Age remains the leading risk factor for neurodegenerative diseases, in which neurons die selectively: some cell types disappear earlier, while others survive for decades. The reasons for this selective vulnerability are unclear, even though aging comes with the accumulation of DNA damage, epigenetic changes, inflammation, and mitochondrial dysfunction. The work sits alongside studies that link aging to the loss of meaning in cellular molecular programs, to brain folds in dementia, and to maps of brain age, but it stands out because it treats aging not as a single process but as the sum of diverging trajectories of individual cell types. The authors use the evolutionary diversity of mammalian lifespan as a natural experiment to find regulatory elements that were selected in long-lived species and are linked to neuronal vulnerability in humans.
The researchers combined two scales. At the species level, they took the genomes of 240 placental mammals from the Zoonomia consortium and predicted chromatin openness in 18 cortical cell types using convolutional neural networks trained on data from four species (human, macaque, mouse, rat). Then, using a phylogenetic linear model, they linked the predicted accessibility of regulatory regions to a longevity coefficient that accounts for body size. At the human level, they used single-nucleus RNA sequencing data from the prefrontal cortex of 69 postmortem samples (age 24-94 years) and developed an iterative elastic net regression model (iENR). The model first predicted chronological age from gene expression, capturing systemic effects, and then iteratively refined its predictions, isolating signals specific to individual cells. This made it possible to separate the intrinsic aging of a cell from the overall aging of the tissue.
The model revealed that genes negatively associated with intrinsic age are enriched in oxidative phosphorylation and mitochondrial electron transport pathways. Neurons with a low intrinsic age (molecularly younger) expressed more nuclear OXPHOS genes. At the same time, in healthy aging and in Alzheimer's disease, it was precisely these neurons that turned out to be vulnerable: their proportion declined with age and in disease. For example, interneurons expressing somatostatin (In-SST) had the lowest intrinsic age and the highest vulnerability. Intrinsic age was more strongly linked to cell loss than systemic age and correlated better with DNA damage, senescence, and epigenetic erosion. At the evolutionary level, 427 regulatory regions that are open in vulnerable neurons and negatively associated with mammalian longevity were named VLR (vulnerability and longevity regions). These regions retained synteny with genes of synaptic function in vertebrates, including the MIR181A1HG locus, in mammals, birds, and reptiles.
The work shows that neuronal aging cannot be reduced to a single trajectory: different cell types age differently, and it is the intrinsic, cell-autonomous programs that determine who survives and who dies. Compared with neighboring studies, which often use bulk data or search for common markers of aging, this work is the first to separate systemic and intrinsic signals at the level of individual cells and link them to evolutionary selection across 240 species. The paradoxical conclusion that neurons that are 'young' by molecular clocks are the most vulnerable changes the view of aging as a simple accumulation of damage: vulnerability may be tied to an inability to switch on protective programs that are already active in 'older' cells.
The topic could move toward larger single-cell cohorts to test the reproducibility of the model and find additional signals. The same iterative approach might be applied to other tissues and diseases where cellular heterogeneity matters. Experimental testing of the discovered VLR regulatory elements looks promising: for instance, editing these regions in neurons or organoids to check their contribution to vulnerability. It would also be worth investigating exactly how mitochondrial dysfunction and oxidative phosphorylation are linked to neuronal survival in neurodegeneration.
The results could influence the development of therapies for Alzheimer's disease and other neurodegenerative diseases aimed at protecting specific types of neurons, such as somatostatin interneurons. The method for separating systemic and intrinsic aging is applicable to any single-molecule data, including cancer, immunology, and regenerative medicine.
Test the model on independent cohorts with a larger number of samples and in other brain regions. Experimentally confirm the role of VLR and genes such as PLCG2, FKBP5, and SST in neuronal vulnerability using knockout or expression modulation.
The work directly addresses the unsolved problem of selective neuronal vulnerability in neurodegeneration: why some cell types die earlier than others. It also links aging to the evolution of longevity, which remains an open question in the biology of aging. The data obtained on mitochondrial dysfunction and epigenetic erosion point to possible mechanisms, but the cause-and-effect relationships have yet to be proven.
In long-lived mammalian species, the regulatory regions linked to neuronal vulnerability are more often closed, meaning evolution seems to 'switch off' dangerous stretches of the genome. The MIR181A1HG locus has kept its connection to its enhancer for hundreds of millions of years across mammals, birds, and reptiles, which points to its ancient and important function.
The idea of molecular clocks that predict cell vulnerability echoes the film Blade Runner, where replicants have a limited lifespan but the vulnerability of different organs and cells is not the same. In Greg Egan's novel Permutation City, the possibility of controlling aging at the cellular level is discussed, which partly reflects the goal of this work.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.