided with enhanced colonic polysaccharide uptake and attenuated efflux and stress responses, providing a potential mechanistic link between gut microbial activity and the stability of circulating amino acid pools. Prior work also indicates that Muribaculaceae may have probiotic potential⁸³ and that tyrosine kinase inhibitors (TKIs) can enhance immunotherapy responses in part by increasing the abundance of M. gordoncarteri and its metabolite urocanic acid⁸⁴. Across human populations, urease capacity in the gut microbiome is generally low yet remains closely linked to host nitrogen recycling and health⁸⁵. Extending these observations, our species- and subpopulation-resolved profiling implicates specific Muribaculaceae members, particularly M. gordoncarteri subpopulations, as potential functional markers in the db/db mouse model and a basis for future studies of host-microbiome metabolic interactions. Mice were fasted for 4 hours prior to metabolomic profiling in this study, which may influence both systemic and luminal metabolite levels in the diabetic context. Future studies under different feeding conditions will be required to further evaluate host-microbiome metabolic interactions.

In these downstream analyses of the expression matrices of microbial single-cell RNA-seq data, we used Seurat⁸⁶ for normalization, dimensionality reduction, and clustering. However, it must be acknowledged that Seurat was originally developed for eukaryotic single-cell RNA-seq data, and its assumptions may not fully match the statistical properties of sparse prokaryotic single-cell transcriptomes. Contemporary prokaryotic single-cell datasets generally have smaller library sizes and lower genome coverage than eukaryotic datasets, which may approach the performance limits of existing algorithms for normalization, variance stabilization, clustering, and differential expression analysis⁸⁷. In this context, comparison with other microbial single-cell studies^{12–16,19,20,40,41} shows that cell filtering thresholds in this field are generally much lower than those used in eukaryotic scRNA-seq. At present, microbial single-cell studies commonly rely on tools developed for eukaryotic scRNA-seq, including Seurat in ProBac-seq¹⁴, BacDrop¹⁵, M3-seq¹⁶, and Scanpy⁸⁸ in microSPLiT¹³. Nevertheless, the development of dedicated computational frameworks optimized for sparse prokaryotic transcriptomes will be essential for improving the robustness and interpretability of future analyses.

Nonetheless, microbial single-cell transcriptomic profiling in this study was performed on a limited number of biological replicates (n = 2), and the cohort was restricted to a single genetic background and housing environment. As such, the observed patterns may vary with strain, age, sex, diet, and the specific T2DM model^89–91. Future improvements in smRandom-seq throughput and scalability should enable larger and more diverse cohorts to further evaluate the functional states identified here. Despite these limitations, the current framework provides a foundation for resolving microbial functional heterogeneity at single-cell resolution in complex gut environments. Extending these findings beyond the current experimental setting, particularly to human systems, will require region-resolved sampling strategies. Capturing spatial heterogeneity along the gastrointestinal tract remains challenging, as multi-segment sampling is difficult to scale longitudinally due to ethical and practical constraints⁹². Recently developed non-invasive ingestible devices provide a feasible path to region-resolved measurements in humans^93,94.

In summary, we introduce a species-level framework that maps microbial transcriptional programs along the mouse gut and links them to host metabolism. This shifts inference from bulk composition to within-species functional states and provides a scalable way to resolve microbial functional heterogeneity and connect single-microbe transcriptional states to host metabolic changes across diverse physiological and metabolic perturbations within the db/db mouse model.

Methods

Mice and samples

A total of 16 eight-week-old male mice on a C57BLKS/J background were used in this study, including C57BLKS/J db/db mice (type 2 diabetes model; n = 8) and age-matched male wild-type littermates (n = 8). Mice were obtained from the Model Animal Research Center of Changzhou Cavens (Jiangsu, China). Mice were housed under specific pathogen-free conditions at 22 ± 2 °C and 40–70% relative humidity, with a 12 h light/dark cycle and free access to water and standard laboratory mouse maintenance chow (Xietong Shengwu, catalogue no. 1010009; 3616 kcal kg⁻¹; 20.6% protein, 12.0% fat and 67.4% carbohydrate, according to the manufacturer’s specification). For metabolomic analyses, mice (db/db: n = 6, WT: n = 6) were fasted for approximately 4 h prior to sample collection. Animals were anesthetized for retro-orbital blood collection, followed immediately by euthanasia via cervical dislocation. Serum was isolated by centrifugation and stored at −80 °C for metabolomic profiling. Luminal contents from the cecum, colon, and rectum were collected immediately after euthanasia and snap-frozen in liquid nitrogen. For microbial single-cell transcriptomic profiling, an independent cohort of mice (db/db: n = 2, WT: n = 2) was processed under the same pre-delivery housing conditions. After euthanasia, luminal contents from the cecum, colon, and rectum were immediately collected for single-cell RNA sequencing of the microbiota. Sex was not considered as a biological variable in the study design or analysis because only male mice were used. All procedures were approved by the Zhejiang University Laboratory Animal Welfare and Ethics Review Committee (Approval No. ZJU20241156).

Escherichia coli BW25113 and Staphylococcus aureus subsp. aureus SA268 were provided by Sir Run Run Shaw Hospital, Zhejiang University School of Medicine and cultured aerobically at 37 °C in fresh LB Lennox medium. Muribaculum gordoncarteri (DSM 108194) and Anaerobutyricum hallii (DSM 3353) were purchased from Mingzhoubio (Ningbo, China) and cultured in CMC medium at 37 °C under anaerobic conditions. For mock community construction, cultures were harvested at OD600 = 0.5, centrifuged at 4 °C (2000 × g for 3 min), washed twice with PBS, and combined prior to smRandom-seq processing.

Microbial single-cell RNA-seq for murine luminal contents

Fresh luminal contents from the mouse cecum, colon, and rectum were collected on ice and processed on the smRandom-seq microfluidic platform, adapting the human feces workflow from the smRandom-seq to murine tissue^17,18. Anatomically defined segments of the cecum, colon, and rectum²⁶ were dissected on ice and immediately transferred intact into a 5 mL tube prefilled with 3 mL ice-cold 4% (wt/vol) paraformaldehyde in PBS (PFA, Sigma-Aldrich, 100496), ensuring the tissue was fully submerged. Within the tube, the segment was opened longitudinally, and luminal contents were gently scraped out with sterile forceps while avoiding contact with the intestinal wall, keeping the tissue fully submerged throughout. All tissue fragments were removed. The luminal suspension was gently mixed and fixed at 4 °C overnight (12-16 h).

Fixed samples were centrifuged at 500 g for 2 min at 4 °C to remove coarse debris and residual tissue fragments. The supernatant was transferred to new 1.5 mL tubes and centrifuged at 3910 g for 5 min at 4 °C to pellet microbial cells. The supernatant containing residual fixative was discarded, and the pellets were combined and resuspended in 1 mL pre-chilled PBS (Thermo Fisher Scientific, 10010072) supplemented with RNase inhibitor (Thermo Fisher Scientific, N8080119, 0.02 U µL⁻¹) and washed twice. The suspension was passed through 8 µm cell strainers (pluriSelect) to remove remaining aggregates, and the filtrate was collected in new 1.5 mL tubes. The filtrate was examined microscopically and counted. The suspension volume was adjusted as needed to transfer approximately 1 × 10⁷ cells to a new 1.5 mL tube for subsequent procedures.

Cells were centrifuged at 3910 g for 5 min at 4 °C, the supernatant was discarded, and the pellet was gently resuspended in 250 µL pre-chilled permeabilization buffer (1× PBS, 0.02 U mL⁻¹ RNase inhibitor, 0.04% (vol/vol) Tween-20, Sangon Biotech, CAS 9005-64-5) and incubated on ice for 3 min. The suspension was centrifuged again at 3910 g for 5 min at 4 °C, the supernatant was discarded, leaving ~100 µL. The pellet was resuspended by gentle pipetting. An enzyme mix (100 µL) was prepared on ice containing 45 µL 1× PBS, 40 µL lyso buffer (1×), 10 µL lysozyme (Thermo Fisher, cat. no. 90082, 50 mg mL⁻¹), 2.5 µL lysostaphin (Sigma-Aldrich, cat. no. 9011-93-2, 1 mg mL⁻¹) and 2.5 µL RNase inhibitor (20 U µL⁻¹). The enzyme mix was combined 1:1 with the ~100 µL cell suspension (final volume ~200 µL), and the mixture was incubated at 37 °C for 15 min with shaking at 800 rpm. Cells were then washed and resuspended in PBS containing RNase inhibitor for downstream smRandom-seq processing.

In situ reverse transcription was performed with the VITApilote-PFT1200 kit (M20 Genomics, R20114124) containing 5× RT buffer, random primers, dNTPs (Thermo Fisher, 18427089), RNase inhibitor, and reverse transcriptase (50 U/µL) using a stepped 8-42 °C program followed by a 30 min extension at 42 °C. After washing, in situ poly(dA) tailing was performed at 37 °C for 30 min using Terminal Transferase (NEB, M0315S), 10× TdT buffer, CoCl₂, and 100 mM dATP (NEB, N0440S).

Microbes (~2,000 cells/µL) were co-encapsulated with 2× extension mix and barcoded hydrogel beads (M20 Genomics, EFT131-1-B) on a VITAcruizer single-cell partitioning system (M20 Genomics, E20000131) using carrier oil (e.g., RAN Biotechnologies 008-FluoroSurfactant-2wtH-50G) and OptiPrep (Sigma-Aldrich, D1556). Droplets were incubated sequentially at 37 °C (60 min), 50 °C (30 min), 60 °C (30 min), and 75 °C (20 min). Emulsions were broken with perfluorooctanol (Adamas, 74014 C), and aqueous phases were purified with magnetic DNA cleanup beads (Vazyme, N411-01/VAHTS RNA Clean Beads N412-01 as needed). qPCR (EvaGreen, Biotium 31000) guided cycle selection to avoid over-amplification. cDNA was amplified with DeepVent/ThermoPol (NEB, M0259S), purified, quantified on a Qubit 4 fluorometer (Thermo Fisher, Q33238), and sized on a Qsep100 (BiOptic).

Libraries were prepared with the VAHTS Universal DNA Library Prep Kit for Illumina V3 (Vazyme, ND607-01) and VAHTS Unique Dual Index Adapters (Vazyme, N342-01) following the manufacturer’s instructions (end-prep, adapter ligation with double-sided bead selection, limited-cycle PCR), quantified/qualified (Qubit/Qsep100), and sequenced on an Illumina NovaSeq 6000 (PE150, 150 bp Paired-End).

Metabolomics profiling and data processing

Untargeted metabolomic profiling was performed by Lianchuan Biotechnology Co., Ltd. (Hangzhou, China). Serum and intestinal luminal-content samples from db/db and WT mice were analysed. For serum metabolomics, 12 biological samples were analysed, including six db/db and six WT mice, together with three pooled QC samples (QC1-QC3). For intestinal luminal-content metabolomics, luminal contents from the cecum, colon and rectum of the same mice were analysed, comprising 36 biological samples in total, together with five pooled QC samples (QC1-QC5). No technical replicate injections were performed for biological samples. Serum and luminal-content samples were thawed on ice. Serum (100 µL) was extracted with 400 µL pre-cooled 80% methanol, and luminal contents (~50 mg) were extracted with 0.5 mL pre-cooled 80% methanol. Extracts were kept at −20 °C for 30 min, centrifuged (20,000 × g, 15 min). Supernatants were vacuum-dried, and the extracts were reconstituted in 100 µL 80% methanol. Pooled QC samples were prepared separately for serum and luminal-content metabolomics by mixing 10 µL from each corresponding extract. Chromatographic separation was performed using an ACQUITY UPLC T3 column (100 mm × 2.1 mm, 1.8 μm, Waters, UK) on an ACQUITY UPLC system (Waters, Milford, USA). The column temperature was maintained at 40 °C, and the flow rate was set at 0.3 mL/min. The mobile phase consisted of A: water (5 mmol/L ammonium acetate + 5 mmol/L acetic acid) and B: acetonitrile. Gradient elution conditions were set as follows: 0–0.8 min, 2% B; 0.8–2.8 min, 2% to 70% B; 2.8–5.6 min, 70% to 90% B; 5.6–6.4 min, 90% to 100% B; 6.4–8.0 min, 100% B; 8.0–8.1 min, 100% to 2% B; 8.1–10 min, 2% B. Mass spectrometry was performed using a Q-Exactive (Thermo Fisher Scientific, Bremen, Germany) in both positive- and negative-ion modes. MS/MS spectra were acquired using a data-dependent acquisition (DDA) strategy⁹⁵, in which each full MS1 scan was followed by MS2 fragmentation of the top 6 most intense precursor ions. Full MS scans covered a range of 70–1050 m/z with a resolution of 70,000 (at m/z 200), AGC target of 3e6, and maximum injection time of 100 ms. MS/MS scans were acquired at a resolution of 17,500 (at m/z 200), AGC target of 1e5, and maximum injection time of 50 ms. Raw LC-MS data were converted to mzXML format using MSConvert (v3.0.24299-a1d9e91) in ProteoWizard. Positive- and negative-ion mode data were processed and annotated separately. Peaks were detected and retention-time aligned using the R package XCMS⁹⁶ (v4.2.3). During peak detection, a signal-to-noise threshold of 6 was applied. In the raw feature intensity matrix exported after peak detection and alignment, missing or undetected peak intensities were recorded as NA. No blank-sample subtraction or multidimensional blank correction was applied. Isotope/adduct annotation was subsequently performed using CAMERA⁹⁷. Within each ionization mode, metabolite identification was performed using metaX⁹⁸ (MS1) and by matching MS2 spectra to an in-house standard product database. Candidates were annotated against Kyoto Encyclopedia of Genes and Genomes (KEGG)³⁸. The raw feature intensity matrices were then merged for downstream preprocessing. Features with high missingness (> 50% in QC samples or > 80% in biological samples) were removed during quality control. Remaining missing values were imputed using a k-nearest-neighbors (KNN)-based approach implemented in the impute.knn function from the impute R package⁹⁹. The default rowmax parameter of the impute.knn function was used. Retained features with > 50% missing entries triggered the impute.knn fallback procedure, in which missing values were imputed using the overall mean per sample rather than nearest-neighbor averaging. After imputation, probabilistic quotient normalization (PQN) was applied. After normalization, feature reproducibility was assessed using pooled QC samples. Features with a coefficient of variation greater than 30% across QC samples were removed. The filtered and normalized feature intensity matrix was then merged with metabolite annotation information by feature ID. MS2 metabolite names were cleaned by removing quotation marks, and entries without MS2-based metabolite annotation were excluded from the curated MS2-annotated metabolite table. For duplicated MS2 metabolite names, the annotation with the highest MS2 matching score was retained for downstream analysis.

Metagenomics profiling

Metagenomics profiling was performed by Lianchuan Biotechnology Co., Ltd. (Hangzhou, China). Total DNA was extracted from cecal contents using the Genome DNA Extraction Kit according to the manufacturer’s protocol. Sequencing libraries were prepared with the TruSeq Nano DNA Library Preparation Kit (Illumina, #FC-121-4001) and sequenced on an Illumina NovaSeq 6000 platform (paired-end 150 bp) at LC-Bio Technology Co., Ltd. (Hangzhou, China). Raw reads were processed with fastp (v0.23.4) to remove adapter-contaminated, low-quality, and undetermined-base reads. Host-derived reads were then removed by aligning paired-end reads to the mouse reference genome (NCBI RefSeq assembly GRCm39) using Bowtie2 (v2.4.5) with the --very-sensitive option, and retaining unmapped read pairs for downstream analysis. For taxonomic profiling and comparison with single-cell RNA sequencing data, host-removed reads were classified using Kraken2⁷⁶ (v2.1.2) with the unrefined m-MGnify-based mouse gut genome catalog used in the single-cell analysis, which did not incorporate the p-RefSeq taxonomy refinement applied in smClassify. Classification was performed in paired-end mode with --confidence 0.1 and --minimum-hit-groups 2. Taxonomic assignments generated by Kraken2 were further refined using Bracken¹⁰⁰ (v2.8). Bracken re-estimates taxon abundances based on k-mer distributions in the same database, allowing more accurate quantification of relative abundances at family, genus, and species. The fraction of reads assigned to each taxon was used as a proxy for its relative abundance.

Preprocessing of microbial single-cell RNA-seq raw data

Raw FASTQ files were first processed following the smRandom-seq¹⁸ and MIC-seq¹⁹ analysis workflow, which includes data pre-processing, count matrix generation, and downstream object processing. All required software dependencies were installed as recommended in the protocol. Cleaned paired-end reads were supplied with Read 1 carrying barcodes/UMIs and Read 2 carrying cDNA. To remove host contamination, clean reads were first aligned to the mouse reference genome (NCBI RefSeq assembly GRCm39) using STAR¹⁰¹ (v2.7.10), and unmapped reads were retained for downstream microbial analysis.

Evaluation of reference genome resources for mouse gut microbial single-cell transcriptomic analysis

To evaluate how different mouse gut microbial reference genome resources affect taxonomic assignment of microbial single-cell RNA-seq data, we compared three reference databases, including p-RefSeq³⁷, MGBC³¹, and m-MGnify³⁴. Key features of these databases, including database size and composition, genome quality, annotation coverage, and taxonomic breadth, are summarized in Suppl. Table 3. We then applied a Kraken2/Bracken-based classification workflow to the preprocessed microbial single-cell datasets from all 12 samples using each of the three reference databases. These databases were used directly and did not incorporate the p-RefSeq-based taxonomy refinement implemented in smClassify. Cleaned and host-depleted paired-end reads were generated as described above, with Read 1 carrying cell barcodes and UMIs and Read 2 carrying cDNA sequences. Cell barcodes were identified from Read 1, and the associated Read 2 sequences were extracted for each barcode. Redundant reads were deduplicated to reduce PCR-derived redundancy. Barcodes were ranked by read count, and the top 200,000 barcodes were retained for downstream classification. Read 2 sequences from retained barcodes were combined into a single FASTQ file for each sample, with cell barcode information encoded in the read identifiers before Kraken2 classification. The pooled FASTQ file was classified against each reference database using Kraken2⁷⁶. Classified reads were then assigned back to individual barcodes according to the barcode information retained in the read identifiers. Kraken2 read assignments were summarized for each barcode. Bracken¹⁰⁰ was used with the Kraken2 output and corresponding Bracken k-mer distribution database to infer species-level support for reads assigned to higher taxonomic ranks. For each barcode, Kraken2 species-level assigned reads and Bracken-estimated species-level reads were combined, and the species with the highest total read support was selected as the barcode-level species. The three reference databases used without smClassify taxonomy refinement were benchmarked based on the overall assignment rate, genus- and species-level assignment rates, and recovered taxonomic diversity, including the numbers of unique genera and species detected from the microbial single-cell data.

Taxonomic classification of microbial single-cell RNA-seq data using smClassify

Host-depleted reads were then aligned to the m-MGnify reference (either full genomes or a CDS reference catalog) using STAR in solo mode for cell barcode and UMI quantification. Cell barcodes and UMIs were extracted from positions 1–20 and 21–28, respectively. Splice junction detection was turned off by setting --alignSJoverhangMin to 1000, as expected for prokaryotic genomes. Alignment quality was controlled using --outFilterScoreMinOverLread 0.5, and strand specificity was specified using --soloStrand Reverse. Gene expression was quantified using the GeneFull mode. Barcode selection thresholds (TopCells 15,000, 20,000, and 30,000) were systematically evaluated by comparison with CellRanger-like barcode rank estimates, including recovered cell numbers, median UMI counts, and detected gene numbers per cell. The top 15,000 barcodes ranked by UMI counts were retained, falling within the range estimated from barcode rank profiles and balancing cell recovery with transcriptome quality. Sequence saturation for each sample was calculated as 1 minus the ratio of unique UMIs to total reads from retained barcodes, where higher values indicate greater sequencing redundancy.

Species assignment of single microbial transcriptomes was performed using a custom R pipeline (smClassify) with Seurat and data.table¹⁰². Uniquely mapped reads from the CDS reference alignment were used for downstream analysis. rRNA/tmRNA-related genes were identified from GTF annotations and excluded prior to analysis. Taxonomic IDs (starting with MGYG) were extracted from m-MGnify Genomes’ gene names and mapped to reference species. Taxonomic ranks were inferred by majority vote based on individual gene-level UMI counts assigned to specific taxonomic IDs, conceptually informed by prior approaches for taxonomic assignment¹⁰³. The following criteria were used to define the confidence of species assignments: at least 50 supporting counts for the top-ranked species, a minimum fraction of 0.6 of UMIs, and a margin of at least 0.2 between the top-ranked and second-ranked species. These parameters were guided by prior studies on confidence score-based filtering in taxonomic classification¹⁰⁴ and further evaluated through sensitivity analyses using the same workflow, including top-species fraction (min_frac = 0.5-0.7), minimum UMI counts (50-100), and margin over the second-ranked species (min_margin = 0.1-0.3). Species names were then harmonized using an MGnify-RefSeq crosswalk derived from ANI comparisons^37,105. When database-specific labels differed, the MGnify genome-based assignment was used as the primary identifier and species names were curated using the crosswalk table. Non-informative species labels (for example “Genus sp.” or “Unknown”) were replaced with RefSeq species names when available; otherwise, the assignment was retained at the genus level. Higher taxonomic ranks (genus and family) were inferred from the harmonized crosswalk table. The resulting harmonized taxonomy was used for all downstream analyses. Barcodes were prefiltered based on UMI counts (≥100) and detected features (≥30) to remove low-quality cells, accounting for the lower RNA content and gene detection rates typical of microbial single-cell transcriptomes compared with mammalian single-cell RNA-seq data.

QC metrics, including sequencing depth, alignment statistics, per-cell UMI and gene counts, and taxonomic assignment performance, were systematically summarized and visualized. Sequencing and alignment metrics based on full-genome and CDS-restricted references are provided in Suppl. Tables 5 and 6, respectively, while smClassify taxonomic assignment metrics are summarized in Suppl. Table 7. Visualizations of margin distribution, doublet frequency¹⁰⁶, and assignment confidence are shown in Suppl. Fig. 5a-c.

Benchmarking of taxonomic profiling methods

Kraken2⁷⁶ and MetaPhlAn⁷⁷ (v4.1.1) were used as external reference methods for taxonomic comparison and benchmarked against smClassify results. These tools were originally developed for bulk metagenomic or metatranscriptomic profiling and do not model cell-level barcode structure, sparse transcript recovery, or within-cell assignment consistency.

For Kraken2-based benchmarking of microbial single-cell data, classification was performed using the same read-level Kraken2 workflow followed by barcode-level summarization described above. Briefly, cleaned and host-depleted Read 2 sequences from retained barcodes were combined into a single FASTQ file for each sample, with cell barcode information encoded in the read identifiers. The pooled FASTQ file was classified using Kraken2 against the m-MGnify catalog built with the standard Kraken2 database-building workflow. This database was used directly and did not incorporate the p-RefSeq-based taxonomy refinement implemented in smClassify. After Kraken2 classification, classified reads were assigned back to individual barcodes using the barcode information retained in the read identifiers. Kraken2 read assignments were then summarized for each barcode. Bracken was used with the Kraken2 output and corresponding Bracken k-mer distribution database to infer species-level support for reads assigned to higher taxonomic ranks. For each barcode, Kraken2 species-level assigned reads and Bracken-estimated species-level reads were combined, and the species with the highest total read support was selected as the barcode-level species.

For MetaPhlAn applied to microbial single-cell data, benchmarking was performed at the sample level, treating each sample’s reads as a pooled library. Cleaned Read 2 sequences were used as input, with parameters --min_mapq_val 2 --read_min_len 30 --stat_q 0.1. Due to low mapping fractions to clade-specific marker genes and sample-level species assignment, MetaPhlAn results were not used for subsequent comparative analyses.

Reproducibility of microbial single-cell profiles using Kraken2 and smClassify was evaluated with Bray-Curtis dissimilarity between biological replicates, computed from species and genus-level abundance profiles using the vegdist function (method = “bray”) in vegan, and paired Wilcoxon tests were used for statistical comparisons. Relative abundances were computed at the genus level after grouping by phenotype and intestinal region. Concordance was assessed using Pearson (log₁₀-transformed abundances) and Spearman correlations.

Comparison of metagenomic and microbial single-cell results

Metagenomic profiles were generated using Kraken2/Bracken with the m-MGnify-based mouse gut genome catalog as described above. Single-cell taxonomic assignments were obtained using smClassify or Kraken2 as described above. Relative abundances of microbial taxa detected by metagenomic sequencing and microbial single-cell RNA-seq were compared at family, genus, and species levels after grouping by phenotype. Concordance was assessed using Pearson correlations on relative and log₁₀-transformed abundances, as well as Spearman correlations, with a pseudo-count of 1 × 10⁻⁸ added prior to log transformation to reduce compositional skew. Overlap and abundance differences were evaluated using shared taxa counts and log₂ fold-change.

Reduction, clustering, and differential gene expression analysis of microbial single-cell RNA-seq data

Microbial single-cell RNA-seq count matrices were processed in R using Seurat⁸⁶ (v5.0). rRNA/tmRNA features were removed, genes were annotated via GTF tables, and those labeled as “hypothetical protein” were excluded (Suppl. Table 11, Suppl. Fig. 5h, i). Gene counts mapping to the same symbol were aggregated, and Seurat objects were created with taxonomic and sample metadata. Gene-level filtering (genes detected in ≥ 5 cells) was applied to reduce sparsity. Cells were filtered to retain those with nCount_RNA ≤ 1000, to limit the influence of cells with disproportionately high transcript counts. Cells were then downsampled to a maximum of 7,000 per sample prior to downstream analyses. This threshold was defined with reference to the sample with the lowest number of cells estimated from CellRanger-like barcode rank profiles (Suppl. Table 4), ensuring balanced representation across samples. Data were normalized using SCTransform (vst.flavor = “v2”), regressing out nCount_RNA to reduce technical variability associated with differences in library size, and selecting 5,000 variable features (using glmGamPoi when available). PCA (50 PCs) was performed, and the top 6 PCs were used for neighbor graph construction (FindNeighbors) and Louvain clustering (FindClusters, resolution = 0.1). UMAP was applied for visualization. To evaluate the impact of this choice, alternative thresholds (8000 and 9000 cells per sample, as well as no downsampling) were processed using the same workflow. Cell number distributions, clustering structure, and functional annotations were compared across settings by mapping cluster labels from the 7000-cell reference dataset based on shared barcodes. Cluster consistency was quantified using retention rates, defined as the fraction of cells within each functional cluster assigned to the dominant Seurat cluster. Differential expression analysis was performed using FindAllMarkers (only.pos = TRUE, logfc.threshold = 1, min.pct = 0.05) based on gene expression at single-cell resolution. The top 50 markers per cluster were identified and annotated. The top 10 markers were used to calculate the average expression across clusters from SCT-normalized data. Comparisons of marker genes or pathway-related genes across groups were visualized using Seurat DotPlot, showing average expression and the fraction of expressing cells in each group, and were interpreted descriptively without formal statistical testing.

KEGG pathway enrichment analysis of microbial single-cell RNA-seq data

Cluster-specific marker genes were mapped to KEGG orthology (KO) terms based on GTF annotations and subjected to pathway enrichment analysis. Over-representation analysis was performed using clusterProfiler¹⁰⁷. Broad pathways (e.g., ko01100) and human disease categories were excluded to highlight functional differences between clusters. Significant KEGG pathways (Benjamini-Hochberg adjusted p-value < 0.05) were visualized with ggplot2 and ggh4x, using gene ratio and -log₁₀(p.adjust) to represent enrichment strength and grouping pathways by functional categories.

Metabolite differential and enrichment analysis

Untargeted metabolomics data from serum and intestinal contents were analyzed using R (version 4.4.3). The filtered and normalized metabolite intensity matrices generated above were log2-transformed for downstream analyses. Principal component analysis (PCA) was performed to evaluate sample distribution and group separation. Partial least squares discriminant analysis (PLS-DA) was conducted with two predictive components and 100 permutations to assess group-specific metabolic variation. Differential metabolite analysis between groups was performed for each tissue using the limma package¹⁰⁸. Metabolites with P < 0.05 and |log₂ fold change | > 0.5 were considered significant. Shared and tissue-specific differential metabolites were visualized using UpSet plots. Volcano plots were generated to highlight global and selected core metabolites. For pathway analysis, differential metabolites were mapped to their corresponding KEGG compound identifiers. In cases where multiple metabolic features (e.g., different adducts or isomers) were assigned to the same KEGG ID, the identifier list was deduplicated. The resulting unique KEGG ID list was then used for pathway enrichment analysis via the clusterProfiler R package. Pathways were further grouped into custom-defined functional categories (e.g., amino acids, fatty acyls/lipids, bile acids, carbohydrates/energy, urea cycle, transporters, signaling). For selected metabolites and metabolite ratios shown in box plots, comparisons between WT and db/db mice within each tissue were performed using two-sided Welch’s t-tests. EPA/AA ratios were calculated from normalized intensity values.

Integration of single-cell transcriptomics and metabolomics

Significantly enriched KEGG pathways from single-cell transcriptomic clusters and metabolomic groups were intersected to identify shared functional signatures. A bipartite network linking clusters and groups was constructed, with edge weights reflecting enrichment strength and significance. The network was visualized on UMAP coordinates using igraph (https://github.com/igraph/rigraph) and ggraph (https://ggraph.data-imaginist.com/). Top KEGG orthologs (KOs) contributing to shared pathways were extracted, and global relationships among clusters, KOs, and groups were visualized using circlize chord diagrams¹⁰⁹. All analyses were performed in R (version 4.4.3) with tidyverse, Seurat, igraph, ggraph, ggrepel, scales, viridis, and circlize packages.

Inter-species correlation analysis of microbial single-cell RNA-seq data

Microbial single-cell RNA-seq data were aggregated by species per sample to build a species abundance matrix, which was then converted to relative abundance. Pairwise inter-species associations were computed using Spearman correlation, and the resulting correlation structure was visualized with heatmaps (pheatmap). Functional pathway activity was quantified by module scoring of defined metabolic and stress gene sets, and summarized across the 20 most abundant species using dot plots to highlight functional relationships.

Subcluster analysis of microbial single-cell transcriptomes

Cells from the three highly correlated species (Muribaculum gordoncarteri, Duncaniella muris, and Bacteroides muris) were subset from the global single-cell dataset for subclustering analysis. Each subgroup was normalized with SCTransform and then subjected to principal component analysis (PCA). The top 10 components capturing the most significant expression variance were used to construct a kNN graph and perform Louvain clustering (resolution 0.1–0.15). Clustering resolution and principal components were adjusted to capture within-species heterogeneity. Subclustering of cells belonging to the genus Duncaniella was performed using PCA (top 6 PCs), and the Louvain clustering resolution was set to 0.1. For Parabacteroides spp., we used the top 5 PCs for kNN graph construction, and we set the Louvain clustering resolution to 0.15. Lever scores and symbolic matrices (+/−/·) were generated and visualized with heatmaps and dot plots to identify functional variation among species and infer cultivation potential. Cells of M. gordoncarteri were reprocessed independently using SCTransform, PCA (top 6 PCs), and Louvain clustering (resolution 0.1).

Pseudotime trajectory inference of microbial single-cell RNA-seq data

To explore lineage dynamics of cells from Muribaculum gordoncarteri, pseudotime trajectory inference was performed using monocle3 (https://cole-trapnell-lab.github.io/monocle3/), importing the existing UMAP embedding from Seurat. Cells were ordered along the trajectory with learn_graph and order_cells. Differential expression along pseudotime was detected with graph_test, and significantly dynamic genes were visualized with plot_genes_in_pseudotime and heatmaps.

Pseudobulk differential expression analysis

For differential expression between WT and db/db groups, raw RNA counts were aggregated at the sample level to generate pseudobulk profiles within each tissue or tissue-by-cluster stratum. Differential expression was performed using edgeR with TMM normalization and a negative binomial quasi-likelihood model. Genes with zero counts across all samples were excluded, and analyses were restricted to strata with at least two biological replicates per group, consistent with current metatranscriptomic analysis practices⁷³.

Statistics & reproducibility

Sample sizes were not determined by formal statistical power calculations, but were chosen based on experimental feasibility, sequencing throughput and availability of biological materials. No data were excluded from the analyses, except for data removed according to the predefined quality-control procedures described in the Methods, including low-quality reads, host-derived reads, low-quality barcodes or cells, low-confidence taxonomic assignments and metabolomic features failing missingness or pooled-QC reproducibility criteria.

The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment. All statistical analyses were performed in R (v4.4.3) unless stated otherwise. Unless indicated, tests were two-sided, and significance was defined as P < 0.05 after the Benjamini-Hochberg multiple-testing procedures. Reproducibility was assessed using independent biological samples, pooled QC samples for metabolomics, biological-replicate comparisons and sensitivity analyses for key single-cell processing parameters.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(12.4MB, pdf)}

Reporting Summary^{(107.6KB, pdf)}

Transparent Peer Review file^{(761.4KB, pdf)}

Source data

Source Data^{(74.9MB, zip)}

Acknowledgements

We thank the core facilities of Zhejiang University and Liangzhu Laboratory for technical support. We also thank Jingyao Chen and Chengcheng Zhang from the Core Facilities, Zhejiang University School of Medicine, for their technical assistance.

Author contributions

Y.W. conceived the study and oversaw project progress. Z.X. and X.Long wrote and revised the manuscript. Z.X., S.Z., X.Li, and F.L. carried out sample collection. Z.X., M.S., S.Z., X.Li, and T.C. coordinated the microbial single-cell RNA-seq experiments. Z.X., X.Long., and T.Z. performed and coordinated the data analysis. All authors reviewed and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

The project was supported by the National Natural Science Foundation of China (No. 82571013, Z.X.), the National Key R&D Program of China (2025YFF0512800, Y.W.), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2026ZD0556100, Y.W.), the Pioneer R&D Programs of Zhejiang Province (2024C03005, Y.W.), the Key R&D Program of Zhejiang (2024SSYS0022, Y.W.), the Major Science and Technology Projects of Shaoxing Yellow Rice Wine (2024-2026) (2024SDXT001-1, Y.W.), and China Postdoctoral Science Foundation (No.2025M770277, S.Z.).

Data availability

The microbial single-microbe RNA-seq data generated in this study have been deposited in the Genome Sequence Archive (GSA) under accession code CRA033019. The metagenomic sequencing data generated in this study have been deposited in the Genome Sequence Archive (GSA) under accession code CRA040709. The full pre-filtered count matrices for all microbial single-cell RNA-seq datasets generated in this study have been deposited in figshare under DOI 10.6084/m9.figshare.31889338 10.6084/m9.figshare.31889338. The public scRNA-seq datasets from mouse cecum and colon used in this study are available in the CNGBdb database under project accession code CNP0005580. Raw LC-MS metabolomics data have been deposited in Zenodo under DOI 10.5281/zenodo.20734580 [https://zenodo.org/records/20734580]. Source data are provided with this paper.

Code availability

The analysis code used in this study is available in the smClassify GitHub repository (https://github.com/WangycLab/smClassify)¹⁰². Previously published reference protocols or workflows used in this study are available at smRandom-seq (https://github.com/WangycLab/smRandom-seq-protocol), and MIC-seq analysis workflow (https://github.com/MIC-seq/MIC-seq-analysis-workflow).

Competing interests

Y.W. is a co-founder of M20 Genomics, and T.Z. is an employee of M20 Genomics, which provides reagents and instrumentation related to the smRandom-seq workflow used in this study. The remaining authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors contributed equally: Ziye Xu, Xin Long.

Contributor Information

Ziye Xu, Email: ziyexu@zju.edu.cn.

Yongcheng Wang, Email: yongcheng@zju.edu.cn.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75546-z.

References

1.The Human Microbiome Project Consortium. Structure, function and diversity of the healthy human microbiome. Nature. 486, 207–214 (2012). [DOI] [PMC free article] [PubMed]
2.Mukhopadhya, I. & Louis, P. Gut microbiota-derived short-chain fatty acids and their role in human health and disease. Nat. Rev. Microbiol. 23, 635–651 (2025). [DOI] [PubMed] [Google Scholar]
3.Fan, Y. & Pedersen, O. Gut microbiota in human metabolic health and disease. Nat. Rev. Microbiol.19, 55–71 (2021). [DOI] [PubMed] [Google Scholar]
4.Li, P., Roos, S., Luo, H., Ji, B. & Nielsen, J. Metabolic engineering of human gut microbiome: Recent developments and future perspectives. Metab. Eng.79, 1–13 (2023). [DOI] [PubMed] [Google Scholar]
5.Meijnikman, A. S., Gerdes, V. E., Nieuwdorp, M. & Herrema, H. Evaluating causality of gut microbiota in obesity and diabetes in humans. Endocr. Rev.39, 133–153 (2018). [DOI] [PubMed] [Google Scholar]
6.Howard, E. J., Lam, T. K. T. & Duca, F. A. The gut microbiome: connecting diet, glucose homeostasis, and disease. Annu. Rev. Med.73, 469–481 (2022). [DOI] [PubMed] [Google Scholar]
7.McCallum, G. & Tropini, C. The gut microbiota and its biogeography. Nat. Rev. Microbiol.22, 105–118 (2024). [DOI] [PubMed] [Google Scholar]
8.Ntekas, I. et al. Spatial transcriptomics maps host-gut microbiome biogeography at high resolution. Nat. Microbiol.11, 1193–1204 (2026). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Pountain, A. W. & Yanai, I. Dissecting microbial communities with single-cell transcriptome analysis. Science (New York, N.Y.)389, eadp6252 (2025). [DOI] [PMC free article] [PubMed]
10.Qin, J. et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature490, 55–60 (2012). [DOI] [PubMed] [Google Scholar]
11.Mei, Z. et al. Strain-specific gut microbial signatures in type 2 diabetes identified in a cross-cohort analysis of 8,117 metagenomes. Nat. Med.30, 2265–2276 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Blattman, S. B., Jiang, W., Oikonomou, P. & Tavazoie, S. Prokaryotic single-cell RNA sequencing by in situ combinatorial indexing. Nat. Microbiol.5, 1192–1201 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Kuchina, A. et al. Microbial single-cell RNA sequencing by split-pool barcoding. Science371, 5257 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
14.McNulty, R. et al. Probe-based bacterial single-cell RNA sequencing predicts toxin regulation. Nat. Microbiol.8, 934–945 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Ma, P. et al. Bacterial droplet-based single-cell RNA-seq reveals antibiotic-associated heterogeneous cellular states. Cell186, 877–891 e814 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Wang, B. et al. Single-cell massively-parallel multiplexed microbial sequencing (M3-seq) identifies rare bacterial populations and profiles phage infection. Nat. Microbiol8, 1846–1862 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Xu, Z. et al. Droplet-based high-throughput single microbe RNA sequencing by smRandom-seq. Nat. Commun.14, 5130 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Xu, Z., Wang, Y., Cai, W., Chen, Y. & Wang, Y. Single microorganism RNA sequencing of microbiomes using smRandom-Seq. Nat. Protoc.21, 160–199 (2025). [DOI] [PubMed] [Google Scholar]
19.Shen, Y. et al. High-throughput single-microbe RNA sequencing reveals adaptive state heterogeneity and host-phage activity associations in human gut microbiome. Protein Cell16, 211–226 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Shen, Y. et al. Single-microbe RNA sequencing uncovers unexplored specialized metabolic functions of keystone species in the human gut. Imeta4, e70035 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Jia, M. et al. Single-cell transcriptomics across 2,534 microbial species reveals functional heterogeneity in the rumen microbiome. Nat. Microbiol.9, 1884–1898 (2024). [DOI] [PubMed] [Google Scholar]
22.Martinez-Guryn, K., Leone, V. & Chang, E. B. Regional Diversity of the Gastrointestinal Microbiome. Cell Host Microbe26, 314–324 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Folz, J. et al. Human metabolome variation along the upper intestinal tract. Nat. Metab.5, 777–788 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Shalon, D. et al. Profiling the human intestinal environment under physiological conditions. Nature617, 581–591 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Hugenholtz, F. & de Vos, W. M. Mouse models for human intestinal microbiota research: a critical evaluation. Cell. Mol. Life Sci.75, 149–160 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Meier, K. H. U. et al. Metabolic landscape of the male mouse gut identifies different niches determined by microbial activities. Nat. Metab.5, 968–980 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Yu, F. et al. Abnormal gut microbiota composition contributes to the development of type 2 diabetes mellitus in db/db mice. Aging11, 10454–10467 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Yu, F. et al. Gut microbiota transplantation from db/db mice induces diabetes-like phenotypes and alterations in Hippo signaling in pseudo germ-free mice. Aging12, 24156–24167 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Lyu, Y. et al. Effects of combination treatment with metformin and berberine on hypoglycemic activity and gut microbiota modulation in db/db mice. Phytomedicine: Int. J. Phytother. Phytopharmacol.101, 154099 (2022). [DOI] [PubMed] [Google Scholar]
30.Liu, R. et al. Study on the mechanism of modified Gegen Qinlian decoction in regulating the intestinal flora-bile acid-TGR5 axis for the treatment of type 2 diabetes mellitus based on macro genome sequencing and targeted metabonomics integration. Phytomedicine: Int. J. Phytother. Phytopharm.132, 155329 (2024). [DOI] [PubMed] [Google Scholar]
31.Beresford-Jones, B. S. et al. The mouse gastrointestinal bacteria catalogue enables translation between the mouse and human gut microbiotas via functional mapping. Cell Host Microbe30, 124–138.e128 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Kieser, S., Zdobnov, E. M. & Trajkovski, M. Comprehensive mouse microbiota genome catalog reveals major difference to its human counterpart. PLOS Comput. Biol.18, e1009947 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Kim, N. et al. MRGM: an enhanced catalog of mouse gut microbial genomes substantially broadening taxonomic and functional landscapes. Gut microbes16, 2393791 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Gurbich, T. A. et al. MGnify genomes: a resource for biome-specific microbial genome catalogues. J. Mol. Biol.435, 168016 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Xiao, L. et al. A catalog of the mouse gut metagenome. Nat. Biotechnol.33, 1103–1108 (2015). [DOI] [PubMed] [Google Scholar]
36.Lesker, T. R. et al. An Integrated Metagenome Catalog Reveals New Insights into the Murine Gut Microbiome. Cell Rep.30, 2909–2922.e2906 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.O’Leary, N. A. et al. Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation. Nucleic Acids Res.44, D733-D745 (2015). [DOI] [PMC free article] [PubMed]
38.Kanehisa, M. & Goto, S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res28, 27–30 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Jain, C., Rodriguez-R, L. M., Phillippy, A. M., Konstantinidis, K. T. & Aluru, S. High throughput ANI analysis of 90K prokaryotic genomes reveals clear species boundaries. Nat. Commun.9, 5114 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Wang, Z. et al. Hosts manipulate lifestyle switch and pathogenicity heterogeneity of opportunistic pathogens in the single-cell resolution. Elife13, RP96789 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Korshoj, L. E. & Kielian, T. Bacterial single-cell RNA sequencing captures biofilm transcriptional heterogeneity and differential responses to immune pressure. Nat. Commun.15, 10184 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Gholami, H., Chmiel, J. A., Burton, J. P. & Maleki Vareki, S. The role of microbiota-derived vitamins in immune homeostasis and enhancing cancer immunotherapy. Cancers15, 1300 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Arumugam, M. et al. Enterotypes of the human gut microbiome. Nature473, 174–180 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Tropini, C. et al. Transient osmotic perturbation causes long-term alteration to the gut microbiota. Cell173, 1742–1754.e1717 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Malaguarnera, G., Giordano, M., Nunnari, G., Bertino, G. & Malaguarnera, M. Gut microbiota in alcoholic liver disease: pathogenetic role and therapeutic perspectives. World J. Gastroenterol.20, 16639–16648 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Raina, S. & Klein, G. Lipopolysaccharide: recent advances in its biosynthesis and controlling cell envelope homeostasis. Int. J. Mol. Sci.26, 7705 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Morrison, D. J. & Preston, T. Formation of short chain fatty acids by the gut microbiota and their impact on human metabolism. Gut Microbes7, 189–200 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Sereti, I. et al. Impaired gut microbiota-mediated short-chain fatty acid production precedes morbidity and mortality in people with HIV. Cell Rep.42, 113336 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Wewalka, M., Patti, M. E., Barbato, C., Houten, S. M. & Goldfine, A. B. Fasting serum taurine-conjugated bile acids are elevated in type 2 diabetes and do not change with intensification of insulin. J. Clin. Endocrinol. Metab.99, 1442–1451 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Serena, C. et al. Elevated circulating levels of succinate in human obesity are linked to specific gut microbiota. ISME J.12, 1642–1657 (2018). [DOI] [PMC free article] [PubMed]
51.Haeusler, R. A., Astiarraga, B., Camastra, S., Accili, D. & Ferrannini, E. Human insulin resistance is associated with increased plasma levels of 12α-hydroxylated bile acids. Diabetes62, 4184–4191 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Duft, R. G. et al. Serum metabolites associated with increased insulin resistance and low cardiorespiratory fitness in overweight adolescents. Nutr., Metab. Cardiovasc. Dis.32, 269–278 (2022). [DOI] [PubMed] [Google Scholar]
53.Li, M. et al. Trends in insulin resistance: insights into mechanisms and therapeutic strategy. Signal Transduct. Target. Ther.7, 216 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Fernández-Veledo, S., Marsal-Beltran, A. & Vendrell, J. Type 2 diabetes and succinate: unmasking an age-old molecule. Diabetologia67, 430–442 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Würtz, P. et al. Metabolic signatures of insulin resistance in 7,098 young adults. Diabetes61, 1372–1380 (2012). [DOI] [PMC free article] [PubMed]
56.Guasch-Ferré, M. et al. Metabolomics in prediabetes and diabetes: a systematic review and meta-analysis. Diab. Care39, 833–846 (2016). [DOI] [PMC free article] [PubMed]
57.Mihalik, S. J. et al. Metabolomic profiling of fatty acid and amino acid metabolism in youth with obesity and type 2 diabetes: evidence for enhanced mitochondrial oxidation. Diabetes Care35, 605–611 (2012). [DOI] [PMC free article] [PubMed]
58.Ling, Z.-N. et al. Amino acid metabolism in health and disease. Signal Transduct. Target. Ther.8, 345 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Anand, U. & Anand, C. V. Connecting links between the urea cycle and the TCA cycle: a tutorial exercise. Biochem. Educ.27, 153–154 (1999). [Google Scholar]
60.Yeh, S.-L., Shih, Y.-M. & Lin, M.-T. in Diabetes (Second Edition) (ed Victor R. Preedy) 255-264 (Academic Press, 2020).
61.Davila, A.-M. et al. Re-print of “Intestinal luminal nitrogen metabolism: Role of the gut microbiota and consequences for the host. Pharmacol. Res.69, 114–126 (2013). [DOI] [PubMed] [Google Scholar]
62.Shen, J. et al. Cross-tissue multi-omics analyses reveal the gut microbiota’s absence impacts organ morphology, immune homeostasis, bile acid and lipid metabolism. Imeta4, e272 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Egalini, F. et al. The Effects of Omega 3 and Omega 6 Fatty Acids on Glucose Metabolism: An Updated Review. Nutrients15, 2672 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Lu, J. et al. Association of Serum Bile Acids Profile and Pathway Dysregulation With the Risk of Developing Diabetes Among Normoglycemic Chinese Adults: Findings From the 4C Study. Diabetes Care44, 499–510 (2020). [DOI] [PubMed]
65.Mozaffarian, D. Saturated fatty acids and type 2 diabetes: more evidence to re-invent dietary guidelines. Lancet Diab Endocrinol.2, 770–772 (2014). [DOI] [PubMed] [Google Scholar]
66.Huang, L. et al. Circulating Saturated Fatty Acids and Incident Type 2 Diabetes: A Systematic Review and Meta-Analysis. Nutrients11, 998 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Lampousi, A.-M. et al. Interaction between plasma phospholipid odd-chain fatty acids and GAD65 autoantibodies on the incidence of adult-onset diabetes: the EPIC-InterAct case-cohort study. Diabetologia66, 1460–1471 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Yao, L. et al. A selective gut bacterial bile salt hydrolase alters host metabolism. eLife7, e37182 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Depommier, C. et al. Supplementation with Akkermansia muciniphila in overweight and obese human volunteers: a proof-of-concept exploratory study. Nat. Med.25, 1096–1103 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Li, W. Z., Stirling, K., Yang, J. J. & Zhang, L. Gut microbiota and diabetes: From correlation to causality and mechanism. World J. Diab11, 293–308 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Takeuchi, T. et al. Gut microbial carbohydrate metabolism contributes to insulin resistance. Nature621, 389–395 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Miyake, S., Ding, Y., Soh, M., Low, A. & Seedorf, H. Muribaculum gordoncarteri sp. nov., an anaerobic bacterium from the faeces of C57BL/6J mice. Int. J. Syst. Evolut. Microbiol.70, 4725–4729 (2020). [DOI] [PubMed] [Google Scholar]
73.Zhang, Y., Thompson, K. N., Huttenhower, C. & Franzosa, E. A. Statistical approaches for differential expression analysis in metatranscriptomics. Bioinforma. (Oxf., Engl.)37, i34–i41 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
74.He, J.-L. et al. Enhanced interactions among gut mycobiomes with the deterioration of glycemic control. Med5, 909–925.e907 (2024). [DOI] [PubMed] [Google Scholar]
75.Huang, H. et al. The mycobiome as integral part of the gut microbiome: crucial role of symbiotic fungi in health and disease. Gut microbes16, 2440111 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Lu, J. et al. Metagenome analysis using the Kraken software suite. Nat. Protoc.17, 2815–2839 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Blanco-Míguez, A. et al. Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4. Nat. Biotechnol.41, 1633–1644 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Delbaere, K. et al. The small intestine: dining table of host-microbiota meetings. FEMS Microbiol. Rev.47, 10.1093/femsre/fuad022 (2023). [DOI] [PMC free article] [PubMed]
79.Steinbach, E. et al. Upper small intestine microbiome in obesity and related metabolic disorders: A new field of investigation. Metabolism-Clin. Exp.150, 10.1016/j.metabol.2023.155712 (2024). [DOI] [PubMed]
80.Karlsson, F. H. et al. Gut metagenome in European women with normal, impaired and diabetic glucose control. Nature498, 99–103 (2013). [DOI] [PubMed] [Google Scholar]
81.Liu, X. et al. High plasma glutamate and low glutamine-to-glutamate ratio are associated with type 2 diabetes: Case-cohort study within the PREDIMED trial. Nutr., Metab. Cardiovasc. Dis.29, 1040–1049 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
82.Dollet, L. et al. Glutamine regulates skeletal muscle immunometabolism in type 2 diabetes. Diabetes71, 624–636 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
83.Zhu, Y. et al. Exploration of the muribaculaceae family in the gut microbiota: diversity, metabolism, and function. Nutrients16, 2660 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Zhang, M. et al. Microbiota-derived urocanic acid triggered by tyrosine kinase inhibitors potentiates cancer immunotherapy efficacy. Cell Host Microbe33, 915–931.e919 (2025). [DOI] [PubMed] [Google Scholar]
85.Yatsunenko, T. et al. Human gut microbiome viewed across age and geography. Nature486, 222–227 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
86.Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol.42, 293–304 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
87.Ostner, J. et al. BacSC: A general workflow for bacterial single-cell RNA sequencing data analysis. 2024.2006.2022.600071, 10.1101/2024.06.22.600071 (2024).
88.Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol.19, 15 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
89.Rossmeisl, M., Rim, J. S., Koza, R. A. & Kozak, L. P. Variation in type 2 diabetes-related traits in mouse strains susceptible to diet-induced obesity. Diabetes52, 1958–1966 (2003). [DOI] [PubMed]
90.Clee, S. M. & Attie, A. D. The Genetic landscape of type 2 diabetes in mice. Endocrine Rev.28, 48–83 (2007). [DOI] [PubMed]
91.Suriano, F. et al. Novel insights into the genetically obese (ob/ob) and diabetic (db/db) mice: two sides of the same coin. Microbiome9, 147 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
92.Gill, S. R. et al. Metagenomic analysis of the human distal gut microbiome. Science. (N. Y., N. Y.)312, 1355–1359 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
93.Nagpal, S. & Srivastava, S. K. Colon or semicolon: gut sampling microdevices for omics insights. npj Biofilms Microbiomes10, 97 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
94.Mohammed, H. et al. Passive intestinal microbiome sampling using an ingestible device with tortuous lattices. Device3, 10.1016/j.device.2025.100904 (2025).
95.Ten-Doménech, I. et al. Comparing Targeted vs. Untargeted MS(2) Data-Dependent Acquisition for Peak Annotation in LC-MS Metabolomics. Metabolites10, 126 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
96.Smith, C. A., Want, E. J., O’Maille, G., Abagyan, R. & Siuzdak, G. XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal. Chem.78, 779–787 (2006). [DOI] [PubMed] [Google Scholar]
97.Kuhl, C., Tautenhahn, R., Böttcher, C., Larson, T. R. & Neumann, S. CAMERA: an integrated strategy for compound spectra extraction and annotation of liquid chromatography/mass spectrometry data sets. Anal. Chem.84, 283–289 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
98.Wen, B., Mei, Z., Zeng, C. & Liu, S. metaX: a flexible and comprehensive software for processing metabolomics data. BMC Bioinforma.18, 183 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
99.Gromski, P. S. et al. Influence of missing values substitutes on multivariate analysis of metabolomics data. Metabolites4, 433–452 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
100.Lu, J., Breitwieser, F. P., Thielen, P. & Salzberg, S. L. Bracken: estimating species abundance in metagenomics data. PeerJ. Computer Sci.3, e104 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
101.Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinforma. (Oxf., Engl.)29, 15–21 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
102.Xu, Z. Microbial Single-Cell Transcriptomics Links Gut Microbiota Functional States to Metabolic Changes in Male Mice,10.5281/zenodo.20740378 (2026). [DOI] [PMC free article] [PubMed]
103.Hanson, N. W., Konwar, K. M. & Hallam, S. J. LCA*: an entropy-based measure for taxonomic assignment within assembled metagenomes. Bioinforma. (Oxf., Engl.)32, 3535–3542 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
104.Liu, Y., Ghaffari, M. H., Ma, T. & Tu, Y. Impact of database choice and confidence score on the performance of taxonomic classification using Kraken2. aBIOTECH5, 465–475 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
105.Almeida, A. et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat. Biotechnol.39, 105–114 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
106.Wolock, S. L., Lopez, R. & Klein, A. M. Scrublet: computational identification of cell doublets in single-cell transcriptomic data. Cell Syst.8, 281–291.e289 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
107.Xu, S. et al. Using clusterProfiler to characterize multiomics data. Nat. Protoc.19, 3292–3320 (2024). [DOI] [PubMed] [Google Scholar]
108.Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res.43, e47 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
109.Gu, Z., Gu, L., Eils, R., Schlesner, M. & Brors, B. circlize Implements and enhances circular visualization in R. Bioinforma. (Oxf., Engl.)30, 2811–2812 (2014). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information^{(12.4MB, pdf)}

Reporting Summary^{(107.6KB, pdf)}

Transparent Peer Review file^{(761.4KB, pdf)}

Source Data^{(74.9MB, zip)}

Data Availability Statement

[CR1] 1.The Human Microbiome Project Consortium. Structure, function and diversity of the healthy human microbiome. Nature. 486, 207–214 (2012). [DOI] [PMC free article] [PubMed]

[CR2] 2.Mukhopadhya, I. & Louis, P. Gut microbiota-derived short-chain fatty acids and their role in human health and disease. Nat. Rev. Microbiol. 23, 635–651 (2025). [DOI] [PubMed] [Google Scholar]

[CR3] 3.Fan, Y. & Pedersen, O. Gut microbiota in human metabolic health and disease. Nat. Rev. Microbiol.19, 55–71 (2021). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Li, P., Roos, S., Luo, H., Ji, B. & Nielsen, J. Metabolic engineering of human gut microbiome: Recent developments and future perspectives. Metab. Eng.79, 1–13 (2023). [DOI] [PubMed] [Google Scholar]

[CR5] 5.Meijnikman, A. S., Gerdes, V. E., Nieuwdorp, M. & Herrema, H. Evaluating causality of gut microbiota in obesity and diabetes in humans. Endocr. Rev.39, 133–153 (2018). [DOI] [PubMed] [Google Scholar]

[CR6] 6.Howard, E. J., Lam, T. K. T. & Duca, F. A. The gut microbiome: connecting diet, glucose homeostasis, and disease. Annu. Rev. Med.73, 469–481 (2022). [DOI] [PubMed] [Google Scholar]

[CR7] 7.McCallum, G. & Tropini, C. The gut microbiota and its biogeography. Nat. Rev. Microbiol.22, 105–118 (2024). [DOI] [PubMed] [Google Scholar]

[CR8] 8.Ntekas, I. et al. Spatial transcriptomics maps host-gut microbiome biogeography at high resolution. Nat. Microbiol.11, 1193–1204 (2026). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Pountain, A. W. & Yanai, I. Dissecting microbial communities with single-cell transcriptome analysis. Science (New York, N.Y.)389, eadp6252 (2025). [DOI] [PMC free article] [PubMed]

[CR10] 10.Qin, J. et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature490, 55–60 (2012). [DOI] [PubMed] [Google Scholar]

[CR11] 11.Mei, Z. et al. Strain-specific gut microbial signatures in type 2 diabetes identified in a cross-cohort analysis of 8,117 metagenomes. Nat. Med.30, 2265–2276 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Blattman, S. B., Jiang, W., Oikonomou, P. & Tavazoie, S. Prokaryotic single-cell RNA sequencing by in situ combinatorial indexing. Nat. Microbiol.5, 1192–1201 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Kuchina, A. et al. Microbial single-cell RNA sequencing by split-pool barcoding. Science371, 5257 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR14] 14.McNulty, R. et al. Probe-based bacterial single-cell RNA sequencing predicts toxin regulation. Nat. Microbiol.8, 934–945 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Ma, P. et al. Bacterial droplet-based single-cell RNA-seq reveals antibiotic-associated heterogeneous cellular states. Cell186, 877–891 e814 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Wang, B. et al. Single-cell massively-parallel multiplexed microbial sequencing (M3-seq) identifies rare bacterial populations and profiles phage infection. Nat. Microbiol8, 1846–1862 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Xu, Z. et al. Droplet-based high-throughput single microbe RNA sequencing by smRandom-seq. Nat. Commun.14, 5130 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Xu, Z., Wang, Y., Cai, W., Chen, Y. & Wang, Y. Single microorganism RNA sequencing of microbiomes using smRandom-Seq. Nat. Protoc.21, 160–199 (2025). [DOI] [PubMed] [Google Scholar]

[CR19] 19.Shen, Y. et al. High-throughput single-microbe RNA sequencing reveals adaptive state heterogeneity and host-phage activity associations in human gut microbiome. Protein Cell16, 211–226 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Shen, Y. et al. Single-microbe RNA sequencing uncovers unexplored specialized metabolic functions of keystone species in the human gut. Imeta4, e70035 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Jia, M. et al. Single-cell transcriptomics across 2,534 microbial species reveals functional heterogeneity in the rumen microbiome. Nat. Microbiol.9, 1884–1898 (2024). [DOI] [PubMed] [Google Scholar]

[CR22] 22.Martinez-Guryn, K., Leone, V. & Chang, E. B. Regional Diversity of the Gastrointestinal Microbiome. Cell Host Microbe26, 314–324 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR23] 23.Folz, J. et al. Human metabolome variation along the upper intestinal tract. Nat. Metab.5, 777–788 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR24] 24.Shalon, D. et al. Profiling the human intestinal environment under physiological conditions. Nature617, 581–591 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Hugenholtz, F. & de Vos, W. M. Mouse models for human intestinal microbiota research: a critical evaluation. Cell. Mol. Life Sci.75, 149–160 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Meier, K. H. U. et al. Metabolic landscape of the male mouse gut identifies different niches determined by microbial activities. Nat. Metab.5, 968–980 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Yu, F. et al. Abnormal gut microbiota composition contributes to the development of type 2 diabetes mellitus in db/db mice. Aging11, 10454–10467 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Yu, F. et al. Gut microbiota transplantation from db/db mice induces diabetes-like phenotypes and alterations in Hippo signaling in pseudo germ-free mice. Aging12, 24156–24167 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Lyu, Y. et al. Effects of combination treatment with metformin and berberine on hypoglycemic activity and gut microbiota modulation in db/db mice. Phytomedicine: Int. J. Phytother. Phytopharmacol.101, 154099 (2022). [DOI] [PubMed] [Google Scholar]

[CR30] 30.Liu, R. et al. Study on the mechanism of modified Gegen Qinlian decoction in regulating the intestinal flora-bile acid-TGR5 axis for the treatment of type 2 diabetes mellitus based on macro genome sequencing and targeted metabonomics integration. Phytomedicine: Int. J. Phytother. Phytopharm.132, 155329 (2024). [DOI] [PubMed] [Google Scholar]

[CR31] 31.Beresford-Jones, B. S. et al. The mouse gastrointestinal bacteria catalogue enables translation between the mouse and human gut microbiotas via functional mapping. Cell Host Microbe30, 124–138.e128 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Kieser, S., Zdobnov, E. M. & Trajkovski, M. Comprehensive mouse microbiota genome catalog reveals major difference to its human counterpart. PLOS Comput. Biol.18, e1009947 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Kim, N. et al. MRGM: an enhanced catalog of mouse gut microbial genomes substantially broadening taxonomic and functional landscapes. Gut microbes16, 2393791 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Gurbich, T. A. et al. MGnify genomes: a resource for biome-specific microbial genome catalogues. J. Mol. Biol.435, 168016 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Xiao, L. et al. A catalog of the mouse gut metagenome. Nat. Biotechnol.33, 1103–1108 (2015). [DOI] [PubMed] [Google Scholar]

[CR36] 36.Lesker, T. R. et al. An Integrated Metagenome Catalog Reveals New Insights into the Murine Gut Microbiome. Cell Rep.30, 2909–2922.e2906 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR37] 37.O’Leary, N. A. et al. Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation. Nucleic Acids Res.44, D733-D745 (2015). [DOI] [PMC free article] [PubMed]

[CR38] 38.Kanehisa, M. & Goto, S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res28, 27–30 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Jain, C., Rodriguez-R, L. M., Phillippy, A. M., Konstantinidis, K. T. & Aluru, S. High throughput ANI analysis of 90K prokaryotic genomes reveals clear species boundaries. Nat. Commun.9, 5114 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR40] 40.Wang, Z. et al. Hosts manipulate lifestyle switch and pathogenicity heterogeneity of opportunistic pathogens in the single-cell resolution. Elife13, RP96789 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR41] 41.Korshoj, L. E. & Kielian, T. Bacterial single-cell RNA sequencing captures biofilm transcriptional heterogeneity and differential responses to immune pressure. Nat. Commun.15, 10184 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR42] 42.Gholami, H., Chmiel, J. A., Burton, J. P. & Maleki Vareki, S. The role of microbiota-derived vitamins in immune homeostasis and enhancing cancer immunotherapy. Cancers15, 1300 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Arumugam, M. et al. Enterotypes of the human gut microbiome. Nature473, 174–180 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Tropini, C. et al. Transient osmotic perturbation causes long-term alteration to the gut microbiota. Cell173, 1742–1754.e1717 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Malaguarnera, G., Giordano, M., Nunnari, G., Bertino, G. & Malaguarnera, M. Gut microbiota in alcoholic liver disease: pathogenetic role and therapeutic perspectives. World J. Gastroenterol.20, 16639–16648 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR46] 46.Raina, S. & Klein, G. Lipopolysaccharide: recent advances in its biosynthesis and controlling cell envelope homeostasis. Int. J. Mol. Sci.26, 7705 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Morrison, D. J. & Preston, T. Formation of short chain fatty acids by the gut microbiota and their impact on human metabolism. Gut Microbes7, 189–200 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Sereti, I. et al. Impaired gut microbiota-mediated short-chain fatty acid production precedes morbidity and mortality in people with HIV. Cell Rep.42, 113336 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR49] 49.Wewalka, M., Patti, M. E., Barbato, C., Houten, S. M. & Goldfine, A. B. Fasting serum taurine-conjugated bile acids are elevated in type 2 diabetes and do not change with intensification of insulin. J. Clin. Endocrinol. Metab.99, 1442–1451 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR50] 50.Serena, C. et al. Elevated circulating levels of succinate in human obesity are linked to specific gut microbiota. ISME J.12, 1642–1657 (2018). [DOI] [PMC free article] [PubMed]

[CR51] 51.Haeusler, R. A., Astiarraga, B., Camastra, S., Accili, D. & Ferrannini, E. Human insulin resistance is associated with increased plasma levels of 12α-hydroxylated bile acids. Diabetes62, 4184–4191 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Duft, R. G. et al. Serum metabolites associated with increased insulin resistance and low cardiorespiratory fitness in overweight adolescents. Nutr., Metab. Cardiovasc. Dis.32, 269–278 (2022). [DOI] [PubMed] [Google Scholar]

[CR53] 53.Li, M. et al. Trends in insulin resistance: insights into mechanisms and therapeutic strategy. Signal Transduct. Target. Ther.7, 216 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Fernández-Veledo, S., Marsal-Beltran, A. & Vendrell, J. Type 2 diabetes and succinate: unmasking an age-old molecule. Diabetologia67, 430–442 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Würtz, P. et al. Metabolic signatures of insulin resistance in 7,098 young adults. Diabetes61, 1372–1380 (2012). [DOI] [PMC free article] [PubMed]

[CR56] 56.Guasch-Ferré, M. et al. Metabolomics in prediabetes and diabetes: a systematic review and meta-analysis. Diab. Care39, 833–846 (2016). [DOI] [PMC free article] [PubMed]

[CR57] 57.Mihalik, S. J. et al. Metabolomic profiling of fatty acid and amino acid metabolism in youth with obesity and type 2 diabetes: evidence for enhanced mitochondrial oxidation. Diabetes Care35, 605–611 (2012). [DOI] [PMC free article] [PubMed]

[CR58] 58.Ling, Z.-N. et al. Amino acid metabolism in health and disease. Signal Transduct. Target. Ther.8, 345 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR59] 59.Anand, U. & Anand, C. V. Connecting links between the urea cycle and the TCA cycle: a tutorial exercise. Biochem. Educ.27, 153–154 (1999). [Google Scholar]

[CR60] 60.Yeh, S.-L., Shih, Y.-M. & Lin, M.-T. in Diabetes (Second Edition) (ed Victor R. Preedy) 255-264 (Academic Press, 2020).

[CR61] 61.Davila, A.-M. et al. Re-print of “Intestinal luminal nitrogen metabolism: Role of the gut microbiota and consequences for the host. Pharmacol. Res.69, 114–126 (2013). [DOI] [PubMed] [Google Scholar]

[CR62] 62.Shen, J. et al. Cross-tissue multi-omics analyses reveal the gut microbiota’s absence impacts organ morphology, immune homeostasis, bile acid and lipid metabolism. Imeta4, e272 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR63] 63.Egalini, F. et al. The Effects of Omega 3 and Omega 6 Fatty Acids on Glucose Metabolism: An Updated Review. Nutrients15, 2672 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR64] 64.Lu, J. et al. Association of Serum Bile Acids Profile and Pathway Dysregulation With the Risk of Developing Diabetes Among Normoglycemic Chinese Adults: Findings From the 4C Study. Diabetes Care44, 499–510 (2020). [DOI] [PubMed]

[CR65] 65.Mozaffarian, D. Saturated fatty acids and type 2 diabetes: more evidence to re-invent dietary guidelines. Lancet Diab Endocrinol.2, 770–772 (2014). [DOI] [PubMed] [Google Scholar]

[CR66] 66.Huang, L. et al. Circulating Saturated Fatty Acids and Incident Type 2 Diabetes: A Systematic Review and Meta-Analysis. Nutrients11, 998 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR67] 67.Lampousi, A.-M. et al. Interaction between plasma phospholipid odd-chain fatty acids and GAD65 autoantibodies on the incidence of adult-onset diabetes: the EPIC-InterAct case-cohort study. Diabetologia66, 1460–1471 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR68] 68.Yao, L. et al. A selective gut bacterial bile salt hydrolase alters host metabolism. eLife7, e37182 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR69] 69.Depommier, C. et al. Supplementation with Akkermansia muciniphila in overweight and obese human volunteers: a proof-of-concept exploratory study. Nat. Med.25, 1096–1103 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR70] 70.Li, W. Z., Stirling, K., Yang, J. J. & Zhang, L. Gut microbiota and diabetes: From correlation to causality and mechanism. World J. Diab11, 293–308 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR71] 71.Takeuchi, T. et al. Gut microbial carbohydrate metabolism contributes to insulin resistance. Nature621, 389–395 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR72] 72.Miyake, S., Ding, Y., Soh, M., Low, A. & Seedorf, H. Muribaculum gordoncarteri sp. nov., an anaerobic bacterium from the faeces of C57BL/6J mice. Int. J. Syst. Evolut. Microbiol.70, 4725–4729 (2020). [DOI] [PubMed] [Google Scholar]

[CR73] 73.Zhang, Y., Thompson, K. N., Huttenhower, C. & Franzosa, E. A. Statistical approaches for differential expression analysis in metatranscriptomics. Bioinforma. (Oxf., Engl.)37, i34–i41 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR74] 74.He, J.-L. et al. Enhanced interactions among gut mycobiomes with the deterioration of glycemic control. Med5, 909–925.e907 (2024). [DOI] [PubMed] [Google Scholar]

[CR75] 75.Huang, H. et al. The mycobiome as integral part of the gut microbiome: crucial role of symbiotic fungi in health and disease. Gut microbes16, 2440111 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR76] 76.Lu, J. et al. Metagenome analysis using the Kraken software suite. Nat. Protoc.17, 2815–2839 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR77] 77.Blanco-Míguez, A. et al. Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4. Nat. Biotechnol.41, 1633–1644 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR78] 78.Delbaere, K. et al. The small intestine: dining table of host-microbiota meetings. FEMS Microbiol. Rev.47, 10.1093/femsre/fuad022 (2023). [DOI] [PMC free article] [PubMed]

[CR79] 79.Steinbach, E. et al. Upper small intestine microbiome in obesity and related metabolic disorders: A new field of investigation. Metabolism-Clin. Exp.150, 10.1016/j.metabol.2023.155712 (2024). [DOI] [PubMed]

[CR80] 80.Karlsson, F. H. et al. Gut metagenome in European women with normal, impaired and diabetic glucose control. Nature498, 99–103 (2013). [DOI] [PubMed] [Google Scholar]

[CR81] 81.Liu, X. et al. High plasma glutamate and low glutamine-to-glutamate ratio are associated with type 2 diabetes: Case-cohort study within the PREDIMED trial. Nutr., Metab. Cardiovasc. Dis.29, 1040–1049 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR82] 82.Dollet, L. et al. Glutamine regulates skeletal muscle immunometabolism in type 2 diabetes. Diabetes71, 624–636 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR83] 83.Zhu, Y. et al. Exploration of the muribaculaceae family in the gut microbiota: diversity, metabolism, and function. Nutrients16, 2660 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR84] 84.Zhang, M. et al. Microbiota-derived urocanic acid triggered by tyrosine kinase inhibitors potentiates cancer immunotherapy efficacy. Cell Host Microbe33, 915–931.e919 (2025). [DOI] [PubMed] [Google Scholar]

[CR85] 85.Yatsunenko, T. et al. Human gut microbiome viewed across age and geography. Nature486, 222–227 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR86] 86.Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol.42, 293–304 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR87] 87.Ostner, J. et al. BacSC: A general workflow for bacterial single-cell RNA sequencing data analysis. 2024.2006.2022.600071, 10.1101/2024.06.22.600071 (2024).

[CR88] 88.Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol.19, 15 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR89] 89.Rossmeisl, M., Rim, J. S., Koza, R. A. & Kozak, L. P. Variation in type 2 diabetes-related traits in mouse strains susceptible to diet-induced obesity. Diabetes52, 1958–1966 (2003). [DOI] [PubMed]

[CR90] 90.Clee, S. M. & Attie, A. D. The Genetic landscape of type 2 diabetes in mice. Endocrine Rev.28, 48–83 (2007). [DOI] [PubMed]

[CR91] 91.Suriano, F. et al. Novel insights into the genetically obese (ob/ob) and diabetic (db/db) mice: two sides of the same coin. Microbiome9, 147 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR92] 92.Gill, S. R. et al. Metagenomic analysis of the human distal gut microbiome. Science. (N. Y., N. Y.)312, 1355–1359 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR93] 93.Nagpal, S. & Srivastava, S. K. Colon or semicolon: gut sampling microdevices for omics insights. npj Biofilms Microbiomes10, 97 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR94] 94.Mohammed, H. et al. Passive intestinal microbiome sampling using an ingestible device with tortuous lattices. Device3, 10.1016/j.device.2025.100904 (2025).

[CR95] 95.Ten-Doménech, I. et al. Comparing Targeted vs. Untargeted MS(2) Data-Dependent Acquisition for Peak Annotation in LC-MS Metabolomics. Metabolites10, 126 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR96] 96.Smith, C. A., Want, E. J., O’Maille, G., Abagyan, R. & Siuzdak, G. XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal. Chem.78, 779–787 (2006). [DOI] [PubMed] [Google Scholar]

[CR97] 97.Kuhl, C., Tautenhahn, R., Böttcher, C., Larson, T. R. & Neumann, S. CAMERA: an integrated strategy for compound spectra extraction and annotation of liquid chromatography/mass spectrometry data sets. Anal. Chem.84, 283–289 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR98] 98.Wen, B., Mei, Z., Zeng, C. & Liu, S. metaX: a flexible and comprehensive software for processing metabolomics data. BMC Bioinforma.18, 183 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR99] 99.Gromski, P. S. et al. Influence of missing values substitutes on multivariate analysis of metabolomics data. Metabolites4, 433–452 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR100] 100.Lu, J., Breitwieser, F. P., Thielen, P. & Salzberg, S. L. Bracken: estimating species abundance in metagenomics data. PeerJ. Computer Sci.3, e104 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR101] 101.Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinforma. (Oxf., Engl.)29, 15–21 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR102] 102.Xu, Z. Microbial Single-Cell Transcriptomics Links Gut Microbiota Functional States to Metabolic Changes in Male Mice,10.5281/zenodo.20740378 (2026). [DOI] [PMC free article] [PubMed]

[CR103] 103.Hanson, N. W., Konwar, K. M. & Hallam, S. J. LCA*: an entropy-based measure for taxonomic assignment within assembled metagenomes. Bioinforma. (Oxf., Engl.)32, 3535–3542 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR104] 104.Liu, Y., Ghaffari, M. H., Ma, T. & Tu, Y. Impact of database choice and confidence score on the performance of taxonomic classification using Kraken2. aBIOTECH5, 465–475 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR105] 105.Almeida, A. et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat. Biotechnol.39, 105–114 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR106] 106.Wolock, S. L., Lopez, R. & Klein, A. M. Scrublet: computational identification of cell doublets in single-cell transcriptomic data. Cell Syst.8, 281–291.e289 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR107] 107.Xu, S. et al. Using clusterProfiler to characterize multiomics data. Nat. Protoc.19, 3292–3320 (2024). [DOI] [PubMed] [Google Scholar]

[CR108] 108.Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res.43, e47 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR109] 109.Gu, Z., Gu, L., Eils, R., Schlesner, M. & Brors, B. circlize Implements and enhances circular visualization in R. Bioinforma. (Oxf., Engl.)30, 2811–2812 (2014). [DOI] [PubMed] [Google Scholar]

PERMALINK

Microbial single-cell transcriptomics links gut microbiota functional states to metabolic changes in male mice

Ziye Xu

Xin Long

Mengdi Song

Sanbao Zhang

Xiaoyue Li

Feifei Lv

Tianyu Zhang

Ting Cao

Yongcheng Wang

Abstract

Introduction

Results

Study design and integrated microbial single-cell transcriptomic-metabolomic analysis framework

Fig. 1. Study design and integrated microbial single-cell transcriptomic-metabolomic analysis framework.

Reference genome choice determines the taxonomic resolution achievable in microbial scRNA-seq

Fig. 2. Impact of reference genome choice on taxonomic resolution in microbial single-cell RNA-seq.

Functional annotation characterizes microbial transcriptional heterogeneity in the mouse gut microbiome

Fig. 3. Functional annotation of microbial single-cell transcriptomes characterizes transcriptional heterogeneity in the mouse gut microbiome.

Cross-omics integration identifies metabolic changes in the mouse gut microbiome

Fig. 4. Cross-omics integration links microbial functional states to gut metabolic changes.

Community-embedded functional states highlight cross-species functional complementation within the gut microbiota

Fig. 5. Community-embedded functional states highlight cross-species functional complementation within the gut microbiota.

Subpopulation heterogeneity of M. gordoncarteri links disrupted nitrogen and stress programs to host nitrogen imbalance

Fig. 6. Subpopulation analysis of M. gordoncarteri shows disrupted nitrogen and stress programs linked to host nitrogen imbalance.

Discussion

Methods

Mice and samples

Microbial single-cell RNA-seq for murine luminal contents

Metabolomics profiling and data processing

Metagenomics profiling

Preprocessing of microbial single-cell RNA-seq raw data

Evaluation of reference genome resources for mouse gut microbial single-cell transcriptomic analysis

Taxonomic classification of microbial single-cell RNA-seq data using smClassify

Benchmarking of taxonomic profiling methods

Comparison of metagenomic and microbial single-cell results

Reduction, clustering, and differential gene expression analysis of microbial single-cell RNA-seq data

KEGG pathway enrichment analysis of microbial single-cell RNA-seq data

Metabolite differential and enrichment analysis

Integration of single-cell transcriptomics and metabolomics

Inter-species correlation analysis of microbial single-cell RNA-seq data

Subcluster analysis of microbial single-cell transcriptomes

Pseudotime trajectory inference of microbial single-cell RNA-seq data

Pseudobulk differential expression analysis

Statistics & reproducibility

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions

Peer review

Peer review information

Funding

Data availability

Code availability

Competing interests

Footnotes

Contributor Information

Supplementary information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases