Some people like to say that age is just a number. Many scientists increasingly agree, arguing that DNA, protein, or other molecular measurements can tell a truer story about a person’s “biological” age than what’s on their birth certificate. Hundreds of so-called aging clocks developed in recent years reflect this idea, and clinical trials have started to use them to assess patients’ responses to putative antiaging treatments. Numerous wellness clinics now claim they can—for a steep price—deduce a person’s actual age, and multiple companies offer biological age testing for worried patients—and even for their dogs.
But scientists don’t agree on which aging clocks work best or how to verify their results. “We need to systematically evaluate them,” says bioinformatician Mahdi Moqri of Harvard Medical School. That’s the motivation for an unprecedented contest, funded by nonprofits and philanthropies and run by an organization called the Biomarkers of Aging Consortium, that is pitting hundreds of the clocks against one another for scientific honors—and some $300,000 in prize money.
Drawing on anonymized data and health information for 500 people, the competitors are vying to produce the most accurate predictions of chronological age, age at death, and “health span”—the time until onset of multiple age-related diseases. “The competition is put up or shut up” for clockmakers, says biogerontologist Steve Horvath of the cell rejuvenation firm Altos Labs, who developed a DNA-based aging assay that kick-started the field more than a decade ago.
This summer, the consortium released the results of the challenge’s first round, in which competitors had to estimate subjects’ chronological ages. Thirty-seven teams submitted more than 550 clocks; the competitors included veteran clockmakers, a company already marketing a test, and rank newcomers. “I thought of it more as a fun project,” says Jakob Träuble, a biotechnology Ph.D. student at the University of Cambridge. He and Stefan Jokiel, a physics master’s degree student at the Ludwig Maximilian University of Munich, came in third, winning a share of the $30,000 purse.
On 1 November, the consortium will announce the results of the $70,000 second round. It entails predicting the ages at death for the 15% to 20% of the subjects who have already died. Contestants in a final round, scheduled to wrap up next year, will face an even tougher challenge: predicting when people in the data set developed multiple age-related diseases.
More accurate, validated clocks could be a boon to clinical trials of potential antiaging treatments. And they could aid a much bigger contest: the XPRIZE Healthspan competition, launched last year, which will parcel out $101 million among scientists who by 2030 come up with strategies that can reverse age-related deterioration in muscles, the brain, and the immune system.
Signs of aging
Over the years researchers have hailed a variety of candidate aging measures. Some use just a single marker, such as the length of telomeres, protective caps at tips of chromosomes. These individual measurements don’t capture the complexity of aging, says life scientist Vadim Gladyshev of Harvard Medical School, a member of the Biomarkers of Aging Consortium. More promising, he says, are clocks that aggregate data, such as the abundances of different proteins in the blood or the patterns of chemical changes to DNA called methylation.
Methylation clocks have raced ahead. Roughly 30 million locations, known as CpG sites, in the human genome can be methylated. The modifications often turn genes on or off, and their distribution across the genome changes as people grow older.
Horvath, who is also part of the consortium, began to use DNA methylation as an aging clock by chance. In 2010, he and his gay brother gave saliva samples for a study that aimed to identify markers of sexual orientation in identical twins. Horvath also lent the study his statistical expertise and became a co-author. The researchers gauged the methylation status of more than 27,000 sites, but the data revealed nothing about sexual orientation. However, Horvath and colleagues realized that measuring methylation for just 88 CpG sites predicted the participants’ ages to within about 5 years, a result they published in 2011.
In 2013, he showed that by expanding the methylation survey to 353 CpG sites in DNA from a variety of tissues, including blood, he could estimate age more precisely, thus launching the eponymous Horvath clock. In 2019, a team led by Horvath and computational scientist Ake Lu, now also at Altos Labs, described an even more powerful clock. The aptly named GrimAge can make forecasts of when people will develop heart disease or cancer, or die.
Researchers have debuted a slew of other methylation-based clocks, including the Hannum clock, PhenoAge, DunedinPACE, and DNAm Age, that can predict chronological age, mortality, and aging rate. “The field has exploded,” Gladyshev says.
Most methylation clocks rely on the variety of artificial intelligence (AI) known as machine learning, which can tease out patterns in huge data sets. To predict chronological age, researchers first train the algorithm by feeding it methylation data from people whose ages are known. Once the clock has learned which methylation sites correlate with age, it can turn an individual’s methylation pattern into an estimate of their biological age—an indication of whether they are aging faster or slower than their peers. By training the AI on data about health and mortality, researchers can also predict time to death or onset of disease.
The competition is put up or shut up.
- Steve Horvath
- Altos Labs
Given that the clocks are relatively simple, “it’s amazing how well they work,” says Lucas Paulo de Lima Camillo, head of machine learning at the biotech Shift Bioscience, which focuses on cell rejuvenation therapies for aging.
But some researchers complain that the connection between methylation clocks and aging is unclear. Scientists don’t know whether changes in methylation patterns drive aging or aging-related illnesses or just correlate with them. That’s one reason clocks that make predictions based on levels of different proteins in the blood are catching on. “Proteins are much closer to the disease,” says molecular epidemiologist Anna Prizment of the University of Minnesota Twin Cities.
Recent studies showcase some of the powers of proteomic clocks. Using data on about 45,000 people enrolled in the UK Biobank, which collects molecular and health information, genetic epidemiologist Cornelia van Duijn of the University of Oxford and colleagues created a clock based on measurements of 204 blood proteins. In Nature Medicine in August, the team revealed that the disparity between a person’s proteomic age estimate and their actual age augured the onset of aging-related diseases and death. People whose predicted age is much greater than their actual age “are racing toward multiple diseases,” van Duijn says.
Compared with people whose biological and chronological ages align, for example, individuals whose disparity was in the top 5% were nearly three times more likely to develop Alzheimer’s disease and twice as likely to fall victim to kidney disease or die.
Proteomic measurements can also detect which organs in the body are breaking down fastest as people get older. In December 2023, neuroscientist Tony Wyss-Coray of Stanford University and colleagues reported in Nature that they could make aging clocks for individual organs based on blood proteins derived from those locations. The researchers ascribed a protein to a particular organ if its level there was at least four times the level in other organs. “We are confident that overall we are capturing something specific to an organ,” says Hamilton Oh, lead author on both studies, who just finished a Ph.D. in Wyss-Coray’s lab.
A bioRxiv preprint the team posted in June takes the proteomic clock analysis a step further, showing that the more “old” organs people have, the greater their odds of dying. Conversely, people whose brain and immune system are “young” for their age are at lower risk of death.
Ready for the clinic?
Researchers are already exploring clinical uses for these clocks. For instance, kidney transplant programs currently use a person’s calendar age to help determine whether they are eligible. Computational biologist Tamir Chandra of the Mayo Clinic and colleagues are testing whether a DNA methylation clock they developed better predicts transplant success. The clock “is amenable for any procedure where physicians take chronological age as an important factor in their thinking,” Chandra says.
Still, even many clock builders say the measures aren’t ready for routine clinical use. “I use GrimAge all the time. It’s a great research tool,” Horvath says. “Should the consumer use it? I’d say no.” His reasoning is that there are no treatments for people whose clocks indicate they are “older” than their chronological age. At best, doctors can recommend lifestyle changes such as better diet and more exercise. The issue comes down to, “Can we give people advice that is more specific than the advice we tell everyone?” van Duijn says.
Other researchers say a grim score on a clock like GrimAge can add more force to that advice, jolting patients into action. “Perhaps we could wake people up for lifestyle interventions,” Prizment says.
Scientists agree, though, that more accurate aging clocks are needed. That’s where the new contest comes in. Contestants can use any data they want to train their clocks, but their algorithms’ performance is tested against a standard set of blood data from the same racially diverse group of 500 people between the ages of 20 and 100.
The initial results were encouraging, Moqri says. The age predictions of the older Horvath and Hannum clocks were off by at least 4.8 years. The best clocks in the challenge shaved more than 2 years off that error. De Lima Camillo, whose methylation clock took second place with an error of 2.55 years, says he obtained more accurate results with deep learning, an AI approach that few published clocks have employed but that can infer more complex relationships between variables than machine learning can. Träuble and Jokiel also adopted deep learning, and incorporated more than 130,000 CpG sites.
Horvath, who has stopped trying to make new clocks but still closely follows the field, is willing to make another prediction: Aging clocks will continue to improve and will eventually become standard for medical checkups. “I will be surprised if it
didn’t happen.”
Time will tell, as it always does for anything connected to aging.