AI nucleosome study reveals graded DNA accessibility, genome control mechanism

Artificial intelligence analysis has shown that more than 85 per cent of nucleosomes contain partially accessible DNA, challenging the long-standing binary model of genome regulation and pointing to a programmable ‘grammar’ of chromatin structure

Each human cell compresses more than six feet of DNA into the microscopic volume inside a cell, which is a feat comparable to fitting an entire house into a single cube of sugar. To achieve this, DNA wraps around nucleosome protein complexes which act as structural units that both compact and regulate access to genetic material. 

For decades, researchers have assumed that DNA bound tightly within nucleosomes remains inaccessible and therefore inactive, while only unwrapped DNA can support gene expression. A study from the University of California, San Francisco affiliated Gladstone Institutes and the Arc Institute of Palo Alto, California, USA, has now challenged that binary framework.

The findings have indicated that nucleosomes rarely exist in either a fully closed or fully open state. Instead, most contain regions of DNA that remain partially accessible, which enables a graded form of gene regulation rather than a simple on–off switch. This observation has introduced a more nuanced model of chromatin organisation and gene control.

In reaching this conclusion, the research team developed an artificial intelligence driven computational method known as ‘Iteratively Defined Lengths of Inaccessibility’ (IDLI). This approach builds on an earlier technique called ‘Single-Molecule Adenine Methylated Oligonucleosome Sequencing Assay’, which has mapped nucleosome positions along individual DNA strands. The new method extends this capability by interrogating the internal structure of each nucleosome, rather than merely its location.

IDLI has analysed sequencing data in two dimensions – along the DNA strand and within the nucleosome itself – to detect structural variation. Each nucleosome comprises eight histone protein subunits. The method has identified cases where these subunits appear partially displaced or loosely associated, which exposes segments of DNA that would otherwise remain concealed.

When applying this approach to chromatin from mouse embryonic stem cells it has revealed that more than 85 per cent of nucleosomes exhibit some degree of structural distortion. These distortions have correlated with varying levels of gene activity, which suggests that chromatin architecture operates along a continuum rather than as a binary system.

“The conception before was that, when it came to nucleosomes, genes were either turned on or off, but we’re finding it’s more like a volume dial,” said Dr. Vijay Ramani of the Gladstone Institutes, who led the study.

“This is a completely novel [understanding of the] organisational code for the genome,” he said.

All cells contain the same genetic sequence, yet distinct cell types express different subsets of genes to fulfil specialised functions. This selectivity depends on regulatory systems that control DNA accessibility. Nucleosomes have long formed a central component of this regulatory architecture and researchers have studied chromatin – the combined complex of DNA and nucleosomes – to infer patterns of gene expression.

The present study has suggested that chromatin regulation involves a richer structural vocabulary than previously recognised. The researchers have identified 14 distinct nucleosome conformations, each associated with specific transcriptional states. These conformations have appeared consistently not only in mouse embryonic stem cells but also in human stem cells directed to adopt liver-like identities and in mature liver cells isolated from mice.

“Before this, our understanding of chromatin was a bit like [trying to read] a text that only had sound and silence – just two states,” said Dr. Hani Goodarzi of the Arc Institute, who co-led the study.

“Now we can see that … there are letters and words and we [have] uncovered a novel kind of grammar that controls them,” he added.

The team has also demonstrated that transcription factors – proteins that regulate gene expression – directly influence nucleosome structure. Experimental removal of two such factors has led to predictable shifts in nucleosome configuration which indicates that these proteins actively modulate chromatin accessibility by stabilising or destabilising nucleosome architecture.

“This adds to the many different ways in which a cell can tune things up and down, by making parts of the DNA more or less accessible,” said Ramani.

The implications extend to disease biology. Many complex disorders, including cancer and neurodegenerative conditions, do not arise from single genetic mutations but from subtle, coordinated changes in gene expression across multiple loci. The graded nucleosome states described in this study may provide a mechanistic basis for such effects, as partial accessibility could allow inappropriate or incomplete gene activation.

“These are precisely the states that end up being quite important in terms of disease relevance.

“Most complex diseases revolve around gradation. Maybe a gene is on but at half the level it would normally be, or maybe it’s on in the wrong cell type,” said Ramani.

The research has also opened avenues in ageing biology. Chromatin structure alters predictably as cells age, and some of these alterations appear reversible. The ability to map nucleosome states at high resolution may therefore allow researchers to track and potentially correct age-associated changes in gene regulation.

“We’re reading the language but, ultimately, we want to learn how to speak it so that we can control and modify it.

“We’re not here just to observe biology; at some point we want to intervene,” concluded Goodarzi.

For further reading please visit: 10.1038/s41586-026-10418-6