Inherited genetic variants shape methylation at 63% of the DNA sites that 13 epigenetic clocks use to assess biological age
Inherited genetic variants shape methylation at 63% of the DNA sites that 13 epigenetic clocks use to assess biological age
On September 15, a preprint from Rice University appeared on arXiv. The author cross-referenced the methylation sites of 13 published aging clocks against the GoDMC catalog, which maps genetic variants that predictably alter blood DNA methylation independently of age. Sites included in the clocks carry such a genetic footprint considerably more often than randomly selected unused sites: 62.7% versus 39.6%. For the 16 sites that appear in five or more clocks simultaneously, the gap is sharper still: 15 out of 16, versus 4 out of 16 in the control set.
Epigenetic clocks estimate age from methylation levels (chemical marks on DNA) at selected genomic sites. A site enters the model because it statistically helps predict age in the training sample, not because its connection to aging has been demonstrated. For some of these sites, an inherited DNA variant sets the methylation level stably throughout life, regardless of age, meaning that part of the clock's reading was determined at birth. When two people of the same chronological age receive different clock readings, the question is whether one actually aged faster, or whether that person simply carried a different baseline methylation at the site because of genotype.
Sean Lim compared the methylation sites of 13 clocks, from Horvath's 2013 clock through GrimAge and DunedinPACE, against GoDMC and the same control set of unused sites. The share of sites carrying a genetic footprint was 62.7% in the clocks versus 39.6% in controls (odds ratio 2.57). The effect held for 11 of the 13 clocks, and across all five variant types. The exception was GrimAge: the clock whose long-term trajectory most accurately predicts mortality showed the most modest enrichment despite having the largest site set.
For these same 16 sites, Lim ran the lead variants through AlphaGenome, a DeepMind model that predicts a variant's effect on gene activity, chromatin accessibility, and regulatory protein binding from DNA sequence alone, without requiring new experiments. By this measure, the variants did not differ statistically from ordinary blood methylation variants. One stood out: rs10190186, a variant in FHL2, a gene encoding an adaptor protein in cardiac and skeletal muscle whose mutations are linked to hereditary dilated cardiomyopathy, a condition in which the heart stretches and weakens. This variant drives the primary genetic signal for two adjacent methylation sites at once: its allele raises methylation at both, in the same direction that methylation already increases with age, and, according to AlphaGenome, increases chromatin accessibility and RNA output of FHL2 itself.
"This is a reason to investigate the locus, not a claim that we have discovered 'FHL2 aging' or that this variant accelerates organismal aging," Lim clarifies.
GrimAge and DunedinPACE already serve as measurement instruments in geroprotector trials, including home-based rapamycin trials and the reanalysis of CALERIE, a multi-year caloric restriction trial. The field is already seeking alternatives: in 2024, another group proposed CausAge, DamAge, and AdaptAge in Nature Aging, clocks that select sites on the basis of a demonstrated causal contribution to aging. Lim's work is the first to quantify how large this same effect is in the older, widely used clocks.
How much genotype shifts a clock's reading can only be determined in cohorts with simultaneous genotype and methylation data. Those same cohorts can test whether rs10190186 alters GrimAge or DunedinPACE readings. For geroprotector trials, this is a practical question: what fraction of the difference in "biological age" between participants reflects the intervention, and what fraction reflects an inherited background that was fixed before the intervention began.