On August 5, 2026, STAT reported that a boy had died in a Chinese gene-editing trial nearly a year earlier. The trial belonged to HuidaGene, a Shanghai company, and tested HG302, a CRISPR-based therapy meant to restore dystrophin in Duchenne muscular dystrophy, the progressive and ultimately fatal muscle disease of childhood. The boy was the fourth and final participant, and he received the highest dose in the study. Two weeks after the disclosure, STAT asked researchers in gene editing and muscle disease to review the company's presentation at the American Society for Gene and Cell Therapy's Presidential Symposium, and their verdicts divided, often sharply.
That division is worth examining, because it is really a disagreement about what a trial is allowed to count. A clinical trial is a measuring instrument. The choices in its design decide what it can see. In this trial, the things that mattered most, the dose, the immune response, the death itself, were the things the measure never held.
Trace dystrophin was the reason to go higher
At the symposium in May 2025, roughly three months before the boy died, HuidaGene's then chief executive, Alvin Luk, presented early data from the first two patients in the study. Both had received the lower dose. The central readout was dystrophin, the muscle protein that Duchenne destroys, and the assay found only trace amounts of it.
Dongsheng Duan, who has spent a decade studying gene editing for Duchenne at the University of Missouri, watched the presentation and was blunt. "I'm not impressed," he said in an email to STAT.
What happened next is the part worth slowing down on. The company's response to a weak signal was not to stop, but to escalate. The measured thing, dystrophin, appeared at the bottom of the ladder in trace quantity. The plan was to climb anyway.
The riskiest rung was the one the fourth patient would take
The study, registered on ClinicalTrials.gov as NCT06594094, was an open-label, dose-escalation trial, the first time HG302 had been given to a human being. It was run at Shanghai Children's Medical Center, part of Shanghai Jiao Tong University School of Medicine, beginning in late 2024. The plan called for about six ambulatory boys between the ages of four and eight. Four were enrolled.
The fourth was the last. He received the highest dose, on the order of 100 trillion viral genomes per kilogram, given intravenously in August 2025. He developed severe activation of the complement system and inflammatory cytokines, then acute respiratory distress syndrome, and died. HuidaGene attributed the death to that immune reaction, and said the other three participants did not experience the same syndrome and remain in long-term follow-up. The study was terminated.
Notice the structure. With four patients and four rungs, the top dose is necessarily reached by the last child enrolled. The signal the design was most afraid of could only appear in the final subject. That is not bad luck; it is the arithmetic of the ladder. A trial that escalates to its ceiling in its fourth and final patient has no room left for the lesson the ceiling teaches.
The dose the company had said it meant to avoid
The company's own words make the gap sharper. Before enrollment, HuidaGene had said its editing system was efficient enough to allow dosing below what other AAV-based Duchenne programs used, and that keeping the dose down was precisely how it intended to limit immune risk. The trial's top dose ended up in the range researchers treat as a warning zone. As researchers in the field describe it, doses around 10 trillion viral genomes per kilogram are the common working zone, and 100 trillion is a level that warrants caution.
Luk has since told STAT the dose was too high and was known to carry serious side-effect risk, while saying the decision was not his alone. He left the company in the summer of 2025; the chief technology officer, TJ Cradick, also departed.
The uncomfortable point is that the trial's own measure encouraged the climb. The readout that justified the next rung, dystrophin, said nothing about the danger of the rung above it. The immune risk was never part of the escalation's math, because the design had no way to see it except by arriving there.
A registry that recorded the trial's end but not its loss
Then came the silence. The death occurred in August 2025. The company did not announce it. In February 2026, the ClinicalTrials.gov record was quietly updated to completed. At the end of July 2026, five days before the disclosure, the record was changed again, to active but not recruiting, with follow-up extended from 26 to 104 weeks. No results were posted, no safety information, nothing about the death. The study, according to STAT's earlier investigation, had no data monitoring committee.
HuidaGene says it reported the serious adverse event through hospital ethics and regulatory channels within required timeframes, and that a complete investigation, including laboratory, immunological, pathological, and autopsy findings, was submitted for peer review in January 2026, with details to be released when the paper is published.
Both things can be true at once. The company may have followed every channel it was given, and the record that the world could see still contained no death for eleven months. The registry was the measure that everyone outside the hospital was allowed to read, and it measured the trial's status, never its loss.
The same reaction had already killed an adult with this disease
The immune catastrophe was not a surprise in the field. In October 2022, Terry Horgan, a 27-year-old man with Duchenne, died eight days after receiving a personalized CRISPR-based therapy delivered by a high dose of AAV9, about 100 trillion viral genomes per kilogram, the same order of magnitude HuidaGene later used. His autopsy, published in the New England Journal of Medicine, concluded that the high vector dose had triggered a severe innate immune reaction that led to capillary leak and acute respiratory distress syndrome.
Horgan's case was public and widely discussed. The risk of high-dose AAV was in the literature before this trial chose its top dose. The point is not that HuidaGene ignored it. The point is that the risk lived outside the trial's measure, in journals and case reports, while the trial's own instruments, the dystrophin assays, the registry, the absence of a monitoring committee, had no slot for it until it arrived as a dead fourth patient.
The other child whose death stayed out of the published data
Two weeks before the HuidaGene disclosure, Science and Retraction Watch reported on a second child. A six-year-old girl with a rare neurodevelopmental disorder died in March 2025, about a week after receiving brain-directed base editing at Xinhua Hospital in Shanghai, also under the investigator-initiated research pathway. The team that treated her later published its results without disclosing the death. Chinese authorities and the university's medical school opened investigations.
The two cases are different diseases, different hospitals, different edits. They share a structure. In both, the outcome that mattered most never entered the record where the field looks: in one, the paper; in the other, the registry. The measure was kept clean, and the signal was the price.
A pathway that hands the safety question to one committee
Both trials ran under China's investigator-initiated research pathway, which requires the approval of a hospital ethics committee rather than the full national regulatory review a company-sponsored drug trial would face. The pathway exists to let researchers test ideas quickly, and it has produced genuinely useful science. But it places the judgment about risk, dose, and disclosure with the institution running the study.
Kiran Musunuru, a gene-editing researcher at the University of Pennsylvania, told STAT the opacity was the deepest problem. "There could be many similar events happening, and we would have no idea," he said, adding that HuidaGene should have released its results as a preprint long before.
The exclusion here is systemic. The pathway measures the researcher's judgment, and nothing checks whether that judgment is sound until after the fact, if ever. In the HuidaGene case, the check that finally forced the death into the open was not a regulator, a committee, or a monitoring board. It was reporters.
The strongest defense of the trial is the disease itself
It is worth stating the case for the trial as its supporters would. Duchenne is unrelenting: boys lose the ability to walk, then to breathe, and most die young. For a family with a child whose mutation no approved therapy covers, a first-in-human trial may be the only hope on offer, and accepting serious risk is not irrational. Dose escalation is standard practice in oncology and gene therapy alike. The company says it followed the reporting rules that applied to it. The other three boys are alive and being followed. The full record is, by the company's account, in peer review.
Among the researchers STAT consulted, some said they were not sure it had been unreasonable to try; others were openly dismayed at what they called the quality of the company's data. This analysis takes no position on whether the trial should have been run. Even granting every defense, the design question remains: a trial that reaches its riskiest dose in its final patient, whose record showed no death for a year, whose published data are still pending, cannot tell the next investigator what it paid so dearly to learn.
What the record owes the boy
The boy is not named in public reporting, and the company says the details will come with the peer-reviewed paper. When they do, the record should include everything the measure left out: the dose rationale, the timeline from infusion to death, the autopsy, the immune findings, and the fact of the death itself, in the registry, in the data, in plain sight. Terry Horgan's autopsy was published because his family wanted it to be, and the field learned from it. The same lesson, at the same dose, in a smaller body, should not have to be pried loose from a registry update.
Families considering a gene therapy trial can ask the questions the record did not answer: how many patients at each dose level, what would stop an escalation, who reviews the safety data as it arrives, and how a serious event would reach the public. The answers, or the silences, are the real data.
Primary sources
- STAT's Jason Mast for the expert reactions to HuidaGene's ASGCT presentation, including Dongsheng Duan's assessment and Kiran Musunuru's transparency concerns, and his earlier investigation documenting the death, the disclosure timeline, the registry amendments, the absence of a data monitoring committee, and Alvin Luk's statements about the dose.
- The ClinicalTrials.gov record for NCT06594094 for the enrollment and study-design details.
- The New England Journal of Medicine case report by Lek and colleagues for the details of Terry Horgan's death and the autopsy conclusion attributing it to an innate immune reaction to a high dose of AAV9.
- Science's report for the details of the death of the six-year-old girl and the published paper that did not disclose it, and additional context from Bloomberg, Ming Pao, and SBS News on the disclosure timeline.