Ataluren and the subgroup that did not travel: a 15-year extension closes without an efficacy endpoint

What was tried

NCT01247207, run by PTC Therapeutics as PTC124-GD-016, was an open-label, single-arm Phase 3 safety study of ataluren in boys with nonsense mutation dystrophinopathy. It enrolled participants who had already received ataluren in an earlier PTC-sponsored trial or treatment plan, along with their siblings, dosed at 10 mg/kg in the morning, 10 mg/kg at midday and 20 mg/kg in the evening. The study opened on 22 November 2010 and closed on 10 February 2026, with 265 male participants and a mean baseline age of 10.9 years (standard deviation 3.99).

The design carried no efficacy endpoint. The sole primary outcome was the number of participants with treatment-emergent adverse events over a period registered as up to approximately 15 years. ClinicalTrials.gov lists the study as terminated for "Sponsor decision", with the record updated on 8 October 2026.

The biological hypothesis

Duchenne muscular dystrophy is caused by loss of dystrophin, encoded by DMD. Open Targets scores the DMD to Duchenne association at 0.869, with a genetic association component of 0.914 and a clinical component of 0.964, and the gene is strongly loss-of-function constrained at an observed-to-expected ratio of 0.191. The genetic case was never the weak link.

The weak link sits one layer down. ChEMBL records ataluren (CHEMBL256997, small molecule, maximum phase 4, first approval 2014) as an 80S ribosome modulator. Ataluren does not bind a DMD gene product. It acts on the translating ribosome to promote readthrough of a premature termination codon, so DMD is the substrate of the hypothesis rather than the drug's binding site. The program inherited the genetic certainty of the disease gene without inheriting any certainty about how much readthrough a given patient achieves, or how much dystrophin restoration translates into preserved ambulation.

What actually happened

Posted results show 265 participants dosed and none completing. Discontinuations were recorded as study termination 111, withdrawal by subject 67, other 32, adverse event 29 and lost to follow-up 16. Ninety of 265 participants had a serious adverse event and 241 of 265 had a non-serious one, most often nasopharyngitis 94, disease progression 59, vomiting 51, diarrhoea 45 and pyrexia 42. Serious cardiac events included cardiac failure in 4 participants, cardiac arrest in 3, cardiomyopathy in 2 and myocarditis in 2. Fourteen deaths were reported across the treatment period. The registry does not attribute cause, and a cohort followed from a mean age of 10.9 years for up to 15 years in a disease with cardiorespiratory mortality is not one in which an all-cause death count can be read as a drug signal. openFDA FAERS holds 29 reports naming ataluren, 11 of them serious and none with a fatal outcome, plus 6 under the Translarna trade name, volumes too small to support disproportionality analysis.

What ended the study was regulatory. The European Commission confirmed the decision not to renew the conditional marketing authorisation for Translarna, and the product stopped being available in France during 2025. PTC withdrew its United States New Drug Application on 12 February 2026, after the FDA indicated the data were unlikely to meet the agency's threshold of substantial evidence of effectiveness. The extension's registered completion date of 10 February 2026 falls two days before that announcement.

Failure mechanism, best guess

This was not a safety failure and not an enrollment failure. It was the closure of a long-running access and safety study whose underlying asset had lost its regulatory basis, which is why the archetype recorded here is sponsor_decision. The informative failure sits upstream.

ACT DMD (NCT01826487) missed its primary endpoint. The least-squares mean change in six-minute walk distance at week 48 was 13.0 m in favour of ataluren, 95% confidence interval -7.4 to 33.4, p = 0.213. A prespecified subgroup with baseline walk distance between 300 and 400 m showed 42.9 m, 95% confidence interval 11.8 to 74.0, p = 0.007, while the subgroup below 300 m showed -7.7 m, p = 0.749.

One observation follows from those two published intervals. The 42.9 m subgroup estimate lies outside the confidence interval of the whole-trial estimate, whose upper bound was 33.4 m. A subgroup effect larger than anything the full trial's interval accommodated is the signature of an estimate shaped partly by sampling noise, and it belonged in the hypothesis column rather than in a registration strategy.

Study 041 then tested a confirmatory population that was not the population the signal came from. In the intention-to-treat set of 359, the rate of walk-distance decline over 72 weeks fell by 21%, p = 0.0248. In the 300 to 400 m subgroup of 169 it fell by 30%, p = 0.0310. In the prespecified population of 185 defined by walk distance of at least 300 m together with a supine-to-stand time of at least 5 seconds, it fell by 9%, p = 0.3626. The population that mattered for registration was the one that did not separate.

How to prevent this next time

Patient-level data and standard errors for Study 041 are not public, so no Bayesian recomputation or power calculation is defensible here. The endpoint-level data still supports four levers.

First, lock the confirmatory enrichment to the exact definition that produced the signal. Study 041's prespecified population was broader and differently constructed than ACT DMD's 300 to 400 m band, and the two gave 9% and 30%.

Second, treat any subgroup point estimate falling outside the whole-trial confidence interval as unreplicated until an independent trial reproduces it in that same stratum.

Third, require a pharmacodynamic bridge. A readthrough mechanism should carry a quantitative dystrophin measure linking exposure to protein restoration to function, so a negative functional result can be diagnosed rather than relitigated by subgroup.

Fourth, attach an explicit stopping rule to open-label extensions at the outset.

The single highest leverage change would have been to run the confirmatory trial in the identical 300 to 400 m stratum that generated the ACT DMD signal, powered on that stratum alone, rather than in a broader prespecified population that diluted it.

What this means for similar programs

The Claidex failure graph now carries DMD at a Mutational Risk Score of 32, in the yellow band. Genetic deficit contributes only 1.97 of a possible 15, because the gene-disease link is about as solid as target validation gets, while saturation contributes 11.97 and phase burden 11.80 against 8 drugs or clinical candidates counted by Open Targets. Risk in this target sits in crowding and late-stage attrition, not in whether the gene matters.

For nonsense suppression specifically, the field has moved. Recent preprints describe an engineered translation release factor 1 that suppresses disease-causing nonsense mutations, and in vivo base editing delivered by myotrophic adeno-associated viruses in dystrophic muscle. Both act on the mutation itself rather than on the probability that a ribosome reads past it, which removes the variability ataluren's clinical record never resolved.

Open questions

Why did the 300 to 400 m stratum replicate in Study 041 while the broader prespecified population did not, and is the difference a progression-rate effect, a measurement-floor effect, or chance in two moderately sized strata?

Is there a dystrophin quantification method precise enough to serve as a go or no-go gate for readthrough compounds?

What follow-on data collection now covers European patients who lost access when the conditional authorisation lapsed?

Sources

1. ClinicalTrials.gov. NCT01247207, PTC124-GD-016, ataluren in nonsense mutation dystrophinopathy, including posted results. https://clinicaltrials.gov/study/NCT01247207 ClinicalTrials.gov. NCT01826487, ACT DMD, Phase 3 ataluren. https://clinicaltrials.gov/study/NCT01826487 McDonald CM, et al. Ataluren in patients with nonsense mutation Duchenne muscular dystrophy (ACT DMD): a multicentre, randomised, double-blind, placebo-controlled, phase 3 trial. Lancet 2017;390:1489-1498.Vlodavets D, et al. Confirmatory long-term efficacy and safety results of ataluren in patients with nmDMD from Study 041. J Comp Eff Res 2025;14(10):e240238.Open Targets Platform. DMD (ENSG00000198947) and Duchenne muscular dystrophy (MONDO_0010679), association and genetic constraint data, accessed 2026-10-10. ChEMBL. Ataluren, CHEMBL256997, mechanism of action 80S ribosome modulator, accessed 2026-10-10. openFDA FAERS. Reports naming ataluren and Translarna, accessed 2026-10-10. European Medicines Agency. Translarna: EMA re-confirms non-renewal of authorisation of Duchenne muscular dystrophy medicine. https://www.ema.europa.eu/en/news/translarna-ema-re-confirms-non-renewal-authorisation-duchenne-muscular-dystrophy-medicine Institut de Myologie. Translarna loses its conditional authorisation in Europe and is no longer available in France in DMD, 1 July 2025. https://www.institut-myologie.org/en/2025/07/01/translarna-loses-its-conditional-authorisation-in-europe-and-is-no-longer-available-in-france-in-dmd/ PTC Therapeutics. Regulatory update on Translarna, 12 February 2026. https://www.prnewswire.com/news-releases/ptc-therapeutics-provides-regulatory-update-on-translarna-302686879.html Preprint. Engineered translation release factor 1 suppresses disease-causing nonsense mutations.Preprint. In vivo base editing via single myotrophic adeno-associated viruses in dystrophic mouse muscle and satellite cells.