What does an epigenetic clock actually measure?

DNA methylation is a chemical tag on DNA that determines whether genes are switched on or off, without altering the genetic code itself. At thousands of specific sites across the genome, these tags change in a predictable way as we age. An epigenetic clock is a mathematical model that converts those patterns of change into an age estimate expressed in years: the so-called methylation age.

That methylation age can come out higher or lower than your actual calendar age. The difference between the two is called age acceleration: being biologically older than your passport suggests, or younger. The first generation of clocks, such as the Horvath clock (353 measurement sites, more than 8,000 samples from 51 tissues and cell types) and the Hannum clock, were trained on calendar age. They do a good job of estimating how old someone is, but are less effective at predicting disease or early death.

Newer clocks such as PhenoAge and GrimAge were trained on clinical measures of health and mortality. PhenoAge predicts all-cause mortality, cancer and physical decline better than earlier versions. GrimAge predicts how long someone will live and when cardiovascular disease or cancer will occur, with statistically strong associations in cohorts of thousands of people. A systematic review of 156 publications found that higher age acceleration is associated with greater mortality, cardiovascular disease, cancer and diabetes. These are associations, however, not proven causal relationships.

A striking peripheral finding: in stem cells that have been reprogrammed to an embryonic state, the clock reads almost zero years. And cancer tissues show an average acceleration of 36 years compared with surrounding tissue. This makes the clock scientifically intriguing, but what these extremes mean biologically is still unclear.

There are also clear limitations. Why exactly those specific measurement sites change with age remains largely unknown. Most clocks were developed using Western or European populations and therefore fit less well in other population groups. Alongside DNA methylation clocks, models based on protein profiles, gene activity and metabolic products are now being developed, but these have not yet been nearly as extensively validated.