Inside every cell in your body, a chemical clock is running. Not ticking exactly, more like drifting: patterns of methyl groups latching onto DNA, accumulating and shifting in ways that reflect everything you’ve done, breathed, eaten, and been exposed to across a lifetime. Researchers have spent the better part of two decades trying to read these patterns as a measure of biological age, building so-called epigenetic clocks that can, in principle, tell you whether your cells are ageing faster or slower than your birth certificate suggests. For the brain, the hope was obvious: if we could track biological ageing in blood, maybe we could predict who would go on to develop dementia. But a large new study suggests that relationship is considerably messier than anyone would like.
The findings, from a team led by Linda McEvoy at the Kaiser Permanente Washington Health Research Institute, are both clarifying and somewhat unsettling. Working with data from nearly 1,200 older women enrolled in the Women’s Health Initiative Memory Study, the researchers measured five widely used epigenetic clocks at baseline and then, roughly eight years later, examined the participants’ brains using structural MRI. What they found challenges a fairly basic assumption in the field: that biological ageing in the blood would map, more or less, onto biological ageing in the brain.
Two Kinds of Brain Ageing
It doesn’t. Or at least, not straightforwardly. The team used two distinct MRI-derived indices to characterise brain change. The first, called SPARE-BA, reflects general brain ageing: it essentially estimates how old your brain looks compared to brains of people at different life stages, producing something like a brain-predicted age. The second, the AD Pattern Similarity Score (AD-PS), is more specific: it measures how closely an individual’s brain structure resembles the volumetric pattern found in people with Alzheimer’s disease. These two measures sound related, perhaps, but they capture rather different things.
None of the five epigenetic clocks were significantly associated with accelerated brain ageing on the SPARE-BA index. Not one. That’s perhaps the most striking null result here: even clocks specifically designed to track biological deterioration showed no relationship with whether the brain was ageing faster than expected for chronological age.
The Alzheimer’s-pattern score told a different story, though only partly. Of the five clocks, just one, GrimAge2, showed a statistically robust association with the AD-PS measure. Women with higher GrimAge2 scores had brain MRIs that looked more similar to the patterns seen in people diagnosed with dementia. Each standard deviation increase in GrimAge2 was linked to roughly a 1.5% higher AD-pattern score, an effect that survived adjustment for a fairly comprehensive list of potential confounders, including education, physical activity, body mass index, diabetes, and cardiovascular disease.
“Epigenetic clocks of biological aging have been associated with cognitive impairment and dementia,” as the researchers note. But why this particular clock, and not the others?
The Smoking Signal in the DNA
The answer seems to come down to one component. GrimAge2 is not a single measure but a composite of ten DNA methylation markers, each corresponding to an ageing-related plasma protein or exposure. When the team unpacked which component was actually driving the AD-pattern association, it was the epigenetic marker of smoking pack-years, not the inflammation markers, not the metabolic markers. The epigenetic signature of cumulative tobacco exposure, burned into DNA methylation patterns, was linked to reduced frontal and temporal lobe volumes. There was no comparable effect in the hippocampus or entorhinal cortex, the regions most strongly implicated in early Alzheimer’s pathology. This is a fairly important distinction: if GrimAge2 were picking up classical Alzheimer’s biology, you’d expect to see it in those regions first. It doesn’t. Which suggests the signal here is more likely vascular in origin, consistent with how smoking damages brain tissue.
What makes this finding particularly interesting is that fewer than 5% of the women in the study were current smokers at baseline. The majority of the effect appears to come from former smokers, and possibly from non-smokers exposed to environmental tobacco or other toxins. Epigenetic changes related to smoking are known to persist for up to 30 years after cessation, and the DNA methylation marker for pack-years has been shown to predict mortality even among people who report never smoking at all. The biological record of exposure is, in a sense, more faithful than self-report.
There are real limits on what this study can tell us. The cohort was restricted to older women, mostly white, enrolled in a hormone therapy trial beginning in the mid-1990s. Men are excluded entirely. Whether the same patterns hold in more diverse samples, in younger populations, or with newer-generation epigenetic clocks, is genuinely unknown. The MRI measure of brain ageing used here, SPARE-BA, is also just one of several available methods; other measures of predicted brain age might yield different results.
The broader implication, though, might be worth sitting with for a moment. The researchers conclude that “measures of epigenetic and brain age acceleration capture different aspects of biological aging.” That’s a somewhat technical way of saying that these clocks are not measuring one unified thing called ageing. They’re measuring particular molecular pathways, each influenced by particular exposures and each relating to particular downstream outcomes. GrimAge2 happens to contain a smoking signal; that signal happens to relate to frontal and temporal atrophy; that atrophy happens to look, on a machine-learning classifier, like early Alzheimer’s-pattern change. The chain of causation is plausible but indirect, and it runs through decades of cellular memory of an exposure that may be long over.
The practical stakes here are not trivial. Epigenetic clocks are increasingly discussed as potential clinical tools, ways to screen for accelerated ageing or to measure the effects of interventions. If blood-based clocks don’t reliably predict brain ageing in the general sense, they may still predict specific pathological pathways, but only if we know which pathways those clocks are actually sensitive to. Getting that wrong, it turns out, means missing the hippocampus entirely.
DOI: https://doi.org/10.18632/aging.206369
Frequently Asked Questions
What is an epigenetic clock, and how is it measured?
An epigenetic clock uses patterns of chemical tags called methyl groups, which attach to DNA over time, to estimate biological age. These patterns are measured from a blood sample and compared to reference patterns from people at different life stages. Some clocks are trained to predict chronological age; others are trained on health outcomes or mortality risk, making them sensitive to different biological pathways.
Why did epigenetic clocks fail to predict general brain ageing in this study?
The researchers found no significant association between any of the five clocks and a composite MRI measure of how old the brain appears relative to chronological age. This suggests that blood-based molecular ageing and structural brain ageing are, at least in part, driven by different biological mechanisms. It’s possible that future clocks trained specifically on brain or cognitive outcomes might perform better.
Is this study saying that smoking causes Alzheimer’s disease?
Not directly, though the connection is notable. The epigenetic marker of cumulative smoking exposure was linked to reductions in frontal and temporal lobe volumes, which produced brain scans that resembled those of people with Alzheimer’s. But the hippocampus, the region most affected in early Alzheimer’s, was not implicated. Researchers think the mechanism may be vascular rather than the amyloid plaques typically associated with classic Alzheimer’s pathology.
Does this apply to people who quit smoking years ago?
Potentially, yes. Epigenetic changes related to smoking are known to persist for up to three decades after cessation. The DNA methylation marker for pack-years also predicts mortality risk in people who have never smoked, suggesting it may capture exposure to second-hand smoke or other environmental toxins. Most of the women in this study were former rather than current smokers at the time of the blood draw.
Could this research change how dementia risk is assessed?
It adds useful nuance. Rather than treating epigenetic clocks as general ageing indicators, this work suggests clinicians and researchers need to understand which specific pathways each clock is measuring. GrimAge2 appears sensitive to smoking-related vascular damage rather than amyloid pathology, which means it might be more useful for flagging vascular contributions to cognitive decline than for predicting classical Alzheimer’s disease.
Discover more from NeuroEdge
Subscribe to get the latest posts sent to your email.