Abstract
Background: Epigenetic clocks, derived from DNA methylation patterns, have revolutionized the estimation of biological age. However, the predictive accuracy of these clocks varies substantially across tissues, and the extent to which tissue-specific methylation signatures improve age prediction remains unclear. Methods: We performed a comparative analysis using publicly available DNA methylation datasets from five human tissues (blood, brain, liver, skeletal muscle, and placenta). A total of 1,247 samples from individuals aged 20–90 years were included. We constructed tissue-specific clock models using elastic-net regression on CpG sites previously identified in pan-tissue clocks [4,23] and compared their performance to a multi-tissue clock [23]. Predictive accuracy was assessed via median absolute error (MAE) and Pearson correlation between DNA methylation age and chronological age. Cross-validation was performed within each tissue. Results: Tissue-specific clocks consistently outperformed the pan-tissue clock within their target tissues, with MAE reductions ranging from 1.2 years (blood) to 4.7 years (brain). The brain-specific clock achieved an MAE of 2.1 years (r=0.96), while the pan-tissue clock yielded an MAE of 6.8 years (r=0.83) for brain samples. The liver clock showed the highest age acceleration variance, suggesting greater metabolic influence. Notably, CpG sites selected for tissue-specific clocks overlapped only 12–18% with pan-tissue sites, indicating distinct epigenetic aging programs. Conclusions: Tissue-specific epigenetic clocks substantially improve age prediction accuracy compared to universal clocks, highlighting the necessity of tailored models for organ-specific aging research. Our findings support the development of clinical tools that integrate tissue-specific methylation markers for precision geroscience and disease risk stratification.
Keywords
epigenetic clocks, DNA methylation, tissue-specific aging, biological age prediction, pan-tissue clock, precision geroscience, elastic-net regression, age acceleration