An improved epigenetic age estimation with TFMethyl Clock reveals DNA methylation changes during aging in transcription factor binding sites

Abstract Methylation-based epigenetic clocks are among the most accurate tools for predicting chronological age. Although DNA methylation (DNAm) at genomic CpG sites is linked to various regulatory mechanisms, the biological interpretability of epigenetic clocks remains surprisingly limited. One primary mechanism by which DNAm is thought to influence gene regulation is by modulating transcription factor binding activity. In this study, we examine established epigenetic clocks to assess the regulatory potential of their predictive CpGs during the aging process. Our analysis reveals that generally most CpG sites used by epigenetic clocks do not overlap known transcription factor binding sites (TFBSs), indicating that changes in TFBS dynamics may not account for prediction accuracy of these models. On the other hand, by identifying age-associated CpGs that overlap TFBSs, we identified transcription factors that may be involved in the aging process. Specifically, the TFBSs of ZBED1, NFE2, CEBPB, FOXP1, EGR1, SP1, PAX5, and MAZ were particularly enriched for age-associated CpGs, while RBPJ, NFIC, RELA, IKZF1, STAT3, and USF2 were significantly protected against methylation changes. By focusing on TFBS-associated CpGs, combined with additional feature selection and engineering steps, we developed an alternative, TFMethyl Clock model, outperforming several existing approaches. Target genes of model-selected, age-predictive CpGs are enriched in the interleukin-1b production and long- chain fatty acid metabolism pathways. In contrast, these CpGs themselves are enriched mainly at binding sites of NR2C2 TF. Furthermore, approximately three-fourths of the target genes downstream of age- predictive CpGs exhibit significant age-related changes, suggesting that our approach captures deeper insights into possible methylation-driven biological aging processes. Our findings demonstrate that incorporating regulatory loci into the design of epigenetic predictors may provide mechanistic insights into the aging process while maintaining or even improving the predictive power.