A human milk oligosaccharide prevents intestinal inflammation in adulthood via modulating gut microbial metabolism

Bacteria that can metabolize HMOs, such as B. infantis and Bifidobacterium breve, are abundant in breastfed infants (21), which persist into adulthood in decreasing but stable abundance (22, 23). To determine the exact impact of 2′-FL on the gut microbial composition in adult mice, whole-genome sequencing of fecal bacteria from mice treated with 2′-FL was performed. A relative abundance taxa plot depicts the top 15 families in control (Con.), 2′-FL 7 days (7D), and 2′-FL 28 days (28D) treatment groups (Fig. 1A). 2′-FL induced the unweighted unifrac principal coordinates analysis (PCoA) coordinates to significantly shift between the three groups (P = 0.003) (Fig. 1B). The relative abundance of several families was altered in the 2′-FL-treated groups compared to the control group (Fig. 1C) (Table S1). The relative abundances of Bifidobacteriaceae and Lactobacillaceae were quantitatively most enriched in the 2′-FL 28D group compared to the control (P = 0.0074 and P = 0.0126, respectively) and the 2′-FL 7D groups (P = 0.0095 and P = 0.0039, respectively). The relative abundance of Eggerthellaceae in 2′-FL-28D group was increased without significance compared to the control (P = 0.0567). The Oscillospiraceae, Rikenellaceae, and Desulfovibrionaceae families demonstrated a similar pattern by increasing significantly in the 2′-FL 7D group compared to the control group (P = 0.0115, P = 0.0244, and P = 0.0097, respectively), but decreased in the 2′-FL 28D group with no significant difference as compared to the control group.

Of the 42 genera identified, three were significantly enriched in the 2′-FL 28D group, but not in the 2′-FL-7D group, compared to the control group: Bifidobacterium (P = 0.0028), Lactobacillus (P = 0.0048), and Enterorhabdus (P = 0.0220). In contrast, the genus Anaerotruncus displayed a significant decrease in relative abundance (P = 0.0059) (Fig. 1D; Table S2). Next, individual species that were contributing to the specific genus- and family-level enrichment with 2′-FL consumption were identified (Table S3). Bifidobacterium pseudolongum (P = 0.0023), Lactobacillus sp. ASF360, reurteri, and johnsonii (P < 0.05 for each), and Enterorhabdus caecimuris P = 0.0431) were significantly enriched, and Anaerotruncus sp. 1XD22-93, P = 0.0151 was significantly diminished in 2′-FL 28D compared to the control groups. No significant changes in these four species were identified in the 2′-FL 7D group (Fig. 1E).

Because we found that 2′-FL pretreatment mitigated colitis in two mouse models of acute colitis, dextran sulfate sodium (DSS)-induced intestinal injury and inflammation, and 2,6,4-trinitrobenzenesulfonic acid (TNBS)-induced Th1-driven colitis in adult mice (Fig. S1), we leveraged the fecal microbiota transplantation (FMT) model to investigate the role of the 2′-FL-modulated gut microbial community on intestinal inflammation. FMT donor mice received 2′-FL in regular drinking water (2′-FL FMT) or regular water (control FMT). FMT recipient mice received antibiotic treatment to decrease the endogenous gut bacteria, followed by FMT and DSS-induced colonic injury and colitis (Fig. 2A). DSS treatment induced acute injury and colitis with massive colon ulceration, crypt damage, and severe inflammation with the injury/inflammation score of 17.18 ± 2.92, which was significantly mitigated by 2′-FL FMT (6.5 ± 1.01; P = 0.0021) but not control FMT (12.2 ± 2.44; P > 0.05) (Fig. 2B and C). DSS-induced mRNA expression of TNF, a proinflammatory cytokine, was significantly decreased by the 2′-FL FMT, as compared to the DSS and DSS plus control FMT groups (P = 0.0186 and P = 0.0126, respectively) (Fig. 2D). The distribution of a tight junctional protein, ZO-1, was detected by immunostaining. The 2′-FL FMT, but not the control FMT, prevented DSS-induced redistribution of ZO-1 from apical tight junctional complexes to the cytoplasmic compartment of colon epithelial cells in adult mice (Fig. 2E), indicating the protective roles of the 2′-FL-modified gut microbiota profile on the intestinal barrier. These results suggest that 2′-FL could lead to a balancing of the gut microbial community that exerts preventive effects on intestinal inflammation.

To further investigate how 2′-FL exerts protective effects through regulation of the gut microbiota, we focused on B. infantis as a bacterial model. B. infantis possesses gene clusters for the metabolism of HMOs (24) and is present in healthy adults in low abundance (22, 23). Multiple B. infantis strains have been shown to be able to grow on 2′-FL as a sole carbon source (25). Consistent with this evidence, 2′-FL had significant effects on promoting the growth of B. infantis ATCC 15702 (Fig. 3A) and ATCC 15697 (Fig. 3B) in Roswell Park Memorial Institute (RPMI) medium (no glucose) compared to the controls. Interestingly, we found that 2′-FL was able to significantly improve the growth of B. infantis 15702 under oxidative stress when B. infantis 15702 was cultured in the presence of 2′-FL and H₂O₂ (P < 0.05) (Fig. 3C).

Next, we applied untargeted metabolomic analysis of bacterial culture supernatants to discover 2′-FL-derived metabolite secretions by B. infantis 15702 cultured in reinforced clostridial medium (RCM). By comparing metabolites in the control medium (MC), medium with 2′-FL (MF), control culture supernatant (C), and 2′-FL culture supernatant (F), we detected four secreted and seven consumed significant metabolites. These significant metabolites were annotated using the metabolite level identification system (i.e., validated, putative, and tentative annotations) (26) and are shown in Fig. 3D. Of the secreted metabolites, pantothenol and isocarbamid were unique to the 2′-FL consumption, and N6-(Delta2-Isopentenyl)-adenine and PE 40:11 were common. Of the consumed metabolites, four were common (adenine, adenosine, thymine, and histidyl-methionine) and three were unique to the 2′-FL medium (glycerol, hypoxanthine, and cytosine). There were no secreted or consumed metabolites unique to the control group. Of the three significant 2′-FL-derived secreted metabolites, pantothenol (alcohol analog of pantothenate; vitamin B5) was the most abundant (>65-fold change compared to control; Fig. 3E) and significantly higher in normalized abundance/peak intensity in the 2′-FL supernatants compared to the control supernatant (P = 2.01 × 10⁻⁷; Fig. 3F). Additionally, the pantothenate biosynthetic process for Bifidobacterium pseudolongum was enriched after 28 days of 2′-FL consumption (identified in all seven mice) compared to the control group (identified in one mouse) and the 2′-FL 7D group (no mice identified) (Fig. 3G). Our results identify the secretion of novel 2′-FL-stimulated B. infantis metabolites, especially pantothenol, thus contributing to our understanding of the mechanisms of HMO-regulated metabolic pathways in gut microbes.

Pantothenate from the diet and gut microbial metabolism is water soluble and can be absorbed by intestinal epithelial cells (27). Pantothenate is required for the biosynthesis of coenzyme A (CoA), which is involved in processes such as the citric acid cycle and lipid metabolism and has shown antioxidant effects due to the formation of glutathione peroxidase, which is essential to protect against free radical-induced apoptosis (28). Pantothenate has been found to accelerate dermal fibroblast migration and proliferation during wound healing (29). Notably, pantothenate levels are reduced in the gut of IBD patients (30). Therefore, we tested if pantothenate could preserve the mucosal barrier and prevent colitis in adult mice.

Mice received pantothenate in drinking water for 2 weeks before induction of colitis. Compared to the intestinal injury and inflammation by DSS (score: 16.75 ± 2.23) and colitis by TNBS (score: 4.17 ± 0.48), pantothenate treatment significantly decreased DSS (score: 4.99 ± 0.49, P < 0.0001) (Fig. 4A and B) and TNBS-induced colitis (score: 1.86 ± 0.67, P = 0.0201) (Fig. 4C and D). To further characterize the effects of pantothenate on DSS-induced disruption of intestinal integrity, we performed an in vivo permeability assay to test intestinal barrier function. DSS-disrupted barrier function, as evidenced by increased FITC-dextran in the plasma was significantly decreased by pantothenate treatment (P = 0.0015, Fig. 4E). Furthermore, we determined the distribution of a protein marker of tight junction structure, ZO-1, using immunostaining. DSS-induced redistribution of this protein from the apical surface to the intracellular compartment was prevented by pantothenate co-treatment in (Fig. 4F).

We next studied how pantothenate could exert direct effects on intestinal epithelial cells to protect the intestinal barrier by using in vitro assays. We evaluated the effect of pantothenate on the oxidative stress-induced disruption of the epithelial barrier. H₂O₂ reduced the paracellular permeability in Caco2 cells measured by transepithelial electrical resistance (TEER) in a time-dependent manner, which was significantly reduced by pantothenate treatment (Fig. 5A). Furthermore, H₂O₂-induced redistribution of ZO-1 from apical tight junctional complexes to the cytoplasmic compartment of YAMC cells was prevented by pantothenate (Fig. 5B). We further studied how pantothenate reduces oxidative stress in YAMC cells by measuring the activity of glutathione peroxidase (GPX), an antioxidant enzyme involved in scavenging free radicals, resulting in a decrease in oxidative stress. Pantothenate at 0.1 mM significantly increased GPX activity in YAMC cells compared to control cells (P = 0.0171, Fig. 5C).

These results suggest that pantothenate plays a role in the preservation of epithelial integrity and prevention of colitis in adult mice, which may be through mitigating the oxidative stress associated with inflammation.

We further studied the impact of 2′-FL consumption on the gut microbial metabolic pathways in adult mice. Functional profiling was performed using the HMP Unified Metabolic Analysis Network (HUMAnN3) program, which profiles the presence/absence and abundance of microbial metabolic pathways (31). In total, 238, 246, and 268 total pathways were identified in the control, 2′-FL 7D, and 2′-FL 28D groups, respectively. By comparing these pathways presence among the three groups, in addition to 233 pathways that were common among all three groups, we identified 34 pathways only present after 2′-FL consumption: 4 pathways only in the 2′-FL 7D group, 9 pathways in the 2′-FL 7D and 28D groups, and 21 pathways only in the 2′-FL 28D group. Furthermore, there are five pathways that were identified in the control and 2′-FL 28D groups (Fig. 6A; Table S4). Further classification of the 34 pathways uniquely enriched by 2′-FL identified common super pathways and pathway products, with individual pathways (identified by BioCyc ID) abundance plotted for each group (Fig. 6B). Of the 34 2′-FL uniquely enriched microbial pathways, several produce metabolites and signaling molecules that have been found to be depleted in or to benefit IBD patients’ symptoms or ameliorate symptoms in rodent models of IBD, including thiamine (32), lactate (33), spermidine/spermine (34), acetate and butanoate (35), and fumarate (36).

Of the 233 common pathways, 99 were statistically different in abundance (P < 0.05) between the 2′-FL groups compared to the control group, the majority of which were enriched by 2′-FL consumption for 28 days as depicted by the heatmap (Fig. 6C). Of the 99 pathways, 98 pathways were significantly different between 28 days 2′-FL vs control, and one pathway was significantly different between 7 and 28 days 2′-FL vs control (Table S5). Classification of the 99 pathways by common functions identified that the majority were involved in degradation/utilization/assimilation (25) and amino acid biosynthesis (22), followed by generation of precursor metabolites and energy (16), cofactor, carrier, and vitamin biosynthesis (8), carbohydrate biosynthesis (6), secondary metabolite biosynthesis (6), nucleoside and nucleotide biosynthesis (5), fatty acid and lipid biosynthesis (2), cell wall biosynthesis (2), and others (7) (Fig. 6D) (Table S6). These results indicate that 2′-FL could shape the metabolic functions of gut microbes which in turn can benefit the host.

To reveal the relevance of 2′-FL-regulated microbial metabolic pathways to IBD, we utilized HUMANn3 to analyze whole-genome sequencing data of human fecal microbiota from 34 patients with UC and 21 non-IBD control subjects collected by the NIH Human Microbiome project, which is available in the online repository (https://ibdmdb.org/tunnel/public/HMP2/WGS/1818/products↗) (Fig. 7A). We identified 372 total microbial metabolic pathways in non-IBD and UC patients (Table S7), of which 82 pathways were significantly different between non-IBD and ulcerative colitis patients (FDR < 0.05) (Table S8). Further analysis of these 82 pathways showed that 33 pathways were significantly positively correlated/enriched with UC (correlation coefficient > 2 and FDR < 0.05), 31 pathways were significantly negatively correlated/enriched with UC (correlation coefficient < −2 and FDR < 0.05) (Fig. 7B). Comparison of the 99 statistically different in abundance pathways by 2′-FL and 34 unique pathways to 2′-FL in mice with the 82 statistically different pathways in human samples identified 26 pathways that were both altered by 2′-FL consumption and in UC (Fig. 7C) (Table S9). Plotting the 26 common pathways based on the coefficient of correlation for 2′-FL (y-axis) vs UC (x-axis) divided the pathways into four distinct cohorts: unique to 2′-FL and depleted in UC (three pathways, green), unique to 2′-FL and enriched in UC (four pathways, blue), enriched in 2′-FL and UC (10, black), and enriched in 2′-FL and depleted in UC (nine pathways, red) (Fig. 7D) (Table S10). A direct comparison of the nine 2′-FL-enriched and UC-depleted pathways identified microbial functions that 2′-FL can restore in UC patients (Fig. 7E), while a biological function-based categorization of the 26 shared pathways identified general functions that pathways are involved in (Fig. 7F). 2′-FL-enriched, or unique, pathways and UC-depleted pathways are primarily involved in the generation of precursor metabolites and energy (1), degradation/utilization/assimilation (4), and amino acid biosynthesis (2), nucleoside and nucleotide biosynthesis (2), carbohydrate biosynthesis (1), and terpenoid biosynthesis (2). In contrast, 2′-FL-enriched, or unique, pathways and UC-enriched pathways are primarily involved in amino acid biosynthesis (3), degradation/utilization/assimilation (3), thiamine biosynthesis (2), generation of precursor metabolites and energy (1), terpenoid biosynthesis (1), phospholipid biosynthesis (1), nucleic acid processing (1), and heme biosynthesis (1). These results demonstrate the potential application of 2′-FL for IBD prevention and treatment.

Male and female wild-type mice on C57BL/6 and Balb/c background were group housed and maintained with a 12-hour light-dark cycle. Food (PicoLab Laboratory Rodent Diet L50D diet, LabDiet) and water were provided ad libitum for all experiments.

Mice at the age of 6–8 weeks received 2′-FL [synthesized as previously described (44)] at 1 mg/mL in the drinking water for 7, 28, or 35 days before induction of colitis or serving as donors for the FMT experiments. Mice at the age of 6–8 weeks received D-calcium pantothenate at 1 mM (Cat#243300050, Fisher Scientific) in drinking water for 14 days. Mice given drinking water were used as a control. Baseline mouse daily water intake and body weight before and after treatments were monitored throughout the experiments.

Recipient mice received antibiotic treatment by oral administration of 200 mg/kg of body weight/mouse (for each antibiotic) of cefoperazone (Cat#C-660-5, Gold Biotechnology, Inc.), metronidazole (Cat#M81000, Research Products International), ampicillin sodium (Cat#A-301-5, Gold Biotechnology, Inc.), and neomycin sulfate (Cat#N-620-10, Gold Biotechnology, Inc.) once a day for four consecutive days, during which time mice were housed in autoclaved cages with autoclaved food and sterile water, which were changed daily. Mice receiving antibiotics were allowed a washout period of 48 hours (no treatment) in sterile cages before FMT (45).

Mice receiving 2′-FL for 35 days served as the 2′-FL-treated donors (2′-FL-donor) and mice receiving water as the control donors (control-donor). Fecal slurry samples were prepared from pooled stool from 2′-FL-donor (N = 3–5) and control-donor (N = 3–5) mice. Feces from individual mice (250 mg feces/mouse) were resuspended and vortexed in sterile PBS with 0.05% cysteine-HCl and pelleted by sequential centrifugation (500 g for 5 minutes, 800 g for 3 minutes, and 2,000 g for 5 minutes). The supernatant was filtered through a 70 µm filter to remove particulates and concentrated by centrifugation at 5,000 g for 20 min. The pellets were then resuspended in PBS with 0.05% cysteine-HCl (pellets from 250 mg feces to 1 mL PBS) for immediate oral gavage into recipient mice. Recipient mice received 100 µL of fecal slurry supernatant samples from 2′-FL or control donors by oral gavage once a day for three consecutive days. Subsequently, recipient mice were allowed 72 hours to promote bacteria colonization before colitis was induced. Antibiotic-treated mice without FMT were used as a control.

For the induction of colitis, mice on the C57BL/6 background were treated with 3% dextran sulfate sodium ( Cat#160110, molecular weight of 36–50 kDa, MP Biomedical, Solon, OH, USA) in drinking water. Mice were euthanized on day 4 of DSS treatment. For mice receiving 2′-FL or pantothenate treatment, 2′-FL and D-calcium pantothenate were provided by oral gavage during DSS treatment to keep 2′-FL and pantothenate intake consistent with pre-treatment consumption.

Colitis induced by 2,6,4-trinitrobenzenesulfonic acid (Cat#92822, Millipore Sigma) in mice on the Balb/c background was performed by intrarectal injection of 100 mL of 70 mM TNBS in 50% ethanol. Control mice received 100 mL of 50% ethanol intrarectally. Mice were euthanized 4 days after ethanol or TNBS treatment.

Paraffin-embedded tissue sections of Swiss-rolled whole colon were stained with hematoxylin and eosin for light-microscopic examination to assess colon injury and inflammation. Samples from the entire colon were examined by a pathologist blinded to treatment conditions. DSS-induced colitis was evaluated by a modified combined scoring system (46) including A. inflammation (scale of 0–3), B. depth of inflammation (scale of 0–3), C. percentage of area involved by inflammation (0–4), D. depth of crypt damage (scale of 0–3), and E. percentage of area involved by crypt damage (0–4). The total score = (A + B) × C + D × E. The scoring system used to assess TNBS-induced colitis included A. lamina propria mononuclear cell and polymorphonuclear cell infiltration, B. enterocyte loss, C. crypt inflammation, and D. epithelial hyperplasia, which were scored from 0 to 3. The total score = A + B + C + D.

Mouse fecal metagenomics data were analyzed through the HUMAnN3 program to profile microbial metabolic pathways. HUMAnN3 pathway data were normalized by reads per kilobase and mapped reads per individual animal. Common and unique pathways between groups were identified using InteractiVenn, (47) and the 233 common pathways were further analyzed using the MaSaLin package (48) in R studio to identify the significantly different pathways (FDR < 0.05). Microbial metabolic pathways (identified by pathway names/BioCyc ID) were used to search the MetaCyc database (49) and to identify super pathways and pathway metabolite products and byproducts. Heatmap of significant pathways was generated using gplots package heatmap.2 function, hierarchical clustering average method. Circos plots were generated using the circlize library in R studio.

We used HUMAnN3 program microbial pathway abundance data for the NIH Human Microbiome WGS sequencing data set and patient metadata from the online repository (https://ibdmdb.org/tunnel/public/HMP2/WGS/1818/products↗). In this data set, 1,638 total stool samples (more than one sample was collected from each individual) with metagenomics analysis were collected, including 40 patients with ulcerative colitis and 27 non-IBD individuals. Patients who were excluded from this study were less than 12 years old, receiving antibiotics in the past 6 months or since the last stool sample collection for the study, being treated with an anti-TNF therapy, chemotherapy, or had immunosuppressants including oral corticosteroids, cancer, immune-mediated diseases, including rheumatoid arthritis, lupus, type 1 diabetes mellitus, acute gastrointestinal infection, or irritable bowel syndrome. This resulted in including 34 ulcerative colitis patients and 21 non-IBD individuals.

Human microbial pathway data (deposited data were normalized by the cpm method) were analyzed using the MaSaLin2 package (48) in R studio (https://huttenhower.sph.harvard.edu/maaslin/↗). For analysis, the non-IBD group was set as the reference, diagnosis as a fixed effect, and patient ID as a random effect to account for variability over the weeks sampled per patient (1–24 weeks of stool sampling per patient). Functional annotations using GO terms were performed through the One Codex database.

FITC-conjugated dextran dissolved in water (4,000 mol wt., Cat#FD4-100MG, Sigma) was administered rectally to mice at 2 mg/10 g body weight. Whole blood was collected via the tail vein 2 hours after FITC-dextran administration. Fluorescence intensity in plasma was analyzed using a plate reader. The concentration of FITC-dextran in plasma was determined by comparing it to the FITC-dextran standard curve.

B. infantis strains 15702 and 15697 were purchased from ATCC. Bacteria from frozen stocks were cultured in reinforced clostridial medium (Cat#DF1801-17-3, ATCC) with 180 RPM at 37°C until OD₆₀₀≈ 1 for experiments. B. infantis 15702 was cultured in sterile glass tubes with rubber stoppers that were heat sealed with a flame, and B. infantis 15697 was cultured in sterile Eppendorf tubes that were paraffin sealed and cultured in a 2.5 L anaerobic jar with the AnaeroPack System (Cat#23-246-376).

For in vitro growth experiments, B. infantis 15702 and 15697 were inoculated in RPMI 1640 medium (no glucose) supplemented with 10, 15, or 20 mg/mL of 2′-FL (1.0%, 1.5%, or 2.0%, wt/vol, respectively). To study the effects of 2′-FL on oxidative stress, ATCC 15702 was cultured in 10 or 20 µM of H₂O₂. ATCC 15697 did not numerically nor statistically respond to H₂O₂ (data not included in this manuscript). OD₆₀₀ was recorded. Independent culture experiments were performed at least three times.

A mouse colonic epithelial cell line, young adult mouse colonic (YAMC) epithelial cells, was isolated from the mouse harboring a thermolabile simian virus 40 (SV40) large T antigen from an SV40 strain, tsA58.21 under the control of an interferon (IFN)-γ-inducible H-2Kb promoter at the permissive temperature (33°C) (50). YAMC cells were maintained in Roswell Park Memorial Institute 1640 medium supplemented with 5% fetal bovine serum (FBS), 5 U/mL murine IFN-γ, 5 µg/mL insulin, 5 µg/mL transferrin, 5 ng/mL selenous acid, and 100 U/mL penicillin and streptomycin at 33°C with 5% CO₂.

The human colonic adenocarcinoma cell lines, Caco-2 (HTB-37, American Type Culture Collection, Rockville, MD, USA), were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% FBS and 100 U/mL penicillin and streptomycin at 37°C with 5% CO₂.

YAMC cells under the culture conditions were treated with D-calcium pantothenate (0.025 or 0.1 mM) for 6 hours before collection. Cells were rinsed with PBS, pelleted, and resuspended in RIPA buffer containing protease and phosphatase inhibitors, sonicated followed by pelleting of cells and collection of the supernatant. Briefly, the GPX activity assay was performed from cell lysates using a coupled assay with hydrogen peroxide as previously reported (38). The reaction mixture consisted of 50 mM potassium phosphate buffer (pH 7), 1 mM EDTA, 1 mM NaN₃, 0.2 mM NADPH, 1 E.U./mL GSSG-reductase, 1 mM GSH, and 0.25 mM H₂O₂ in a total volume of 1 mL. All reagents except the enzyme source and peroxide were combined at the beginning of each day. Protein lysates were added to 0.8 mL of the above mixture and incubated at room temperature for 5 minutes before the initiation of the reaction with the addition of 0.1 mL peroxide solution. Absorbance at 340 nm was recorded for 5 minutes, and GPX activity was calculated from the slope as micromolar NADPH oxidized per minute. Experiments were independently performed three times.

Caco2 cells were seeded in transwells (Cat#3413, Corning Incorporated) and grown to TEER as 200–400 ohms, using EVOM² Epithelial Voltohmmeter (World Precision Instruments, Inc.). After serum starvation in DMEM containing 0.5% FBS and 1% penicillin/streptomycin overnight, cells were pretreated with D-calcium pantothenate at 20 and 40 µg/mL for 2 hours followed by H₂O₂ treatment (50 µM) for 20 hours. Baseline TEER reading (0 hours) was performed before treatment, and subsequent readings were performed after the addition of H₂O₂. Readings were taken in three locations for each well and averaged for each time point. Experiments were independently performed three times.

All data are presented as mean ± standard error of the mean. A P < 0.05 was defined as statistically significant. Statistical significance was determined using one-way ANOVA analysis followed by Tukey’s multiple comparisons test or unpaired t test for comparing data from two samples using GraphPad Prism 9.0 (GraphPad Software, Inc. San Diego, CA, USA). For microbial sequencing data analysis, significance for comparison between groups was set at P < 0.05. For the PCoA, a PERMANOVA analysis was performed. Analysis of the HUMAnN3 data for the mouse and human data sets was performed using the MaSaLin2 package, with significance set at FDR < 0.05. Metabolomics data significance was set at FDR < 0.05.

Group assignment and mouse number in each group for colitis models and methods for mouse fecal microbiota analysis, untargeted metabolomic analysis, immunostaining, and RT-PCR assay are provided in Supplementary Methods.