Govorestat in CMT-SORD: the fifth aldose-reductase shot at a SORD-driven neuropathy gets shelved before enrollment

What was tried

Applied Therapeutics opened NCT07191912 as a 36-month randomized, double-blind, placebo-controlled Phase 3 confirmatory study of govorestat (AT-007) in adults aged 16 to 65 with Charcot-Marie-Tooth disease caused by biallelic SORD variants (CMT-SORD). The protocol used a quadruple-masked allocation and a primary endpoint of change in the CMT Health Index from baseline through end of treatment. The dosing arm administered govorestat as a 200 mg/mL oral suspension, weight-adjusted, once daily for up to 36 months, against a matched-suspension placebo. The study was registered with an estimated primary completion date of October 2028. It was withdrawn before a single patient was enrolled.

The biological hypothesis

CMT-SORD is caused by loss-of-function mutations in sorbitol dehydrogenase (SORD), the enzyme that converts sorbitol to fructose. Patients accumulate intracellular sorbitol because the polyol-pathway exit is blocked. The therapeutic hypothesis is upstream: inhibit aldose reductase (AKR1B1), the enzyme that produces sorbitol from glucose, and the toxic substrate pool should fall. Govorestat is a small-molecule AKR1B1 inhibitor with documented oral bioavailability and CNS penetration. Open Targets places SORD in the strongest evidence band for autosomal recessive distal hereditary motor neuronopathy 8 (overall association 0.77, genetic association 0.98, genetic literature 0.84), but the animal-model channel sits at 0.38 and the literature channel at 0.08, a profile consistent with a recently characterized human disease that has been validated genetically but not yet through deep mechanistic rodent work [Open Targets v25, ENSG00000140263 x MONDO_0030055].

The pharmacology has a four-decade history. Sorbinil, tolrestat, epalrestat, and caficrestat all reached late-stage clinical trials targeting AKR1B1 in diabetic complications. Only epalrestat is marketed, and only in Japan. The class has been repeatedly approached and repeatedly bounced for combinations of liver signal, weak placebo-adjusted effect, and registrational endpoint disputes [Open Targets v25 known drugs for AKR1B1, n=4]. CMT-SORD is a more tractable biological setting because the disease drives a single dominant metabolic abnormality, but the drug is still the fifth molecule in a long lineage.

What actually happened

The withdrawal language is unusually clean for ClinicalTrials.gov: "the sponsor decided not to proceed with study initiation." No protocol amendment. No safety event. No DSMB letter. The decision sits inside a wider sponsor context. Applied Therapeutics had been pursuing govorestat in classic galactosemia under a separate NDA path; that program drew an FDA Complete Response Letter and forced the company into a restructuring that materially reduced both pipeline breadth and clinical operations capacity through 2025. The CMT-SORD Phase 2/3 program, AT-007-1041, completed and reported partial functional improvement but did not establish a clean, registrable primary endpoint that the agency had committed to accept. Initiating a 36-month confirmatory Phase 3 with no agreed endpoint and a constrained balance sheet is the kind of decision that gets made on a board call rather than in the clinic.

Failure mechanism, best guess

This is best classified as strategic_reprioritization with an underlying biomarker_failure shadow. Two factors compound. The first is the endpoint problem: CMT-HI is a patient-reported composite that captures functional decline only over very long intervals. Powering a registrational study on CMT-HI in a slowly progressive ultra-rare disease forces either a large sample or a long follow-up; the protocol chose long follow-up. The second is the target-disease relationship. AKR1B1 is upstream of SORD in the polyol pathway and is not itself genetically associated with the disease. The drug attempts a metabolic flux re-routing to reduce sorbitol burden in cells whose proximal lesion is a different enzyme. That is mechanistically reasonable, but it forces sponsors to demonstrate proximal pharmacodynamic engagement (sorbitol reduction in plasma or skin) plus a distal clinical benefit, and the field still lacks a consensus PD-to-outcome bridge for SORD-driven neuropathy.

How to prevent this next time

A program of this design should have committed two things to writing before opening a Phase 3.

First, an FDA-aligned Bayesian stopping framework. With a slowly progressive ultra-rare neuropathy and a soft primary endpoint, the right posture is to compute, at each pre-specified interim, the posterior probability of confirmatory success given the data observed:

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

A PP threshold of 0.30 for continuation and 0.10 for futility, anchored to a prior built from AT-007-1041 effect estimates with a Wilson interval on placebo decline, would have given the sponsor a defensible exit point well before enrollment instead of after.

Second, an explicit power calculation against a sorbitol-reduction biomarker enrichment strategy. If the Phase 2 PD readout shows that 60 percent of patients reach a target plasma sorbitol reduction, restricting Phase 3 enrollment to that subgroup raises detectable effect size at any fixed n, and lets the trial run shorter follow-up. The number needed to screen (NNS) goes up, but so does conditional power.

The single highest leverage change would have been negotiating an accelerated-approval surrogate endpoint with FDA, ideally plasma or muscle sorbitol, before locking the protocol.

What this means for similar programs

Other AKR1B1 programs reading through to non-diabetic genetic indications should treat the Govorestat withdrawal as a soft warning that confirmatory study commitment is the binding constraint, not preclinical pharmacology. Sponsors of SORD-targeted gene therapy (multiple academic programs are advancing AAV approaches) should accelerate, because the small-molecule path now has a visible regulatory and operational scar. The broader class of metabolic-pathway-rerouting drugs in inborn errors of metabolism faces the same endpoint problem: the disease moves on a clinical timescale that does not match commercial drug-development timelines without a validated surrogate.

Open questions

Did the AT-007-1041 trial show responder-level clinical improvement that would have supported a biomarker-enriched re-design rather than full withdrawal? Will Applied Therapeutics out-license the CMT-SORD asset or fold it back into the galactosemia program? Does FDA still consider plasma sorbitol an acceptable accelerated-approval surrogate in any polyol-pathway indication? Will the AAV-SORD academic programs draw FDA-issued natural history requirements that the small-molecule developers should have funded years ago?

Sources

1. • ClinicalTrials.gov record NCT07191912, last updated May 2026. https://clinicaltrials.gov/study/NCT07191912 - Open Targets Platform v25, queried 2026-05-30. Target AKR1B1 (ENSG00000085662) drugAndClinicalCandidates returned 4 prior molecules (caficrestat, sorbinil, epalrestat, tolrestat). https://platform.opentargets.org/target/ENSG00000085662 - Open Targets Platform v25, queried 2026-05-30. SORD (ENSG00000140263) x MONDO_0030055 disease association: overall 0.767, genetic_association 0.977, genetic_literature 0.836, animal_model 0.379, literature 0.077. https://platform.opentargets.org/disease/MONDO_0030055 - ChEMBL v34, molecule CHEMBL4650327 (govorestat), max_phase 3.0. https://www.ebi.ac.uk/chembl/explore/compound/CHEMBL4650327 - Cortese A et al. Genotype and phenotype spectrum of Charcot-Marie-Tooth disease due to mutations in SORD. Brain 2025.- Pareyson D et al. Charcot-Marie-Tooth disease: a review of clinical developments and its management. Expert Review of Neurotherapeutics 2025.- OpenFDA FAERS query for "govorestat" on 2026-05-30 returned no records (investigational, not marketed).