Cell paper details how a high dose CRISPR therapy edited Duchenne muscle and killed one boy in 81 hours

Duchenne muscular dystrophy is defined by the absence of dystrophin, the protein whose loss causes muscle to break down under ordinary use, and that gap has made the disease a proving ground for in vivo gene editing. The appeal is that editing targets the cause. Instead of adding a shortened gene, a CRISPR payload given systemically is meant to correct the underlying mutations so that muscle cells resume making their own dystrophin. Adeno-associated virus vectors have carried much of the field's hope because they reach skeletal muscle after an intravenous infusion.

HG302, the candidate at the center of the new publication, is built around that logic. It packs a compact nuclease called hfCas12Max, together with its guide RNA, inside a single AAV9 vector, a design meant to keep the entire editing payload within the cargo limit of one virus particle. HuidaGene Therapeutics, the developer, is based in Shanghai, and the program holds orphan drug designation from the US Food and Drug Administration. The organisations around the program include Shanghai Children's Medical Center and the Chinese Academy of Sciences Shanghai Institute of Materia Medica.

The clinical vehicle was HG302-01, an investigator-initiated first-in-human study registered as NCT06594094. Its design was small and, on paper, cautious. Four ambulatory boys with Duchenne, aged 4 to 8, each received a single intravenous infusion at 5.0x10^13 or 1.0x10^14 vector genomes per kilogram. The protocol had planned for four to six boys, with dosing staggered by at least 12 weeks and then by six weeks, and each dose requiring written advice from a safety review committee.

The study delivered both the proof of concept the field wanted and its worst case. Biopsies showed on-target genome editing and detectable dystrophin, and two boys who completed 12 months of follow-up walked farther and scored higher on a standard motor scale. The fourth participant died. The full account appeared in Cell on October 26, 2026, roughly 14 months after the death and nearly three months after disclosure.

Key Facts

Cell published the paper on October 26, 2026, under the title In Vivo Genome-Editing Therapy for Duchenne Muscular Dystrophy: Preclinical Studies and First-in-Human Trial. It reports on-target editing and detectable dystrophin expression in post-treatment muscle biopsies, the molecular evidence that the payload reached human muscle and cut where intended, and it documents the treatment-related fatal serious adverse event in participant 4, including the clinical course and causality assessment.

At day 91, DNA editing in the first three boys measured between 1.02 percent and 4.31 percent, and dystrophin reached roughly 2.29 percent to 3.83 percent of normal. The two boys with a full year of follow-up gained 8 and 10 points on the North Star Ambulatory Assessment and walked 106 metres and 162 metres farther in six minutes. VCBeat Health reported on October 27, 2026, that the results are the first published clinical data for an in vivo CRISPR-based therapy given systemically for Duchenne, the proof of concept the field had been waiting for.

Nature reported on October 27, 2026, in a piece by Rachel Fieldhouse, that the boy developed sudden severe abdominal pain within 24 hours of dosing, then a fever that responded only partly, followed by discoloured skin from lack of oxygen and a slow, irregular heartbeat before his heart stopped. An autopsy found that he died of acute respiratory distress syndrome related to the treatment. Ian Woodcock, a paediatric neurologist at Murdoch Children's Research Institute in Melbourne, told Nature that the speed of the decline pointed to a reaction to the dose, adding that even though AAV does not cause disease in humans, the amount of foreign material involved really triggers the immune system.

NetEase reported on October 27, 2026, that the complement marker sC5b-9 rose above 1,800 nanograms per millilitre and that interleukin 6 was markedly elevated, leading investigators to attribute the death to strong innate immune activation. On 5 June 2025 the third boy became the first to receive 1.0x10^14 vector genomes per kilogram and developed grade 3 thrombocytopenia, grade 2 oxygen desaturation and complement activation, with sC5b-9 at 1,568 nanograms per millilitre on day 7 and still elevated on day 21, though those findings resolved and were not judged dose limiting. The safety review committee met on 25 July 2025, approved continued enrolment on 28 July, and the fourth boy was dosed on 30 July. His body mass index was 22.93, the highest in the cohort, so weight-based dosing may have given his muscle a higher exposure.

HuidaGene announced the death on 5 August 2026, about a year after it happened in August 2025, and the delay drew criticism. The company said a manuscript on the death and post-mortem findings was submitted in January after what it called a comprehensive investigation. Co-author Hui Yang, a HuidaGene co-founder, said immunosuppressants, animal testing and staggered dosing after safety review were insufficient, and that the most important lesson is that severe systemic immune toxicity after AAV administration remains difficult to predict.

Analysis

What this really means is that the field's safety machinery failed at the precise point where it was supposed to be strongest. The step to a dose twice as high followed a protocol, a waiting period and a formal review, and the boy who died was treated two days after that review cleared the way. That review, as the paper acknowledges, was conducted by the principal investigator, a sub-investigator and the sponsor's medical monitor, with no independent members. A panel drawn from the trial's own staff and the company's own payroll is not an independent check.

The second failure is arithmetic. Dosing by total body weight assumes mass predicts the tissue that receives the vector, and in a child with Duchenne that assumption is weakest where it counts, because a higher body mass index does not mean more lean muscle. The fourth boy's body mass index of 22.93 was the highest in the cohort, which by the authors' hypothesis may have pushed his muscle exposure above that of the third boy, who received the same dose and survived. A dose that is safe in one child and lethal in another at the same label is a dosing rule problem.

The third failure is predictive. The company itself said the findings underscore the limits of preclinical models and existing risk mitigation in anticipating severe immune mediated toxicity after systemic AAV administration. That is a striking admission from a developer whose candidate had passed the usual animal work. The third boy's complement spike, 1,568 nanograms per millilitre on day 7 and still abnormal on day 21, was visible evidence of the mechanism that later killed the fourth boy, and it was read as transient. Whether that reading was defensible is now the central question for anyone else running a high-dose AAV trial.

The bigger picture here is that the Duchenne editing field now has proof of molecular activity and proof of catastrophic risk in the same short paper. Editing rates of 1.02 percent to 4.31 percent and dystrophin at 2.29 percent to 3.83 percent of normal are real signals, not artifacts, but they remain modest in absolute terms, and with no control arm the functional gains in two boys are suggestive rather than conclusive. The safety question is now the binding constraint.

Why It Matters

For patients and families, the risk conversation has changed shape. A single infusion that edits muscle is still the goal, but the trial shows that the immune system, not the editing machinery, sets the near term ceiling. The three surviving boys remain in long-term safety follow-up, and their data will show whether that 1.02 percent to 4.31 percent window can be raised.

For developers, the paper is a warning about how much confidence to place in animal models and in review structures that look rigorous on paper. HuidaGene's own conclusion, that severe systemic immune toxicity after AAV administration remains difficult to predict, applies well beyond Duchenne. Every high-dose systemic AAV program uses similar escalation logic and similar committees, and the same blind spot travels with them.

For regulators and journals, the timing matters as much as the science. A death in August 2025 became public in August 2026, and the full paper arrived in October 2026. That interval is defensible only if the investigation behind it was genuinely independent, and the paper's own description of the review committee suggests otherwise.

Next Up

HuidaGene announced on October 27, 2026, that dosing in HG302-01 is complete and that the three surviving participants remain in long-term safety follow-up. The company has not outlined a next cohort or a revised dose. The paper calls for continued refinement of safety assessment and clinical monitoring for systemically delivered AAV therapies.

The field now faces a narrower question than whether editing can work in muscle. It knows that it can, at least at the low single digit percentages reported here. The question is whether anyone can predict, before an infusion, which child will mount the complement driven response that killed the fourth boy.