Gut microbiota-derived GlcNAc-MurNAc is a TLR4 agonist that protects the host gut

Abstract

Gut microbiota-derived peptidoglycan fragments (PGNs) are key signaling molecules that regulate multiple aspects of the host’s health. Yet the exact structures of natural PGNs in hosts have not been fully elucidated. Herein, we developed an LC-HRMS/MS analytical platform for global quantification and profiling of natural PGN subtypes in host gut and sera, unexpectedly revealing the abundance of PGN-derived saccharide moieties that do not resemble canonical ligands of mammalian NOD1/2 receptors. Focusing on the disaccharide GlcNAc-MurNAc (GM), which does not activate NOD1/2 yet still exhibits immunostimulatory effects in host immune cells, we established GM as a mild TLR4 agonist, illustrating an alternate PGN sensing mechanism other than NOD signaling. Importantly, the administration of GM mitigates colonic inflammation in the DSS-induced colitis model in mice via a TLR4-dependent manner, highlighting the in vivo significance of gut microbiota-derived PGN saccharides in maintaining host intestinal homeostasis.

Introduction

The gut microbiota has a profound impact on host health. The resident gut bacteria bestow a rich source of microbial metabolites and molecules, known as microbe-associated molecular patterns (MAMPs), which are recognized by the host’s innate immune pattern recognition receptors (PRRs)^1,2. Proper interactions between commensal bacterial MAMPs and host PRRs are crucial for maintaining intestinal homeostasis, including the development and maturation of isolated lymphoid follicles³, reinforcement of gut barrier functions^4,5, and protection against colonic injury and pathogenic infections^6,7. Conversely, gut microbiota dysbiosis can inflict dysregulated immune responses in hosts, leading to chronic inflammatory diseases, such as inflammatory bowel diseases (IBD), rheumatoid arthritis, and asthma^8,9,10,11. Understanding how the host recognizes and responds to gut microbial ligands is a key step toward developing immuno-therapeutics to target the gut microbiota-host interface.

Peptidoglycan, the major bacterial cell wall component, represents a well-known MAMP that stimulates the host’s innate immune system^12,13. A mesh-like layer that surrounds the bacterial cytoplasmic membrane, peptidoglycan is composed of repeating N-acetylglucosamine-β−1,4-N-acetylmuramic acid (GlcNAc-MurNAc) disaccharide with a stem peptide connected to the lactoyl group of each MurNAc unit. Adjacent peptidoglycan strands are cross-linked via the appended stem peptides¹⁴. Despite a conserved scaffold, peptidoglycan varies across bacterial species in terms of stem peptide composition, type and degree of cross-linking, and glycan modifications, giving rise to complex and heterogeneous peptidoglycan polymeric structures (i.e., peptidoglycome)¹⁵. During bacterial cell wall remodeling and turnover, soluble peptidoglycan fragments (PGNs) are generated and released into the milieu. Some PGN fragments activate mammalian NOD1/2 receptors, leading to the production of proinflammatory cytokines¹⁶. Notably, NOD1 and NOD2 recognize distinct minimal PGN motifs: NOD1 detects the dipeptide D-γ-Glu-mDAP (iE-DAP)^17,18, while NOD2 recognizes the muramyl dipeptide N-acetylmuramyl-L-Ala-D-isoGln (MDP or M-AQ)^19,20. Biological studies of bacterial PGNs have predominantly focused on these canonical NOD1/2 ligands. Given the diversity of the gut microbiota peptidoglycome, these ligands likely cannot embody the repertoire of natural gut microbiota-derived PGNs in hosts, which may exhibit bioactivity via NOD-independent pathways. For instance, GlcNAc derived from peptidoglycan was shown to trigger NLRP3 inflammasome formation by inhibiting the metabolic enzyme hexokinase in primed immune cells without engaging NOD1/2²¹. Moreover, we recently identified an anti-inflammatory 1,6-anhydro-PGN motif from the probiotic Bifidobacterium that does not activate NOD1/2²², suggesting alternate PGN sensing mechanisms in hosts. Despite increasing recognition of gut microbiota-derived PGNs as key effector molecules in hosts, the fundamental questions regarding the structures of natural PGNs, their biological roles, and host responses beyond NOD1/2 signaling are yet to be determined.

Traditionally, the detection of PGNs has largely relied on cell-based reporter assays that selectively respond to canonical NOD1/2 ligands²³. We have developed a monoclonal antibody 2E7 that specifically recognizes MDP, which enabled an indirect competitive enzyme-linked immunosorbent assay (icELISA) for PGN detection. This advancement has proved the presence of gut microbiota-derived PGNs in host systemic circulation; however, the exact structures of natural PGNs remain unknown²⁴. Wheeler et al. recently employed radiolabeled ligands to elucidate the biodistribution of natural PGNs in hosts²⁵. Their research revealed that PGN-derived saccharide moieties preferentially accumulate in the host’s brain and adipose tissue, suggesting the physiological relevance of the non-canonical PGN motifs. Thus, it is imperative to uncover the natural forms of gut microbiota-derived PGNs to better understand their biological roles in hosts.

In this study, we developed an LC-HRMS/MS-based analytical platform to quantify and profile gut microbiota-derived PGNs in hosts. Contrary to the conventional notion, we did not identify canonical MDP (i.e., M-AQ) in the host gut; instead, we found that the PGN-derived saccharide moieties dominated the gut PGN pool. Unexpectedly, the PGN disaccharide motif, GlcNAc-MurNAc (GM), which does not activate NOD1/2, still elicits mild immuno-stimulatory effects. Supported by cellular and biochemical evidence, we established that GM is a toll-like receptor 4 (TLR4) agonist, highlighting PGN-TLR4 interactions in microbiota-host crosstalk. Importantly, we showed that GM administration ameliorates DSS-induced colitis in mice via TLR4-dependent mechanisms. Our work underscores the biological significance of gut commensal-released PGN saccharides in safeguarding host intestinal homeostasis.

Results

Muramic acid (MurN) analysis affords global quantification of soluble PGNs in hosts

Acid hydrolysis of PGNs readily releases muramic acid (MurN), an invariant bacterial constituent that is absent in mammalian metabolites, rendering it an ideal unit for PGN quantification. Upon proper sample cleanup, we detected MurN (m/z: 252.1078) in both sera and feces of mice and humans (Fig. 1a–c), supporting the ubiquitous existence of gut microbiota-derived PGNs in hosts. Quantification of MurN was achieved based on the abundance of its featured daughter ions (m/z: 126.0549 and 144.0655) using the parallel reaction monitoring (PRM) mode of LC-HRMS/MS (Fig. 1c). Calibration curves were established with serial dilutions of N-acetyl-MurNAc in water and spiked pooled sera or feces (Fig. S1). Next, we quantified endogenous MurN in different biological samples with adjustments for the respective matrix effects. Ceca and feces from specific pathogen-free (SPF) mice contain a considerable amount of MurN, while feces from germ-free (GF) mice show virtually undetectable levels (Fig. 1d). Healthy human stools manifest higher MurN abundance but with considerable individual variability. Healthy human sera also display a wide range of MurN concentrations, averaging approximately 200 nM, which is comparable to the amount of MurN in two brands of fetal bovine serum (FBS). Notably, previous icELISA studies also revealed similar PGN concentrations in human sera²⁴, supporting the results from our quantification workflow. Together, our LC-HRMS/MS-based MurN analysis provides a quantitative perspective of gut microbiota-derived soluble PGNs in hosts.

Fig. 1: HPLC-MS/MS analysis of muramic acid (MurN) enables global quantification of gut microbiota-derived peptidoglycan fragments (PGNs) in biological samples.

a Scheme of sample preparation for MurN analysis ([M + H]⁺: 252.1078). b–c Extracted ion chromatograms (EICs) and MS/MS spectra of MurN detected in host samples compared to the MurN standard. d MurN concentrations in biological samples, including SPF mice ceca (n = 4) and feces (n = 25), GF mice feces (n = 8), human stools (n = 12) and sera (n = 11), and commercial fetal bovine serum (FBS) from two different brands. Data are presented as mean values ± s.e.m. with indicated P values determined using two-tailed unpaired t test. All experiments were repeated twice with similar results. Source data are provided as a Source Data file.

Saccharide moieties represent predominant PGN subtypes in the host gut

For structural profiling of gut microbiota-derived PGNs, we directly subjected the cleaned-up sera or stool samples to an untargeted analysis using the full-scan (FS) and data-dependent analysis (FSddA) mode of LC-MS/MS. Facilitated by our recently developed in silico PGN_MS2 spectral library²², we extensively evaluated the experimental LC-MS/MS data, where hits were prioritized according to matching criteria including MS1 accuracy, isotopic patterns, and MS/MS fragmentation patterns (Fig. 2a). To validate potential PGN hits, we chemically synthesized a panel of PGNs as authentic standards (Fig. 2b–d). With this workflow, we found that the diversity and abundance of PGNs were significantly higher in the human stool than in sera, which is consistent with the rapid excretion and overall biodistribution of gut microbiota-derived PGNs in hosts²⁶. Interestingly, the major PGN subtypes in human stools and mouse feces are highly similar (Fig. 3), suggesting that closely related gut bacteria may be the primary contributors of soluble PGNs or that both mammalian hosts encode similar enzymes for processing PGNs.

Fig. 2: Profiling of gut microbiota-derived PGNs reveals abundant saccharide moieties in the host gut.

a Scheme of HPLC-HRMS/MS workflow for PGN identification. b-c EICs of major PGN subtypes in the host feces: ahM, M, GM, and M-AE (b) and their chemical structures (c). d Chemical synthetic route of disaccharide GM.

Fig. 3: Profiling unravels the composition of gut PGN reservoirs.

a Pie charts showing the relative abundance of detectable PGN subtypes in mouse feces (n = 6) and human stools (n = 12). Data are presented as mean values. b Profiling of gut microbiota-derived PGNs in feces and ceca of individual mice reveals distinct PGN compositions along the GI tract (n = 5). Source data are provided as a Source Data file.

The soluble PGNs in the host gut can be classified into two groups: saccharide moieties that lack any stem peptide, and typical muropeptides containing both glycan backbone and stem peptide (Figs. 2b, c, 3). The PGN-derived saccharide moieties, including MurNAc (M), 1,6-anhydro-MurNAc (ahM), GlcNAc-MurNAc (GM), and GlcNAc-ahMurNAc (GahM) are present in every sample of the cohort. In contrast, variable forms of muropeptides are detected across individuals, yet they mostly belong to the iE-DAP-containing PGNs. In addition, we revealed the abundance of muramyl-dipeptides GM-AE and M-AE in mice ceca, which are close analogs of the widely used MDP (i.e., M-AQ) but differ in the amidation state of the stem peptide (Figs. 3, S2). Furthermore, muropeptides with diacetylated GlcNAc or MurNAc moieties were detected, which represent peptidoglycan modifications that confer lysozyme resistance in many gut bacteria²⁷. LC-MS quantification of these major PGN subtypes showed that the saccharide moieties account for nearly 90% of all soluble PGNs in host feces. The diverse spectrum of natural PGNs in the host gut underscores the need to explore PGN bioactivities beyond the well-studied model ligands.

Profiling serum PGNs proved technically challenging due to low signal-to-noise ratios in LC-MS/MS analysis. Nonetheless, in two different cohorts of pooled sera from healthy human individuals and in commercial FBS, we identified PGN-derived monosaccharides M and ahM, but not any muropeptides (Fig. S3). The presence of M and ahM in serum coincides with their prominent abundance in the host gut (Fig. 3), while the striking difference in the overall quantity and diversity of soluble PGNs between the gut and serum suggests that the systemic dissemination of gut microbiota-derived PGNs is a selective process²⁵. In a Caco-2 monolayer-coated transwell system, we showed that PGN saccharides M and GM translocated significantly faster than the larger muropeptide M-AEKAA (Fig. S4A, B). Furthermore, we reasoned that the absence of intact muropeptides in sera may be attributed to peptidoglycan-processing enzymes encoded by the host^13,28,29,30. For instance, mammalian PGRP2 is a serum amidase that specifically cleaves the N-acetyl-muramyl-L-Ala bond of soluble PGNs³¹, whose activity may deplete muropeptides and contribute to the PGN-derived monosaccharides in host sera. Consistently, we showed that the incubation of 10% FBS with the muropeptide M-AEKAA reliably yields the expected degradation products: M and AEKAA (Fig. S4C). Taken together, our LC-MS/MS profiling workflow reveals that saccharide moieties are the predominant gut PGN subtypes in hosts.

GM is immunologically active via NOD-independent mechanisms

Intrigued by the abundance of gut microbiota-derived PGN saccharides in hosts at steady-state, we questioned if they exhibit any potential biological effects. To obtain sufficient PGNs of high purity for biological assays, we chemically synthesized a panel of PGNs of interest, including M-AE, GM, and M (Fig. 2c, d). As mammalian NOD1/2 represent canonical PGN sensors, we first assessed these PGNs in HEK-Blue^TM NOD reporter cells (Fig. 4a). As expected, the natural muropeptide M-AE elicited potent NOD2 activation similar to the model ligand MDP (i.e., M-AQ); yet the saccharide moieties, such as GM and M, are unable to activate NOD1/2, which is unsurprising given their lack of the required structural motifs for NOD1/2 recognition²⁰. At this juncture, we sought to explore the potential effects of PGN saccharides in immune cells that express diverse PRRs, which are more physiologically relevant. Unexpectedly, the disaccharide GM exhibits dose-dependent immuno-stimulatory activity in innate immune cells, including murine macrophage RAW264.7, primary murine bone-marrow-derived macrophages (BMDMs), and human monocytes THP-1, as evidenced by upregulated expression and production of proinflammatory cytokines such as TNF-α and IL-6 (Fig. 4b–d). In comparison, the monosaccharide M appears less potent (Fig. 4c, d), suggesting that the disaccharide structure may be crucial for immunoactivity. Furthermore, GM upregulates CD80/86 surface marker expressions in the type-II conventional dendritic cell (cDC2) population of murine bone marrow-derived dendritic cells (BMDCs) (Figs. 4e and S5). Its ability to promote DC maturation ex vivo suggests its potential role in regulating adaptive immune responses in hosts. Importantly, we confirmed that the synthetic GM is endotoxin-free, as it tested negative in the LAL assay and its activity was unaffected by polymyxin B neutralization (Figs. 4f and S6). In addition, no glycosidase-cleaved monosaccharides (G and M) were detected in GM-treated cells, indicating that the immunoactivity likely originates from the intact disaccharide GM rather than its degradation products (Fig. S7). With these observations, we concluded that the disaccharide GM is a bioactive PGN motif that acts via NOD1/2-independent pathways.

Fig. 4: The disaccharide GM exhibits immuno-stimulatory effects in vitro independent of NOD1/2 signaling.

a SEAP reporter readouts from HEK-Blue^TM NOD1/2 reporter cells stimulated with MDP (200 nM), M-AE (200 nM), GM (20 μM), or M (20 μM) for 18 h. Data are presented as mean values (n = 3). b-g PGNs induce cytokine production in macrophages and monocytes and activate dendritic cells. The respective immune cells were stimulated with the indicated PGNs, followed by cytokine analysis by RT-qPCR (c) or ELISA (d). RAW264.7 cells were treated with GM (4 mM, 2 mM, 1 mM), M (4 mM), or MDP (0.4 mM) for 24 h (n = 4). Murine bone marrow-derived macrophages (BMDMs) and THP-1 cells were treated with GM (8 mM, 4 mM, 2 mM), M (8 mM), or MDP (0.4 mM). n = 3 or 4, respectively. e Mean fluorescent intensity (MFI) of surface CD80/86 in the cDC2s population of murine bone marrow-derived dendritic cells (BMDMs) that were treated with GM (50, 100, and 200 μM) for 18 h. f-g Assays using polymyxin B (PB) confirmed that GM was not contaminated with trace endotoxin. f ELISA analysis of the TNF-α levels in BMDMs stimulated with GM (4 mM), LPS (100 ng/mL), or polymyxin B resin (PB-resin)-neutralized GM (4 mM) or LPS (100 ng/mL) for 24 h. g The TLR4 inhibitor specifically suppresses the GM-induced TNF-α production. The RAW_Dual^TM cells were pre-incubated with the respective TLR inhibitors (10 μM) for 1 h followed by GM stimulation (2 mM) for 24 h. TLR2i, TLR4i, and TLR7/9i refer to CU_CPT22, TAK-242, and AT791, respectively. (h) SEAP reporter readouts from HEK-Blue^TM mTLR4 reporter cells stimulated with GM (1 mM) or LPS (100 ng/mL) titrated with polymyxin B (0, 0.1, 1, 10 kU) for 24 h (n = 4). Data are presented as mean values ± s.e.m. (n = 3 unless otherwise noted) with indicated P values determined using one-way ANOVA followed by Dunnett’s multiple comparison for (c-d, f-g) or two-way ANOVA followed by Tukey’s multiple comparison for (e, h). All experiments were repeated twice with similar results. Source data are provided as a Source Data file.

TLR4 mediates GM-triggered immune responses

Apart from the NOD sensors, TLRs represent another important class of mammalian PRRs that detect various MAMPs, many of which contain glycan motifs³². To assess whether GM is recognized by TLRs in immune cells, we pretreated RAW264.7 cells with a panel of small-molecule TLR inhibitors, then stimulated them with GM and evaluated secreted cytokine levels using ELISA. Remarkably, the addition of TAK-242, a selective TLR4 inhibitor³³, suppressed GM-triggered TNF-α production in RAW264.7 cells, while inhibition of TLR2 or TLR7/9 did not interfere with GM’s bioactivity (Figs. 4g and S8). Our results implicate TLR4, a well-recognized innate immune receptor of bacterial LPS³⁴, in sensing the gut microbiota-derived disaccharide GM. Importantly, we demonstrated that the bioactivity of GM in HEK-Blue^TM-mTLR4 reporter cells is unaffected by polymyxin B titration, which effectively neutralizes LPS/endotoxin (Fig. 4h).

To establish the role of TLR4 in the immunostimulatory effects of GM, we next utilized RAW-Dual^TM wildtype (WT) and Tlr4^-/- reporter cells for assays. Notably, TLR4 activation is known to trigger two distinct signaling transduction pathways, NF-κB- and IRF-dependent pathways^35,36,37, which can be evaluated in parallel via colorimetric and luciferase readouts in the dual reporter cells (Fig. 5a). As expected, GM induces dose-dependent activation of both NF-κB and IRF signaling in WT cells but exhibits minimal activity in Tlr4^-/- cells, which supports the essential role of TLR4 in GM-induced immune responses. Interestingly, the regio-isomer MurNAc-GlcNAc (MG) shows similar TLR4-dependent immune activation (Fig. 5b).

Fig. 5: The immuno-stimulatory activity of GM is TLR4-dependent.

a-b WT and Tlr4^-/- RAW_Dual^TM reporter cells were used to evaluate GM-induced NF-κB and IRF signaling. Cells were treated with respective PGNs at indicated concentrations for 24 h for quanti-blue analysis (top, n = 3) or quanti-luciferase analysis (bottom). LPS (100 ng/mL), MDP (0.4 mM), or Poly(I: C) (0.1 mg/mL) were used as controls. Concentrations of GM, MG, and M tested were indicated in the figures. c-e The RNAseq analysis and validation confirm that the GM-induced immune responses in BMDM require TLR4. c Heatmap plot of RNAseq results showing the mean values of differentially expressed genes (p < 0.01, and logFC > 1) in WT and Tlr4^-/- BMDMs treated with GM (8 mM) or MDP (0.4 mM) for 24 h (n = 3). d-e Validation of the RNAseq results by RT-qPCR (d) and ELISA (e). WT and Tlr4^-/- BMDMs were treated with GM (4 mM) or MDP (0.4 mM) for 24 h. Data are presented as mean values ± s.e.m. (n = 4 unless otherwise noted) with indicated P values determined using one-way ANOVA followed by Dunnett’s multiple comparison for (d-e) or two-way ANOVA followed by Tukey’s multiple comparison for (b). All experiments were repeated twice with similar results. Source data are provided as a Source Data file.

To gain deeper insights into the GM-induced transcriptional effects, we performed whole-genome RNA sequencing (RNAseq) in WT and Tlr4^-/- BMDMs that were stimulated with GM for 24 h. Gene expression levels were analyzed with DESeq2 (Fig. 5c)³⁸. KEGG pathway analysis of the GM-treated WT BMDM identified several upregulated pathways associated with immune response and autoimmune diseases, such as Toll-like receptor signaling, NF-κB signaling pathway, and IBD, which are not enriched in the GM-treated Tlr4^-/- BMDMs (Fig. S9A, B). Specifically, some cytokine and chemokine genes, including Il12a/b, Il6, Il1a/b, Csf3, Ccl5, and Cxcl1, are significantly upregulated in GM-stimulated WT BMDMs. In contrast, GM treatment fails to induce the expression of these genes in Tlr4^-/- BMDMs (Fig. S10). The transcriptional changes of cytokines and chemokines were further validated by RT-qPCR (Figs. 5d, S11). Notably, elevated expression of Tlr4 was also observed in the GM-treated WT BMDMs (Fig. S11). Collectively, our results demonstrate that TLR4 mediates the immunological effects of GM.

Additionally, we also analyzed RNA-seq data from WT and Tlr4^-/- BMDMs treated with MDP, the model NOD2 ligand (Fig. 5c). As expected, MDP-induced pathways---including TNF signaling, NOD-like receptor signaling, and NF-κB signaling---are similarly enriched in WT and Tlr4^-/- BMDMs (Fig. S9C, D). Upon closer analysis of cytokines and chemokines, we observed that while many genes, such as Il12 and Il6, are highly stimulated by MDP in both types of cells, a subset of MDP-induced genes, including Ccl5, Cxcl10, and Tnfa, appear to be TLR4-dependent (Figs. 5d, S11). This discovery suggests that TLR4 may also play a role in modulating MDP-triggered NOD2 signaling. However, while MDP increases expression of Nod2 in WT BMDMs as expected, it does not upregulate Tlr4 (Fig. S11).

Lastly, we also evaluated cytokine production in treated Tlr4^-/- BMDMs by ELISA. The disaccharide GM, like LPS, fails to elicit increased levels of TNF-α and IL-6 in Tlr4^-/- BMDMs; in contrast, other ligands, such as CpG and MDP still trigger comparable proinflammatory cytokine production in Tlr4^-/- BMDMs as in WT BMDMs (Figs. 5e, S12). Of note, the lack of immune-stimulatory response in GM-treated Tlr4^-/- BMDMs aligned with the absence of p65 and p38 phosphorylation, which was in contrast to the mild increase in phosphorylation in GM-treated WT BMDMs (Fig. S13).

GM directly binds to TLR4

To further validate TLR4 as a receptor for GM, we sought biochemical evidence of GM binding to TLR4. First, in an in vitro pulldown assay, the immobilized GM (GM-1-resin) enriched recombinant mTLR4 protein in a dose-dependent manner, whereas the control resin immobilized with GlcNAc (G-1-resin) did not exhibit specific interaction with mTLR4. Pre-incubation of mTLR4 with excess GM as competitors led to a dose-dependent decrease in the amount of mTLR4 bound to GM-1-resin (Figs. 6a, b and S14). To solicit additional evidence for direct GM-mTLR4 interaction, we next performed the surface plasmon resonance (SPR) assay, in which LPS binds to mTLR4 in a dose-dependent manner (with Kd of ~20 ± 10.4 µM) as reported³⁹, while MDP does not interact with it, serving as appropriate controls for the SPR setup. As anticipated, GM binds to mTLR4 under such setup with an estimated Kd of ~383 µM (Figs. 6c and S15). The weak affinity of GM to mTLR4 in vitro is consistent with its mild immunostimulatory effects observed in the cellular assays. Our data confirms GM is a TLR4 ligand.

Fig. 6: GM binds to mTLR4 in vitro.

a-b In vitro enrichment of recombinant mTLR4 protein by immobilized GM. a Scheme of the pulldown assay, with GM- or G-immobilized agarose beads. b Coomassie blue-stained SDS-PAGE analysis of mTLR4 proteins bound or unbound to the resin. Excess GM was added to compete with mTLR4 bound to GM-immobilized beads. c Surface plasmon resonance (SPR) demonstrates direct binding of GM and mTLR4. Representative SPR sensorgrams of mTLR4 binding to MDP (left) and GM (middle) are shown. Corresponding plots and fitted data of steady-state binding for MDP (blue) and GM (black), with the estimated respective K_D values, are shown (right). ND stands for not detected. The data shown are the representative results from 3 independent experiments. Source data are provided as a Source Data file.

MD-2 is essential for TLR4 recognition of GM

Of note, MD-2 is an essential co-receptor of TLR4 for extracellular LPS sensing⁴⁰, where specific residues in TLR4/MD-2 are known to stabilize the hydrophobic lipid chains and the polar glycan core of lipid A⁴¹. To investigate if MD-2 is also required for TLR4 recognition of GM, we evaluated the NF-κB activity in GM-stimulated HEK293T cells that transiently express either mTLR4, mMD-2, or both (Fig. 7a). The expression levels of the transfected genes were validated by Western blot (Fig. S16). Similar to LPS, GM exhibits dose-dependent activation of NF-κB signaling only in cells that co-express mTLR4 and mMD-2, indicating MD-2 is also essential for GM recognition by TLR4. Considering GM resembles the disaccharide core of Lipid A, we predict that the TLR4/MD-2 residues involved in polar interactions are likely crucial for GM recognition. As expected, mutations of mTLR4_K263 and Q339, both of which form hydrogen bonds with the glycan core of Lipid A^41,42, abolished GM-triggered NF-κB activity; whereas mutating mMD-2_F119, a residue that stabilizes the hydrophobic acyl chains in LPS^41,42, did not significantly impair the activity of GM. In addition, mutation of mMD-2_F126, which blocks ligand-induced TLR4 dimerization^41,42, abrogated NF-κB activation by both GM and LPS, suggesting that dimerization of the TLR4/MD-2 complex is required for GM-induced signal transduction. While our study elucidated several molecular interactions in GM recognition, a detailed understanding of how the disaccharide GM engages with the TLR4/MD-2 complex requires further structural investigations.

Fig. 7: TLR4/MD-2 specifically recognizes GM.

a Analysis of specific residues in TLR4/MD-2 required for GM-induced NF-κB activity. The various mTLR4/mMD2 variants were transiently expressed in HEK-Blue^TM reporter cells and treated with GM (0.4 mM, 0.2 mM) or LPS (100 ng/mL) for 24 h. The activation of NF-κB was measured based on the SEAP reporter signals. For each transfected group, the readouts were normalized to the H₂O-treated samples. Data are presented as mean values ± s.e.m. (n = 3). b-c Structural-activity relationship (SAR) analysis of disaccharide analogs using HEK-Blue^TM mTLR4 reporter cells. Cells were treated with respective disaccharides at 0.5 or 1 mM for 24 h. LPS (20 ng/mL and 100 ng/mL) was used as controls. Data are presented as mean values (n = 2). All experiments were repeated twice with similar results. Source data are provided as a Source Data file.

Structure-activity-relationship (SAR) analysis of GM-induced TLR4 signaling

Given the mild stimulation by GM and its weak affinity to TLR4, we questioned the specificity of TLR4 recognition of disaccharide ligands. In HEK-Blue^TM mTLR4 reporter cells, we tested a panel of synthetic disaccharides bearing various structural modifications, including deacetylation of GlcNAc (A1), alterations of the lactoyl moiety on MurNAc (A2-A4), 1,6-anhydro-terminus (GahM), and the regioisomer MG, as well as chitobiose and diacetylated chitobiose (Fig. 7b, c). Remarkably, only GM and MG stimulate TLR4, while other structural changes render the disaccharides inactive (Fig. 7c), suggesting the specific TLR4 recognition of PGN-derived disaccharides. In addition, we found that disaccharide-muropeptides, such as GM-A and GM-AE, also do not induce TLR4 signaling (Fig. S17A). Notably, GM does not activate HEK-Blue^TM mTLR2 reporter cells (Fig. S17B). However, this assay alone cannot completely rule out the possibility that GM might also act on TLR2. Lastly, we demonstrated that GM antagonizes LPS-stimulated TLR4 signaling in reporter cells, likely by competitively binding to the TLR4 receptor (Fig. S17C).

GM mitigates DSS-induced colitis in mice in a TLR4-dependent manner

Upon establishing GM as a bioactive TLR4 ligand in vitro, we asked whether this naturally released PGN disaccharide has any biological significance in vivo. Given previous work showing that MDP-mediated NOD2 signaling protects against DSS-induced colitis in mice⁴³, we sought to explore the potential efficacy of GM in this context. In our timeline, GM was administered intraperitoneally to mice daily from two days before and throughout a week during which 3% DSS was supplemented in their drinking water. For the control group, PBS was injected in place of GM, following the same protocol (Fig. 8a). Importantly, we validated that intraperitoneal injection of GM temporally increased its abundance in the mouse gut, providing grounds for daily injections to maintain elevated levels of gut PGN saccharides (Fig. S18A). The administration of GM (without DSS treatment) did not result in any morbidity in mice. More gratifyingly, GM alleviated DSS-induced colitis in WT mice, as evidenced by reduced body weight loss and colonic shortening (Fig. 8b–d). In H&E-stained colonic tissue images, the GM-treated DSS group showed reduced epithelial erosion and mononuclear cell infiltration (Figs. 8e and S18B). Correspondingly, infiltrating CD45⁺ immune cells, including monocytes and neutrophils, in the colonic lamina propria were significantly reduced in GM-treated DSS mice (Figs. 8f and S18C). Consistently, colonic expressions of proinflammatory genes, such as Tnfa, Il1b, and Ccl2 were significantly suppressed, supporting attenuated colonic inflammation upon GM supplementation (Fig. 8g). On the other hand, cytoprotective factors of colonic damage such as Keap1 and Hspb1 were upregulated in GM-treated colitis mice, although no significant change in Il10 was observed (Fig. S18D). To further assess systemic inflammation in mice, we conducted multiplex immunoassays to quantify 13 cytokines and chemokines in sera of WT mice at the timeline endpoint (Fig. S18E). DSS-induced colitis mice manifested a global elevation of inflammatory markers, including IL-6, IFN-β, IL-27, IL-10, and IL-23. GM supplementation restored circulating proinflammatory markers to near baseline levels, consistent with its ability to mitigate DSS-induced inflammation in mice. Importantly, GM alone does not significantly alter circulating cytokines and chemokines in mice.

Fig. 8: Administration of GM reduces colon inflammation in DSS-induced colitis in WT mice but not Tlr4^-/- mice.

a Timeline of the study, where GM was intraperitoneally administered at 10 mg/kg to mice (WT and Tlr4^-/-) that were provided with 3% DSS or normal drinking water. I.p. injections of PBS were given to the negative control group. b-c Measurements of mice body weight (b) (n = 8, 6, 10, 10, 9, 8, 10, 10, respectively) and colon length (c) (n = 8, 6, 14, 14, 8, 8, 15, 14, respectively) at the endpoint of the timeline. d-e Representative images of colons (d) and H&E-stained colon sections (e) from each group (n = 5). f Flow cytometry analysis of immune cell infiltration in the colons of each treatment group. The percentage of the CD45⁺ population (n = 8, 6, 12, 13, 7, 8, 12, 12, respectively) was calculated within pre-gated live cells. The percentage of Ly6C⁺ MHCII^- (n = 8, 6, 10, 10, 7, 8, 10, 8, respectively) and Ly6C⁺ Ly6G⁺ (n = 8, 6, 8, 8, 7, 8, 11, 9, respectively) populations was calculated within the CD45⁺ population. g RT-qPCR analysis of colonic cytokines demonstrates the significantly reduced inflammation in the GM-supplemented DSS-induced colitis in WT mice but not Tlr4^-/- mice. Tnfα, n = 10, 10, 9, 9, 5, 8, 8, 8, respectively. Il1β, n = 12, 12, 10, 13, 5, 8, 8, 7, respectively. Ccl2, n = 10, 10, 9, 9, 5, 8, 6, 6, respectively. Data are presented as mean values ± s.e.m. (n as indicated) with indicated P values determined using two-way ANOVA followed by Tukey’s multiple comparison for (c, f-g). All experiments were repeated twice with similar results. Source data are provided as a Source Data file.

To determine whether GM’s in vivo protective effects are TLR4-dependent, we performed the aforementioned experiments in Tlr4^-/- mice (Fig. 8a). In this case, GM administration failed to prevent weight loss and colonic shortening in DSS-administered Tlr4^-/- mice (Fig. 8b–d). In addition, similar levels of colonic crypt damage were observed in the colitis Tlr4^-/- mice, regardless of GM supplementation (Fig. 8e). The results of immune cell infiltration and colonic proinflammatory gene expression were consistent with the loss of GM’s protective effects against DSS-induced colitis in Tlr4^-/- mice (Fig. 8g). No upregulation of cytoprotective responses in GM-treated colitis Tlr4^-/- mice was observed (Fig. S18D). Collectively, our data suggest that the gut microbiota-derived disaccharide GM maintains host gut homeostasis via TLR4-dependent mechanisms.

Discussion

Using the LC-MS/MS-based PGN analysis platform, we discovered that 90% of the gut microbiota-derived PGNs in the host gut consist of saccharides, while the remaining muropeptides are primarily of the mDAP type. Given that the gut bacterial peptidoglycome encompasses both Lys- and mDAP-type peptidoglycan from Gram-positive and Gram-negative bacteria¹⁴, the soluble muropeptides in the host gut milieu appear mostly derived from Gram-negative bacteria. We speculated that gut commensals such as Bacteroides spp. that possess mDAP-containing peptidoglycan could be the major contributors to gut PGNs⁴⁴. Moreover, we did not identify the model ligand MDP (M-AQ) in hosts; instead, we found that its non-amidated analogs, such as M-AE and GM-AE, are prevalent in the mice ceca. These natural muramyl dipeptides are likely the cleavage products of commensal bacteria-secreted peptidoglycan endopeptidases, such as SagA and LPH. Previous studies have shown that elevated levels of these enzymes or their corresponding cleaved muropeptide products benefit gut homeostasis and enhance cancer immunotherapy in mice^7,45,46,47. On the other hand, we only detected a minute amount of PGN monosaccharide (e.g., M and ahM) in healthy host sera, which could indicate selective systemic dissemination of gut PGNs²⁵ or PGN degradation by serum amidase, such as PGRP2³¹. Alternatively, circulating PGNs may be retained by serum proteins or carried by vesicular transporters^25,48,49, potentially evading detection by LC-MS/MS. Nevertheless, unlike existing PGN detection assays that rely on protein-based recognition of particular PGN motifs, our LC-HRMS/MS analysis directly reveals PGN structures in host samples.

Intrigued by the abundance of PGN saccharides that do not resemble classic NOD1/2 ligands in the host gut, we were prompted to investigate the potential bioactivity of these non-canonical PGN moieties. Wolf et al. previously reported that cytosolic GlcNAc (G) released from the phagosome degradation of bacterial peptidoglycan is detected by hexokinase that leads to inflammasome activation in LPS-primed immune cells, unveiling a NOD1/2-independent PGN sensing mechanism in the host²¹. Conversely, MurNAc (M) was previously shown to suppress inflammation in LPS-induced macrophages in vitro⁵⁰, and a recent report suggests it may help alleviate colorectal cancer⁵¹. Herein, we demonstrate that the gut microbiota-derived disaccharide GM exerts mild immuno-stimulatory effects in the absence of LPS priming, under which condition the monosaccharides G and M are barely active. Importantly, we established that GM directly binds to TLR4 and activates its downstream NF-κB and IRF pathways. Of note, bacterial LPS is the canonical TLR4 ligand, whose Lipid A motif constitutes a β−1,6-linked glucosamine disaccharide with phosphate groups and multiple acyl chains⁵². Apart from LPS, endogenous ligands such as heat shock proteins, hyaluronan, and monosodium urate crystals, as well as synthetic small molecules like neoseptins, were reported to stimulate TLR4 signaling, demonstrating the structural diversity of TLR4 ligands^53,54,55. Our discovery of the gut microbiota-derived disaccharide GM as a TLR4 agonist expands the scope of natural TLR4 ligands in hosts. Structurally, GM may resemble the disaccharide core of Lipid A for TLR4 recognition; however, the lack of hydrophobic acyl chains in GM renders it a mild TLR4 agonist. Interestingly, the presence of a stem peptide in disaccharide (GM)-muropeptides, such as GM-A and GM-AE, diminishes TLR4 activation. Hence, the fact that TLR4 selectively recognizes disaccharide PGNs (i.e., GM and its regioisomer MG) but not closely related fungal cell wall fragments (such as GG) underscores the potential biological significance of natural PGNs in gut microbiota-host interactions. However, with the current reporter assays, we cannot completely rule out the potential involvement of TLR2 in GM-induced immunological effects.

Notably, TLR4 activation by metabolites from gut commensal microflora under steady-state conditions is essential for maintaining gut homeostasis and protecting against colonic injury⁵⁶. For instance, LPS supplementation rescues the severity of DSS-induced colitis in commensal-depleted mice via TLR4 stimulation^56,57. Our discovery of the naturally abundant gut PGN-disaccharide GM as a TLR4 ligand suggests its physiological relevance in hosts. The administration of GM protects against DSS-induced colitis in mice via a TLR4-dependent manner. We offer several hypotheses accounting for its in vivo protection. First, GM may act to prime TLR4 to downregulate its inflammatory responses to subsequent triggers in hosts. Supportively, we observed that GM antagonizes LPS-induced activation of TLR4-mediated NF-κB signaling in reporter cells in vitro. Second, colonic epithelium TLR4 activation confers cytoprotection and repair against DSS-induced tissue injury and damage⁵⁶. To this end, we observed that GM supplementation upregulated cytoprotective genes, such as Keap1 and Hspb1 in colonic tissues of the DSS-treated WT but not Tlr4^-/- mice. However, GM did not seem to promote the production of the anti-inflammatory cytokine IL-10 in vitro and in vivo. Lastly, GM-induced TLR4 activation may, in turn, regulate other signaling pathways to provide protective effects in vivo. For example, NOD2 activation by MDP suppresses DSS-induced colitis in mice via the induction of the negative regulator IRF4⁴³. The transcriptomics analysis in this work uncovered the intricate effects of TLR4 on MDP-mediated NOD2 signaling, as indicated by the reduced expression of specific genes in MDP-treated Tlr4^-/- BMDMs compared to MDP-treated WT BMDMs. Extrapolating from these in vitro observations, we postulate that GM supplementation stimulates TLR4 signaling, which may promote NOD2 activation by endogenous gut microbiota-derived muropeptides like M-AE and GM-AE to protect against DSS-induced colitis in mice. Consistently, GM triggers augmented Nod2 expression in WT BMDMs but not in the Tlr4^-/- counterparts, implying that TLR4-mediated NF-κB signaling may regulate NOD2 expression. While in-depth investigations into the in vivo protective mechanisms should be carried out in future studies, the gut microbiota-derived disaccharide GM, a mild TLR4 agonist that does not induce acute inflammation or morbidity in mice, emerges as a promising candidate as postbiotics or adjuvants. Fine-tuning the TLR4-mediated immune responses holds potential for treating gut microbiota-associated inflammatory diseases.

Methods

Human samples

Nanyang Technological University (NTU) Institutional Review Board approval was obtained (IRB-2021-03-030) to collect human stool and serum samples from healthy participants. Informed consent was obtained from each volunteer. For study enrollment, volunteers from both genders need to meet the exclusion criteria as follows: 1) age between 22–65; 2) not pregnant; 3) not taking any antibiotic drugs during sample collection; and 4) no existing medical conditions. The serum was isolated and separated immediately after blood collection and aliquoted in 1.5 mL tubes for storage at −80 °C. Stool specimens were frozen at −80 °C and homogenized in sterile dH₂O before PGN analysis.

Mice

Mice were age-, weight-, and sex-matched for all analyses. SPF Tlr4^-/- C57BL/6 J mice (both genders) were purchased from Jackson Laboratory (USA) at 7–9 weeks of age for in-house breeding. SPF WT C57BL/6 J mice (both genders) were obtained from NTU animal facilities. Mice were housed under standard conditions in a climate-controlled environment with 12 h light and dark cycles and fed ad libitum with a maintenance diet and water. Paired mice (1 male and 2 females per cage) were provided with a breeding diet. Mice were allowed to acclimate for 1 week prior to any studies. Male mice at 7–8 weeks of age were used in the DSS-induced colitis model. Animal experiments were approved by the Institutional Animal Care and Use Committee of NTU (Approval No: A23003 and A23039).

Cell culture

Caco-2 (HTB-37™) and RAW264.7 (TIB-71™) cells were obtained from the American Type Culture Collection. THP-1 cells were gifted by Prof. Veit Hornung (Ludwig Maximilian University, Germany). HEK-Blue™ Null1 / Null2 / hNod1 / hNod2 cells were gifted by Prof Yue Wang (A*STAR Infectious-Diseases Labs, Singapore). HEK-Blue™ mTLR4 cells were gifted by Prof Zhang Guodong (National University of Singapore, Singapore). HEK-Blue™ mTLR2 cells (hkb-mtlr2), RAW-Dual™ (rawd-ismip), and RAW-Dual™ Tlr4^-/- cells (rawd-kotlr4) were purchased from InvivoGen. Caco-2 cells were maintained in high-glucose Dulbecco’s Modified Eagle Medium (DMEM) with sodium pyruvate (Gibco). RAW264.7 WT were cultured in high-glucose Iscove’s Modified Dulbecco’s Medium (IMDM) containing HEPES (Gibco). THP-1 cells were maintained in Roswell Park Memorial Institute 1640 (RPMI 1640, Gibco). All reporter cells (HEK-Blue™ mTLR4 / mTLR2 / Null1 / Null2 / hNod1 / hNod2 cells, RAW-Dual™ and RAW-Dual™ Tlr4^-/- cells) were grown in high-glucose DMEM supplemented with 100 µg/mL Normocin (InvivoGen) and selection antibiotics as required. Specifically, HEK-Blue™ Null1 and Null2 cells were selected under 100 µg/mL Zeocin (InvivoGen); HEK-Blue™ hNod1 and hNod2 cells were selected under 100 µg/mL Zeocin (InvivoGen) and 30 µg/mL Blasticidin (InvivoGen); HEK-Blue™ mTLR4 and mTLR2 cells were supplemented with 1× HEK-Blue™ Selection (InvivoGen). RAW-Dual™ and RAW-Dual™ Tlr4^-/- cells were supplemented with 200 µg/mL Zeocin (InvivoGen) every other passage. Selective antibiotics were excluded during testing. Primary cells isolated from mice were sustained in IMDM with corresponding differentiation ligands. All abovementioned culturing media contain L-glutamine and were supplemented with 100 units/mL penicillin, and 100 µg/mL streptomycin (Cytiva) unless otherwise noted. DMEM and RPMI 1640 were mixed with 10% (v/v) heat-deactivated FBS (FBS, Gibco) while IMDM was supplemented with 5% (v/v) heat-inactivated FBS instead. Cells were routinely cultured in FBS- and antibiotics-supplemented media in the humidified incubator at 37 °C with 5% CO₂.

Human serum and stool collection

The freshly collected blood was kept still for 15 min at room temperature for spontaneous agglutination, followed by centrifugation at 3000 rpm for 15 min with a low accelerating and decelerating speed at 4 °C. The serum at the top layer was collected and aliquoted into new 1.5 mL tubes for sample cleanup for LC-MS analysis directly or stored at −80 °C. (Note: for optimal serum PGN detection by LC-MS, we recommend performing serum protein removal within the same day of blood collection. We found that PGN signals gradually decreased in serum samples that were frozen immediately upon collection without protein removal processing).

Sterile stool sample collection kits were distributed to the volunteers within the week of blood donation. Stool samples were weighed and resuspended in sterile dH₂O followed by homogenization aided by autoclaved stainless-steel beads using a PowerLyzer 24 Homogenizer (Qiagen) at 1000 rpm for 1 min. The homogenized stool samples were centrifuged at 13,500 × g for 5 min, and the supernatant was collected into new tubes. The reminder stool pellet was further resuspended with sterile dH₂O by vortexing for 30 s followed by centrifugation (13,500 × g for 5 min). The supernatant was combined and diluted to a final concentration of 50 mg stool per 1 mL sterile dH₂O. The combined stool supernatant was subjected to centrifugation at 16,000 × g for 5 min to remove any debris carried over from the previous steps. The stool supernatants (50 mg/mL) were saved at −80 °C as aliquots for further analysis.

Serum and stool sample cleanup

For MurN quantification, 100 µL concentrated HCl (12 M stock) was added to a 200 µL sample (fecal supernatant or standard MurNAc) to achieve 4 M HCl-containing solution. The mixture underwent hydrolysis by heating at 90 °C for 3 h, followed by acid removal via a rotatory evaporator. The concentrated samples were reconstituted in 1% (v/v) TFA in dH₂O and were subjected to cleanup with a pre-equilibrated Strata C18-E solid phase extraction (SPE) cartridge (Phenomenex), where the eluted fractions of 0–20% acetonitrile in water were collected, lyophilized, and resuspended in 200 µL dH₂O. Upon centrifugation at 16,000 × g for 5 min to remove any sediments, the clear supernatant was collected for MurN quantification. For the serum sample (1 mL), methanol (9 mL, HPLC grade) was added to precipitate serum proteins. After centrifugation at 16,000 × g for 5 min, the clear supernatant was collected, dried by rotatory evaporation, and reconstituted in 200 µL dH₂O (i.e., 5x original serum concentration), which was then subjected to acid hydrolysis and SPE cleanup as described above. The processed serum was eventually resuspended in 100 µL dH₂O for MurN quantification (i.e., 10× original serum concentration). We found that concentrating serum samples by 10× is necessary to enhance MurN signals in LC-MS analysis. To note, FBS from Capricorn (Cat. No. S1810) and Biowest (Cat. No. FBS-11A) were subjected to MurN analysis following the same protocols.

For PGN subtype profiling, 1 mL fecal supernatant (50 mg feces/mL dH₂O) was directly subjected to SPE cleanup, where eluted fractions of 0–20% acetonitrile in water were collected, lyophilized, and resuspended in 100 µL dH₂O for subtype screening (i.e., 10× original concentration). 1 mL of serum was subjected to protein precipitation with methanol (9 mL) before SPE cleanup. The collected elute fractions from the SPE column were dried and reconstituted in 100 µL dH₂O for subtype profiling (i.e., 10× original concentration). We found that sample concentration was necessary for PGN subtype analysis. For quantification of PGN subtypes, 1 µg/mL sialic acid (Sigma-Aldrich, Cat. No. A0812) was spiked into the final reconstituted samples as an internal standard in LC-MS analysis.

LC-MS/MS workflow for PGN quantification and profiling

PGN analysis was performed with the Vanquish HPLC system equipped with Orbitrap Exploris 120 mass spectrometer (Thermo Fisher Scientific). The autosampler was controlled at 18 °C, and the column oven was at room temperature, with the sample injection volume set as 10 µL for each run. Gradient separation was achieved with a Phenomenex C18 Kinetex EVO18 RP column (2.0 × 150 mm, 100 Å) with solvent A (0.05% formic acid in H₂O, LC/MS grade, Fisher Scientific) and solvent B (0.05% formic acid in acetonitrile, LC/MS grade, Fisher Scientific) at a constant flow rate of 0.2 mL/min. For MurN quantification, the LC program starts with 1% B holding for 2 min, followed by a linear gradient to 30% B at 5 min, then up to 90% B at 6 min, which is held for 2.5 min. The solvent gradient is then converted to 1% B within 0.1 min and kept for 2.4 min as pre-equilibration for the next run. For PGN subtype profiling, the LC method starts with 1% B for 2.5 min, followed by a steady linear increase to 40% B at 27.5 min, which is kept for 2.5 min. Next, the gradient increases to 90% B at 31 min, which was held for 4 min, followed by the return to 1% B at 36 min and kept for 4 min till the end as pre-equilibration for the next run. Elutes of the first minute are diverted to waste to avoid solvent peak suppression and high salt contamination.

The OT Exploris 120 mass spectrometer was operated in positive ion mode using electrospray ionization and internal calibration. The spray voltage was 3500 V with a vaporizer temperature of 300 °C and an ion transfer tube temperature of 325 °C. Gas was statically flowing at 50, 10, and 0 (Arb) for sheathing, auxiliary, and sweeping gases, respectively. For MurN quantification, only MS2 spectra were acquired in PRM mode for the target precursor ion at m/z = 252.1078. MS1 spectra were collected in FS mode, ranging from 120–1200 m/z with a resolution of 60,000. MS2 spectra were acquired for the fragmentation of assigned precursors in FSDDA mode when the intensity exceeded the threshold of 1.0e5. For the acquisition of MS2 spectra, 4 MS/MS scans were acquired for each precursor ion (within the isolation window of ±1 m/z) with a higher energy collision dissociation energy of 30% and a resolution of 15,000.

MurN analysis and calculation of ion suppression factors in biological samples

A serial dilution of MurNAc (0–5000 ng/mL) was utilized for the establishment of the standard calibration curve. Areas under the curve (AUCs) of extracted ion chromatograms of designated daughter ions were plotted against concentrations of MurNAc, resulting in linear regression. To determine the ion suppression factors in biological samples brought by matrix, a serial dilution of MurNAc (0–2500 ng/mL) was spiked into pooled sera or feces, which were then subjected to an acid hydrolysis workflow. According to the standard calibration curve, apparent MurN concentration in individual spiked samples was calculated based on the observed AUCs. The serial ‘‘calculated concentrations’’ (y-axis) were then plotted against the known spiked-in concentration of MurNAc (x-axis). The inverse of the slope(s) of the linear regression fitting plot(s) reflects the signal suppression by the biological matrix. For MurN quantification in individual biological samples, the detected MurN concentration was corrected with the respective matrix suppression factor.

Transwell assay

Caco-2 cells were seeded in Transwell inserts (24-well format; Corning, CLS3470) using complete media (DMEM + 10% FBS). The media was changed every two days until an intact monolayer was formed. The monolayer was then washed twice with PBS and incubated with DMEM (serum-free) for 2 h prior to the assay experiment. The compound was mixed with DMEM (serum-free) in a 1:9 ratio, as indicated in the figure legends. The media in the inserts were completely removed, and 150 µL of the compound mixture was introduced into each insert. The wells were filled with 600 µL of PBS. At specified time intervals described in the figure legends, 50 µL of PBS was collected. Samples were twofold diluted in filtered dH₂O and centrifuged at 16,000 × g for 5 min. Supernatants were transferred to the vials for LC-MS analysis. For the permeability test of the monolayer, FITC-Dextran (1% v/v, 20 kDa; Sigma, FD20) was added to the inserts, with or without cells. PBS samples were collected for fluorescence measurement using a BioTek Synergy HTX multi-mode reader.

Compound uptake assay

HEK-Blue^TM hNOD2 reporter cells were seeded as 50,000 cells/well in 24-well plates for 24 h, followed by GM (0.4 mM) treatment for 24 h at 37 °C. 1 mL complete media was added in each well. Cells were harvested and centrifuged at 200 × g for 5 min. Cells were washed with PBS and then resuspended in ice-cold 90% methanol, followed by lysis at 100 °C for 5 min. The lysates were shaken at 200 rpm at 4 °C for 30 min before being centrifuged at 16,000 × g for 5 min at room temperature. The supernatants were collected and dried out by SpeedVac vacuum concentrator (Thermo Fisher Scientific) at 30 °C. The samples were reconstituted in 50 μL dH₂O, briefly vortexed and centrifuged at 16,000 × g for 5 min. The supernatants were subjected to HPLC-MS analysis.

Acquisition was performed on the same equipment. Gradient separation was realized by Phenomenex Luna Omega Polar C18 column (100 × 4.6 mm, 3 μm, 100 Å) at a constant flow rate of 0.5 mL/min. The LC program starts with 1% B (0.05% formic acid in acetonitrile, LC/MS grade, Fisher Scientific) holding for 3 min, followed by a linear gradient to 50% B at 8 min and holding for 2 min, then up to 80% B at 11 min, which was held for 4 min. The solvent gradient was then converted to 1% B within 0.5 min and kept for 1.5 min as pre-equilibration for the next run. MS1 spectra were collected in FS mode, ranging from 120–1200 m/z with a resolution of 60,000.

Reporter cell culture and assay

RAW-Dual™, and RAW-Dual™ Tlr4^-/- cells were harvested and seeded as 100,000 cells/well in 24-well plates for 24 h before PGN treatment. 1 mL of complete media was added to each well without any selection antibiotic. THP-1 cells were harvested and seeded as 100,000 cells/well in half-serum media in 24-well plates and immediately proceeded with ligand stimulation. Time and concentration were indicated in the corresponding figure legends.

HEK-Blue™ Null1 / Null2 / hNod1 / hNod2 / cells were seeded as 50,000 cells/well in flat-bottom 96-well plates and used immediately for PGN stimulation. 0.2 mL of complete media without selection antibiotics was added to each well. HEK-Blue™ mTLR4 / mTLR2 cells were seeded at 50,000 cells/well in complete media in flat-bottom 96-well plates. 0.2 mL of complete media without selection antibiotics was added to each well. After 6 h, replace the media with 0.2 mL/well DMEM (serum-free) and proceed with PGN stimulation. Time and concentration were indicated in the corresponding figure legends.

40 µL of supernatants from treated reporter cells were mixed with 160 µL Quanti-Blue reagent (InvivoGen) and incubated at 37 °C for 6 h. The absorbance was measured at λ=620 nm after 10 s shaking by a Thermo Multi-Scanner Microplate reader. The readouts were blanked with negative control (i.e., sterile dH₂O) and also subtracted by the readouts of the respective parental cell line (if any).

20 µL of supernatants from treated reporter cells were mixed with 50 µL of Quanti-Luc 4 reagent (InvivoGen) and luminescence was detected immediately by CYTATION5 imaging reader (BioTek) and recorded by Gen3.2 software. Auto gate and default luminescence filter were set for recording.

Murine bone marrow cell differentiation and compound stimulation

Femur and tibia were harvested from euthanized SPF C57BL/6 J female mice. 10 mL complete media was used to flush a pair of femur and tibia through a 27 G needle. Cells were centrifuged at 300 × g for 5 min and then resuspended in red blood cell lysis buffer (0.89% NH₄Cl, w/v, 10 mM KHCO₃) for 8 min. Lysis was quenched by complete media. White blood cell pellets were resuspended in complete media, followed by filtration and counting.

Cells were diluted to 2,000,000 cells/mL in BMDC media (complete media supplemented with FLT3 ligand-containing cell supernatants) and seeded at around 20,000,000 cells in a 10 cm culturing dish for DC differentiation. 10 mL of complete media was added to the same dish on days 3 and 6. DCs were harvested on day 9⁵⁸, and 2,000,000 cells/well were seeded in a 24-well plate. 1 mL of BMDC media was added to each well. Cells were stimulated with commercially available ligands or chemically synthesized compounds for 18 h.

Cells were diluted to 200,000 cells/mL in BMDM media, complete media supplemented with 20 ng/mL recombinant GM-CSF (Biolegend, Cat. No. 576306) or 10% L929 cell supernatants, for macrophage differentiation. 40,000 cells/well were seeded in a flat-bottom 96-well plate for 6 days. 0.2 mL of BMDM media was added to each well. The media was replaced by freshly prepared BMDM media on day 3. 800,000 cells/well were seeded in a 12-well plate for 6 days. 1 mL of BMDM media was added to each well. Another 1 mL of BMDM media was added to each well on day 3. Cells were ready by day 6⁵⁹. Media were replaced by IMDM (serum-free) by day 6. Compounds of various concentrations (as indicated in figure legends) were added to 0.2 mL IMDM in a 96-well plate or to 1 mL IMDM in a 12-well plate. Upon ligand stimulation for 24 h, cells were harvested for western blot, RT-qPCR, or RNAseq analysis. Culture supernatants were collected for ELISA. Commercial ligands and small-molecule inhibitors used in cell treatment are listed in Table S3.

RNAseq and RT-qPCR analysis

The total RNA from cells was extracted by Monarch Total RNA Miniprep Kit (New England Biolabs) according to the vendor’s recommended protocols. Tissues were first homogenized in TRIzol reagents (Ambion, Invitrogen) and stored at −80 °C until further processing. Chloroform was added to the TRIzol-tissue complex and sat for 5 min after vortexing, followed by centrifugation (16,000 x g for 10 min) at 4 °C. The top layer containing RNA/DNA was separated. Total tissue RNA was extracted by the same kit according to the recommended protocols. RNA was eluted by RNase-free ddH₂O, and its concentration was determined by NanoDrop™ One Microvolume UV-Vis Spectrophotometer (Thermo Fisher Scientific). Samples were preserved at −80 °C for RNAseq (Novogene) or RT-qPCR analysis. RNA (500–1000 ng) was utilized for reverse transcription of each sample (normalized within each batch) by ProtoScript II First Strand cDNA synthesis kit (New England Biolabs) or M-MLV reverse transcriptase (Promega) to yield cDNA according to the recommended protocols. Necessary reagents, including Oligo-dT (5’-TTT TTT TTT TTT TTT-3’, 100 µM in IDTE buffer, Integrated DNA Technologies), dNTP mix (0.5 mM, Thermo Fisher Scientific), RiboLock RNase Inhibitor (1 U/µL, Thermo Fisher Scientific) were mixed with M-MLV reverse transcriptase as required. cDNA samples were then diluted if needed or directly used as templates for quantitative PCR (YouSeq Tetra™ OneStep 2× qRT-PCR Mastermix or Bio-Rad SsoAdvanced Universal SYBR Green Supermix). The readouts were obtained from the Eco Real-Time PCR System (Illumina) or CFX Duet Real-Time PCR System (Bio-Rad). Relative gene expression of target genes was determined utilizing 2^-ΔΔCt method and normalized to the housekeeping gene β-act. Primers used (customized design and ordered from Integrated DNA Technologies) are listed in Table S1.

ELISA

All samples used for ELISA were centrifuged at 16,000 × g for 5 min to remove any potential sediments or cell debris right before assays. ELISA MAX™ standard sets (Biolegend) were utilized to evaluate cytokine (human IL-8, mouse IL-6, IL-10, and TNF-α) in culture supernatants following the manufacturer’s recommended protocols. Samples were prediluted with blocking buffer (1% BSA in PBS, pH 7.4) if required. BSA (Bovine Serum Albumin, powder, Sigma-Aldrich) stock solution (1%, w/v) was dissolved in sterile PBS (pH 7.4) followed by 0.22 μm filtration. The substrate TMB set (Biolegend) was added, and the reaction was stopped by 1 M sulfuric acid. The final absorbance at λ = 450 nm was recorded by Thermo Multi-Scanner Microplate and SkanIt software (Thermo Scientific).

Molecular docking of GM into murine TLR4-MD2 model

The chemical structure of GM was converted from its SMILES notation [CC( = O)NC1C(OC2C(CO)OC(O)C(NC(C) = O)C2OC(C)C( = O)O)OC(CO)C(O)C1O] to PDB file using OpenBabel. The model of ligand-free murine TLR4-MD2 protein complex was obtained from UniProt (PDB: 5IJB)⁴¹. To generate the ligand and receptor for docking, water and other molecules were removed from each of the GM and mTLR4-MD2 models and converted to individual PDBQT files. Docking simulation was conducted using AutoDock-Vina by fitting GM into the reported LPS binding site of mTLR4-MD2 in a 30 × 30 × 30 Å grid box. The docked complex with the lowest free-binding energy was selected for visualization and analysis using BIOVIA Discovery Studio.

Murine TLR4-MD2 co-transfection in HEK293T and PGN treatment

Recombinant plasmid pFlag-CMV-1::mTLR4 harboring FLAG-tagged murine toll-like receptor 4 (mTLR4) cDNA was obtained from Addgene (Cat. No. 13087). c-Myc-tagged murine MD2 was de novo synthesized as gBlock fragment (Integrated DNA Technologies) according to the cDNA sequence from NCBI (accession number NM_016923) and cloned into pcDNA3.1-Zeo backbone between KpnI and EcoRI sites, yielding pcDNA3.1-Zeo::mMD2-Myc. E. coli DH5α was transformed, and colonies were selected with carbenicillin. mTLR4-MD2 mutants were generated according to the Q5 site-directed mutagenesis kit (New England Biolabs) using SDM primers (listed in Table S2). Plasmid was extracted and purified using FavorPrep™ Plasmid DNA Extraction Kit (Favorgen Biotech Corp.) according to the manufacturer’s protocol.

Variant mTLR4-mMD2 complex was co-transfected into HEK-Blue™ Null1 (seeding density: 50,000 cells/mL complete media, 1 mL/well in 24-well plate) using Lipofectamine 3000 according to the manufacturer’s protocol. In each well, cells were transfected with a total of 500 ng of pFlag-CMV-1::mTLR4 and pcDNA3.1-Zeo::mMD2-Myc (mixed at 1:1) for 16 h. Water-Lipofectamine mixture instead of plasmids was added as mock-transfected cells. Culture media was replaced with fresh complete DMEM the next day, and the transfected cells were then incubated with various PGN compounds for 24 h. Culture supernatants were collected to measure NF-κB activity based on SEAP-reporter assay.

Immunoblotting

WT and Tlr4^-/- BMDMs were directly stimulated with LPS (500 ng/mL) or chemically synthesized GM (5 mM) for indicated time in the figure legends. The lysis buffer (66 mM Tris-Cl pH 7.4, 2% SDS, 0.5% Triton X-100) was freshly supplemented with protease inhibitor (Roche Life Science), phosphatase inhibitor cocktail (Nacalai Tesque), and 10 mM DTT. Cells were washed with cold PBS twice and lysed by complete lysis buffer and boiled for 15 min. Supernatants were collected after centrifugation at 14,000 × g for 5 min and boiled with 1×SDS-loading buffer supplemented with 50 mM DTT for 15 min followed by SDS-PAGE and PVDF membrane transfer. The membranes were blocked with 3% skim milk in TBST at room temperature for 1 h. Each membrane was incubated with primary antibodies (dilution factor 1:1000) corresponding to rabbit anti-mouse GAPDH or total/phosphorylated NF-ĸB p65 and MAPK p38 (Cell Signaling Technology) at 4 °C overnight after washing. The secondary antibody (HRP goat anti-rabbit, Invitrogen) was added to each membrane (dilution factor 1:5000) and incubated for 1 h at room temperature after washing. For all washing steps, the membrane was rotated in 10 mL TBST for 5 min for three rounds. After the final washing steps, Amersham™ ECL Select™ detection reagent (Cytiva) was added to the membranes, and protein bands were visualized and documented using ImageQuant 800 (Cytiva). Quantification of the band darkness was conducted by ImageJ.

mTLR4 pulldown by immobilized GM-agarose beads

Recombinant mouse TLR4 (26-638)-His₆ protein (Cusabio, Cat. No. CSB-YP023603MO) was reconstituted in water and aliquoted for storage at −80 °C. 40 μL alkyne agarose beads (Vector Laboratories, Cat. No. CCT-1032) were washed with 400 μL dH₂O twice before usage. Washed beads were collected by centrifugation at 750 × g for 45 s. To immobilize GM onto the beads, GM-1-N₃ (2.5 mM) was added together with 1.2 mM sodium ascorbate, 0.5 mM CuSO₄, and 0.5 mM THPTA to a total volume of 200 μL for the click reaction at 4 °C for 3 h with rotation. For negative control, click reagents were added, but without the GM-1-N₃ probe. Following the click reactions, the GM-immobilized agarose beads and control alkyne-agarose beads were washed, and incubated with recombinant mTLR4(26-638)-His₆ protein (100 or 50 ng) in a total volume of 200 μL in dH₂O for overnight incubation at 4 °C. The beads were centrifuged (at 750 × g for 45 s) and the supernatant was collected as flow-through, washed with 400 μL dH₂O (wash), and subjected to boiling in 40 μL 5X SDS-loading buffer containing 50 mM DTT for 15 min (elute). 20 μL of boiled samples were loaded onto an SDS-PAGE gel for analysis. For Western blot analysis, the transferred PVDF membrane was blocked with 5% skim milk in TBST at room temperature for 1 h, washed (10 mL TBST for 5 min x 3), and incubated with HRP-conjugated anti-His Ab (H-3, 1: 1000 dilution, Santa Cruz Biotechnology) for 1 h at room temperature. After final washing (10 mL TBST for 5 min x 3), the Amersham™ ECL Select™ detection reagent (Cytiva) was added onto the membrane for visualization with ImageQuant 800 (Cytiva).

Surface plasmon resonance

SPR analysis was conducted on a BIACORE T200 system (GE Healthcare) at 20 °C. mTLR4 was covalently immobilized on a CM5 sensor chip via a standard amine coupling protocol provided by an amine coupling kit (comprising NHS, EDC, and ethanolamine hydrochloride). Specifically, mTLR4 at 30 μg/mL in 10 mM sodium acetate (pH 4.5) was immobilized onto the active flow cell (Cell 2) to achieve approximately 3000 response units (RU), while the reference flow cell (Cell 1) was activated with NHS/EDC and subsequently blocked with ethanolamine without protein, serving as the substrate reference to facilitate accurate baseline subtraction. Prior to compound injection, three startup cycles with running buffer (0.02 M HEPES, pH 7.4, 0.15 M NaCl, 0.05% Tween 20) were performed to stabilize the baseline. Compounds were then injected in multi-cycle mode at 10 two-fold dilutions up to a maximum concentration of 500 μM, with an association phase of 120 s at a flow rate of 30 μL/min, followed by a dissociation phase of 240 s. Each cycle concluded with sensor surface regeneration using 10 mM glycine-HCl (pH 3.0) delivered at 30 μL/min for 30 s and a subsequent stabilization period of 30 s. The resulting sensorgrams were reference subtracted and fitted to a Steady State affinity binding model using Biacore T200 Evaluation Software to determine the binding parameters.

The Steady State Affinity binding model equation:

$${R}_{{eq}}=\frac{{R}_{\max }\times [C]}{{K}_{D}+[C]}$$

where R_eq is the equilibrium response signal, Rmax is the maximum binding capacity of the sensor surface, [C] is the concentration of the analyte, and K_D is the equilibrium dissociation constant indicating the affinity between the ligand and the analyte.

Dextran sulfate sodium (DSS)-induced colitis model

3% DSS (MP Biomedicals) drinking water (w/v) was given to SPF C57BL/6 J male mice for a continuous 7 days, followed by euthanasia. GM (10 mg/kg) or PBS (5 mL/kg) was intraperitoneally injected into mice daily starting from 2 days before the DSS water administration until the end of the experiments. The body weight of each mouse was recorded daily.

Colon immune cell isolation

The colon was harvested, cut open, and rinsed with cold PBS (pH 7.4) to remove the luminal contents. To remove the epithelium, the colon was incubated in 15 mL complete HBSS (Hanks’ balanced salt solution without Ca²⁺ and Mg²⁺ supplemented with 1.3 mM EDTA and 0.35 g/L sodium bicarbonate, Sigma-Aldrich) at 37 °C for 20 min with shaking. The buffer was replaced with another 15 mL of fresh complete HBSS, followed by shaking for 25 min at 37 °C. The colon was then washed with IMDM complete media and cut into small pieces, which were subsequently digested in IMDM complete media containing 1 mg/mL collagenase D (Roche) and 20 U/mL DNase I (Life Technologies) at 37 °C for 70 min. The digested tissue was then gently mashed through a 100 µm cell strainer. The leukocytes were enriched using a 70–30% Percoll gradient (GE Healthcare Life Science), where cells at the interface were collected and stained for flow cytometry analysis.

Flow cytometry

Cells collected from in vitro culture or isolated from mice were suspended in flow solution (PBS + 2% FBS). Cells were stained and subsequently analyzed using a BD LSRFortessa or FACSymphony A3 five-laser flow cytometer (BD Biosciences). Data were analyzed using FlowJo v10 (TreeStar) according to the gating strategy shown in Supplementary Figs. Cells were blocked with a homemade Fc-blocker (1:350 dilution). BMDCs were stained with antibodies in flow solution with indicated dilution factor as listed for 20 min: anti-MHC-II (Biolegend, Cat. No. 107632, 1:1000), anti-SiglecH (Biolegend, Cat. No. 129604, 1:600), anti-CD11c (Biolegend, Cat. No. 117318, 1:600), anti-CD11b (Invitrogen, Cat. No.45-0112-82, 1:600), anti-CD103 (Invitrogen, Cat. No. 17-1031-82, 1:600), anti-CD80 (Biolegend, Cat. No. 104708, 1:600) and anti-CD86 (Biolegend, Cat. No. 105008, 1:600), followed by Live/Dead Ghost dye Violet 510 (Tonbo Biosciences Cat. No. 50-105-2993) 1:1000 diluted in PBS for 5 min. Cells isolated from the colon were stained with antibodies in flow solution with indicated dilution factor as listed for 20 min: anti-CD45 (BD biosciences, Cat. No. 748371, 1:600), anti-MHC-II (Biolegend, Cat. No. 107639, 1:1000), anti-CD11c (Biolegend, Cat. No. 117318, 1:600), anti-CD11b (Biolegend, Cat. No. 101226, 1:600), anti-CD64 (Biolegend, Cat. No. 139306, 1:600), anti-Ly6C (Biolegend, Cat. No. 128006, 1:600), anti-Ly6G (BD biosciences, Cat. No. 563978, 1:600), and anti-F4/80 (Biolegend, Cat. No. 123110, 1:600), followed by DAPI (Biolegend, Cat. No. 105008) 1:1000 diluted in PBS for 5 min. Cells were finally resuspended in flow solution before subjection to flow cytometry for analysis.

The inflammatory panel of mouse sera was evaluated according to the manufacturer’s protocol. The beads were finally reconstituted in the washing buffer offered in the kit and analyzed using the FACSymphony A3 five laser flow cytometer (BD Biosciences) attached to a high-throughput sampler (BD Biosciences). Data were uploaded and analyzed using the corresponding software QOGNIT (Biolegend).

Histology

The colon was separated immediately after the sacrifice. The lumen was cut open and flushed with PBS (pH 7.4, on ice) to remove excretions. The colon was further washed with cold PBS (pH 7.4) for five rounds with vortexing for 10 s per round. The clean colon was dipped into ethanol, glacial acetic acid fixative for at least 30 min. Ethanol and acid were washed away with 1× PBS (pH 7.4). The colon was rolled and placed in the cassette in 4% (w/v) paraformaldehyde solution diluted from stock (10% w/v aq. soln., methanol-free, Thermo Scientific) for long-term preservation. The samples were sent to the A*STAR, AMPL for histology service, where paraffin block preparation, slide preparation, and Hematoxylin & Eosin staining were completed. Three sections at different depths were collected on a slide for each sample. The images were taken by Nikon Eclipse 80i at magnification of 4× and 20×, respectively with default auto white compensation during imaging.

Statistical analysis

Graphs were prepared with statistical analysis using GraphPad Prism 10. Results were illustrated as mean values ± the standard error of the mean (s.e.m.) of biological replicates unless otherwise noted. Statistical analysis was performed using unpaired t test, one-way analysis of variance followed by Dunnett’s multiple comparison tests, or two-way ANOVA followed by Tukey’s multiple comparison tests wherever appropriate. The statistical tests used, exact p values, and sample sizes (biological replicates) were indicated in each figure legend. Data were representative of at least two independent experiments.

Materials availability

All unique/stable reagents generated in this study are available from the corresponding author with a completed Materials Transfer Agreement.

Reporting summary

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