They Turned Up the Healthy Copy of an Autism Gene. It Worked. In a Dish.

A team at the University of California, Los Angeles, has used CRISPR to turn up two autism risk genes in lab-grown human neurons and cortical organoids — pea-sized clumps of brain tissue grown from stem cells. The group reports that the genes became more active only in cells carrying a mutation, and that several of the differences the mutation causes moved back toward normal.

Nothing in the work was tested in an animal or in a person. Worth stating early, because the headline will not.

The authors write that about 15% of the genetic susceptibility to autism comes from new mutations that knock out one of a gene's two copies, leaving the other to work at roughly half strength. Their approach leaves the broken copy alone and raises output from the healthy one, using the cell's own controls. The paper was published Nov. 13, 2026, in *Nature Neuroscience*.

The group mapped stretches of regulatory DNA called enhancers around the two genes, CHD8 and SCN2A, and picked ones predicted to be active during fetal brain development. Aiming the same activator at a gene's promoter instead raised expression in healthy cells too — the overshoot they wanted to avoid.

In the mutant lines, CHD8 ran at about half the normal level. Mutant organoids grew larger, with too many of the dividing cells that go on to make neurons. The authors report that treated organoids came back toward normal size, with fewer of those cells and more neurons, and that overall gene activity shifted toward the unmutated pattern.

Only about a quarter of the cells in a treated organoid carried the activating protein. That is not a trivial detail. A quarter of cells is not the whole tissue, and the effect still showed up at the organoid level — which suggests something beyond cell-autonomous rescue, or a measurement that deserves a closer look. The paper reports what it reports.

SCN2A encodes a sodium channel neurons use to fire. Treated neurons fired at close to normal rates, including neurons grown from two patients' own cells.

One treatment lasted months in culture. Timing mattered: the increase faded later on, and organoids first treated at 300 days old showed none.

So the intervention has a window. Treat too late, and nothing happens. That is a useful finding and an inconvenient one for anyone imagining a straightforward therapeutic.

The authors call the work a proof of principle and offer the method as a workflow for other autism risk genes. They report no rise in the activity of neighboring genes — one of the off-target checks they ran.

Other checks presumably exist. The paper is a proof of principle in organoids, not a treatment, not an animal study, not a clinical trial. The cells are human, the tissue is lab-grown, and the "rescue" is measured in firing rates and organoid size, not in behavior, cognition, or anything a person would recognize as a life.

CRISPR activation of the healthy copy of two autism risk genes partially normalized several cellular and electrophysiological phenotypes in human organoids. The effect was mutation-specific, enhancer-targeted, time-dependent, and achieved with only a quarter of cells carrying the activator. No animals. No people.

That is a solid piece of mechanistic work. It is also the kind of result that gets rewritten somewhere between the journal and the group chat as "scientists reverse autism in a dish." They did not reverse autism. They nudged some cellular readouts toward normal in tissue grown from stem cells. The distinction matters, and the paper itself is careful about it. The retelling usually is not.