A statistical analysis of chemical tags on DNA may help unify disparate theories of aging
Ingrid Wickelgren
Originally published in Quanta Magazine on August 17, 2022. Republished here with the permission of the Simons Foundation, copyright holder.
This time a year ago, Steve Horvath was looking for pangolin DNA. The ancient scaly anteater would be a first for his collection, which was then about 200 mammals strong. “I didn’t have any of that order, which is why I desperately wanted them,” he recalled.
Since the summer of 2017, Horvath, who until recently was an anti-aging researcher at the University of California, Los Angeles, has spent as much as 10 hours a day penning emails to zoos, museums, aquariums and laboratories. He has attended talks on bats and Tasmanian devils to meet their keepers. He has reached out to the far corners of the world, begging for the DNA of flying foxes, vervet monkeys, minipigs and bowhead whales.
With that vast menagerie of samples, he has built computational clocks that can calculate the age of creatures as diverse as shrews, koalas, zebras, pigs and “every whale you can name,” he said, just by looking at their DNA. But those were merely steps toward the completion of Horvath’s ambitious moonshot of a project: a universal clock that could measure the biological age of any mammal.
Measuring age might seem to be no harder than using the nearest clock or calendar. But chronological age is an imperfect metric since some individuals and tissues show the effects of age more rapidly than others. For decades, scientists have searched for an objective and versatile way to measure biological aging, the changes in healthy function over time. “You want to have a biomarker that accurately measures ages in many different tissues and cell types,” said Horvath, who left UCLA this year to become a principal investigator at Altos Labs, a biotechnology startup working toward the rejuvenation of cells.
Horvath and his colleagues completed a version of the pan-mammalian clock earlier this year. Now he and other researchers are hoping to identify the molecular processes common to diverse creatures that make such a clock possible. Understanding why clocks like this one work, Horvath believes, could help lead us to what he calls “the true root cause of aging.”
His clocks are based on analyses of the chemical tags called methyl groups that hang on DNA like charms on a bracelet and help control gene activity. They are products of epigenetics (literally, “above genetics”), the field that studies heritable information not written in the genetic code. A dozen years ago, Horvath and his colleagues began applying their know-how to building the clocks, first to assess the age of DNA from saliva, and later to determine the age of blood, liver and other individual tissues.
Many biologists were skeptical at first because the clocks were rooted in statistics rather than an understanding of biomolecular mechanisms. Yet the accuracy of the clocks stood up to tests and sent ripples through the biomedical community. Scientists began using Horvath clocks in their research to measure the aging of cells because the clocks were better arbiters of the state of the body and the risk of disease than chronological age. “Epigenetic clocks are closer to the actual process of aging than any other biomarkers,” said Vadim Gladyshev, a biochemist at Brigham and Women’s Hospital and Harvard Medical School who studies cancer and aging. Now the clocks are leading some scientists to rethink their ideas about what aging is, as well as its connection to diseases.
“I now have collaborators that work a lot in breast cancer and [are] starting to think about, ‘If you have advanced biological aging, is that also informative for breast cancer?’” said Sara Hägg, a molecular epidemiologist at the Karolinska Institute in Stockholm, Sweden. If the clocks can usefully illuminate how to stop the aging process from triggering age-related disorders, she added, “we could prevent not just one disease but many.”
Seeing a Signal
Time and again in past decades, biological researchers thought a clock for aging was within reach. For example, they learned in the early 1960s that cells growing in culture aren’t immortal but instead die after only 40-60 rounds of replication, which suggested that cells harbor a kind of aging clock. In 1982, researchers thought they might have found the clock’s mechanism when they isolated telomeres, DNA-protein complexes at the ends of chromosomes that shorten each time a cell divides; when telomeres become critically short, cells die.
But telomeres did not pan out as an aging clock. The correlation of telomere length with age and mortality is weak in humans and nonexistent in some other species. “Telomere [length] does not actually track age. It just tracks cell proliferation,” said Ken Raj, a principal investigator at Altos Labs.
As an alternative to telomere length, in 2009 Horvath began working on a clock based on the RNA transcripts of a cell’s active genes, the templates for the proteins that define a cell and allow it to function. For the next two years he tried to make that approach work, to no avail: The transcription data was just too noisy.
But in 2010, Horvath answered a request for help from a colleague at UCLA. To study possible connections between sexual orientation and epigenetics, the researcher was collecting saliva from identical twins who differed in sexual orientation, with the hypothesis that the DNA in their saliva cells might reveal some consistent differences in methylation patterns. Horvath’s twin brother is gay; Horvath is heterosexual. They supplied their spit.
The study’s analysis looked at sites in the DNA where cytosine bases are located and checked which of them were methylated. (Cytosines are the only bases to which methyl groups attach.) A recently introduced lab-on-a chip technology made it easy to test tens of thousands of cytosine sites in each cell’s DNA. When the colleague needed a statistician to analyze the data, Horvath volunteered his services.
He did not find what they were looking for. “There was no signal whatsoever for homosexuality,” Horvath said. “But because the data were on my computer, I said, let me look at aging effects,” since the ages of the twins in the study spanned decades.
Until then, Horvath had steered clear of epigenetic data in his own research. The relationship of methylation patterns to gene expression is messy and indirect, and it had seemed unlikely to show much useful connection to aging. But now that he had this windfall of epigenetic data at his disposal, there seemed no harm in looking.
Horvath started matching the methylation patterns with the ages of the twins. In any one saliva sample — or any sample from any tissue — not all the cells will show the same methylation pattern. But the proportion of cells that are methylated at a given cytosine in DNA can be measured. In one sample, for instance, 40% of cells might be methylated at a certain position; in another, that proportion might be 45% or 60%.
To his surprise, Horvath found a strong correlation between age and the proportion of cells with methylation, even when he looked at just one site in the DNA. Looking at more locations boosted the accuracy.
“This changed everything for me,” he said. “Once I looked at the signal for aging, it blew me away.”
Horvath built a model that predicted a person’s age from the methylation status of about 300 cytosines across millions of cells in a saliva sample. “You spit in a cup, and we can measure your age,” he said.
Soon he was building epigenetic clock models for evaluating the biological ages of blood, liver, brain and various other tissues. First, he measured the proportions of cells in each sample that showed methylation at specific sites. From that data, he created profiles of the tissues that described the proportions of cells methylated at each site.
To build a clock, he fed a computer thousands of epigenetic profiles along with the age of each tissue profiled. Through machine learning, the computer linked ages to methylation patterns. It also narrowed down the number of sites needed to predict age. The computer then weighted the significance of each site’s methylation in its calculations to create the best predictive formula for age, which Horvath tested on a separate set of samples of known ages.
Within two years, he had combined their separate tissue aging clocks into one formula for a “pan-tissue” clock, published in 2013. The pan-tissue clock was “the game changer,” said Daniel Belsky, an epidemiologist at the Columbia Mailman School of Public Health. The formula applied to any and all human cells containing DNA. And anyone could use it — Horvath put the software on the internet. By uploading their own methylation data, biologists could find out how much of a toll time had taken on cells in their samples.
Quantifying Decline
Horvath’s pan-tissue clock was miraculously accurate at predicting chronological age. It also seemed to reflect important underlying differences between chronological and biological age. Researchers discovered that when the epigenetic clock estimated that someone’s age was greater than their chronological age, the person faced a higher risk of disease and death. When the clock estimated that someone was younger, their risk went down. Even though the epigenetic clock was derived from chronological age data, its algorithm predicted mortality better than age did.
So in late 2014, Horvath set out to track biological age explicitly. He and his colleagues, including Morgan Levine (a pathology researcher at Yale University who recently joined Altos Labs) and Luigi Ferrucci of the National Institute on Aging, trained an algorithm on a composite measure that included chronological age as well as the results of nine blood chemistry tests that foretell disease and mortality. The data came from the blood of more than 9,900 adults in the National Health and Nutrition Examination Survey. The resulting clock, DNAm PhenoAge, published in 2018, predicted overall mortality and the risk of cardiovascular disease, lung disease, cancer and diabetes, among other outcomes. A year later, Horvath and a team led by Ake T. Lu of UCLA released an even more precise predictor of time to death, GrimAge, which looked at a person’s sex, chronological age, smoking history and blood-protein mortality markers.
A new tool from Belsky and his colleagues, introduced in 2020 and updated earlier this year, acts as an aging speedometer. In creating their Pace of Aging biomarker, they quantified the rate of change in 19 markers of organ function at four ages, compiled them into a single index, and modeled it with methylation. “We’re actually quantifying the ongoing process of age-related decline and system integrity,” Belsky said. Those who age faster by this measure die younger, he said, adding that it predicts mortality about as well as GrimAge and may forecast stroke and dementia even better.
Age-Old Question
In 2017, representatives of the Paul G. Allen Family Foundation approached Horvath after one of his talks. They liked his work and suggested he dream big, because the foundation supports high-risk endeavors. Find a project that nobody else would fund, they said.
It didn’t take Horvath long to suggest an aging clock that would apply to all vertebrates. The proposal passed — it was outlandish enough — but as Horvath came to realize the magnitude of what it would involve, the plan morphed into a relatively restrained clock for all mammals.
By January 2021, Horvath had methylation data from 128 mammalian species, and he posted his clock on the preprint server biorxiv.org. “The same math formula, the same cytosines for a mouse or a rat or a dog or a pig. We can measure aging in all of these species,” Horvath said. Still, he scoured the globe for more.
By late summer of last year, Horvath was in contact with Darren Pietersen, a pangolin expert at the Tikki Hywood Foundation in Harare, Zimbabwe, offering him supplies for collecting data from pangolins and several other species. No one even knew for sure how long pangolins live. Some official accounts said 15 to 20 years, but Pietersen thought at least some types live longer. “The one animal that we aged recently was about 34 years old (although with a fairly wide margin of error),” he wrote.
From the supplied tissue data, Horvath built a pangolin clock, one more life span timer to add to his collection. “You want a pig clock, I have a pig clock. I have a clock for kangaroos and for elephants,” Horvath said. Each species-specific clock was a boon for scientists in the field. Elephant researchers, for instance, wanted the elephant clock so that they could ascertain the age structure of wild populations to aid conservation efforts.
But a clock that merges all of them can help answer a more basic question: What is aging? One view is that your body ages like your shoes, gradually fading and falling apart from wear. But the successful predictions from the pan-mammalian clock imply that something also causes cells to fail on a certain timetable, perhaps because of developmental genes that do not switch off when their work is done. “This suggests an element of determinism in aging,” said Raj, one of the clock’s more than 100 builders.
Data from methylation clocks suggests that aging starts very early, long before the body breaks down. In a 2021 paper, Gladyshev and his colleagues describe a methylation clock that dates stages of mammalian development. They found that during early embryogenesis in mice, a rejuvenation of sorts dials back the embryo’s age to zero. Biological aging then proceeds apace, even though human children are arguably growing stronger, not weaker, during this time, and mortality in humans declines until about age 9. “That is to me very profound because it nails this question of aging down to a process that is inextricable from the process of development,” Raj said.
Two recent studies of the naked mole rat, a rodent with an improbably long 37-year life span, show that the animal ages epigenetically, even though its chances of dying do not rise with chronological age. “I think mortality rate is not the best measure of aging,” said Gladyshev, who led one of the studies. “Aging is an unavoidable consequence of being alive.”
Aging still reflects the effects of experience, behavior and the environment, of course. Smoking and sun exposure, for example, can accelerate it, as measured by methylation and other markers, and exercise or a low-calorie diet can stall it. In work published last March, an epigenetic clock tailored to marmots showed that hibernation slows down aging, and a paper published last week showed that the same is true for bats. A clock made for rhesus macaques suggests that in 2017 Hurricane Maria sped up aging in a colony of the monkeys on an island off the coast of Puerto Rico.
Original Sin
No one knows entirely why the clocks work. Some but not all of the genes and molecular pathways involved have been identified, and researchers are still learning how methylation patterns affect the behaviors and health of cells, tissues and organs. “It comes back to what I call ‘the original sin of the construction,’” Horvath said. “It is based on a [statistical] regression model that is at some level agnostic to biology.”
To atone for this sin, Raj and Horvath have begun seeking biological correlates for epigenetic aging. Perturbations of the biochemical pathways the body uses to sense its need for nutrients slow aging, they recently discovered, in accordance with the effects of calorie-restricted diets on aging. Derailing the workings of mitochondria speeds it up. The clock also tracks the maturation of stem cells. If these processes are connected at a deeper level, epigenetic clocks may reveal unifying mechanisms for aging, the authors wrote in a 2022 paper in Nature Aging.
What those unifying mechanisms might be or why methylation status tracks aging so well, however, is yet to be fully determined. “We don’t really know if epigenetic clocks are causally linked with aging,” Hägg said.
Even if they are, epigenetic clocks are almost certainly measuring only part of what occurs during aging, said Matt Kaeberlein, a researcher at the University of Washington School of Medicine in Seattle who studies the biology of aging. “Whether they are actually measuring more than a single dimension of biological age is not clear,” he said. “This is part of the problem here — the conflation of epigenetic age with biological age. Those are not equivalent in my view.”
Raj speculates that the methylation changes reflect a loss of cellular identity with age. All cells in the body have the same DNA, so what makes a liver cell a liver cell and a heart cell a heart cell is the pattern of gene expression, which epigenetics controls. As changes in methylation accumulate with age, some of those controls might be lost, replaced by re-emerging developmental programs that should be switched off, Raj suggests.
Although methylation clocks may be the most accurate monitors of biological age for now, some studies suggest there is room for improvement. A more precise predictor might combine quantifiable cellular properties — say, protein, metabolite or gene expression levels — with physiological signals and an index of frailty. “We can measure so many things in a human being now,” Hägg said. “The more of these things that you quantify, the more accurately you will capture your aging.”
Methylation clocks also have limited clinical uses, Hägg warned. People can buy a readout of their biological age from various commercial sources, but not only are the results often inconsistent, they lack clinical relevance because the clocks were meant for group-level analyses in research. “They are not built to be predictive at an individual level,” she said.
And even if someone does change their lifestyle in a way that lowers their biological age as measured by these clocks, will they have a longer life or a lower risk of disease? “We don’t know that yet,” Kaeberlein said.
Horvath is now preparing a paper about his pan-mammalian clock for submission to a journal. Although he has seemingly reached his goal, gaps in his collection still nag at him. In May, he corresponded with senior curators at a museum in Australia about acquiring tissue from marsupial moles, a small, largely blind creature that spends most of its time underground. “We already generated data from 348 mammalian species, but we would like to add more,” he said.
When Horvath proposed this project, he set out to analyze 30 species, but 30 soon became 50, then 100, then more than three times that. “I need to pace myself,” he said, “because I have this impulse to collect more.”