Research Paper Volume 8, Issue 2 pp 394—401

Figure 1. Epigenetic age-predictions correlate with mortality. (A) Scheme for the study design. Epigenetic age was estimated based on our previously published models with 99 CpGs, 3 CpGs, or individual CpGs thereof [8]. The offset of predicted age and chronological age was subsequently correlated with all-cause mortality in the Lothian Birth Cohort 1921 (LBC1921) and LBC1936. (B) The 99-CpG model and (C) the 3-CpG model were specifically trained for Illumina HumanMethylation450 BeadChip data using Hannum dataset (GSE40279) and subsequently validated using additional 12 DNAm datasets of blood samples. (D) The 99-CpG model was then applied on DNAm profiles of LBC1921 and the deviation of predicted and chronological age (Δ_age) was determined for each sample. Samples in the lowest (Q1) and highest quartiles (Q4) of Δ_age are depicted in navy and red, respectively. (E) Kaplan-Meier plots (K-M) of LBC1921 participants classified by respective quartiles indicate lower mortality for those with increased Δ_age.. (F, G) In analogy the same analysis was performed for the 3-CpG model, but there was no significant association between these age-predictions and mortality. (H, I) Alternatively, we tested association of DNAm at individual age-associated CpGs with mortality. To this end, linear models were trained for each CpG site using Hannum dataset and then applied to the LBC1921 cohort. cg17861230 (associated with PDE4C) reveals a significant association with mortality. The calculated Δ_age is subject to survival analysis with adjustment for chorological age and gender.