A Six-Year-Old Died in a Gene-Editing Trial. Nobody Was Told.

Some stories are about whether the science works. This one is about whether the system around the science works, and it is grimmer. A joint investigation by Science and Retraction Watch, published today, documents the case of a six-year-old girl, identified by the pseudonym Mei, who died in late March 2025, seven days after receiving an experimental base-editing therapy at Xinhua Hospital in Shanghai. Her death was never made public, and sixteen months later the public record still says the trial is going fine^[1]. Her condition was never going to kill her.

Mei carried a mutation in a gene called CHD3, a single wrong letter of DNA that causes Snijders Blok-Campeau syndrome, a neurodevelopmental condition diagnosed in only about 237 people worldwide. Her case was mild, and it was not life-threatening. What she received instead was a world first: a base editor, a CRISPR-derived tool that rewrites individual DNA letters, packaged into trillions of AAV9 viruses and infused into her spinal fluid on March 24, 2025^[1]. Within days she developed fever, kidney failure, and plummeting platelets. The cause of death was thrombotic microangiopathy, a known and sometimes fatal reaction to high-dose AAV, the virus family used in several approved gene therapies. The hospital's own ethics board later judged her death "definitely related" to the treatment^[2].

Gene therapy has been here before, and everyone in the field knows the name. In 1999, an eighteen-year-old named Jesse Gelsinger died four days after an adenovirus vector triggered an overwhelming immune response in a University of Pennsylvania trial. He had ornithine transcarbamylase deficiency in a mild, managed form, and he was not dying of it when he volunteered. His death froze the field for the better part of a decade, and the lesson it left, as the Science History Institute's retrospective on the case lays out, was specific: the more survivable the disease, the less toxicity a first-in-human dose is permitted to carry. Mei received a different vector, a different payload, and a quarter century of additional technique. What she shared with Gelsinger is the part that killed them both, a high dose of virus given to a patient who was not in mortal danger.

Every safeguard failed in sequence

The details here matter more than the outrage, so let me be precise. The trial ran under China's investigator-initiated track, a category that requires no review by the national drug regulator. The hospital ethics committee approved it on January 2, 2025, six weeks before the primate toxicology report was even finished. That report, dated February 17, showed moderate to severe liver damage in all four monkeys dosed. Four animals is a small toxicology cohort, which cuts both ways: it is too few to characterize a risk, and four out of four is not a rate anyone gets to average away. The consent form the family signed never mentioned death as a possible outcome, and it stated the sponsor would cover all costs. That was false: Mei's parents paid $860,000, including $130,000 wired to a researcher's personal account, for an unproven therapy that charging patients for is illegal in China^[1].

I keep a mental checklist for first-in-human stories: regulator review, independent ethics approval working from complete data, an honest consent form, no pay-to-participate, mandatory reporting of serious adverse events. This case failed all five, and not subtly. When Stanford bioethicist Hank Greely says "death should always be mentioned in a first-in-human trial," he is describing the floor, not the ceiling^[2]. Steven Gray, a gene therapy researcher at UT Southwestern, was blunter: "This shouldn't have gone to trial"^[1].

So why the hurry? Not the medical record; the calendar. In February 2025, a month before Mei's infusion, a team at Children's Hospital of Philadelphia dosed an infant with a base editor built for his mutation alone, a case written up in the New England Journal of Medicine and carried around the world as a first. Chinese genetics, meanwhile, has spent years trying to be first at something in this field that is not He Jiankui's edited babies, and that pressure runs from funding bodies down to individual labs. None of it excuses anything. It does explain the shape of the failure: the guardrails at Xinhua were not bypassed by sloppiness. They were bypassed by people racing for a headline, and a headline is what they got.

The paper that cleaned it up

The aftermath adds a second scandal on top of the first. The team behind the trial, led by neuroscientist Zilong Qiu, published the preclinical work in Nature this year, and the paper shows mice rescued from the same mutation, with no trace of Mei's treatment, her family's money, or her death^[3]. State broadcaster CCTV praised the research as "the first ray of hope." Nature says it was unaware of the death. Seven outside experts who reviewed the paper at Science's request raised data-integrity concerns, and several believe the problems may warrant retraction. The consequences, meanwhile, read like satire: a fine of roughly $3,600 for the hospital, verbal counseling for the overseeing physician, and no action against Qiu from his university^[1].

Then there is the safety net that was supposed to catch all of this. The trial was registered publicly, which turns out to be worse than if it had never been listed at all. The ClinicalTrials.gov entry was last updated on June 26, 2025, three months after Mei died, and it still presents the study as active and ongoing, with no death, no serious adverse event, and no halt anywhere in it^[4]. A registry entry is not transparency by itself. It is a claim about reality that somebody has to keep true, and this one was actively maintained as a fiction for months after the child it described was dead.

What this does and does not mean

The obvious reading is "gene editing kills," and it is the wrong one. AAV-based therapies have real, verified wins, and the reaction that killed Mei is a known risk that dosing discipline and honest consent exist to manage. The harder reading is about implementation. Joy Zhang, a sociologist at the University of Kent, put it cleanly: the case shows "the gap between what is intended and what has been put in place"^[2]. A regulatory track built to let hospitals try things became a track that let one hospital skip every review that would have said no.

Mei's father, also speaking under a pseudonym, told the investigators: "We did not realize how unusual and dangerous many of the arrangements were"^[1]. That sentence belongs over the door of every ethics committee on earth. The entire apparatus of human-subjects protection exists because families cannot be expected to spot an unusual arrangement. Spotting it is the institution's job. The institution did not do it, the paper trail was laundered into a prestigious journal, and a child is dead. Gene editing will move on from this, and on the evidence of its better days, it should. The record of what happened here has to move with it.