Genome-wide methylation profiles reveal quantitative views of human aging rates

Abstract

The ability to measure human aging from molecular profiles has practical implications in many fields, including disease prevention and treatment, forensics, and extension of life. Although chronological age has been linked to changes in DNA methylation, the methylome has not yet been used to measure and compare human aging rates. Here, we build a quantitative model of aging using measurements at more than 450,000 CpG markers from the whole blood of 656 human individuals, aged 19 to 101. This model measures the rate at which an individual's methylome ages, which we show is impacted by gender and genetic variants. We also show that differences in aging rates help explain epigenetic drift and are reflected in the transcriptome. Moreover, we show how our aging model is upheld in other human tissues and reveals an advanced aging rate in tumor tissue. Our model highlights specific components of the aging process and provides a quantitative readout for studying the role of methylation in age-related disease.

Figure 1. Global data on the aging methylome

(A) A density plot of methylation fraction values for the marker cg16867657, separated by young (green) and old (blue) individuals. (B) A histogram of the age distribution for all individuals. (C) A heat map of the age-associated methylation markers, sorted by the magnitude of association (regression coefficient). The individuals are ordered youngest to oldest. A specific example of an age-associated region is shown in Figure S1. Annotation coincidence tables are also provided as Tables S1, S2.

Figure 2. Model predictions and clinical variables

(A) A flow chart of the data (green boxes) and analyses (red ovals) used to generate aging predictions (blue boxes). (B) A comparision of predicted and actual ages for all individuals based on the aging model. (C) Out-of-sample predictions for individuals in the validation cohort. (D) Apparent methylomic aging rate (AMAR) for each individual, based on the aging model without clinical variables. The distribution of aging rates shows faster aging for men than women. A table of the markers used in the aging model are provided as Table S3.

Figure 3. Genetic effects on methylomic aging

(A) We surveyed genomic variants for an association with age-associated methylation markers. 8 genetic variants, corresponding to 14 meQTLs, were chosen for validation. Of these, 7 were significant in the validation cohort and two showed an association with AMAR. (B) A plot of the trend between the methylation marker cg27367526 (STEAP2) and age. The state of variant rs42663 (GTPBP10) causes an offset in this relationship. (C) A second example for cg18404041 and rs2230534 (ITIH1, NEK4). A table of confirmed genetic associations is provided as Table S4.

Figure 4. Multi-tissue support

(A) Predictions of age made by the full aging model on the TCGA control samples. There is a high correlation between chronological and predicted age, but each tissue has a different linear intercept and slope. (B) After adjusting the intercept and slope of each tissue, the error of the model is similar to that of the original whole blood data. Age predictions made on cancer samples are presented as Figure S2. (C) Age predictions made on matched normal and tumor samples from TCGA. Predictions are adjusted for the linear offset of the parent tissue (breast, kidney, lung, skin). (D) Tumor samples show a significant increase in AMAR.

Figure 5. Age associations for methylation fraction and deviance

(A) Methylation fraction values for are shown for the marker cg24724428. Over any subset of the cohort, we consider two group methylation statistics: the mean and variance. Marker variance is a measure of the mean methylation deviance, which is defined as the squared difference between an individual’s methylation fraction and their expected methylation fraction. (B) A density plot showing the change in mean methylation with age for the marker cg24724428. Young and old groups are based on the top and bottom 10%. (C) A histogram of the significance of association between the methylation fraction of all markers and age. P-values are signed such that positive values represent an increase of methylation with age. Markers which exceeded the FDR < 0.05 threshold are grouped into the most extreme bins. (D) A density plot showing the change in methylation deviance with age for the marker cg24724428. (E) A histogram in the same form as ‘D’, of the significance of association between the methylation deviance of all markers and age. Aging trends are mapped for CpG islands in Figure S3.

Figure 6. Methylome-wide trends with age

(A) Aggregate regression lines for all methylation markers that increased with age (red) and decreased with age (blue). The darkest color represents the median regression line and the bounds represent the 25% and 75% quantile. Both increasing and decreasing markers trend toward moderate methylation fraction values. (B) An entropy aging rate was calculated as the mean Shannon entropy of age-associated methylation markers divided by chronological age. This was strongly associated with AMAR.

Figure 7. Transcription aging model

(A) We built an aging model using mRNA expression data for genes which showed an aging trend in the methylome. It’s standard error (RMSE = 7.22 years) is increased due to the rounding of ages to the nearest five-year interval in the dataset. (B) Similar to the methylome, the transcriptome shows an increased aging rate for men as compared to women (P < 10⁻⁴).