Indole-3-lactic acid associated with Bifidobacterium-dominated microbiota significantly decreases inflammation in intestinal epithelial cells

Previously, we characterized a cohort of Bangladeshi breast fed infants revealing a high (dominating) level of B. infantis occurring frequently among this cohort (1, 23). To examine if this high bifidobacterial content was associated with specific metabolic products, we binned fecal samples for high and low bifidobacterial content and performed mass spectrometry-based metabolomics. Fecal samples were selected based on total Bifidobacterium and B. infantis relative abundance with 9 samples containing the highest abundance and 9 with the lowest abundance [23]. The low Bifidobacterium samples contained an average of 0.96% Bifidobacterium (median 1%; standard deviation of 0.7%), and will be referred to as ‘low Bifidobacterium” group. The high Bifidobacterium samples had an average of 93.56% (median 92%; standard deviation 2.55%) Bifidobacterium and will be referred to as ‘high Bifidobacterium’ group. In this high Bifidobacterium group, the next most abundant genera were represented by Streptococcus, Lactobacillus and Enterococcus. Conversely, the ‘low Bifidobacterium’ fecal samples were populated with various genera, including predominantly Streptococcus and Enterococcus genera (Fig. 1a). In the samples with high Bifidobacterium abundance, the mean relative abundance of B. infantis represented 97.22% (median 1.0; standard deviation 6.53%) from the total Bifidobacterium population. The gut microbial weighted (betadisper: F.model = 4.96, P = 0.048; ADONIS: F.model = 34.0, R2 = 0.680, P < 0.01) and unweighted (betadisper: F.model = 0.446, P = 0.50; ADONIS: F.model = 2.87, R2 = 0.152, P < 0.01) β diversity was significantly different between high and low Bifidobacterium groups (Fig. 1b and c).

Short chain fatty acids and the indole derivative ILA were measured in the 18 infant fecal samples using mass spectrometry. The concentration of ILA was significantly elevated in fecal samples from infants with a high versus low Bifidobacterium-dominated microbiome (Fig. 1d, e; P < 0.0001). The B. infantis-dominated microbiome population produced a greater than 19-fold increase in ILA concentration compared to the low-Bifidobacterium dominated infant microbiome (Fig. 1d, e;). The concentration of acetic and lactic acid was significantly increased in the fecal samples from the high versus low Bifidobacterium groups (Fig. 1d; P = 0.0353, P = 0.0095, respectively). There was also a significant decrease in the concentration of propionic acid in the high versus low Bifidobacterium groups (P = 0.049). There was no significant difference in the concentrations of the other detectable SCFA (butyric, formic, isobutyric, and valeric) between the high versus low bifidobacterial groups.

The observation of increased ILA from B. infantis-dominated feces suggested this metabolite emanated from B. infantis. One mechanism by which B. infantis dominates the infant gut is through the consumption of complex human milk oligosaccharides (HMOs) in milk (15). Production of acetate, lactate and formate by Bifidobacterium species has previously been shown to be dependent on the carbon source in the growth medium [24] but how growth on HMOs influences production of metabolites from B. infantis has not been described. In order to determine the differences in the metabolites produced by B. infantis grown on two different carbon sources, we performed untargeted metabolomic analysis of the supernatant from B. infantis grown in chemically defined RPMI 1640 medium supplemented with either lactose or HMO. B. infantis grown on HMO or lactose produced a large number of metabolites, including amino acids, organic acids and carbohydrates (Supplemental Fig. 1). The data show significantly increased production of acetate, glutamate, glycerol, glycine, indole-3-lactate, leucine, pyruvate, and valine in HMO-supplemented compared to lactose-supplemented medium (Supplemental Fig. 1; all P < 0.0001). HMO-supplemented growth of B. infantis produced greater than 4-fold more ILA compared to bacteria grown in lactose-supplemented media (Supplemental Fig. 1, insert; P < 0.0001), suggesting that production of ILA is increased when B. infantis is consuming milk glycans.

Recently, it was shown that supplementing the diet of weanling piglets with tryptophan decreased expression of inflammatory markers in the large intestine [25] which may be due to the increased production of tryptophan metabolites. ILA is a known catabolite of tryptophan [26] and ILA was significantly elevated when B. infantis was grown on HMOs and in the feces containing high levels of B. infantis. Therefore, we investigated whether addition of tryptophan would alter metabolite composition when B. infantis was grown in RPMI 1640 supplemented with either HMO or lactose in the presence of increasing concentrations of tryptophan. Metabolomics analysis revealed that tryptophan produced an increase in the production of glutamate, glycerol, ILA, leucine, and valine in HMO-supplemented compared to lactose-supplemented medium (Supplemental Fig. 1; P = 0.0001, P = 0.0077, P = 0.0009, P = 0.0001, P = 0.0040, respectively). Increasing the tryptophan also significantly increased the catabolism of 3’FL, 6′SL, cysteine, glucose and LNnT in HMO (Fig. 2: P = 0.0037; P < 0.0001; P = 0.0003; P < 0.0001; P < 0.0001, respectively).

Given the differential amounts of ILA between feces containing high versus low abundance of B. infantis, combined with the increase in ILA production from B. infantis grown on HMO, and its increase concentration in the presence of tryptophan in the growth medium, ILA appeared to be an appropriate target for further study. ILA production differs from the other significant metabolites that were also increased by growth on HMOs compared to lactose as it is not directly involved in energy metabolism. ILA and other tryptophan metabolites are known to produce beneficial effects in the host [27–29]. We hypothesized that ILA may be responsible, at least in part, for the anti-inflammatory effects associated with B. infantis.

To test for a potential anti-inflammatory role of ILA in the intestinal mucosa, studies were performed in a macrophage cell line and intestinal epithelial cell lines in vitro. Initial experiments to show a potential anti-inflammatory action was performed in the mouse macrophage reporter cell line RAW blue cells. Subsequent experiments were performed in the intestinal epithelial cell lines Caco2 and HT-29. Two different cell lines were used to ensure that the anti-inflammatory response to ILA could be reproduced in both cell lines. Experiments on dissecting the AhR and Nrf2 pathway were performed in HT-29 cells.

Given the central role NFκB plays in the activation of innate immune responses [30], we first evaluated the ability of ILA to modulate LPS-induced immune activation using the mouse macrophage reporter cell line RAW blue cells. Overnight treatment of the mouse RAW blue cells with LPS (10 μg/ml) significantly increased activation of NFκB (Fig. 2a, P < 0.0001); pretreatment with ILA (0.1 mM–10 mM) significantly inhibited LPS-induced NFκB activation in a dose-dependent manner (Fig. 2a; P < 0.0001). Similar anti-inflammatory effects were seen in Caco-2 cells; pretreatment with ILA significantly decreased LPS-induced production of IL-8 (Fig. 2b; P = 0.0006; P = 0.0021).

Similar anti-inflammatory actions were seen in HT-29 cells. LPS is a potent stimulus for release of the pro-inflammatory cytokine tumor necrosis factor-alpha (TNF-α) [31], therefore, we determined the ability of ILA to decrease inflammatory responses to TNF-α. HT-29 cells were incubated with TNF-α (20 ng/ml) for 1 h in the absence (control) or presence of ILA (0.1-10 mM). Cellular responses were analyzed using qPCR for IL-8 (immune activation), serotonin transporter (SERT) and tryptophan hydroxylase 1 (TPH1) (ILA is an indole derivative which have previously been shown to interact with serotonin receptors), and β-defensin 2 (hBD2; antimicrobial defense protein previously shown to be activated by bifidobacteria). TNF-α induced a robust increase in IL-8 mRNA expression (P < 0.0001) but there was no change in expression of SERT, hBD2, or TPH1 (Fig. 3). The increase in expression of IL-8 induced by TNF-α was significantly decreased by treatment with ILA (Fig. 3a; P < 0.0001 for 0.1, 1, and 10 mM ILA); there was no significant difference in expression of IL-8 between cells treated with TNF-α in the presence of ILA and those treated with vehicle alone (Fig. 3a, P = 0.061, P = 0.90 P = 0.59 for 0.1, 1, and 10 mM ILA, respectively). Further, expression of SERT and hBD2 were significantly increased in cells treated with ILA (10 mM) (Fig. 4b, c, P = 0.016 and P = 0.042). Expression of TPH-1 was not affected by treatment with TNF-α in the presence or absence of ILA (Fig. 3d).

There is evidence that tryptophan catabolites contribute to gut homeostasis via modulating mucosal immune responses through an AhR signaling pathway [26, 32]. AhR is a nuclear receptor that when activated translocates to the nucleus to increase transcription of AhR-regulated genes. Previous work has indicated that AhR changes its cellular location based on cell density [33]. To confirm this occurs in intestinal cells, we measured protein expression of AhR by western blot in cytosolic and nuclear extracts from Caco-2 cell cultured at either low (50% confluency) or high (100% confluency). Consistent with published data from other cell types [28], intestinal epithelial cells grown at high density have increased cytosolic AhR compared to cells plated at low density (Fig. 4a, P = 0.01). For a preliminary test that ILA can activate AhR, AhR expression was measured in the nuclear and cytosolic fractions in Caco-2 cells grown at high density in response to ILA. There was a trend for increased AhR expression in the nucleus and decreased AhR expression in the cytosol, indicating that ILA may activate AhR (Fig. 4c, d).

In order to further understand the potential role of AhR in the anti-inflammatory response to AhR, we used HT-29 cells. Treatment of HT-29 cells with ILA (0.1–10 mM) induced a robust increase of mRNA transcript of AhR regulated gene cytochrome p450 1A1 (CYP1A1), which was significantly decreased by pretreatment with the AhR antagonist (CH-223191; 10 μM) (Fig. 5a, P = 0.0000003) or Nrf2 antagonist (SML1833; 4 μM) (Fig. 5a, P = 0.0001066). There is considerable evidence for cross talk between the AhR and Nrf2 pathways, which led us to test activation of the Nrf2 pathway [34]. Upon exposure to ILA, there were dose-dependent increases in Nrf2 activated genes glutathione peroxidase 2 (GPX2) and super oxide dismutase 2 (SOD2) and an increase in NAD(P) H dehydrogenase (NQO1) was observed. Inhibition of either AhR or Nrf2 completely suppresses ILA activation of GPX2, SOD2, and NQO1 genes (Fig. 5a-d, all P < 0.001).

To test if the AhR or Nrf2 pathway is involved in the ILA-induced suppression of pro-inflammatory IL-8 production, HT-29 cells were treated with TNF-α (20 ng/ml) and ILA (0.1-10 mM) in the presence of either an AhR antagonist or a Nrf2 inhibitor. ILA treatment produces a dose-dependent decrease in TNF-α-induced IL-8 expression which is inhibited by pretreatment with either an AhR antagonist or a Nrf2 inhibitor (Fig. 5e, all P < 0.0001).

We utilized 18 fecal samples (9 with highest and 9 with lowest Bifidobacterium abundance) originally collected as part of a larger trial in Bangladesh (clinicaltrials.gov identifier: NCT01583972↗ and NCT02027610↗). Detailed study design, sample collection procedures, and initial results have been reported earlier [23, 58–60]. Briefly, parents of infants born at the Maternal and Child Health Training Institute in Dhaka, Bangladesh were approached during the third trimester of pregnancy and informed consent was obtained. Infants were supplemented with a high dose vitamin A within 48 h of birth and stool samples were collected at 6 weeks of age and stored at − 70 C, as part of the original study. The study was approved by the Ethical Review Committee of the International Centre for Diarrheal Disease Research, Bangladesh (ICDDR,B) and by the Human Studies Committee of the World Health Organization.

Whole stool samples were collected and were frozen at − 70 °C until assay. Detailed of the 16S assay including randomization, V4 amplicon library preparation, and bioinformatics analysis have been reported previously [59]. In brief, stool DNA samples were randomized, total fecal DNA was extracted and mixed template 16S V4 amplicon library was prepared using the primer sets (515F and barcoded 806R) [61] and sequenced using the Illumina MiSEQ platform with 2x250bp paired-end sequencing. Raw paired-end sequences were analyzed using the open-source software Quantitative Insights Into Microbial Ecology (QIIME) version 1.8 [62]. Taxonomy was assigned using the Greengenes May 2013 reference database [63] at the threshold of 97% pairwise identity. To estimate Bifidobacterium species and subspecies, we used Bif-TRFLP (Bifidobacterium-specific terminal restriction fragment length polymorphism) assay and BLIR (Bifidobacterium longum subsp. longum-infantis Ratio) assay, respectively, as has been reported earlier [23]. The relative abundance of the infant’s gut microbiota presented as combined results from 16S QIIME, Bif-TRFLP, and BLIR assay. The 16S sequence data is available at NCBI Sequence Read Archive (SRA) database under the BioProject ID “PRJNA636907” and “PRJNA636905”.

Pre-weighed fecal samples are diluted, homogenized, and centrifuged before extraction of the fecal supernatants. 20 μL of supernatant is reacted with 20 μL of 200 mM N-(3-Dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride (1-EDC HCl) in 5% pyridine and 40 μL of 100 mM 2-nitrophenylhydrazine (2-NPH) in 80% ACN/H₂O (v/v) with 50 mM HCl. The samples were reacted for 30 min at 40 °C before being diluted with 400 μL of 10% ACN/H₂O. Samples were centrifuged for 20 min before transferring to a 96-well injection plate for QqQ LC-MS/MS analysis. Dynamic MRM method was used in positive mode for detection and quantitation of the compounds. The quantification is absolute concentration and are shown as mmol/kg of original fecal sample wet weight [64].

Single colony isolates of B. infantis ATCC 15697 were grown from frozen stocks on de Mann, Rogosa, and Sharpe medium (MRS) supplemented with 2% (wt/vol) carbon source and 0.25% (wt/vol) L-cysteine (Sigma-Aldrich) for 18 h, centrifuged for 10 min at 4000 rpm, resuspended in glucose-free RPMI 1640 (Life Sciences) then inoculated in three biological replicates at 1:100 ratio in RPMI 1640 (chemically defined) containing 2% of human milk oligosaccharides (HMO) previously purified in the UCD Milk Processing Lab or lactose (control). To obtain lactose-free HMOs, a purification approach of pooled donor milk based on lactose hydrolysis, yeast consumption of the resulting monosaccharides and further clean-up and concentration by sequential membrane filtration was used [65]. The freeze-dried powder was characterized by mass spectrometry and analyzed by high-performance anion-exchange chromatography with pulsed amperometric detection using a previously published method tailored for oligosaccharide quantification in milk [66] (Supplemental Table 1). B. infantis was incubated at 37 °C until early stationary phase under anaerobic conditions (5% carbon dioxide, 5% hydrogen, and 90% nitrogen) in a vinyl anaerobic chamber (Coy Laboratory Products). Bacterial growth was monitored by optical density (OD₆₀₀) and an aliquot of the bacterial suspension was taken for determination of the bacterial counts by plating on MRS Agar. Sample preparation for exometabolome analysis was performed as described by Smart et al [67]. Cell-free supernatants were obtained by centrifuging the bacterial cultures at 4000 rpm for 15 min, followed by filtration using 0.22 μm filters (Millipore) to remove microbial cells and addition of 0.2 μmol per sample of isobutyric acid as internal standard. Samples were distributed in 100uL aliquots and stored at − 80 °C until processing. Uninoculated growth media was collected and used as controls for exometabolome analysis.

Frozen supernatant samples were thawed and filtered through 3000 MW cutoff filters to remove lipids and proteins. Samples were prepared by addition of 65 μL of an internal standard containing 5 mM DSS-d6 (2,2,3,3,4,4-d6–3-(trimethylsilyl)-1-propane sulfonic acid), 0.2% NaN₃ in 99% D₂O to 585 μL of supernatant. Sample pH was subsequently adjusted to 6.8 through the addition of small amounts of NaOH or HCl. 600 μL aliquots were transferred to 5 mm Bruker NMR tubes, and stored at 4 °C until NMR acquisition which was within 24 h of sample preparation. NMR spectra were acquired as previously described by He et al. [68] using a Bruker Avance 600 MHz NMR equipped with a SampleJet autosampler using a NOESY-presaturation pulse sequence (noesypr) at 25 °C. Once acquired, all spectra were processed as described [68] using Chenomx NMR Suite v6.1 Processor (Chenomx Inc., Edmonton, Canada). Metabolite quantification was achieved using the 600-MHz library from Chenomx NMR Suite v6.1 Profiler, which uses the concentration of a known reference signal (in this case DSS) to determine the concentration of individual compounds. Metabolites were quantified in micro-molar (μM) units and exported from Chenomx for statistical analysis. Metabolite concentrations were normalized to bacterial CFU and were calculated relative to the concentration found in the growth media whereby a positive number indicates microbial production and a negative indicates consumption.

RAW-Blue cells (InvioGen), a mouse macrophage cell line engineered with secreted embryonic alkaline phosphatase (SEAP) reporter system, was used to evaluate LPS-induced NFκB activation in the presence or absence of ILA in vitro. ILA was added to RAW-Blue cells for 1 h before being exposed to LPS. After incubating for 4 h, SEAP levels were quantified by developing supernatant with QUANTI-Blue substrate for 1 h and reading absorption at 620 nm as stated in the manufacturer’s instructions (InvioGen). Caco-2 and HT-29 cells were purchased from ATCC and grown to the specifications. To assess IL-8 production in response to LPS, Caco-2 cells were pre-incubated with ILA for 1 h before overnight exposure to LPS. For gene expression experiments, cells were incubated with ILA in the presence of TNF-α for 1 h or pretreated with AhR or Nrf2 inhibitors for 1 h before 1 h exposure to ILA. Experiments were performed for a total of 6–8 biological replicates. ILA, AhR antagonist CH-223191, Nrf2 antagonist SML1833, TNF-α and LPS 0111:B4 were purchased from Sigma. GAPDH, anti-rabbit HRP and anti-biotin HRP antibodies were purchased from Cell Signaling Technology. AhR antibody was purchased from Invitrogen. HDAC1 antibody was from Santa Cruz Biotechnologies.

ELISA Duoset was used to measure IL-8 levels in Caco-2 supernatants according to manufacturer’s instructions (R&D Systems).

Nuclear and cytosolic fractions were prepared as described previously [69]. Cell lysates were evaluated for total protein content using Bradford assay and overall concentration for each samples adjusted with deionized water and boiled with 2x Laemmli reducing buffer. Individual samples were resolved in SDS-PAGE as per standard protocol. Signals were developed and detected with appropriate antibodies and bands were analyzed using BioRad Imager. GAPDH and HDAC1 were used as internal loading controls for cytosolic and nuclear fractions.

Total RNA was extracted from human Caco-2 or HT-29 using TRIzol reagent (Invitrogen) following the manufacturer’s instructions. RNA was reverse transcribed with Bio Rad iScript cDNA synthesis kit. cDNA template was combined with PowerUp SYBR Green Master Mix (ThermoFisher) and quantitative PCR was run using Applied Biosystems QuantStudio 6 Real-Time PCR System. Primers used to amplify were cytochrome p450 1a1, glutathione peroxidase 2, superoxide dismutase 2, NAD(P) H dehydrogenase, interleukin-8, beta-defensin2, serotonin transporter, tryptophan-hydroxylase 1, 60S acidic ribosomal protein P13A. Primer sequences are provided in Supplemental Table. 2

Student’s t test or One-way ANOVA with Bonferroni Multiple comparison test was performed using GraphPad Prism. A P value of < 0.05 was considered statistically significant. For ILA dose curves, the mean ± SEM of relative expression values were fitted with a 3-paramenter dose response curve vs the log-transformed ILA concentration (Prism, v. 8.0, GraphPad Software, San Diego, CA). Statistical significance of the effect of each treatment on the means of relative expression of each target gene at the highest ILA concentration (10 mM) was determined using a one-way ANOVA analysis with a TukeyHSD posthoc comparison (R, version 3.5.2 [70]). For IL-8, the statistical differences between the means in the absence of ILA was as also determined by this method. Differences in the homogeneity of microbiota between high and low Bifidobacterium groups were determined by using the PERMDISP2 function of R Package Vegan [71] with 999 permutations. Differences in microbial community β-diversity were tested by ADONIS (perMANOVA) in the R Package Vegan. Principal coordinate (PCoA) analysis was carried out by PhyloSeq [72]. Correlation analysis was carried out by Spearman correlation. Statistical analyses were performed by using R version 3.5.2 [70] and Prism, v. 8.0 (GraphPad Software, San Diego, CA). Graphs were prepared by GGplot2 [73] and Prism, v. 8.0 (GraphPad Software, San Diego, CA).

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.