Study Finds 26% of Fetal Neurons Carry Big DNA Errors, 6% After Birth

Before birth, the human brain makes more neurons than it keeps, and many die around the time of birth. A new study suggests that this cull is not random: it may clear out neurons whose DNA is badly scrambled.

The work, published in Nature on 28 October, was led by Diane Shao of Boston Children’s Hospital and Yifan Zhao of Harvard Medical School, with Peter Park of Harvard Medical School and Christopher Walsh of Boston Children’s Hospital and the Howard Hughes Medical Institute as senior authors.

What the team measured

The researchers looked for copy-number variants, or CNVs. These are stretches of DNA that have been duplicated or deleted, so a cell ends up with too many or too few copies of some genes. Inherited CNVs are a known risk factor for a range of neurodevelopmental and psychiatric conditions.

Rather than read DNA from bulk tissue, the team read the genomes of single neurons from the cerebral cortex, using donated brain tissue from deceased individuals aged from 14 weeks of gestation to over 90 years. In total they sequenced 2,357 neurons. They used a new low-cost method to read 1,526 neurons from 16 people at shallow depth, then checked the result with a second, deeper method on 831 neurons from 12 people.

A quarter before birth, about 6% after

With the first method, an average of 25.9% of neurons before birth had what the authors call widespread CNVs, involving several chromosomes. Across fetal samples the share ranged from 12.7% to 46%. After birth, it fell 4.2-fold to about 6.2%. The second method gave 33.1% before birth and 7.2% after, a 4.6-fold fall. Taken together, the paper says the share dropped about 4.5-fold.

The share then held fairly steady through life: 7.1% in infants up to two years old and 4.8% in people over 60. Neurons with smaller numbers of CNVs declined more slowly, over decades.

“We were totally unprepared for how common these were in early development,” Walsh said in Harvard Medical School’s news release.

Why the authors think the brain removes them

The fall is timed roughly with the known wave of programmed cell death around birth, when the brain sheds a large share of its neurons. The paper also found that the CNVs left in neurons after birth were depleted of genes that are sensitive to dosage and genes linked to neurodevelopmental disorders. That pattern fits the idea that neurons carrying changes in those genes were removed.

To look for a mechanism, the team analysed 44,861 cell nuclei for gene activity. Neurons with heavy CNV burdens showed abnormal activity in genes that build synapses, the connections between neurons. Developing neurons compete to form synapses, and those that fail tend to die. The authors suggest that faulty genomes may leave some neurons unable to compete. “The brain is selecting for genomic integrity,” Shao said in the release.

What the authors say about autism and other conditions

The paper does not show that this process fails in any disease. Its abstract says only that eliminating defective neuronal genomes “might represent an important developmental genome quality-control process that may be vulnerable in neurodevelopmental disease”.

Shao put it as a possibility: “If that quality control system can be disrupted, that could be part of what goes wrong in neurodevelopmental disorders like autism.” The release says the team is now looking at whether the process differs in the brains of people with autism and other neurodevelopmental disorders. It lists epilepsy among the conditions to which CNVs in general may contribute, but the paper reports no epilepsy finding. Walsh’s group is also asking whether the same process plays a part in adult neurodegenerative diseases such as Alzheimer’s.

Single-cell reading changes the question

This is TSN’s analysis. The main advance here is the method. Standard genetic tests read DNA pooled from many cells, often from blood, so changes that arise in single brain cells during development are invisible to them. A cheap way to read thousands of single neurons lets researchers ask how common such changes are, and when they disappear. As single-cell and genome tools get cheaper, from portable sequencers that run on a desktop computer to AI agents mining large brain atlases, this kind of census becomes possible for other tissues too. The finding that the developing brain may run its own quality control is good news for basic biology: it suggests a high early error rate is normal, and that the brain has a way to deal with it.

Small samples and post-mortem tissue

The limits matter. The data come from post-mortem tissue from fewer than 30 donors across the two methods, so each age group rests on a handful of brains, and the authors cannot watch the same neurons die. The link to disease is a hypothesis. Earlier tools for autism, such as the FDA-cleared assessment aid TSN covered, work on behaviour, not on this biology, and any test or treatment built on this work is a long way off. The next step is the one the team describes: comparing brains from people with and without neurodevelopmental conditions. Work on protein misfolding in the brain, such as the Johns Hopkins-led ARPA-H programme, shows how long the path from mechanism to diagnosis tends to be. For more science with a public benefit, see TSN’s Tech for Good hub.

Sources

• Shao, D. D., Zhao, Y. et al., “Perinatal elimination of genetically aberrant human cortical neurons”, Nature, 28 October 2026
• Harvard Medical School (via Newswise), “Abnormal Neurons Are Common in the Developing Brain. The Body Has a Mechanism to Eliminate Them”, 28 October 2026

Related stories

• OpenAI Says Its GPT-Rosalind Agents Mined a Mouse-Brain Atlas in 32 Minutes
• You Can Now Sequence the Human Genome on a Mac Mini
• Johns Hopkins-Led Team Wins Up to $52.1m to Model Protein Misfolding