Figure 1. Epigenetic clock age predictions in the Amboseli baboons.
Predicted ages are shown relative to true chronological ages for (A) females (Pearson’s r = 0.78, p=6.78×10−30, N = 142 samples) and (B) males (r = 0.86, p=5.49×10−41, N = 135 samples). Solid lines represent the best-fit line; dashed lines show the line for y = x. (C, D) Predictions for individuals with at least two samples in the data set (N = 30; 14 females and 16 males). In 26 of 30 cases (87%), samples collected later were correctly predicted to be from an older animal.
Figure 1—figure supplement 1. Characteristics of the RRBS data set.
(A) Proportion of the 458,504 evaluated CpG sites that overlapped annotated features of the Panu2 genome. (B) Proportion of annotated features in the Panu2 genome that overlapped at least one of the 458,504 evaluated CpG sites. (C) Distribution of mean DNA methylation levels for CpG sites within annotated features of the Panu2 genome. Each white box represents the interquartile range, with the median value depicted as a black horizontal bar. Whiskers extend to the most extreme values within 1.5× the interquartile range. As expected, CpG sites tended to be highly methylated genome-wide and have lower average methylation in promoters, enhancers, and CpG islands.
Figure 1—figure supplement 2. Comparison of clock performance across alternative values of alpha.
Alpha was set via grid search across possible values from 0.1 to 1, in steps of 0.1, and chosen based on the highest R2 value between predicted age and known chronological age (red lines). The blue lines show the median absolute difference between predicted and true age (lower is better) and exhibits roughly inverse behavior to R2. (A) For each clock generated with a different alpha value, the total number of CpG sites included in the clock is shown on top, and the number of clock sites that overlap the final clock used in this study (N = 573 sites, alpha = 0.1) is given in parentheses immediately below. (B, C) As in (A), but with results shown specifically for males (B) versus females (C).
Figure 1—figure supplement 3. Enrichment of the epigenetic clock CpG sites by genomic compartment.
The log2(odds ratio) of CpG sites in the epigenetic clock, relative to all 458,504 CpG sites initially evaluated, in (A) annotated genomic regions and (B) in loci with putative regulatory activity or in or near genes that are responsive to age or immune stimulation. Regions of regulatory activity were identified with the massively parallel reporter assay, mSTARR-Seq (Lea et al., 2018a), following a liftover from the human genome to the baboon genome to identify putatively orthologous coordinates. Age differentially methylated regions (DMR) and genes responsive to lipopolysaccharide (LPS) were previously identified from blood samples from the same baboon population (Lea et al., 2015a; Lea et al., 2018b). Two-sided Fisher’s exact tests were performed separately for epigenetic clock sites that increased (positive clock sites: N = 459) or decreased (negative clock sites: N = 134) in DNA methylation levels with age. See Supplementary file 2 for a complete list of the genomic locations of the 573 epigenetic clock sites. *p<0.05, ***p<0.005.
Figure 1—figure supplement 4. Association between age and DNA methylation level for individual clock CpG sites.
(A) Volcano plot of the effect size (βage) versus the –log10(p-value) of age effects on DNA methylation for males (blue) and females (red), based on estimates from a binomial mixed-effects model designed for bisulfite sequencing data (Lea et al., 2015a). Results for the 534 sites that could be modeled using this approach are shown. Other predictor variables in the model included a fixed effect for sample batch and a random effect that controlled for kinship (estimated via Queller and Goodnight’s r and multilocus microsatellite genotype data in the program coancestry
Wang, 2011). Dashed line corresponds to a nominal p-value of 0.01. (B) Age effects on DNA methylation estimated separately in males and females are highly correlated (R2 = 0.83, p=3.35×10−204). The dashed line indicates the y = x line. The solid black line indicates the best-fit line.
Figure 1—figure supplement 5. Comparison of the performance of the epigenetic clock to other predictors of chronological age.
Performance measures of age predictors are presented separately for females (A, C, E) and males (B, D, F) except for differential white blood cell counts (blood smears), where males and females were combined. Predictors are ordered in the same fashion in all panels (epigenetic clock to the left, and then following highest to lowest R2 in females). The breakpoint to define youthful versus aged animal BMI was 10 and 8 years old for females and males, respectively. (A, B) Adjusted R2 between predicted age and true chronological age. (C, D) Absolute difference between the y = x line (slope of one) and the slope of the best-fit line of predicted age as a function of true chronological age. This metric captures bias in age prediction estimates (values that are lower on the reverse-coded y-axis are more biased). (E, F) Median absolute difference (MAD) between each individual’s predicted age and true chronological age (values that are lower on the reverse-coded y-axis have higher MAD).