Research ArticleHEALTH AND MEDICINE

Modulation of lymphocyte-mediated tissue repair by rational design of heterocyclic aryl hydrocarbon receptor agonists

See allHide authors and affiliations

Science Advances  15 Jan 2020:
Vol. 6, no. 3, eaay8230
DOI: 10.1126/sciadv.aay8230

Abstract

Aryl hydrocarbon receptor (AHR) is an essential regulator of gut immunity and a promising therapeutic target for inflammatory bowel disease (IBD). Current AHR agonists are inadequate for clinical translation due to low activity, inadequate pharmacokinetics, or toxicity. We synthesized a structurally diverse library and used integrated computational and experimental studies to discover mechanisms governing ligand-receptor interaction and to design potent drug leads PY109 and PY108, which display physiochemical drug-likeness properties, desirable pharmacokinetic profiles, and low toxicity. In a murine model of dextran sulfate sodium–induced colitis, orally administered compounds increase interleukin-22 (IL-22) production and accelerate mucosal healing by modulating mucosal adaptive and innate lymphoid cells. AHR and IL-22 pathway induction was confirmed using RNA sequencing and characterization of the lymphocyte protein-protein interaction network. Significant induction of IL-22 was also observed using human T cells from patients with IBD. Our findings support rationally designed AHR agonists for IBD therapy.

INTRODUCTION

Inflammatory bowel disease (IBD) is the consequence of a sustained inflammatory response to commensal microorganisms in a genetically susceptible host with excessive production of pro-inflammatory cytokines. Recent evidence now suggests that the exacerbated inflammatory response observed in IBD is initiated and maintained by loss of gut epithelial integrity manifest by increased barrier permeability, impaired mucin production, and reduced secretion of antimicrobial peptides with an ensuing dysbiosis (1, 2). Immune-modulating agents that promote mucosal healing and microbial homeostasis hold promise as an important new class of disease-modifying therapeutic for IBD. The aryl hydrocarbon receptor (AHR) is an essential regulator of the gut innate immune system and mediates processes, including expression of interleukin-22 (IL-22), which are responsible for gut barrier function and microbial homeostasis (3, 4). AHR belongs to the basic helix-loop-helix/Per-Arnt-Sim (PAS) family of transcription factors, which is bound to several co-chaperones and present in an inactive form in the cytosol. In response to small-molecule ligand binding, AHR translocates to the nucleus, dissociates from its chaperones, and dimerizes with the AHR nuclear translocator (ARNT). The heterodimer, in turn, binds to xenobiotic response elements (XRE) and induces gene transcription (5). The critical role of AHR in barrier protection and mucosal immunity is highlighted by the phenotype of AHR-deficient mice, which exhibit increased susceptibility to epithelial damage, dysbiosis, and bacterial infection and colitis (3, 4, 6, 7). This defect in gut barrier function is also associated with diminished γδ T cells (6), group 3 innate lymphoid cells (ILC3) (7), and intestinal lymphoid follicles (3), suggesting an indispensable role of AHR signaling in maintaining the gut innate lymphoid cell population and mucosal immunity. Furthermore, genome-wide association studies have identified AHR as a susceptibility locus for IBD (8).

All told, the AHR pathway represents a promising therapeutic target for restoration and maintenance of intestinal integrity; however, several barriers have continued to limit the development of AHR modulators for clinical evaluation. While a variety of AHR agonists have been reported, including those that reduce disease severity in murine models of colitis (911), none of these tool compounds are suitable for clinical translation because of their lack of drug-like properties, including inadequate pharmacokinetics (PK) and compound-specific toxicity (1216). There also remains an incomplete understanding of receptor structure and ligand recognition. The reported AHR ligands have been discovered by serendipity rather than by systemic lead optimization and represent structurally unrelated compounds, which infers a promiscuous ligand binding domain (LBD) inadequate for rational and structural-based drug design. Last, competing mechanistic explanations regarding the precise role of AHR activation in ameliorating colitis underscores our imperfect understanding of signal transduction and the associated in vivo mechanism of action (9, 10).

In this report, a focused library of structurally diverse indole and indazole compounds was synthesized and screened, resulting in the characterization of a number of highly potent AHR modulators. Integrated computational and experimental studies have enabled the identification of stringent structural requirements for ligand-induced activation of AHR and led to the generation of lead compounds, PY109 and PY108, which displayed sufficient biostability to achieve a therapeutic effect at very low oral dosing and appropriate plasma clearance to avoid persistent AHR activation. Moreover, their drug-likeness properties are consistent with high-quality clinical drug candidates. Mechanistically, our studies confirm that these orally administered compounds potently increase murine IL-22 production by a direct effect on the gut innate and adaptive immune system and locally expand ILC3 and γδ T cell subpopulations. In the process, mucosal healing is accelerated, gut antimicrobial barrier function is reestablished, and intestinal integrity is restored. Significant induction of IL-22 was also observed using human T cells from both healthy participants and patients with active IBD. Our findings support rationally designed multi-heterocyclic AHR agonists as a novel class of disease-modifying therapeutic for IBD.

RESULTS

Identification of indole-derived AHR agonists and structure-activity relationships

A focused library of indoles and indazoles was synthesized, and compounds were screened for their ability to activate AHR in an XRE luciferase assay using transfected human HepG2 or mouse Hepa-1c1c7 cells. A hit-to-lead selection cascade was established to guide optimization and selection of potent agonists (Fig. 1A). The structural diversity of the initial compound library is highlighted by the broad range of aryl moieties that include pyridine/pyrazine (PY), benzene (BN), and thiazole (TZ) derivatives (table S1). In addition to a number of moderately active hits [half-maximal effective concentration (EC50), <1000 nM, BN114 and TZ142], methyl 6-(1H-indole-3-carbonyl) pyridine-2-carboxylate, designated PY10, was discovered as an exceptionally potent AHR agonist (EC50, 3.7 nM), with activity similar to 2-(1′H-indole-3′-carbonyl)-thiazole-4-carboxylic acid methyl ester (ITE; EC50, 4.6 nM) (Fig. 1, B and C). Notably, the regioisomeric methyl 6-(1H-indole-3-carbonyl) pyridine-3-carboxylate PY11 and the benzene analog BN202 [methyl 3-(1H-indole-3-carbonyl) benzoate] were completely inactive (EC50, >10,000 nM; Fig. 1D). Subsequent experimental studies using PY10 as a tool compound identified stringent structural requirements for ligand-induced activation of AHR (Fig. 1E and table S2). Initial structural modifications revealed a direct correlation between activity and size of the pyridine-2-carboxylate (─CO2R) ester group (Fig. 1F and table S2). Esters with increased bulkiness displayed a substantial loss of AHR activity, and even the slightly bulkier ethyl ester analog PY147 was only marginally active (EC50, 1108 nM). Similarly, the bulkier N,N-dimethylamide analog PY152 was >100-fold less active than PY10. Interrogation of the structure-activity relationship (SAR) of the indole moiety also indicated remarkable sensitivity to substitution patterns (Fig. 1G and table S2). For example, chlorination (Cl) at the 4-position of the indole moiety led to an inactive analog PY198, while Cl at the 5-position in PY163 led to a nearly 10-fold decrease in activity. In contrast, the EC50 of 6- and 7-chloroindole analogs PY164 (EC50, 1.7 nM) and PY186 (EC50, 1.4 nM) was lower than that of PY10.

Fig. 1 Identification of AHR agonist and SAR.

(A) Experimental design demonstrating AHR agonist hit-to-lead selection cascade. (B) Structure of focused library and lead compound PY10. (C) Activity of compound PY10 and ITE, measured by XRE-driven luciferase reporter assay in the human HepG2 cell line (n = 3). (D) Library screen reveals stringent structural requirements for AHR activation. (E) General scheme for SAR-based modifications of the PY10 backbone. Series I represents interrogation of the pyridine-2-carboxylate ester group and series II represents interrogation of the indole moiety. (F) Pyridine esters with increased bulkiness displayed substantial loss of AHR activity. (G) AHR activity is sensitive to indole substitution. (H) Superimposed AHR homology model before (green) and after (gray) MD simulation. Secondary structures including four α-helices (Cα, Dα, Eα, and Fα) and four β-strands (Aβ, Gβ, Hβ, and Iβ) are labeled. The molecular surface (Connolly) volume of the LBD is displayed in blue. (I) Detailed view of the ligand binding domain (LBD) in complex with PY10 of the molecular dynamics (MD) homology model. Amino acid residuals forming the LBD are labeled. (J) Detailed view of the water-mediated H-bond network near the indole moiety of PY10 (red line). (K) Detailed view of the Met340 subdomain occupied by the methyl ester group of PY10. Color for heteroatoms are yellow for sulfur (S), red for oxygen (O), and blue for nitrogen (N). Data represent mean ± SEM.

Molecular modeling of AHR-bound ligands

Homology modeling and molecular dynamics (MD) simulations were performed to investigate the unique structural requirements for AHR activation. While the crystal structure of a truncated AHR mutant without the PAS-B LBD domain has been reported (17), the three-dimensional structure of the AHR PAS-B LBD has not been resolved. Because the PAS-B domains of human hypoxia-inducible factor–2α (HIF-2α) and AHR are structurally similar, we used known co-crystal structures of the human HIF-2α PAS-B domain in complex with a small-molecule ligand [Protein Data Bank (PDB): 3H82, 3H7W, and 3F1O] to construct homology models of the human AHR-LBD using MODELLER (fig. S1A). One hundred potential models were developed and scored for fitness using the discrete optimized protein energy (DOPE) method, and the top 20 scoring homology models were used for subsequent docking studies with the AHR agonist PY10 (fig. S1, B and C). Docking poses were refined by explicit solvent energy minimization, and the AHR-LBD/PY10 binding complex with the lowest potential energy was removed from the MD trajectory and used to model PY10 bound within the AHR-LBD (Fig. 1, H and I). The MD model was stabilized by a H-bond network involving two water molecules, which participated in H-bonding interactions with His291, Thr289, H2O, Gln383, Ser365, and Val350 (Fig. 1J). These amino acid residues appear to form a rigid boundary that defines the ligand-accessible region of the LBD. PY10 occupies a cavity with a molecular surface (Connolly) volume of 581.2 Å3 in the MD model, which is similar in size to human and mouse AHR homology models reported elsewhere (18, 19).

Homology modeling indicated a lack of H-bonding interactions between PY10 and the LBD of AHR. None of the indole NH, pyridine N, or carboxylate O atoms formed H-bonds with the surrounding amino acids in the ligand-binding domain. Rather, hydrophobic interactions were observed with residues that surround the indole moiety and those that enclose the pyridyl-6-carboxylate methyl ester. Notably, PY10 adopted a low-energy conformer, defined by a 51° torsional angle between the indole and the pyridine aromatics, positioning the pyridyl-6-carboxylate methyl ester within a deep pocket, confined by His337, Met340, Ala367, Ile379, and Val381. This pocket, referred to as the Met340 subdomain, positions the methyl carbon atom in close proximity to Met340 and Ala379 at the base of the pocket (Fig. 1K and fig. S1D). Conformational analysis of a series of closely related structural analogs confirmed that inactive compounds were characterized either by steric clashes or misorientation, which limited suitable docking within the hydrophobic pocket, or by the lack of a methyl ester or similarly sized group that could appropriately occupy the pocket. For example, conformational analyses of active PY10 and inactive BN202 along the center C═O linker revealed that PY10 is in a favorable low-energy conformer with a small torsional angle (<90°), which is required to project the methyl ester group into the M340 subdomain (fig. S2A). In contrast, BN202 has an unfavorable high energy caused by the steric clash between the hydrogen atoms at the 2-position of the indole and the 2-position of the 3-benzoyl methyl ester (fig. S2A). The result is a near-perpendicular geometry between the indole and benzene rings with a torsional angle of 104°, which projects the methyl ester moiety away from the Met340 subdomain (fig. S2B). A similar effect is observed for inactive PY144, where the methyl group at the 2-position of the indole clashes with the pyridine ring (fig. S2B). Further, LBD surface area analysis revealed that a sufficient cavity exists in the region of the indole 7-position to accommodate a chloro (Cl) substituent without detrimental effect on activity (PY186; EC50, 1.4 nM). In contrast, the 4-position is in close contact with neighboring residues, which prohibits substitution at this position (PY198; EC50, >10,000 nM). Similarly, the methyl ester moiety is in close contact with Met340 subdomain, which prohibits a bulkier substitution (fig. S2C) and is consistent with experimental observations (Fig. 1F).

Lead optimization for plasma and metabolic stability

Ester groups are metabolically labile and readily hydrolyzed by plasma esterases or hepatic enzymes. We observed that ITE and PY10 were highly unstable when incubated in plasma (Fig. 2A) or in the presence of liver microsomes (Fig. 2B). After a 30-min incubation in plasma, liquid chromatography–tandem mass spectrometry (LC-MS/MS) analysis revealed the predominant presence of a much less active carboxylic acid metabolite PY145 (EC50, 725 nM) rather than the parent compound PY10 (fig. S3, A and B). Seeking to improve bioavailability by increasing metabolic stability, lead optimization was directed at the replacement of the labile ester group of PY10. Molecular modeling supported the notion that the methyl ester group of PY10 contributed to ligand-receptor stabilization primarily by space filling but not through H-bonding interactions. In addition, the MD model corroborated SAR studies, which revealed a requirement for a moiety with limited steric bulkiness. As a consequence, we analyzed a series of analogs in which the methyl ester was replaced with a nonhydrolyzable substituent (table S2). As expected, analogs with bulkier substituents were less active with the phenyl derivative, PY113, and the N-morpholino derivative, PY110, displaying an EC50 of 10 and 62 nM, respectively. In contrast, substitution with cyano (─CN) or trifluoromethyl (─CF3) groups resulted in the identification of PY108 (EC50, 2.5 nM) and PY109 (EC50, 1.2 nM) as among the most potent AHR agonists in this series (Fig. 2, C to E). PY108 and PY109 were stable in plasma and demonstrated an observed half-life (t1/2) in the liver microsomal assay of 69 and 60 min, respectively (Fig. 2, A and B). Docking analysis of PY108 and PY109 in the AHR homology model revealed interactions for ─CN and ─CF3 groups that were similar to those observed for the methyl ester group of PY10 (Fig. 2, F and G, and fig. S3C). These results further validated the MD homology model of the AHR LBD.

Fig. 2 Characterization of compounds with improved stability.

(A) Compound stability in mouse plasma and (B) liver microsomes was measured by liquid chromatography–tandem mass spectrometry (LC-MS/MS; n = 3). Plasma t1/2 (in minutes): ITE (44); PY10 (45); PY108 (>60); PY109 (>60). Liver microsome t1/2 (in minutes): ITE (11); PY10 (17); PY108 (69); PY109 (60). (C) Structure of lead compounds PY108 and PY109. (D and E) Activity of PY108 and PY109 measured by XRE-driven luciferase reporter assay in human HepG2 and mouse Hepa-1c1c7 cell lines. (F) Docking pose (delta E = 10.4 kcal/mol) of PY109 in the MD homology model. (G) Docking pose (delta E = 10.4 kcal/mol) of PY108 in the MD homology model. Data represent mean ± SEM. The 95% confidence interval of reported EC50 values is available in table S2.

Activation of AHR by lead compounds PY108 and PY109

Rapid nuclear enrichment of AHR was induced by PY108 and PY109 in both human HepG2 and mouse Hepa-1c1c7 cells (Fig. 3A and fig. S4A). This finding is consistent with a change in the preferred aryl-hydrocarbon receptor-interacting protein (AHR) binding partner from AIP to ARNT in Hepa-1c1c7 cells, as revealed by immunoprecipitation (IP)–MS and Western blotting (Fig. 3, B and C, and fig. S4, B and C). In contrast, BN202, which is structurally similar to PY108 and PY109, but inactive in the AHR luciferase assay, had minimal effect on AHR association with binding partners, supporting specific on-target activation of AHR by identified lead compounds.

Fig. 3 Biological characterization of lead compounds in vitro and after oral administration.

(A) Nuclear translocation of AHR induced by ITE and lead compounds PY108 and PY109 at 90 min. AHR (green), actin (red), and nucleus (blue). (B and C) Hepa-1c1c7 was incubated with 1 μM compound, followed by immunoprecipitation of AHR complex. AHR-associated binding partner level was assessed by LC-MS/MS for ARNT (B) or Western blots for ARNT and AIP (C) (n = 3). (D) Biological and in silico physiochemical parameters of PY108 and PY109. *Polar surface area (PSA) and octanol/water partition coefficient (cLogP) were calculated using ChemAxon software. LLE, lipophilic ligand efficiency; calculated from EC50 in human HepG2 cells by the equation: LLE = pEC50 − cLogP (pEC50 = −log10EC50). (E) Lead compounds PY108 and PY109 induced AHR-dependent Cyp1a1 expression in mouse Hepa-1c1c7 cells (n = 4). (F) Quantitative polymerase chain reaction (qPCR) analysis of Cyp1a1 induction in liver and colon, 12 hours after oral gavage of PY109 (n = 3). (G) Plasma concentration of PY109 and PY108 after oral administration of 1 mg (PO) of PY109 and PY108. Compounds were quantified in plasma using LC-MS/MS (n = 4). Data represent mean ± SEM. *P < 0.05 versus control, #P < 0.05 versus ITE by Student’s t test (D and F) or ANOVA and Bonferroni’s multiple comparison test (B).

In vitro activity and in silico derived physiochemical parameters for PY108 and PY109 are summarized in Fig. 3D. PY109 was equally potent in AHR XRE luciferase assays conducted in mouse Hepa-1c1c7 cells and human cells, while PY108 demonstrated preferential activation of human AHR by nearly 40-fold. The weaker activity of PY108 in mouse cells was confirmed in a Cyp1A1 induction assay with mouse Hepa-1c1c7 cells (Fig. 3E). Cell permeability and the efflux ratio (ER) were assessed using a Caco2 monolayer equilibrium dialysis assay. PY108 and PY109 exhibited a high apparent permeability (Papp) and a low ER (PY108: Papp, 412 nm/s and ER, 0.74; PY109: Papp, 358 nm/s and ER, 1.07), consistent with favorable drug-likeness properties as predicted by in silico parameters: molecular weight (Mw), calculated logarithm of the octanol/water partition coefficient (cLogP), and polar surface area (PSA). Lipophilic ligand efficiency (LLE) is a widely used parameter that correlates with the identification of high-quality clinical drug candidates (20). Derived from an analysis of U.S. Food and Drug Administration–approved oral drugs, an LLE ≥ 5, based on a dissociation constant < 10 nM, cLogP < 3, and Mw < 500 Da, along with a heavy atom count of ≤38, is predictive of a successful drug candidate (20). In this regard, PY108 and PY109 are promising drug leads for oral administration (Fig. 3D).

PK and PD profiles

Given the predicted suitability of these compounds for oral dosing, PK and pharmacodynamics (PD) profiles were examined in C57BL/6 mice after oral administration. For PD studies, tissue induction of Cyp1A1 was selected as a marker of AHR activation. Liver, colon, spleen, and blood were harvested 12 hours after administration of PY109 (0.1 to 1000 μg). Dose-dependent Cyp1A1 induction was observed in liver and colon tissue, while Cyp1A1 expression was noted in splenic tissue and blood only at the highest administered doses (Fig. 3F and fig. S4D). For PK studies, mice were administered 1 mg of PY109 or PY108 and blood was collected over 24 hours for plasma quantification via LC-MS/MS. An elimination t1/2 of 12 and 11 hours were observed for PY109 and PY108, respectively (Fig. 3G). Elevated dosing regimens of up to 30 mg/kg were also investigated to characterize toxicity parameters, with no adverse events recorded (fig. S5). Together, these data provide strong support that PY108 and PY109 are sufficiently stable to activate AHR in vivo after oral dosing with appropriate plasma clearance to avoid persistent AHR activation.

DSS-induced colitis is attenuated by PY109 and PY108

To investigate lead compounds in the treatment of IBD, a mouse model of dextran sulfate sodium (DSS)–induced colitis was established by administering C57BL/6 mice water supplemented with 3% (w/v) DSS for 7 days followed by a 13-day recovery period. Mice received 1 μg of PY109 or PY108 in 100 μl of vehicle [1% hydroxypropyl methylcellulose (HPMC) and 1% polysorbate 80 in water] or vehicle alone by daily oral gavage, beginning on day 0. Body weight and disease activity index (DAI) were recorded daily. On days 7 and 11, representing the acute and recovery phase of colitis, respectively, mice were euthanized and the colon was harvested for length measurement, histology, RNA isolation, and lamina propria mononuclear cell (LPMC) extraction.

Treatment with PY109 significantly improved survival, reduced weight loss, preserved colon length, and reduced overall disease activity (Fig. 4, A to D). PY108 attenuated colitis in a similar manner (fig. S6, A to C). Histological examination revealed a significant reduction in colonic inflammation and crypt loss (Fig. 4, E and F, and fig. S6, D to F), with preserved goblet cell number (Fig. 4, G and H). RNA sequencing (RNA-seq) was used to examine colon transcriptomes at day 11. Three groups were enrolled, including healthy control, DSS-induced vehicle control, and DSS-induced PY109 treatment group. Analysis of differentially expressed transcripts in colons from untreated mice and DSS-treated mice identified a “DSS colitis transcriptome” of 1978 genes. Of these 1978 genes in the DSS transcriptome, PY109 affected the expression of 171 transcripts, with the majority (163) demonstrating counter-regulation with restoration of a healthy genotype (Fig. 4I). Antigen presentation and liver X receptor/retinoid X receptor (LXR/RXR) activation were identified as the top enriched pathways for PY109-mediated rescue or suppression of expression. Specifically, PY109 rescued expression of the antigen presentation pathway, which was severely down-regulated in DSS-treated mice, while the LXR/RXR pathway, elevated in DSS mice, was significantly down-regulated in PY109-treated mice (table S3). Dysregulation in both pathways has been linked to human IBD (21, 22). We also observed that PY109 mediated up-regulation of prototypical AHR target genes, including Cyp1a1 and Ido1 and Il22 and Il17a (Fig. 4J). We did not observe PY109-mediated regulation of Foxp3, the lineage defining transcription factor for regulatory T cells or IL-10 (Fig. 4K). Consistent with a role for IL-22 and IL-17A in barrier function, PY109-treated mice demonstrated enhanced expression of IL-22/IL-17A–regulated genes involved in mucin production (Muc4 and B3gnt7), epithelial function (Mfsd2a and Ccnd1), and antimicrobial defense (Oas1, Oas1, Zbp1, and Nos2) (Fig. 4K) (23, 24).

Fig. 4 PY109 attenuates DSS-induced colitis in mice.

Mice in colitis model were recorded for survival (A), body weight (B), and disease activity score (C). Data represent three independent trials with n = 7 per group per trial. (D) Colon lengths at day 11 (n = 12 to 13). (E) Hematoxylin and eosin staining (100×) and (F) histological score at day 11 (n = 8). (G and H) Colon goblet cells at day 11 (n = 7 to 8). (I) RNA sequencing (RNA-seq) analysis of colon from untreated mice (n = 4) and dextran sulfate sodium (DSS)–treated mice ± PY109 (n = 5) at day 11. Heatmap demonstrates colitis transcriptome counter-regulated by PY109. (J) Volcano plot of differentially expressed genes by PY109. Red, increased genes; green, decreased genes; blue, AHR target genes. (K) Heatmap of reported AHR target mediators and Il22 or Il17a regulated barrier genes with red denoting increase and green denoting decrease. (L and M) qPCR expression level of Il22 and Il17a in the whole colon at days 7 and 11 of the DSS model. (N to Q) Flow cytometry staining of LPMC. IL-22+ and IL-17A+ were stained in combination with T helper staining (CD3+CD4+; N to O) or ILC3 staining (CD3RORrt+; P and Q). (R and S) Absolute number of ILC3 (R) or intraepithelial CD3+CD45+ TCR γδ+ cells (S). (T to V) qPCR of the epithelial fraction was performed for detection of antimicrobial peptides S100a8 (T), S100a9 (U), and RegIIIγ (V). qPCR data represent n = 4 to 8 per group. Flow cytometry data represent n = 4 to 10 per group. Data represent mean ± SEM, *P < 0.05 versus control (Student’s t test).

AHR activation and induction of cytokine expression were confirmed in colonic tissue by qPCR and flow cytometry. While Ahr expression was not affected by PY109 treatment, Cyp1a1 was significantly elevated (fig. S6, G and H). Consistent with RNA-seq data, PY109 treatment groups demonstrated a 10-fold increase in Il22 expression and 2-fold increase in Il17a expression by day 11 (Fig. 4, L and M). We did not observe PY109-regulated expression of Foxp3 or Il10 (fig. S6, I and J). A modest, nonsignificant suppression of Tnf and Ifnγ expression was observed at day 7, with signals further decreasing at day 11 (fig. S6, K and L). Likewise, flow cytometry analysis of colonic LPMCs revealed that PY109-treated mice demonstrated a 40% reduction of tumor necrosis factor–α–positive (TNF-α+) and interferon-γ–positive (IFN-γ+) cells at day 7, with no impact on regulatory T cells (CD3+CD4+Foxp3+) (fig. S6, M to O). T helper and ILC3 cells are two major effector cell types that secrete IL-22 and IL-17A. Treatment with PY109 induced IL-22 and IL-17A secretion from both CD4+ T helper and CD3RORγt+ innate lymphoid cells. The absolute number of CD4+IL-22+, CD4+IL-17A+, IL-22+ILC3, and IL-17A+ILC3 cells doubled in response to drug treatment (Fig. 4, N to Q). AHR signaling is essential for the maintenance of ILC3 and intraepithelial γδ T cells, both of which play an important role in maintaining gut barrier integrity (6, 7). The absolute number of both cell types among colonic LPMCs increased in response to PY109, suggesting a direct effect on the innate lymphoid cell population (Fig. 4, R and S). Intraepithelial γδ T cells are also known to produce IL-22 and IL-17A (25). qPCR analysis of the epithelial fraction revealed increased expression of both cytokines, suggesting that γδ T cells might be a third effector cell type that secretes IL-22 and IL-17A in response to PY109 (fig. S6, P and Q). Consistent with activation of the IL-22 pathway, qPCR analysis of the colon epithelial fraction also demonstrated increased expression of antimicrobial peptides S100a8, S100a9, and RegIIIγ at days 7 and 11, along with a modest increase in the expression of barrier glycoproteins Muc1 and Muc3 and the anti-apoptotic protein Bcl2l1 (Fig. 4, T to V, and fig. S6, R to T).

PY109 promotes epithelial antimicrobial peptide expression via lymphocyte-derived IL-22 and IL-17A

To confirm the ability of PY109 to induce lymphocyte expression of IL-22 and IL-17A, CD4+ T cells were purified from the spleen and activated in the presence of Th17 polarization medium in vitro. Consistent with prior reports, Ahr expression was up-regulated with Th17 polarization, supporting its active regulatory role in this T helper subtype (fig. S7A) (26). PY109 induced a more pronounced expression of Cyp1a1 in Th17 cells, as compared to unpolarized cells (fig. S7B). PY109 significantly increased the expression of both Il22 and Il17a mRNA in Th17 cells (fig. S7, C and D), and flow cytometry revealed increased intracellular staining for both IL-22 and IL-17A (Fig. 5, A to C). Enzyme-linked immunosorbent assay (ELISA) measurements of secreted IL-22 and IL-17A confirmed a dose-dependent response to drug levels (Fig. 5, D and E). To further characterize the effect of PY109 in colon-derived CD4+ T cells and innate ILC3 cells, colonic LPMCs were isolated and cultured under Th17 polarization conditions or in the presence of IL-1β and IL-23, which promotes IL-22 and IL-17A secretion in colonic ILC3 cells (7, 25). PY109 significantly increased secretion of IL-22 and IL-17A from both CD3+CD4+ T helper cells (Fig. 5, F and G) and CD3RORγt+ ILC3 cells (Fig. 5, H and I).

Fig. 5 PY109 promotes epithelial AMP expression via lymphocyte-derived IL-22 and IL-17A.

(A) Flow cytometry scatterplots of IL-22 and IL-17A in spleen CD4+ T cells activated in Th17 media with or without PY109. (B and C) IL-22+ and IL-17+ frequency in CD4+ cells (n = 3). (D and E) Enzyme-linked immunosorbent assay (ELISA) measurement of IL-22 and IL-17A levels in the supernatant of Th17-polarized CD4+ T cells (n = 3). (F and G) IL-22+ and IL-17+ frequency in CD4+ cells from colon-derived LPMCs activated in Th17 media (n = 3). (H and I) IL-22+ and IL-17+ frequency in (CD3RORrt+) ILC3 cells from colon-derived LPMCs treated with IL-1β and IL-23 (n = 3). (J) Western blot detection of AHR, AIP, and ARNT in immunoprecipitated AHR complex from spleen-derived Th17 cells treated with or without PY109 for 1 hour (n = 3). (K) Quantification of band intensity in Western blots. (L) IP-MS determination of AHR-associated proteins. Proteins with more than twofold change in response to PY109 and 5% intensity of AHR are highlighted. (M) qPCR detection of S100a9 levels in mouse colon epithelial CMT-93 cells treated with IL-22 or IL-17A or both (n = 4). (N) qPCR detection of S100a9 level in CMT-93 cells treated with conditioned medium from spleen-derived CD4+ cells polarized in Th17 with or without PY109 and blocking antibodies (n = 4). Data represent mean ± SEM, *P < 0.05 versus control by Student’s t test (B to I) or ANOVA and Bonferroni’s multiple comparison test (K, M, and N).

To validate traditional AHR activation in PY109-treated Th17 cells, we first used IP–Western blot to confirm that PY109 treatment induced the preferential binding of AHR from its cytosolic binding partner AIP to its nuclear partner ARNT (Fig. 5, J and K). LC-MS/MS was subsequently applied for unbiased identification of AHR-associated proteins in Th17 cells treated with or without PY109. Binders that exhibited both high abundance and large fold change in response to PY109 are highlighted in the volcano plot (Fig. 5L), which suggests the presence of additional noncanonical networks that may mediate the effect of PY109 in Th17 cells.

We sought to determine whether PY109-induced expression of IL-17A and IL-22 in Th17 lymphocytes would up-regulate antimicrobial peptide expression in colon epithelial cells. The colon epithelial cell line, CMT-93, was treated with exogenous IL-17A, IL-22, and conditioned media from spleen-derived Th17 cells treated with PY109. Although IL-17A induced robust S100a9 expression, exposure to IL-22 alone did not. As previously reported, IL-22 together with IL-17A displayed a strong synergistic effect (Fig. 5M). Significant induction of S100a9 was also observed when colon epithelial cells were incubated in PY109-treated Th17-conditioned media, which contains both IL-22 and IL-17A. This effect was reduced by blocking antibodies to IL-17A or IL-22 and completely abrogated in the presence of both antibodies (Fig. 5N and fig. S7E). These studies support the ability of PY109 to promote normal epithelial barrier function by induction of IL-22 and IL-17A from T cells.

PY109 induces IL-22 in human T lymphocytes

Given previous reports that have demonstrated AHR-dependent induction of IL-22 in human Th17 lymphocytes, we examined the ability of PY109 to activate AHR and promote IL-22 production in CD4+ T cells from healthy participants and patients with active IBD (26, 27). CD4+ T cells were isolated from circulating peripheral blood mononuclear cells (PBMCs) and activated in Th17 media with or without PY109 for 5 days, followed by exposure to stimulation cocktail and harvest for qPCR and flow cytometry. Healthy participants and patients with IBD expressed similar levels of AHR, and this expression was not affected by PY109. Marked induction of CYP1A1 was observed in the presence of PY109, confirming activation of the AHR signaling pathway in human CD4+ lymphocytes (Fig. 6, A and B). Cells were subsequently examined for IL22 and IL17A expression, and we observed significant PY109-dependent up-regulation of IL22 mRNA and protein, while IL17A levels were generally suppressed (Fig. 6, C to G). Notably, we observed no difference in PY109-dependent responses between healthy control and IBD cohorts.

Fig. 6 PY109 induces IL-22 in human T cells from healthy donors and patients with IBD.

CD4+ T cells were purified from PBMCs of healthy participants (H.P.) and patients with IBD. Cells were activated with anti-CD3 and anti-CD28 antibodies in the presence of Th17-polarizing cytokines (TGF-β1, IL-1β, and IL-6) with or without 5 μM of PY109 for 5 days. Cells were re-stimulated with PMA, ionomycin, and brefeldin A and RNA expression of AHR (A), CYP1A1 (B), IL22 (C), and IL17A (D) analyzed by qPCR using 18S as an internal control. (E) Representative scatterplots of IL-22 and IL-17A staining from one donor. Flow cytometry quantification of IL-22 (F) and IL-17A (G). Dot plots represent 7 to 9 healthy donors and 10 to 12 patients with IBD. *P < 0.05, PY109 versus control (paired t test).

DISCUSSION

AHR is an essential regulator of the gut innate immune system, mediating processes responsible for microbial homeostasis and maintenance of mucosal integrity (3, 4, 7, 9, 11). Endogenous low-affinity AHR agonists are derived from a variety of dietary metabolites through microbial- or host-mediated metabolism (5, 11, 28). The beneficial effects of Lactobacillus species as a commensal organism are thought to be achieved by metabolic production of AHR ligands (11, 28). Emerging evidence now suggests that among patients with IBD, host- and microbial-derived AHR ligands are reduced, which may contribute to disease severity (28). Consistent with these observations, genome-wide association studies have identified polymorphisms of CARD9 that are associated with reduced microbial production of AHR ligands and confer susceptibility to IBD (21, 28).

Although the metabolic route for many indole-related AHR ligands has not been well defined, 6-formylindolo[3,2-b]carbazole (FICZ), ITE, indole-3-aldehyde, indole-3-carbinol, and supplemental dietary tryptophan ameliorate 2,4,6-trinitrobenzene sulfonic acid (TNBS)–induced, DSS-induced, and T cell transfer–induced colitis (6, 911, 29). IL-22 was increased in several studies (9, 11, 29), with beneficial effects abrogated by treatment with an IL-22 blocking antibody or an AHR antagonist (9). These reports demonstrate the therapeutic potential of indole-based AHR agonists. Nonetheless, the effectiveness of these metabolites and related first-generation derivatives has been limited by low activity, poor PK, and off-target effects (1416).

In an effort to design potent indole-based AHR agonists, we constructed a homology model based on crystal structures of HIF-2α, which shares partial homology to the AHR LBD. While the apo form of HIF-2α PAS-B has been used in previous studies (18), we used three ligand-bound (holo) crystal structures of the HIF-2α PAS-B domain. A recent study has demonstrated that model reliability can be enhanced by combining multiple holo crystal structures, with associated ligand-induced conformational rearrangements, to improve statistical representation of bound forms (19). An initial screen of a focused library of 3-aroyl-indoles and 3-heteroaroyl-indoles identified PY10 as a synthetic AHR agonist with low nanomolar potency (EC50, 3.2 nM). We postulated that MD simulation of the AHR-LBD/PY10 binding complex would converge on an optimal ligand binding pocket upon model equilibration. The validity of the homology model constructed using this strategy is supported by a number of observations. First, several of the amino acids involved in ligand stabilization have been experimentally validated in mutagenesis studies (30, 31), including Met340 and Val381, which form a deep binding pocket within the AHR-LBD. Second, the homology model accounted for nearly all of the observed SARs derived from experimental analysis of our initial compound library. In addition, in the process of lead optimization, the constructed homology model provided important insights to guide drug design. Specifically, the absence of hydrogen bonding interactions from the labile methyl ester group of PY10 led to the design of highly potent AHR agonists PY108 and PY109 that were suitable for oral administration with improved metabolic stability. Importantly, both compounds display physiochemical and PK properties, consistent with high-quality clinical drug candidates. In contrast to ITE and FICZ that are metabolically unstable (16, 32) and require intraperitoneal administration (9, 10, 28) or 2,3,7,8-tetrachlorodibenzodioxin (TCDD), which perpetually activates AHR, PY108 and PY109 exhibited enhanced biostability, with a t1/2 of 12 hours after oral administration, ideal for once daily oral dosing, while minimizing safety risks due to large fluctuations in drug exposure (33). Both compounds were active in a murine disease model following daily microgram oral dosing, in contrast to low-activity dietary ligands that have typically been administered at 1000-fold higher doses (6, 11, 29).

In promoting the restoration of gut barrier integrity, we observed that PY109 increased colonic expression of IL-22 and IL-17A and downstream antimicrobial peptides in vivo and in vitro. A number of investigations have previously documented that IL-22 promotes microbial homeostasis and intestinal healing (9), with more recent reports also supporting a protective role for IL-17A (34). While IL-17A is up-regulated in Crohn’s disease and was presumed to function as a pro-inflammatory mediator (35), clinical trials have demonstrated that blockade of IL-17A or its receptor among patients with Crohn’s disease increases adverse events without clinical benefit (36). Of note, AHR activation may conversely regulate IL-22 and IL-17A in human T lymphocytes. Similar to previous reports, we observed enhanced IL-22 production in PY109-treated human CD4+ T cells, while IL-17A expression was down-regulated (27). In addition to regulating the secretion of IL-22 and IL-17A, AHR activation is required for maintenance of gut ILC3 and γδ T cells, as demonstrated by decreased γδ T cells (6) and ILC3 cells (7), as well as intestinal lymphoid follicles (3) in AHR-deficient mice. Our study reveals that PY109 expands both mucosal γδ T cell and ILC3 subpopulations, with γδ T cells providing an additional source of IL-22 and IL-17A. Although not addressed in this report, ILC3 cells are also known to promote fucosylation (37) and regulate intestinal lymphoid follicles as a site for immunoglobulin A (IgA) production (38), which are factors that promote repair of the intestinal barrier and microbial homeostasis.

Some limitations apply to the current study. First, while DSS-induced colitis is a common murine model of IBD, it is best suited to examine potential therapeutics directed at the restoration of gut barrier integrity. Determining the effect of novel agents on adaptive immune processes would be better examined through other models, such as the T cell transfer model. Regardless, there are inherent limitations in cross-species comparisons, and no mouse model fully recapitulates human IBD. Second, ligand-induced IL-22 and IL-17A expression has the potential to elicit undesirable inflammatory effects. Before clinical translation, the benefit of ligand-regulated IL-22 and IL-17A secretion would require further verification in complementary animal models of IBD and human tissue.

In summary, inspired by the structure of endogenous ligands, through computational modeling and systematic lead optimization, we have rationally designed a unique class of AHR agonists. In contrast to natural ligands with low activity and poor stability, lead compounds exhibit potent nanomolar activity, favorable physiochemical drug-likeness properties, desirable PK profiles, and oral bioavailability. These lead compounds serve as potent AHR-based immune modulators and inhibit DSS-induced murine colitis by inducing IL-22 and IL-17A production from major mucosal immune cell populations, promoting barrier protection, and limiting barrier-mediated chronic inflammatory disease. We believe that these studies represent an important step toward the development of AHR therapeutics.

MATERIALS AND METHODS

Chemical synthesis and characterization

Detailed methods and compound characterization can be found in Supplementary Materials and Methods.

Molecular modeling

Primary sequences of three HIF-2α PAS-B domains (PDB: 3H82, 3H7W, and 3F1O) were aligned, relative to the human AHR sequence (Uniprot) using MODELLER v9.16 (39), and alignment was visualized using Chimera v1.9. One hundred potential homology models were developed and scored for relative fitness using the DOPE method (40). Ligands from all three co-crystal structures were included during model building but were excluded from subsequent ligand-docking studies using the top 20 scoring homology models.

Rigid docking, which prohibited atom movements of the protein domain but allowed complete ligand flexibility, was performed using Autodock Vina v1.1.2 (41). Ligands were prepared for docking using Open Babel v2.3.2 to generate protonation states and an initial three-dimensional structure (42). Further ligand preparation, such as calculating partial charges, was performed using AutoDockTools v1.5.6 (43). Docking volumes encompassed the entire binding cavity (40 Å × 30 Å × 40 Å) with high exhaustiveness (1024) to ensure comprehensive sampling. The top 200 binding poses, composed of 10 binding poses derived from each of the top 20 homology models, were optimized by gradient descent minimization subject to a maximum of 50,000 iterations using GROMACS v5.1.2 in an explicit solvent box. The complex was solvated using the TIP3P water model in a cubic water box, containing 0.15 M NaCl, extending 12 Å away from any protein atom. Energy minimization used the AMBER99SB force field with the general AMBER force field (gaff) used for ligands. Gaff parameters were calculated with antechamber using the ACPYPE package included in AmberTools14. The lowest energy complex was selected for MD simulation. Ligand-bound complexes were heated to 300°K and equilibrated for 50,000 iterations (2 fs), initially at constant volume (NVT) followed by constant pressure (NPT). Ten independent 100-ns MD simulations were performed, each equilibrated separately.

To estimate the free energy of the complex, metadynamics simulations were also performed using GROMACS, in combination with the PLUMED v2.1.3 plugin (44). Following equilibration, 20-ns trajectories were generated for each ligand and free-energy landscapes were calculated. Among identified AHR ligands, rotatable bonds adjacent to a central carbonyl group were each considered a collective variable with a Gaussian width and height of 0.5 rad and 10.0 kJ/mol, a bias factor of 10, and a stride of 20 ps. The lowest free-energy conformer was identified from the trajectory, extracted, and overlaid within the complex.

Luciferase reporter assay

Human HepG2 or mouse Hepa-1c1c7 cells were seeded in 96-well plates at 75,000 cells per well in full medium [Dulbecco’s modified Eagle’s medium (DMEM) and 10% fetal bovine serum (FBS)] for 24 hours. Transfection was performed with 150 ng of XRE-luciferase plasmid at a transfection reagent (FuGENE, Promega)/DNA ratio of 4:1 for 18 hours in Opti-MEM. Transfected cells were treated with test compound (10−5 to 10−12 M) for 4.5 hours in full medium. Cells were lysed and luciferase was activated using the Luciferase Assay System (Promega). Luminescence signals were measured with a FluroStar Omega microplate reader. Luminescence signals were normalized to a vehicle [dimethyl sulfoxide (DMSO)]–treated control group and plotted as a concentration-dependent fold change. For compounds that displayed a saturated activity curve, EC50 (95% confidence interval ) and Emax were calculated by least-square curve fitting (GraphPad). For compounds with low activity, EC50 was recorded as >10 μM and Emax was presented as percentage of ITE activity at 10 μM.

Nuclear translocation assay

Mouse Hepa-1c1c7 or human HepG2 cells were seeded onto eight-well chamber slides followed by incubation in tryptophan-deficient media [Hanks’ balanced salt solution (HBSS) with Ca2+ and Mg2+, 10% heat-inactivated FBS, 1× nonessential amino acids, 1× sodium pyruvate, and glucose (4.5 g/liter)] for 24 hours. Cells were subsequently treated with vehicle control (DMSO), 10 μM ITE, PY108, or PY109 for 90 min in tryptophan-deficient media. Cells were then fixed in 4% formaldehyde and processed according to a standard immunofluorescence staining protocol. Primary anti-AHR antibody was applied at 1:100 dilution (Enzo, BML-SA210) and Alexa Fluor 488–conjugated secondary antibody was applied at 1:500 dilution (Cell Signaling, 4412S). Actin was stained using Alexa Fluor 568–conjugated phalloidin (Thermo Fisher Scientific). Samples were mounted in anti-fade mounting medium supplemented with 4′,6-diamidino-2-phenylindole for nuclear staining and imaged with confocal microscopy (Leica SP5× MP, 63× oil lens).

IP-MS and Western blot analysis

Confluent Hepa1c1c7 cells were incubated in tryptophan-deficient media for 24 hours to deplete endogenous ligands and treated with vehicle control (DMSO) or 1 μM test compound for 90 min. In a separate experiment, in vitro differentiated mouse spleen–derived Th17 cells were treated with DMSO or 1 μM PY109 for 60 min. Hepa1c1c7 cells or Th17 cells were lysed in buffer containing 10 mM Hepes, 420 mM NaCl, 0.1% NP-40, and complete protease inhibitor (Roche) followed by cyclic sonication and benzonase nuclease treatment (Sigma-Aldrich). Supernatants were collected after centrifugation at 14,000g for 30 min. Immunoprecipitation was performed by adding 2 μl of anti-AHR antibody (Enzo, BML-SA210) or IgG control, with overnight incubation at 4°C followed by the addition of 25 μl of magnetic beads (Thermo Fisher Scientific) for 2.5 hours at 4°C. Protein complex bound beads were washed twice in 1 ml of phosphate-buffered saline, and after each wash, samples were transferred into a set of new tubes to reduce background.

For mass spectrum determination of co-immunoprecipitated proteins, proteins were digested on beads with 1 μg of trypsin (Thermo Fisher Scientific) in 100 mM triethylammonium bicarbonate overnight at 37°C. Peptides were desalted using a C18 ZipTip (P10; Millipore Corporation) and subjected to reverse-phase LC separation on a 60-min gradient and analyzed on a Q Exactive HF-X (Thermo Fisher Scientific). Data-dependent fragmentation used collision-induced dissociation. RAW files were searched using MaxQuant under standard settings using the UniProt mouse database, allowing for two missed trypsin cleavage sites, variable modifications for N-terminal acetylation, and methionine oxidation. Ion tolerances of 20 and 6 parts per million were set for first and second searches, respectively. Candidate peptides and protein identifications were filtered on the basis of a 1% false discovery rate threshold based on searching of the reverse sequence database. The intensities of identified proteins were normalized to the intensity of AHR to examine how drug treatment affects levels of proteins bound to AHR. To remove potential contaminants, we eliminated proteins detected in IgG control group and required that a protein be identified in all three replicates and by at least two unique peptides. This threshold generated a list of 264 proteins. Proteins with at least twofold changes in response to PY109 and 5% intensity of AHR were highlighted in volcano plots.

For Western blots, bead-bound protein complexes were denatured by boiling with loading buffer containing sodium dodecyl sulfate (SDS) and dithiothreitol (DTT). SDS–polyacrylamide gel electrophoresis was performed to separate proteins, with transfer to a nitrocellulose membrane using a Bio-Rad mini trans-blot cell. Monoclonal mouse anti-AHR (RPT-1, Novus Biologicals), anti-ARNT (NB100-110, Novus Biologicals), and anti-AIP antibodies (35-2, Novus Biologicals) were applied at 1:1000 dilution, followed by incubation with alkaline phosphatase–conjugated secondary antibodies at 1:7500 dilution (Promega). Blots were imaged with Western Blue stabilized substrate (Promega). Band intensity was quantified using Image J.

Cyp1a1 induction

Cyp1a1 induction was evaluated in mouse Hepa-1c1c7 cells cultured in 24-well plates to confluence in full medium followed by a 24-hour incubation period in tryptophan-deficient media. Cells were incubated with vehicle control (DMSO), ITE, PY108, or PY109 (10 nM to 10 μM) for 8 hours, followed by cell lysis in TRIzol for RNA isolation.

Plasma and microsomal stability assays

Drug stability in plasma was evaluated by incubating 5 μM PY10, PY109, PY108, or ITE in mouse plasma for 0, 5, 15, 30, 45, and 60 min at 37°C with constant shaking at 225 rpm. Reactions were terminated by the addition of twofold (v/v) acetonitrile, followed by centrifugation at 14,000 rpm for 15 min to remove particulates, and the supernatant was analyzed by LC-MS/MS.

Drug metabolic stability was examined using a liver microsomal assay (45). Briefly, 5 μM PY10, PY109, PY108, or ITE was added to pooled mouse liver microsomes (0.5 mg/ml) in 100 mM phosphate buffer containing MgCl2 (Thermo Fisher Scientific). The assay was activated by the addition of an NAPDH (reduced nicotinamide adenine dinucleotide phosphate)–regenerating system containing NADP+, glucose-6-phosphate, and glucose-6-phosphate dehydrogenase (Sigma-Aldrich). Samples were incubated for 0, 5, 15, 30, and 45 min at 37°C, with constant shaking (225 rpm). Reactions were terminated by adding twofold (v/v) acetonitrile, samples were centrifuged at 14,000 rpm for 15 min, and the supernatant was analyzed using LC-MS/MS. t1/2 values were calculated according to the equationt1/2=0.693/kwhere k is the elimination rate constant, obtained by fitting C = initial × exp.k × t.

Caco-2 permeability assay

A Caco-2 permeability assay was performed (Alliance Pharma) to determine apparent cellular permeability and ER. Caco-2 cells were cultured to form a mature polarized enterocyte monolayer on a Transwell insert and 10 μM PY109 or PY108 was added to the apical or basolateral sides of the Transwell. Permeability coefficients (Papp) from apical to basolateral Papp A > B or from basolateral to apical Papp B > A were calculated. Efflux ratio was defined as Papp B > A/Papp A > B with active efflux represented by a ratio >2. Propranolol and digoxin were included as reference compounds.

PK and PD analysis

After oral administration of 1 mg/100 μl of PY108, PY109, or vehicle control (1% HPMC, 1% polysorbate 80), blood was collected into lithium heparin tubes at 0, 0.25, 0.5, 1, 2, 4, 12, and 24 hours. Samples were centrifuged to collect plasma (1500g for 10 min, room temperature) and mixed with cold acetonitrile (1:4) containing caffeine (5 mg/ml). After centrifugation at 17,000g for 20 min at 4°C, 80 μl of supernatant was isolated and dried in vacuo. Samples were reconstituted in 32 μl of NANOpure water, and 2 μl was injected onto a C18 trap column (5 mm by 0.3 mm, 5 μm) with 0.05% (v/v) trifluoroacetic acid and 2% (v/v) acetonitrile in water as the eluent. A binary gradient was applied to a C18 analytical column (150 μm by 75 μm, 2 μm) at a flow rate of 0.4 μl/min using 0.1% (v/v) formic acid in water (A) and in acetonitrile (B) as follows: 0 to 3 min, 40% (B); 3 to 8 min 75% (B); 8 to 16 min, 90% (B); 16 to 23 min, 90% (B); 23 to 24 min, 40% (B); and 24 to 34 min, 40% (B). Spectra were acquired in the positive mode using an Orbitrap Fusion Lumos mass spectrometer under the following settings: MS automatic gain control, 4 × 105; MS maximum injection time, 50 ms; MS/MS automatic gain control, 5 × 104; MS/MS maximum injection time, 60 ms; precursor ion isolation width, 1.6 Da; and higher-energy collisional dissociation normalized collision energy, 37. A standard curve was generated by adding varying concentrations of PY108 or PY109 to plasma followed by subsequent extraction in cold acetonitrile with caffeine. The response was plotted as the ratio of caffeine to PY108 or PY109. A linear regression curve was fit to the data and used to determine concentrations of PY108 or PY109. PD analysis consisted of measuring Cyp1a1 induction in response to PY109 (0.1 to 1000 μg) in 100 μl of vehicle, administered by oral gavage. After 12 hours, liver, colon, spleen, and leukocytes were harvested and RNA was isolated by TRIzol reagent method (Invitrogen) for qPCR analysis.

Dextran sodium sulfate induced murine colitis

C57BL/6 mice (8 to 10 weeks, male) were administered drinking water supplemented with 3% (wt/vol) DSS (MP Biomedicals) for 7 days followed by a 13-day recovery period, with standard drinking water. Test compounds, PY109 or PY108, were suspended in vehicle (1% HPMC and 1% polysorbate 80 in water) and administered by daily oral gavage (1 μg per mouse in 100 μl), beginning at day 0. Control mice were administered 100 μl of vehicle. Body weight and DAI were recorded daily. DAI was determined by combining scores related to weight loss, stool consistency, blood in stool, activity, and appearance as previously described with modification (46). Scores were determined as follows: change in weight (0, <1%; 1, 1 to 10%; 2, 10 to 18%; 3, 18 to 20%); stool consistency (0, normal; 1, loose stools; 2, diarrhea); stool blood (0, negative; 1, trace blood; 2, bloody stool; 3, gross bleeding); activity (0, normal; 1, low activity; 2, hiding; 3, no response to gentle stimuli); and appearance (0, normal; 1, rigor; 2, hunched posture). Additional cohorts of mice were enrolled to gain further insight into the mechanism of drug action. Mice were sacrificed on days 7 and 11, the colon was removed for length measurement and histological examination, and both RNA and LPMCs were isolated.

Histological analysis

The harvested colon was cut longitudinally and prepared using the Swiss roll method with the distal side in the center. Colons were fixed overnight in 10% neutral-buffered formalin, processed for paraffin embedding, and sectioned at a thickness of 5 μm. To characterize the inflammatory response and crypt loss, hematoxylin and eosin staining was performed. Sections were cleared with xylene, rehydrated through an alcohol gradient, stained with hematoxylin, and washed with acid and ammonium alcohol. Sections were then rinsed in water and exposed to eosin, followed by dehydration through graded alcohols and cleared in xylene. All sections were mounted using Permount (Thermo Fisher Scientific) and visualized using a BX41 microscope (Olympus). For histological scoring, three transverse sections of the Swiss roll were chosen and nine images (100×) were captured for each section. Images were analyzed for inflammation and crypt structure and scored by a blinded observer using the following schemes: inflammation (0, rare inflammatory cells in the lamina propria; 1, increased numbers of inflammatory cells in the lamina propria; 2, confluence of inflammatory cells extending into the submucosa; 3, transmural extension of the inflammatory infiltrate); and crypt loss (0, none; 1, basal one-third damaged; 2, basal two-third damaged; 3, only surface epithelium intact; 4, entire crypt and epithelium lost). Inflammation and crypt loss were scored separately and multiplied by a factor of 1 to 4 to account for the estimated involved area within the tissue section (1, 0 to 25%; 2, 25.1 to 50% affected; 3, 50.1 to 75% affected; 4, > 75% affected) (46). To examine goblet cell number, sections were stained with periodic acid–Schiff and Alcian blue (Thermo Fisher Scientific). Three transverse sections were selected for each colon, and five random images were acquired in the distal region at 200×. Goblet cell number was determined for each image by a blinded observer and normalized to colon length using image J.

Transcriptome profiling using RNA quantification sequencing and data analysis

RNA derived from untreated, DSS-affected colon with or without PY109 at day 11 were subjected to next-generation sequencing to generate deep coverage RNA-seq data. For each treatment group, sequencing was performed on four to five mice. Sequencing libraries of ribosomal digested RNA (NEBNext rRNA depletion kit, E6310) were generated using the NEBNext Ultra II directional RNA library prep kit for Illumina (E7760) according to the manufacturer’s protocol. Library quality control was checked using the DNA High Sensitivity Chip on the Agilent Bioanalyzer and qRT-PCR. High-quality libraries were sequenced on NextSeq 500. To achieve comprehensive coverage for each sample, we generated ~40 to 50 million paired-end reads of 150 nucleotides. The raw sequencing data were processed to remove any adaptor, PCR primers, and low-quality transcripts using FASTQC and Cutadapt. These provide a very comprehensive estimate of sample quality on the basis of read quality, read length, GC (guanine-cytosine) content, uncalled based, ratio of bases called, sequence duplication, adaptor, and PCR primer contamination. These high-quality, clean reads were aligned against the mouse mm10 genome using the Hisat2 algorithm. Gene expression measurement was performed from aligned reads by counting the unique reads. The read count–based gene expression data were normalized on the basis of library complexity and gene variation using the voom method. The normalized count data were compared among groups in a paired manner using a Linear Models for Microarray (Limma) approach to identify differentially expressed genes. The differentially expressed genes were identified on the basis of multiple test–corrected P value and fold change. Genes were considered significantly differentially expressed if the multiple test–corrected P value was ≤0.05 and the absolute fold change was ≥2.

Ingenuity Pathway Analysis (IPA 9.0, Qiagen) was used to identify the pathways that are significantly affected by differentially expressed genes. The knowledge base of this software consists of functions, pathways, and network models derived by systematically exploring the peer reviewed scientific literature. A detailed description of IPA analysis is available at the Ingenuity Systems’ web site (http://www.ingenuity.com). It calculates a P value for each pathway according to the fit of users’ data to the IPA database using one-tailed Fisher exact test. The pathways with P values <0.01 were considered significantly affected.

Isolation of colonic epithelial and lamina propria cells

Briefly, the colon was dissected and washed with HBSS to remove fecal contents, opened longitudinally, and cut into 2-cm segments. Epithelium was isolated by incubation in RPMI medium supplemented with 5 mM EDTA and 1 mM DTT at 37°C for 15 min with constant shaking. Crypt-enriched supernatants were combined, allowed to sediment at 37°C for 15 min, and washed once in HBSS. qPCR analysis of epithelial barrier gene expression required lysis of crypt-enriched supernatants in TRIzol. Intraepithelial lymphocytes were isolated and purified for flow cytometry by collecting the cell fraction interphase of 40 and 80% Percoll (GE Healthcare) after centrifugation at 500g for 20 min. The remaining colon fragments were minced into 1-mm pieces and digested in RPMI medium containing collagenase (1 mg/ml; Sigma-Aldrich) and DNase I (0.1 mg/ml; Sigma) at 37°C under constant shaking for 45 min. Digested cell mixtures were filtered through a 70-μm cell strainer, and lamina propria cells were purified by collecting the cell fraction at the interphase of 40 and 80% Percoll (GE Healthcare) after centrifugation at 500g for 20 min. Cells were washed and suspended at 106 cells/ml in RPMI medium containing 10% FBS, 1× nonessential amino acids, 1× sodium pyruvate, and 5 μM 2-mercaptoethanol. Intracellular cytokine staining for flow cytometry was performed by treating cells with Cell Stimulation Cocktail (00-4970-03, Thermo Fisher Scientific) containing phorbol 12-myristate 13-acetate (PMA), ionomycin, and brefeldin A for 8 hours.

Flow cytometry

Cells were washed, blocked with anti-CD16/32 (2.4G2, eBioscience), and stained in permeabilization buffer (eBioscience) with isotype control or antibodies against murine CD3 (17A2, BD Pharmingen), CD4 (RM4-5, eBioscience), TCR γδ (GL-3, eBioscience), RORγt (B2D, eBioscience), Foxp3 (FJK-16s, eBioscience), IL-22 (1H8PWSR, eBioscience), IL-17A (17A2, BD Pharmingen), IFN-γ (XMG1.2, BD Pharmingen), and TNF-α (MP6-XT22, eBioscience). For human T cells, the following antibodies were used: CD4 (RPA-T4, Invitrogen), IL-22 (#142928, R&D Systems), and IL-17A (eBio64DEC17, eBioscience). Analysis was performed on an LSRII flow cytometer (BD Bioscience). Lymphocytes were gated using forward scatter and side scatter, and within the lymphocyte gate, the populations were identified as T helper cells (CD3+CD4+), regulatory T cells (CD3+CD4+Foxp3+), innate lymphoid cell type 3 (ILC3, CD3RORγt+), or γδ T cells (CD3+ TCR γδ+). Cytokine expression levels were further quantified within defined cell populations.

qPCR analysis

RNA was extracted using TRIzol (Invitrogen) and 0.1 μg of RNA was converted to complementary DNA (cDNA) using a high-capacity cDNA reverse transcription kit (Invitrogen). qPCR was performed on the Applied Biosystems 7900 using TaqMan Universal PCR Master Mix (Life Technologies) with TaqMan (Life Technologies) primers against murine Cyp1a1 (Mm00487218_m1), Ahr (Mm00478932_m1), Il22 (Mm01226722_g1), Il17a (Mm00439618_m1), Foxp3 (Mm00475162_m1), Il10 (Mm01288386_m1), Tnf (Mm00443258_m1), Ifng (Mm01168134_m1), S100a8 (Mm00496696_g1), S100a9 (Mm00656925_m1), Reg3g (Mm00441127_m1), Muc1 (Mm00449604_m1), Muc3 (Mm01207064_m1), and Bcl2l1 (Mm00437783_m1). Fold increase was calculated using the ΔΔCT method with Rn18s (Mm00446186_m1) as an internal reference gene. For study of human genes, the following primers were used: 18S (Hs99999901_s1), AHR (Hs00169233_m1), CYP1A1 (Hs01054797_g1), IL17A (Hs00174383_m1), and IL22 (Hs01574154_m1).

In vitro differentiation of murine Th17 cells for analysis of cytokine expression

The ability of test compound to induce expression of IL-22 and IL-17A was examined in spleen-derived CD4+ T cells. Spleens were harvested from C57BL/6 mice and fragmented, red blood cells were lysed by osmotic shock, and single-cell suspensions were collected using a 70-μm cell strainer. CD4+ T cells were isolated by negative selection using a T cell isolation kit (Miltenyi Biotec) and stimulated with plate-bound CD3 and soluble CD28 antibody (5 μg/ml) in Th17 polarization medium [IL-6 (20 ng/ml), TGF-β (1 ng/ml), IL-1β (10 ng/ml), anti–IFN-γ (10 μg/ml) in Iscove’s modified Dulbecco’s medium supplemented with 2 mM glutamine, 100 U of penicillin, 100 U of streptomycin, 50 μM 2-mercaptoethanol, and 5% FBS] for 4 days (26). Cells were treated with vehicle (DMSO) or PY109 (100 nM to 10 μM) beginning on day 0 and gene expression was evaluated on day 3. Cytokine expression was evaluated by flow cytometry on day 4 after re-stimulation with phorbol dibutyrate (500 ng/ml), ionomycin (500 ng/ml), and brefeldin A (10 μg/ml) for 4 hours. IL-22 and IL-17A were measured in culture medium on day 4 by ELISA (R&D Systems).

Cytokine expression in murine LPMC-derived Th17 cells and ILC3 cells

LPMCs were isolated from C57BL/6 mice, as previously described. To activate and differentiate CD4+ T cells, LPMCs were cultured in Th17 polarization medium with CD3/CD28 ligation. To activate ILC3 cells, LPMCs were cultured with IL-1 (10 ng/ml) and IL-23 (10 ng/ml). Cells were treated with 10 μM PY109 for 3 days, after which cells were re-stimulated for 8 hours with PMA, ionomycin, and brefeldin A, and intracellular IL-22 and IL-17A staining was performed, along with T helper markers (CD3+CD4+) or ILC3 markers (CD3-RORγt+).

Antimicrobial peptide expression in CMT-93 colon epithelial cells

The colonic epithelial cell line, CMT-93, was purchased from the American Type Culture Collection and used between passages 1 to 3. CMT-93 cells were cultured until 80% confluency in DMEM with 10% FBS. Antimicrobial peptide expression was measured after incubation with IL-22 and IL-17A. Cells were treated with recombinant mouse IL-22 (10 ng/ml; R&D Systems) or recombinant mouse IL-17A (50 ng/ml; R&D Systems) or both for 48 hours in full medium. Conditioned medium from Th17-polarized T cells treated with or without 10 μM PY109 was collected after 4 days, frozen in aliquots, and stored at −80°C. CMT-93 cells were incubated with a 1:1 mix of conditioned medium and DMEM with 10% FBS for 48 hours. Th17 polarization medium in the absence of T cells with or without 10 μM PY109 served as a negative control. In select experiments, anti–IL-22 (IL-22JOP, eBioscience or AF582, R&D Systems) or anti–IL-17A (eBioMM17F3, eBioscience) neutralizing antibodies or the corresponding IgG isotype was added to culture medium at a concentration of 4 μg/ml.

Human T cell isolation and polarization

Peripheral blood was drawn from 10 healthy donors and 13 patients with active IBD recruited from the Gastroenterology Division of Beth Israel Deaconess Medical Center. The institutional review board approved all the procedures described in this work. Of these patients, eight had ulcerative colitis and five had Crohn’s diseases. At the time of study, six patients were receiving infliximab or adalimumab, three were on steroids, and four were treated with mesalamine. PBMCs were separated using Ficoll-Paque PLUS (GE Healthcare) and subjected to negative selection using the pan-CD4+ T cell isolation kit (130-096-533, Miltenyi Biotec). Cells were stained for CD4 (RPA-T4, Invitrogen) and 97% purity was achieved. Th17 polarization was conducted by supplementation with IL-6 (30 ng/ml), IL-1β (10 ng/ml), TGF-β (1 ng/ml), anti–IFN-γ (1 μg/ml), and anti–IL-4 monoclonal antibody. Vehicle or 5 μM PY109 was added at the beginning of the culture period. Cell density, antibody, and cytokine concentrations were titrated to optimize proliferation and IL-17A production. Cells were re-stimulated after 5 days in culture with stimulation cocktail (00-4975-03, Invitrogen) containing PMA, ionomycin, and brefeldin A for 5 hours and harvested for qPCR and flow cytometry.

Statistical analysis

Data are expressed as mean ± SEM for the indicated experimental groups. All data represent measurements taken from distinct samples including independent cell cultures and independent in vivo tissue and cell samples. Experiments were repeated in triplicate and a representative data set was presented. Statistical differences were assessed using a two-tailed unpaired Student’s t test. For human T cell data, statistical differences were assessed using a two-tailed paired Student’s t test to account for individual baseline differences. For multiple comparisons with independent variables (Figs. 3B and 5, K, M, and N, and fig. S14), one-way analysis of variance (ANOVA) and post hoc Student’s t test with Bonferroni’s modification were performed (GraphPad Prism 5). A P value of less than 0.05 was considered statistically significant.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/3/eaay8230/DC1

Supplementary Materials and Methods

Fig. S1. AHR homology model.

Fig. S2. Conformational analyses of ligands.

Fig. S3. Lead optimization for metabolic stability.

Fig. S4. Biological characterization of lead compounds in vitro and after oral administration.

Fig. S5. Toxicity study of PY109.

Fig. S6. Lead compounds attenuate DSS-induced colitis in mice.

Fig. S7. Effect of PY109 on Th17 and CMT-93 cells.

Table S1. Representative SARs for screening library.

Table S2. SARs of pyridine series.

Table S3. Pathway analysis for PY109 counter-regulated genes.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: We acknowledge support from NIHT32 HL007734 fellowship for D.J.W. and NIHT35 HL 110843 fellowship for R.C.D.L.R. and A.M.P. (Program Director, F. W. LoGerfo). We thank the members of the laboratories of E.L.C. and L.S. for helpful discussions, as well as input from M. S. Longhi regarding the DSS colitis model and J. C. Tigges of the BIDMC flow cytometry core. Author contributions: J.C., C.A.H., L.S., and E.L.C. designed experiments; S.K.K., Y.W., and J.E.C. performed chemical synthesis and characterization of AHR agonists; F.E.J., and L.S. performed molecular modeling; J.C., D.J.W., R.K., D.D.P., R.C.D.L.R., A.M.P., and L.L. conducted biological experiments; J.E.C., J.K., and A.E. performed and analyzed immunoprecipitation-MS results; B.T., S.B., and M.B. performed and analyzed RNA-seq results; E.P. and A.C.M. contributed to the design and collection of clinical samples; J.C., C.A.H., L.S., and E.L.C. wrote the manuscript. Competing interests: L.S. and E.L.C. are inventors on a pending patent related to this work filed by the Beth Israel Deaconess Medical Center (International Patent Application No.: PCT/US2019/025993; based on U.S. Patent Application No. 62/653257, filed 5 April 2018). The authors declare no other competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data are available from the authors upon request. The compound PY109 can be provided by E.L.C. and L.S. pending scientific review and a completed material transfer agreement. Requests for PY109 should be submitted via email to E.L.C. or L.S.
View Abstract

Stay Connected to Science Advances

Navigate This Article