Aging clocks can be produced from large databases of near any form of biological data that changes with age via the use of machine learning techniques that identify weighted combinations of measures that correlate strongly with chronological age, remaining life expectancy, mortality risk, and so forth. The true goal of developing clock technology is to find a way to rapidly assess the quality of any new potential therapy to slow or reverse aspects of aging, and thus greatly accelerate research and development, focusing attention on the best possible approaches. Unfortunately no clock can be trusted to perform accurately in the face of a new approach to treat aging. A clock could underestimate, overestimate, or be unaffected by biological changes that do have meaningful effects on aging. It is unknowable in advance how this will play out, and the only way to determine what will happen is to first calibrate the clock against the therapy using lengthy studies, which defeats the point of the exercise. Nonetheless, researchers are engaged in producing as much data as possible on the way in which clocks behave, in an attempt to at least try to predict the degree of trust one should have in a giving clock assessing a given therapy.
Epigenetic clocks have emerged as potential biomarkers of aging due to their association with morbidity and mortality. They have recently gained traction as outcome measurements for anti-aging interventions. However, as the number of new epigenetic clocks continues to grow, different clocks may not agree on the effect of an intervention. This raises the question of when researchers can trust epigenetic clock results. In this study, we find that when extending previously published single-clock intervention studies to include additional clock models, no additional clocks reach statistical significance.
We then analyze clock behavior across positive control datasets (age-accelerating events) to propose three criteria that increase confidence in a result: (1) reliable principle component based clocks respond, (2) mortality-trained or pace-of-aging clocks respond, and (3) multiple clocks respond in the same direction. To explain why genuine effects would appear across multiple clocks, we perform a simulation study to demonstrate how reliable clocks are more sensitive to true effects, and use longitudinal control datasets to show that reliable clock changes are correlated with each other. We then show examples of interventions that do satisfy these criteria.
These findings provide a framework for interpreting epigenetic clock results in anti-aging intervention studies, and suggest that some previously published single-clock findings may warrant reexamination.
Link: https://doi.org/10.1111/acel.70756
View the full article at FightAging
