AI-based drug developer Insilico Medicine detailed in a peer-reviewed paper how it applied its Chemistry42 platform to develop a treatment for inflammatory bowel disease (IBD) that is in early clinical trials and could be the first to treat the disease by inhibiting the prolyl hydroxylase domain (PHD).
In the paper, published in Nature Biotechnology, Insilico founder and CEO Alex Zhavoronkov, PhD, and 22 co-authors based at the company shared how the company used its Chemistry42 generative chemistry engine to design and optimize novel gut-restricted inhibitors of PHD1 and PHD2, protein targets implicated in biological processes that include aging and regeneration.
PHD has been shown to play a critical role in regulating the stability and transcriptional activity of hypoxia-inducible factor-1α (HIF-1α), leading to the expression of those gut barrier protective genes.
“PHD is not a super novel target but it is novel for IBD. Our company wanted to go into IBD as it is a very large patient population and unmet medical need,” Zhavoronkov told GEN Edge. “Most other companies focus on the inflammatory cytokine pathways, immune cell trafficking mechanisms, or intracellular signaling nodes like JAK. Most of these just treat or ameliorate the symptoms. We wanted to produce a lasting disease-modifying effect.”
Insilico said it has advanced the best of its PHD inhibitors into the clinic as ISM5411, an oral, PHD-specific inhibitor designed to treat IBD by inducing the expression of gut barrier protective genes. ISM5411 is designed to directly regulate the intestinal immune microenvironment, so that the drug simultaneously reduces the gut inflammation seen in IBD while also promoting repair of the intestinal mucosal barrier.
With that dual purpose in mind, Insilico researchers identified and prioritized the PHD1/2-HIF-1α axis as a therapeutic target for IBD, using its PandaOmics target identification engine to prioritize multiple targets for mucosal barrier repair.
We also wanted to make sure that the target is implicated in aging as many of our targets are “dual-purpose” targeting aging and disease at the same time,” Zhavoronkov explained.
He cited the practice of several longevity drug developers in prioritizing PHD for aging. BioAge Labs, for example, had licensed the small molecule oral HIF-PH inhibitor BGE-117 from Taisho Pharmaceutical for development to treat anemia, launching two Phase II studies (NCT05152641 and NCT04815603 ) that were both withdrawn in 2022—due to a shift in program priority and not safety, BioAge stated on ClinicalTrials.gov.
Going after multiple diseases
“We wanted to go after multiple diseases including IBD,” Zhavoronkov explained. “The beauty of having a mucosal barrier repair molecule is that it can be combined with multiple anti-inflammatory and other drugs to deliver even greater benefit.”
In terms of the mechanism, intestinal PHD1/2-HIF-1α signaling is critical for maintaining epithelial barrier integrity by regulating processes such as mucus production, tight junction assembly, immunosuppressive metabolite synthesis, and wound repair.
Having zeroed in on inhibiting PHD, researchers then employed Chemistry42 and its generative chemistry, Alchemistry, and ADMET analysis modules to design PHD inhibitors with the properties they sought.
Chemistry42 is an automated, machine-learning de novo drug design and scalable engineering platform, which according to the company, enables users to find novel lead-like molecules in as little as a week. Drawing on large numbers of compounds and molecular fragments, Chemistry42 uses generative AI to create new drug-like molecules optimized to have specific properties.
Using the platform’s Aichemistry module, Insilico’s researchers initiated fragment growth by pre-establishing a priority fragment, the benzonitrile group, and six key pharmacological points based on the reported PHD complex structures. By enhancing their novelty and synthetic availability filters, the researchers generated and screened hit compounds with activity deemed to be promising, then used Structure-Activity Relationship (SAR) analysis to optimize the properties of potential candidates. AIchemistry helped researchers assess the potency of those generated compounds to compute binding free-energy estimates for PHD2-ligand complexes.
Next, investigators used the ADMET module to collect key predicted properties, prioritizing low solubility and high clearance to minimize systemic exposure while retaining its moderate permeability for cell penetration.
“The generative chemistry module identified a novel PHD inhibitor candidate scaffold, the Alchemistry module optimized its potency by refining the Insilico Medicine Restricted (Level 3) molecular structure, and the ADMET profiling module modeled the physiochemical and biological properties that led us to select an effective and safe gut-restrictive molecule,” Zhavoronkov and colleagues reported in their paper.
The selected candidate, ISM5411, was nominated for the IBD program after showing gut-restrictive pharmacokinetic (PK) profiles and favorable safety in preclinical studies. ISM5411 also demonstrated what Insilico characterized as significant anti-colitis activity at lower doses by restoring intestinal barrier function and reducing gut inflammation in multiple “clinically relevant” murine colitis models, without the systemic side effects typically associated with PHD inhibition.
Ultimate molecular profile
“We thought about the ultimate molecular profile for a perfect PHD inhibitor and also looked at all the previous attempts and possible reasons for failure. One of the reasons for prior failures may be systemic toxicity,” Zhavoronkov reasoned.
Current PHD inhibitors in development, such as 2-oxoglutarate (2-OG) analogs and Fe (II) chelators, lack specificity, increasing their likelihood of off-target side effects.
This challenge surfaced in 2022, when Gossamer Bio acknowledged the failure of the Phase II SHIFT-UC trial (NCT04556383) assessing its Fe (II) chelating PHD inhibitor GB004 in adults with active ulcerative colitis. The trial missed its primary endpoint, the proportion of participants with clinical remission at week 12. Gossamer later ended clinical development of GB004, a drug designed to treat ulcerative colitis by stabilizing HIF-1α. GB004 was licensed from Aerpio Pharmaceuticals (since merged with Aadi Bioscience), which had initially developed the drug as AKB-4924.
“Chemistry42 reduces the potential for side effects in PHD inhibitors by addressing limitations of earlier inhibitors like GB004, which lack specificity and have systemic exposure risks,” Zhavoronkov said. “On-target toxicities of PHD inhibitors are well-documented and can be monitored through established biomarkers such as erythropoietin (EPO) and vascular endothelial growth factor (VEGF).
To address this challenge, Insilico sought properties for a potent PHD inhibitor that exhibited gut-restricted PK properties, excellent activity, and specificity against PHDs.
“Novel and synthetically accessible compounds were developed, demonstrating strong hydroxylase inhibitory activity against PHD2, with a half-maximal inhibitory concentration (IC50) of 4 nM. Structure-activity relationship (SAR) optimization refined the properties of candidate compounds. Alchemistry was used to estimate binding free energies of PHD2-ligand complexes, while ADMET profiling predicted solubility, permeability, and clearance,” Zhavoronkov recalled. “These assessments prioritized low solubility and high clearance to minimize systemic exposure while maintaining moderate permeability for effective cell penetration, ultimately yielding ISM5411.”
Binding, targeting, distribution
Zhavoronkov said ISM5411 displayed:
- A distinct binding mechanism from 2-OG by inducing a conformational change in PHD2 to form an expanded inner sub-pocket—a mechanism that according to Insilico enhances specificity and reduces off-target binding. Researchers then evaluated the drug’s barrier-protective functions and gut-restricted distribution.
- Selective targeting of PHD isoforms: ISM5411 was designed to interact specifically with PHD isoforms, particularly PHD1 and PHD2, without significantly affecting other Fe(II)- and 2-OG-dependent enzymes, as shown by its lack of inhibition of non-PHD targets even at concentrations up to 10 μM.
- Gut-restricted distribution shown to reduce systemic exposure: ISM5411 was engineered with properties that favor low solubility and high clearance, ensuring its distribution remained largely confined to the gastrointestinal tract. Insilico cited PK studies in mice and rats showing a 67-fold enrichment in the colon relative to plasma, minimizing systemic exposure and associated risks like cardiovascular and tumorigenic effects.
- Localized efficacy: The gut-restricted distribution ensured that ISM5411 acts primarily at the site of inflammation, where it stabilized HIF-1α to promote barrier repair and reduce inflammation without triggering widespread physiological effects.
“The overarching generative chemistry techniques used are tunable and translatable to other molecular design tasks that may be of immediate value to the biomedical research community,” Zhavoronkov and colleagues concluded.
12-month process
Insilico said the process of designing ISM5411—from launching Chemistry42 studies to nominating a preclinical candidate—took just 12 months, during which approximately 115 molecules were synthesized and screened.
By contrast, Insilico has said it needed 18 months to bring its lead pipeline candidate, the TNIK inhibitor ISM001-055 (formerly ISM018_055), a treatment for idiopathic pulmonary fibrosis (IPF), from target discovery to preclinical candidate nomination, and took another 12 months to advance the drug into Phase I. The company plans to enter ‘055 into a potentially pivotal study after generating positive topline results in a Phase IIa trial showing encouraging clinical efficacy.
Last year, ‘055 became the first AI-discovered drug candidate to receive the FDA’s Orphan Drug Designation. Insilico detailed the development of ISM001-055, from AI alogrithms to Phase II trials, in a paper also published in Nature Biotechnology.
“Our average time to preclinical candidate is now around 13 months,” Zhavoronkov said.
Zhavoronkov said Insilico’s fastest start to preclinical candidate record was under nine months with the company’s small molecule glutaminyl-peptide cyclotransferase-like protein (QPCTL) inhibitor program, which the company is developing as a cancer immunotherapy for cold tumors. The program—partnered with Fosun Pharma—is now progressing through Phase I.
ISM541 is under study in two Phase I trials—one in Australia (NCT06012578) and the other conducted in parallel in China. Both studies were designed to evaluate the safety, tolerability, pharmacokinetics, and food effects of ISM5411 at increasing oral doses in healthy participants.
“Both Australia and China studies have been completed,” Zhavoronkov said. “We expect to publish the topline data soon—hopefully, in a few weeks. We are not a public company, so we may be able to time the announcement around a major conference.”
Phase II plans and sites for ISM5411 have not been confirmed, “but are being actively discussed as the results of the Phase I are processed and analyzed,” he added.
Insilico plans to further evaluate ISM5411 in larger populations through global multi-center proof-of-concept efficacy studies in patients with ulcerative colitis.
ISM5411 is one of 31 programs within Insilico’s pipeline. The company has nominated 21 preclinical candidates for development since 2021, two years after being able to start its own drug discovery efforts after raising its first significant financing of $37 million. Insilico has since progressed 10 to clinical stage following Investigational New Drug (IND) clearance.
“The reason for the decreasing timelines is improved Chemistry42 capabilities, streamlined workflows, a focused design strategy tailored to the specific therapeutic targets, efficient preclinical validation, an experienced team, and many more factors,” Zhavoronkov added.
