Biosynthetic enzyme analysis identifies a protective role for TLR4-acting gut microbial sulfonolipids in inflammatory bowel disease

Older, Ethan A.; Zhang, Jian; Ferris, Zachary E.; Xue, Dan; Zhong, Zheng; Mitchell, Mary K.; Madden, Michael; Wang, Yuzhen; Chen, Hexin; Nagarkatti, Prakash; Nagarkatti, Mitzi; Fan, Daping; Ellermann, Melissa; Li, Yong-Xin; Li, Jie

doi:10.1038/s41467-024-53670-y

Download PDF

Article
Open access
Published: 30 October 2024

Biosynthetic enzyme analysis identifies a protective role for TLR4-acting gut microbial sulfonolipids in inflammatory bowel disease

Nature Communications volume 15, Article number: 9371 (2024) Cite this article

9243 Accesses
28 Citations
5 Altmetric
Metrics details

Subjects

Abstract

The trillions of microorganisms inhabiting the human gut are intricately linked to human health. While specific microbes have been associated with diseases, microbial abundance alone cannot reveal the molecular mechanisms involved. One such important mechanism is the biosynthesis of functional metabolites. Here, we develop a biosynthetic enzyme-guided disease correlation approach to uncover microbial functional metabolites linked to disease. Applying this approach, we negatively correlate the expression of gut microbial sulfonolipid (SoL) biosynthetic enzymes to inflammatory bowel disease (IBD). Targeted chemoinformatics and metabolomics then confirm that SoL abundance is significantly decreased in IBD patient data and samples. In a mouse model of IBD, we further validate that SoL abundance is decreased while inflammation is increased in diseased mice. We show that SoLs consistently contribute to the immunoregulatory activity of different SoL-producing human microbes. We further reveal that sulfobacins A and B, representative SoLs, act on Toll-like receptor 4 (TLR4) and block lipopolysaccharide (LPS) binding, suppressing both LPS-induced inflammation and macrophage M1 polarization. Together, these results suggest that SoLs mediate a protective effect against IBD through TLR4 signaling and showcase a widely applicable biosynthetic enzyme-guided disease correlation approach to directly link the biosynthesis of gut microbial functional metabolites to human health.

Microbiome and metabolome features in inflammatory bowel disease via multi-omics integration analyses across cohorts

Article Open access 06 November 2023

Identification of gut microbial species linked with disease variability in a widely used mouse model of colitis

Article Open access 01 April 2022

Colitis-associated intestinal microbiota regulates brain glycine and host behavior in mice

Article Open access 29 September 2022

Introduction

The human gut microbiome, composed of trillions of microorganisms, is intricately linked to human health¹. At the species abundance level, numerous human microbes have been rigorously correlated with disease phenotypes, however, the mechanisms by which these microbes influence human health remain largely unkown^2,3,4. One important mechanism is through the biosynthesis of functional metabolites by microbes^5,6,7,8,9. Microbial functional metabolites are in direct contact and constant exchange with human cells, granting them inherent biological activity in complex host-microbe interactions^5,6,8. Accordingly, human microbiome research has begun moving towards revealing these microbial functional metabolites and their corresponding molecular mechanisms that drive specific disease phenotypes^10,11.

Human microbiota-derived lipids are a prolific class of functional metabolites. While many studies have focused on common microbe-derived lipids, such as short-chain fatty acids (SCFAs) and phospholipids^12,13, there are a significant number of underexplored lipids which may be equally capable of influencing human health¹⁴. One such class of microbe-derived lipids is sulfonolipids (SoLs), unique molecules that bear striking structural similarity to both bacterial and endogenous sphingolipids (SLs) which are known for their role in mediating immune signaling in humans¹⁵. SoL-producing bacteria do not biosynthesize SLs but instead produce SoLs in high abundance^16,17, suggesting that SoLs may replace SLs as functional metabolites with similar but distinctly different functions. In fact, two SoL-producing genera, Alistipes and Odoribacter, have been negatively correlated with the two primary forms of IBD, ulcerative colitis (UC) and Crohn’s disease (CD)^18,19,20,21, with some species shown to ameliorate IBD symptoms^20,21. We have found that sulfobacin A (SoL A), a representative SoL produced by Chryseobacterium gleum F93 DSM 16776, exhibits unusual immunoregulatory activity in vitro by modulating inflammatory cytokine production, especially through suppression of the lipopolysaccharide (LPS)-induced inflammatory response²² which has been reported as a key contributor to the progression of IBD^23,24,25,26. We have also elucidated the biosynthetic pathway of SoLs and shown that this pathway involves critical enzymes that are specifically involved in SoL biosynthesis, allowing it to be distinguished from the SL biosynthetic pathway²². Whether SoLs produced by Alistipes and Odoribacter, two genera negatively associated with IBD^18,19, represent functional metabolites in this negative correlation is unknown. Furthermore, the molecular target(s) of SoLs as a whole class of unique and abundant lipids is also unknown.

To probe the associations between microbial functional metabolites and disease, many studies have used untargeted metabolomics. However, this approach alone faces challenges such as the complexity of the metabolome, the lack of reference databases for identification, trace metabolite amounts, and a high degree of variability between different metabolomes^27,28,29, all of which lead to difficulty in revealing specific gut microbial functional metabolites as drivers of molecular mechanisms in disease. In contrast, disease-related sequencing datasets such as the Human Microbiome Project³⁰, DIABIMMUNE³¹, and numerous others hosted in the NCBI Sequence Read Archive³² are widely available, higher quality, less dimensional, and less variable. This presents an alternative approach to elucidating the potential connection between SoLs and disease. Thus, we developed an approach that leverages the critical SoL biosynthetic enzymes to examine their differential prevalence and expression in disease, allowing genomic and transcriptomic-level correlations between the biosynthesis of SoLs by human microbes and disease. Informed by this biosynthetic enzyme-guided disease correlation, we can further examine the abundance of SoLs in human disease-related metabolomics datasets to validate these genomic and transcriptomic-level associations at the metabolomic level.

In this work, we used this informatics-based, functional metabolites-focused approach to reveal a negative correlation between SoL biosynthesis and human IBD pathogenesis, followed by targeted chemoinformatic and metabolomic analysis. We then experimentally validated this multi-level correlation using a mouse model of IBD. Through bioactive molecular networking, we determined that SoLs consistently contribute to the immunoregulatory activity of SoL-producing human gut commensals. Using cell-based assays, we also revealed that SoLs primarily mediate their immunomodulatory activity through interaction with TLR4. Specifically, SoLs binds directly to TLR4 via the accessory protein myeloid differentiation factor 2 (MD-2) and displace LPS from MD-2 at higher concentrations, leading to suppression of TLR4 signaling pathways and macrophage M1 polarization. Together, these results demonstrate our biosynthetic enzyme-guided disease correlation approach to uncovering the chemical basis and molecular mechanisms of intricate host-microbe interactions and outline a potential mechanism by which gut microbial SoLs exert a protective effect against IBD progression through disruption of LPS-mediated TLR4 signaling.

Results

Biosynthetic enzyme-guided disease correlation analysis reveals a negative correlation between SoL biosynthesis and IBD

We began by systematically investigating the biosynthetic potential of SoLs from 285,835 human gut bacterial reference genomes including single amplified genomes (SAGs) and metagenome-assembled genomes (MAGs)³³. Based on sequence homology with experimentally verified SoL biosynthetic enzymes^22,34,35,36 (Supplementary Fig. 1, Supplementary Data 1), we identified a total of 562,214 homologous enzyme sequences, including 469,012 cysteate synthases (CYS), 33,486 cysteate fatty acyltransferases (CFAT), and 59,716 short-chain dehydrogenases/reductases (SDR) (Supplementary Fig. 2a). Uncovering phylogenetic trends, we found that these three enzymes were widely distributed in 255,572 genomes (Supplementary Fig. 2a) across 21 phyla, with the majority belonging to Bacteroidota and Firmicutes_A (Supplementary Fig. 2b). A subset of 6.21% (15,863/255,572) of the genomes was found to encode all three putative SoL biosynthetic enzymes (Supplementary Fig. 2a). To prioritize them for further analysis, we filtered the homologs on the basis of three rules: (1) the homology of both CFAT and CYS must equal or exceed 50% sequence similarity with experimentally validated CFATs and CYSs (Supplementary Data 1), as these enzymes are the first two specific enzymes in the biosynthetic pathway of SoLs that distinguish the biosynthesis between SoLs and SLs^22,34,35,36; (2) the homologous regions of CYS, CFAT, and SDR should include protein domains with Pfam IDs PF00291, PF00155, and PF00106, respectively (hit score >50); (3) a set of homologous enzymes, especially SDR enzymes that show variable sequence similarities, should come from the same genome encoding all three enzymes as all three are required for SoL biosynthesis, thus ensuring co-occurrence. Applying these rules, we prioritized 9,731 CYS (1384 unique sequences), 9740 CFAT (917 unique sequences), and 10,319 SDR enzymes (1,076 unique sequences) (Fig. 1a, Supplementary Data 2) from 9,633 bacterial genomes. The prioritized enzymes were distributed among 42 species from Bacteroidota (99.99%, 9632/9633, 95% confidence interval: 99.94% ~ 100%) and one species from Firmicutes_A (0.01%, 1/9633, 95% confidence interval: 0.0018% ~ 0.059%) (Fig. 1b, Supplementary Data 3). Of note, among the 42 species from Bacteroidota, 71% (30/42) of them belong to bacterial families that have been previously reported to produce SoLs (Fig. 1b) including Rikenellaceae (containing genera Alistipes and Alistipes_A)^16,37, Marinifilaceae (containing genus Odoribacter)¹⁶, and Weeksellaceae (containing genus Chryseobacterium B)^22,38.

**Fig. 1: The presence and expression profiles of SoL biosynthetic enzymes and the production of SoLs differ in IBD subjects versus healthy controls.**

To determine whether there is a link between gut microbial capacity to produce SoLs and IBD incidence, we conducted a comparative analysis of metagenomic and metatranscriptomic data obtained from the Inflammatory Bowel Disease Multi’omics Database (IBDMDB)^19,39. We began by generating sequence similarity networks with a 90% sequence identity threshold to group enzymes with similar functions. Consequently, we categorized the prioritized biosynthetic enzymes into 214 subfamilies (79 CYS subfamilies; 25 CFAT subfamilies; and 110 SDR subfamilies) for the subsequent analyses (Supplementary Data 4). Looking for the presence of the prioritized 214 subfamilies in IBD cohorts, we identified 154 subfamilies in 667 metagenome samples (182 healthy samples and 485 IBD disease samples), of which 116 subfamilies were detected in ≥5% of samples (Supplementary Fig. 2c). Beta diversity of the presence of these 116 subfamily biosynthetic enzymes indicated that the overall composition of SoL biosynthetic enzyme subfamilies was significantly different between the healthy and IBD cohorts (Fig. 1c, Jaccard distance, PERMANOVA p = 0.001). Of note, 57 subfamilies had a significantly higher prevalence (two-sided Fisher’s exact test p < 0.05) in healthy individuals as compared to IBD cases (Supplementary Data 5), among which 35 subfamilies (18 CYS subfamilies, 2 CFAT subfamilies, and 15 SDR subfamilies) further show a difference of prevalence > 10% (Fig. 1d).

To further examine the difference between the expression profiles of SoL biosynthetic enzymes between the IBD and healthy groups, we extended our comparative analysis to the metatranscriptomic level. We found that 132 SoL biosynthetic enzyme subfamilies were expressed in 777 metatranscriptomic samples (193 healthy samples and 584 IBD disease samples), with about 42% (55/132) detected in at least 5% of samples (Supplementary Fig. 2d). Beta diversity of the expression profiles of SoL biosynthetic enzymes suggested that the overall expression of these enzyme subfamilies is significantly different between the healthy and IBD cohorts (Fig. 1e, Bray-Curtis distance, PERMANOVA p = 0.001). To capture more detail, we compared the prevalence and abundance differences of each enzyme subfamily in the metatranscriptomic samples. Nine subfamilies had higher prevalence (two-sided Fisher’s exact test p < 0.05, varying from 9% ~ 17%) in the non-IBD group than the IBD group (Supplementary Data 6). We further identified 8 subfamilies (6 CYS, 1 CFAT, and 1 SDR) as significantly different in abundance (expression) profiles between the healthy controls and IBD cases (Fig. 1f, two-sided Mann-Whitney U test, adjusted p < 0.05). Notably, 7 of the 8 subfamilies had a higher prevalence (Fig. 1f, upper panel, two-sided Fisher’s exact test p < 0.05) and a higher abundance (Fig. 1f, lower panel, one-sided Mann-Whitney U test, adjusted p < 0.05) in the non-IBD group than in the IBD group.

We finally looked for metabolomic evidence in the differences of detectable SoLs, the products of the biosynthetic enzymes mentioned above, among IBD and non-IBD groups from publicly accessible metabolomics datasets. We expected that the increased expression of SoL biosynthetic enzymes would correspond with an increased abundance of stool SoLs in non-IBD groups compared to IBD groups after possible uptake by the host. Using metabolomics data from two independent datasets (dataset 1: IBDMDB^19,39, corresponding to the same dataset used for metagenomics and metatranscriptomics analysis; dataset 2: PRISM⁴⁰), we identified metabolite features putatively corresponding to specific sulfonolipids^16,41 (Supplementary Data 7), by exact mass comparison with mass error less than 5 ppm (Supplementary Data 8). Within each dataset individually, we indeed found that metabolomic features potentially corresponding to SoLs were decreased in stool samples of IBD groups compared to non-IBD groups. Since there were no MS/MS data available in either dataset, we utilized additional complementary approaches to confirm these features as SoLs including analysis of in-source fragmentation, correlation of co-eluting metabolomic features, and retention time matching between the dataset and data recreated using our own instrument, followed by experimental validation using targeted metabolomics with an additional set of samples from independent IBD disease case and control cohorts.

To validate the presence of SoLs in these datasets, we first tried to identify SoLs using in-source fragments (ISFs) of metabolites based on an established set of criteria⁴². We initially identified six groups of co-eluting metabolomic features as potential ISFs which showed peak-to-peak intensities highly correlated with putative SoL features (Supplementary Fig. 3, Pearson correlation coefficients ≥0.9). We then examined the reference MS/MS spectra of our isolated and literature-reported SoLs^22,41 matching putative SoL masses, which contained limited m/z values corresponding to potential ISFs we identified. However, their relatively low intensity was not conclusive enough to classify them as high-confidence ISFs⁴². Thus, we proceeded with a complementary correlational approach to identify the putative SoL features. Among the originally identified six groups of co-eluting metabolomic features, five members were detected in both datasets mentioned above (Supplementary Fig. 3c–f, features highlighted in bold). Based on exact mass matching, these features corresponded to SoL analogs: SL 34:1;2 O, SL 17:0;O/16:1;O, SL 33:1;2 O | SL 17:0;O/16:1;O, SL 34:1;2 O | SL 17:0;O/17:1;O, and SL 32:0;O | SL 17:0;O/15:0⁴³. In addition, these features had higher peak area-peak area correlation with each other (Pearson correlation ≥0.9; Supplementary Fig. 3c–f). Notably, SoLs are often detected in metabolomics as a series of analogs with consecutive additions of CH₂ and H₂ moieties within the class and with different numbers of oxygens between classes^16,43. Thus, these features likely represent a series of analogs chemically modified from a common parent metabolite, or co-produced by a specific microbe, which is consistent with SoL analogs. These metabolomic features were further positively correlated with species of the prolific SoL-producing genus Alistipes: A. putredinis, A. finegoldii, A. indistinctus, A. shahii, A. onderdonkii, and Bacteroidales bacterium ph8 (which belongs to A. obesi) with Spearman correlation coefficients ≥ 0.5 (Supplementary Figs. 4–6). This positive correlation indicated that the abundance of these metabolites increased with the increase in these species, supporting that these species likely produced these molecules. Furthermore, these metabolomic features had significantly higher abundance in non-IBD groups compared to IBD groups in both IBD datasets (Fig. 2a, Supplementary Fig. 7, Wilcoxon rank sum test, p < 0.05, one-sided), which was consistent with our exact mass matching analysis.

**Fig. 2: SoL abundances are decreased in IBD metabolomics datasets and independent IBD patient samples.**

To further validate our identification of SoLs in these datasets, we acquired one of the columns used to generate the original data^19,40. We then selected several standard compounds used in dataset 2 and the candidate SoL B feature that we identified by exact mass matching, and subsequently analyzed the retention times of the standard compounds alongside our own standard SoL B using our in-house HPLC-MS instrument. Due to the inherent variability of retention time between instruments^44,45, we calculated the relative retention time (RRT)^46,47 using each of the standard’s retention time relative to that of our SoL B standard and compared these values to the RRTs calculated using the corresponding dataset standards and candidate SoL B. We found that the RRT values using our SoL B standard and the RRT values using the candidate SoL B shared a linear relationship (Supplementary Fig. 8, R2 = 0.9915), indicating that the shift in retention time was linear and thus suggesting that the candidate SoL B feature was SoL B.

Finally, to experimentally validate our informatic analysis, we obtained deidentified stool samples collected from an independent cohort of IBD patients (n = 40) and healthy controls (n = 40), and analyzed their SoL abundance by targeted HPLC-MS/MS. We detected SoLs B, C, and F as major SoLs and found that all their abundances were decreased in IBD samples compared to non-IBD samples (Fig. 2b), with SoL B and F being significantly decreased (Wilcoxon rank sum test, p < 0.05, one-sided). This independent validation was consistent with our bioinformatic analysis which also showed that major SoLs including SoL B were significantly decreased in IBD metabolomes and further supported our identification of SoLs in the metabolomics datasets as well as our chemoinformatic analysis showing decreased abundance of SoLs in stool samples of IBD.

Thus, our metagenomic analysis reflected that SoL biosynthetic enzymes were more prevalent in the non-IBD group than the IBD group, metatranscriptomics suggested that genes encoding these enzymes are more actively transcribed in the non-IBD group, and chemoinformatics and metabolomics indicated that representative SoLs are in higher abundance in stool samples from the non-IBD group. We further validated the metabolomics data in an independent cohort of IBD patient samples which showed that SoL abundance was indeed significantly decreased in IBD compared to non-IBD samples. Altogether, our findings establish a negative correlation directly between SoL biosynthesis and IBD, consistent with the previously reported negative association between SoL-producers, namely Alistipes and Odoribacter, and IBD^22,23.

An experimental model of colitis demonstrates the negative correlation between SoL production and IBD progression

Encouraged by our informatically predicted negative correlation between SoL biosynthesis and IBD, we sought to further experimentally validate our prediction using a well-established mouse model of IBD. We used Il10-deficient (Il10^–/–) mice that are genetically susceptible to developing intestinal inflammation and chronically treated them with the non-steroidal anti-inflammatory drug piroxicam, which induces the development of colitis through the disruption of the gut mucosal barrier in inflammation-susceptible hosts^48,49. We selected this model due to its stability, as Il10^–/– mice generally will not develop colitis when born and raised under specific pathogen-free conditions unless induced by external stimuli such as piroxicam treatment. This allowed us to more confidently ensure that the effects observed were dependent on the induction of colitis and not due to the Il10 deficiency. Stimulation of mucosal TLRs stemming from mucosal barrier breakdown was another factor in our selection of this model, as we have previously shown that SoL A suppresses LPS-induced inflammation and LPS is well-known to activate TLR signaling^22,50. As has been previously reported^48,49, we observed that the colonic tissues were inflamed in the piroxicam-treated (IBD) group of Il10^–/– mice when compared to the control (pre-IBD) group as indicated by gross pathology and blinded histopathology analyses (Fig. 3a–c, Supplementary Data 9 and 10).

**Fig. 3: SoLs are decreased in a mouse model of colitis concurrent with increased expression of inflammatory markers.**

To explore the link between SoL production and IBD, we collected fecal material from piroxicam-treated Il10^–/– (IBD) mice (n = 7, female) and pre-IBD control Il10^–/– mice (n = 4, female), extracted metabolites, and measured the abundance of SoLs by targeted metabolomics using high-performance liquid chromatography (HPLC)-high resolution mass spectrometry (HRMS) (Fig. 3d). We detected metabolites with m/z corresponding to major SoLs, specifically SoLs A (SL 34:0;2 O | SL 17:0;O/17:0;O) and B (SL 32:0;O | SL 17:0;O/15:0), in all fecal samples tested and unambiguously determined their identities by HPLC-MS/MS (Fig. 3e, Supplementary Fig. 9a, b). We then determined that the abundances of both SoLs A and B were significantly decreased in feces from piroxicam-treated mice compared to control (Fig. 3f, g). This result confirms our above-described informatic analysis and directly establishes a negative correlation between SoL production and colitis progression in the mouse model. In addition, we also observed significantly increased expression of the NF-κB-regulated inflammatory markers TNFα, NOS2, IL-6, and IL-1β in the IBD mouse group (Mann-Whitney U test, p ≤ 0.005; Fig. 3h–k), further indicating a negative correlation between SoL production and these inflammatory markers. Given our previously observed anti-inflammatory activity of SoL A against LPS²², a natural ligand of TLR4, this negative correlation suggests a potential role of SoLs in regulating IBD that may involve suppressing TLR4-mediated NF-κB activation. To exclude any differences caused by sex, we performed another independent study with male mice using the same model and observed the same negative correlation between SoL production and IBD progression (Supplementary Fig. 10).

Constant identification of SoLs’ contribution to immunomodulatory activity

We next examined the production of SoLs and their contribution to immunomodulatory activity in different human gut commensals. Unlike C. gleum F93 DSM 16776, which we experimentally investigated for its functional metabolites in relation to inflammatory activity²², the prolific SoL-producers Alistipes and Odoribacter had not yet been thoroughly chemically investigated to identify the biologically active components associated with remediation of IBD. In addition, Alistipes and Odoribacter produce a mixture of other SoLs¹⁶ and are likely to produce a multitude of other functional metabolites, both of which may complicate the potential immunomodulatory activity of these genera’s metabolites with respect to their bioinformatically predicted negative association with IBD. Thus, we conducted bioactive molecular networking of three Alistipes and two Odoribacter strains (Supplementary Data 11) to identify the constant contributor(s) to biological activity. We fractionated crude extracts of the Alistipes and Odoribacter strains and determined the biological activity of each fraction using a cell-based assay that measured the suppression of LPS-indued TNFα production (Fig. 4a). We simultaneously analyzed each fraction by untargeted HPLC-HRMS/MS to generate molecular networks using the Global Natural Products Social (GNPS) feature-based molecular networking (FBMN) pipeline⁵¹. We then correlated the relative expression of TNFα in each fraction with the relative peak area of molecular features across all fractions to generate a bioactivity score reflecting the contribution of specific features to the activity of the fractions. Bioactivity scores and relative peak areas were then mapped onto the molecular network to visualize these contributions. The resulting bioactive molecular network generated from Alistipes timonensis DSM 27924 is presented in Fig. 4b. The SoL-containing cluster contained the most abundant and most active molecular features, as indicated by the node size and color intensity compared to other clusters in the network. Additionally, this cluster contained several known SoLs but many more unannotated SoLs, suggesting that the family of biologically active SoLs is larger than what is currently known. In all other SoL-producers tested, we consistently identified SoLs as a major contributor in the active fractions of each strain (Supplementary Fig. 11). To exclude the possibility of observed SoL activity being influenced by LPS contamination, we confirmed the absence of leftover LPS in the SoL samples using a chromogenic LAL assay (Supplementary Fig. 12). Narrowing down the immunosuppressive activity of each of the strains to SoLs guided us to isolate pure SoLs A and B from A. timonensis DSM 27924 (structures confirmed by NMR spectroscopy; Supplementary Tables 1 and 2, Supplementary Figs. 13–24), as well as from each of the other Alistipes and Odoribacter strains tested. We thus reinforced the contribution of this class of lipids to the observed biological activity of Alistipes and Odoribacter.

**Fig. 4: Bioactive molecular networking leads to the identification of SoLs as major bioactive components of a SoL-producer.**

SoL A mediates dual immunomodulatory activity through TLR signaling

While the causes of IBD remain largely unknown, IBD progression has been linked to aberrant TLR signaling⁵⁰. TLRs are pattern recognition receptors (PRRs) that initiate a variety of host processes, especially inflammatory responses, through the recognition of pathogen-associated molecular patterns (PAMPs) and other non-pathogenic microbial factors^50,52,53. Specifically, TLR2 and TLR4 are well-known to recognize PAMPs in the gut microbiome⁵⁰. In addition, their expression is significantly increased in IBD pathogenesis, reflecting a state of aberrant activation^50,52. Thus, we expected that SoLs may interact with TLR4 or TLR2 to mediate their immunomodulatory activity. We treated primary mouse macrophages collected from wild-type C57BL/6 mice with SoL A (as a representative of SoLs) either alone or together with LPS (an agonist of TLR4) or Pam3CSK4 (an agonist of TLR2/1) and measured the expression of three inflammatory cytokines (IL-6, TNFα, and IL-1β). By itself, SoL A exhibited a mild to moderate effect on the expression of pro-inflammatory cytokines compared to control (Fig. 5), generally consistent with our previous finding²². As expected, the TLR ligands, LPS and Pam3CSK4, both showed significant induction of all three cytokines compared to control (Student’s t-test, p ≤ 0.0001) (Fig. 5). Notably, SoL A was found to significantly suppress the expression of all three cytokines induced by LPS (p ≤ 0.0001) (Fig. 5). Together, SoL A’s mild pro-inflammatory activity by itself and strong inhibition against LPS-induced inflammation constitute its dual immunomodulatory activity. SoL A also inhibited Pam3CSK4-induced IL-6 and TNFα to a smaller extent while increasing IL-1β expression induced by Pam3CSK4 (p ≤ 0.05) (Fig. 5). This result indicates that SoL A primarily affects LPS-induced inflammation and implies that interaction with TLR4 may be involved in SoL A’s mechanism of action. Interestingly, SoL A’s partial suppression of Pam3CSK4-induced inflammation suggests that SoL A-related anti-inflammatory activity may also extend to the TLR2/1 pathway and warrants further investigation. After identifying that SoL A’s primary effect is through TLR4, we further examined the biological activity of SoL B against LPS-induced inflammation. We found that SoL B also inhibited LPS-induced inflammation albeit to a lesser extent than SoL A (Supplementary Fig. 25). This activity is consistent with previous reports that SoL B exhibits anti-inflammatory activity both in vitro and in vivo in mouse models of acute inflammation⁵⁴.

**Fig. 5: SoL A primarily suppresses LPS-induced TLR4 activation.**

Molecular docking and ELISA displacement assay suggest SoLs binding to TLR4/MD-2 complex

LPS stimulation of TLR4 occurs through a series of interactions ultimately resulting in LPS binding to MD-2, which forms a complex with TLR4 and induces dimerization to initiate signaling^55,56,57. The TLR4/MD-2 heterodimer recognizes structurally diverse LPS molecules, giving it the flexibility to detect different LPS-related PAMPs in the human gut microbiome⁵⁷. Interestingly, the TLR4/MD-2 complex was recently found to recognize human sulfatides, sphingolipid derivatives which bear a sulfated saccharide head group and dual acyl chains, presumably mimicking the disaccharide core and multiple acyl chains of LPS⁵⁸. Comparing the chemical structure of SoL A to those of sulfatides and lipid A (the immunogenic portion of LPS) (Fig. 6a), we noted structural similarity in the negatively charged head groups and multiple acyl chains. We thus considered if multiple molecules of SoL A might bind to MD-2 in a similar configuration as sulfatides and lipid A. Inspired by sulfatides that bind in triplicate to MD-2⁵⁸, we used molecular docking to model the binding of three molecules of SoL A to MD-2. Our analysis predicted three molecules of SoL A indeed bind in the hydrophobic pocket of MD-2 (Fig. 6b), where lipid A is known to bind, with a docking score of -8.9 kcal/mol, better than that of lipid A which had a docking score of -6.2 kcal/mol. Additionally, SoL A was predicted to make hydrophobic contacts with several amino acids including I117, F119, I52, and F121 (Supplementary Data 12), all of which are also reported to contact the acyl chains of lipid A⁵⁷. Notably, SoL A is also predicted to contact residues including R264 and R90 (Supplementary Data 12), consistent with contacts between these residues and the phosphate groups of lipid A⁵⁷. This suggests that SoL A may bind directly to the TLR4/MD-2 complex and possibly compete for binding with LPS, allowing it to suppress LPS-induced activation of the TLR4 pathway. A critical aspect of lipid A binding to MD-2 is the exclusion of one acyl chain from the hydrophobic pocket of MD-2 which forms a bridge with TLR4 and is involved in inducing dimerization⁵⁷. Likewise, we observed one acyl chain of SoL A excluded from the hydrophobic pocket in our docking analysis (Fig. 6b), further suggesting that SoL A mimics LPS as a ligand for TLR4. After successfully docking SoL A we then tested SoL B which lacks an extra hydroxy group, (Fig. 6a) potentially increasing its interactions with the hydrophobic binding pocket of MD-2. Our docking analysis indeed showed that SoL B also binds to MD-2 (Fig. 6b), with similar contacts as SoL A (Supplementary Data 12) but higher affinity (docking score of −9.6 kcal/mol) as predicted.

**Fig. 6: SoLs bear structural similarity to both lipid A and sulfatide and bind with MD-2 to block LPS binding.**

To experimentally determine if SoLs A and B bind to MD-2 and to what extent the SoLs compete with LPS for binding to MD-2, we conducted an ELISA-based displacement assay. Taking advantage of biotinylated LPS, which retains the activity of unconjugated LPS⁵⁹, we measured absorbance generated by an HRP-linked streptavidin probe to measure the relative amount of MD-2 which was bound with biotin-LPS as opposed to MD-2 bound with SoL A or B. We administered 0.1, 1.0, and 10 μM concentrations of SoL A or B and 1 ng/mL biotin-LPS to MD-2 in three sequences: 1) SoL first followed by LPS 1 h later, 2) LPS first followed by SoL 1 h later, and 3) both SoL A or B and LPS at the same time. After 1 h of incubation, we found that at all concentrations, when SoL A or B was added first, there was a marked decrease in percent absorbance as compared to when LPS was added first and when the two compounds were added together (Fig. 6c). This suggests that SoLs A and B both bind and occupy some sites of MD-2, preventing LPS from fully binding when it is added 1 h after SoL A or B. Furthermore, when moving from low to high concentrations of SoL A or B, we observed that the percent absorbance decreased dramatically. This indicates that with increasing concentration of SoL A or B, less LPS binds to MD-2, implying that SoLs indeed compete with LPS for binding to MD-2. Taken together, these results indicate that SoLs A and B can bind directly to MD-2 and more importantly compete with LPS for binding to this target, thus providing a potential molecular mechanism underlying SoL A’s pro-inflammatory activity by itself as well as its strong activity in suppressing LPS-induced inflammation which likely also expands to other members of the SoL family.

SoLs suppress LPS-induced TLR4 signaling to regulate macrophage polarization

Upon LPS binding, TLR4 initiates downstream signaling, such as through the NF-kB and MAPK pathways, resulting in the induction of inflammatory cytokine expression⁵². If a SoL binds to MD-2, it will block LPS-induced activation of the TLR4 pathway. Therefore, we investigated whether the addition of SoL A or B affected the phosphorylation of TLR4-downsteam signaling molecules, ERK1/2, and p38, or the degradation of IκBα, which are all critical for LPS-induced cytokine expression⁵² (Fig. 7a). We treated macrophages with LPS in the presence of increasing concentrations of SoL A or B (from 0 to 20 μM), then performed western blot analysis to examine the TLR4-downstream signaling pathways. We found that both SoLs reduced LPS-induced phosphorylation of p38 and ERK1/2 in a concentration-dependent manner. At the concentration of 20 μM, SoLs A and B almost completely blocked LPS-induced phosphorylation of p38 and ERK1/2 (Fig. 7b). Western blot also showed that SoLs concentration-dependently suppressed LPS-induced IκBα degradation (Fig. 7b). These results support that SoLs exert their anti-inflammatory effect by blocking LPS-mediated phosphorylation of downstream TLR4 proteins, effectively negating LPS activation of the TLR4 pathway. Also, SoL A or B alone at the concentration of 20 μM slightly enhanced the phosphorylation of certain signaling molecules (e.g., ERK1/2) compared to control, consistent with our observations that SoLs alone induced a mild pro-inflammatory effect on cytokine expression (Fig. 5) and supporting the dual immunomodulatory activity of SoLs which provides further opportunities to regulate homeostatic immune responses.

**Fig. 7: SoLs suppress LPS-induced activation of TLR4 signaling pathway and macrophage M1 polarization.**

Because TLR4 signaling leads to macrophage polarization which has been shown to contribute to IBD^60,61,62, we also examined the effects of SoL A on macrophage polarization. We treated THP-1 monocytes with IFN-γ and LPS to induce M1 polarization or IL-4 and IL-13 to induce M2 polarization. Successful induction of M1 and M2 polarization was confirmed by morphology changes and subsequent RT-qPCR quantification of cytokine profiles. When 10 μM of SoL A was added alongside the respective inducing agents, our relative cytokine expression results showed that SoL A significantly reduced the production of M1-polarized macrophage markers IL-6, CXCL10, IL-12β, and TNFα, compared to macrophages treated without SoL A (Fig. 7c) but had a mostly non-significant effect on M2 polarization (Fig. 7d). This suggested that SoL A suppresses macrophage M1 polarization, which supports our aforementioned result that SoLs interfere with TLR4 signaling potentially leading to inhibition of TLR4-mediated IBD.

Discussion

The results of this study have described two major points which are summarized here and discussed in detail below. First, we have leveraged our biosynthetic enzyme-guided disease correlation approach to directly connect the biosynthesis and production of a class of abundant yet underexplored human microbial metabolites, SoLs, to IBD, an existing human health condition with complex and poorly understood etiology, followed by an independent IBD patient cohort and mouse model of IBD to validate this informatically predicted negative correlation. Second, we have revealed that SoLs A and B, two representative gut microbial SoLs, modulate host immune responses through the TLR4/MD-2 complex and inhibit LPS-induced TLR4 activation and macrophage M1 polarization, which provides a mechanistic explanation for SoLs’ potential protective activity against IBD.

Through both bioinformatic and chemoinformatic analyses, we revealed that the expression of SoL biosynthetic enzymes and abundance of SoLs in the gut metabolome are negatively correlated with IBD status in humans. Our IBD model with male and female piroxicam-treated Il10^–/– mice supported this sex-independent negative correlation with a concurrent negative association between SoL production and TLR-4-related inflammatory markers TNFα, NOS2, IL-6, and IL-1β. Our findings were consistent with literature reports which have shown that the bacterial genera of abundant SoL production, Alistipes and Odoribacter, are associated with the remediation of IBD symptoms^20,21. Considering that these SoL-producers are commensal members of the human gut microbiome^16,63, their constant production of SoLs in the human gut may help to maintain intestinal immune homeostasis, thereby preventing IBD. Both Alistipes and Odoribacter are also known to produce SCFAs, which have been implicated in reducing intestinal inflammation^{12,21,63,64,65}. Besides SCFAs, there are other gut microbial functional metabolites such as secondary bile acids and indole derivatives which have also been linked to modulating host inflammation and immunity^{66,67,68,69,70}. Thus, the role SoLs play individually and/or synergistically with other factors in IBD pathogenesis remains interesting and awaits further investigation.

Toward understanding the mechanism of the immunomodulatory activity of SoLs, we showed that the representative SoLs A and B primarily target the TLR4 pathway and that they both block LPS binding to MD-2 to suppress TLR4 signaling. Our analysis indicated that SoL A preferentially suppresses LPS-induced inflammation as compared to Pam3CSK4-induced inflammation, suggesting that TLR4 activation is more strongly affected by SoL A than TLR2/1 activation. The selectivity for TLR4 may stem from SoL A’s structural similarity to LPS, a hypothesis supported by the recently reported human TLR4 ligands, sulfatides, which share highly similar structural features to SoLs⁵⁸. As microbial functional metabolites, it is reasonable that SoLs are directly recognized by TLR4, which has evolved specifically to recognize PAMPs⁷¹. Our molecular docking also suggested SoLs A and B’s recognition by TLR4, indicating that three molecules of either SoL indeed bind with stronger predicted binding score compared to lipid A and make contacts with important amino acids in the pocket of MD-2, consistent with lipid A binding as well as structurally-related sulfatides^57,58. Our ELISA-based displacement analysis then confirmed that both SoLs A and B directly bind to MD-2, block LPS binding when added prior to LPS, and could displace bound LPS from MD-2 at higher concentrations. We further demonstrated that increasing concentration of SoLs A and B suppressed LPS-induced TLR4 signaling pathways in a dose-dependent manner and SoL A significantly suppressed M1 polarization of macrophages, further indicating their capacity to reduce downstream TLR4 signaling responses. Notably, increased numbers of M1 macrophages is a characteristic feature of IBD⁶² and this suppressive effect may represent one explanation of how SoL-producing bacteria are able to remediate symptoms of IBD. Taken together, our results represent the first report of SoLs A and B’s binding to MD-2 and establish that SoLs are likely to mediate their dual immunomodulatory activity by occupying the hydrophobic binding pocket of MD-2 (pro-inflammatory by SoLs alone) but primarily by blocking LPS binding to the TLR4/MD-2 complex (anti-inflammatory against LPS-induced inflammation). This discovery suggests SoLs’ mechanistic role in regulating a multitude of TLR4-related inflammatory conditions, most notably IBD which is associated with dysregulation of TLRs, especially TLR4^23,24,25,26, leading to aberrant macrophage activation^62,72.

While many studies have focused on the association of certain microbes with specific diseases by analyzing the abundance and distribution of microbial strains^{18,19,73,74,75,76}, we describe here an approach which leverages disease-related sequencing data to connect the biosynthesis of functional metabolites directly to human health conditions. Our approach takes advantage of critical biosynthetic enzymes specifically required for the biosynthesis of their corresponding functional metabolites. Increased or decreased expression of these enzymes, reflecting the differential production of a specific functional metabolite, can be correlated with different disease states to reveal positive or negative associations. Through these correlations, our biosynthetic enzyme-guided disease correlation approach can reveal trends in functional metabolite biosynthesis in the context of human health conditions, enabling a more focused, targeted chemoinformatic analysis to rapidly fish out metabolites of interest and further confirm their association with disease. By applying this approach, we have identified a negative relationship between SoL biosynthesis and IBD directly from patients’ data followed by verification using an IBD mouse model, which has guided us to further reveal a molecular target and a potential mechanistic explanation for the protective effect that SoLs and SoL-producing bacteria exert against IBD. We are now further characterizing the biological role and therapeutic potential of SoLs through elucidating the mechanism of SoL delivery and uptake in the host and probing their effects on IBD pathogenesis. Our future studies include colonization of SoL-producers in the Il10^–/– mouse model, determining whether SoLs are gut-restricted microbial metabolites or not, and exploring the role of microbial delivery systems such as outer membrane vesicles. Furthermore, with the increasing availability of disease-related sequencing data^32,77 and the rapidly advancing investigation of microbial biosynthetic pathways^78,79,80, we envision that our approach can be broadly applied to uncover previously unknown human microbial metabolites with potentially important human health implications and facilitate the investigation of their complex effects on human health.

Methods

Identification of SoL biosynthetic enzymes from human gut bacterial reference genomes

We collected experimentally validated enzymes involved in SoL biosynthesis as reference amino acid sequences of CYS, CFAT, and SDR (Supplementary Data 1). Reference amino acid sequences were used as seed sequences to search for homologs in the human gut bacterial reference genomes using the DIAMOND blastp model⁸¹ with an e-value threshold of 10⁻⁵. We then investigated the taxonomic distribution of the resulting homolog sets based on the taxonomy annotation for each genome³³. To prioritize SoL biosynthetic enzymes, the homologs of CFATs and CYSs from genomes which encode copies of CFAT, CYS, and SDR were first used to generate sequence similarity networks with experimentally validated CFAT and CYS at a threshold of 50% similarity using MMseqs2⁸². The prioritized CFATs and CYSs meet co-occurrence with SDRs in the same genome. The filtered enzymes were further subjected to Pfam domain analysis by hmmsearch (HMMER v3.3) with default parameters against the Pfam-A database (v33.1). Enzymes containing corresponding Pfam domains with hit score >50 were selected as prioritized SoL biosynthetic enzymes. Prioritized enzymes were used to generate CYS, CFAT, and SDR subfamilies using sequence similarity networks with at least 90% similarity by MMseqs2 clustering⁸². Maximum-likelihood trees were generated using the representative genome (Supplementary Data 3) of each species by GTDB-Tk (v2)⁸³ with the following parameters (refer to the GTDB-Tk user guide: https://ecogenomics.github.io/GTDBTk/commands/index.html): 1. GTDB-Tk reference data release 207 was used. 2. Parameters of classify: gtdbtk classify_wf --genome_dir --out_dir --force. 3. Parameters of infer tree from multiple sequence alignment: gtdbtk infer --msa_file MSA_FILE --out_dir OUT_DIR. 4. Using the convert_to_itol command to make the tree suitable for visualization in iTOL: gtdbtk convert_to_itol --input_tree --output_tree. Finally, we annotate their phylogenetic trends using iTOL⁸⁴.

Quantification of enzyme abundances in metagenomic and metatranscriptomic samples

Metagenomic and metatranscriptomic whole-genome sequencing datasets of human gut microbiomes related to IBD were downloaded from the Inflammatory Bowel Disease Multi’omics Database^19,39 (IBDMDB, https://ibdmdb.org/tunnel/public/summary.html) For both metagenomic and metatranscriptomic samples, reads were quality filtered and adapter removed using bbduk.sh with the following parameters: qtrim=rl ktrim=r mink=11 trimq=10 minlen=40 (read quality cutoff is 10, read length cutoff is 40). High-quality reads were mapped to the nucleotide sequences of corresponding contigs containing the SoL biosynthetic genes using the BWA mem algorithm⁸⁵ with default parameters. The reads mapped to genes were counted by featurecounts⁸⁶ with the following parameters: -f -t CDS -M -O -g transcript_id -F GTF -s 0 -p --fracOverlap 0.25 -Q 10 –primary. Enzymes encoded or expressed in at least 5% of samples were considered as common distribution in humans and were included in the comparative analysis. Transcripts per million (TPM) were calculated for each SoL biosynthetic enzyme-encoding gene. The abundance of each subfamily was calculated by the sum of the relative abundances of all genes in the subfamily. Beta diversity was performed to quantify the prevalence and relative abundance differences in the overall composition of SoL biosynthetic enzymes between the IBD and the control groups. PERMANOVA was performed to show the encoding and expression profile differences of SoL biosynthetic enzymes between IBD and control groups. Both beta diversity and PERMANOVA were performed using the R package vegan^87,88. To explore the differences between IBD and control groups of single SoL biosynthetic enzyme subfamilies, we used the Shapiro–Wilk test to evaluate the normality of a specific gene family’s relative abundance. We then calculated the significance of relative abundance between healthy and IBD individuals using either two-sample Student’s t-test (for normally distributed data) or a two-sample Wilcoxon rank sum test (for not normally distributed data). Significance tests were performed in Python using packages Pandas and SciPy^89,90,91. The Benjamini-Hochberg method was used to adjust p-values to correct for multiple testing⁹². SoL biosynthetic enzyme subfamilies were considered differential if the adjusted p-value was less than 0.05. For differential Sol biosynthetic enzymes, we performed a two-sided Fisher’s exact test to explore their difference in prevalence across IBD and non-IBD groups.

Analysis of publicly available metabolomics datasets

Processed per-subject metabolomic feature tables and microbial species relative abundance profiles were collected from two publicly available IBD datasets: IBDMDB^19,39 (https://ibdmdb.org/downloads/html/rawfiles_MBX_2017-03-01.html) and PRISM⁴⁰ (https://www.nature.com/articles/s41564-018-0306-4#Sec31). Known SoL-related metabolites were collected from MS-DIAL4⁴¹ and reference¹⁶ (Supplementary Data 7). Search parameters were set to the exact mass of SoLs using a 5-ppm match tolerance for parent ions. Identified metabolomic features were considered SoL candidates. Next, features co-eluting with SoL candidates (within ± 0.05 min retention time window) were selected to build a feature similarity network (Pearson correlation). Relative abundance was normalized using the equation below following the calculation described in dataset 2:

$${{{\rm{Metabolomic}}}}\; {{{\rm{feature}}}}\; {{{\rm{abundance}}}}=\frac{{Ni}}{{{{\rm{Sum}}}}(Ni)}*{10}^{6}$$

(1)

Here, Ni represents the absolute intensity of a metabolomic feature, Sum(Ni) represents the total absolute intensity of all features in the sample. In this step, all the metabolomic samples were included (dataset 1: 546 samples; dataset 2: 220 samples). The correlation coefficients ≥0.9 between SoL candidates and other metabolomic features were subjected to further analysis hereafter. If the co-eluting features could be identified in SoL candidates’ MS/MS spectra in MS-DIAL4⁴¹ and a previous in-depth study¹⁶, these features were considered SoL candidates’ in-source fragments. If not, these metabolites might be co-produced by a specific microbe, or chemically modified from a common parent metabolite, and these metabolites were considered as SoL analog candidates. Of note, each data type, detected using the same LCMS method^19,39,40, was analyzed separately in each cohort.

To further examine whether SoL candidates correlated with the abundance of any taxa, the Spearman correlation between species-level relative abundances and SoL candidates’ relative abundance was calculated. To do so, a series of SoL analog candidates (SoL candidates correlated with other SoL candidates) were selected for further analysis. In total, 472 samples collected from dataset 1 and 220 samples collected from dataset 2 were included. Of note, corresponding microbial species’ relative abundance profiles and metabolomic features data were paired in these samples. Species that were present in less than 5% of samples were excluded. Finally, the relative abundance of SoL analogs was used to calculate differential abundances between non-IBD and IBD samples. The one-sided Wilcoxon rank sum test was used to measure statistical significance with the hypothesis that the abundance of SoL was higher in the non-IBD group than in UC or CD.

Animals

B6.129P2-Il10^tm1Cgn/J (Il-10 KO; Il10^–/–) mice from the C57BL/6 background were originally purchased from The Jackson Laboratory. Mice were bred and maintained in specific pathogen-free conditions at the University of South Carolina and were maintained on a 12-hour light/dark cycle with unlimited access to water and food (Inotiv Teklad, 8604, https://insights.envigo.com/hubfs/resources/data-sheets/8604-datasheet-0915.pdf). All animal protocols were approved by the University of South Carolina Institutional Animal Care and Use Committee.

Piroxicam-accelerated Il10 ^–/– mouse IBD model and analysis

At 8-12 weeks of age, male (n = 5 mice distributed in 2 cages) and female (n = 7 mice distributed in 2 cages) Il10^–/– mice were switched to a normal chow diet (Inotiv Teklad, 8604) supplemented with 100 ppm of piroxicam (Cayman Chemical, 13368; Inotiv Teklad, TD.210442) to induce colitis development as previously described⁴⁹. In parallel to these mice, male (n = 3 in 1 cage) and female (n = 4 in 1 cage) Il10^–/– mice were maintained on the control diet for the duration of the experiment to serve as pre-colitic controls. Mice were euthanized at 18 days to collect tissues for inflammation assessment and intestinal contents for quantification of SoLs. At necropsy, colitis severity was first grossly assessed, which included qualitative evaluations of cecal atrophy (0–5), thickening of cecal (0–5) and colon tissues (0–5), the extent of content loss in the cecum (0–4) and stool consistency/diarrhea (0–3). For histopathology, segments of the colon were first washed in PBS and then fixed in 10% neutral buffered formalin. The tissues were embedded in paraffin, cut into 5-mm sections, and stained with hematoxylin and eosin (H&E) at the Instrumentation Resource Facility at the University of South Carolina School of Medicine. Inflammation scores of colon sections were blindly assessed as previously described⁹³ using an Echo Revolve light microscope and accompanying software. Briefly, colitis severity was assessed based on the following histopathological features: length measurements in microns of crypt hyperplasia converted to a score from 0-4, qualitative assessment of goblet cell loss (0–5), crypt abscesses per 10X field counts converted to a score from 0-4, and qualitative assessment of submucosal edema (0-3). RNA isolations and RT-qPCR were performed as previously described⁹⁴. Briefly, RNA was isolated from snap-frozen cecal tissues using the TriZol method (Thermo Fisher Scientific). cDNA was synthesized using SuperScript III reverse transcriptase (ThermoFisher Scientific). qRT-PCR was performed at the Functional Genomics Core at the University of South Carolina. The relative abundance of mammalian mRNA transcripts was calculated using the ΔΔC_T method and normalized to Eef2 levels. The oligonucleotides used for qRT-PCR were: Eef2 forward: TGTCAGTCATCGCCCATGTG, reverse: CATCCTTGCGAGTGTCAGTGA; Tnfa forward: AGCCAGGAGGGAGAACAGAAAC, reverse: CCAGTGAGTGAAAGGGACAGAACC; Nos2 forward TTGGGTCTTGTTCACTCCACGG, reverse: CCTCTTTCAGGTCACTTTGGTAGG; Il6 forward: GAAATGATGGATGCTACCAAACTG, reverse: CTCTCTGAAGGACTCTGGCTTTG; Il1b forward: CTCAATGGACAGAATATCAACCAAC, reverse: GGCTGTGCCGTCTTTCATTAC. Fecal samples were collected and immediately flash-frozen and stored at –80° C.

SoL extraction from mouse and patient fecal material

For SoL extraction and quantification, frozen mouse or human fecal samples were lyophilized to remove remaining water and resuspended in methanol (Thermo Fisher Scientific). Fecal sample suspensions were vortexed for 1 min prior to sonication for 10 min. The methanol extract was collected by centrifugation at 20,000 x g for 10 min and dried under a gentle stream of nitrogen. The resulting residue was then redissolved in methanol + 0.1% ammonium hydroxide (ThermoFisher Scientific) and filtered through a 0.22 μm filter prior to analysis.

Liquid chromatography and mass spectrometry analysis

High-resolution mass spectra were collected using a ThermoFisher Scientific Q-Exactive HF-X hybrid Quadrupole-Orbitrap mass spectrometer using electrospray ionization in negative mode. Liquid chromatography (LC) used a ThermoFisher Scientific Vanquish HPLC coupled to the aforementioned mass spectrometer. LC was performed using a Waters Xbridge BEH C18 XP column (2.1 x 100 mm) with alkaline mobile phase (pH 10.7) consisting of solvent A (water + 0.1% ammonium hydroxide) and solvent B (acetonitrile + 0.1% ammonium hydroxide) in a gradient starting from 35% B and increasing to 100% B over 10 min, hold at 100% B for 5 min, then re-equilibration at 35% B for 5 min at a flow rate of 0.4 mL/min. MS scans were obtained in the orbitrap analyzer which was scanned from 500 to 2000 m/z at a resolution of 60,000 (at 200 m/z). Targeted MS/MS fragmentation was conducted for SoLs in a ± 0.5 Da window around their expected m/z. MS data were analyzed by Xcalibur (ThermoFisher Scientific 4.2.47). Relative retention time was calculated by dividing RT of the analyte of interest by the RT of a standard compound.

Anaerobic culture and bioactive molecular networking

Three Alistipes (A. putredinis DSM 17216, A. timonensis DSM 25383, and A. timonensis DSM 27924) and two Odoribacter strains (O. laneus DSM 22474 and O. splanchnicus DSM 20712) were cultured in Reinforced Clostridial Medium (RCM, BD Biosciences) under anaerobic conditions at 37 °C. After three days of growth, cultures were harvested by centrifugation at 12,000 x g for 30 min. The resulting cell-free supernatant was extracted with an equal volume of methyl ethyl ketone and the cell pellets were extracted by resuspension in methanol and sonication before both extracts were combined and concentrated in vacuo. The combined crude extract was then fractionated on a silica gel column using a stepwise gradient of dichloromethane and methanol (DCM:MeOH; 15:1, 7:1, 5:1, 3:1, 1:1). Each fraction was then used in an in vitro cell-based assay measuring the suppression of LPS-induced TNFα expression. Simultaneously, samples of the fractions were subjected to untargeted HPLC-HRMS/MS as described above. MS/MS was conducted using data-dependent acquisition with a resolution of 30,000, isolation window of 2.0 m/z, and dynamic exclusion time of 15 s. HPLC-HRMS/MS data was processed using MZmine3 following the GNPS FBMN workflow with minimal changes⁹⁵. Molecular networks were constructed using the quickstart GNPS FBMN setting with no changes⁵¹. Raw data used for this analysis was deposited in the University of California, San Diego Center for Computational Mass Spectrometry MassIVE database (https://doi.org/doi:10.25345/C5028PP9T; ftp://massive.ucsd.edu/MSV000091884/). Bioactivity scores were assigned using a custom R script which calculated Pearson correlation coefficients between each molecular feature and the activity of each fraction⁹⁶. Finally, bioactive molecular networks were visualized in Cytoscape v3.9.1⁹⁷.

Purification of SoLs A and B

Fractions containing SoLs were further purified by Sephadex LH-20 run in 1:1 DCM:MeOH. Finally, pure SoLs A and B were isolated by semi-preparative scale HPLC running an isocratic solvent composition of 47% water + 0.1% ammonium hydroxide and 53% acetonitrile + 0.1% ammonium hydroxideon a ThermoFisher Scientific Ultimate 3000 semi-preparative scale HPLC equipped with a Waters Xbridge Prep C18 5 μm OBD column (19 x 100 mm) with a flow rate of 5.0 mL/min. ¹H, ¹³C, ¹H-¹³C HSQC, ¹H-¹³C HMBC, and ¹H-¹H COSY NMR spectra for SoLs A and B were acquired in methanol-d4 on a Bruker Avance III HD 400 MHz spectrometer with a 5 mm BBO 1H/19F-BB-Z-Gradient prodigy cryoprobe. Data were collected and reported as follows: chemical shift, integration multiplicity (s, singlet; d, doublet; t, triplet; m, multiplet), coupling constant. Chemical shifts were reported using the methanol-d4 resonance as the internal standard for ¹H-NMR methanol-d4: δ = 3.31 ppm and ¹³C-NMR methanol-d4: δ = 49.0 ppm. Pure SoLs A and B were confirmed to be free of LPS using a Chromogenic Endotoxin Quant Kit (Pierce).

Preparation and treatment of macrophages

Primary mouse macrophages were prepared by first introducing 3 mL of 3% thioglycolate to mice via intraperitoneal injection. After 3 days, 10 mL of chilled PBS was introduced intraperitoneally to flush out macrophages. The cell suspension was then separated by centrifugation at 300 x g for 5 min. Cells were seeded in culture dishes containing DMEM with 10% FBS for 1 h before being washed with serum-free DMEM two times to remove unattached cells. The cells were incubated in serum-free DMEM for 16 h before treatment. To treat the macrophages, cells were incubated for 6 to 24 h in DMEM without FBS with addition of LPS (Sigma-Aldrich), Pam3CSK4 (Invivogen), or SoL A. Cells were finally washed twice with Dulbecco’s phosphate-buffered saline before being lysed for total RNA or protein extraction.

mRNA extraction and RT-qPCR in macrophage-based assays

Treated mouse macrophage cells were lysed with TriZol (Invitrogen) and total RNA was extracted from the cell lysate using a Direct-zol RNA miniprep kit (Zymo Research) according to the manufacturer’s protocol. The quality and quantity of RNA was then determined using a nanodrop and 1000 ng of mRNA from each sample was used for cDNA synthesis using a First-strand cDNA Synthesis System (Marligen Bioscience). qPCRs reactions were prepared in a 20 uL final volume containing Fast Start Universal SYBR Green Master (Rox) (Roche Applied Science), cDNA template, deionized water, and primers and probes for IL-1β, TNFα, IL-6, and the 18S rRNA which was used as a housekeeping gene Cycling conditions were 95 °C for 10 min followed by 40 cycles of 95 °C for 10 s, 60 °C for 15 s, and 68 °C for 20 s, then a melting curve analysis from 60 °C to 95 °C every 0.2 °C was obtained. Amplifications were performed on an Eppendorf Realplex Mastercycler (Eppendorf). Relative gene expression levels were calculated using the ΔΔC_T method and expression levels of 18S were used to normalize the results.

Molecular docking

The crystal structure of the TLR4/MD-2 comlpex was retrieved from the Protein Data Bank (PDB ID: 3FXI)⁵⁷ and prepared using AutoDock Tools⁹⁸. Molecular structures of SoL A, SoL B, sulfatide, and lipid A were constructed, and energy minimized using Marvin version 21.17.0, ChemAxon (https://www.chemaxon.com). Models of SoL A, SoL B, sulfatide, and lipid A were also prepared using AutoDock Tools and docked against the TLR4/MD-2 complex using AutoDock Vina^99,100 in a 32x32x32 angstrom box surrounding the MD-2 monomer. Docking results were visualized using PyMol¹⁰¹.

ELISA displacement assay

Solid-phase sandwich ELISA kits were purchased from Invitrogen. The ELISA experiments were performed according to the kit instructions, using 50 nM hMD-2 (Novus Biologicals), 1 ng/mL LPS-EB Biotin (Invivogen, Cat. No.: tlrl-lpsbiot, derived from E. coli 0111:B4), and 0.1, 1.0, and 10 μM purified SoL A. SoL A was added to the assay 1 h before, 1 h after, or simultaneously with LPS-EB Biotin. Absorbance was measured at 450 nm using a BioTek microplate reader.

Western blot

Macrophages were treated with 100 ng/mL of LPS and 5 or 20 μM SoL A for 30 min. Following treatment, all cellular protein was extracted using MPER lysis buffer (Thermo Scientific). Protein samples were loaded onto SDS-PAGE gels for separation, then transferred to nitrocellulose membranes (Amersham Biosciences). Primary antibodies and HRP-conjugated secondary antibodies (Cell Signaling Technology) were used to detect target proteins. Antibodies used included p38 (Cell Signaling Technology, Cat. No. 8690, 1:1000 dilution), phospho-p38 (Cell Signaling Technology, Cat. No. 9211, 1:1000 dilution), ERK1/2 (Cell Signaling Technology, Cat. No. 4695, 1:1000 dilution), phospho-ERK1/2 (Cell Signaling Technology, Cat. No. 9101, 1:1000 dilution), IκBα (Cell Signaling Technology, Cat. No. 4814, 1:1000 dilution), and β-actin (Cell Signaling Technology, Cat. No. 3700, 1:1000 dilution). Signal was detected using an ECL kit (Thermo Scientific).

Macrophage polarization

THP-1 monocytes were maintained in RPMI1640 with 10% heat inactivated FBS, 1% Penicillin-Streptomycin-Amphotericin B, and 50 μM 2-mercaptoethanol prior to differentiation. The cells were differentiated into macrophages with 150 nM PMA for 48 h. M1 polarization was induced by adding 20 ng/mL IFN-γ and 100 pg/mL LPS for 24 h. M2 polarization was induced by adding 20 ng/mL IL-4 and 20 ng/mL IL-13 for 24 h. In all tests, 10 μM SoL A was added at the same time as M1 and M2 differentiation agents. After 24 h of treatment, total RNA was collected, and RT-qPCR was performed as described above.

Statistics and reproducibility

Statistical tests in the biosynthetic enzyme-disease correlation portion of this study used publicly available data, thus aspects including sample size, data exclusion, randomization, and blinding were not applicable to our re-analysis. For all other experiments, sample sizes were chosen to achieve statistical significance, and data from outliers were excluded based on Grubbs’ test for outliers. For animal experiments, randomization was applied to assign animals to groups and cages and blinding was applied in the analysis of histopathology data.

Reporting summary

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

Data availability

285,385 human microbial reference genomes were obtained from a previously published collection³¹. Whole-genome sequencing datasets of human gut microbiomes related to IBD used for biosynthetic enzyme-guided disease correlation were downloaded from the NCBI SRA (SRA Accessions: PRJNA398089 and PRJNA389280). Targeted metabolomic analysis performed in this work was based on the data obtained from two publicly available metabolomics datasets downloaded from IBDMDB and Metabolomics Workbench (Accession: PR000677 [https://doi.org/10.21228/M85H44]). Tables for the processed biosynthetic enzymes and metabolite abundance are available in Supplementary Data 2, 3, and 8. The mass spectrometry data generated in this study for bioactive molecular networking have been deposited in the UCSD CCMS MassIVE database under accession MSV000091884 [https://doi.org/10.25345/C5028PP9T]¹⁰². Source data are provided with this paper.

Code availability

The code used for analysis in support of our findings are available in the GitHub repository: https://github.com/ZHANGJianArya/SoL. The code can also be accessed in Zenodo with the https://doi.org/10.5281/zenodo.13896673¹⁰³.

References

Donia, M. S. & Fischbach, M. A. Small molecules from the human microbiota. Science 349, 1254766 (2015).
Article PubMed PubMed Central Google Scholar
Haiser, H. J. & Turnbaugh, P. J. Developing a metagenomic view of xenobiotic metabolism. Pharmacol. Res. 69, 21–31 (2013).
Article CAS PubMed Google Scholar
Flint, H. J., Scott, K. P., Duncan, S. H., Louis, P. & Forano, E. Microbial degradation of complex carbohydrates in the gut. Gut. Microbes 3, 289–306 (2012).
Article PubMed PubMed Central Google Scholar
Dai, H. et al. Recent advances in gut microbiota-associated natural products: structures, bioactivities, and mechanisms. Nat. Prod. Rep. 40, 1078–1093 (2023).
Article CAS PubMed Google Scholar
Lavelle, A. & Sokol, H. Gut microbiota-derived metabolites as key actors in inflammatory bowel disease. Nat. Rev. Gastroenterol. Hepatol. 17, 223–237 (2020).
Article PubMed Google Scholar
Skelly, A. N., Sato, Y., Kearney, S. & Honda, K. Mining the microbiota for microbial and metabolite-based immunotherapies. Nat. Rev. Immunol. 19, 305–323 (2019).
Article CAS PubMed Google Scholar
Spanogiannopoulos, P., Bess, E. N., Carmody, R. N. & Turnbaugh, P. J. The microbial pharmacists within us: a metagenomic view of xenobiotic metabolism. Nat. Rev. Microbiol. 14, 273–287 (2016).
Article CAS PubMed PubMed Central Google Scholar
Cao, Y. et al. Commensal microbiota from patients with inflammatory bowel disease produce genotoxic metabolites. Science 378, eabm3233 (2022).
Article CAS PubMed PubMed Central Google Scholar
Yao, L. et al. A biosynthetic pathway for the selective sulfonation of steroidal metabolites by human gut bacteria. Nat. Microbiol. 7, 1404–1418 (2022).
Article CAS PubMed PubMed Central Google Scholar
Fischbach, M. A. Microbiome: Focus on causation and mechanism. Cell 174, 785–790 (2018).
Article CAS PubMed PubMed Central Google Scholar
Chaudhari, S. N., McCurry, M. D. & Devlin, A. S. Chains of evidence from correlations to causal molecules in microbiome-linked diseases. Nat. Chem. Biol. 17, 1046–1056 (2021).
Article CAS PubMed PubMed Central Google Scholar
Koh, A., De Vadder, F., Kovatcheva-Datchary, P. & Bäckhed, F. From dietary fiber to host physiology: Short-chain fatty acids as key bacterial metabolites. Cell 165, 1332–1345 (2016).
Article CAS PubMed Google Scholar
Chiurchiù, V., Leuti, A. & Maccarrone, M. Bioactive lipids and chronic inflammation: Managing the fire within. Front. Immunol. 9, 38 (2018).
Article PubMed PubMed Central Google Scholar
Bae, M. et al. Akkermansia muciniphila phospholipid induces homeostatic immune responses. Nature 608, 168–173 (2022).
Article ADS CAS PubMed PubMed Central Google Scholar
MacEyka, M. & Spiegel, S. Sphingolipid metabolites in inflammatory disease. Nature 510, 58–67 (2014).
Article ADS CAS PubMed PubMed Central Google Scholar
Walker, A. et al. Sulfonolipids as novel metabolite markers of Alistipes and Odoribacter affected by high-fat diets. Sci. Rep. 7, 11047 (2017).
Article ADS PubMed PubMed Central Google Scholar
Pitta, T. P., Leadbetter, E. R. & Godchaux, W. Increase of ornithine amino lipid content in a sulfonolipid-deficient mutant of Cytophaga johnsonae. J. Bacteriol. 171, 952–957 (1989).
Article CAS PubMed PubMed Central Google Scholar
Durack, J. & Lynch, S. V. The gut microbiome: relationships with disease and opportunities for therapy. J. Exp. Med. 216, 20–40 (2018).
Article PubMed Google Scholar
Lloyd-Price, J. et al. Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature 569, 655–662 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Dziarski, R., Park, S. Y., Kashyap, D. R., Dowd, S. E. & Gupta, D. Pglyrp-regulated gut microflora Prevotella falsenii, Parabacteroides distasonis and Bacteroides eggerthii enhance and Alistipes finegoldii attenuates colitis in mice. PLOS ONE 11, e0146162 (2016).
Article PubMed Google Scholar
Lima, S. F. et al. Transferable immunoglobulin A–coated Odoribacter splanchnicus in responders to fecal microbiota transplantation for ulcerative colitis limits colonic inflammation. Gastroenterology 162, 166–178 (2022).
Article CAS PubMed Google Scholar
Hou, L. et al. Identification and biosynthesis of pro-inflammatory sulfonolipids from an opportunistic pathogen Chryseobacterium gleum. ACS Chem. Biol. 17, 1197–1206 (2022).
Article CAS PubMed Google Scholar
Pasternak, B. A. et al. Lipopolysaccharide exposure is linked to activation of the acute phase response and growth failure in pediatric Crohnʼs disease and murine colitis. Inflamm. Bowel Dis. 16, 856–869 (2010).
Article PubMed Google Scholar
Pastor Rojo, O. et al. Serum lipopolysaccharide-binding protein in endotoxemic patients with inflammatory bowel disease. Inflamm. Bowel Dis. 13, 269–277 (2007).
Article PubMed Google Scholar
Im, E., Riegler, F. M., Pothoulakis, C. & Rhee, S. H. Elevated lipopolysaccharide in the colon evokes intestinal inflammation, aggravated in immune modulator-impaired mice. Am. J. Physiol.-Gastrointest. Liver Physiol. 303, G490–G497 (2012).
Article CAS PubMed Google Scholar
Stephens, M. & von der Weid, P.-Y. Lipopolysaccharides modulate intestinal epithelial permeability and inflammation in a species-specific manner. Gut. Microbes 11, 421–432 (2020).
Article PubMed Google Scholar
Chaleckis, R., Meister, I., Zhang, P. & Wheelock, C. E. Challenges, progress and promises of metabolite annotation for LC–MS-based metabolomics. Curr. Opin. Biotechnol. 55, 44–50 (2019).
Article CAS PubMed Google Scholar
Cui, L., Lu, H. & Lee, Y. H. Challenges and emergent solutions for LC-MS/MS based untargeted metabolomics in diseases. Mass Spec. Rev. 37, 772–792 (2018).
Article ADS CAS Google Scholar
Schrimpe-Rutledge, A. C., Codreanu, S. G., Sherrod, S. D. & McLean, J. A. Untargeted metabolomics strategies—challenges and emerging directions. J. Am. Soc. Mass Spectrom. 27, 1897–1905 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar
Proctor, L. M. et al. The integrative human microbiome project. Nature 569, 641–648 (2019).
Article ADS Google Scholar
Vatanen, T. et al. Genomic variation and strain-specific functional adaptation in the human gut microbiome during early life. Nat. Microbiol. 4, 470–479 (2019).
Article CAS PubMed Google Scholar
Katz, K. et al. The sequence read archive: a decade more of explosive growth. Nucleic Acids Res. 50, D387–D390 (2022).
Article ADS CAS PubMed Google Scholar
Almeida, A. et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat. Biotechnol. 39, 105–114 (2021).
Article CAS PubMed Google Scholar
Liu, Y. et al. Identification and characterization of the biosynthetic pathway of the sulfonolipid capnine. Biochemistry 61, 2861–2869 (2022).
Article CAS PubMed Google Scholar
Vences‐Guzmán, M. Á. et al. Identification of the Flavobacterium johnsoniae cysteate‐fatty acyl transferase required for capnine synthesis and for efficient gliding motility. Environ. Microbiol. 23, 2448–2460 (2021).
Article PubMed Google Scholar
Radka, C. D., Miller, D. J., Frank, M. W. & Rock, C. O. Biochemical characterization of the first step in sulfonolipid biosynthesis in Alistipes finegoldii. J. Biol. Chem. 298, 1–13 (2022).
Article Google Scholar
Radka, C. D., Frank, M. W., Rock, C. O. & Yao, J. Fatty acid activation and utilization by Alistipes finegoldii, a representative Bacteroidetes resident of the human gut microbiome. Mol. Microbiol. 113, 807–825 (2020).
Article CAS PubMed Google Scholar
Kamiyama, T. et al. Sulfobacins A and B, novel von Willebrand factor receptor antagonists: I. Production, isolation, characterization and biological activities. J. Antibiotics 48, 924–928 (1995).
Article CAS Google Scholar
Schirmer, M. et al. Dynamics of metatranscription in the inflammatory bowel disease gut microbiome. Nat. Microbiol. 3, 337–346 (2018).
Article CAS PubMed Google Scholar
Franzosa, E. A. et al. Gut microbiome structure and metabolic activity in inflammatory bowel disease. Nat. Microbiol. 4, 293–305 (2019).
Article CAS PubMed Google Scholar
Tsugawa, H. et al. A lipidome atlas in MS-DIAL 4. Nat. Biotechnol. 38, 1159–1163 (2020).
Article CAS PubMed Google Scholar
Guo, J., Shen, S., Xing, S., Yu, H. & Huan, T. ISFrag: De novo recognition of in-source fragments for liquid chromatography–mass spectrometry data. Anal. Chem. 93, 10243–10250 (2021).
Article CAS PubMed Google Scholar
Folz, J. et al. Human metabolome variation along the upper intestinal tract. Nat. Metab. 5, 777–788 (2023).
Article CAS PubMed Google Scholar
Witting, M. & Böcker, S. Current status of retention time prediction in metabolite identification. J. Sep. Sci. 43, 1746–1754 (2020).
Article CAS PubMed Google Scholar
García, C. A., Gil-de-la-Fuente, A., Barbas, C. & Otero, A. Probabilistic metabolite annotation using retention time prediction and meta-learned projections. J. Cheminformatics 14, 33 (2022).
Article Google Scholar
Sun, L. et al. A simple method for HPLC retention time prediction: linear calibration using two reference substances. Chin. Med. 12, 16 (2017).
Article PubMed Google Scholar
Wang, Y. et al. A simple method for peak alignment using relative retention time related to an inherent peak in liquid chromatography-mass spectrometry-based metabolomics. J. Chromatographic Sci. 57, 9–16 (2019).
Article CAS Google Scholar
Hale, L. P., Gottfried, M. R. & Swidsinski, A. Piroxicam treatment of IL-10-deficient mice enhances colonic epithelial apoptosis and mucosal exposure to intestinal bacteria. Inflamm. Bowel Dis. 11, 1060–1069 (2005).
Article PubMed Google Scholar
Berg, D. J. et al. Rapid development of colitis in NSAID-treated IL-10-deficient mice. Gastroenterology 123, 1527–1542 (2002).
Article CAS PubMed Google Scholar
Cario, E. Toll-like receptors in inflammatory bowel diseases: a decade later. Inflamm. Bowel Dis. 16, 1583–1597 (2010).
Article PubMed Google Scholar
Nothias, L.-F. et al. Feature-based molecular networking in the GNPS analysis environment. Nat. Methods 17, 905–908 (2020).
Article CAS PubMed Google Scholar
Kawasaki, T. & Kawai, T. Toll-like receptor signaling pathways. Front. Immunol. 5, 1–8 (2014).
Article CAS Google Scholar
Kawai, T. & Akira, S. The role of pattern-recognition receptors in innate immunity: update on Toll-like receptors. Nat. Immunol. 11, 373–384 (2010).
Article CAS PubMed Google Scholar
Maeda, J. et al. Inhibitory effects of sulfobacin B on DNA polymerase and inflammation. Int. J. Mol. Med. 26, 521–527 (2010).
Google Scholar
Shimazu, R. et al. MD-2, a molecule that confers lipopolysaccharide responsiveness on Toll-like receptor 4. J. Exp. Med. 189, 1777–1782 (1999).
Article CAS PubMed PubMed Central Google Scholar
Miyake, K. Roles for accessory molecules in microbial recognition by Toll-like receptors. J. Endotoxin Res. 12, 195–204 (2006).
CAS PubMed Google Scholar
Park, B. S. et al. The structural basis of lipopolysaccharide recognition by the TLR4–MD-2 complex. Nature 458, 1191–1195 (2009).
Article ADS CAS PubMed Google Scholar
Su, L. et al. Sulfatides are endogenous ligands for the TLR4–MD-2 complex. Proc. Natl Acad. Sci. 118, 1–12 (2021).
Article Google Scholar
Luk, J. M., Kumar, A., Tsang, R. & Staunton, D. Biotinylated lipopolysaccharide binds to endotoxin receptor in endothelial and monocytic cells. Anal. Biochem. 232, 217–224 (1995).
Article CAS PubMed Google Scholar
Mosser, D. M. & Edwards, J. P. Exploring the full spectrum of macrophage activation. Nat. Rev. Immunol. 8, 958–969 (2008).
Article CAS PubMed Google Scholar
Zhang, Y. et al. ECM1 is an essential factor for the determination of M1 macrophage polarization in IBD in response to LPS stimulation. Proc. Natl Acad. Sci. USA 117, 3083–3092 (2020).
Article ADS CAS PubMed Google Scholar
Zhang, X. & Mosser, D. Macrophage activation by endogenous danger signals. J. Pathol. 214, 161–178 (2008).
Article CAS PubMed Google Scholar
Parker, B. J., Wearsch, P. A., Veloo, A. C. M. & Rodriguez-Palacios, A. The genus Alistipes: Gut bacteria with emerging implications to inflammation, cancer, and mental health. Front. Immunol. 11, 906 (2020).
Article CAS PubMed Google Scholar
Chang, P. V., Hao, L., Offermanns, S. & Medzhitov, R. The microbial metabolite butyrate regulates intestinal macrophage function via histone deacetylase inhibition. Proc. Natl Acad. Sci. USA 111, 2247–2252 (2014).
Article ADS CAS PubMed Google Scholar
Trapecar, M. et al. Gut-liver physiomimetics reveal paradoxical modulation of IBD-related inflammation by short-chain fatty acids. Cell Syst. 10, 223–239.e9 (2020).
Article CAS PubMed Google Scholar
Bhattarai, Y. et al. Bacterially derived tryptamine increases mucus release by activating a host receptor in a mouse model of inflammatory bowel disease. iScience 23, 101798 (2020).
Article ADS CAS PubMed Google Scholar
Dodd, D. et al. A gut bacterial pathway metabolizes aromatic amino acids into nine circulating metabolites. Nature 551, 648–652 (2017).
Article ADS CAS PubMed PubMed Central Google Scholar
Paik, D. et al. Human gut bacteria produce Τ_Η17-modulating bile acid metabolites. Nature 603, 907–912 (2022).
Article ADS CAS PubMed Google Scholar
Sato, Y. et al. Novel bile acid biosynthetic pathways are enriched in the microbiome of centenarians. Nature 599, 458–464 (2021).
Article ADS CAS PubMed Google Scholar
Li, W. et al. A bacterial bile acid metabolite modulates Treg activity through the nuclear hormone receptor NR4A1. Cell Host Microbe 29, 1366–1377.e9 (2021).
Article CAS PubMed PubMed Central Google Scholar
Rakoff-Nahoum, S. & Medzhitov, R. Toll-like receptors and cancer. Nat. Rev. Cancer 9, 57–63 (2009).
Article CAS PubMed Google Scholar
Moreira Lopes, T. C., Mosser, D. M. & Gonçalves, R. Macrophage polarization in intestinal inflammation and gut homeostasis. Inflamm. Res. 69, 1163–1172 (2020).
Article CAS PubMed Google Scholar
Bhattarai, Y., Muniz Pedrogo, D. A. & Kashyap, P. C. Irritable bowel syndrome: a gut microbiota-related disorder? Am. J. Physiol.-Gastrointest. Liver Physiol. 312, G52–G62 (2016).
Article PubMed PubMed Central Google Scholar
Wang, J. et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490, 55–60 (2012).
Article ADS PubMed Google Scholar
Thomann, A. K. et al. Depression and fatigue in active IBD from a microbiome perspective-a Bayesian approach to faecal metagenomics. BMC Med. 20, 366 (2022).
Article CAS PubMed Google Scholar
Mars, R. A. T. et al. Longitudinal multi-omics reveals subset-specific mechanisms underlying irritable bowel syndrome. Cell 182, 1460–1473.e17 (2020).
Article CAS PubMed Google Scholar
Mukherjee, S. et al. Twenty-five years of Genomes OnLine Database (GOLD): data updates and new features in v.9. Nucleic Acids Res. 51, D957–D963 (2023).
Article CAS PubMed Google Scholar
Blin, K. et al. antiSMASH 6.0: improving cluster detection and comparison capabilities. Nucleic Acids Res. 49, W29–W35 (2021).
Article CAS PubMed Google Scholar
Blin, K. et al. The antiSMASH database version 2: a comprehensive resource on secondary metabolite biosynthetic gene clusters. Nucleic Acids Res. 47, D625–D630 (2019).
Article CAS PubMed Google Scholar
Terlouw, B. R. et al. MIBiG 3.0: a community-driven effort to annotate experimentally validated biosynthetic gene clusters. Nucleic Acids Res. 51, D603–D610 (2023).
Article CAS PubMed Google Scholar
Buchfink, B., Reuter, K. & Drost, H.-G. Sensitive protein alignments at tree-of-life scale using DIAMOND. Nat. Methods 18, 366–368 (2021).
Article CAS PubMed PubMed Central Google Scholar
Mirdita, M., Steinegger, M., Breitwieser, F., Söding, J. & Levy Karin, E. Fast and sensitive taxonomic assignment to metagenomic contigs. Bioinformatics 37, 3029–3031 (2021).
Article CAS PubMed PubMed Central Google Scholar
Chaumeil, P.-A., Mussig, A. J., Hugenholtz, P. & Parks, D. H. GTDB-Tk v2: memory friendly classification with the genome taxonomy database. Bioinformatics 38, 5315–5316 (2022).
Article CAS PubMed PubMed Central Google Scholar
Letunic, I. & Bork, P. Interactive Tree Of Life (iTOL) v5: An online tool for phylogenetic tree display and annotation. Nucleic Acids Res. 49, W293–W296 (2021).
Article CAS PubMed Google Scholar
Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows–Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
Article CAS PubMed PubMed Central Google Scholar
Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014).
Article CAS PubMed Google Scholar
R. Core Team. R: A Language and Environment for Statistical Computing. https://www.R-project.org/ (2021).
Oksanen, J. et al. Vegan: Community Ecology Package. https://CRAN.R-project.org/package=vegan (2020).
Van Rossum, G. & Drake, F. L. Python 3 Reference Manual. (CreateSpace, Scotts Valley, CA, 2009).
McKinney, W. Data structures for statistical computing in python. in Proceedings of the 9th Python in Science Conference 51–56 https://doi.org/10.25080/Majora-92bf1922-00a (2010).
Virtanen, P. et al. SciPy 1.0: Fundamental algorithms for scientific computing in python. Nat. Methods 17, 261–272 (2020).
Article CAS PubMed PubMed Central Google Scholar
Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc.: Ser. B (Methodol.) 57, 289–300 (1995).
Article MathSciNet Google Scholar
Erben, U. et al. A guide to histomorphological evaluation of intestinal inflammation in mouse models. Int J. Clin. Exp. Pathol. 7, 4557–4576 (2014).
PubMed PubMed Central Google Scholar
Ellermann, M. et al. Endocannabinoids Inhibit the Induction of Virulence in Enteric Pathogens. Cell 183, 650–665.e15 (2020).
Article CAS PubMed PubMed Central Google Scholar
Schmid, R. et al. Integrative analysis of multimodal mass spectrometry data in MZmine 3. Nat. Biotechnol. 41, 447–449 (2023).
Article CAS PubMed PubMed Central Google Scholar
Nothias, L.-F. et al. Bioactivity-based molecular networking for the discovery of drug leads in natural product bioassay-guided fractionation. J. Nat. Prod. 81, 758–767 (2018).
Article CAS PubMed Google Scholar
Shannon, P. et al. Cytoscape: A software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003).
Article CAS PubMed Google Scholar
Morris, G. M. et al. AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility. J. Comput Chem. 30, 2785–2791 (2009).
Article CAS PubMed PubMed Central Google Scholar
Trott, O. & Olson, A. J. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. J. Computational Chem. 31, 455–461 (2010).
Article CAS Google Scholar
Eberhardt, J., Santos-Martins, D., Tillack, A. F. & Forli, S. AutoDock Vina 1.2.0: New docking methods, expanded force field, and python bindings. J. Chem. Inf. Model. 61, 3891–3898 (2021).
Article CAS PubMed Google Scholar
Schrödinger, L. L. C. The PyMOL molecular graphics system, version 2.0. (2015).
Li, J. MassIVE MSV000091884 - GNPS - Bioactive molecular networking for sulfonolipids - MZmine Processing for FBMN: This dataset contains the raw data used for MZmine processing and Feature-based molecular networking of semi-purified fractions obtained from A. timonensis used for bioactive molecular networking to reveal contribution of sulfonolipids to observed biological activity. MassIVE https://doi.org/10.25345/C5028PP9T (2023).
Article Google Scholar
Zhang, J. ZHANGJianArya/SoL: initial release for citation. Zenodo https://doi.org/10.5281/ZENODO.13896673 (2024).
Article Google Scholar

Download references

Acknowledgements

This work is partially funded by a National Institutes of Health (NIH) grant P20GM103641 awarded to M.N. and J.L., an NIH grant 1R35GM150565 awarded to J.L., a National Science Foundation grant 2239561 awarded to J.L., a Hong Kong Research Grants Council ECS grant HKU27107320 awarded to Y.X.-L., and the Hong Kong Branch of Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou) grant (SMSEGL20SC02) awarded to Y.X.-L. We acknowledge Dr. Kristen Hogan and Dr. Maria Marjorette Pena of the University of South Carolina Center for Colon Cancer Research Mouse Core Facility for assistance with mouse microbiome studies, as well as Dr. Michael Walla of the University of South Carolina Mass Spectrometry Facility for assistance in acquiring HRMS and HRMS/MS data. We also acknowledge Emily Quinn, Dorathea Lee, and Andrew Campbell for assistance with SoL purification.

Author information

These authors contributed equally: Ethan A. Older, Jian Zhang.

Authors and Affiliations

Department of Chemistry and Biochemistry, University of South Carolina, Columbia, South Carolina, 29208, USA
Ethan A. Older, Zachary E. Ferris, Dan Xue, Michael Madden & Jie Li
Department of Chemistry and The Swire Institute of Marine Science, The University of Hong Kong, Pokfulam Road, Hong Kong, China
Jian Zhang, Zheng Zhong & Yong-Xin Li
Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou), Guangzhou, China
Jian Zhang, Zheng Zhong & Yong-Xin Li
Department of Biological Sciences, University of South Carolina, Columbia, South Carolina, 29208, USA
Mary K. Mitchell, Hexin Chen & Melissa Ellermann
Department of Cell Biology and Anatomy, School of Medicine, University of South Carolina, Columbia, South Carolina, 29209, USA
Yuzhen Wang & Daping Fan
Department of Pathology, Microbiology and Immunology, School of Medicine, University of South Carolina, Columbia, South Carolina, 29209, USA
Prakash Nagarkatti & Mitzi Nagarkatti

Authors

Ethan A. Older
View author publications
Search author on:PubMed Google Scholar
Jian Zhang
View author publications
Search author on:PubMed Google Scholar
Zachary E. Ferris
View author publications
Search author on:PubMed Google Scholar
Dan Xue
View author publications
Search author on:PubMed Google Scholar
Zheng Zhong
View author publications
Search author on:PubMed Google Scholar
Mary K. Mitchell
View author publications
Search author on:PubMed Google Scholar
Michael Madden
View author publications
Search author on:PubMed Google Scholar
Yuzhen Wang
View author publications
Search author on:PubMed Google Scholar
Hexin Chen
View author publications
Search author on:PubMed Google Scholar
Prakash Nagarkatti
View author publications
Search author on:PubMed Google Scholar
Mitzi Nagarkatti
View author publications
Search author on:PubMed Google Scholar
Daping Fan
View author publications
Search author on:PubMed Google Scholar
Melissa Ellermann
View author publications
Search author on:PubMed Google Scholar
Yong-Xin Li
View author publications
Search author on:PubMed Google Scholar
Jie Li
View author publications
Search author on:PubMed Google Scholar

Contributions

E.A.O., J.Z., Y.-X.L., and J.L. designed the study; E.A.O., J.Z., Z.E.F., D.X., Z.Z., M.K.M, M.M., and Y.W. collected the data; E.A.O., J.Z., H.C., D.F., M.E., Y.-X.L., and J.L. analyzed the data; P.N., M.N., Y.-X.L., and J.L. funded the study; E.A.O., J.Z., Y.-X.L., and J.L. prepared the manuscript; All authors participated in revising the manuscript.

Corresponding authors

Correspondence to Yong-Xin Li or Jie Li.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Alexander Rodriguez-Palacios, and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Additional information

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

Supplementary information

Supplementary Information (download PDF )

Description of Additional Supplementary Files (download PDF )

Supplementary Data 1-7 (download XLSX )

Supplementary Data 8 (download XLSX )

Supplementary Data 9-12 (download XLSX )

Reporting Summary (download PDF )

Transparent Peer Review file (download PDF )

Source data

Source Data (download XLSX )

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Older, E.A., Zhang, J., Ferris, Z.E. et al. Biosynthetic enzyme analysis identifies a protective role for TLR4-acting gut microbial sulfonolipids in inflammatory bowel disease. Nat Commun 15, 9371 (2024). https://doi.org/10.1038/s41467-024-53670-y

Download citation

Received: 10 May 2023
Accepted: 18 October 2024
Published: 30 October 2024
Version of record: 30 October 2024
DOI: https://doi.org/10.1038/s41467-024-53670-y

This article is cited by

Genetics-mediated regulation of intestinal gene expression on microbiome contributes to human disease heritability
- Haochuan Wang
- Chengyu Li
- Xushen Xiong
Molecular Systems Biology (2025)