A study of 409 thousand UK Biobank participants finds that accelerated organ aging and the risk of death from organ disease are distinct, genetically unrelated measures
A study of 409 thousand UK Biobank participants finds that accelerated organ aging and the risk of death from organ disease are distinct, genetically unrelated measures
A Research Square preprint published on 30 September 2026 describes how a University of Hong Kong team built aging clocks for 14 individual organs using data from UK Biobank, the United Kingdom’s largest biobank. Blood metabolomic and proteomic measurements were available for 26 thousand participants, and genetic data for 409 thousand. The main genetic findings were also tested in an independent group of 383 thousand people. For each organ, the researchers trained three separate models to predict chronological age, the risk of death from diseases of that organ, and the risk of death from any cause. They then compared these estimates within individuals.
Ranking people by how quickly their liver is aging produces a largely different list from ranking them by their risk of dying from liver disease. Across organs, the overlap between the groups flagged by the two measures was no more than 27%, and for some organs it was just 12%.
The prostate provides the clearest example. Clocks trained to predict death from any cause failed to detect prostate cancer risk: their estimates showed no difference from average risk. Prostate cancer accounts for only 8.4% of deaths in the sample, so its signal is lost against the background of heart attacks and strokes. Clocks trained specifically to predict death from prostate disease, however, showed a strong association with the same cancer: people with the highest scores had 2.6 times the risk. For benign, nonfatal prostate enlargement, the third measure, age acceleration, showed the stronger association, with a 1.7-fold increase in risk. The three models answered three different questions about the same organ and produced three different answers.
The same pattern appeared in the brain, pancreas, lungs, and kidneys. Clocks trained to predict death from diseases of these organs predicted Alzheimer’s disease, type 2 diabetes, chronic obstructive pulmonary disease, and chronic kidney disease, respectively. People with high scores had 7 to 8 times the risk of those with average scores. All three models were trained on the same set of molecules, but each had a different prediction target. The authors conducted genome-wide searches for associations with all the measures. Age acceleration and mortality risk shared few genetic associations: of 172 independent loci, 114 were significantly associated with only one of the three measures. Age acceleration was more often associated with the FADS gene cluster, which is involved in fatty acid metabolism. Organ-specific mortality risk was associated with proteins that mark tissue damage (GDF15, NPPB, HAVCR1, and others) and with APOE, the Alzheimer’s risk gene, which accounted for almost all of the genetic signal for the brain clock’s mortality measure.
Genetic variants are fixed at birth and cannot result from a disease that has already begun. This separation at the DNA level therefore provides much stronger evidence than differences between scores alone. Mendelian randomization uses these inherited variants to test whether one factor causes another rather than merely accompanying it. The analysis also indicated the direction of the effects: excess weight causally increased mortality scores for thirteen organs, while vigorous physical activity lowered these scores for eleven organs. Neither affected the age acceleration measure.
The same instability had already appeared in clocks retrained separately by sex: in men, the associations between the liver and spleen clock scores and mortality risk reversed direction.
If organ-specific mortality risk and the rate of molecular aging are distinct dimensions with different genetic bases, a geroprotector trial that assesses slowing of aging solely through age acceleration could entirely miss an effect on the actual risk of dying from a particular disease. Conversely, measuring mortality risk alone could miss an effect on molecular aging.