Aging is the leading risk factor for chronic diseases, but its molecular drivers remain incompletely understood. Epigenetic changes, especially in chromatin, are seen as a key mechanism behind the loss of cellular identity. Existing aging clocks based on DNA methylation are accurate but hard to interpret at the level of regulatory programs, and they usually work at the tissue level rather than in individual cells. This work stands somewhat apart: nearby studies examine cell aging through methylation, brain age maps, and white matter analysis, but no one had built cell-type-specific clocks from chromatin accessibility for the brain.
The authors used the largest human prefrontal cortex snATAC-seq dataset: 1.5 million cells, 521,000 chromatin accessibility peaks, and 357 donors aged 15-100 years. Cells were pseudobulked by donor and cell type (excitatory and inhibitory neurons, astrocytes, microglia, oligodendrocytes, OPCs). Ridge regression models were trained to predict age separately for all cells and for each cell type. To test transferability, peak coordinates were aligned across species (hg38, mm10, danRer11) and the models were applied to independent cohorts: human hippocampus, mouse hypothalamus, mouse and zebrafish retina, mouse liver with SIRT6 overexpression, and data from Alzheimer's and Parkinson's diseases. Interpretation relied on SHAP and motif analysis.
Within the training cohort, the clock predicted age with a Pearson correlation of 0.79-0.91 and a mean absolute error of 6.7-11.6 years. Age was predicted best from excitatory neurons and worst from astrocytes. In the independent hippocampus, the correlation was 0.65-0.89. In the mouse hypothalamus, correlations reached 0.82 for inhibitory neurons but were weak for oligodendrocytes. In mouse and zebrafish retina, the all-cell clock showed a positive association with age in nearly all cell types. The clock captured slowed aging in mice with SIRT6 overexpression: predicted age was lower in transgenic animals (p=0.03 in old ones). In Alzheimer's and Parkinson's diseases, age residuals (the difference between predicted and chronological age) positively correlated with pathology severity, most strongly in microglia. In women with high Alzheimer's pathology, aging acceleration was more pronounced than in men, especially in oligodendrocytes and OPCs. Top age-related peaks overlapped with loci of methylation clocks and were linked to genes of inflammation, metabolism, and histones.
This work is the first to show that cell-type-specific clocks based on chromatin accessibility, trained on a single brain region, transfer to other regions, tissues, and species without retraining. This sets it apart from neighboring studies that either use DNA methylation, build clocks from the transcriptome, or are limited to one tissue. Here, the result is a tool that simultaneously reads biological age and yields interpretable regulatory elements, opening a path to a mechanistic understanding of aging and to screening anti-aging interventions.
Other interventions (diet, exercise, parabiosis) and risk factors (toxins, stress, smoking) can be tested. It would be promising to expand the training sample to other brain regions and species, and to move to nonlinear models that capture nonlinear aging dynamics. It is important to test causal relationships through targeted perturbations of specific regulatory elements.
The method will influence gerontology, the neurobiology of aging, drug development, and personalized medicine, because it allows quantitative assessment of the biological age of individual cell types and the effects of interventions.
Test the clock on additional cohorts and interventions. Run experiments with targeted editing of the identified regulatory elements to establish causality.
The connection to unsolved problems in the neurobiology of aging: which cell types are most vulnerable and why, what molecular drivers underlie the heterogeneity of aging, and how to separate universal mechanisms from context-specific ones. The work provides a tool for quantifying cell-type-specific biological age, but it does not settle the question of causality in epigenetic changes.
A clock trained on human prefrontal cortex cells managed to predict age in the zebrafish retina, an animal whose last common ancestor with humans lived more than 400 million years ago.
The idea of an "aging clock" that can be turned back echoes Oscar Wilde's novel The Picture of Dorian Gray, in which aging is transferred to a portrait, as well as the film Cloverfield Paradox, where accelerated aging is used as a dramatic device.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.