Commensal-derived acetylcholine enhances mucosal immune education Authors: Deguang Song, Brianna Duncan-Lowey, Varnica Khetrapal, Randy Hamchand, Tong Deng, Hailey Brown, Anchi Wu, Anjelica L. Martin, Kaylyn M. Bauer, Yanyu Zhao, Mytien T. Nguyen, Nicole D. Sonnert, Shana R. Leopold, Qihao Wu, Jason M. Crawford, Noah W. Palm Published: 2026 ABSTRACT The microbiota produces thousands of potentially bioactive small molecules1–3. High-throughput bioactivity screens of in vitro commensal cultures have exposed microbiota metabolites that shape host physiology by activating diverse G-protein-coupled receptors (GPCRs)4–7. However, owing to technical limitations, the GPCRome-wide bioactivities of in vivo metabolomes, which result from complex diet–microorganism–host interactions, remain unclear. Here we used a multiplexed GPCR screening technology to assess GPCRome-wide bioactivities of 100 commensal strains grown in vivo in monoassociated germ-free mice or in vitro in bacterial culture medium. In vivo and in vitro commensal metabolomes exhibited distinct GPCR activation patterns due to (1) host-mediated metabolite degradation; (2) in vivo microbial metabolic reprogramming; and (3) biotransformation of dietary substrates. Notably, we found that multiple commensal strains produced acetylcholine (ACh) in vivo through the conversion of dietary choline, including select Bifidobacterium strains that dominate the microbiome in early life and a probiotic Pediococcus strain. Mechanistically, we identified and characterized the bacterial enzymes that mediate this biotransformation in Bifidobacterium breve and Pediococcus pentosaceus, and generated an isogenic mutant B. breve strain lacking ACh production. Mice colonized with ACh-producing B. breve exhibited enhanced intestinal immunoglobulin A (IgA) production, altered microbiota composition and increased resistance to enteric infection. These findings underscore the profound impacts of the in vivo environment on microbiota metabolism and reveal a diet–microbiome–host axis that strengthens mucosal immune defences and reinforces host–microbiota mutualism. Main The human gut microbiota produces thousands of small-molecule metabolites through de novo synthesis and through the biotransformation of chemicals derived from our diet or produced and secreted by the host8–10. These microbiota metabolites have essential roles in regulating diverse physiological functions, including energy metabolism, inflammation and mucosal barrier integrity11–13. Previous studies from our laboratory and others have used high-throughput GPCR screens of supernatants from commensal strains grown in vitro (that is, in vitro commensal metabolomes) to identify impacts of microbiota metabolites on host physiology4,6. However, the chemical diversity and physiological relevance of these in vitro metabolomes are constrained by the simplicity of the cultivation medium and the unavoidable artificiality of in vitro growth conditions. By contrast, in vivo microbiota metabolomes reflect the dynamic environment of the mammalian gastrointestinal tract, in which commensal microorganisms are exposed to complex environmental cues and precursor chemicals from the host, diet and other microorganisms. However, due to the requirement for large sample volumes, it is not possible to perform unbiased receptorome-wide GPCR screening of in vivo microbiota metabolomes using conventional approaches. To overcome this limitation, we recently developed a highly multiplexed GPCR screening platform (PRESTO-Salsa) that enables receptorome-wide assessment of nearly all non-olfactory GPCRs (>300 receptors) in a single well of a 96-well plate5 and used this technology to directly compare bioactivities of in vitro and in vivo commensal metabolomes. In vitro and in vivo metabolome screens The microbial compositions of commensal communities grown in vitro differ substantially from those grown in vivo, making it infeasible to compare complex microbial communities grown under these two conditions. Thus, to enable direct comparisons between in vitro and in vivo metabolomes, we compared metabolomes from individual strains grown either in standard culture medium or in monocolonized gnotobiotic mice. We selected 100 phylogenetically diverse microbiota strains from public and internal strain collections (Supplementary Table 1). For in vitro bacterial metabolome libraries, we cultured each bacterial strain in medium and extracted using three different solvents known to enrich for distinct metabolite classes. For in vivo metabolomes, germ-free (GF) C57BL/6 mice were monoassociated with individual strains for 2 weeks, after which caecal contents were collected and extracted using the same solvents. We then screened the resulting 597 enriched commensal metabolomes (297 from in vitro cultures and 300 from mono-associated mice) against nearly all conventional GPCRs using PRESTO-Salsa (Fig. 1a and Supplementary Table 2).Fig. 1In vivo and in vitro commensal metabolomes exhibit distinct GPCR activation patterns.a, Schematic of generating commensal metabolomes from bacteria cultured in vitro in medium and in vivo in monoassociated GF mice, followed by GPCRome interaction screening using the multiplexed GPCR screening technology PRESTO-Salsa. b, The number of in vitro or in vivo commensal metabolomes activating each GPCR. c–e, Representative models of in vivo-mediated alterations in GPCR-active commensal metabolites. c, Host MAOs degrade bacterially derived PEA in vivo. MAOi, MAO inhibitor; RLU, relative luciferase units. d, In vivo metabolic reprogramming suppresses succinate or histamine production. e, Diet-derived choline enables bacterial ACh production in vivo. In c–e, GPCR activity data from in vitro samples are representative of three independent experiments (n = 3 biologically independent samples per group in each experiment); GPCR activity data from in vivo samples are representative of two independent experiments (n = 3 mice per group in each experiment). For c–e, data are mean ± s.e.m. Statistical significance was assessed using two-sided Welch’s t-tests (d and e and in vitro comparison in c) and Welch’s analysis of variance (ANOVA) followed by Games–Howell post hoc test (in vivo comparison in c). The diagrams in a and c–e were created using BioRender; Song. D https://BioRender.com/x0hrt5l (a) and https://BioRender.com/sqnbzth (c–e) (2026).Source data In vivo and in vitro commensal metabolomes exhibit distinct GPCR activation patterns. a, Schematic of generating commensal metabolomes from bacteria cultured in vitro in medium and in vivo in monoassociated GF mice, followed by GPCRome interaction screening using the multiplexed GPCR screening technology PRESTO-Salsa. b, The number of in vitro or in vivo commensal metabolomes activating each GPCR. c–e, Representative models of in vivo-mediated alterations in GPCR-active commensal metabolites. c, Host MAOs degrade bacterially derived PEA in vivo. MAOi, MAO inhibitor; RLU, relative luciferase units. d, In vivo metabolic reprogramming suppresses succinate or histamine production. e, Diet-derived choline enables bacterial ACh production in vivo. In c–e, GPCR activity data from in vitro samples are representative of three independent experiments (n = 3 biologically independent samples per group in each experiment); GPCR activity data from in vivo samples are representative of two independent experiments (n = 3 mice per group in each experiment). For c–e, data are mean ± s.e.m. Statistical significance was assessed using two-sided Welch’s t-tests (d and e and in vitro comparison in c) and Welch’s analysis of variance (ANOVA) followed by Games–Howell post hoc test (in vivo comparison in c). The diagrams in a and c–e were created using BioRender; Song. D https://BioRender.com/x0hrt5l (a) and https://BioRender.com/sqnbzth (c–e) (2026). Source data Consistent with our previous findings, select in vitro metabolomes activated adrenergic receptors (ARs), dopamine receptors (DRDs), histamine receptors (HRHs), serotonin receptors (HTRs), formyl peptide receptors (FPRs) and the succinate receptor4,5 (SUCNR1; Fig. 1b). Extraction-based enrichments also revealed activities at additional receptors that were not detected previously with crude supernatants, including the melatonin receptors (MTNR1A and MTNR1B), cysteinyl leukotriene receptor 2 (CYSLTR2), hydroxycarboxylic acid receptor 3 (HCA3), bradykinin receptor B1 (BDKBR1) and GPR120 (Fig. 1b and Supplementary Table 3). Compared with in vitro metabolomes, corresponding in vivo-derived commensal metabolomes exhibited distinct GPCR activation patterns. Among the GPCRs activated by in vitro metabolomes, most were also activated by in vivo metabolomes, including DRDs, ARs, HRHs, FPRs, HTRs, SUCNR1 and GPR120 (Fig. 1b), although the in vivo metabolomes of fewer strains exhibited these activities. Notably, select strains exhibited unique receptor engagements in vivo but not in vitro, particularly for the ARs α2A (also known as ADRA2A), ADRA2C, DRD2, HRH3 and HTR4 (Fig. 1b, Extended Data Fig. 1a,b and Supplementary Table 3). We created a publicly accessible interactive website (https://nwplab.shinyapps.io/commensal-vitro-vivo-metabolome-gpcr/) that highlights these differences and enables exploration of all screening data. Overall, the major differences in GPCR activation patterns between in vitro and in vivo metabolomes fell into at least three non-mutually-exclusive potential mechanistic categories (additional details and examples are provided in the Supplementary Note). First, host enzymes can degrade lumenal microbiota-derived metabolites, limiting their availability in vivo. For example, Morganella morganii NP17-derived phenethylamine (PEA) activated DRDs and ARs in vitro but not in vivo unless host monoamine oxidases (MAOs) were inhibited (Fig. 1c and Extended Data Fig. 2). Second, in vivo environments can induce reprogramming of microbial metabolism. For example, SUCNR1 activation was frequently observed in vitro but often diminished in vivo, in part due to dietary vitamin-B12-dependent metabolic rewiring in taxa such as Akkermansia muciniphila (Fig. 1d and Extended Data Fig. 3). A similar in vitro-to-in vivo shift was observed for histamine-related activities. Some commensals activated HRHs in vitro but not in vivo (Fig. 1d and Extended Data Fig. 4), and transcriptional analysis of Citrobacter portucalensis NWP28 revealed reduced expression of genes involved in amino acid decarboxylation and histidine uptake after gut colonization (Extended Data Fig. 5). Finally, biotransformations of diet- or host-derived substrates that are present in vivo and absent from standard culture medium can lead to the emergence of unique in vivo bioactivities. For example, multiple in vivo commensal metabolomes, but not their corresponding in vitro metabolomes, activated the muscarinic ACh receptors (CHRMs) (Fig. 1b,e), suggesting that microbiota strains may produce CHRM ligands by biotransforming host- or diet-derived precursors. Select commensals produce ACh in vivo Our GPCRome-wide screening studies revealed that in vivo metabolomes from multiple bacterial strains activated the CHRMs (as represented by CHRM4), including Bifidobacterium longum NWP308, Bifidobacterium adolescentis NWP330, Bacteroides caccae NWP468, Bacteroides fragilis NWP229, Bacteroides dorei NWP244, Ruminococcus gnavus ATCC29149, P. pentosaceus NP38 and Streptococcus parasanguinis NWP114 (Fig. 2a). Notably, none of these strains activated CHRM4 when cultured in vitro (Fig. 2a).Fig. 2Select gut commensal strains produce ACh.a, CHRM4 activation (indicated by log2-transformed fold change (log2[FC])) by in vitro (bottom) and in vivo (top) commensal metabolomes as measured using PRESTO-Salsa. n = 3 biologically independent samples. b, CHRM4 activation by in vitro versus in vivo metabolomes from representative bacteria measured using PRESTO-Tango. c, Mass spectrometry quantification of ACh in the bacterial metabolomes analysed in b. d, CHRM4 activation by representative bacteria cultured in vitro with 1 mM choline. e, Mass spectrometry quantification of ACh in in vitro bacterial cultures analysed in d. f, CHRM4 activation by diverse Bifidobacterium strains cultured in vitro with 1 mM choline. In b–f, CHRM4 activity data from in vitro samples are representative of three independent experiments (n = 3 biologically independent samples per group in each experiment); CHRM4 activity data from in vivo samples are representative of two independent experiments (n = 3 mice per group in each experiment). ACh quantification data are from one representative independent experiment. Data are mean + s.e.m in a, and mean ± s.e.m in b–f. Statistical significance was assessed using one-way ANOVA followed by Tukey’s post hoc test (b (in vitro)) and Welch’s ANOVA followed by Games–Howell post hoc test for (b (in vivo), d and f). In b–f, statistical comparisons were performed between each bacterial in vitro culture and the medium control, and between monocolonized groups and the GF control. Exact P values are shown in the figure when P < 0.05; all other P values are provided in the Source data. n.d., not detected.Source data Select gut commensal strains produce ACh. a, CHRM4 activation (indicated by log2-transformed fold change (log2[FC])) by in vitro (bottom) and in vivo (top) commensal metabolomes as measured using PRESTO-Salsa. n = 3 biologically independent samples. b, CHRM4 activation by in vitro versus in vivo metabolomes from representative bacteria measured using PRESTO-Tango. c, Mass spectrometry quantification of ACh in the bacterial metabolomes analysed in b. d, CHRM4 activation by representative bacteria cultured in vitro with 1 mM choline. e, Mass spectrometry quantification of ACh in in vitro bacterial cultures analysed in d. f, CHRM4 activation by diverse Bifidobacterium strains cultured in vitro with 1 mM choline. In b–f, CHRM4 activity data from in vitro samples are representative of three independent experiments (n = 3 biologically independent samples per group in each experiment); CHRM4 activity data from in vivo samples are representative of two independent experiments (n = 3 mice per group in each experiment). ACh quantification data are from one representative independent experiment. Data are mean + s.e.m in a, and mean ± s.e.m in b–f. Statistical significance was assessed using one-way ANOVA followed by Tukey’s post hoc test (b (in vitro)) and Welch’s ANOVA followed by Games–Howell post hoc test for (b (in vivo), d and f). In b–f, statistical comparisons were performed between each bacterial in vitro culture and the medium control, and between monocolonized groups and the GF control. Exact P values are shown in the figure when P < 0.05; all other P values are provided in the Source data. n.d., not detected. Source data We selected a representative strain from each phylum to validate the observed CHRM4 activity using CHRM4-Tango: B. longum NWP308 (Actinobacteria), B. caccae NWP468 (Bacteroidetes), P. pentosaceus NP38 (Firmicutes) and Bacteroides ovatus NWP195 as a non-CHRM4-activating control. We monocolonized groups of GF mice with each strain and prepared caecal metabolome libraries 2 weeks after colonization. Caecal metabolomes from mice colonized with putative CHRM4-activating strains triggered CHRM4 activation, while metabolomes from GF or B. ovatus NWP195-colonized mice lacked this activity (Fig. 2b). Consistent with our initial screening results, in vitro metabolomes from all strains failed to activate CHRM4 (Fig. 2b). Mass spectrometry analysis of caecal contents from P. pentosaceus NP38, B. longum NWP308 and B. caccae NWP468-colonized mice revealed high levels of ACh, whereas caecal contents from B. ovatus NWP195-colonized and GF mice showed no detectable ACh (Fig. 2c). Commensal strains that produce ACh in vivo may lack this activity in vitro due to an absence of required chemical precursors in standard culture medium. Notably, all of the in vivo CHRM4-active bacteria identified in our original screen were cultured in gut microbiota medium (GMM)14, which lacks the essential dietary nutrient and ACh precursor choline (Supplementary Table 4). Accordingly, supplementation of GMM with 1 mM choline restored robust ACh production and CHRM4 activation in all tested strains aside from the negative control strain B. ovatus NWP195, with P. pentosaceus NP38 showing the highest activity (Fig. 2d,e). These results indicate that the ACh biosynthetic pathway is functional in vitro but limited by substrate availability under standard culture conditions. Consistent with this, the standard mouse diet used in our facility contains approximately 1,200 mg per kg choline, providing an abundant source of choline to commensal microorganisms grown in vivo. We noted that multiple Bifidobacterium species, which dominate the infant gut microbiota15–17 (Supplementary Fig. 2a,b), activated CHRM4 in our screen. To broadly explore ACh synthesis across different bifidobacteria, we tested 13 representative Bifidobacterium strains from our human gut commensal collections for CHRM4 activation. When cultured in the presence of choline, all B. breve strains, B. longum strains and one Bifidobacterium dentium strain activated CHRM4, with B. breve NWP289 showing the strongest activity (Fig. 2f). B. breve and B. longum are among the most prevalent Bifidobacterium strains in infant microbiomes, while B. dentium is less prevalent but is highly abundant in some individuals15 (Supplementary Fig. 2c). Human breast milk is enriched in choline-containing compounds (approximately 1,400 μM total choline, including 140 μM free choline)18. All tested Bifidobacterium strains converted free choline into ACh; by contrast, most Bifidobacterium strains did not efficiently convert glycerophosphocholine, phosphocholine or sphingomyelins into ACh, although B. dentium NWP234 showed evidence of phosphatidylcholine utilization (Supplementary Fig. 3). Together, these data suggest that infant microbiomes are dominated by ACh-producing Bifidobacterium species exposed to high levels of choline in breast milk or formula, creating a favourable environment for high-level production of microbiota-derived ACh. This prompted us to examine the mechanisms by which commensal species produce ACh and the potential impacts of microbial ACh production on host biology. Mechanisms of commensal ACh synthesis ACh synthesis in mammals is catalysed by choline acetyltransferase (ChAT), which transfers an acetyl group from acetyl-CoA to the hydroxyl substituent of choline (Extended Data Fig. 6a). We could not identify any homologues of mammalian ChAT in the B. breve NWP289 genome. Thus, we next searched for acetyltransferases using rapid annotations using subsystems technology (RAST)19 and identified 14 full-length putative acetyltransferases (Supplementary Table 5). Nine sequences annotated as N-acetyltransferases were excluded from further study. We cloned and recombinantly expressed the other five putative acetyltransferases. After purification (Extended Data Fig. 6b), we tested their ability to catalyse ACh synthesis from acetyl-CoA and choline in vitro (Extended Data Fig. 6c). Products from putative acetyltransferase 2 exhibited potent CHRM4 activity and high ACh levels, while the products of putative acetyltransferases 1 and 5 exhibited only very weak activity and no detectable ACh (Fig. 3a,b). Thus, putative acetyltransferase 2 is the primary enzyme that catalyses ACh synthesis in B. breve NWP289; hereafter, we refer to this enzyme as B. breve ChAT (BbChAT).Fig. 3Gut microorganisms use hexapeptide repeat proteins to synthesize ACh.a,c, CHRM4 activation by products from in vitro enzyme assays using recombinant putative acetyltransferases from B. breve NWP289 (a) or P. pentosaceus NP38 (c) incubated with choline and acetyl-CoA. b,d, Mass spectrometry quantification of ACh in the assay products from a and c, respectively. e, AlphaFold2 models of BbChAT and PpChAT, and crystal structure of EcMoAT (PDB: 1OCX). f, Sequence alignment of hexapeptide repeat acetyltransferase family proteins showing the conservation of the catalytic histidine. B. anthracis, Bacillus anthracis; C. difficile, Clostridioides difficile; G. kaustophilus, Geobacillus kaustophilus; S. aureus, Staphylococcus aureus. g, CHRM4 activation by products from WT and mutant BbChAT and PpChAT assayed under the same conditions as in a and c. h, Phylogenetic tree of the prokaryotic hexapeptide repeat protein superfamily. i, CHRM4 activation by assay products from selected hexapeptide repeat proteins (highlighted in h). j, CHRM4 activation by WT and Δchat B. breve cultured with and without 1 mM choline. k, Mass spectrometry quantification of ACh in cultures of WT and Δchat B. breve grown with 1 mM choline. CHRM4 activity (a, c, g, i and j) and ACh quantification (b, d and k) are representative of two independent experiments (n = 3 biologically independent samples per group in each experiment). Data are mean ± s.e.m. Statistical significance was assessed using Welch’s ANOVA with Games–Howell post hoc test (a, c, d and i), Kruskal–Wallis tests followed by Dunn’s multiple-comparison test (b), two-sided Welch’s t-tests (g) and one-way ANOVA with Tukey’s post hoc test (j). Statistical comparisons were performed between each candidate enzyme and the no-protein control (a–d and i), and between each bacteria culture and the medium control (j). Exact P values are shown in the figure when P < 0.05; all other P values are provided as Source data.Source data Gut microorganisms use hexapeptide repeat proteins to synthesize ACh. a,c, CHRM4 activation by products from in vitro enzyme assays using recombinant putative acetyltransferases from B. breve NWP289 (a) or P. pentosaceus NP38 (c) incubated with choline and acetyl-CoA. b,d, Mass spectrometry quantification of ACh in the assay products from a and c, respectively. e, AlphaFold2 models of BbChAT and PpChAT, and crystal structure of EcMoAT (PDB: 1OCX). f, Sequence alignment of hexapeptide repeat acetyltransferase family proteins showing the conservation of the catalytic histidine. B. anthracis, Bacillus anthracis; C. difficile, Clostridioides difficile; G. kaustophilus, Geobacillus kaustophilus; S. aureus, Staphylococcus aureus. g, CHRM4 activation by products from WT and mutant BbChAT and PpChAT assayed under the same conditions as in a and c. h, Phylogenetic tree of the prokaryotic hexapeptide repeat protein superfamily. i, CHRM4 activation by assay products from selected hexapeptide repeat proteins (highlighted in h). j, CHRM4 activation by WT and Δchat B. breve cultured with and without 1 mM choline. k, Mass spectrometry quantification of ACh in cultures of WT and Δchat B. breve grown with 1 mM choline. CHRM4 activity (a, c, g, i and j) and ACh quantification (b, d and k) are representative of two independent experiments (n = 3 biologically independent samples per group in each experiment). Data are mean ± s.e.m. Statistical significance was assessed using Welch’s ANOVA with Games–Howell post hoc test (a, c, d and i), Kruskal–Wallis tests followed by Dunn’s multiple-comparison test (b), two-sided Welch’s t-tests (g) and one-way ANOVA with Tukey’s post hoc test (j). Statistical comparisons were performed between each candidate enzyme and the no-protein control (a–d and i), and between each bacteria culture and the medium control (j). Exact P values are shown in the figure when P < 0.05; all other P values are provided as Source data. Source data We similarly analysed the P. pentosaceus NP38 genome and selected five candidate acetyltransferases for characterization (Supplementary Table 5). We cloned and purified all selected enzymes and evaluated their ability to produce ACh and activate CHRM4 in vitro (Extended Data Fig. 6d). Among these enzymes, putative acetyltransferase 4 (hereafter PpChAT) demonstrated the highest activity, while putative acetyltransferase 5 showed only very mild activity (Fig. 3c,d). To better characterize these two ACh-producing enzymes, we modelled their structures using AlphaFold2 and searched for homologues within the Protein Data Bank (PDB; Fig. 3e). While BbChAT and PpChAT share only approximately 30% amino acid identity, structural predictions revealed that they both belong to the hexapeptide repeat acetyltransferase family, showing high structural homology to Escherichia coli maltose O-acetyltransferase (EcMoAT; DALI Z score: PpChAT, 29.6; BbChAT, 26.4; Fig. 3e), another member of the hexapeptide repeat acetyltransferase family that has been structurally resolved20,21. Alignment of BbChAT and PpChAT revealed conservation at the putative active site histidine (Fig. 3f), which was required for ACh synthesis by these enzymes (Fig. 3g and Extended Data Fig. 6b,e). The hexapeptide repeat protein superfamily contains hundreds of thousands of sequences in InterPro. To contextualize the commensal-encoded ChAT enzymes identified here, we generated a phylogenetic tree of all prokaryotic hexapeptide repeat proteins in the PDB with experimentally determined structures (Fig. 3h). These proteins were primarily annotated as acetyltransferases but varied widely in their predicted substrate specificity21. Proteins clustered in the same clade generally had related annotated functions, for example, cell wall biosynthesis, antibiotic resistance, L-lysine biosynthesis, L-cysteine biosynthesis and lipid-A biosynthesis (Fig. 3h). PpChAT and BbChAT localized to a clade of maltose/galactoside acetyltransferases that also included EcMoAT and E. coli LacA. To examine the potential ChAT activities of proteins in this superfamily, we selected three additional proteins from the maltose/galactoside acetyltransferase clade and one protein each from the neighbouring antibiotic resistance and L-lysine biosynthesis clades (Supplementary Table 5) and assessed their ability to convert choline to ACh as measured on the basis of CHRM4 activation. The newly described ChATs from Bifidobacterium or Pediococcus exhibited the strongest activity, while the LacA (E. coli) and CatB8 (Acinetobacter baumannii) samples exhibited very weak CHRM4 activation, and MoAT (E. coli), DAPD (Enterococcus faecalis) and HrcAT (Vibrio cholerae) exhibited minimal activity (Fig. 3i), showing that ChAT activity is unique and not a universal feature of these related maltose/galactoside acetyltransferases or hexapeptide repeat proteins more broadly. Given that BbChAT was the most active ChAT from B. breve NWP289, we next examined whether this gene was required for B. breve ACh biosynthesis. We constructed a BbChAT mutant (B. breve::pFREM28-Bb_chat, hereafter Δchat B. breve), in which the coding sequence of BbChAT was disrupted (Extended Data Fig. 6f–h). We then cultured wild-type (WT) B. breve NWP289 (hereafter WT B. breve) and Δchat B. breve in the presence or absence of 1 mM choline. Δchat B. breve did not produce ACh or activate CHRM4 (Fig. 3j,k), demonstrating that BbChAT is essential for ACh synthesis in B. breve NWP289. Commensal ACh enhances intestinal IgA To test the potential impacts of commensal ACh on gut physiology, we monocolonized GF mice with WT or Δchat B. breve and then treated these animals with or without the FDA-approved acetylcholinesterase inhibitor (AChEi) rivastigmine to enhance the potential effects of commensal ACh. WT and Δchat B. breve strains colonized the gut at similar levels both in the presence and absence of AChEi under monocolonization conditions (Extended Data Fig. 7a,b); furthermore, consistent with our previous bioactivity-based assessments, luminal contents from WT B. breve-colonized mice contained approximately 6 µg per g ACh, whereas ACh was undetectable in Δchat B. breve-colonized mice (Extended Data Fig. 7c,d). As the major parasympathetic neurotransmitter, ACh regulates gut motility and epithelial fluid secretion22. Accordingly, we observed that mice colonized with WT B. breve and treated with AChEi exhibited significantly increased gut motility compared with those colonized with the Δchat strain (Fig. 4a), whereas this difference was blunted without AChEi supplementation (Extended Data Fig. 7e). B. breve-induced increases in gut motility were abolished by treatment with the CHRM3 antagonist 4-DAMP (Fig. 4a), indicating that bacterial-derived ACh can increase gut motility through a CHRM3-dependent mechanism. By contrast, we did not observe any measurable differences in faecal water content between WT and Δchat B. breve-colonized mice (Extended Data Fig. 7f,g).Fig. 4ACh-producing B. breve increases gut motility and enhances intestinal IgA responses.a, Whole-gut transit time of WT or Δchat B. breve-monocolonized mice treated with AChEi, with or without 4-DAMP (1 mg per kg, intraperitoneal (i.p.)). n = 5, 4, 5 and 4 mice per group. b, Schematic of bulk RNA-seq of LP immune cells from WT or Δchat B. breve-monocolonized mice treated with AChEi. c, Differentially expressed genes in LP immune cells (DESeq2, two-sided Wald tests with Benjamini–Hochberg correction). d, GSEA of differentially expressed genes. e–h, Representative flow cytometry plots (e) and the frequency (f) and number (g) of colonic LP IgA plasma cells, and faecal IgA concentrations (h) in monocolonized mice. n = 8 and 7 (f and g) and n = 8 (h) mice per group. i, Schematic of colonization of antibiotic-treated SPF mice with WT or Δchat B. breve. j–m, Representative flow cytometry plots (j) and the frequency (k) and number (l) of colonic LP IgA plasma cells, and faecal IgA concentrations (m) in colonized SPF mice. n = 5 mice per group. n–q, Representative flow cytometry plots (n and p) and the frequency of IgA-coated bacteria (o and q) in monocolonized mice (n and o; n = 8 mice per group) and colonized SPF mice (p and q; n = 5 mice per group). r–u, Representative flow cytometry plots (r) and the frequency (s) and number (t) of colonic LP IgA plasma cells, and faecal IgA concentrations (u) in Δchat and WT B. breve-colonized mice treated with atropine or mecamylamine (25 mg l−1 each in drinking water). n = 5, 4, 4 and 4 mice per group. Data are mean ± s.e.m. and representative of two (a, j–m and p–u) or three (e–h, n and o) independent experiments. Statistical significance was assessed using unpaired two-sided t-tests (a, f–h, k–m, o and q) and one-way ANOVA followed by Tukey’s post hoc test (s–u). The diagrams in b and i were created using BioRender; Song. D https://BioRender.com/8t518es (b) and https://BioRender.com/vhqblos (i) (2026).Source data ACh-producing B. breve increases gut motility and enhances intestinal IgA responses. a, Whole-gut transit time of WT or Δchat B. breve-monocolonized mice treated with AChEi, with or without 4-DAMP (1 mg per kg, intraperitoneal (i.p.)). n = 5, 4, 5 and 4 mice per group. b, Schematic of bulk RNA-seq of LP immune cells from WT or Δchat B. breve-monocolonized mice treated with AChEi. c, Differentially expressed genes in LP immune cells (DESeq2, two-sided Wald tests with Benjamini–Hochberg correction). d, GSEA of differentially expressed genes. e–h, Representative flow cytometry plots (e) and the frequency (f) and number (g) of colonic LP IgA plasma cells, and faecal IgA concentrations (h) in monocolonized mice. n = 8 and 7 (f and g) and n = 8 (h) mice per group. i, Schematic of colonization of antibiotic-treated SPF mice with WT or Δchat B. breve. j–m, Representative flow cytometry plots (j) and the frequency (k) and number (l) of colonic LP IgA plasma cells, and faecal IgA concentrations (m) in colonized SPF mice. n = 5 mice per group. n–q, Representative flow cytometry plots (n and p) and the frequency of IgA-coated bacteria (o and q) in monocolonized mice (n and o; n = 8 mice per group) and colonized SPF mice (p and q; n = 5 mice per group). r–u, Representative flow cytometry plots (r) and the frequency (s) and number (t) of colonic LP IgA plasma cells, and faecal IgA concentrations (u) in Δchat and WT B. breve-colonized mice treated with atropine or mecamylamine (25 mg l−1 each in drinking water). n = 5, 4, 4 and 4 mice per group. Data are mean ± s.e.m. and representative of two (a, j–m and p–u) or three (e–h, n and o) independent experiments. Statistical significance was assessed using unpaired two-sided t-tests (a, f–h, k–m, o and q) and one-way ANOVA followed by Tukey’s post hoc test (s–u). The diagrams in b and i were created using BioRender; Song. D https://BioRender.com/8t518es (b) and https://BioRender.com/vhqblos (i) (2026). Source data ACh has also been shown to modulate innate and adaptive immunity23–28. To examine the impacts of commensal ACh on the mucosal immune system, we performed single-cell RNA sequencing (scRNA-seq) analysis of gut lamina propria (LP) immune cells from mice monocolonized with WT and Δchat B. breve. We did not observe any marked differences in cellular frequencies; however, ileal plasma cell and naive/memory B cell populations exhibited transcriptional differences in mice colonized with WT versus Δchat B. breve (Extended Data Fig. 8a). Pseudobulk DESeq2 analysis revealed increased Igha gene expression in plasma cells and naive/memory B cells from mice colonized with WT versus Δchat B. breve (Extended Data Fig. 8b–g); increased Igha expression was also apparent in a reclustered UMAP of plasma cell, naive/memory B cell and germinal centre B cell subsets (Extended Data Fig. 8h). Next, we treated monocolonized mice with AChEi to enhance the potential effects of commensal ACh and performed bulk RNA-seq analysis of colonic and ileal LP immune cells (Fig. 4b). Mice colonized with WT B. breve showed significantly higher colonic expression of Aicda, Igha, Jchain, Mzb1, Bhlha15, Tnfrsf17, Mef2b and Ccl21d, compared with those colonized with Δchat B. breve (Fig. 4c). Ileal transcripts exhibited similar trends (Fig. 4c and Supplementary Fig. 4). Gene set enrichment analysis (GSEA) revealed that colonic LP immune cells from mice colonized with WT B. breve were enriched for pathways related to immunoglobulin production, immunoglobulin-mediated immune response, B-cell-mediated immunity, production of molecular mediator of immune response and adaptive immune response (Fig. 4d). We next analysed colonic LP immune cells using flow cytometry (Supplementary Fig. 5a). Consistent with the transcriptomic data, WT B. breve colonization led to a higher frequency and absolute number of plasma cells in the colonic LP (Fig. 4e–g). Similarly, we observed a significant increase in faecal IgA levels in WT versus Δchat B. breve-colonized mice (Fig. 4h). To assess whether this effect was evident in a complex microbial context, we colonized specific-pathogen free (SPF) mice with WT or Δchat B. breve by pretreating SPF mice with antibiotics and then gavaging with B. breve for 5 weeks to establish stable colonization (Fig. 4i). Consistent with our monocolonization results, colonization with WT B. breve induced increased colonic plasma cell frequency and faecal IgA levels as compared to colonization with Δchat B. breve (Fig. 4j–m). However, in this context, the Δchat strain exhibited significant colonization defects compared with the WT control (Δchat, around 1% relative abundance; WT, around 11% relative abundance) (Extended Data Fig. 7h). These results suggest that ACh-producing B. breve can promote enhanced IgA production and that B. breve ChAT increases bacterial fitness in the presence of a complex SPF microbial community. IgA is secreted into the gut lumen, where it binds to commensal bacteria at steady state29–31. WT mice monocolonized with B. breve and treated with or without AChEi exhibited a higher proportion of IgA-coated faecal bacteria than Δchat-colonized mice (Fig. 4n,o, Extended Data Fig. 9a,b and Supplementary Fig. 5b). We observed similar effects in SPF mice colonized with WT or Δchat B. breve (Fig. 4p,q). To assess the specificity of these IgA responses, we stained in vitro-grown B. breve NWP289 and ten unrelated bacterial species with faecal IgA from mice monoassociated with WT or Δchat B. breve (Extended Data Fig. 9c). We observed limited faecal IgA binding to in vitro-grown non-B. breve strains and, notably, also to in vitro-grown B. breve across all groups (Extended Data Fig. 9d–f). On the basis of these data, we hypothesized that in vitro- and in vivo-grown B. breve may differ in the cell-surface structures targeted by IgA. To test this possibility, we monocolonized Rag1-deficient (Rag1−/−) mice with B. breve and used faecal bacteria from these mice as a source of non-antibody-bound in vivo-grown B. breve. We found that concentration-normalized faecal IgA from WT B. breve-colonized mice bound to in vivo-cultivated B. breve at a higher level than IgA from Δchat B. breve-colonized mice, while binding to a defined community of non-B. breve strains grown in Rag1−/− mice was similar between groups (Extended Data Fig. 9g–l). Together, these data suggest that ACh-producing B. breve preferentially enhances the production of B. breve-reactive IgA. Finally, to discern whether muscarinic versus nicotinic ACh receptors mediated B. breve-induced IgA enhancement, we supplemented atropine (a pan-muscarinic ACh receptor antagonist) or mecamylamine (a pan-nicotinic ACh receptor antagonist) in the drinking water of mice monoassociated with B. breve. Mecamylamine treatment abolished WT B. breve-induced increases in colonic plasma cells and faecal IgA levels, whereas atropine had no significant effect (Fig. 4r–u). These findings suggest that ACh-producing B. breve enhances intestinal IgA responses primarily through nicotinic ACh receptors. ACh+ B. breve shapes gut ecology As ACh-producing B. breve induced enhanced mucosal immune responses, we next tested the impacts of B. breve ACh production on microbiota composition and resistance to enteric infection. To assess the potential effects of commensal ACh on microbial community assembly, we monocolonized mice with WT or Δchat B. breve for 5 weeks and then exposed these groups of mice to a consortium of culturable human gut commensals (Fig. 5a and Extended Data Fig. 10a). While both groups displayed similar microbiota compositions 2 days after colonization, and WT and Δchat B. breve strains colonized at similar levels in this context (Extended Data Fig. 10b), the surrounding microbial communities in WT versus Δchat B. breve-colonized mice diverged over time and differed significantly by 28 days after community colonization (Fig. 5b). For example, Staphylococcus sciuri, unclassified Bacillaceae and Enterococcus sp. were more abundant in mice colonized with Δchat B. breve than in WT B. breve-colonized controls, while Clostridium aldenense, Eubacterium dolichum and an unclassified Ruminococcaceae were more abundant in WT B. breve-colonized mice (Fig. 5c–h and Extended Data Fig. 10c).Fig. 5ACh-producing B. breve shape gut microbiota community composition and enhance resistance to S. Tm challenge.a, Timeline of precolonization with WT or Δchat B. breve (with AChEi supplementation) followed by gavage with a simplified human gut bacterial community. Faecal samples were collected weekly for 4 weeks after community colonization. b, Principal coordinates analysis (PCoA) of faecal microbiome beta diversity based on Canberra distance. Each point represents one sample; the colours indicate the group and the shapes indicate the timepoint. c–h, The relative abundance of the indicated taxa in WT and Δchat B. breve-colonized mice 28 days after community colonization. n = 5 mice per group. i, Schematic of S. Tm ΔssaV challenge in GF mice colonized with WT or Δchat B. breve (with AChEi supplementation). j, Faecal S. Tm ΔssaV burdens 20 h after infection. n = 4 and 5 mice per group. CFU, colony-forming units. k, Schematic of S. Tm ΔssaV challenge in WT or Δchat B. breve-colonized SPF mice. SPF mice were pretreated with an antibiotic cocktail on day −1 to deplete native bacteria, then gavaged daily with either WT or Δchat B. breve from day 0 to day 35 with AChEi supplementation, followed by S. Tm ΔssaV challenge. l, Faecal S. Tm ΔssaV burdens were measured from day 1 to day 6 after challenge. n = 5 mice per group. Data are mean ± s.e.m. and are representative of two independent experiments. Statistical significance was assessed using two-sided Wilcoxon rank-sum tests (c–h and l (day 5)) and an unpaired two-sided t-test (j). The diagrams in a, i and k were created using BioRender; Song. D https://BioRender.com/ybcn422 (a), https://BioRender.com/38cj9fj (i) and https://BioRender.com/bysgxos (k) (2026).Source data ACh-producing B. breve shape gut microbiota community composition and enhance resistance to S. Tm challenge. a, Timeline of precolonization with WT or Δchat B. breve (with AChEi supplementation) followed by gavage with a simplified human gut bacterial community. Faecal samples were collected weekly for 4 weeks after community colonization. b, Principal coordinates analysis (PCoA) of faecal microbiome beta diversity based on Canberra distance. Each point represents one sample; the colours indicate the group and the shapes indicate the timepoint. c–h, The relative abundance of the indicated taxa in WT and Δchat B. breve-colonized mice 28 days after community colonization. n = 5 mice per group. i, Schematic of S. Tm ΔssaV challenge in GF mice colonized with WT or Δchat B. breve (with AChEi supplementation). j, Faecal S. Tm ΔssaV burdens 20 h after infection. n = 4 and 5 mice per group. CFU, colony-forming units. k, Schematic of S. Tm ΔssaV challenge in WT or Δchat B. breve-colonized SPF mice. SPF mice were pretreated with an antibiotic cocktail on day −1 to deplete native bacteria, then gavaged daily with either WT or Δchat B. breve from day 0 to day 35 with AChEi supplementation, followed by S. Tm ΔssaV challenge. l, Faecal S. Tm ΔssaV burdens were measured from day 1 to day 6 after challenge. n = 5 mice per group. Data are mean ± s.e.m. and are representative of two independent experiments. Statistical significance was assessed using two-sided Wilcoxon rank-sum tests (c–h and l (day 5)) and an unpaired two-sided t-test (j). The diagrams in a, i and k were created using BioRender; Song. D https://BioRender.com/ybcn422 (a), https://BioRender.com/38cj9fj (i) and https://BioRender.com/bysgxos (k) (2026). Source data Next, we examined the impacts of WT and Δchat B. breve on enteric infection. We infected mice that were previously monocolonized with WT and Δchat B. breve for 35 days (Extended Data Fig. 10d) with Salmonella enterica subsp. enterica serovar Typhimurium (S. Tm) ΔssaV32, an attenuated S. enterica strain lacking the SPI-2 type III secretion system (Fig. 5i). After 20 h, Δchat B. breve-colonized mice showed markedly higher S. Tm ΔssaV burdens compared with WT B. breve-colonized mice, while both groups showed similar levels of the inflammatory marker LCN2 (Fig. 5j and Extended Data Fig. 10e). To test whether this finding would extend to a complex community setting, we colonized antibiotic-treated SPF mice with WT or Δchat B. breve for 5 weeks before S. Tm ΔssaV challenge (Fig. 5k). Mice colonized with WT B. breve were protected from S. Tm ΔssaV infection, whereas Δchat B. breve-colonized mice showed robust infection on all days (Fig. 5l and Extended Data Fig. 10f,g). Accordingly, faecal LCN2 was elevated in Δchat B. breve-colonized mice with high S. Tm burdens but remained undetectable in WT B. breve-colonized mice (Extended Data Fig. 10h). Together, these results demonstrate that ACh-producing B. breve can shape gut microbiota composition and enhance resistance to enteric infection. Discussion Microbial metabolites have crucial roles in shaping host physiology, including immune responses that reinforce host–microbiota mutualism. Here we used a multiplexed GPCR screening technology to compare receptorome-wide bioactivities of in vitro and in vivo metabolomes from diverse human commensals and found that many in vivo metabolome bioactivities differed substantially from their in vitro counterparts. Notably, multiple in vivo commensal metabolomes, but not matched in vitro metabolomes, activated ACh receptors. Select commensals, including Bifidobacterium strains enriched in early life and a probiotic Pediococcus strain, could convert dietary choline to ACh. We identified and characterized the bacterial enzymes responsible for ACh production and found that ACh-producing B. breve promoted gut motility, enhanced IgA responses, altered gut microbiota composition and increased resistance to Salmonella infection. Previous GPCR-based studies of commensal metabolites were restricted to in vitro metabolomes4–6. By combining extraction-based metabolite enrichment with PRESTO-Salsa5, we compared in vitro and in vivo commensal metabolomes at scale. These analyses revealed broad differences in receptor engagement, including for ARs, DRDs, HRHs, HTRs and SUCNR1. These data highlight the considerable impact of the in vivo environment on microbial metabolomes and provide a valuable resource for examining how human gut commensals influence in vivo biology. Human commensal microorganisms are known to produce diverse neurotransmitters33, including gamma-aminobutyric acid, serotonin, histamine, PEA and tyramine. Although certain fungi and fermented food-associated bacteria (such as Lactobacillus plantarum) have been reported to produce ACh34–37, we and others had not previously observed strong cholinergic receptor activation by in vitro human commensal metabolomes4–6. Here we found that multiple phylogenetically diverse commensals produced ACh in vivo through the conversion of dietary choline. Although ACh production by L. plantarum was first described nearly eight decades ago34, and previous studies showed that bacterial ChAT is resistant to human ChAT and carnitine acetyltransferase (CarAT) inhibitors36, the metabolic pathway that catalyses ACh synthesis in bacteria remained undetermined. We identified two previously uncharacterized hexapeptide repeat family acetyltransferases from B. breve and P. pentosaceus that catalysed this reaction. Although these enzymes shared only modest amino acid identity, they exhibited structural similarity and fell within a broader clade of O-acetyltransferases, in which robust ACh-synthesizing activity appeared to be relatively rare. These findings therefore highlight the importance of functional approaches in defining microbiota metabolite production. Many of the human commensals that produced ACh were Bifidobacterium species, which are enriched in infant microbiomes, and both Bifidobacterium strains and other potent ACh-producing taxa (such as P. pentosaceus) have reported probiotic functions. Using an isogenic Δchat B. breve mutant strain, we found that ACh-producing B. breve induced enhanced intestinal IgA responses in both monocolonization and complex community settings, including increased IgA coating of gut microorganisms in vivo in both AChEi-treated and non-treated animals. While WT and Δchat B. breve colonized the gut at similar levels in monocolonization and simplified community settings, WT B. breve exhibited a significant colonization advantage in the setting of an SPF gut microbial community, suggesting that ChAT expression enhances B. breve competitive fitness. Although the physiological effects of microbiota-derived ACh were previously unexplored, previous studies have revealed diverse impacts of neuronal and non-neuronal ACh production on the immune response. For example, ACh derived from vagal neurons or ChAT+ T cells suppresses macrophage inflammatory responses through the α7 nicotinic ACh receptor (CHRNA7)23,25. Recent studies have also revealed impacts of T-cell-derived ACh on B cell responses, for example, by promoting splenic plasma cell differentiation through CHRNA9 during stress26,38. Our data suggest that commensal ACh can induce enhanced intestinal IgA responses through nicotinic receptors. However, further research will be necessary to determine the individual receptor(s) responsible for these effects and whether ACh enhances intestinal IgA production by acting directly on B cells or indirectly through neuronal, immunological or epithelial intermediaries. Mice lacking IgA or with defective mucosal antibody responses often display features of dysbiosis and increased susceptibility to enteric infections39,40. We observed that mice monocolonized with WT or Δchat B. breve, then exposed to a simplified human microbial community, developed divergent microbiota compositions. In particular, WT B. breve colonization restricted colonization by or expansion of potentially harmful taxa, such as S. sciuri, and unclassified Enterococcus, while supporting the persistence of E. dolichum, unclassified Ruminococcaceae and C. aldenense. These observations are consistent with the dual roles of IgA: restricting the overgrowth of potentially harmful bacteria through immune exclusion, while fostering stable colonization by mutualistic microorganisms through immune inclusion41,42. We also observed that both monocolonized mice and SPF mice colonized with WT B. breve exhibited enhanced resistance to infection with S. Tm ΔssaV41. However, disentangling the relative importance of IgA versus microbial ecology on these phenotypes will require further research. Overall, these data suggest that commensal ACh production enhances intestinal immune education and reinforces host–microbiota mutualism. We speculate that coupling IgA production to dietary choline availability may allow the host to tailor engagement of this energetically costly defence pathway to nutrient status. Our in vivo metabolome screening pipeline enabled extensive exploration of bioactive metabolites produced by the microbiota; however, these studies have several limitations. First, to enable direct comparisons between in vitro and in vivo metabolomes, we only interrogated metabolomes from monoassociated mice. Thus, these studies do not capture metabolites that arise only in complex community settings, such as metabolites produced through interspecies chemical exchange or synthesized by individual microorganisms in response to chemical cues from other microorganisms. Furthermore, due to the limited volume of in vivo material, we could not perform deep chemical fractionations of gut contents to identify rare metabolites produced in low abundance. Finally, because most of our studies used AChEi treatment to enhance the potential impacts of commensal ACh, additional work will be required to determine the precise immunological effects and biodistribution of ACh in intestinal tissues, both in the presence and absence of AChE inhibition. The vast metabolic arsenal encoded by our indigenous bacteria has a pivotal role in shaping host physiology, yet in vitro cultures only scratch the surface of this universe, often failing to capture the full spectrum of metabolic activities that occur within their native niches. These studies underscore the importance of developing approaches to mine microbial metabolomes under physiologically relevant settings to reveal how microbial chemicals influence human health and disease. Methods Bacterial growth and generation of in vitro bacterial metabolite libraries Based on phylogenetic diversity, 100 bacterial strains were picked from the HMP bacterial culture collection and our laboratory’s internal human gut bacterial culture collections (Supplementary Table 1). Anaerobic bacteria were cultured in a vinyl anaerobic chamber (Coy Laboratory Products), containing 4% H2, 10% CO2 and 86% N2. Aerobic bacteria were cultured in an aerobic microbiological incubator. All bacteria were cultured in triplicate at 37 °C for 48 h. To extract bacterial metabolites with an acetonitrile-methanol-water (2:2:1) mixture (ACN-MeOH-water), the bacterial cultures were dried overnight at room temperature under vacuum (GeneVac). Then, the dried matter was resuspended in ACN-MeOH-water with intermittent sonication. The suspensions were centrifuged at 4,000g for 10 min, and the supernatant was dried overnight at room temperature under a vacuum (GeneVac). The dried extracts were dissolved in sample buffer (PBS containing 10% DMSO, v/v). To extract bacterial metabolites with butanol or ethyl acetate, the bacterial cultures were mixed with two volumes of butanol or ethyl acetate and shaken vigorously for 10 min in a horizontal shaker at room temperature. The organic phase was collected and dried overnight at room temperature under a vacuum (GeneVac), and the dried extracts were dissolved in sample buffer. The three types of extracts comprise the bacterial in vitro metabolome libraries. Mouse experiments All animal protocols were approved by the Yale University Institutional Animal Care and Use Committee (IACUC Protocol 11513). Mice were on a dark–light cycle of 12 h–12 h with temperature (22 °C) and humidity (50%) controlled. Gnotobiotic mice and generation of in vivo bacterial metabolite libraries GF C57BL/6 mice were maintained in flexible film isolators (Class Biologically Clean) with all bedding, chow (Teklad, 2018S) and water autoclaved before import. All GF breeding isolators were regularly monitored for bacteria (culture-dependent and -independent techniques). For the monoassociation gnotobiotic experiment, male GF mice (aged 6 to 8 weeks) were transferred from the isolators to positive pressure ventilated microisolator cages (Tecniplast, ISO72P) and gavaged with 0.2 ml of bacterial culture. Bacterial inocula from all 100 strains were freshly prepared and aliquoted in 2 ml screw cap tubes (Sarstedt, 72.693.005). After 2 weeks of colonization, the caecal contents were collected after euthanization. To prepare the bacteria in vivo metabolome libraries, the caecal contents were resuspended using ACN-MeOH-water solvent. After vigorously shaking in the horizontal shaker at room temperature, the supernatant was collected after centrifuging at 4,000g for 10 min. The supernatant was divided equally into three parts. They were directly dried overnight at room temperature under a vacuum, and the dried extracts were resuspended in sample buffer. One of them was saved as ACN-MeOH-water extracted metabolomes. The other two parts were further extracted using either butanol or ethyl acetate, respectively, followed by drying and dissolving using the same method described for preparing the in vitro metabolome libraries. These three types of extracts comprise the bacteria in vivo metabolome libraries. Chemical supplementation in mouse drinking water To support B. breve WT (NWP289) and Δchat colonization and ACh production, mice in these experiments received drinking water supplemented with 1.5% (w/v) HMOs (LayerOrigin SuperHMO Prebiotic Mix) and 0.25% choline (Sigma-Aldrich, C1879). To enhance the potential impacts of commensal ACh, the AChEi rivastigmine (TCI America, R0093) was supplemented in the mouse drinking water (40 mg l−1). Atropine (Cayman Chemical, 12008) and mecamylamine (Cayman Chemical, 14602) were provided in the drinking water at 25 mg l−1. Water was replenished weekly. Serum sample preparation To prepare serum samples, blood was collected in serum tubes (SAI Infusion Technology, PMTS-SG-1.1). After centrifugation at 4,000g for 5 min, serum samples were collected and stored at −80 °C until use. To collect brain and liver samples, mice were perfused with 10 ml of ice-cold PBS after euthanization. The liver and brain samples were then collected, flash–frozen and stored at −80 °C until use. SPF mouse experiments Wild-type C57BL/6J male mice (aged 6 weeks) were ordered from Jackson Laboratory (000664) and housed in the mouse facility at Yale School of Medicine. To deplete the endogenous microbiota, mice were orally gavaged with an antibiotic cocktail twice on the same day with a 12 h interval. The antibiotic cocktail43 comprised 10 mg ml−1 ampicillin (Sigma-Aldrich, A6140), 10 mg ml−1 neomycin (Sigma-Aldrich, N1876), 10 mg ml−1 metronidazole (Sigma-Aldrich, M1547) and 5 mg ml−1 vancomycin (Gold Biotechnology, V-200-25). Rag1−/− mouse experiment GF Rag1−/− male mice (aged 6–8 weeks old) were maintained under the same conditions as GF C57BL/6 mice described above and used for the indicated in vivo bacterial cultivation and IgA-binding experiments. Sample size, randomization and blinding No statistical methods were used to predetermine sample size. Sample sizes were estimated based on preliminary experiments. Sex- and age-matched mice were randomly assigned to experimental groups for all mouse experiments. Blinding was not used because the primary readouts in this study were based on objective measurements, including sequencing-based datasets, GPCR activity assays, metabolite quantification and flow cytometry, for which investigator awareness of group assignment was unlikely to influence data collection or analysis. Multiplexed GPCRome bioactivity screening The multiplexed GPCRome bioactivity screening experiment was performed using our recently developed GPCR screening technology, PRESTO-Salsa5. The PRESTO-Salsa reporter cell line was maintained in DMEM supplemented with 10% FBS (GeminiBio, 100-106), 1% penicillin–streptomycin (Gibco, 15140122) and 2 µg ml−1 puromycin (Sigma-Aldrich, P8833). Cells were tested and confirmed free of mycoplasma. When the cell density reached approximately 90% confluency, the reporter cells were seeded into 96-well flat-bottom plates and grown to around 90% confluence. For each plasmid pair, 100 ng of GPCR plasmid and 100 ng of PRESTO-Salsa reporter plasmid were mixed with 600 ng PEI Max (Polysciences, 24765) in DMEM. After 20 min of incubation at room temperature, the transfection mixtures were added to the cell culture. Then, 6 h after transfection, the cell medium was discarded. The wells containing unique pairs of GPCR/PRESTO-Salsa reporter plasmids were trypsinized, resuspended (using DMEM with 10% FBS and 1% penicillin–streptomycin), and pooled to form the PRESTO-Salsa reporter cell library. The cell library was then seeded back to new 96-well flat bottom poly-d-lysine-coated (Thermo Fisher Scientific, A3890401) plates and allowed to grow for 12 h. The medium was replaced with 180 μl of DMEM with 1% penicillin–streptomycin and 20 mM HEPES (AmericanBio, AB06021). Then, 2 h later, 20 μl of each in vitro and in vivo bacterial metabolome samples were added to the wells in triplicate, followed by 9 h of stimulation. The total RNA was extracted using the Qiagen RNeasy 96 kit (Qiagen, 74181) according to the manufacturer’s instructions, and RNA concentration was quantified using a NanoDrop 8000 spectrophotometer. 1 µg of RNA from each well was used for cDNA synthesis with the iScript cDNA Synthesis Kit (Bio-Rad, 1708891) according to the manufacturer’s instructions. Next, 1 µl of the cDNA product from each well was used as a template for generating the amplicons by a standard 2-step PCR protocol. The second-round PCR products were pooled and purified by cutting the 264 bp band on an agarose gel, followed by purification using the QIAquick Gel Extraction Kit (Qiagen, 28704) to form the PRESTO-Salsa screening amplicon library. The library was then sequenced on the Illumina NovaSeq 6000 system. Multiplexed GPCR screening data analysis The sequencing data were preprocessed to generate the GPCR read counts for each sample based on the method described previously5. A metadata file associating plate-well identifiers previously designated in the sequencing sample sheet with sample names and assigned plate-specific controls was also generated. GPCR read count table and metadata file were input into DESeq2 (v.1.42.1)44. GPCRs with less than 100 reads were filtered out. log2[FC] values adjusted for effect-size shrinkage values for each GPCR-ligand combination were calculated using apeglm (v.1.26.0)45; adjusted P values were calculated using the Benjamini–Hochberg method, with a significance cut-off of 0.05. Hit calling was based on the threshold of adjusted log2[FC] of more than 0.8 and an adjusted P value of less than 0.05. If a given receptor–sample interaction had a hit in at least one of the three types of extracted metabolomes (ACN-MeOH-water, butanol, ethyl acetate), the interaction was considered to be positive. The screening quality was monitored using the positive-control wells on each plate. To facilitate exploration of the screening dataset, the GPCR screening results have been made publicly available through an interactive Shiny app. PRESTO-Tango assay PRESTO-Tango assay was performed as previously described4. In brief, the HTLA cell line, the PRESTO-Tango reporter cell line that stably expresses β-arrestin/TEV and tTA-luciferase (a gift from G. Barnea), was maintained in DMEM supplemented with 10% FBS, 1% penicillin–streptomycin, 100 µg ml−1 hygromycin B (Invitrogen, 10687010) and 2 µg ml−1 puromycin. Cells were tested and confirmed free of mycoplasma. When cell density reached approximately 90% confluency, 200 ng of GPCR plasmid was transfected into HTLA cells with PEI Max. Then, 20 h after transfection, the medium was replaced with DMEM supplemented with 1% penicillin–streptomycin and 20 mM HEPES, and the cells were stimulated with 20 µl of each sample. Next, 20 h after stimulation, the cell supernatant was discarded, and 50 µl of Bright-Glo luciferase substrate solution (Promega, E2620) was added to each well. After 15 min of incubation at room temperature, luminescence was measured using a SpectraMax i3x with SoftMax Pro 6.5 and SoftMax Pro 7 (Molecular Devices). Bacterial metabolite quantification Chemical quantification of neurotransmitters Serum samples were thawed and mixed by pipetting. Then, 20 µl of serum was mixed with 20 µl of methanol and centrifuged at 18,000g for 10 min. Next, 10 µl of the resulting supernatant was subjected to analysis using high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (MS) (HPLC–QTOF). Caecal content and brain samples were resuspended in ACN-MeOH-water solvent, with occasional sonication and heating (65 °C). The samples were centrifuged at 18,000g for 10 min, and 10 μl of supernatant was subjected to HPLC–QTOF analysis. In vitro samples were dried overnight under vacuum (GeneVac) and resuspended in ACN-MeOH-water with sonication. The samples were centrifuged at 18,000g for 10 min and 10 μl of supernatant was subjected to HPLC–QTOF analysis. For all samples, standard curves for each neurotransmitter (tyramine, PEA, histamine and ACh) were generated in tandem to quantify their concentrations in biological samples. Chemical quantification of organic acids Liver and caecal samples were resuspended in ACN-MeOH-water with occasional sonication and heating (65 °C). The samples were centrifuged at 18,000g for 10 min, 30 µl of the resulting supernatant was mixed with 15 µl of 3-nitrophenylhydrazine (dissolved in 50% aqueous acetonitrile) and 15 µl of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (dissolved in 50% aqueous acetonitrile + 6% pyridine). The solutions were incubated at 42 °C for 30 min and subsequently diluted with 150 µl of 10% aqueous acetonitrile. The samples were centrifuged again at 18,000g for 10 min and 10 µl of the supernatant was subjected to HPLC–QTOF analysis. Serum samples were thawed and mixed by pipetting, and 30 µl of the sample was functionalized with 3-nitrophenylhydrazine and subjected to HPLC–QTOF analysis as described above. For all samples, standard curves for each short-chain fatty acid (propionate and succinate) were generated in tandem to quantify their concentrations in biological samples. These standards were subjected to the same 3-nitrophenylhydrazine functionalization protocol as the samples. HPLC–QTOF analysis High-resolution electrospray ionization MS (HRMS) was performed on an Agilent iFunnel 6550 quadrupole time-of-flight (QTOF) MS instrument coupled to an Agilent 1290 Infinity HPLC system equipped with a Phenomenex Kinetex C18 (100 Å) 5 µm (250 × 4.6 mm) column. Data were acquired using MassHunter Workstation Data Acquisition v.B.05.01 (Agilent). Chromatography was conducted using a linear water–ACN gradient containing 0.1% formic acid at a flow rate of 0.7 ml min−1 over 25 min for the aromatic neurotransmitters, or 30 min for the derivatized short-chain fatty acids. For the aromatic neurotransmitters, the gradient proceeded from 5% ACN to 50% ACN, followed by a 5 min wash with 100% ACN. For the derivatized short-chain fatty acids, the gradient proceeded from 10% ACN to 100% ACN, followed by a 5-min 100% ACN wash. Mass spectra were recorded in positive ionization mode with a mass range of 100 m/z to 1,700 m/z. To measure the concentrations of specific compounds in biological samples, extracted ion chromatograms (EICs) were constructed with a 10 ppm error limit of the exact monoprotonated [M+H]+ masses of compounds to be quantified. The peaks in the EICs were integrated using the MassHunter Qualitative Analysis v.B.06.00. Peak areas were compared to standard curves generated with commercially available standards to determine concentration. Standard curves were run in the same worklist as the corresponding samples. The metabolomics data generated in this study have been deposited in the NIH Common Fund’s National Metabolomics Data Repository, the Metabolomics Workbench46. Bacterial whole-genome sequencing and annotation Whole-genome sequencing Single bacterial colonies were picked from agar plates and grown in broth overnight. The bacterial genomic DNA was isolated using the DNeasy UltraClean Microbial Kit (Qiagen, 12224-250) according to the manufacturer’s instructions. Sequencing libraries were prepared using the Illumina Nextera XT library preparation kit (Illumina, FC-131-1024) and barcoded using the Nextera XT Index Kit V2 (Illumina, FC−131-2001). The libraries were sequenced with NovaSeq 6000 at a coverage of 150×. De novo genome assembly Genome assembly and annotation were performed as previously described47. All Illumina paired-end reads were filtered and trimmed using Trimmomatic v.0.38 with the following parameters: ILLUMINACLIP: NexteraPE-PE.fa:2:30:12:1:true LEADING:3 TRAILING:3 MAXINFO:40:0.994 MINLEN:36 (ref. 48). The output files after trimming included two (forward and reverse) FASTQ files with paired reads and two FASTQ files with unpaired reads. All four files from each strain were assembled into contigs using SPAdes (v.3.13.0)49 with the default parameters for paired-end libraries. Genome coverage was calculated using BBMap (v.39.06; https://sourceforge.net/projects/bbmap/). Genome annotation For each genome assembly, scaffolds longer than 2,000 bp were uploaded to the RAST server (v.2.0)19 for annotation (using the default RASTtk pipeline). Protein expression and purification Recombinant putative acetyltransferases were purified using methods previously described50. In brief, the putative acetyltransferases were cloned into an N-terminal 6×His-SUMO2-tagged pET vector and transformed into BL21-RIL E. coli (Agilent Technologies, 230245). Large-scale cultures were grown for approximately 5 h at 37 °C, then induced with IPTG overnight at 16 °C. Bacterial pellets were resuspended and sonicated in lysis buffer (20 mM HEPES-KOH pH 7.5, 400 mM NaCl, 30 mM imidazole, 10% glycerol and 1 mM dithiothreitol) and purified using Ni-NTA resin (Qiagen, 30230). Ni-NTA resin was washed with lysis buffer supplemented to 1 M NaCl and eluted with lysis buffer supplemented to 300 mM imidazole. The eluted proteins were concentrated using a 3 kDa cut-off concentrator (Millipore, UFC500324). Protein gels were imaged on an Azure c200 imaging system using Azure c200 imaging software (Azure Biosystems). Samples were then flash-frozen in liquid nitrogen and stored at −80 °C. Acetyltransferase in vitro assay The acetyltransferase in vitro assay was performed as previously described51. The reaction mixtures containing 50 mM potassium phosphate (pH 7.4), 1 mM EDTA, 20 mM choline chloride, 250 mM sodium chloride, 0.1 mM acetyl-CoA and 2 μg of recombinant putative acetyltransferase protein were incubated for 15 min at 37 °C. The reaction was terminated by immediate incubation on ice and followed by separation with a 3-kDa cutoff concentrator (Millipore, UFC500324) to remove the proteins at 4 °C. Acetyltransferase activity was evaluated by measuring CHRM4 activation and ACh concentration. Bioinformatic analyses of hexapeptide repeat proteins Structure models were created using AlphaFold2 (v.2.3.0)52, with multimeric states chosen manually based on homology. Models were compared to the PDB and Z scores were calculated using DALI53 (accessed in 2024). The sequences of proteins in the hexapeptide repeat family (G3DSA:2.160.10.10) with experimentally determined structures were accessed from the PDB. Sequence alignments were generated and a phylogenetic tree were constructed using MAFFT (v.7)54 after manually removing eukaryotic sequences, then visualized with iTOL (v.7)55. Construction of B. breve ChAT disruption strain Generation of plasmid pFREM28-Bb_chat as a non-replicative vector for creating B. brevechat gene insertion mutant The pFREM28-Bb_chat non-replicative plasmid was constructed in the pFREM28 vector backbone using standard cloning techniques56. In brief, an internal 400 bp fragment of the Bbchat gene was amplified from B. breve NWP289 genomic DNA using primer pair (Bb_chat+138_EcoRI_fwd: TAGGCCGAATTCTCTCCAGCTTGAACTCAACG; Bb_chat_-536_HindIII_rev: TAGGCCAAGCTTACCAGCGAGTTGGCGCCAAT). The amplified product was cloned into the pFREM28 vector at the EcoRI and HindIII restriction sites using sticky-end cloning. The construct was transformed and propagated in E. coli EC1000 by electroporation and grown on LB plates with kanamycin and erythromycin. The plasmid was isolated from 100 ml culture using ZymoPURE II Plasmid Midiprep Kit (Zymo, D4201) and concentrated using Amicon 30 kDa Ultra Centrifugal Filter (Sigma-Aldrich, UFC503024). Generation of B. breve chat gene insertion mutant The Bbchat gene insertion mutant was created using the non-replicative pFREM28-Bb_chat plasmid using the method described previously57 with minor modifications. B. breve NWP289 cells were made electrocompetent. A single colony of B. breve NWP289 was grown in 10 ml RCM supplemented with 0.05% cystine and 1% lactose, anaerobically at 37 °C for 18–20 h. The overnight culture was diluted 1:50 in fresh 10 ml RCM supplemented with 0.05% cystine and 1% lactose for about 16–18 h. Then the culture was diluted in 50 ml MRS supplemented with 0.05% cystine and 1% lactose to adjust the starting optical density (OD) to around 0.3. The cells were grown to a final OD600 of 0.8–0.9. The cells were chilled on ice for 20 min and washed three times with the chilled electroporation buffer (sucrose-citrate buffer; 0.5 M sucrose and 1 mM ammonium citrate, pH 5.8). Then, 50 μl of the electrocompetent cells was transformed by electroporation with 6–7 µg of pFREM28-Bb_chat plasmid DNA and recovered overnight in RCM broth at 37 °C. The recovered cells were plated on Gifu agar supplemented with 5 µg ml−1 erythromycin. All of the steps were carried out in an anaerobic chamber (except electroporation) with pre-reduced materials. Plasmid insertion was confirmed by PCR amplification and sequencing using primer pairs (Bb_chat_+38_SphI-fwd: TAGGCCGCATGCATGATGCGACGCGCGAGAAA; Bb_chat_-623_KpnI-rev: TAGGCCGGTACCTAGTAGACGTCGTCGCGTTC). Bacterial RNA-seq In vitro culture total RNA isolation C. portucalensis overnight culture was inoculated 1:100 into fresh medium, cells were collected when the density reached OD600 = 0.32. 3 ml of bacteria culture was mixed with 6 ml of RNA protect reagent (Qiagen, 76506) and incubated at room temperature for 5 min. Then the bacterial cells were digested with 120 µl of digestion buffer (40 µl TE buffer, 60 µl of 50 mg ml−1 lysozyme, 20 µl of 20 mg ml−1 proteinase K) for 15 min at room temperature. 700 µl of buffer RLT (containing 1% of β-mercaptoethanol) was then added, and bead-beaten with lysing matrix B beads for 10 min. After centrifugation at 13,000g for 10 s, the supernatant was then transferred into a new tube and an equal volume of 70% ethanol was added. Then total RNA was isolated using RNeasy mini kit (Qiagen, 74106) following the instructions. Monocolonized bacterial total RNA isolation Caecal contents from mice that had been monocolonized with C. portucalensis NWP28 for 2 weeks were collected and resuspended in 5 ml PBS. After centrifuging at 50g for 5 min, 3 ml of the top fraction (containing the bacterial cells) was collected and mixed with 6 ml of RNA protect reagent. The samples were then processed using the same protocol as described for the in vitro cultured bacteria. Library preparation Ribosomal RNA was depleted from total RNA, followed by ethanol precipitation and fragmentation. First-strand cDNA was synthesized using random hexamer primers, and second-strand cDNA synthesis was performed using dUTPs to allow strand specificity. The libraries were then prepared through end repair, A-tailing, adapter ligation, size selection, PCR amplification and purification. Library quality and concentration were assessed using Qubit fluorometry, quantitative PCR and the Agilent Bioanalyzer. Sequencing was performed on the Illumina NovaSeq X Plus platform. Data analysis Trimmed reads were generated using Cutadapt (v.5.0)58 with adapter removal and quality trimming. Trimmed reads were aligned to the C. portucalensis NWP28 reference genome using BWA-MEM (v.0.7.17-r1188)59. The resulting BAM files were sorted and indexed with SAMtools (v.1.21)60. Gene-level read counts were obtained using featureCounts (v.2.0.3)61 with the GFF genome annotation. Differential gene expression analysis was performed in R using DESeq2 (v.1.42.1)44, comparing in vivo and in vitro conditions. Functional enrichment analysis of differentially expressed genes was performed with fgsea (v.1.28.0)62 using KEGG-orthology-based gene sets. Whole-gut transit time The whole-gut transit time experiment was performed as described before with minor modifications63. Mice were gavaged with 150 µl of 6% carmine solution, and the time of gavage was recorded. Immediately after gavage, mice were single-housed in microisolator cages placed on an IsoRack. The cages were monitored for the appearance of the first red faecal pellet without disturbing the animals. Whole-gut transit time was calculated as the time interval between the gavage and the first observed red faecal pellet. To assess impacts of CHRM3 antagonist on gut motility, 1 mg per kg 4-DAMP (Cayman Chemical, 14574) was administered by intraperitoneal injection 1 h before carmine gavage. Faecal water content Four to five faecal pellets were collected into tubes and were dried in the oven at 80 °C overnight. The faecal water content was calculated according to the equation: (fresh faeces weight − dried faeces weight)/fresh faeces weight × 100%. LP immune cell isolation After euthanization, the ileal and colon tissues were collected, and the Peyer’s patches and the remaining mesentery/fat were excised. Tissues were kept in cold DMEM on ice until all tissues were collected. The gut tissues were opened longitudinally with a ball-tipped scissors and rinsed sequentially in three beakers containing ice-cold PBS. The tissue was cut into 1 cm pieces into 50 ml conical tubes containing 10 ml of PBS (without Ca2+ and Mg2+), 1 mM dithiothreitol and 30 mM EDTA. Tubes were shaken at 37 °C, 250 rpm for 10 min, and the tubes were placed almost horizontally. The tissues were drained with a tissue strainer and washed with 10 ml of PBS on the strainer. The tissues were then transferred into a new 50 ml conical tube containing 10 ml of PBS and shaken at 37 °C, 250 rpm for 10 min. Tissues were drained on a tissue strainer and rinsed with 10 ml of PBS. Tissues were transferred into six-well plates and minced before adding digestion enzyme buffer (2 ml of DMEM supplemented with 10 mM HEPES, 2% FBS, 2 mg ml−1 type IV collagenase (Gibco, 17104019), 0.5 mg ml−1 DNase I (Sigma-Aldrich, 10104159001)). Tissues were digested at 37 °C, 120 rpm for 30 min, followed by vigorous pipette mixing. The digestion was terminated by adding 5 ml of DMEM with 2% FBS. The digested suspensions were passed through a nylon mesh (100 μm pore size). Cells were pelleted and applied to 80% and 40% Percoll gradient to separate immune cells by centrifugation at 500g, room temperature for 20 min (without brake). Immune cells at the middle layer were collected for either bulk RNA-seq or flow cytometry analysis. LP immune cell scRNA-seq Library preparation and sequencing Immune cells at the middle layer were collected and sorted for live CD45+ cells. Cell densities were adjusted to 1000 cells per μl and submitted for single-cell library construction using Chromium Next GEM Single Cell 5’ Reagent Kits v2 (10x Genomics, PN-1000263). The single cell libraries were sequenced with Illumina NovaSeq X Plus platform. scRNA-seq data analysis The sequencing data were processed with Cell Ranger (10x Genomics Cloud Analysis) to generate a merged digital expression matrix using Cell Ranger (v.7.2.0) aggregation. Cell clustering and annotation The merged digital expression matrix from Cell Ranger was further analysed using Seurat (v.5.1.0) following the instructions64. In brief, cells were filtered to have more than 200 features and less than 5% mitochondrial transcripts. Variable features were then identified using the FindVariableFeatures function. The ScaleData function was used to regress out the sequencing depth for each cell. The identified variable features were used in principal component analysis (PCA) to reduce the data dimension. Then, 20 principal components were used in the UMAP analysis to further reduce the dimension to 2. The cells were then grouped into different cell clusters with a resolution of 0.5 using the function of FindClusters. Cluster marker genes were then found for each cluster using the FindAllMarkers function. Cell types were manually annotated based on the cluster marker genes. The proportion of each cell type was calculated by dividing the number of cells in each cluster by the total number of cells in the corresponding sample. UMAP plots were generated with monocle3 (v.1.3.1). Pseudobulk expression analysis Cells were grouped by cell type and colonization group (colonization condition) to create pseudobulk profiles using the AggregateExpression function in Seurat. Raw count data was extracted from the pseudobulk Seurat object using the GetAssayData function. A DESeq2 dataset was created using DESeqDataSetFromMatrix. DESeq2 was used to normalize the data and perform differential expression analysis. Differential expression results were obtained by comparing groups (WT versus Δchat B. breve). Differentially expressed genes were visualized using volcano plots. LP immune cell RNA-seq RNA isolation Immune cells isolated from the gut LP were lysed with buffer RLT (Qiagen, 79216) containing 1% β-mercaptoethanol, and the total RNA was isolated using the RNeasy Mini kit (Qiagen, 74106) following the instructions. RNA quality was assessed using the NanoPhotometer spectrophotometer for RNA purity and using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system. Library preparation mRNA was purified from total RNA using poly(T) oligo-attached magnetic beads and subsequently fragmented. First-strand cDNA was synthesized using random hexamer primers, followed by second-strand synthesis incorporating dTTP. The resulting double-stranded cDNA was processed through end repair, A-tailing, adapter ligation, size-selection, PCR amplification and purification. Library quality and concentration were evaluated using Qubit fluorometry and real-time PCR, and fragment size distribution was analysed on a Bioanalyzer 2100 system. Libraries were then pooled and sequenced on the Illumina NovaSeq X Plus platform. Data analysis Trimmed reads were generated using Cutadapt (v.5.0)58 with adapter removal and quality trimming. Trimmed reads were quantified at the transcript level using Salmon (v.1.4.0)65 with germinal centre (GC) and sequence bias correction enabled, using the Mus musculus reference transcriptome (GRCm39, Ensembl). Transcript abundances were summarized to the gene level using the tximport package (v.1.30.0)66 in R. Differential gene expression analysis was performed with DESeq2 (v.1.42.1)44. GSEA was conducted with clusterProfiler (v.4.10.1)67 using GSEA on a ranked list of genes. Flow cytometry LP immune cell flow cytometry LP immune cells were prepared as described above. Cells were stained sequentially with Near-IR live/dead (Invitrogen, L34976, 1:1,000) for 10 min, Fc blocker (BioLegend, 156604, 1:100) and antibodies against CD45 (BioLegend, 103149, 1:200), B220 (BioLegend, 103243, 1:200), CD19 (BioLegend, 115534, 1:200), TACI (BioLegend, 133403, 1:200) and CD138 (BioLegend, 142507, 1:200) for 20 min. Then cells were permeabilized with Cytofix/Cytoperm Fixation and Permeabilization solution (BD, 554722) and stained intracellularly for IgA (Thermo Fisher Scientific, 17-4204-82, 1:200) and IgM (eBioscience, 11-5790-81, 1:200) for 2 h. Stained cells were analysed using CytoFLEX (Beckman Coulter). Data were acquired using CytExpert v.2.1 and analysed using FlowJo software (v.10.9.0). Intracellular staining of IgA to distinguish IgA plasma cells was based on direct comparison between intracellular versus surface IgA staining, where intracellular staining revealed modest but significantly higher proportions of IgA+ cells (Supplementary Fig. 6). Bacterial flow cytometry Bacterial cells recovered from faecal samples were blocked with 1% BSA and stained for IgA (Thermo Fisher Scientific, 17-4204-82) for 20 min. Cells were then resuspended with SYTO9 green fluorescent nucleic acid stain (Thermo Fisher Scientific, S34854) and analysed using CytoFLEX (Beckman Coulter). Bacterial cells recovered from in vitro cultures or from Rag1−/− mouse faeces were stained with supernatant (containing free IgA) from faecal suspension before blocking. Data were analysed using FlowJo (v.10.9.0). IgA ELISA Mouse faecal pellets were weighed and homogenized in 1 ml of PBS with lysing matrix D (MP, 116913050-CF) for 15 s in BeadBeater (Biospec). The mixtures were sonicated for 30 s and thoroughly vortexed. Faecal water samples were collected after centrifuging at 10,000g for 5 min. Immuno MaxiSorp Plate (Thermo Fisher Scientific, 439454) were coated for 30 min at 37 °C with 50 μl anti-mouse IgA capture antibody (Abcam, ab97231). After washing with TBS-T (20 mM Tris, 150 mM NaCl, 0.1% Tween-20, pH 7.4), the plates were blocked with 150 μl of 3% BSA (in PBS) for 2 h at room temperature. After washing with TBS-T, 50 μl of mouse IgA standards (BD Biosciences, 553476) or samples were added to the plates and incubated for 2 h at room temperature. After washing with TBS-T, plates were labelled with 50 μl of HRP-conjugated goat anti-mouse IgA (Sigma-Aldrich, A4789) for 30 min at room temperature. Plates were then washed with TBS-T and developed with 100 μl of TMB substrate (Thermo Fisher Scientific, 34022) for 10 min at room temperature. Development was stopped with 100 μl of 2 N H2SO4 (Avantor, H381-05), and the plates were read at 450 nm using a SpectraMax i3x. LCN2 ELISA LCN2 ELISA was performed using the Mouse Lipocalin-2 DuoSet ELISA kit (R&D Systems, DY1857) according to the manufacture’s instructions. Mouse faecal samples were processed using the same method described for IgA ELISA. The Immuno MaxiSorp Plates (Thermo Fisher Scientific, 439454) were coated with 4 µg ml−1 of capture antibody at room temperature for 2 h. After washing with TBS-T (20 mM Tris, 150 mM NaCl, 0.1% Tween-20, pH 7.4), the plates were blocked with 150 μl of 1% BSA (in PBS) for 1 h at room temperature. After washing with TBS-T, 100 µl of diluted samples and standards were added to plates and incubated at room temperature for 2 h. After washing with TBS-T, 100 µl of 50 ng ml−1 detection antibody was added to each well and incubated at room temperature for 2 h. After washing with TBS-T, the plates were labelled with 100 µl of Streptavidin-HRP for 20 min at room temperature. After washing with TBS-T, the plates were developed with 100 µl of TMB substrate at room temperature for 10 min. The development was stopped with 50 µl of 2 N H2SO4 (Avantor, H381-05) and read at 450 nm using the SpectraMax i3x system. Bacterial 16S rRNA-seq Bacterial genomic DNA was extracted using MagAttract PowerSoil DNA EP kit (Qiagen, 27100-4-EP). The 16S rRNA V4 region was amplified from the genomic DNA using dual index multiplexing strategy as described previously68. Libraries were quantified using KAPA Library Quantification Kit (Roche Diagnostics, KK4835), and sequenced on an Illumina MiSeq platform using a MiSeq Reagent Kit v2 (500 cycles) (Illumina, MS-102-2003). Sequencing data were collected through MiSeq System Suite v.4.1.0, and processed using QIIME 2 (v.2024.2). S. Tm ΔssaV infection S. Tm ΔssaV (a gift from J. Galán laboratory)32 was streaked on LB agar plate supplemented with 50 µg ml−1 streptomycin and cultured at 37 °C overnight. A colony was picked and grown in LB broth for 6 h at 37 °C with shaking at 220 rpm. We next adjusted the bacteria density to 106 colony-forming units (CFU) per ml, and gavaged each mouse with 200 µl (equivalent to 2 × 105 CFU each mouse). For enumerating faecal S. Tm ΔssaV CFUs, faeces were weighed, homogenized by bead beating, serially diluted in sterile PBS and plated on LB agar plates supplemented with 50 µg ml−1 streptomycin. Plates were incubated for 12 h at 37 °C before enumeration. Statistical analysis All statistical analyses were performed using GraphPad Prism (v.10.1.1) or R (v.4.3.0). PRESTO-Salsa screening data, mouse LP immune cell bulk RNA-seq data, bacterial RNA-seq data and pseudobulk differential expression analysis of scRNA-seq data were analysed using DESeq2 (v.1.42.1) in R. scRNA-seq data were processed using Seurat (v.5.1.0) and Monocle3 in R. For comparisons between two groups, unpaired two-sided t-tests or Welch’s t-tests were used as appropriate. For comparisons among more than two groups, one-way ANOVA followed by Tukey’s post hoc test, Welch’s ANOVA followed by Games–Howell post hoc test or Kruskal–Wallis test followed by Dunn’s multiple-comparison test was used as appropriate. For selected nonparametric comparisons, two-sided Wilcoxon rank-sum tests were used. For repeated-measures experiments, a two-sided mixed-effects model followed by Šidák’s multiple-comparison test was used. For DESeq2 analyses, statistical significance was assessed using two-sided Wald tests with Benjamini–Hochberg correction for multiple comparisons. The statistical test used for each experiment is indicated in the corresponding figure legend and source data. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10592-7. Supplementary information Supplementary InformationSupplementary Note, Supplementary Figs. 1–6 and Supplementary References. Supplementary Note: a detailed description of in vivo versus in vitro commensal metabolome GPCR activity differences (related to the ‘In vitro and in vivo metabolome screens’ section). Reporting Summary Supplementary Table 1List of commensal strains, culture condition and origins. Supplementary Table 2Fold change and P value data from the PRESTO-Salsa screen of commensal in vivo and in vitro metabolomes against 314 conditional GPCRs. Supplementary Table 3Summary of GPCR hits activated by commensal in vivo and in vitro metabolomes. Supplementary Table 4GMM recipe. Supplementary Table 5Information of acetyltransferase from bacteria. Supplementary DataSource data for Supplementary Figs. 2, 3 and 6. Supplementary Note, Supplementary Figs. 1–6 and Supplementary References. Supplementary Note: a detailed description of in vivo versus in vitro commensal metabolome GPCR activity differences (related to the ‘In vitro and in vivo metabolome screens’ section). Reporting Summary List of commensal strains, culture condition and origins. Fold change and P value data from the PRESTO-Salsa screen of commensal in vivo and in vitro metabolomes against 314 conditional GPCRs. Summary of GPCR hits activated by commensal in vivo and in vitro metabolomes. GMM recipe. Information of acetyltransferase from bacteria. Source data for Supplementary Figs. 2, 3 and 6. Source data Source Data Fig. 1 Source Data Fig. 2 Source Data Fig. 3 Source Data Fig. 4 Source Data Fig. 5 Source Data Extended Data Fig. 1 Source Data Extended Data Fig. 2 Source Data Extended Data Fig. 3 Source Data Extended Data Fig. 4 Source Data Extended Data Fig. 6 Source Data Extended Data Fig. 7 Source Data Extended Data Fig. 8 Source Data Extended Data Fig. 9 Source Data Extended Data Fig. 10 Source Data Fig. 1 Source Data Fig. 2 Source Data Fig. 3 Source Data Fig. 4 Source Data Fig. 5 Source Data Extended Data Fig. 1 Source Data Extended Data Fig. 2 Source Data Extended Data Fig. 3 Source Data Extended Data Fig. 4 Source Data Extended Data Fig. 6 Source Data Extended Data Fig. 7 Source Data Extended Data Fig. 8 Source Data Extended Data Fig. 9 Source Data Extended Data Fig. 10