Inflammatory Bowel Disease: A Pressing Need for Alternative, Restorative Therapies
Inflammatory bowel disease (IBD) is an umbrella term for a set of autoimmune disorders affecting the digestive tract– Crohn’s disease (CD) and ulcerative colitis (UC). Most cases of the estimated >6.8 million individuals globally affected by IBD1 suffer from moderate to severe disease2, and these patients are most likely to initiate treatment with advanced therapies, particularly anti-TNF biologic therapy3. Unfortunately, as many as 60 or 85% of patients receiving biologic advanced therapy exhibit insufficient response3,4, with a lower time on treatment for subsequent treatments5. The high rate of failure for these advanced therapies speaks to the need for additional therapeutic options in moderate to severe IBD.
Moreover, the current standard-of-care IBD therapies have traditionally not been developed specifically to resolve inflammation of the gut at the cellular level, often referred to as histologic healing or histologic remission, but rather to ease symptoms and restore patient quality-of-life6. Histologic remission, characterized by restoration of the gut epithelial barrier, tissue repair, and reduction of immune cell infiltration into the gut tissue, however, is associated with a lower incidence of long-term relapse risk, hospitalization, surgical intervention, and colorectal cancer than the traditional goal of endoscopic remission6,7. New therapies, therefore, would benefit from targeting such deep remission to most effectively help IBD patients who remain at high risk.
Drug development timeline for ISM012-042
Figure 1: A timeline covering the full scope of our AI-driven study. Following target discovery in 2019, ISM012-042 was first synthesized and validated as a potent, safe, and selective PHD1/2 inhibitor in under 12 months. Its efficacy was tested in four in vivo murine fibrosis models, including both prophylactic and post-onset settings, and is currently being tested in two Phase 1 clinical trials in Australia and China.
Before Taking A Shot, You Have To Find The Target
Targeted therapy has emerged as a mainstay of IBD treatments to augment or supplant broadly acting immunosuppressive agents. Targets of today’s assortment of therapies, for both biologics and small-molecules, include cytokines like TNF, IL-23, and IL-12/IL-23; lymphocyte activation kinases JAK1 and JAK1/3; and lymphocyte homing receptors integrin α4β7 and S1PR. All of these agents exert their effects in a relatively straightforward way by blocking inflammatory pathways via well-established disease-associated effectors and receptors.
Our group instead takes a top-down approach to re-prioritize disease-associated molecular targets, leveraging the enormous corpus of published and publicly available “omics” data sets to identify disease targets de novo that may not have been identified or prioritized previously. Our artificial intelligence (AI)-powered drug discovery methods can uniquely integrate not only high-dimensional multi-omic data sets, such as gene expression analyses, epigenetic profiling, and proteomics, but also text-mining natural language processing models to learn disease-gene associations from publications, grants, patents, clinical trial databases, and other sources. Wrapped in a single commercially available platform, our PandaOmics8 AI tool employs 23 disease-specific models to derive gene interaction networks and knowledge graphs to ultimately rank genes for probability of successful pharmacologic targeting.
Each step in the drug discovery process doesn’t exist in a vacuum but as part of a multifactorial development pipeline that interfaces with the economic landscape of funding and drug commercialization. PandaOmics factors in these considerations with customizable models that integrate characteristics of potential genetic targets predictive of drug development success, such as druggability, safety, novelty, and commercial tractability to nominate targets for therapies most likely to reach patients.
We aimed our PandaOmics engine at IBD to begin the process of developing a maximally effective and safe small-molecule drug for the patients who still face uncertain prognosis with current therapies. The AI platform ranked the PHD1/2-HIF1α signaling axis as a top potential target.
HIF1α is a well-known transcription factor subunit that activates expression of a variety of metabolism- and angiogenesis-related genes under hypoxic conditions9. Under normoxic conditions, HIF1α is rapidly degraded following hydroxylation by PHD1-3, which utilizes oxygen (O2) as a co-factor, resulting in low steady-state expression of these genetic programs. Hypoxic conditions, and the subsequent lack of oxygen as a co-factor, prevents PHD activity, leading to stabilization of HIF1α and downstream transcriptional activation9,10. The healthy intestinal epithelium has a well-structured oxygen gradient, spanning from the highly hypoxic lumen to the vascularized and immune-rich lamina propria. IBD is characterized by a loss in gut epithelium integrity, causing leakiness of the hypoxic and microbe-rich intestinal lumen into the surrounding gut tissue, leading to microbial infiltration, immune activation, and chronic inflammation with a positive-feedback loop driving maintenance of hypoxia and a disrupted hypoxia gradient. Giving the tissue a chance to repair and restore gut integrity by dampening this inflammatory response is a key goal of the immunosuppressing treatment options today. Mouse models of colitis exhibited reduced disease severity with increased HIF1α expression and decreased PHD expression10, but therapeutic targeting of the HIF1α pathway via the PHDs has so far been underexplored, with only one clinical-stage drug, GB004, that ended up failing to demonstrate efficacy in phase 2 testing for UC10,11.
Diagram of IBD
Figure 2. IBD-affected gut tissue is characterized by a breakdown in epithelial barrier integrity, infiltration and activation of inflammatory immune cells, and a dysregulated hypoxia gradient. Inhibition of PHD factors stabilizes HIF1α, allowing it to activate transcription of barrier-protective genes.
AI on the Prize: A Novel Gut-Restricted PHD1/2 Inhibitor for IBD
Serendipitously, we previously published our medicinal chemistry work developing a novel class of PHD inhibitor for the treatment of anemia12. Chronic kidney disease (CKD)-associated anemia is frequently driven by decreased production of erythropoietin (EPO) by the kidney. As one of the key transcriptional targets of HIF1α under hypoxic conditions, induction of EPO expression by stabilization of HIF1α is one strategy for treating anemia for patients with CKD. To this end, several pan-PHD inhibitors have been approved for the treatment of CKD-associated anemia globally13.
Before being repurposed to treat IBD, our PHD inhibitor had to be further developed to address IBD-specific shortcomings of other investigative PHD inhibitors, such as low efficacy and high toxicity due to systemic activation of HIF1α targets like EPO and VEGF. To achieve this, we again turned to Chemistry4214, our AI-powered generative chemistry platform, which had previously been used to generate our top hit PHD inhibitor for anemia12. Just like the PandaOmics target discovery platform learns features from sources beyond disease biology and genomics datasets (such as novelty or commercial tractability) to nominate therapeutic targets, the Chemistry42 platform can integrate data types beyond biophysical models of target-inhibitor interactions to nominate candidate drug molecules. Thus, we leveraged the modular nature of Chemistry42 to focus on optimizations of the candidate molecule with Alchemistry, the module for fine-tuning potency by binding free-energy estimation, and ADMET Profiling, the module for predicting key properties such as solubility, permeability, and toxicities. Prior generation of the original PHD inhibitor hit also included assessment and prioritization of other features, including synthetic accessibility, novelty, and diversity.
We hypothesized that the key criteria to generate a maximally effective and safe PHD inhibitor for IBD included minimizing solubility, maximizing clearance, and maintaining moderate epithelial permeability, such that the drug would remain in the digestive tract following oral administration, penetrate into the adjacent epithelial and lamina propria tissue most typified and damaged by chronic inflammation in the setting of IBD, and be excreted quickly to reduce exposure to non-gastrointestinal tissues. In the end, our optimizations resulted in ISM012-042, a selective PHD1 and PHD2 inhibitor with impressive on-target IC50 of 1.9 and 2.5 nM, respectively, and nearly two orders of magnitude enrichment in colon tissue compared to plasma of both healthy and colitis mouse models.
What may be most impressive about the discovery of ISM012-042 is actually the accelerated discovery timeline that the generative AI platforms afforded. Traditional drug discovery pipelines rely on massive chemical screens and biology-guided prioritization of genetic and chemical space for both target and compound discovery, taking an average of three years from hit generation to nomination of a preclinical candidate molecule. Using PandaOmics and Chemistry42, we went from project initiation through preclinical candidate nomination in just 12 months, synthesizing only 115 molecules for hands-on testing. This is yet another example in only the past couple years of an extremely condensed drug development timetable enabled by generative AI-powered drug discovery platforms, following on the heels of our preclinical discovery of a safe and effective TNIK inhibitor for the treatment of idiopathic pulmonary fibrosis (IPF)15. The field of AI-driven drug design (AIDD) has been searching for the first “wins” in the race to reduce the notoriously high costs and long development times for any single new drug, and these recent preclinical successes suggest we may have reached an inflection point in tipping the scales to that end.
Please check out our previous Behind the Paper article discussing the exciting development of our clinical-stage TNIK inhibitor rentosertib (ISM001-055), described in another Nature Biotechnology publication.
ISM012-042: In Vitro and In Vivo Efficacy Against Models of IBD
Our new PDH inhibitor, ISM012-042, ended up acing the gamut of preclinical assays we ran it through to test for potential efficacy and safety. Low-nanomolar in vitro IC50 indicated promising on-target inhibition provided by our structure-activity relationship optimizations, while changing the concentration of iron in the cell culture media had no effect, in stark contrast to GB004, which exhibited significantly decreased activity and target specificity in high-iron conditions, indicative of adverse off-target effects due to its iron-chelating activity. ISM012-042 protected the integrity of Caco-2 cell monolayers from barrier-disrupting dextran sodium sulfate (DSS), a known animal colitis induction agent. ISM012-042 also reduced the expression of inflammatory cytokines, such as IL-12 and TNFα, in activated mouse bone marrow-derived dendritic cells (BMDCs) as effectively as PHD inhibitor Roxadustat, suggestive of strong HIF1α-mediated anti-inflammatory efficacy.
Encouraged by the positive in vitro studies, we moved to in vivo evaluation of ISM012-042 in animal models of IBD. In mice and rats, ISM012-042 exhibited extremely high restriction to the gut, proving that the fine-tuning of the molecular design to maximize exposure exclusively in the digestive tract was successful. In both prophylactic and post-onset models of 2,4,6-trinitrobenzene sulfonic acid (TNBS)- and oxazolone-induced colitis in mice, ISM012-042 dose-dependently attenuated colitis progression, promoted remission from colitis symptoms with lower IBD histopathology scores, increased HIF1α staining intensity along the luminal aspect of colon tissues, and increased expression of HIF1α-induced epithelial barrier-protective genes like Tff3, Tjp1, and Nt5e. ISM012-042-treated colitis mice had broadly reduced cytokines and chemokines in isolated gut tissue, including IL-17 pathway, TNFα, IL-6, and IL-12 genes, as well as reduced pro-inflammatory immune cell types, such as monocytes, neutrophils and T cells expressing TNF, IFN-γ , and/or IL-17A. Further, in a very clear visualization of the treatment’s efficacy, excised colon tissue from colitis mice treated with ISM012-042 shows striking resolution of gross inflammation and colon shortening indicative of IBD-associated tissue damage in under one week of treatment.
ISM012-042 treatment mitigated TNBS-induced colitis pathology in mouse models of IBD when administered after disease onset.
Figure 3. ISM012-042 treatment mitigated TNBS-induced colitis pathology in mouse models of IBD when administered after disease onset. ISM012-42 was administered once-daily starting on day 2 after colitis induction, and dissected tissue was collected on day 7.
Importantly, ISM012-042 reduced body weight loss or even increased body weight of colitis mice. Not only functioning as a metric for alleviated IBD severity, this demonstrates the absence of systemic toxicity with treatment, further evidenced by no increase in levels of circulating EPO and VEGF, a sign of gut-restricted activation of HIF1a and rapid systemic clearance of the drug.
Looking to the Future of ISM012-042 and AIDD
Our work demonstrating gut-restricted activity of ISM012-042 and resolution of IBD without significant toxicity represents the latest confirmation that achieving the promise of AIDD is within reach. From program initiation to preclinical candidate nomination in 12 months, this study highlights the efficiency– in terms of time and money– brought to the drug discovery process by AI tools integrated at every step. Ultimately, the benefit to patients in need of affordable therapeutic options is the driving force in delivering pharmaceutical solutions more quickly and less expensively. The transformation of drug development pipelines that AI tools bring is immensely exciting, and we look forward eagerly to the ever-growing number of such drugs entering clinical testing and thus to patients in need.
Insilico Medicine recently announced that ISM012-042 (formerly ISM5411) has entered phase I clinical trials to assess safety, tolerability, pharmacokinetics, and food effects in healthy volunteers.
Figure 4. Insilico Medicine recently announced that ISM012-042 (formerly ISM5411) has entered phase I clinical trials to assess safety, tolerability, pharmacokinetics, and food effects in healthy volunteers.
ISM012-042 is progressing through a phase 1 clinical trial (NCT06012578) assessing safety, tolerability, pharmacokinetics, and food effects in healthy volunteers, having now completed dosing in two cohorts. This trial adds to the growing body of clinical studies for which Insilico Medicine has provided end-to-end drug development capabilities. Our recent report of positive top-line results for a phase 2a trial of rentosertib (TNIK inhibitor) for the treatment of IPF15,16 shows that AI-powered and -accelerated drug discovery has the potential to bring solutions to previously intractable clinical problems and expand the clinical toolkit for hard-to-treat disease. Modular and customizable AI platforms like PandaOmics and Chemistry42 can be taken by any pharmaceutical development group and tailored to any disease of interest to design drugs with properties uniquely advantageous to that disease. We hope others can use our software suite and implement their own tweaks and customizations to target, discover, and design novel therapeutics that can bring relief to patients in need across diseases. The work is now only beginning!
References
1. Alatab, S. et al. The global, regional, and national burden of inflammatory bowel disease in 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. The Lancet Gastroenterology & Hepatology 5, 17–30 (2020).
2. Burisch, J. & Munkholm, P. The epidemiology of inflammatory bowel disease. Scandinavian Journal of Gastroenterology 50, 942–951 (2015).
3. Gibble, T. H. et al. Identification of inadequate responders to advanced therapy among commercially-insured adult patients with Crohn’s disease and ulcerative colitis in the United States. BMC Gastroenterology 23, 63 (2023).
4. Bokemeyer, B., Picker, N., Wilke, T., Rosin, L. & Patel, H. Inadequate Response, Treatment Patterns, Health Care Utilization, and Associated Costs in Patients With Ulcerative Colitis: Retrospective Cohort Study Based on German Claims Data. Inflammatory Bowel Diseases 28, 1647–1657 (2022).
5. Brady, J. E., Stott-Miller, M., Mu, G. & Perera, S. Treatment Patterns and Sequencing in Patients With Inflammatory Bowel Disease. Clinical Therapeutics 40, 1509-1521.e5 (2018).
6. Pai, R. K. et al. Histologic assessments in ulcerative colitis: the evidence behind a new endpoint in clinical trials. Expert Review of Gastroenterology & Hepatology 18, 73–87 (2024).
7. Bryant, R. V., Winer, S., Spl, T. & Riddell, R. H. Systematic review: Histological remission in inflammatory bowel disease. Is ‘complete’ remission the new treatment paradigm? An IOIBD initiative. Journal of Crohn’s and Colitis 8, 1582–1597 (2014).
8. Kamya, P. et al. PandaOmics: An AI-Driven Platform for Therapeutic Target and Biomarker Discovery. J. Chem. Inf. Model. 64, 3961–3969 (2024).
9. Weidemann, A. & Johnson, R. S. Biology of HIF-1α. Cell Death Differ 15, 621–627 (2008).
10. Bhat, S. & Rieder, F. Hypoxia-Inducible Factor 1-Alpha Stabilizers in the Treatment of Inflammatory Bowel Diseases: Oxygen as a Novel IBD Therapy? J Crohns Colitis 16, 1924–1932 (2022).
11. GB004, Inc. A Phase 2, Randomized, Double-Blind, Placebo-Controlled, Multi-Center Study to Evaluate GB004 in Adult Subjects With Mild-to-Moderate Active Ulcerative Colitis. https://clinicaltrials.gov/study/NCT04556383 (2023).
12. Xu, J. et al. Discovery of Novel and Potent Prolyl Hydroxylase Domain-Containing Protein (PHD) Inhibitors for The Treatment of Anemia. J. Med. Chem. 67, 1393–1405 (2024).
13. Galkin, F. et al. Roles of hypoxia-inducible factor-prolyl hydroxylases in aging and disease. Ageing Research Reviews 102, 102551 (2024).
14. Ivanenkov, Y. A. et al. Chemistry42: An AI-Driven Platform for Molecular Design and Optimization. J. Chem. Inf. Model. 63, 695–701 (2023).
15. Ren, F. et al. A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models. Nat Biotechnol 1–13 (2024) doi:10.1038/s41587-024-02143-0.
16. Nature, R. C. by S. Preliminary Phase 2a Readout for a Novel Drug Discovered and Designed Using Generative AI Sets a Major Milestone in AI-Powered Drug Discovery. Research Communities by Springer Nature https://communities.springernature.com/posts/preliminary-phase-2a-readout-for-a-novel-drug-discovered-and-designed-using-generative-ai-sets-a-major-milestone-in-ai-powered-drug-discovery (2024).