Intestinal mucosal barrier repair and immune regulation with an AI-developed gut-restricted PHD inhibitor

Abstract

Hypoxia-inducible factor prolyl hydroxylase (PHD) inhibitors have been approved for treating renal anemia yet have failed clinical testing for inflammatory bowel disease because of a lack of efficacy. Here we used a multimodel multimodal generative artificial intelligence platform to design an orally gut-restricted selective PHD1 and PHD2 inhibitor that exhibits favorable safety and pharmacokinetic profiles in preclinical studies. ISM012-042 restores intestinal barrier function and alleviates gut inflammation in multiple experimental colitis models.

Fig. 1. Development of a specific and gut-restricted potent PHD inhibitor, ISM012-042.

a, AI-powered workflow of ISM012-042’s discovery and design. MCF, medicinal chemistry filter; SA, synthetic accessibility; SOM, self-organizing map. b, Crystal structure of ISM042-012 with PHD2 (Protein Data Bank (PDB) 8Y0V, 2.5 Å). c, Comparison of crystal structures of PHD2 in complex with ISM012-042 and 2-OG. d,e, PHD1 hydroxylase (d) and PHD2 hydroxylase (e) inhibition of ISM012-042 relative to DMSO control. f,g, Luminescent HiBiT-tagged HIF1α accumulation in Caco-2 cells treated with serially diluted ISM012-042 (f) or AKB-4924 (g) for 6 h in the presence of Fe(II) sulfate. Values in d–g are the mean ± s.d. (n = 2 biological replicates per treatment condition); the representative plot from three independent experiments is shown. NA, not applicable. h, TEER of DSS-induced Caco-2 cell monolayers pretreated with ISM012-042 (mean ± s.d.; n = 5–6 biological replicates per group). P values at the 96-h time point were calculated using a two-way ANOVA with post hoc Tukey’s multiple-comparisons test. i, mRNA expression of IL-12 and TNF in LPS-stimulated or baseline murine BMDCs following 24-h treatment with ISM012-042 or roxadustat (mean ± s.d.; n = 6 biological replicates per group). P values calculated using a one-way ANOVA with Dunnett’s test. j,k, LC–MS/MS-based quantitation of ISM012-042 presence in the colon and plasma of healthy C57BL/6 (j) or oxazolone-induced colitis (k) mice at 2 days after intracolonic induction by orally administered 10 mg kg⁻¹ ISM012-042 at the indicated time points (n = 3 healthy or 5 colitis mice per time point). Data represent the mean ± s.d. l, Compound tissue distribution at the indicated time points following a single oral dose of 30 mg 100 μCi⁻¹ kg^{−1 14}C-labeled ISM012-042 in healthy SD rats. Data below the limit of quantification (LOQ) labeled in red were input as the LOQ/2 of the respective tissue (n = 6 per time point; n = 3 for reproductive organs per time point). Comparisons with no P value are nonsignificant (NS). Source data

Fig. 2. ISM012-042 resolves colitis, restores intestinal barrier function and reprograms the colonic immune environment.

a–c, ISM012-042 or mesalamine was orally administered once daily to TNBS colitis mice starting on day −1. a, Daily DAI scores. Data represent the mean ± s.d. (n = 8 mice per group for the sham, vehicle and mesalamine groups; n = 14 mice per group). P values on day 7 were calculated using two-way ANOVA with a post hoc Dunnett’s test. b, HIF1α staining of colonic sections from sham-treated, vehicle-treated and ISM012-042-treated TNBS colitis mice. Green arrowheads are HIF1α-positive signals in luminal epithelial cells. Scale bars, 300 μm (mean ± s.d.; n = 6 mice per group). P values were calculated using Dunnett’s test following a one-way ANOVA. c, qPCR of mucosal epithelial HIF1α target genes (mean ± s.d.; n = 8 mice per group). P values were calculated using post hoc Dunnett’s test following a one-way ANOVA. d–f, ISM012-042 or mesalamine was administered once daily to TNBS colitis mice starting on day 2 after induction. d, Daily DAI scoring (mean ± s.d.; n = 8 mice per group). P values on day 7 were calculated using Dunnett’s test following a two-way ANOVA. e, Plasma levels of FITC–dextran 4000 (FD4) on day 7 after induction (mean ± s.d.; n = 8 mice per group). P values were calculated using Dunnett’s test following a one-way ANOVA. f, Flow cytometric analysis of immune cell subsets in colonic lamina propria isolated on day 7 after induction (mean ± s.d.; n = 4 per group). P values were calculated using a two-sided unpaired t-test. g–j, ISM012-042 was administered to oxazolone-induced colitis mice starting on day −1. g, Daily DAI scoring (mean ± s.d.; n = 11 mice per sham and vehicle groups and 26 mice per ISM012-042-treated group). P values were calculated using Dunnett’s test following a two-way ANOVA. h, Cytokine profiling of whole-colon tissue collected 4 h after final dosing. i,j, Plasma IL-1β (i) and fecal LCN2 (j) at 4 h after final dosing (mean ± s.d.; n = 8 mice per group). P values were calculated using Dunnett’s test following a one-way ANOVA. k–m, ISM012-042 was administered to oxazolone-induced colitis mice starting on day 2 after induction. k, Daily DAI scoring (mean ± s.d.; n = 8 mice per group). P values on day 7 were calculated using Dunnett’s test following a two-way ANOVA. The arrow indicates the first dosing of ISM012-042, mesalamine or vehicle. l, Plasma levels of FD4 on day 7 after oxazolone induction (mean ± s.d.; n = 8 mice per group). m, Colon weight and length on day 7 after oxazolone induction (mean ± s.d.; n = 8 mice per group). QD means the indicated compound was administered once daily. P values in l,m were calculated using Dunnett’s test following a one-way ANOVA. Comparisons with no P value are NS. Source data