Soticlestat in Dravet and Lennox-Gastaut syndrome: a CH24H bet that cleared Phase 2 and missed in Phase 3

What was tried

Takeda ran ENDYMION 2 (NCT05163314), a Phase 3, prospective, open-label, single-group extension that assessed long-term safety and tolerability of soticlestat as adjunctive therapy in Dravet syndrome and Lennox-Gastaut syndrome. Soticlestat (TAK-935, ChEMBL CHEMBL4298172) is an inhibitor of cholesterol 24-hydroxylase, the enzyme encoded by CYP46A1. The study enrolled 352 participants aged 2 to 56 and tracked treatment-emergent adverse events, suicidality, and growth measures as primary safety outcomes, with a primary completion date of 24 September 2025. As an extension protocol, its purpose was to carry responders from the pivotal studies and document long-term tolerability rather than to test efficacy on its own (ClinicalTrials.gov, NCT05163314).

The biological hypothesis

CYP46A1 is the brain-specific enzyme that converts cholesterol to 24S-hydroxycholesterol, the principal route for cholesterol turnover in neurons (Open Targets function annotation, and Brain, 2024). 24S-hydroxycholesterol is a positive allosteric modulator of the NMDA receptor, so inhibiting CYP46A1 was proposed to lower 24S-hydroxycholesterol, dampen glutamatergic overexcitation, and reduce seizures. Preclinical work supported the idea, with soticlestat modifying acute seizure burden and chronic epilepsy phenotypes in mouse models (Neuropharmacology, 2023). The hypothesis was attractive for developmental and epileptic encephalopathies, where standard antiseizure medications leave high residual seizure frequency and where Dravet syndrome in particular carries severe unmet need (Epilepsy Research, 2022). Open Targets scores the CYP46A1 to Dravet syndrome association at 0.0936, weighted toward clinical evidence (0.152) over literature (0.041), with no human genetic association datatype score. The CYP46A1 to Lennox-Gastaut syndrome association is higher at 0.363, again clinically rather than genetically anchored.

What actually happened

The Phase 2 signal did not hold in Phase 3. The ELEKTRA randomized Phase 2 study had reported a reduction in convulsive seizure frequency in Dravet syndrome and drop seizure frequency in Lennox-Gastaut syndrome (Epilepsia, 2022), which justified the pivotal program. The Phase 3 studies then returned negative results. The posted termination reason for the extension states plainly that because of negative results from the Phase 3 SKYLINE and SKYWAY studies, and unrelated to patient safety, supplementary data from the TAK-935-3003 study was no longer necessary, so Takeda closed the study (ClinicalTrials.gov, NCT05163314). The extension that was supposed to house long-term responders lost its reason to exist once the pivotal efficacy readouts failed. OpenFDA FAERS holds 10 soticlestat reports, with drug ineffective the single most frequent term (4 of 10), followed by seizure (3) and intellectual disability (2), alongside 8 serious reports and no death reports. That count is too small for disproportionality and is noted only because the dominant term matches the efficacy outcome.

Failure mechanism, best guess

This was an efficacy failure, and the most likely explanation is that the Phase 2 to Phase 3 transition overcounted the strength of the original signal. Three factors plausibly contributed. First, regression to the mean in smaller early studies inflates effect sizes that then shrink under larger pivotal conditions. Second, the mechanism is indirect. Lowering 24S-hydroxycholesterol is one input into NMDA receptor tone, and in heterogeneous encephalopathies with diverse genetic drivers a single metabolic lever may be insufficient to move a hard seizure endpoint across the whole population. Third, the target lacks human genetic validation for these syndromes. The Open Targets association of 0.0936 for Dravet syndrome rests on clinical and literature inference, not on genetics that tie CYP46A1 variation to seizure burden, which weakens the prior that strong target modulation will translate into clinical benefit. The CYP46A1 biology is also broader than seizures, with recent work implicating the enzyme in alpha-synuclein pathology (PLoS Biology, 2025), a reminder that modulating a central cholesterol node has effects beyond the intended pathway.

How to prevent this next time

Two quantitative tools would have sharpened the Phase 3 decision. First, a Bayesian posterior probability of success should have been computed from the ELEKTRA data before committing to two pivotal studies. The posterior probability of final success integrates the interim or prior evidence over the plausible effect sizes:

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

A prior that discounted the Phase 2 effect for regression to the mean, combined with a historical base-rate adjustment for adjunctive antiseizure medications in developmental epileptic encephalopathies, would have produced a lower posterior than the raw Phase 2 estimate implied, flagging the gap between the observed effect and the effect size the pivotal studies were powered to detect. Second, an explicit power and number-needed-to-treat calculation tied to a biomarker of target engagement would have helped. Measuring plasma 24S-hydroxycholesterol suppression and powering the pivotal studies on the subgroup achieving a prespecified suppression threshold, rather than on the full intent-to-treat population, would have tested whether the mechanism works where it is engaged. The single highest leverage change would have been a pre-specified Bayesian go/no-go on the pivotal studies that explicitly discounted the Phase 2 effect for regression to the mean and required a plasma 24S-hydroxycholesterol engagement threshold for the primary analysis population.

What this means for similar programs

Soticlestat is, by ChEMBL annotation, the only molecule with a recorded mechanism of action on CYP46A1, so this program effectively was the clinical test of the target in epilepsy. The Claidex Mechanism Risk Score for CYP46A1 is 46 of 100 (yellow), with the largest contributions from the Phase 3 burden component (15.74 of 40) and the genetic deficit (13.60 of 15). Because soticlestat is the only molecule with a recorded mechanism of action on CYP46A1, there is no independent program to corroborate or rescue the mechanism. Programs built on indirect metabolic modulation of excitability should treat a single positive Phase 2 as a hypothesis to stress-test, not a green light.

Open questions

Did the Phase 3 SKYLINE and SKYWAY studies show any signal in prespecified subgroups or in participants who achieved strong 24S-hydroxycholesterol suppression? Was the Phase 2 effect size formally reconciled against the Phase 3 powering assumptions before the pivotal studies launched? Does CYP46A1 inhibition retain a role in a genetically defined epilepsy subset, even if it failed across the broad Dravet and Lennox-Gastaut populations?

Sources

1. • ClinicalTrials.gov, NCT05163314, soticlestat extension (ENDYMION 2) in Dravet and Lennox-Gastaut syndrome, study record and termination reason. https://clinicaltrials.gov/study/NCT05163314 - Hahn CD et al. A phase 2, randomized, double-blind, placebo-controlled study of soticlestat (ELEKTRA). Epilepsia, 2022. https://- Nishi T et al. Soticlestat, a novel cholesterol 24-hydroxylase inhibitor, modifies acute seizure burden and chronic epilepsy phenotypes. Neuropharmacology, 2023. https://- Petrov AM, Pikuleva IA. Brain cholesterol and Alzheimer's disease: challenges and opportunities in probe and drug development. Brain, 2024. https://- Strzelec M et al. Dravet syndrome: advances in etiology, clinical presentation, and treatment. Epilepsy Research, 2022. https://- The cholesterol 24-hydroxylase CYP46A1 promotes alpha-synuclein pathology in Parkinson's disease. PLoS Biology, 2025. https://- Open Targets Platform, CYP46A1 (ENSG00000036530) to Dravet syndrome (MONDO_0100135) association, score 0.0936. To Lennox-Gastaut syndrome (MONDO_0016532), score 0.363. https://platform.opentargets.org - ChEMBL, soticlestat (CHEMBL4298172), maximum phase 3. CYP46A1 target (CHEMBL4523510), 1 molecule with annotated mechanism of action. https://www.ebi.ac.uk/chembl - OpenFDA FAERS, drug event query for soticlestat, 10 reports; top reaction terms and seriousness counts. https://api.fda.gov/drug/event.json.