The most quietly devastating sentence in the proceedings is not about a death or a dose. It is about aim. A phase 2, multicentre, sham-surgery-controlled trial delivered nerve growth factor into the cholinergic basal forebrain of people with Alzheimer's using an adeno-associated virus serotype 2 vector. It showed no cognitive benefit. Then the investigators looked at where the vector had actually gone, and found that more than 80 percent of the AAV2-NGF injections had missed their intended targets. A therapy cannot fail a brain region it never reached. That finding, along with a survey of every other genetic therapy programme in dementia its members could account for, was put before the first EU-US CTAD Task Force on genetic therapies in Alzheimer's disease, convened at the Clinical Trials in Alzheimer's Disease meeting in October 2024 and published on 9 July 2025 by Jakabek, Isaacs, De Strooper, Tuszynski, Lane, Uspenskaya, McDade, Rafii and Mummery, a group drawn from University College London, VIB-KU Leuven, UC San Diego, Ionis, Prevail Therapeutics (a Lilly subsidiary), Washington University in St. Louis and the University of Southern California.
They met in a field with a body count behind it. A companion editorial published in the same journal on 31 July 2025 sets out the background plainly: deaths associated with gene therapy prompted the US Food and Drug Administration to hold a special public meeting in 2020, and since then fatalities have been reported following treatment with the spinal muscular atrophy gene therapy Zolgensma, and in clinical trials in Duchenne muscular dystrophy, Rett syndrome and myotubular myopathy. Those events, the editorial notes, appear associated with high-dose intravenous administration of AAV vectors. Dementia programmes have largely not gone that route, for the blunt reason that they cannot: getting these agents to the brain still means intrathecal or intracisternal delivery, or a needle placed in the tissue itself. The hope raised is that engineered capsids, the redesigned protein shells around the viral genome, might eventually reduce liver damage and adverse immune responses. That is a hope, not a result.
The clearest window into what happens when an AAV does reach the human brain came from frontotemporal dementia. In people with GRN mutations, progranulin falls through haploinsufficiency, so the logic of replacing it is about as clean as neurology gets. Preliminary results from an intracisternal AAV9 programme, presented to the task force and described in the proceedings as preliminary, showed a dose-dependent increase in progranulin in all participants, sustained at 12-month follow-up, with the intracisternal administration procedure well tolerated. But cerebrospinal fluid pleocytosis, an influx of inflammatory cells, was seen in most patients. It was asymptomatic, with one exception: a participant with symptomatic sensorineural hearing loss, ongoing but recovering at the time of the meeting. The pleocytosis was expected and attributed to transgene over-expression in the dorsal root ganglia, and the inflammatory response appeared responsive to corticosteroids. Development is continuing, with eligibility widened to earlier symptomatic disease (0.5 to 1 on the CDR plus NACC FTLD global score), alongside other registered approaches using intrathalamic AAV9 delivery (NCT06064890) and intracisternal AAV1 (NCT04747431).
Dorsal root ganglia toxicity is the recurring tax on this modality. The proceedings single it out as the concern of particular note for central nervous system gene therapies, because it is dose-limiting: it is thought to arise from transgene over-expression and the inflammatory response that follows, which means the more you give, the more you provoke. Human studies suggest responsiveness to immunosuppression; animal studies suggest less. For a family weighing whether to enrol, the practical translation is uncomfortable but honest. The ceiling on how much therapy can be given is set not by the disease but by the delivery system, and the trade is between enough gene product to matter and enough to inflame.
The response to the missed-target problem has been engineering rather than abandonment. New MRI-guided infusion methods were developed after the AAV2-NGF failure, and they are being used in a current trial of AAV2-BDNF in mild Alzheimer's and mild cognitive impairment (NCT05040217), in which six patients with mild Alzheimer's had been safely treated at the time of the report, with enrolment now extending to mild cognitive impairment. Six people is six people. The supporting evidence for brain-derived neurotrophic factor remains animal work: improved memory and rebuilt synapses in APP-overexpressing mice, and improved memory, prevented cell death and activated cell function in aged and lesioned rats and non-human primates. None of that has yet been shown to change the course of anyone's illness.
What the task force did beyond the vector arithmetic matters for readers more than the vector arithmetic does. It looked at who is being left out and who gets consulted. In autosomal dominant Alzheimer's, PSEN mutations are predominantly heterozygous, so work is under way to design antisense targets that silence the pathological allele while leaving the healthy one expressing; in PSEN1 L435F knock-in mice, introducing wild-type human PSEN1 by AAV restored gamma-secretase activity, a mouse result and nothing more. The proceedings also flag that people with Down syndrome, among those at highest genetic risk of Alzheimer's, were largely ignored until recently, carry a higher proportion of cerebral amyloid angiopathy, and are not yet being prescribed anti-amyloid antibodies because of concern about amyloid-related imaging abnormalities. And the group put the involvement of patients in trial design on its list of unique challenges, next to adverse events, monitoring and timing of therapy, which is an admission that the last twenty years of this work did not do that well enough.