The prefrontal cortex mapped cell by cell, across 1,494 brains

On 23 September, nine papers appeared at once across Nature, Nature Genetics, Nature Medicine, Nature Communications and Scientific Data. They come from the PsychAD Consortium, a project led from Mount Sinai in New York and funded by the US National Institute on Aging since 2019, and they describe a single dataset: the gene activity of 6.3 million individual cells taken from the brains of 1,494 people after death.

The donors include 500 people with Alzheimer’s disease, 112 with Lewy body disease, 85 with vascular dementia, 40 with Parkinson’s disease, 120 with schizophrenia, 55 with bipolar disorder, and a large group with no diagnosis. The ages run from before birth to over 100. Every sample comes from the same place, the dorsolateral prefrontal cortex, a patch of brain behind the upper forehead that handles working memory, planning and self-control, functions that falter in almost every one of those conditions.

What makes this more than a large number is the resolution. Most earlier studies of diseased brains ground up a piece of tissue and measured the average activity of every gene in it. That average blends neurons with the far more numerous support cells around them, so a change in one rare cell type disappears into the mix. PsychAD used single-nucleus RNA sequencing: it pulls out the nucleus of each cell and reads which genes that one cell was transcribing into RNA, the working copies a cell makes of the genes it is using. Do that six million times and you can sort cells into types (the consortium distinguishes 27 subclasses) and then ask how each type differs between a healthy brain and a sick one, or between a brain of 20 and one of 80.

A timetable for the cortex

The finding that will travel furthest is about time, not disease. A lifespan paper within the package, built on 1.3 million nuclei from 284 donors aged 0 to 97, describes the prefrontal cortex’s molecular life in three acts. Childhood and adolescence are turbulent: cells are still being made and matured, and gene activity shifts quickly. Around the mid twenties the pace drops sharply, and the authors single out age 24 as an inflection point after which the cell mix settles. Then comes a long plateau through middle adulthood, when little changes. After about 60 the curve rises again, and this time the change is carried mostly by glia, the non-neuronal cells that insulate, feed, clean and defend the neurons, and in particular by immune cells.

One detail from that late phase is striking. Genes that follow a daily rhythm in neurons appear to lose their rhythm after 60, while immune cells pick up rhythmic activity of their own, tied to stress and inflammation. How strongly to trust that depends on method, since nobody’s brain can be sampled across a day; rhythms in post-mortem tissue can only be reconstructed across many donors.

The cross-disorder papers then place risk genes on that timeline, and the pattern is clean enough to sketch.

Genes linked to schizophrenia and bipolar disorder are most active early in development, years or decades before anyone shows symptoms. Genes linked to Alzheimer’s disease are expressed mainly in ageing glial cells rather than in neurons. Across the seven conditions the consortium reports both common pathways and disease-specific signatures spread over neuronal, glial, immune and vascular cells, which is to say the disorders are neither one disease wearing different masks nor seven unrelated ones.

Why it matters

Three consequences follow, in rising order of ambition.

The first is genetic. Large studies have found hundreds of stretches of DNA that raise the risk of these disorders, but most of them sit outside genes, in switches that regulate other genes, and nobody knew which cell the switch acts in. A Nature Genetics paper in the package maps thousands of such switches at single-cell resolution and finds their effects are highly cell-type specific. A risk variant that looks inert in a tissue average can be doing something decisive in one cell population. The package also deliberately includes 439 donors of non-European or mixed ancestry, which matters because genetic findings drawn from European cohorts often transfer poorly.

The second is about individuals. A Nature Medicine paper introduces a machine-learning tool, PASCode, that picks out cell states associated not just with Alzheimer’s pathology but with depression and cognitive decline within the disease, and with resilience: people whose brains carry the damage without the symptoms. Another paper models donor-specific networks and finds substantial variation between patients who share a diagnosis. That is the evidence base for the recurring idea that “Alzheimer’s” names several processes rather than one.

The third is therapeutic, and it is the most promissory. Mount Sinai says a robotic platform is already screening more than 1,000 compounds a week against molecular changes the atlas identified. That is a statement about activity, not results.

What the atlas cannot show

Every brain in this dataset is a dead one, and that shapes everything above. A post-mortem sample is the end state of a life: a brain after decades of disease, medication, and whatever killed its owner. Changes seen in an Alzheimer’s brain may be causes, consequences or bystanders, and a single snapshot cannot tell them apart. The three acts of the lifespan are also inferred from different people at different ages, not from anyone followed over time, so the “plateau” is a pattern across a population, not a guarantee about any one brain. Seth Ament of the University of Maryland, commenting to The Transmitter, called such databases “mainly useful for generating hypotheses”.

The coverage has gaps. One region of cortex stands in for the whole brain, which is a real limitation for Parkinson’s disease, whose defining damage lies deep in the midbrain; with 40 Parkinson’s donors, conclusions about it are the thinnest in the set. RNA in a nucleus is not protein at a synapse, and a gene being read is not proof its product is doing anything. And the lifespan paper has circulated as a preprint since late 2024, so its headline shape is not new to specialists, even if the full package is.

None of that diminishes what was built. What was published last week is a reference: a description, at the level of individual cell types, of what a prefrontal cortex looks like at each age and in each of seven conditions, released openly for others to test. The consortium says it intends to grow the resource to about 10,000 people. The claims that will matter most, about which cells to treat and when, are the ones that still have to be checked in living patients.

Further reading

• The PsychAD Consortium collection, the nine papers gathered by Nature.
• Lifespan single-cell transcriptomic atlas of the human prefrontal cortex, the three-act paper in Nature.
• Prefrontal cortex has three transcriptomic acts, The Transmitter’s coverage, with comment from outside researchers.
• Nine studies led by Mount Sinai investigators, the institutional summary of each paper.