Chemogenetic gene therapy cuts monthly seizures by 92.6 percent in three epilepsy patients

Three adults whose epilepsy had resisted drug treatment for years are now having far fewer seizures after doctors infused a gene therapy vector directly into the tissue that was generating their seizures and then activated it with an oral pill. The intervention, described in The New England Journal of Medicine, is the first clinical use of chemogenetics in the human brain, a technique that until now had been confined to laboratory animals. During the two months used to assess efficacy, the mean monthly seizure count across the three participants fell by 92.6 percent.

Epilepsy is among the most common serious neurological conditions, and medication controls seizures for many people. But about 30 percent of patients cannot achieve adequate seizure control with existing antiseizure drugs, according to Xuanwu Hospital. For those whose mesial temporal lobe epilepsy is driven by hippocampal sclerosis on one side of the brain, the usual next step is surgery to remove or disconnect the affected tissue, an irreversible operation that carries risks to memory and language function.

The new approach offers a different route. Instead of destroying the seizure focus, it tries to dial it down. Surgeons deliver a gene that instructs excitatory neurons in the hippocampus to build an engineered receptor. That receptor is inert on its own and responds only when the patient takes clozapine, an antipsychotic drug acting here as the switch. The receptor, hM4D(Gi), is an inhibitory DREADD, a designer receptor activated exclusively by a designer drug. When clozapine binds it, the targeted neurons become less excitable, and the circuit that generates seizures is dampened rather than cut out.

The work was led by Guoguang Zhao, a neurosurgeon at Xuanwu Hospital of Capital Medical University in Beijing, with Minmin Luo of the Chinese Institute for Brain Research Beijing as co-senior author. The vector, GA002, is an adeno-associated virus carrying the hM4D(Gi) gene under the CaMKIIalpha promoter, a genetic element that restricts expression to excitatory cells. The findings appeared in volume 395, issue 18, of The New England Journal of Medicine, at pages 1867 to 1869, on November 5.

Key Facts

The trial enrolled three adults with drug-resistant mesial temporal lobe epilepsy associated with unilateral hippocampal sclerosis. Nature reported on November 4 that all three had been diagnosed between 10 and 25 years earlier, had developed resistance to at least two antiseizure drugs, and had averaged at least two seizures per month in the three months before enrolling.

Under intraoperative MRI guidance, the team used convection-enhanced delivery to infuse GA002 into the head, body and tail of the hippocampus on the side where seizures originated. Capital Medical University reported on November 5 that imaging confirmed adequate coverage with no reflux of the vector into the ventricles or the subarachnoid space.

Two weeks after surgery, participants began a low-dose oral clozapine regimen every eight hours while continuing their existing antiseizure medication. During the efficacy window from week 5 to week 12 after surgery, mean monthly seizure frequency fell 92.6 percent, from 4.33 to 0.33 seizures per month. Individually, the counts went from 4.5 to 0.5, from 4.0 to zero, and from 4.5 to 0.5.

The New England Journal of Medicine reported on November 5 that the primary endpoint was safety, while seizure frequency, seizure severity, epileptiform discharges, 18F-FDG PET metabolism and quality of life were secondary endpoints. All recorded adverse events were grade 1 or grade 2. One patient had an asymptomatic elevation of liver enzymes that resolved without intervention, and cognitive testing found no clinically meaningful decline.

Follow-up has reached 12 months, 8 months and 8 months for the three participants. The first two remained seizure free. The third had two seizures during an unplanned interruption of clozapine and none after resuming it. Seizure severity scores fell in all three patients, and video EEG recordings showed fewer interictal epileptiform spikes.

Analysis

What this really means is that a neurological circuit can become a controllable drug target after a single surgical delivery, with the control exercised by a pill taken at home. Chemogenetics has been a workhorse of animal neuroscience for years, valued precisely because it lets researchers silence or activate defined cell populations. Moving it into people required clearing three hurdles at once: a viral vector that expresses only in the right cells, a surgical technique that spreads it through a defined volume of tissue, and a ligand safe enough at the doses needed.

The design choices are worth noting. Expression was restricted to excitatory neurons by the CaMKIIalpha promoter, leaving inhibitory neurons untouched, which matters because indiscriminate silencing inside the hippocampus could worsen rather than relieve the imbalance that produces seizures. Convection-enhanced delivery, rather than a simple injection, was used to spread the vector through the head, body and tail of the hippocampus. And the ligand is clozapine, a drug with a long clinical history, given at low dose.

The caveats are substantial and the authors do not hide them. The study had three participants and no control group, because it was designed mainly to test safety. Nature reported on November 4 that Matthew Walker, a neurologist at University College London, called the effect dramatic but cautioned that the study is small and that long-term use of high-dose clozapine, an antipsychotic, has been associated with cardiovascular problems and other negative effects. Science reported on November 6 that Gabriele Lignani, an epilepsy neuroscientist at the same institution, called the results very promising and an important first step, while stressing that further studies in larger groups are needed.

The bigger picture here is that this is a platform result as much as a therapy result. A small, open-label study cannot establish how much of the 92.6 percent reduction is attributable to the gene therapy, to the surgical procedure, to shifts in concurrent medication, or to the placebo response that is well documented in epilepsy trials. What it can do, and does, is show that the mechanism engages in humans, that the delivery is anatomically feasible, and that the safety profile in three people was mild. Those are the questions a first-in-human study is built to answer.

Why It Matters

For patients with unilateral hippocampal sclerosis, the alternative is often resection. A treatment that can be adjusted, paused or stopped changes the risk calculus, particularly for people whose seizure focus sits close to tissue they cannot afford to lose. The third participant's course is instructive: seizures returned when clozapine was interrupted and stopped when it resumed, evidence that the effect depends on continued occupancy of the engineered receptor rather than on a one-time lesion.

Regulatory momentum is already visible. PiQ Markets reported on November 6 that GA002 completed investigational new drug filings in both China and the United States on October 31, 2026, and that a phase 1/2 registration trial led by Xuanwu Hospital is recruiting. Science reported on November 6 that Luo's team has approvals from the US Food and Drug Administration and the Chinese Centre for Drug Evaluation for phase 1 trials, and that the group says it has treated 15 or 16 patients across epilepsy and pain studies, with safety described as very good so far.

The wider significance is that chemogenetics could be pointed at other circuits. Parkinson's disease, chronic pain and psychiatric conditions all involve defined populations of neurons that might be modulated rather than ablated. None of that is demonstrated here, and each would need its own vector, its own surgical target and its own safety package. But the proof that the architecture works in a human brain is the part other programs have been waiting for.

Next Up

The immediate task is replication at scale. The recruiting phase 1/2 trial will need to establish a dose-response relationship for clozapine, define how much hippocampal coverage is required, and track safety over years rather than months, including the metabolic and cardiovascular risks that accompany long-term antipsychotic exposure. The team has said it will continue long-term safety follow-up and evaluate the strategy in larger clinical trials.

There are also questions about durability. AAV-mediated expression can persist for years, but it is not known whether it fades, and there is currently no way to switch the receptor off if a patient needs to stop. For now, the pill is the off switch, and the third patient's experience suggests that a lapse in dosing is punished quickly. Whether that counts as a safety feature or a fragility depends on data the field does not yet have.