Alzheimer's Bends the Shape of the Genome Itself, a New Single-Cell Map Finds

A Pittsburgh and Carnegie Mellon team read gene activity and DNA folding in the same 23,825 brain cells from 20 postmortem donors, and found the architecture of the genome disturbed in Alzheimer's. The work is a map, not a mechanism, and every cell in it came from a brain after death.

Two metres of DNA are packed into a nucleus a few microns wide, and the packing is not random. It loops, it clusters, it sorts itself into neighbourhoods of busy DNA and quiet DNA, and it holds the switches that turn genes on close to the genes they govern. For years the single-cell atlases of Alzheimer's have counted which genes are loud and which have gone quiet in which brain cells. A study published in Science on 23 July 2026 asked a different question: what does the folding look like in those same cells, in the same moment, in the same nucleus?

The answer, from a team led by Jian Ma at Carnegie Mellon with Hansruedi Mathys at the University of Pittsburgh School of Medicine and Zhijun Duan at the University of Washington, is that the folding is changed. The researchers used GAGE-seq, a method that measures the transcriptome and the three-dimensional chromatin contacts from the same single cell, on prefrontal cortex tissue from 20 donors in the Religious Orders Study and Rush Memory and Aging Project: 10 with Alzheimer's pathology and 10 age-matched donors without it. After quality control, 23,825 nuclei survived. The clinical records from ROSMAP confirmed the groups differed as expected, with higher plaque and tangle burden and lower global cognition scores in the Alzheimer's group. To place the findings back in intact tissue, the team added Xenium spatial transcriptomics from four donors, 213,271 cells after filtering, and integrated matched chromatin accessibility data from the same participants.

The signature they describe has a memorable name: increased compartment mingling. The genome normally sorts itself into large active and inactive compartments that keep their distance from each other. In cells from the Alzheimer's donors those boundaries blurred, across several cell types. Short-range contacts thinned out and long-range contacts increased, while contacts between genes and the nearby regulatory elements that control them weakened. Greater mingling tracked with lower overall gene activity, and the patterns lined up with the biology the field already recognises: reduced neuronal and synaptic programmes, altered metabolic and stress responses, and senescence-related programmes in microglia. "The genome's 3D structure is a fundamental regulatory layer that helps to connect DNA sequence to gene activity," Ma said in a Carnegie Mellon statement.

The team then built Hicformer, a deep learning model that predicts gene expression from DNA sequence plus one- and two-dimensional chromatin features, and compared it with a sequence-only version. The genes whose expression the structural model predicted far better than sequence alone were, in Alzheimer's microglia, enriched for type I interferon signalling, cytokine production and immune regulation. That is the study's sharpest point: the folding is carrying information that the letters of the DNA do not. Alzheimer's has long been read as a sequence story, with risk written into variants such as APOE. This is a layer above the sequence, and it is disturbed in disease.

What this means for a family living with a diagnosis is honest and modest. Nothing here changes a treatment, a test, or a prognosis, and nothing here is a drug. It is a map of where to dig. If specific weakened contacts between a gene and its regulatory element turn out to drive the loss of synaptic programmes rather than merely accompany it, those contacts become candidate targets, and the authors say plainly that determining which structural changes contribute directly to disease is work still to be done. "Our study represents a major advance in understanding what goes wrong in Alzheimer's disease," Mathys said in a University of Pittsburgh release, and the phrase to hold onto in that sentence is understanding.

There is also a caution worth stating in the same breath as the excitement. Twenty donors is a small number for a disease that varies enormously between people, all the tissue came from one brain region, and every measurement is a single frozen instant taken after death. A nucleus cannot be watched folding and unfolding over the twenty years a dementia takes to unfold with it. What the study offers is a new axis on which to look, and a public framework, Hicformer, for others to test it.

The next question is causal and cannot be answered in postmortem tissue: whether the folding shifts first, or follows the damage. That takes experiments in living systems, and those have not been done.