Key Takeaways
- DHEAS serves as a crucial biomarker for estimating biological age, highlighting the differences between chronological and biological aging.
- Researchers created a model using routine blood tests that predicts biological age with about six years of accuracy, which is practical compared to DNA methylation methods.
- The study introduces a gut microbiome model that also estimates age by analyzing bacterial species, showing a strong correlation with blood parameters.
- Both biological age models suggest shared underlying processes of aging, indicating potential pathways for interventions that could influence biological age.
- Further validation on diverse populations is necessary before applying these models clinically, but they offer promising insights into aging mechanisms and health monitoring.
DHEAS doesn’t care about your birthday. The hormone, produced by the adrenal glands and gradually declining across adult life, has no particular interest in the number printed on your driving licence.
What it tracks instead is something harder to measure and considerably more useful: the actual functional state of the organism it’s running through. Which is why, when Anastasia Kobelyatskaya and her colleagues at the Russian Clinical Research Center for Gerontology sat down to build a neural network that could estimate biological age from a small panel of routine blood tests, DHEAS ended up doing a lot of the heavy lifting. In both men and women, it exerted the strongest individual pull on predicted age, nudging the estimate younger in people whose levels are healthy and older in those whose adrenal output has dropped off. A single molecule, doing what birthdays cannot.
The broader project Kobelyatskaya’s team was tackling is one of the more pressing puzzles in gerontology right now. Chronological age, the kind you celebrate, is a lousy predictor of how old a body actually is. Two 60-year-olds can differ by a decade or more in their biological clocks, shaped by genetics, diet, stress, inflammation, and dozens of other variables that calendars simply don’t record. Researchers have been building so-called biological clocks for some years now, mostly using DNA methylation patterns, but those require specialised equipment and are difficult to translate into clinical settings. The question Kobelyatskaya’s group asked was simpler: could you get there with standard blood work?
The answer, published in the journal Aging this March, turns out to be yes, roughly. Their biochemical model uses just seven markers, selected from an initial panel of 85 blood parameters on the basis of how strongly each one correlates with age. Three apply to everyone: cystatin-C, which reflects kidney filtration capacity and rises with age; IGF-1, a growth hormone mediator that declines and contributes to loss of muscle mass; and DHEAS, with its neurosteroid activity and links to cognitive slowing when it drops. The remaining four are sex-specific. Women’s model adds homocysteine (a cardiovascular risk factor), urea (another kidney indicator), glucose, and zonulin, a protein that regulates gut barrier permeability. Men’s model swaps those out for HbA1c (chronic blood sugar), NT-proBNP (a marker of cardiac wall stress), hs-CRP (the classic inflammaging signal), and free testosterone.
What’s notable is how coherent these lists are, even if you didn’t already know they were tracking age. They read, essentially, as a summary of systemic decline: the kidneys slow, muscle mass retreats, inflammation creeps up, hormones drop, glucose regulation frays. The body’s systems degrade in concert, and the blood apparently records it in real time if you know what to look for.
Chronological age is simply how long you’ve been alive; biological age reflects the functional state of your body, shaped by genetics, lifestyle, and environment. Researchers measure it by tracking biomarkers that change predictably with age, such as hormone levels, inflammatory proteins, and kidney function markers, and using those values to estimate where your body sits on the aging continuum. Two people with the same birthday can differ considerably in biological age, which is part of why it’s increasingly seen as a more clinically meaningful number than the one on your passport.
The gut microbiome shifts in consistent ways across a lifespan, with certain beneficial species declining and others associated with inflammation or metabolic disruption becoming more prevalent. Researchers found that the relative abundance of 45 bacterial species could predict age with roughly the same accuracy as a blood panel, and that the two models agreed closely, suggesting they’re both picking up the same underlying biological deterioration through different windows. The practical limitation is that measuring gut bacteria currently requires sequencing that most clinics can’t easily provide.
The blood and microbiome models both achieved mean absolute errors of about six years, meaning their age predictions land within six years of actual age on average. Established clocks based on DNA methylation can get closer to three years, but require expensive specialist equipment. The new models use standard clinical blood tests, which makes them more accessible, even if slightly less precise. Compared to the widely used PhenoAge algorithm, the new models are a little less accurate but have the advantage of not requiring chronological age as one of their inputs.
That’s part of the researchers’ explicit motivation. A biological age estimate that responds to lifestyle changes, drug treatments, or dietary interventions would give clinical trialists an early, interpretable signal of whether an intervention is actually slowing aging, rather than waiting years for downstream health outcomes. The explainability layer, which shows the contribution of each individual marker to the final prediction, is designed precisely to make that kind of monitoring possible.
The study was conducted on a Caucasian-only population, which means the models may not generalise accurately to other ethnic groups without further validation. The microbiome model requires sequencing technology that isn’t widely available in clinical settings. Both models were tested on a relatively modest dataset of 637 people, and the researchers call for validation in larger, more ethnically diverse cohorts before these tools move into clinical practice.
Trained on data from 637 patients aged 18 to 99, the blood model predicted age with a mean absolute error of roughly six years. That’s not quite as sharp as DNA methylation clocks, some of which get under three years, but it’s considerably more practical. Getting a full epigenetic workup requires equipment most hospitals don’t have. A standard biochemistry panel does not. The team also ran their neural network with Shapley Additive Explanation values, a technique borrowed from machine learning that breaks open the black box and shows how much each variable is contributing to any individual prediction, in years added or subtracted from baseline.
The second model the researchers built runs on gut microbiome data rather than blood. It uses the relative abundance of 45 bacterial species, measured by full-length gene sequencing from stool samples, and produced similarly accurate results. Twenty-nine of those 45 species are negatively correlated with age, meaning their populations shrink as we get older. Among the most informative: Blautia obeum and Butyricicoccus pullicaecorum, both of which show the clearest gradient of decline across age groups. Losing them, it seems, is a fairly reliable sign that something in the gut is shifting toward an older biological profile.
What makes the dual-model approach interesting isn’t just that both clocks work, but that they agree with each other to a degree the researchers found somewhat striking. When you run both models on the same person, the estimates correlate at better than 0.84. Since the gut and the blood are measuring quite different biological systems, that agreement suggests they’re picking up something shared, some deeper underlying process that expresses itself through multiple physiological pathways. The team identified three candidate axes connecting them: a chronic inflammation pathway (hs-CRP rising, pathogenic gut species proliferating), a metabolic axis involving insulin resistance (high glucose and HbA1c correlating with declining butyrate-producing bacteria), and a gut barrier dimension where zonulin levels and protective microbiome species track together. Each axis is arguably not just a marker of aging, but a potential driver of it.
There are real limits here. The cohort was entirely Caucasian, which could skew the models in ways that won’t become clear until they’re tested on other populations. The microbiome model requires sequencing, which most clinics can’t yet offer, so its near-term utility is probably confined to research settings. And compared to the established PhenoAge algorithm, which achieves a mean absolute error of around four years using a slightly different marker panel, both new models are a notch less precise (though they don’t require chronological age as an input, which PhenoAge does, meaning they’re arguably measuring something more independent).
The researchers acknowledge all of this and call for external validation in larger, more diverse cohorts before anyone starts using these predictions clinically. Reasonable enough. But the direction of travel seems worth noting. The ambition of the field has shifted from simply asking how old a body is to asking why it ages the way it does, and then to tracking whether interventions can change the answer. The explainability layer matters there: a model that can show you, in years, exactly how much a particular marker is contributing to a high biological age estimate is a model that could, in principle, tell you which lever to pull.
Whether gut microbiome profiles will eventually become as standard as a cholesterol test is hard to say. The technical barriers are real, and the practical questions about what patients should actually do with that information remain mostly unanswered. But Kobelyatskaya’s team frames the longer-term goal clearly: these aren’t just clocks. “As the proposed models possess both global and local explainability,” they write, “they hold future potential for application in monitoring the effectiveness of various interventions in clinical trials.” The biological age number, in other words, isn’t the destination. It’s a signal to follow.
DOI / Source: https://doi.org/10.18632/aging.206360
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.