AMT-130 four-year follow-up: functional signal, primary endpoint not significant

September 29 result / updated October 4

Four-year results have now been disclosed. The prospective sections retained below describe what was known earlier, not an outstanding September event. The prespecified 48-month cUHDRS comparison was not statistically significant; the functional TFC signal is nominally significant. This is a mixed readout, distinct from the September 2 application submission and from regulatory acceptance or approval. [6]

Population, biology and delivery

AMT-130 uses gene silencing delivered directly to the striatum. Its intended disease-modifying mechanism does not remove the need for a credible untreated comparator. NCT04120493 and NCT05243017 originally included US sham-controlled and European open-label cohorts; the long-term estimates below instead use matched natural-history controls. [6]

At the June 30, 2026 cutoff, the first two cohorts comprised 17 high-dose and 12 low-dose recipients. Fifteen high-dose participants contributed at 36 months, twelve at 48 months; each low-dose timepoint had twelve. The updated high-dose external dataset contains 1,337 records and 53% missingness at 48 months. These are not randomized arms or interchangeable denominators.

Absolute results and version reconciliation

The table preserves the updated external-control analysis. Earlier 2025 values remain below and should not be overwritten: the population and comparator version changed. The source-reported 36-month cUHDRS contrast of 1.12 differs slightly from subtraction of rounded means; no correction is invented. Confidence intervals and complete missing-data sensitivity models were not supplied in the announcement. [6]

Statistical interpretation / external-control sensitivity

Using the prior comparator post hoc produced nominal p=.041 for cUHDRS and p=.001 for TFC at 48 months. It is a sensitivity analysis, not a successful replacement of the prespecified primary test. The sponsor attributes part of the attenuation to survivor bias in the updated controls; this explanation is plausible but is not independently established by aggregate summaries. [6]

Analysis: propensity matching addresses measured baseline variables, not unmeasured selection, differential assessments or informative loss to follow-up. A larger external sample can narrow model-based uncertainty while leaving causal bias intact. A useful confirmatory comparison must prespecify the estimand, assessment schedule, handling of deaths and missing visits, and interpretation of correlated clinical scales.

The separate high-versus-low-dose paragraph reports different raw changes (cUHDRS −0.91/−1.94; TFC −0.30/−0.70). It answers a different comparison and is not combined with the external-control estimates. Without randomized dose allocation and a formal interaction analysis, this pattern is supportive rather than proof of dose response.

Safety / inflammation and late follow-up

The release reports five high-dose participants with resolved treatment-related CNS inflammatory SAEs, but labels this 17% while also stating 17 high-dose participants in the first two cohorts. The safety denominator is therefore unresolved; neither 5/17 nor 5/29 is silently substituted. One low-dose suicide occurred about five years after treatment and was judged unrelated by the investigator. Attribution does not mean the death should be omitted. [6]

Total TEAEs, all-cause SAEs, discontinuations and exposure-adjusted rates by dose were not provided. A neurosurgical intervention is not reversible like stopping an oral drug. Procedure effects, immunosuppression effects and delayed vector-related toxicity need distinct surveillance; historical statements about no new related events cannot replace a cumulative safety table.

Capital and execution

The July financial disclosure reported $810.3 million of cash, equivalents and current investments at June 30, with management projecting runway into 2030. This is a dated liquidity snapshot, not a current fully diluted capitalization. The new results were not part of the originally submitted BLA according to the September release. [6] [7]

Analysis: available funding supports follow-up and confirmatory execution but cannot validate the comparator. The decision-relevant next evidence is a transparent randomized confirmatory design and regulator assessment of the totality of data. The favorable TFC pattern should be presented alongside, not instead of, the nonsignificant four-year primary endpoint.

Biology / Target

AMT-130 is a one-time gene-therapy approach intended to reduce huntingtin expression. Its clinical claim concerns slowing decline in Huntington’s disease, not reversing every established deficit or curing the genetic disorder. [1] [3]

In a progressive disease, stabilization may be useful even without an absolute score increase. But the counterfactual—what the same patient would have experienced untreated—must be credible. Biological rationale alone cannot establish that counterfactual.

Concordant functional, motor and biomarker directions can strengthen an interpretation, but correlated endpoints are not independent replications. A composite scale and its components may share information, so several favorable p-values should not be counted as several unrelated experiments.

Molecule / Delivery and PK–PD

The program uses direct neurosurgical delivery into the striatum. The US study included a sham-controlled core period, but the widely cited 36-month comparison uses external natural-history controls. Those are different comparison designs. [1] [3]

For a locally delivered gene therapy, dose, distribution and viable target tissue can matter as much as administered vector quantity. CSF biomarker behavior is supportive but not a direct spatial map of target suppression throughout the relevant brain structures.

Higher-dose and lower-dose cohorts may differ in timing, procedures and eligibility. A stronger high-dose result is compatible with a dose effect, but it is not a clean randomized dose-response estimate unless these differences and uncertainty are addressed.

Clinical evidence: actual denominators and comparator

The 2025 analysis included 29 treated patients: 17 high-dose and 12 low-dose. Twelve in each dose group had reached Month 36. The matched external-control datasets comprised 940 and 626 respectively; these are not randomized controls. [1]

A large external dataset cannot compensate for a small treated cohort’s susceptibility to selection and individual trajectories. Precision conditional on a matching model is different from protection against unmeasured confounding.

Four-year follow-up / Statistical interpretation

The July 2026 update guided September four-year data for 24 patients, 12 per dose. It also described FDA discussions on a confirmatory study and a potential accelerated-approval submission. These are company-reported regulatory discussions, not an approval decision. [2]

The four-year question is whether the separation persists under the same analysis rules. Compare the same participants over time; adding a different subset can alter an average even when no individual trajectory changes. Report losses, deaths, missed visits and analysis exclusions explicitly.

Propensity matching balances measured variables, not unknown prognostic factors. Check age, disease burden, baseline function, site assessments, calendar era and eligibility overlap. Sensitivity to alternative external datasets is helpful, but cannot reproduce the protection of random assignment.

An apparent percentage slowing can become unstable when the comparator change is small. Absolute changes, confidence intervals and individual trajectories should therefore lead the interpretation. No four-year efficacy estimate is invented before verified disclosure.

Safety / Procedure and delayed effects

The 2025 release reported no new drug-related serious events since December 2022. “No new” over a specified interval does not mean none occurred in the entire program. [1]

Separate surgical complications, immune suppression, vector-related inflammation and background disease events. A persistent intervention needs long-term surveillance even after an initially manageable safety profile. Missing total SAE or death tables should remain missing, not be converted into zero.

Capital structure / Management

The operational dependency is a credible confirmatory study alongside long-term follow-up. A proposed regulatory route can require additional enrollment and manufacturing work; its feasibility should be assessed separately from an encouraging early efficacy pattern. [2]

Capitalization is not independently reconciled here. Management evidence should focus on the consistency of regulatory communications, whether complete data are published and whether the confirmatory design can genuinely resolve the external-control limitations.

September 16 update: submission occurred; four-year analysis remains separate

The earlier submission guidance above is preserved. uniQure announced a US BLA and UK MAA submission on September 2, supported by the three-year analysis. Priority review was requested, not reported as granted. The separate submission History report is linked in the references. This existing record continues to track the four-year results: the submission announcement still guided that analysis for the third quarter and did not disclose it. [4] [5]

Sources / References

1. uniQure · Phase I/II topline results · September 24, 2025
2. uniQure · Q2 2026 update · SEC Exhibit 99.1, July 29, 2026
3. uniQure · AMT-130 clinical trial design
4. uniQure · BLA and UK MAA submission announcement · September 2, 2026; verified September 16
5. Separate September 2 submission event and clinical evidence
6. uniQure · Four-year disclosure · September 29, 2026
7. uniQure · Q2 financial and program update · July 29, 2026