Do Aging Clocks Actually Work?

There are tests now that claim to measure how old you really are. Not the number on your birth certificate, but your biological age, the idea being that two people born on the same day can be aging at different speeds, and that a blood or saliva sample can tell them apart. It starts with a discovery from fifteen years ago, that chemical marks on your DNA shift as you get older. Since then, the clocks built on it have moved fast, from reading your age, to forecasting how long you have left, to scoring each organ, to scoring individual cell types from a single vial of blood. So what does one of these numbers actually measure, how good have they become, and would one handed to your doctor change anything about what happens to you?

0:00 – Can You Really Measure Your Age?
1:04 – The Cheek Swab That Started It
4:00 – Reading Death, Not Birthdays
5:45 – The Hidden Flaw
7:51 – Your Organs Age Differently
8:36 – 11 Ages, One Blood Draw
10:06 – 40 Clocks, One Vial
11:36 – The Missing Proof
14:15 – The Verdict

*Citations can be found on my website

Transcript

Copy Transcript Copied

SPEAKER_00 0:00

There are cats now that claim to measure how old you really are. Not the number on your birth certificate, your biological age. The idea that two people born on the same day can be aging at different speeds, and that this number catches the difference. They're starting to pop up everywhere. Longevity clinics seldom, supplement companies promise to lower them. People retest every few months, hoping to see if their habits are working, and they rest on something that's actually real. Fifteen years ago, researchers found that as you age chemical marks on your DNA shift in a predictable pattern. These marks are called methylation tags, tags that sit on your gene and change over your lifetime. Read the pattern, and you can estimate someone's age with real accuracy. That became the first aging clock. But the accuracy came with a catch built into how the first clocks were made. The story of the last decade is scientists trying to evolve it, turning a tool that read your age into one that could read your aging. Question is whether aging clocks actually work. It turns out that depends on what you're actually asking them to do. It started with a cheek swab. Around 2011, researchers noticed that if you looked at just a couple of methylation marks and a sample of saliva, you could guess a person's age within about five years, a handful of chemical tags, and you could estimate how long someone had been alive, closer to a forensic parlor trick than an actual medical tool, but the marks weren't random. They moved with time in a direction that you could read. Two years later, in 2013, that hint became a field. A scientist named Steve Corbeth took thousands of blood and tissue samples, each from a person whose age was already known, and handed them to an algorithm, a model that learns a pattern from enough examples. The task was narrow, find the methylation marks that move most reliably with age, and read them well enough to guess how old each person is. It found 353 of them, and it worked almost unreasonably well. Give the finished clock a fresh blood sample it had never seen, and it landed within about three and a half years of that person's real age. No birth certificate and not a gray hair, a chemical pattern on DNA read by a formula landing within a few years of someone's real age. That result has been reproduced many times and it became the foundation everything else is built on. And it wasn't alone. The same year a second teen built another clock from blood, 71 marks, and it landed just as close. Two groups, two clocks, the same finding. Pattern was real and it was readable. But look at what these clocks were asked to do. They were trained to predict chronological aid, the number of years since they were born. That was the answer key they were graded against, rewarded over and over again for getting as close as possible to the age on a birth certificate, which exposes the problem hiding inside the result. If a tool's entire job is to reproduce a number you already know, what has it learned about aging? Nothing that you didn't already know. You know your own age. Useful for a forensic case, maybe, but nearly useless in a doctor's office. The useful question is a different one. Two people share a birthday. At 60, one runs half marathons on no medication, the other is managing high blood pressure and two other conditions and feels a decade older than the calendar says. Chronological age hands them the same number, anyone can see they aren't aging at the same speed. That gap between the calendar and the body is what people actually want measured, not how many years you've lived, how much those years have cost you. That gap has a name. The clock reads one age, your true age sits underneath, and the distance between them is the deviation, the part everyone is actually after. Everything a clock is hoped to tell you lives in that deviation, not in how well it guesses your birth year, and a clock train to nail the birthday looks straight past that gap by design because a mark that reflects how fast you're wearing down is just noise to a model graded only on the calendar. The first clocks proved DNA carries a readable record of time. They were also, for the question people care about, aimed just to the sight of it. The term came out of the same lab that built the first clock. In 2018, Steve Horvath's group at UCLA tried something subversive. Train a clock to guess the birthday, and you teach it the birthday is the right answer, so any mark reflecting how worn down a person is gets thrown out as error. You train away the one thing you wanted instead of aiming at age, they aimed at the body's condition. They took nine ordinary blood work markers, inflammation, blood sugar, kidney and liver function, boiled them into a single measure of how a body was doing, and trained a methylation clock to read that off DNA. They called it pheno age, and it did what the first clocks couldn't. Among people the exact same age, the ones that flagged as older were more likely to get sick and die sooner. The first clocks told you the date, this one told you the toll. Now the DNA clock wasn't even the best predictor in the study. The plain cheap blood work it was built from, no DNA at all, predicted death better than the methylation version, trained to copy it. Something always gets lost, translating the body's real condition into molecular marks. And that loss matters later. Now, a year later, the same group pushed the idea as far as it goes. They trained a clock directly on death, on the actual time between a person's blood draw and the day they died, chemical marks on your DNA, standing in for damaged proteins in your blood, together predicting how close you are to the end. They named it Grim Age, and it delivered, followed across thousands of people for years. It was the strongest predictor of how long a person had left that anyone had built. The people it flagged as fast agers, the ones it read as older than their birthday, faced roughly twice the risk of dying as others their exact age, the slowest about half. And its makers were careful about what it was. One line in the primary Grimage paper says it out loud. Technically, the authors write, it's a mortality risk estimator. Metaphorically speaking, it estimates biological age, the most powerful clock yet built, and the people who built it drew the line in print between predicting death and measuring age. Then the simplest possible tests. Take one person's blood, split it into two tubes, run the same clock on both. You'd expect the same answer, but the two tubes came back different. For the original Horvat clock, the two halves of one sample disagreed by a median of three years, sometimes as much as eight. Same person, same moment, two numbers, most of a decade apart. You see a methylation clock reads specific spots on the DNA and asks how heavily you just tag, and most of those spots are noisy. Measure the same one twice and the reading wobbles on its own. One large analysis scored hundreds of thousands of them and found the average reliability sat in the range scientists formally called poor. The clocks work anyway by averaging hundreds of shaky signals, so the wobbly part cancels. Partly. Enough survives that the final number drifts run to run on the same blood, which matters because of what the number actually is. It's two things added together: your chronological age, which the clock nails, plus a deviation from it. That deviation is the whole point. The part that's supposed to mean something, and for the mortality clocks it does, but retest months later, and your birthday hasn't moved. So every bit of change is change in the deviation, and the deviation is where the noise lives. Run one unchanged sample twice, and the clock can land three years apart. A two-year drop between your visit could be your biology or the assay landing somewhere else. You can't tell which. And here's the part that should stop you. This noise wasn't uncovered years later. It was published in 2022. People using these outside of primary research settings were retesting and drawing conclusions from swings that may have been nothing but measurement error. The same 2022 work also fixed it. Rebuild the clocks to lean on the stable shared signal across thousands of spots instead of the shaky individual ones, and they get dramatically steadier, climbing to near perfect agreement between the two tubes of the same blood. Newer and steadier clocks exist now, but the steadier clocks didn't fix the deeper problem. The one under every methylation clock going back to the first. It gives you a single number for your whole body, and a body does not age all at once. You know this from watching people get older. One person's knees go first, another's heart, another stay sharp into their 90s and their lungs give out. Aging isn't one clock ticking, it's dozens running at different speeds in different parts of the same person. In 2023, researchers put numbers to that. Using scans and physical measures from tens of thousands of people, they built separate age estimates for the brain and seven body systems, and found that inside one person those ages pull apart. A heart can run old while the brain runs young, and the systems aging fastest predicted the diseases that someone went on to get, which naturally raised a question. If you want to know how one organ is doing, why read off DNA methylation at all? Methylation is a shadow of the machinery. The machinery is proteins, the molecules that do the body's actual work and organs shed them into the blood. Measure the right proteins, then you read an organ condition far more directly than any PAG on a gene. The clocks changed what they were made of. That same year, a team led by a Stanford scientist called Tony Weiss Corey did exactly that. They measured close to 5,000 proteins in the blood of more than 5,000 people, sorted them by which organ they came from, and built a separate clock for each of 11 organs. And the organ ages barely tracked each other. Knowing someone's heart age told you almost nothing about their brain age. Nearly one in five had a single organ aging much faster than the rest, and the fast heart group faced two and a half times the risk of heart failure. These clocks were still only trained to get chronological age, and they still saw disease coming as sharply as the Grim Age clock, a full-body protein version built from 200 markers held up across Britain, China, and Finland, predicting 18 diseases. Another group trained organ clocks directly on death and put it plainly, a chronic disease is what it looks like when one organ system ages faster than the rest of the youth. By 2025, run on nearly 45,000 people, an unusually old brain bred from a2 to blood carried as much Alzheimer's risk as inheriting a copy of the strongest gene known for the disease. And then this year the resolution dropped one level further, as far down as it currently goes, not the whole body, not the organ, the single cell type. A team again from Weiss Corey's group out of Stanford measured more than 7,000 proteins in the blood of over 60,000 people, traced each protein back to the cell type that released it, and built aging clocks for more than 40 of them. Neurons, asthrocytes, muscle cells, immune cells, each with its own age, and one ordinary blood draw. It's the sharpest version of everything the field had been circling. Aging isn't only uneven between organs, it's uneven between the cell types inside them and specific cells aging fast lined up with specific diseases. People whose muscle cells were aging fastest were nearly 13 times more likely to develop ALS, the disease that kills the nerve, driving muscle. And the signal showed up years before diagnosis. People whose astrocytes, a support cell in the brain, were aging fastest, carried Alzheimer's risk on the order of the main genetic risk factor. And the reverse is the part that stays with you here. Among people carrying two copies of that Alzheimer's gene, the highest risk group there is, not one with youthful astrocytes developed the disease. The same aging at predictable illness run backwards look like protection. It reaches past the brain too. Aging in the cells lining the airways flags smokers at markedly higher lung cancer risk on top of smoking itself, and across every cause and death, the best cell level predictor of dying was the age of a person's muscle cells. Fifteen years ago, one number for your whole body, and all it could reliably tell you was your birthday. Today, a map of how fast 40 kinds of cells are wearing down from a single vial of blood, naming diseases, before they arise. Which brings us to a major question. Not whether this is scientifically interesting, whether one of these clocks handed to your doctor would change what happens to you. That's a higher bar, and it's worth being exact about what clearing it takes. A test earns a place in medicine by clearing four things. It has to give the same answer twice on the same sample. It has to tell you something you don't already know that predicts what actually happens to you, not just restate your aid. It has to move when your body changes. And the hard one, moving the number has to change your fate. The number falling has to mean disease is being prevented, not just that a number fell. Hold the clocks against those four. Reproducibility was the methylation clock's original sin, and it's largely fixed now. In the rebuilt methylation clocks and the newer protein clocks that hold steady across labs and machines, predicting your future better than the calendar is what the best of these clocks plainly do. Off a single reading, one deviation, not a trend line, on those two counts, they work. Now back to the third point, the clocks having to move when your body changes. In 2023, a control trial had a group of people cut their calories for two years and track them with a clock built to measure the pace of aging rather than a fixed age. That clock slowed, the first hard evidence from an actual experiment that the speed of aging can be moved at all. But the trial is honest about its ceiling, and that ceiling is the fourth bar, the one nothing has cleared. The clock moved, no one has yet shown that when it moves, the person gets fewer diseases or lives longer, or that they even feel better. In other words, if they had symptoms relevant to the system in question whose clocks had moved, did they improve at all? Moving the number and changing your fate aren't yet the same thing. At least we can't say that with confidence. The study that would tie them together, following people for decades, hasn't been published yet. So are these clocks useful now? For reading your risk off a single measurement? Yes. And more so every year. The neweth clocks tell you not just that you're aging fast, but where, which organ, which cell type, exactly what a doctor would need to act on. But as something you treat, a number you drive down over months and trust you bought yourself time, no, that proof doesn't exist yet. Will it soon? The reproducibility problem is solved and strip down clocks, a brain clock from 10 proteins, a whole body one from 20, already keep most of their accuracy while being cheap enough to run scale. What's missing isn't the technology anymore, it's time. The trials that prove moving a clock changes and outcome take years by definition, and they're only just now beginning, which lets us answer the question we open with. Do aging clocks work? It depends entirely on what you ask them to do. Ask them to read your age, and they work almost eerily well, which is the trap because you already knew your age. Ask them to predict your risk of disease and death better than the calendar, and the good ones genuinely do. Ask them to measure your own personal speed of aging, and they're not quite there. The snapshot is real, a single deviation that genuinely carries signal, and they compare it to a crowd. What we can't yet say is that they are a speedometer, a number you watch move on yourself and trust, because that movement is exactly where the noise and the missing proof both sit. But the direction is not in doubt. In 15 years, these clocks went from a parlor trick that told you only what you already knew to a blood test that reads the aging of individual cell types and named the diseases coming for you. They still don't do the one thing the person retesting every few months wants most to watch a number and know that they're winning. The science is moving toward exactly that and moving fast. The clocks that couldn't tell you anything you didn't know have quietly become clocks that can tell you things you'd rather not. Thank you so much for joining me on today's episode. If you learned something today, please like, comment, and subscribe wherever it is you're listening or watching. Until next time, stay healthy.