Blood-protein organ aging clocks predicted death less accurately than brain MRI, lung function, cognitive tests, and DNA methylation in their first head-to-head comparison
Blood-protein organ aging clocks predicted death less accurately than brain MRI, lung function, cognitive tests, and DNA methylation in their first head-to-head comparison
On October 1, the journal Aging Cell published a study by researchers at the University of Edinburgh. Using data from 861 Scots in the Lothian Birth Cohort 1936 (all born in 1936), the authors brought proteomic aging clocks for 11 organs together with other measures of biological age in a single statistical framework for the first time, to determine which best predicts who will die.
The idea behind organ clocks seemed compelling: in any given person, certain organs age faster than the rest, so those organs' ages should predict mortality risk more precisely than a general index. One of the creators of organ clocks has already warned that the market for such tests remains a "Wild West" without a shared standard, and recently the clocks were used to measure semaglutide's effect on heart and kidney age. But until now, organ clocks had never been compared directly against other biomarkers in the same individuals. Such a comparison requires a rare cohort with decades of data on DNA methylation, the proteome, brain MRI, and cognitive testing. The Lothian Birth Cohort 1936 turned out to be that cohort. For the first time, all of these measures were evaluated side by side against the strictest criterion: who among the participants died over 17 years of follow-up.
All 11 organ clocks were associated with mortality: one standard deviation of acceleration raised the risk by 16–43%, most strongly for the liver, immune system, and heart. Yet they proved weaker than their competitors. The GrimAge2 epigenetic clock, based on DNA methylation, raised risk by 62%; reduced brain volume on MRI raised it by 52%; impaired lung function, by 51%; lower cognitive test scores, by 46%. In a joint model that included all measures simultaneously, the organ clocks lost independent significance entirely. Only brain volume, white matter lesion volume, cognitive score, and walking speed retained predictive power; together, these raised the accuracy of mortality prediction nearly sevenfold compared to chronological age and sex alone. The organ clock effect sizes here match the original 2023 validation in order of magnitude (17–53%): the clocks performed as designed; they were simply less accurate than the competing methods.
The explanation lies in how GrimAge2 was built. These clocks were trained from the outset not on chronological age but directly on predicting death, with surrogates of nine blood proteins built into the formula, including GDF15 (a protein associated with cellular stress and inflammation) and cystatin C (a standard marker of kidney function). GrimAge2 already carries part of the same signal that organ clocks attempt to capture from scratch, which is why it outperforms them.
Separately, the authors tested all 9,703 individual blood proteins, adding adjustments step by step to distinguish proteins that predict death on their own from those whose signal depends on kidney function or lifestyle. After adjusting for age and sex, 368 proteins were significantly associated with mortality; GDF15 showed the strongest association. After further adjustment for kidney function, smoking, alcohol, and body weight, the list narrowed to 202 proteins, with cystatin C and collagen COL18A1 rising to second and third place. Cystatin C predictably weakened once its signal was adjusted for the very kidney function it measures. GDF15 alone remained in first place across every variant of the model.
For a market seeking a surrogate mortality endpoint to avoid years of waiting for geroprotector trial outcomes, an expensive and heavily marketed tool proved inferior in direct testing to a combination of brain MRI, lung function testing, cognitive assessment, and a well-established, cheaper DNA methylation analysis.