DNA methylation variability emerged as a distinct signal of aging with little overlap with mean methylation levels: a mortality biomarker combining both signals outperformed the Horvath, Hannum, and GrimAge clocks in two independent cohorts

DNA methylation variability emerged as a distinct signal of aging with little overlap with mean methylation levels: a mortality biomarker combining both signals outperformed the Horvath, Hannum, and GrimAge clocks in two independent cohorts

Researchers at Simon Fraser University, the US National Institute on Aging, and Dalhousie University examined blood DNA methylation in 1445 Canadians aged 45–85. Alongside the usual genomic sites where mean methylation levels change as health declines, they looked for sites where methylation variability changes among people of the same age. They found a largely separate set of sites. A biomarker combining both signals predicted all-cause mortality in two independent cohorts more accurately than either signal alone or existing aging clocks.

The paper was published on September 17 in Journal of Gerontology: Series A. The idea grew out of the team's 2024 study, which found that several physiological measures have an optimal middle range associated with health, with deviations in either direction associated with poorer health. Earlier observations suggested a similar principle: healthy, long-lived people have less variation in telomere length across cells than typical older adults, while increased methylation variability in tumors indicates genomic dysregulation. These three observations suggested that the stability of a measure itself carries information about health. The team therefore tested whether this also applies to methylation during normal aging.

The results revealed two largely nonoverlapping genomic patterns: changes in mean methylation levels occurred in genes involved in immunity and inflammation, with the strongest signal at SLC1A5, which encodes a glutamine transporter. Changes in variability occurred in CpG islands, regulatory regions at the start of genes.

“Variability may reflect an age-related decline in the maintenance of epigenetic stability... methylation variability reflects epigenetic dysregulation structured around specific regions rather than diffuse noise across the genome,” the authors write.

The researchers combined 789 sites into a single score. In a held-out subset of the Canadian sample, it predicted all-cause mortality more accurately than either signal alone and five established tools: the frailty index (developed by study coauthor and geriatrician Kenneth Rockwood), the Horvath and Hannum clocks, PhenoAge, and GrimAge v2. None of these tools showed a significant association with mortality in this sample. The advantage was also confirmed in an independent cohort, the Baltimore Longitudinal Study of Aging in the US (728 participants). The new biomarker was associated with smoking and the inflammatory markers C-reactive protein and interleukin-6. Its heritability was only about 41%: lifestyle and environment accounted for most of the variation, with genes contributing a smaller share.

Of the five age-related diseases examined, cancer, cardiovascular disease, chronic kidney disease, diabetes, and chronic obstructive pulmonary disease, the biomarker significantly predicted an increased risk only of chronic obstructive pulmonary disease. Testing five diseases required a stricter significance threshold to limit chance findings, and only this result met it.

A few weeks earlier, researchers independently found a similar effect in the Generation Scotland cohort: a set of sites measured in blood where methylation variability between people increases with age while mean methylation levels remain unchanged predicted mortality on its own. For a decade, aging clocks had read the methylome along a single dimension: mean methylation levels. This paper shows that the same data contain a second, independent dimension, variability. Including it improves mortality prediction, so leaving it out means a measurable loss of predictive accuracy.