Apazunersen missed its Phase 3 cognitive endpoint in Angelman syndrome while the target hypothesis stayed intact

What was tried

Ultragenyx ran NCT06617429, the Phase 3 Aspire study of apazunersen, also known as GTX-102, in pediatric Angelman syndrome. The design was randomized parallel assignment, quadruple masked. One arm received apazunersen by lumbar puncture, the comparator a sham needle prick on the lower back before crossing to open-label dosing. Actual enrollment was 129.

Eligibility required genetic confirmation of a full maternal UBE3A deletion in the 15q11.2 to q13 region and age 4 through 17 years. The primary endpoint was change from baseline in the Bayley-4 cognitive raw score without caregiver input at Day 338. Secondary endpoints included net response on the Multidomain Responder Index and Bayley-4 receptive communication.

The study ran from 3 December 2024 to a recorded completion of 30 October 2026, with actual primary completion on 10 July 2026. ClinicalTrials.gov was updated on 3 November 2026 with the stopped reason given as lack of efficacy, and no results have been posted. Two companion registrations carry the same stopped reason and update date: NCT07157254, the Phase 2 Aurora study with 55 participants, and NCT06415344, the long-term extension with 66. The program closed as a unit.

The biological hypothesis

UBE3A is expressed from the maternal allele in neurons, while the paternal allele is silenced by the UBE3A-ATS antisense transcript. Loss of maternal UBE3A function causes Angelman syndrome. An antisense oligonucleotide that degrades UBE3A-ATS removes that silencing and lets the intact paternal allele produce UBE3A protein. Every patient carries a usable copy of the gene, which is why the mechanism has drawn three clinical programs.

The genetic case is close to the strongest available. Open Targets, queried on 4 November 2026, gives UBE3A and Angelman syndrome an overall association score of 0.843, with the genetic association datatype at 0.9647, genetic literature at 0.7974, text-mined literature at 0.6266 and animal model evidence at 0.5003. Contributing datasources include ClinVar derived evidence at 0.9657, UniProt variants at 0.9608 and Genomics England at 0.8776, and UBE3A sits in the most loss-of-function constrained bin. The same release lists no drug acting directly on the UBE3A protein, and tractability flags show only a medium-quality small molecule pocket, consistent with an oligonucleotide rather than a small molecule or antibody.

Preclinical support was specific. Milazzo and colleagues reported that antisense treatment rescued UBE3A expression and multiple phenotypes in an Angelman mouse model, and Dindot and colleagues described an antisense therapy directed at a conserved region at the start of the UBE3A-AS transcript.

What actually happened

Ultragenyx announced on 2 September 2026 that Aspire missed its primary endpoint and the key secondary endpoint of net response on the Multidomain Responder Index. No treatment difference, p-value or confidence interval has been released for either, and no endpoint-level data appear in the registry, so no effect size is quoted here. The company stated that safety remained consistent with earlier Phase 1/2 findings and that it would evaluate the future of the program including Aurora. The registry updates of 3 November 2026 answered that for all three studies. An openFDA FAERS query on 4 November 2026 returned no reports for apazunersen or GTX-102, the expected result for an agent never marketed.

Failure mechanism, best guess

The archetype recorded here is efficacy failure, because that is what the sponsor and registry state. The useful question is which of three mechanisms produced it, and the public record does not separate them.

The first is insufficient paternal UBE3A reactivation at tolerated intrathecal doses. The second is adequate reactivation delivered outside a developmental window in which it can still change cognition, which matters in a trial restricted to ages 4 through 17 in the deletion genotype, the most severe form. The third is real reactivation that the Bayley-4 cognitive raw score cannot detect across 48 weeks.

The competing program makes the first candidate uncomfortable as a default. In the TANGELO Phase 1 study, 61 children aged 1 to 12 received the UBE3A-ATS oligonucleotide rugonersen, and the exploratory analysis showed dose-dependent partial normalization of the Angelman-associated EEG delta abnormality, so target engagement by this modality is measurable in humans. The Aspire registry entry lists no pharmacodynamic marker among its outcomes. Without a reactivation or EEG measure from Aspire, a reader cannot tell a drug that failed to reach its target from one that reached it and was measured with the wrong instrument.

How to prevent this next time

No endpoint-level data have been released, so no posterior, power recalculation or subgroup estimate can be computed from this trial, and none is offered. The available levers are qualitative and specific.

The first is a pharmacodynamic gate before the registrational readout. Scalp EEG delta power is an established Angelman biomarker, and Spencer and colleagues built a longitudinal natural history model from 204 recordings in 56 subjects, then used it to compute the sample and effect sizes needed to detect a treatment effect at 80% power. In the matched mouse work, deviations from predicted delta power correlated with Ube3a expression at P less than 0.001. A program carrying that measurement into Phase 3 can separate a dose problem from a measurement problem on the day it reads out.

The second is endpoint feasibility testing in the actual population. Hagenaar and colleagues assessed candidate measures in 28 children with Angelman syndrome against a bar of 70% completion with acceptable data quality in 60% of completers. Eye tracking and air-displacement plethysmography cleared it, while functional near-infrared spectroscopy, indirect calorimetry and bioimpedance did not. Protest behavior and inability to sit still drove early termination, the same behaviors a cognitive assessment must survive.

The third is genotype and age stratification declared in advance. Aspire enrolled only full maternal deletions while Aurora reached ages 1 through 65 and nondeletion genotypes, so the program held the data needed to test a developmental window hypothesis and closed both studies before reporting it.

The single highest leverage change would have been to run the registrational trial with a prespecified EEG delta power pharmacodynamic endpoint alongside the Bayley-4 cognitive score, so that a negative cognitive result would have identified whether the drug reached its target.

What this means for similar programs

Two other UBE3A-ATS oligonucleotides remain in development: rugonersen, CHEMBL5095469, and obudanersen, also known as ION-582, CHEMBL6068612, both at Phase 3 stage. Of the 15 agents Open Targets associates with the disease, these two and apazunersen are the only ones aimed at the causal mechanism. Both survivors inherit the endpoint problem Aspire has demonstrated, and neither inherits a reason to doubt the target.

The Claidex Mechanism Risk Score for UBE3A is 35 of 100, in the yellow band, and the components explain why it is not higher. Genetic deficit contributes only 2.35 of a possible 15 points because the association is 0.843. Phase burden contributes 11.80 of 30 from one Phase 3 failure, saturation 6.77 of 15 from three competing programs, and recency 4.25 of 15 because the failure is days old. A target can be entirely correct and still produce a terminated Phase 3.

Open questions

What paternal UBE3A reactivation and what change in EEG delta power were measured in Aspire, and do they differ from the TANGELO range? Was a prespecified analysis by age at first dose planned? Will Ultragenyx post results rather than leaving the record at a stopped reason? Did the Aurora termination follow its own readout or the Aspire result?

Sources

1. • ClinicalTrials.gov. A Phase 3, Randomized, Double-blind, Sham-controlled Study Investigating the Efficacy and Safety of GTX-102 in Pediatric Subjects With Angelman Syndrome. NCT06617429. Record updated 3 November 2026. https://clinicaltrials.gov/study/NCT06617429 - ClinicalTrials.gov. A Safety and Efficacy Study of GTX-102 in Subjects With Deletion- or Nondeletion-type Angelman Syndrome. NCT07157254. Record updated 3 November 2026. https://clinicaltrials.gov/study/NCT07157254 - ClinicalTrials.gov. Long-term Extension of GTX-102 in Angelman Syndrome. NCT06415344. Record updated 3 November 2026. https://clinicaltrials.gov/study/NCT06415344 - Open Targets Platform. UBE3A (ENSG00000114062) and Angelman syndrome (MONDO_0007113) association evidence. Queried 4 November 2026. https://platform.opentargets.org/evidence/ENSG00000114062/MONDO_0007113 - Hipp JF, Bacino CA, Bird LM, et al. The UBE3A-ATS antisense oligonucleotide rugonersen in children with Angelman syndrome: a phase 1 trial. Nat Med. 2025;31(9):2936-2945. https://- Spencer ER, Shi W, Komorowski RW, et al. Longitudinal EEG model detects antisense oligonucleotide treatment effect and increased UBE3A in Angelman syndrome. Brain Commun. 2022;4(3):fcac106. https://- Milazzo C, Mientjes EJ, Wallaard I, et al. Antisense oligonucleotide treatment rescues UBE3A expression and multiple phenotypes of an Angelman syndrome mouse model. JCI Insight. 2021;6(15):e145991. https://- Dindot SV, Christian S, Murphy WJ, et al. An ASO therapy for Angelman syndrome that targets an evolutionarily conserved region at the start of the UBE3A-AS transcript. Sci Transl Med. 2023;15(688):eabf4077. https://- Hagenaar DA, Bindels-de Heus KGCB, van Gils MM, et al. Outcome measures in Angelman syndrome. J Neurodev Disord. 2024;16(1):6. https://- Yang L, Shu X, Mao S, et al. Genotype-phenotype correlations in Angelman syndrome. Genes (Basel). 2021;12(7):987. https://- Ultragenyx apazunersen Phase 3 announcement of 2 September 2026, as reported by NeurologyLive, 3 September 2026. https://www.neurologylive.com/view/ultragenyx-apazunersen-fails-primary-end-point-phase-3-angelman-trial - Angelman Syndrome Foundation. Ultragenyx releases Phase 3 results. https://angelman.org/news-blog/ultragenyx-releases-phase-3-results/ - openFDA drug event (FAERS) API. Queries for apazunersen and GTX-102 returned no matching reports. Queried 4 November 2026. https://api.fda.gov/drug/event.json.