Alzheimer's isn't only amyloid and tau. DNA's 3D shape is also disrupted in affected brain cells.

Alzheimer's has been told for decades as a story about misfolded proteins: amyloid plaques that clump between neurons, and tau tangles that build up inside them. A new study in Science adds a different layer. Inside specific brain cell types from donors who had Alzheimer's, the 3D shape DNA folds into is also disrupted, and that reorganization changes which genes are switched on.

The work, published September 12 in Science (DOI 10.1126/science.adz1652), is the first cell-type-resolved map of 3D genome reorganization in Alzheimer's postmortem brain tissue. It comes from a collaboration between Carnegie Mellon's School of Computer Science, the University of Pittsburgh School of Medicine, and the University of Washington, and it positions genome folding as a regulatory axis in the disease that sits alongside, not instead of, the long-standing amyloid and tau story.

"This study gives us a new layer of Alzheimer's biology to investigate," said Jian Ma, the Ray and Stephanie Lane Professor of Computational Biology at CMU's School of Computer Science, in a CMU release. Ma's lab has spent years building tools to read the genome's physical structure inside individual cells, and the new paper is its most direct test of how that structure changes in disease.

DNA does not lie flat inside a cell. Two meters of it get packed into a nucleus only a few microns across, and the way it loops and folds determines which genes are accessible to the machinery that switches them on. That packing is not just packaging. The same gene can be active or silent depending on which 3D neighborhood it sits in. Reorganize the neighborhood, and you change the cell's behavior.

The team used postmortem prefrontal cortex samples from donors who had Alzheimer's and from controls who had not, all drawn from a long-running dementia study cohort. They applied a method called GAGE-seq, which measures gene expression and 3D genome contacts in the same single cells, and combined it with spatial mapping of intact tissue. A new deep learning model pulled the three views together: folding, cell state, and where each cell sits in the brain.

The result is a cell-by-cell picture of how 3D genome architecture shifts in specific brain cell types in Alzheimer's, and which of those shifts line up with altered gene activity and with the tissue's structural changes. The team frames the reorganization as a previously underexplored regulatory layer, one that could eventually point to mechanism-driven targets distinct from amyloid and tau.

The scope is narrower than the framing sometimes suggests. This is postmortem correlation from a donor cohort drawn from a single long-term dementia study. It cannot tell whether the folding changes are cause, consequence, or both, and it does not yield a treatment, a clinical timeline, or a drug. Ma's own framing, in a companion CMU news piece, is that the value now is mechanistic: a new layer of the disease to investigate, with new things to test.

For a reader whose mental model of Alzheimer's is plaques and tangles, the update is that the protein story is not wrong, but it is no longer the whole story. The genome's 3D architecture inside specific brain cell types is also disrupted in the disease, and that disruption is now measurable cell by cell. Replication in larger cohorts, separation of cause from consequence, and the search for druggable nodes are the open tests this map leaves behind.

Drug Target Review's coverage and Ma's publication list place the paper in a longer thread of single-cell 3D-genome work from the lab, including tools that this study now applies to a disease question.