Dietary Indole-3-Carbinol Activates AhR in the Gut, Alters Th17-Microbe Interactions, and Exacerbates Insulitis in NOD Mice

NOD/ShiLtJ (NOD) mice were obtained from The Jackson Laboratory and maintained in the specific pathogen-free animal facility at Oregon State University. NOD.AhR-/- mice were generated by backcrossing B6.129-AHRtm1Bra/J onto the NOD/LtJ background for more than 13 generations. All experiments used female littermate-matched mice. All animal procedures were carried out following protocols approved by the Institutional Animal Care and Use Committee at Oregon State University.

Cl-BBQ (11-chloro-7H-benzimidazo[2,1-a]benzo[de]Iso-quinolin-7-one; Chembridge) or I3C (Sigma) were dissolved in DMSO-Cremaphor-Peceol (30%:20%:50%) by sonicating in a 37 degree water bath for 1 h. Female NOD mice were administered 200 ul of Cl-BBQ (45 mg/kg), I3C (250 mg/kg) or the vehicle control by oral gavage. Mice were sacrificed at 6 and 24 h after oral gavage to measure AhR activation.

NOD mice were fed normal chow until 7 weeks of age when they were transitioned to either a synthetic diet (AIN93M; Research Diets) alone or supplemented with 2,000 ppm I3C. Mice were maintained on these diets through the remainder of the study (12 weeks). Food consumption was tracked daily by weighing the remaining food for the first week of the study and then weekly from weeks 8 to 12. Body weight was likewise measured daily for the first week and then once a week for the remainder of the study. Two independent experiments containing three or four individual birth cohorts were conducted.

The liver and small intestine were removed from mice and stored in RNAlater for immediate stabilization prior to RNA isolation using RNeasy Mini Kit (Qiagen). cDNA was synthesized using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). qPCR reactions were performed on an Agilent Stratagene Real-Time PCR system (Applied Biosystems) using SYBR Green/ROX Master Mix from SA Biosciences. Cyp1a1 levels were normalized to Actb using primers from SA Biosciences.

Insulitis was scored on sequential hematoxylin and eosin (H&E) stained pancreas sections separated by 200 μM with at least 50 islets scored per pancreas. Islets were scored as no infiltration, less than 50% infiltration or greater than 50% infiltration.

At 12 weeks of age, mice were sacrificed, pancreatic lymph node (PLN), spleen, and the small intestine were excised. Single cell suspensions were prepared from the spleen, PLN, and Peyer’s Patches by mechanical disruption between frosted slides. Splenoyctes were further processed by hypotonic red blood cell lysis. Isolation of lamina propria (LP) cells and intraepithelial lymphocytes (IEL) from the small intestine was performed as previously described (29). Briefly, Peyer’s patches were excised and the epithelial layer of the small intestine was isolated through sequential washes in a 5 mM EDTA and 0.145 mg/ml DTT solution while stirring at 37°C. LP cells were isolated from the remaining intestinal tissue by mincing and digesting the tissue using 0.2 U/ml Liberase TM and 0.05% DNase while shaking at 37°C. The digested tissue was washed three times with a 3% FBS solution and successively filtered after each wash through two 70 μm filters and one 40 μm filter in preparation for cell staining.

Cells were stained with fixable viability dye (eBioscience), Fc receptors were blocked with rat IgG (Jackson ImmunoResearch) and the cells were stained with the following antibodies CD45 (30-F11), CD4 (RM4-4), Nrp1 (3DS304M) from eBioscience, Lag3 (C9B7W), CD25 (PC61) and CD210 (1B1.3a), CD8 (53-6.7) from BD Biosciences and Tim3 from Biolegend. For intracellular staining, cells were fixed and permeabilized using the Foxp3 Fixation/Permeabilization buffer (eBioscience) and stained with Tbet (4B10) and Foxp3 (FJK168) from eBioscience, and RORγt (Q31-378) from BD Biosciences. For cytokine staining in splenoyctes, cells were stimulated with PMA, ionomycin, brefeldin A, and monensin (eBioscience) for 4 h in culture and stained with IFNγ (XMG1.2), IL-10 (JES5-16ES), IL-22 (1H8PWSR) from eBioscience and IL-17 (TC11-18H10) from BD Biosciences.

Data were acquired on a Cytoflex flow cytometer (Beckman Coulter). Data were compensated and analyzed using FlowJo (Treestar) software. Fluorescence minus one (FMO) controls were used for setting gates for analysis.

For microbial measurements, fresh stool pellets were collected at 7, 8, and 12 weeks of age and immediately stored at -80°C. To extract microbial DNA, frozen fecal pellets were resuspended in 1.4 ml ASL buffer (Qiagen) and homogenized with 2.8 mm ceramic beads followed by 0.5 mm glass beads using an OMNI Bead Ruptor (OMNI International). DNA was extracted from the entire resulting suspension using QiaAmp mini stool kit (Qiagen) according to manufacturer’s protocol. DNA was quantified using Qubit broad range DNA assay (LifeTechnologies).

The primers 319F and 806R were used to amplify the V3-V4 domain of the 16S rRNA using a two-step PCR procedure. In step one of the amplification procedure, both forward and reverse primers contained an Illumina tag sequence (bold), a variable length spacer (no spacer, C, TC, or ATC for 319F; no spacer, G, TG, ATG for 806R) to increase diversity and improve the quality of the sequencing run, a linker sequence (italicized), and the 16S target sequence (underlined). In step two, each sample was barcoded with a unique forward and reverse barcode combination using forward primers (AATGATACGGCGACCACCGAGATCTACACNNNNNNNNTCGTCGGCAGCGTC) with an Illumina P5 adapter sequence (bold), a unique 8 nt barcode (N), a partial matching sequence of the forward adapter used in step one (underlined), and reverse primers (CAAGCAGAAGACGGCATACGAGATNNNNNNNNGTCTCGTGGGCTCGG) with an Illumina P7 adapter sequence (bold), unique 8 nt barcode (N), and a partial matching sequence of the reverse adapter used in step one (underlined). The final product was quantified on the Qubit instrument using the Qubit Broad Range DNA kit (Invitrogen) and individual amplicons were pooled in equal concentrations. The pooled library was cleaned utilizing Ampure XP beads (Beckman Coulter) then the band of interest was further subjected to isolation via gel electrophoresis on a 1.5% Blue Pippin HT gel (Sage Science). The library was quantified via qPCR followed by 300-bp paired-end sequencing using an Illumina MiSeq instrument in the Genome Center DNA Technologies Core, University of California, Davis. The QIIME 2 (30) bioinformatics pipeline (v. 2018.8.0) was used to demultiplex and quality filter the forward-end fastq files. Denoising was performed using DADA2 (31). The raw data can be accessed at NCBI Sequence Read Archive (SRA) (Accession #PRJNA679964).

Host parameters were first normalized by birth cohort. Host parameters that were analyzed in both independent experiments were subsequently normalized across the two studies. For microbial data, after excluding amplicon sequence variant (ASV) singletons, a threshold of 99.5% cumulative abundance across all samples in both experiments was used to select the most abundant ASVs for downstream analysis. Following this, ASVs were relativized per million and quantile normalized (per experiment), then log2 transformed for downstream analysis.

The network analysis has been performed as described in previous papers (32, 33), except for the minor modifications outlined below. Spearman rank correlations were conducted for all experimental groups (Control and I3C for each of the two experiments) between all pairs of ASVs and host parameters. Edges that did not demonstrate the same sign of correlation direction (positive or negative) across all groups and those that were not compliant with causality principles (34) were removed. Similarly, those edges that did not contain nodes consistent in fold change direction across experiments were also removed. The metacor function of the meta package [v 4.12-0 (35);] in R (v 3.5.1) was used to calculate the fixed effect model p-values based on Fisher’s Z-Transformation of proportions of the correlations across the four groups. Prior to calculating FDRs, the correlations between microbiota and phenotypes were first filtered for those that had an individual p-value <0.6 across all groups, with the exception of the PP_RORgt+Foxp3-<==>ASV21 and IEL_RORgt+Foxp3-<==>ASV290 edges (FDR = 0.135 for both). These two additional edges were kept in the network despite not passing individual p-value thresholds due to the high biological importance of the RORγt⁺Foxp3^– cell populations. Therefore, we checked for edges connecting to that cell population that barely missed making the network. FDR was then calculated on the correlation p-values for each of the following groups: (1) between microbes; (2) between microbes and host parameters; (3) between host parameters. Microbe-host parameter edges with an FDR < 0.15 were retained in the network. Edges between microbiota were required to have an FDR < 0.05 while those between host parameters were required to have an FDR <0.1.

Networks were visualized using Cytoscape v3.7.2 (36). Silva IDs were used in labeling the important microbes (as determined by their high normalized bipartite betweenness centrality [BiBC] and degree). The Python module NetworkX v2.2 was used to calculate BiBC and degree between groups, as well as to randomly generate the 10,000 networks used in validating the BiBC and degree results. BiBC values were calculated as previously described (32), then normalized by the number of nodes in each group. The 10,000 binomial random (Erdos-Renyi) networks were generated from the G(m,n) ensemble with m = 221 (to match the number of edges in the real network) and n = 109 (the number of nodes). The 2D contour histogram BiBC-degree distribution was plotted using the online tool Plotly (https://plot.ly/↗). Probability density was used as a measurement of the likelihood of randomly finding a node with the given BiBC and degree (or higher). A large value (i.e. a dark-colored space in the contour map) indicates that a node in that area typically occurs in random networks size-matched to the network generated from the data.

With the exception of the network analyses, statistical analyses were performed using Graphpad Prism. For insulitis and host parameters, data were normalized by birth cohort/littermates. For comparing two groups, a Student’s t test was performed. For multiple comparisons, one-way ANOVA with Tukey’s test was used. P <0.05 was considered statistically significant. All plotted data points represent an individual mouse.

Activation of the AhR by I3C supplementation has not been previously studied in NOD mice. To determine the extent by which I3C can induce AhR activation, we first measured activation following oral gavage with I3C as compared to Cl-BBQ, a high affinity AhR ligand previously shown to suppress T1D in NOD mice (10, 12). In C57BL/6 mice, 250 mg/kg of I3C (or dietary equivalent) is on the higher range of previously reported doses used to activate the AhR (22–24, 26, 27, 37) [; higher doses (> 500 mg/kg) of I3C are associated with neurotoxicity and increased mortality (20)]. To account for the AhR^d allele, which has a ~10 fold lower sensitivity to AhR ligands (12, 38), a dose 250 mg/kg I3C was chosen to maximize AhR activation and minimize toxicity.

NOD mice were administered I3C by oral gavage and Cyp1a1 induction was measured in the liver, pancreatic lymph node, and small intestine (duodenum, jejunum, and ileum). AhR response elements upstream of Cyp1a1 make it a highly sensitive target of AhR and, as a result, the induction of Cyp1a1 is a commonly used biomarker for AhR activation (39). The positive control, Cl-BBQ, resulted in sustained Cyp1a1 induction in the liver and pancreatic lymph nodes at 6 and 24 h post-oral gavage. In contrast, I3C only resulted in transient Cyp1a1 induction, which was 4- and 20-fold lower in the liver and pancreatic lymph node, respectively, at 6 h post-oral gavage. By 24 h after I3C administration, Cyp1a1 was back to baseline (Figure 1A). Furthermore, I3C did not induce Cyp1a1 in the duodenum, jejunum, and ileum of the small intestine following oral gavage (Figure 1B).

In contrast to gavage administration, one week of diet supplementation with I3C (2,000 ppm, equivalent to 250 mg/kg/day based on food consumption) was capable of inducing strong Cyp1a1 in the small intestine (median of ~1,000-fold increase; Figure 1C). However, as with oral gavage, AhR activation was likewise limited systemically following dietary exposure. Given the apparent lack of systemic AhR activation following dietary I3C supplementation, it was difficult to predict how it would impact T1D pathogenesis. However, the ability of I3C to induce Cyp1a1 locally in the intestine (Figure 1D), provided an unexpected opportunity to determine how AhR activation in the gut alters the development of T1D.

To determine if strong AhR activation in the intestine was sufficient to suppress insulitis, NOD mice were fed a diet supplemented with 2,000 ppm I3C from seven to twelve weeks of age, during which time insulitis is known to progress. Initiating the dietary regimen at this timepont was selected to correspond to the treatment timing in our previous studies with Cl-BBQ and TCDD (10). At seven weeks of age, NOD mice were randomized by litter, cage, and weight then assigned to a control or I3C-diet fed group (Figure 2A). Mice fed an I3C-supplemented diet showed no difference in food consumption or body weight compared to control mice (Figures 2B, C). Mice averaged 2–2.5 g of food/week resulting in an average exposure of 190–260 mg/kg/day of I3C.

At 12 weeks of age, mice were euthanized and islet infiltration was scored in the sectioned pancreas. Unexpectedly, mice given the diet supplemented with I3C had significantly increased insulitis. The increase in insulitis was primarily a reflection of the significant increase in the percentage of islets per pancreas that were heavily infiltrated (50% infiltration, p < 0.05) in mice fed the I3C-supplemented diet; the percentage of islets with <50% infiltration was unchanged between treatment groups. This pattern was observed in two independent experiments (Figure 2D).

Th17 cells and Tregs (Foxp3⁺ and Tr1) are targets of AhR activation and implicated in promoting and inhibiting the autoimmune destruction of beta cells, respectively (16, 18, 19, 40–42). Therefore, alterations in markers representing CD4⁺ T cell helper subsets were examined in the spleen, pancreatic lymph node, Peyer’s patches, the small intestine lamina propria, and the small intestine intraepithelial cells. Significant increases were found in CD4⁺Foxp3^–RORγt⁺ Th17 cells in the lamina propria (2.3-fold increase), intraepithelial layer (2.7-fold increase), and Peyer’s Patches (1.6-fold increase; Figures 3A–D). A subset of lamina propria CD4⁺Foxp3^– cells expressing high levels of RORγt expressed the T cell activation marker CD25⁺ and were significantly increased as well (1.7-fold increase; Figure 3E). Confirming that dietary I3C increased Th17 cells, LPL cells from a subset of mice were stimulated with PMA/ionomycin and IL-17 was measured by flow cytometry (Figure 3F). No significant changes in Tr1 cells, Th1 cells, and Th17 cells were found in the spleen or pancreatic lymph nodes, although a trend toward increased Th17 cells was observed in the PLN (Supplementary Figure 1), consistent with the minimal AhR activation at these sites. While the majority of significant changes in immune cell markers were identified in the intestine, some minor yet significant populations were changed in the spleen. Significant changes in all analyzed populations are listed in Figure 3G.

Since modulation of immune cell populations in the gut by dietary I3C can arise from AhR-mediated changes in gut microbiota diversity, we examined whether dietary I3C altered gut microbial composition in NOD mice. Stool samples were collected prior to the dietary intervention at 7 weeks of age, after one week of the diet, and at the time of sacrifice at 12 weeks of age.

Within one week of starting the new diets, there was a strong shift in the microbial communities of both groups of mice fed the synthetic control diet or the I3C-supplemented diet (Figures 4A, B). Samples from mice in the control diet group and, to a greater extent, the I3C-supplemented diet group separated from the 7 week samples on the PC1 axis, which accounted for 50–58% of diversity in two independent experiments. Samples collected from the control diet group separated from the I3C-supplemented group on the PC2 axis, which accounted for 12–17% of diversity. Interestingly, changes in microbial diversity were established within the first week of the synthetic diet (Figure 4A).

At the phylum level (Figure 4C), the synthetic diet, regardless of supplementation, reduced the abundance of Tenericutes and increased Verrucomicrobia. The samples from mice fed the I3C-supplemented diet diverged from the control-diet group with an increase in Proteobacteria and Verrucomicrobia. In both NOD mice and patients with T1D, an increased ratio of Bacteriodetes to Firmicutes is implicated in the development of T1D (5). Consistent with dietary I3C increasing insulitis, there was a significant increase in the ratio of Bacteriodetes : Firmicutes at 12 weeks of age in mice fed the I3C supplemented diet (Figure 4D). This corresponded with a positive trend (p = 0.06) between the Bacteriodetes : Firmicutes ratio and percentage of islets with greater than 50% infiltration (Figure 4E).

To begin to identify which ASVs may be implicated in the immunological changes that arose following dietary I3C, we constructed a transkingdom network. Transkingdom networks have been successfully used to tease out causal interactions between mammalian host phenotypes and their colonizing microbiota (43). Here, the transkingdom network (Figure 5A) was constructed by computing correlations between host factors (hexagons) that significantly changed in mice fed the I3C-supplemented diet (Figure 3G) and ASVs identified at 12 weeks of age (circles). The network retained 108 nodes (red: increased and blue: decreased in I3C treated mice) and 221 edges, of which there were 179 positive correlations (red lines) and 42 negative correlations (blue lines).

The network was then interrogated to predict which host factors and microbes were most likely to play a role in the response to dietary I3C. We calculated two properties of the nodes, degree and bipartite betweenness centrality (BiBC) (Figure 5B); degree is the number of nodes each individual node interacts with, and measures the direct impact of one node has on other parameters in the system; bipartite betweenness centrality calculates the number of times the node lies in the shortest path connecting two groups of nodes (host parameters and ASVs). A node with a high BiBC acts as a bottleneck and thus is a regulator between processes of the system. Collectively nodes with high degrees and centrality are predicted to be critical in defining host-microbe interactions that are involved in the response to I3C. By plotting bipartite betweenness centrality and degree, we identified three ASVs, one with a high BiBC (Intestinimonas) and two with both high BiBC and high degree (Ruminiclostridium 9 and unclassified Lachnospiraceae) which likely regulate the abundance of other gut microbes. Intestinimonas, Ruminiclostridium, and Lachnospiraceae were decreased in I3C treated mice and are members of phylum Firmicutes. On the host side, two parameters showed high BiBC, intraepithelial CD4⁺RORγt⁺Foxp3^– cells (Figure 3D) and CD4⁺CD25⁺Foxp3-RORγt(high) lamina propria cells (Figure 3E) and therefore likely influence or are influenced by microbial interactions.

The statistical importance of these five nodes was further validated by randomly generating 10,000 networks and comparing the degree and BiBC of these specific nodes to those in a randomly generated model. Indeed, the degrees and BiBC of the five nodes had a low probability of occurring randomly (Figure 5C).

While Ruminiclostridium 9, Intestinimonas, and Lachnospiraceae likely play a role in host interactions, it is unclear if their change in abundance was due to, or resulted in, changes in the intestinal immune response after I3C treatment. Therefore, we looked at the temporal abundance of these microbes (Figures 5D–F). Within one week of switching from the normal chow diet to the synthetic diet there was an increase in abundance in both Ruminiclostridium 9 and Intestinimonas regardless of I3C supplementation. However, the increase in abundance of Intestinimonas was larger in the control group in comparison to the I3C-diet group, and this pattern remained through 12 weeks of age. In contrast, the comparatively lower abundance of Ruminiclostridium 9 was not significant until 12 weeks of age. The transition from the normal diet to the synthetic diet did not significantly alter the abundance of the unclassified Lachnospiraceae ASV, however supplementation with I3C depleted the abundance of this family at both 8 and 12 weeks of age. The immediate differential abundance in Intestinimonas and Lachnospiraceae may reflect a primary response to AhR-mediated changes in the gut immune system, whereas the delayed reduced abundance in Ruminiclostridium 9 may be secondary to changes in microbial composition.

Since I3C could alter colonization with Firmicutes directly or indirectly of AhR intestinal activation, we verified if the abundance of these bacteria were affected by genetic manipulation of AhR expression. To ensure relevance of this experimental group for effect on diabetes, we used NOD AhR knockout mice at 12 weeks of age when approximately 60% of islets are infiltrated, but prior to clinical manifestations. Among the three above mentioned microbes, only Intestinimonas showed lower abundance with gene dosage. Accordingly, gut colonization by the only member of the Intestinimonas genus detected in these mice, was found in 33% of wildtype mice and 50% of heterozygous mice were colonized with Intestinimonas, 80% of their knockout littermates were colonized (Figures 5G, H). Taken together, Intestinimonas abundance is negatively regulated by AhR and therefore the decrease in this genus is likely to be an effect of I3C-mediated AhR signaling in the intestine.

The data presented in the study are deposited in the NCBI Sequence Read Archive, accession number PRJNA679964.