Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Feb 13.
Published in final edited form as: Nat Microbiol. 2025 Oct 10;10(11):2797–2807. doi: 10.1038/s41564-025-02141-1

Human gut bacteria produce structurally related monoglycolipids with contrasting immune functions

Ji-Sun Yoo 1, Da-Jung Jung 1, Byoungsook Goh 1, Kyoo Heo 1, Wen Zheng 2, ChangWon C Lee 2, Jeong In Seo 1, Naama Geva-Zatorsky 2, Meng Wu 2,5, Seung Bum Park 3, Dennis L Kasper 2, Sungwhan F Oh 1,4,6
PMCID: PMC12896134  NIHMSID: NIHMS2124289  PMID: 41073663

Abstract

Gut symbiont Bacteroides fragilis can produce alpha-galactosylceramides (BfaGCs); sphingolipids with immunomodulatory functions that regulate colonic natural killer T (NKT) cells. However, their synthesis pathway and whether other human gut bacteria can produce them is unclear. Using genetic and metabolomic approaches, we mapped the sphingolipid biosynthesis pathway of B. fragilis and identified that alpha-galactosyltransferase (agcT) is essential and sufficient for colonic NKT cell regulation in mice. The distribution of agcT is restricted to only a few species among Bacteroidales. However, structural homologs of AgcT, such as BgsB, are widely distributed in gut microbiota and produce alpha-glycosyl diacylglycerols (aGDGs), particularly in Enterococcus. Analysis of infant gut metagenomes revealed that B. fragilis predominantly accounts for agcT abundance regardless of the cohort, but bgsB-encoding bacteria were taxonomically diverse and showed dynamic changes with host age. Additionally, aGDGs from bgsB-encoding species act as antagonistic ligands for BfaGC-mediated NKT cell activation in vitro and in vivo. Our findings highlight the relevance of immunoreactive glycolipid-producing symbionts in the human gut microbiome, particularly in early life.

Introduction

The gut microbiota is a unique ecosystem, exhibiting immense taxonomic, genetic, and molecular diversity. The collective metagenome of gut symbionts exceeds the host genome by 2 to 3 orders of magnitude, enabling the production of a vast array of structurally diverse metabolites 5. These microbial metabolites can profoundly influence host health and disease through a variety of mechanisms. However, a significant portion of the metagenome remains unannotated or uncharacterized; as a result, the discovery of novel molecular structures often lacks corresponding understanding of their biosynthesis mechanisms. This challenge is further compounded by extensive redundancy at the gene and pathway levels across microbial taxa, making it difficult to resolve these mechanisms at the level of the microbial community. Moreover, the gut microbiota is highly dynamic, with temporal fluctuations in composition and function, adding another layer of complexity to determining the biological relevance of microbial activity within the host-microbiota context.

Among microbiota-derived metabolites, glycosphingolipids act as ligands for MHC class I-like molecule CD1d, enabling them to modulate the host immune system6. Notably, BfaGCs are among the first identified symbiont-derived (endobiotic) glycosphingolipids with such immunomodulatory effects2. BfaGCs are originally reported as ‘inhibitory’ against potent NKT cell agonists such as KRN700017, the best characterized ligand of CD1d, along with their regulatory activity on colonic NKT cell proliferation. Further structure-activity relationship studies revealed that not only BfaGCs can compete against strong CD1d ligands, but also can induce characteristic immunomodulatory responses, distinct from those of previously known Th1- or Th2-skewed activators. Notably, host dietary factors can dictate the structure of BfaGCs produced in the gut lumen, confering structure-specific immunomodulatory actions on NKT cells3.

Despite their unique functionality, because BfaGC (and more broadly, bacterial alpha-galactosylceramides (aGCs)) biosynthesis was only partially elucidated, it has remained unclear whether B. fragilis is the sole producer of aGCs within the gut microbiome, or if other species also can synthesize aGCs or other structurally related glycolipids. Such limitations are common for microbiota-derived metabolites: even when their molecular structures have been characterized, the corresponding biosynthetic pathways often remain unidentified, rendering them orphan metabolites. This presents a major barrier to systematic, microbiome-level investigations. To overcome these challenges, we applied a multi-faceted strategies to characterize metabolites with previously uncharacterized biosynthetic origins. We conducted forward- and reverse-genetic approaches to identify the genes responsible for sphingolipid biosynthesis in B. fragilis and related species. Building on these genetic insights, we further elucidated the in vivo immunological functions of symbiont-derived metabolites and explored metagenomic variation in key biosynthetic genes across diverse human cohorts.

Results

Characterization of BfaGC biosynthesis confirms its necessity for NKT cell activation and colonic regulation in early life.

To identify the gene responsible for aGC synthesis in B. fragilis, we utilized forward genetics-based, targeted metabolomic approaches (Extended Data 1a). We generated a genome-wide transposon mutant library7, performed a high-throughput targeted lipidomic screening, and identified a mutant unable to synthesize BfaGCs (Tn8A7, with the transposon inserted in BF9343_3069, Extended Data 1a-c). Based on its protein motifs and putative function as two-component sensor/kinase response regulator, we hypothesized that BF9343_3069 functions as a regulator rather than a glycosyltransferase. To identify the downstream glycosyltransferase regulated by BF9343_3069, we conducted a transcriptomic analysis of BF9343_3069 knockout mutant Δ (Extended Data 1d). Among 20 significantly downregulated glycosyltransferases, we excluded those located within capsular polysaccharide operons and those classified as β-anomeric inverting glycosyltransferases8–10. BF9343_3149 was the only remaining candidate predicted to be an α-anomeric retaining (α→α) glycosyltransferase regulated by BF9343_3069 (Extended Data 1e). We generated an isogenic mutant of BF9343_3149 (described as ΔagcT; alpha galactosyl ceramide transferase) and confirmed the complete loss of BfaGC production (Fig. 1a).

Figure 1. BfaGC is necessary and sufficient to modulate colonic NKT cells.

Figure 1.

Open in a new tab

a) Sphingolipid biosynthesis pathway of B. fragilis confirmed with individual gene knockout strains. CerS produces 3-keto-4,5-dihydro-ceramide (ketoCer) as a transient intermediate, which is further reduced by CerR to produce dihydroceramide (Cer) and complex sphingolipids such as BfaGCs. Extracted ion chromatograms (XICs) of 3-keto-dihydrosphingosine (KDS) (17:0), ketoCer (34:0), Cer (34:0), and alpha-galactosylceramides (aGC) (34:0) from B. fragilis wild type (WT), Δspt, ΔcerS, ΔcerR, and ΔagcT were shown.

b) Deleting individual genes involved in BfaGC biosynthesis abolishes CD1d-mediated IL-2 induction in NKT cells elicited by B. fragilis total lipids. NKT cell hybridomas (24.7) were incubated with CD1d-lipid complex and IL-2 secretion was measured by ELISA (biologically independent replicates, n = 4 per group). KRN700017 (n=3) was included as a positive control.

c) Heterologous expression of agcT in ceramide-producing species (biologically independent replicates, n=4 for each group) enabled induction of IL-2 in NKT cell hybridomas (24.7).

d) BfaGC production in vivo depends on agcT. C34:0 BfaGC levels were measured in stool samples from 5-week-old female C57BL/6 mice monocolonized with B. fragilis wild-type (n=3) or agcT mutant strains (n=4) and compared to GF (n=4).

f) Selective loss of BfaGC by deletion of agcT or its upstream regulator Δ3069 abrogated colonic NKT cell modulation by B. fragilis. At 6 weeks of age, colonic NKT cell frequencies were analyzed in male and female Swiss Webster mice monocolonized from birth with B. fragilis wild type (n=5), Δ3069 (n=4), or ΔagcT (n=5), alongside SPF (n=13) and GF (n=6) mice. The representative gating strategy is shown in Fig. S3c-d.

a-d) Each dot indicates a biologically independent replicate. Results represent two independent experiments showing similar trends. Statistical analysis was performed using one-way ANOVA followed by Dunnett’s multiple comparisons test in a-c, andone-way ANOVA followed by Tukey’s multiple comparisons test in d. Bars and error bars represent the mean ± s.e.m.

In parallel to loss-of-function screening, we conducted a gain-of-function screening by generating a transformant library of the whole B. fragilis genome (Extended Data 1a). We introduced the B. fragilis genomic DNA library into Phocaeicola vulgatus (P. vulgatus), which produces ceramides but no aGCs. Targeted lipidomic analysis identified one pooled colony producing aGCs and two clones isolated from the pool were confirmed to produce aGCs and express agcT (Extended Data 1a, f). To confirm the identified gain-of-function transformant, we heterologously expressed B. fragilis agcT in P. vulgatus and B. thetaiotaomicron. Production of aGCs was observed in both species (Extended Data 1g-i), providing orthogonal evidence that AgcT is a sufficient enzyme to produce aGCs from ceramides, consistent with the biochemical study of the gene 11.

In B. fragilis, only the serine-palmitoyl transferase (spt) gene, encoding the first enzyme in the ceramide synthesis pathway, has been functionally characterized2. To fully map the BfaGC biosynthetic pathway, we further characterized the entire BfaGC biosynthesis pathway genes using targeted approaches. From a homology search with recently annotated bacterial ceramide pathway genes (Caulobacter crescentus12), we identified putative ceramide synthase (CerS; BF9343_4218) and ceramide reductase (CerR; BF9343_4240) and generated isogenic knockout mutants of each B. fragilis genes. Sphingolipid profiles of each mutant, in addition to Δspt and ΔagcT were compared (Fig. 1a, Extended Data 1j, 2). As reported, Δspt could not produce sphingolipids at all. Deletion of cerS resulted in the accumulation of 3-ketosphinganine (3-keto-dihydrosphingosine or KDS), and the loss of cerR caused the accumulation of 3-keto-4,5-dihydro-ceramide (ketoCer), the immediate upstream intermediates of each respective gene. Of note, B. fragilis genes in ceramide biosynthesis did not constitute a biosynthetic gene cluster, as spt (BF9343_2380), cerS (BF9343_4218), cerR (BF9343_4240), and agcT (BF9343_3149) are scattered across the genome.

Characterizing its biosynthesis pathway, we next investigated the role of BfaGC on NKT cell functions at cellular and in vivo levels. To assess NKT cell activation, we employed an in vitro antigen presentation assay13–15 using CD1d molecules loaded with lipid extracts from BfaGC-deficient and -sufficient bacterial strains (Extended Data Fig. 3a). As expected, lipid extracts from B. fragilis mutants lacking BfaGC biosynthetic genes (Δspt, ΔcerS, ΔcerR, or ΔagcT) failed to elicit IL-2 secretion from NKT cell hybridoma (Fig. 1b). On the other hand, lipid extracts from agcT-transformed P. vulgatus, which produce aGCs, induced IL-2 production, demonstrating that agcT and aGCs are necessary and sufficient for NKT cell activation (Fig. 1c).

We also examined the in vivo role of agcT and BfaGCs using gnotobiotic mice. Stool from mice monocolonized with B. fragilis wild type contained significantly higher levels of BfaGCs compared to that from ΔagcT-colonized or germ-free (GF) mice (Fig. 1d). To further assess the effect of BfaGC on NKT cell levels, we monocolonized GF mice from birth with either B. fragilis WT or BfaGC-deficient mutants and analyzed colonic NKT cell frequencies at 6 weeks of age (Fig. 1e and Extended Data 3b,c). As previously reported, GF mice exhibited elevated NKT cell numbers, whereas B. fragilis WT-colonized mice showed normalized NKT cell levels, comparable to specific-pathogen-free (SPF) mice2. However, colonization with BfaGC-deficient strains (Δ3069 or ΔagcT) failed to normalize NKT cell numbers, which remained elevated and similar to GF mice. These results confirmed that BfaGC is required for the early-life regulation of colonic NKT cell abundance.

Metabolomic and genomic analysis revealed a narrow distribution of aGCs among gut symbionts.

To elucidate whether additional gut symbionts beyond B. fragilis can produce aGCs, we selected 25 prominent human gut microbial species encoding the spt gene and performed targeted lipidomic analysis (Fig. 2a-b). Among these 25 species, 24 species produced ceramides, but only B. fragilis and B. salyersiae synthesized aGCs with essentially identical structures (Fig. 2b and Extended Data 4a).

Figure 2. Lipidomic and genomic analyses revealed a narrow distribution of agcT among gut Bacteroidales.

Figure 2.

Open in a new tab

a) Most Bacteroidales species in the human gut produce ceramides. The heatmap shows LC-MS/MS intensities of ceramide species detected in lipid extracts from individual bacterial cultures with background values determined from blank samples.

b) aGC was only found in B. fragilis and B. salysiae.

c) Sequence homology analysis revealed distinct conservation patterns between upstream ceramide synthesis enzymes (Spt, CerS, CerR) and AgcT. The sequence identity was determined by searching species genomes against UniRef90 proteins; Spt (A0A0K6BUE2), CerS (E1WSI8), CerR (A0A2M9VAV5) and AgcT (A0A380YRQ3).

d) Analysis of agcT homologs across 24,562 strain genomes in the UHGG catalogue revealed a dichotomous distribution. B. fragilis and B. salyersiae consistently carried agcT, while other species showed little to no conservation. The bar plot shows the proportion of strains harboring agcT homologs relative to the total number of strains per species.

Protein sequence homology analysis of sphingolipid biosynthetic enzymes, using B. fragilis genes as references, supported the lipidomic results. Nearly all tested species encoded homologs of ceramide synthesis genes (spt, cerS, and cerR), but only B. fragilis and B. salyersiae harbored agcT (Fig. 2c). Additionally, analysis of strain genomes from the Unified Human Gastrointestinal Genome (UHGG18) catalogue confirmed that the vast majority of B. fragilis and B. salyersiae strains possessed agcT (Fig. 2d). In contrast, 14 other species exhibited no conservation (0%), and 9 species exhibited low conservation (0.1~15%) of agcT. Consistently, all tested B. fragilis and B. salyersiae strains produced aGCs, whereas P. copri and B. nordii stratins did not (Extended Data 4b). Functional assays also confirmed that NKT cell activation was dependent on the presence of aGCs (Extended Data 4c). In addition to aGCs, α-glucuronosyl ceramides (aGlcCers), known CD1d ligands capable of activating NKT cells, have been previously reported in Sphingomonas species19–21. Since Sphingomonas is neither prevalent nor abundant in the human gut microbiome and is not represented in the UHGG catalogue, we examined whether any of the selected Bacteroidales species produced aGlcCers, using two Sphingomonas as positive controls. None of the symbiotic Bacteroidales species produced aGlcCers, while Sphingomonas species did (Extended Data Fig. 4d).

Metagenomic exploration revealed widespread structural homologs of AgcT across microbiota

Several classes of α-anomeric glycolipids have been identified as microbial CD1d ligands22,23, prompting us to investigate whether additional symbiotic gut bacteria harbor AgcT-related proteins. To broaden our search for potential microbial CD1d ligands, we focused on AgcT and its conserved domain cd03817, a retaining glycosyltransferase responsible for producing characteristic α-glycosyl moiety. We compiled a list of 1,461 gut bacterial species from 18,780 human gut metagenome samples24 and retrieved their genomes from the National Center for Biotechnology Information (NCBI) database. Using both sequence homology and conserved domain structure searches14,15, we identified potential AgcT homologs (Fig. 3a). The primary sequence-based search revealed only 16 close homologs including B. salyersiae (Fig. 2b), all restricted to the order Bacteroidales. In contrast, the cd03817 domain-based search uncovered 288 putative homologs widely distributed across the gut microbiota (Fig. 3a and Extended Data 5a). Given that microbial regulation of NKT cell levels occurs within a critical early-life window25,26, we focused on 179 prevalent (>1%) and abundant (>0.01%) species in the infant microbiome. Within this subset, AgcT close homologs were found in 3 species, while cd03817 domain-containing homologs were found in 39 species. Most of these cd03817 domain-containing homologs were found in the phylum Bacilliota, especially in the Lactobacillales order such as Streptococcus, Lactobacillus, and Enterococcus (Fig. 3b-c). Notably, Enterococcus faecalis (E. faecalis) emerged as the most abundant species with cd03817 domain-containing proteins (Fig. 3b). These cd03817 domain-containing homologs shared classification with B. fragilis AgcT across multiple databases, including Carbohydrate-Active enZYmes (CAZy), Kyoto Encyclopedia of Genes and Genomes (KEGG) orthology, and Enzyme Commission (EC) classes (Extended Data 5b). Furthermore, in silico–generated 3D structural modeling of several Lactobacillales cd03817 proteins revealed substantial structural similarity to B. fragilis AgcT, despite their relatively low sequence identity (Fig. 3d).

Figure 3. Structural homologs of B. fragilis AgcT are widely distributed in gut microbiota.

Figure 3.

Open in a new tab

a) Workflow for identifying sequence and structural homologs of B. fragilis AgcT was presented. Protein homology searches were conducted using DIAMOND BLAST27 for primary sequence similarity and HMMER hmmsearch28 for domain-based detection of homologs.

b) Taxonomic cladogram with homolog search results showed the distribution of AgcT homologs among prominent infant gut bacterial species.

c) Phylogeny of 42 cd03817 proteins from gut bacterial species showed a clear separation between Lactobacillales and Bacteroidales clusters. The maximum likelihood tree was generated using IQ-TREE229.

d) The superimposed structures revealed structural similarity between AgcT (A0A380YRQ3) and cd03817 proteins of various Lactobacillales bacteria.

Structural homologs of AgcT produce alpha-glycosyl-diacylglycerols.

Multiple AgcT structural homologs in Lactobacillales are annotated as 1,2-diacylglycerol 3-alpha-glycosyltransferase and are predicted to produce alpha-glycosyldiacylglycerols (aGDGs)30. To investigate this, we cultured several representative species and analyzed their lipid profiles. LC-MS/MS profiling revealed varying levels of C34:1 aGDG, with MS/MS spectra that were essentially identical to the synthetic standards (Fig 4a and Extended Data a-b). Additional chain length variants of aGDGs were also detected across species (Extended Data 6c). For further analysis, we focused on Enterococcus species, which are among the most abundant species encoding cd03817-family proteins in the infant microbiota (Fig. 3b). All tested strains of Enterococcus produced aGDGs (Extended Data 6d), and BgsB was identified as the cd03817-family glycosyltransferase. To confirm its role, isogenic bgsB knockout mutants were generated in E. faecalis and E. faecium, confirming that bgsB is essential for aGDG biosynthesis in bacterial culture (Fig. 4b-c). Using a gnotobiotic mouse model, we compared aGDG levels in mice monocolonized with either E. faecalis WT or bgsB mutant and confirmed that bgsB is required for aGDG production in vivo (Fig. 4d and Extended Data 6e). Notably, the levels of glycosyl-diacylglycerols were found to be almost exclusively dependent on the presence of the gut microbiota. (Extended Data 6f).

Figure 4. Characterization of cd03817 family proteins in gut symbionts producing aGDGs.

Figure 4.

Open in a new tab

a) Multiple Lactobacillales bacteria produce aGDGs. The abundance of aGDGs in individual bacterial cultures was assessed by LC-MS analysis. Each dot represents biologically independent replicates (n=2 for each species). Data are presented as mean values +/− SEM.

b)

c) bgsB is essential for aGDG production in Enterococcus. XICs and abundance of aGDGs in Enterococcus WT and bgsB mutants were shown.

d) bgsB is responsible for aGDG production in vivo. C34:1 aGDG levels were measured in stool samples from 4-week-old male and female C57BL/6 mice monocolonized with E. faecalis WT (n=10) or bgsB mutants (n=6) and compared to GF mice (n=7). Each dot represents an individual mouse. Data are presented as mean values +/− SEM. Data represent two independent experiments with consistent trends. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test.

agcT and bgsB exhibit distinct compositional patterns across development.

To explore the distribution of two structurally related but functionally distinct glycosyltransferases, we analyzed the abundance and composition of agcT- and bgsB-encoding species in early-life human gut microbiota using multinational, longitudinal infant metagenome cohorts (TEDDY and DIABIMMUNE)31–34 (Extended Data 7a). Homologs were identified based on ≥40% protein sequence identity from species genomes (Extended Data 7b). In both cohorts, over 95% of agcT-encoding species were B. fragilis, whereas bgsB-encoding species were more taxonomically diverse and exhibited dynamic changes during early development (Fig. 5a-b). In the DIABIMMUNE cohort, although Russian infants displayed higher overall bgsB abundance compared to those from Estonia and Finland, the contrasting compositional patterns—uniform agcT and diverse bgsB— were consistent across all three countries (Extended Data Fig. 7c). Notably, Enterococcus species contributed 10–20% to total bgsB abundance before one year of age, but this proportion declined with age.

Figure 5. Longitudinal metagenomic landscapes of agcT and bgsB show distinct patterns in infant gut.

Figure 5.

Open in a new tab

a-b) agcT and bgsB show differences in composition. B. fragilis consistently dominates agcT abundance regardless of host age, whereas bgsB abundance is distributed among diverse species that undergo significant compositional shifts over developmental time. The abundance plot shows the relative abundance of agcT- and bgsB-encoding species across time, visualized using LOESS smoothing in ggplot233.

We further expanded our analysis to additional early-life and adult gut metagenomic datasets 24,35–40(Extended Data 7d-e). Across all cohorts, the composition of bgsB-encoding species remained highly variable, while agcT was consistently dominated by B. fragilis, highlighting distinct compositional patterns of these two genes.

aGDGs antagonize BfaGC actions on NKT cells.

Certain aGDGs from pathogenic bacteria have been identified as NKT cell ligands41–43. To assess whether aGDGs produced by bgsB-encoding symbionts function similarly, we tested their ability to activate NKT cells using plate-bound CD1d assays (Extended Data 3a). To our surprise, total lipid extracts from aGDG-producing species did not induce NKT cell activation (Fig. 6a). However, when applied together with BfaGCs during the CD1d loading step, these lipid extracts antagonized BfaGC-induced NKT cell activation in a dose-dependent manner (Fig. 6b and Extended Data 8a). Lipid extracts from bsgB knockout mutants failed to inhibit NKT cell activation (Fig. 6c), confirming that the antagonistic activity is specific to aGDGs. Since this antagonism occurred at the CD1d loading step, we assessed the competitive binding between BfaGCs and aGDGs using a CD1d enrichment assay44. The CD1d enrichment assay revealed that aGDGs compete with BfaGCs for CD1d binding (Extended Data 8b).

Figure 6. aGDGs inhibit the actions of BfaGCs on NKT cells.

Figure 6.

Open in a new tab

a–c) Lipid extracts from B. fragilis, alone or in combination with lipid extracts from other bacterial species, were loaded onto CD1d molecules and NKT cell activation was assessed as described in Fig. S3b. Each dot represents a biologically independent replicate. Data are presented as mean values +/− SEM. Data are representative of three independent experiments showing consistent trends. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test.

d) Lipid extracts from bgsB-encoding species antagonized to B. fragilis-mediated NKT cell activation.

e) E. faecalis lipid extracts suppressed BfaGC-induced NKT cell activation in a dose-dependent manner.

f) The antagonistic effect of bacterial lipids on NKT cell activation is dependent on the presence of bgsB.

d-e) Female 7-week-old C57BL/6 mice (n=5 per group) were intraperitoneally injected with 5 μg synthetic BfaGC (SB2217) with or without 10 μg synthetic aGDGs (BbGL-2). After 19 hours, serum and spleens were collected for analysis. Each dot represents an individual mouse, and data represent three independent experiments. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test.

e) aGDGs reduced BfaGC-induced serum IFN-γ levels.

f) aGDGs suppressed BfaGC-induced upregulation of co-stimulatory molecules on splenic dendritic cells.

g) Monocolonization of E. faecalis from birth failed to normalize colonic NKT cell development. Pregnant Swiss Webster mice were monocolonized with E. faecalis WT or bgsB mutant. Colonic NKT cells were analyzed in male and female offspring at 6 weeks of age (n=10 for E. faecalis WT, n=6 for E. faecalis bgsB mutant) alongside SPF (n=10) and GF (n=10) Swiss Webster mice of both sexes.

h) Experimental overview for h–i. Swiss Webster neonates were gavaged with synthetic lipids (250 ng SB2217 and 500 ng BbGL-2) on days 3, 6, 9, 12, and 15 after birth. Colonic NKT cells were analyzed in male and female offspring at 6 weeks of age.

i) BfaGCs reduced colonic NKT cell levels. Swiss Webster GF mice gavaged with synthetic BfaGCs (SB2217) (n = 4) exhibited reduced colonic NKT cells compared with GF mice (n = 8) and levels comparable to SPF mice (n = 7).

j) aGDGs suppressed the BfaGC-mediated reduction of colonic NKT cell levels. Swiss Webster GF mice gavaged with synthetic BfaGCs (SB2217) plus synthetic aGDGs (BbGL-2) (n = 10) exhibited elevated colonic NKT cells compared with B. fragilis monocolonized mice (n = 11) and levels comparable to GF mice (n = 11).

f–i) Each dot represents an individual mouse, and data represent two independent experiments. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test.

To assess the antagonistic function in vivo, we intraperitoneally injected BfaGCs alone or in combination with aGDGs and measured serum IFN-γ levels and co-stimulatory molecule expressions on splenic dendritic cells. While aGDGs alone did not elicit serum cytokine production or dendritic cell activation, it significantly attenuated the immune response induced by BfaGCs (Fig. 6d-e). In addition, we tested antagonistic action of aGDGs to BfaGC-restricted NKT cell proliferation during early life using gnotobiotic mice model. Germ-free mice monocolonized with E. faecalis from birth did not exhibit reduced colonic NKT cell numbers, unlike those colonized with B. fragilis (Fig. 6f). Furthermore, the colonic NKT cell-normalizing activity of the synthetic BfaGC in GF neonates was blunted by the co-administration of synthetic aGDGs (Fig. 6g,h,i). Together, these results demonstrate that aGDGs act as in vitro and in vivo antagonists of BfaGCs, interfering with both NKT cell activation and early-life colonic NKT cell regulation.

Discussion

Having co-evolved over millions of years, the host immune system has developed sophisticated machinery to distinguish subtle structural differences in microbial metabolites. While synthetic aGCs are known to elicit proinflammatory responses6, symbiont-derived BfaGCs can induce immunomodulatory functions, with activity determined by specific structural features. In this report, we elucidated the biosynthesis of BfaGCs in the gut microbiota by identifying the responsible synthase and characterizing its metagenomic homologs, that produce structurally related yet functionally distinct glycolipids (Extended Data 9).

Even though sphingolipid-producing species are ubiquitous in the human gut, agcT was found almost exclusively in B. fragilis, suggesting selective acquisition by certain commensals to occupy a specialized gut niche. In contrast, structural homologs, such as bgsB in E. faecalis, are broadly distributed across taxa. Comparative metagenomic analysis across multiple cohorts showed that agcT-encoding species follow a “specialist” pattern—produced by a few taxa—whereas bgsB-encoding species follow a “generalist” pattern, being widespread in the community. These contrasting patterns highlight the complexity of microbial metabolite production in multi-species communities.

Of note, aGDGs, the products of BgsB, antagonize BfaGC-mediated NKT cell activation both in vitro and in vivo. Despite sharing domain classifications in protein family databases, agcT and bgsB differ markedly in function, underscoring the limitations of gene-centric annotations and the importance of molecular-level validation.

In addition, both agcT- and bgsB-encoding species are more abundant during early life, raising questions about the roles of their glycolipid products in shaping neonatal gut colonization and immune development. Further studies are needed to clarify how these metabolites influence colonization dynamics and host physiology. Overall, our work provides a generalizable framework to characterize biochemical pathways of structurally related metabolites, highlighting the value of integrating genomics, structural biology, and immunology to better understand symbiotic microbiota and their roles in host-microbe interactions.

Methods

Mice

All animal procedures were supervised by the Harvard Center for Comparative Medicine (IS187–06) and Brigham Women’s Hospital Center for Comparative Medicine (2019N000009) and maintained by the Institutional Animal Care and Use Facilities. Experimental groups were age and sex matched. All mice were housed under 12-hour light-dark cycle and controlled climate (temperature: 21 °C, humidity: 50%).

Germ-free Swiss-Webster mice were bred and maintained in inflatable plastic isolators on LabDiet 5K67. B. fragilis monocolonized mice were prepared by gavage of breeding pairs with a single bacterial strain (B. fragilis NCTC9343 wild-type, BF9343-Δ3069, BF9343-ΔagcT) and maintained in isolators to obtain offspring (F1 and later generations). E. faecalis monocolonized mice were prepared by gavage of a pregnant female and maintained in iso-P cages to obtain offspring. Stool samples from GF and monocolonized mice in isolators were regularly streaked onto plates and grown in both aerobic and anaerobic conditions to confirm sterility and absence of contamination.

SPF Swiss-Webster mice and C57BL/6 mice for in vivo assays were purchased from Taconic.

Bacterial culture

Individual bacterial cultures (Supplmentary Table 1), maintained as frozen stocks, were first streaked onto plates. A single colony was picked and inoculated into PYG liquid medium (2% proteose peptone, 0.5% yeast extract, 0.5% NaCl supplemented with 0.5% d-glucose, 0.5% K2HPO4, 0.05% l-cysteine, 5 mg/L hemin and 2.5 mg/L vitamin K1), and cultured either in an aerobic incubator or an anaerobic chamber. For Lactobacillus culture, MRS medium (HIMEDIA MV369) was used. After overnight growth, cultures were centrifuged, and bacterial pellets were stored –80°C until extraction.

Chemical compounds

Synthetic BfaGC (SB2217) was prepared by total organic synthesis3. Perdeuterated (d35) β-galactosylceramide (Matreya #1914), Sphinganine-d7 (d18:0) (Cayman #27145), N-omega-CD3-octadecanoyl-phytosphingosine (Matreya #2210), 3-keto sphinganine (d18:0) (Cayman #24380), bGluDAG (Avanti #840529), BbGL-2 (1-oleoyl-2-palmitoyl-3-(α-D-galactosyl)-sn-glycerol, Avanti #840528), MGlc-DAG (Avanti #840522p), and KRN7000 (Avanti #867000) were obtained from commercial vendors.

Lipidomic analysis

For bacterial lipid profiling, a methyl tert-butyl ether extraction method was utilized2,3. Perdeuterated (d35) beta-galactosylceramide (Matreya LLC), N-omega-CD3-Octadecanoyl-phytosphingosine (Matreya LLC), and Sphinganine-d7 (d18:0) (Cayman Chemical) were used as internal standards.

UHPLC-MS/MS condition.

An UHPLC-MS/MS system (Thermo Scientific Vanquish RP-UPLC connected to a Q Exactive Orbitrap) was used for bacterial lipid profiling. Positive and negative ion mode method was established with parameters of spray voltage, 3.80 kV (positive ion mode) or 3.25 kV (negative ion mode); sheath gas, 40 AU; auxiliary gas, 8 AU; capillary temperature, 350°C; Aux gas heater temperature, 350°C; S-lens RF level, 65.0 AU; mean collision energy, 22.5 AU. YMC-Triart C8 column (2.1 mm × 100 mm × 3 μm, 200 μl min−1) was used for the gradient liquid chromatography elution; 50% 2-propanol/10% acetonitrile/0.05% formic acid (0–2 mins), linearly increased to 85% 2-propanol/10% acetonitrile/0.05% formic acid (2–5 mins), held (5–12 mins), and then returned to the initial condition for 0.1 min, and held for 7.9 min at 40°C. The detection of fecal lipids was conducted employing UHPLC-MS/MS condition as previously reported.3

Targeted lipidomics.

A high-resolution (R=70,000 @ m/z 200) MS1 scan (250–1000Da followed by top 3 data-dependent acquisition (R=17,500 @ m/z 200, isolation window: 1.0 m/z) to acquire MS and MS/MS spectra by Xcalibur 4.0 (Thermo Fisher Scientific). MS/MS spectra of biogenic and synthetic molecules were acquired and directly compared. Relative quantitation of individual bacterial lipids was done by quantitation of area under MS1 peaks with normalization by internal standard recovery.

Transposon screening and targeted high-throughput LC-MS/MS lipidomics

Transposon insertion library of B. fragilis 638R were generated by conjugation with E. coli S17-lambda-pir harboring pSAM-Bt (Addgene_112497, deposited by Goodman and Gordon).7 A single colony of transposon-inserted mutants was isolated on BHI plate containing erythromycin 10 mg/mL and inoculated to 96-well plate. Individual samples were centrifuged, and the pellets were dissolved and extracted with isopropyl alcohol. Supernatant was transferred to V-bottom 96-well plate and analyzed by LC-MS/MS. A high-throughput LC method (Kinetex C8 30mmx2.1mmx2.6um, column temperature at 40 C) was applied: 80% acetonitrile / 0.1% formic acid (0–0.1min), ramped to 85% 2-propanol / 0.1% formic acid (0.1–0.5 min) and hold (0.5–2.5 mins), then returned to the initial condition for 1.5 min to re-equilibrate.

Metagenomic analysis.

The species-level profiles of multiple human gut metagenome datasets were obtained from curatedMetagenomicData v3.6.2.34 Reference genomes of gut bacterial species within the profiles were downloaded using NCBI Entrez Direct and NCBI Datasets tools. Searches for homologs of Spt, CerS, CerR and AgcT in reference genomes of gut bacterial species were performed using DIAMOND blastp.45 To search AgcT structural homologs, the cd03817 representative sequences (NCBI Conserved Domain Database) were used to construct a profile hidden Markov model. Subsequently, HMMER was employed to search for matches among gut microbial proteins, and these hits were further validated using Reverse Position-Specific BLAST (RPS-BLAST 46,47). 45,46Multiple sequence alignment of resulting homologs of AgcT was conducted by MAFFT,48 and trimmed by TrimAl49. The tree was generated by IQ-TREE 250 and visualized by ggtree51.

Metagenomic data of the TEDDY study were acquired from the dbGaP database. Human DNA contamination was removed using the Bowtie2 software and assessed species-level taxonomy profiles with MetaPhlAn 4 52. For the DIABIMMUNE and other early life stage metagenome data, species-level profiles were sourced from the curatedMetagenomicData v3.6.2.34 For identified gut bacterial species, reference genomes were downloaded using the NCBI Datasets tool. To explore bacterial genes homologous to AgcT, and BgsB, we conducted searches within these reference genomes using the DIAMOND BLASTp tool,45 setting the cutoff for percent identity > 40%. We analyzed and visualized the contributions of various species using local polynomial regression fitting via the ggplot2 package in R.

Structure superimposition.

AlphaFold structures were obtained from UniProt database53,54. The superimposition of AlphaFold structures of AgcT homologs were performed using ChimeraX matchmaker55, with cross-validation performed by TM-align56 and Foldseek.57

RNA sequencing

Total RNA of B. fragilis NCTC 9343 and Δ3069 were extracted using Trizol (ThermoFisher Scientific #15596026), and genomic DNA was removed by TURBO DNase (AM2238) treatment. The absence of gDNA contamination was confirmed by PCR of 16S rRNA DNA region. mRNA was enriched using a rRNA Depletion Kit (Bacteria) (NEB E7850). The sequencing library was constructed with Ultra™ II Directional RNA Library Prep with Sample Purification Beads (NEB E7765) and sequenced on a sequencing machine. Paired-end reads were trimmed with Trimmomatic, and trimmed reads were aligned to reference genome with Bowtie2.58 Read count matrix was obtained with featureCounts (v2.0.0).59 Gene counts were processed with DESeq2 (v1.32.0)60 to identify differentially expressed genes.

Bacterial genetic manipulation for mutagenesis and heterologous expression

Bacteroides mutant generation and heterologous expression.

B. fragilis NCTC 9343 mutants were constructed with a counterselection vector pLGB13, a gift from Laurie Comstock (Addgene 126618).61 Heterologous expression of agcT in P. vulgatus and B. thetaiotaomicron was conducted with a chromosome-integrated and inducible vector pNBU2 erm-TetR-P1T_DP-GH023, a gift from Andrew Goodman (Addgene 90324).62 Recombinant vectors were generated using HiFi DNA Assembly Cloning Kit (NEB E5520). During cloning procedures, Bacteroides were grown in brain heart infusion broth (3.7% brain heart infusion powder supplemented with five mg/L hemin and 2.5 mg/L vitamin K1) or brain heart infusion agar plates. Deletion or integration of the targeted locus was confirmed by PCR (Supplementary Table 2).

Enterococcus mutagenesis.

Enterococcus bgsB mutants were generated using the counterselection vector pLT06(Supplementary Table 2).63 Recombinant vectors were constructed using HiFi DNA Assembly Cloning Kit (NEB E5520) and transformed by electroporation. Single crossover colonies were selected on Todd-Hewitt broth-containing chloramphenicol plate, and double crossover colonies were selected on Todd-Hewitt broth-containing p-chloro-phenylalanine plate.

In vitro, APC-free NKT activity assay

24.7 NKT hybridoma cells64 were cultured in RPMI 1640 containing 10% FBS, sodium pyruvate, β-mercaptoethanol, NEAA at 37°C with 5% CO2.64 Biotinylated murine CD1d monomers (NIH tetramer facility) were mixed with synthetic ligands or bacterial lipids in 50 mM pH 6.2 citrate buffer 0.25% CHAPS. For antagonism test, B. fragilis lipids extracts in combination with other bacterial lipid extracts to CD1d molecules. After overnight incubation at 37°C, ligand-loaded CD1d (0.25 ug per well) was bound onto streptavidin-coated plates (R&D Systems #CP003). Plates were washed with PBS three times and then 24.7 NKT cells were added (4 × 104 - 105 per well). After overnight incubation, IL-2 of supernatants were analyzed by ELISA (BioLegend #431004).

CD1d enrichment assay

Biotinylated murine CD1d monomers (NIH tetramer facility) were mixed with SB2217 and BbGL-2 in 50 mM pH 6.2 citrate buffer 0.25% CHAPS. After overnight incubation at 37°C, ligand-loaded CD1d (0.25 ug per well) was bound onto streptavidin-coated plates (R&D Systems #CP003). Plates were washed with PBS three times and CD1d-bound lipids were extracted by methanol. Lipids were analyzed by UPLC-MS condition as described in lipidomic analysis.

Colonic NKT cell analysis

Colonic lamina propria lymphocyte isolation.

Conventional (SPF), GF, and monocolonized mice were euthanized. The large intestines were collected, and fat tissue removed. The intestine was opened longitudinally, and after fecal content was removed, cut into 1-inch pieces, and shaken in HBSS containing 2 mM EDTA for 50 min at 37°C. After the removal of epithelial cells, the intestines were washed in HBSS and incubated with RPMI 1640 containing 10% FBS, collagenase type VIII (1 mg/mL), and DNase I (0.1 mg/mL) (Sigma-Aldrich) for 45 min at 37°C under constant shaking. The digested tissues were mixed with FACS buffer (PBS with 2% FBS and 1 mM EDTA), filtered twice through strainers (mesh size, 70 and 40 μm), and used for flow cytometry.

FACS analysis.

For staining with the indicated dilution, APC-labeled mouse CD1d tetramer—unloaded or loaded with PBS-57 (1:500; NIH Tetramer Core Facility)—as well as anti-mouse CD3–FITC (1:400), TCRβ–PE (1:400), CD45–PerCP–Cy5.5 (1:200; Biolegend) and cell viability dye (Fixable Viability Dye eFluor™ 780, 1:1000; ThermoFisher) were used. Individual samples were stained for 20 min at 4°C and washed with cold FACS buffer. FACS analysis was performed with a BD FACSCanto system (BD Biosciences), pre-gated with forward and side scatters, a singlet population, and viable cells. The frequencies of CD3+/CD1d tetramer+ cells from the gated total CD45+ population were enumerated using FlowJo V10 software (BD Biosciences).

Splenic dendritic cell analysis

Splenic lymphocyte isolation.

Spleens were collected from mice and cut into 5 pieces. Tissues were digested in RPMI 1640 containing 10% FBS, collagenase type IV (1 mg/mL), and DNase I (0.1 mg/mL) (Sigma-Aldrich) for 30 min at 37°C with constant shaking. After digestion, red blood cells were lysed using RBC lysis buffer (BioLegend #420301) for 3 min. The resulting cell suspensions were mixed with FACS buffer, passed through strainers (mesh size, 70 μm), and used for flow cytometry analysis.

FACS analysis.

Cells were stained for 20 minutes at 4°C with the following antibodies and reagents at the indicated dilutions: anti-mouse CD3–PE (1:700; Biolegend), MHCII–V500 (1:700; Biolegend), CD11c–BUV395 (1:700; Biolegend), CD40-FITC (1:500), CD80-PE-Cy7 (1:700; Biolegend), CD86-APC (1:700; Biolegend) and cell viability dye (Fixable Viability Dye eFluor™ 780, 1:1200; ThermoFisher) were used. After staining, cells were washed with FACS buffer and analyzed with a BD FACS symphony system (BD Biosciences). Data were enumerated using FlowJo V10 software (BD Biosciences).

Statistical analysis

Data represent two or more independent experiments with similar trend. To assess significance, student’s T-test was carried out for direct comparison of two groups. For more than three groups, adjusted P Value One-way ANOVA Dunnett’s multiple comparisons test was conducted. Significances were shown individually.

Extended Data

Extended data 1.

Extended data 1

Open in a new tab

Extended Data 1. Characterization of screened gene targets and generation of isogenic knockout and transformant strains.

a) Loss-of-function screening identified a transposon insertion mutant (Tn8A7) with depleted BfaGC production. The relative abundance was quantified by LC-MS/MS based on peak area.

b) BfaGC production was abrogasted in the B. fragilis Δ3069 mutant. The abundance of BfaGC (C34:0) was quantified by LC-MS/MS based on peak area.

c) RNA-seq volcano plot comparing B. fragilis Δ3069 to wild type revealed multiple genes downregulated in Δ3069 mutant.

d) The expression of agcT is reduced in B. fragilis Δ3069. Transcript levels of agcT, normalized to leuB, were measured by qRT-PCR. Significance was determined by an unpaired two-tailed t-test.

e) Gain-of-function screening identified P. vulgatus transformant producing aGCs. The relative abundance of aGC/Cer was quantified by LC-MS/MS based on peak area.

f) agcT-transformed P. vulgatus produced molecules identical to BfaGC. XICs of aGCs from B. fragilis, P. vulgatus wild type and P. vulgatus expressing agcT were displayed. Chromosome-integrated expressing system pNBU2-tetR was used to expressing agcT in P. vulgatus.

g) MS/MS spectra mirror plot demonstrated a match between aGCs from P. vulgatus expressing agcT and those from B. fragilis.

h) B. thetaiotaomicron and P. vulgatus heterologously expressing agcT produced aGCs. Chromosome-integrated expressing system pNBU2-tetR was used to expressing agcT in B. thetaiotaomicron and P. vulgatus.

i) Relative abundance of KDS, ketoCer, Cer, and BfaGC in B. fragilis wild type, Δspt, ΔcerS, ΔcerR, and ΔagcT were shown. The abundance was quantified by LC-MS/MS based on peak area.

Extended data 2.

Extended data 2

Open in a new tab

Extended Data 2. Tandem mass spectra analysis determined sphingolipid intermediate structures synthesized by B. fragilis. To assign the structures unambiguously, we chose chain length variants of major sphingolipid species, whose synthetic version was available.

a) MS/MS spectra of B. fragilis KDS (C18:0) matched those of the synthetic standard.

b) MS/MS spectra of synthetic C35:1 ceramide (m/z 568.53) was displayed.

c) The parent ion with an m/z value of 570.53 from B. fragilis was structurally assigned as C35:0 dihydroceramide.

d) The parent ion of m/z 568.53 from B. fragilis ΔcerR was assigned as C35:0 ketoCer. Comparison with isobaric C35:1 ceramide (Panel b) exhibits a distinct fragmentation pattern, as shown by the presence and absence of m/z 264 (300-2H2O).

e-f) MS/MS spectra of C34:0 ketoCer and dhCer from B. fragilis were shown.

Extended data 3.

Extended data 3

Open in a new tab

Extended Data 3. In vitro and in vivo assays for NKT cell modulation by BfaGCs.

a) Schematic process illustrates the in vitro antigen presentation assay. Bacterial lipid extracts were loaded onto biotinylated CD1d molecules, which were subsequently immobilized on streptavidin-coated plates. After washing, NKT cell hybridomas (24.7) were added and incubated overnight. IL-2 secretion was measured by ELISA to assess NKT cell activation.

b) Representative gating strategies for colonic NKT cell analysis by flow cytometry are shown. NKT cell refers to the CD1d tetramer+ cells within the CD3+CD45+TCRβ+ T cell population.

c) Representative flow cytometry plots of unloaded CD1d tetramer controls are shown.

Extended data 4.

Extended data 4

Open in a new tab

Extended Data 4. B. fragilis and B. salyersiae produce aGCs and induce NKT cell activation.

a) A MS/MS mirror plot of B. fragilis and B. salyersiae C34:0 aGC species showed an exact match.

b) All B. fragilis and B. salyersiae strains, including type strains and clinical isolates, synthesize aGC, in contrast, no tested B. nordi, P. gordornii, and P. copri strains produce aGCs.

c) Only aGC-producing bacterial lipid extract can elicit NKT cell activation. NKT cell hybridomas (24.7) were incubated with CD1d-lipid complex and IL-2 secretion was measured by ELISA. Each dot indicates a biologically independent replicate. Bars and error bars depict the mean ± s.e.m. Data are representative of three independent experiments showing consistent trends. Statistical analysis was performed using one-way ANOVA.

d) Symbiotic Bacteroidales species did not produce aGlcACers, whereas Sphingomonas species served as positive controls and produced aGlcACers.

Extended data 5.

Extended data 5

Open in a new tab

Extended Data 5. Homologs of AgcT share conserved structures.

a) The histogram displays the distribution of e-values from the search for proteins containing cd03817 domain. The red dashed vertical line indicates the cutoff threshold used for identifying cd03817 family hits by HMMER search.

b) An AgcT homolog from B. salyersiae and a cd03817 family protein from E. faecalis belong to the same protein family and have comparable structures.

Extended data 6.

Extended data 6

Open in a new tab

Extended Data 6. Targeted metabolomic analysis confirms aGDG production by cd03817-encoding bacteria.

a) Mirror plot of MS/MS spectra of E. faecalis aGDGs and a commercially available standard (Avanti #840522p) showed a match.

b) The retention time of aGDGs of E. faecalis matches that of standard molecule (Avanti #840522p).

c) aGDGs from cd03817-domain-containing gut bacterial species exhibit chain-length variation. XICs of aGDGs from Streptococcus mitis, Lacticaseibacillus rhamnosus, and E. faecalis were shown.

d) Various Enterococcus strains produce aGDGs. XICs of C34:1 aGDGs were displayed.

e) E. faecalis WT and bgsB mutant exhibit comparable colonization abilities in C57BL/6 mice.

f) aGDG levels in stool samples from C57BL/6 SPF, conventionalized, and GF mice were shown. “Conventionalized” indicates GF mice conventionalized via cohousing with SPF mice.

Extended data 7.

Extended data 7

Open in a new tab

Extended Data 7. Metagenomic profiles of agcT and bgsB in multiple cohorts.

a) Skimming process to search species with homologs of AgcT or BgsB was shown.

b) Histograms show sequence identity distributions from the search for homologs of AgcT or BgsB, with vertical dashed lines indicating identity cutoffs.

c) In the DIABIMMUNE cohort, B. fragilis consistently dominates agcT abundance, while bgsB abundance shows dynamic changes and is contributed by a taxonomically diverse set of species. Pie charts show the proportion of species abundance in the dataset.

d-e) Across multiple infant and adult cohorts, agcT abundance is consistently driven by B. fragilis, whereas bgsB is distributed among a broader array of species.

Extended data 8.

Extended data 8

Open in a new tab

Extended Data 8. aGDGs species inhibit the effects of BfaGCs on NKT cells.

a) Bacterial lipid extracts from various Lactobacillales bgsB-encoding species exhibit dose-dependent antagonism to SB2217-induced NKT cell activation.

b) aGDGs competed with BfaGCs for CD1d binding in a dose-dependent manner. After loading the lipids onto CD1d, CD1d-lipids complexes were purified, and the bound lipids were extracted and analyzed by LC-MS.

c) Representative flow cytometry plots show the gating strategy used for analysis of splenic dendritic cells.

Extended data 9.

Extended data 9

Open in a new tab

Extended Data 9. Graphical overview of forward genetics-based metabolomic screening to characterize AgcT and explore related metabolites in the gut microbiome.

Unannotated metabolites were linked to their biosynthetic gene agcT through forward genetics-based metabolomic screening. Protein structural homologs were identified in gut symbionts, revealing structurally related glycolipids with contrasting immunological activities. Species-level contributions of agcT and its homologs bgsB were profiled in human gut metagenomes.

Supplementary Material

Supp Table 1-2
Supp Fig Legends

Acknowledgements

This work was supported by the National Institutes of Health (K01-DK102771, R01-AT010268, R01-AI165987: S.F.O.) and the National Research Foundation of Korea (2021R1A6A3A14044113: J-S.Y.; RS-2024-00411992 :B.G.; 2021R1A6A3A14039202: D-J.J.; RS-2024-00348702 :K.H.; RS-2023-00217123: J.I.S.; 2014R1A3A2030423 and 2012M3A9C4048780: S.B.P.). CD1d tetramers were provided by the NIH Tetramer Core Facility (contract number 75N93020D00005). We thank Tsering Yanostang for technical support, and helpful discussion and comments from Dr. Neil Surana (Duke University).

Footnotes

Competing Interests

S.F.O. and D.L.K. filed a patent on the functions of BfaGCs and related structures (US patent 10,329,315).

S.F.O., S.B.P. and D.L.K. filed a patent on the functions of BfaGCs and related structures (US patent application 17/427,756).

The remaining authors declare no competing interests.

Data Availability

Lipidomics and RNA-sequencing datasets supporting this study are available on Harvard Dataverse (https://doi.org/10.7910/DVN/DYJJQJ). TEDDY microbiome data are available from dbGaP (phs001443.v1.p1) under dbGaP-controlled access. Curated human microbiome data are accessible via curatedMetagenomicData (doi:10.18129/B9.bioc.curatedMetagenomicData). The RNA-seq data are available in the NCBI database under BioProject PRJNA1309292.

References

  • 1.Dorrestein PC, Mazmanian SK & Knight R Finding the Missing Links among Metabolites, Microbes, and the Host. Immunity 40, 824–832 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.An D et al. Sphingolipids from a Symbiotic Microbe Regulate Homeostasis of Host Intestinal Natural Killer T Cells. Cell 156, 123–133 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Oh SF et al. Host immunomodulatory lipids created by symbionts from dietary amino acids. Nature 2021 600:7888 600, 302–307 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Theilacker C et al. Deletion of the glycosyltransferase bgsB of Enterococcus faecalis leads to a complete loss of glycolipids from the cell membrane and to impaired biofilm formation. BMC Microbiol 11, 67 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Skelly AN, Sato Y, Kearney S & Honda K Mining the microbiota for microbial and metabolite-based immunotherapies. Nature Reviews Immunology vol. 19 305–323 Preprint at 10.1038/s41577-019-0144-5 (2019). [DOI] [PubMed] [Google Scholar]
  • 6.Brutkiewicz RR CD1d Ligands: The Good, the Bad, and the Ugly. The Journal of Immunology 177, 769–775 (2006). [DOI] [PubMed] [Google Scholar]
  • 7.Goodman AL et al. Identifying genetic determinants needed to establish a human gut symbiont in its habitat. Cell Host Microbe 6, 279–89 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Comstock LE et al. Analysis of a capsular polysaccharide biosynthesis locus of Bacteroides fragilis. Infect Immun 67, 3525–32 (1999). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Coyne MJ, Chatzidaki-Livanis M, Paoletti LC & Comstock LE Role of glycan synthesis in colonization of the mammalian gut by the bacterial symbiont Bacteroides fragilis. Proc Natl Acad Sci U S A 105, 13099–13104 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Moremen KW & Haltiwanger RS Emerging structural insights into glycosyltransferase-mediated synthesis of glycans. Nature Chemical Biology 2019 15:9 15, 853–864 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Okino N et al. Two bacterial glycosphingolipid synthases responsible for the synthesis of glucuronosylceramide and α-galactosylceramide. Journal of Biological Chemistry 295, 10709–10725 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Stankeviciute G et al. Convergent evolution of bacterial ceramide synthesis. Nature Chemical Biology 2021 18:3 18, 305–312 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Naidenko OV et al. Binding and antigen presentation of ceramide-containing glycolipids by soluble mouse and human CD1d molecules. Journal of Experimental Medicine 190, (1999). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Brennan PJ et al. Invariant natural killer T cells recognize lipid self antigen induced by microbial danger signals. Nat Immunol 12, (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Zeissig S, Olszak T, Melum E & Blumberg RS Analyzing antigen recognition by natural killer T cells. Methods in Molecular Biology 960, (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Olszak T et al. Microbial Exposure During Early Life Has Persistent Effects on Natural Killer T Cell Function. Science (1979) 336, 489–493 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Morita M et al. Structure-Activity Relationship of .alpha.-Galactosylceramides against B16-Bearing Mice. J Med Chem 38, 2176–2187 (1995). [DOI] [PubMed] [Google Scholar]
  • 18.Almeida A et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat Biotechnol 39, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Kawahara K et al. Chemical structure of glycosphingolipids isolated from Sphingomonas paucimobilis. FEBS Lett 292, (1991). [DOI] [PubMed] [Google Scholar]
  • 20.Naka T et al. A novel sphingoglycolipid containing galacturonic acid and 2-hydroxy fatty acid in cellular lipids of Sphingomonas yanoikuyae. J Bacteriol 182, (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Kinjo Y et al. Recognition of bacterial glycosphingolipids by natural killer T cells. Nature 434, 520–525 (2005). [DOI] [PubMed] [Google Scholar]
  • 22.Mayassi T, Barreiro LB, Rossjohn J & Jabri B A multilayered immune system through the lens of unconventional T cells. Nature 595, 501 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Oh SF, Jung D-J & Choi E Gut Microbiota-Derived Unconventional T Cell Ligands: Contribution to Host Immune Modulation. Immunohorizons 6, 476–487 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Pasolli E et al. Accessible, curated metagenomic data through ExperimentHub. Nature Methods vol. 14 Preprint at 10.1038/nmeth.4468 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Olszak T et al. Microbial exposure during early life has persistent effects on natural killer T cell function. Science (1979) 336, (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.An D et al. Sphingolipids from a symbiotic microbe regulate homeostasis of host intestinal natural killer T cells. Cell 156, (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Buchfink B, Reuter K & Drost HG Sensitive protein alignments at tree-of-life scale using DIAMOND. Nat Methods 18, (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Eddy SR Accelerated profile HMM searches. PLoS Comput Biol 7, (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Minh BQ et al. IQ-TREE 2: New Models and Efficient Methods for Phylogenetic Inference in the Genomic Era. Mol Biol Evol 37, (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Hölzl G & Dörmann P Structure and function of glycoglycerolipids in plants and bacteria. Prog Lipid Res 46, 225–243 (2007). [DOI] [PubMed] [Google Scholar]
  • 31.Stewart CJ et al. Temporal development of the gut microbiome in early childhood from the TEDDY study. Nature vol. 562 583–588 Preprint at 10.1038/s41586-018-0617-x (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Vatanen T et al. Variation in Microbiome LPS Immunogenicity Contributes to Autoimmunity in Humans. Cell 165, 842–853 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Vatanen T et al. The human gut microbiome in early-onset type 1 diabetes from the TEDDY study. Nature 562, 589–594 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Pasolli E et al. Accessible, curated metagenomic data through ExperimentHub. Nature Methods 2017 14:11 14, 1023–1024 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Bäckhed F et al. Dynamics and stabilization of the human gut microbiome during the first year of life. Cell Host Microbe 17, (2015). [DOI] [PubMed] [Google Scholar]
  • 36.Brooks B et al. Strain-resolved analysis of hospital rooms and infants reveals overlap between the human and room microbiome. Nat Commun 8, (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Shao Y et al. Stunted microbiota and opportunistic pathogen colonization in caesarean-section birth. Nature 574, (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Yassour M et al. Strain-Level Analysis of Mother-to-Child Bacterial Transmission during the First Few Months of Life. Cell Host Microbe 24, (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Turnbaugh PJ et al. The Human Microbiome Project. Nature 449, 804–810 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Lloyd-Price J et al. Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature 569, 655–662 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Kinjo Y et al. Natural killer T cells recognize diacylglycerol antigens from pathogenic bacteria. Nat Immunol 7, 978–986 (2006). [DOI] [PubMed] [Google Scholar]
  • 42.Oh SF, Jung D-J & Choi E Gut Microbiota-Derived Unconventional T Cell Ligands: Contribution to Host Immune Modulation. Immunohorizons 6, (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Yoo JS & Oh SF Unconventional immune cells in the gut mucosal barrier: regulation by symbiotic microbiota. Experimental and Molecular Medicine vol. 55 Preprint at 10.1038/s12276-023-01088-9 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Huang S et al. CD1 lipidomes reveal lipid-binding motifs and size-based antigen-display mechanisms. Cell 186, 4583–4596.e13 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Baldridge MT et al. Commensal microbes and interferon-λ determine persistence of enteric murine norovirus infection. Science 347, 266–9 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Eddy SR Accelerated Profile HMM Searches. PLoS Comput Biol 7, e1002195 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Lu S et al. CDD/SPARCLE: the conserved domain database in 2020. Nucleic Acids Res 48, D265–D268 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Katoh K & Standley DM MAFFT Multiple Sequence Alignment Software Version 7: Improvements in Performance and Usability. Mol Biol Evol 30, 772–780 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Capella-Gutiérrez S, Silla-Martínez JM & Gabaldón T trimAl: a tool for automated alignment trimming in large-scale phylogenetic analyses. Bioinformatics 25, 1972–1973 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Minh BQ et al. IQ-TREE 2: New Models and Efficient Methods for Phylogenetic Inference in the Genomic Era. Mol Biol Evol 37, 1530–1534 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Yu G, Smith DK, Zhu H, Guan Y & Lam TT ggtree : an r package for visualization and annotation of phylogenetic trees with their covariates and other associated data. Methods Ecol Evol 8, 28–36 (2017). [Google Scholar]
  • 52.Blanco-Míguez A et al. Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4. Nature Biotechnology 2023 41:11 41, 1633–1644 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Jumper J et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021 596:7873 596, 583–589 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Varadi M et al. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res 50, D439–D444 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Pettersen EF et al. UCSF ChimeraX : Structure visualization for researchers, educators, and developers. Protein Science 30, 70–82 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Zhang Y & Skolnick J TM-align: a protein structure alignment algorithm based on the TM-score. Nucleic Acids Res 33, 2302–2309 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.van Kempen M et al. Fast and accurate protein structure search with Foldseek. Nature Biotechnology 2023 1–4 (2023) doi: 10.1038/s41587-023-01773-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Langmead B & Salzberg SL Fast gapped-read alignment with Bowtie 2. Nat Methods 9, 357–359 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Liao Y, Smyth GK & Shi W featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014). [DOI] [PubMed] [Google Scholar]
  • 60.Love MI, Huber W & Anders S Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.García-Bayona L & Comstock LE Streamlined genetic manipulation of diverse bacteroides and parabacteroides isolates from the human gut microbiota. mBio 10, (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Lim B, Zimmermann M, Barry NA & Goodman AL Engineered Regulatory Systems Modulate Gene Expression of Human Commensals in the Gut. Cell 169, 547–558.e15 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63.Thurlow LR, Thomas VC & Hancock LE Capsular polysaccharide production in Enterococcus faecalis and contribution of CpsF to capsule serospecificity. J Bacteriol 191, 6203–6210 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64.Behar SM, Podrebarac TA, Roy CJ, Wang CR & Brenner MB Diverse TCRs Recognize Murine CD1. The Journal of Immunology 162, 161–167 (1999). [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Supp Table 1-2
NIHMS2124289-supplement-Supp_Table_1-2.pdf (133.1KB, pdf)
Supp Fig Legends
NIHMS2124289-supplement-Supp_Fig_Legends.docx (22.7KB, docx)

Data Availability Statement

Lipidomics and RNA-sequencing datasets supporting this study are available on Harvard Dataverse (https://doi.org/10.7910/DVN/DYJJQJ). TEDDY microbiome data are available from dbGaP (phs001443.v1.p1) under dbGaP-controlled access. Curated human microbiome data are accessible via curatedMetagenomicData (doi:10.18129/B9.bioc.curatedMetagenomicData). The RNA-seq data are available in the NCBI database under BioProject PRJNA1309292.

RESOURCES