DNA methylation and aging : a longitudinal study of old Swedish twins

Abstract

DNA methylation is a well-known biomarker of aging. Many previous studies have reported the change of DNA methylation patterns with age, and analyzed DNA methylation in association with aging outcomes. However, most publications were based on cross-sectional data while longitudinal evidence was largely missing. Hence, in this thesis, we used longitudinal measures of DNA methylation from the Swedish Adoption/Twin Study of Aging (SATSA) to comprehensively study the role of DNA methylation in aging.

The first three studies in this thesis focus on different mechanisms of DNA methylation related to aging, including methylation level, methylation variability and epigenetic mutation. In Study I, we investigated the longitudinal change of methylation level with age from an epigenome-wide association study (EWAS) using a mixed effect model. We identified 1316 age-related CpGs and successfully validated them in two external cohorts. Further, we analyzed the methylation difference between paired twins at the same time-point, and found it increased with age. We also identified genetic effect on age-associated CpGs, but the effect was independent on age. In Study II, we first developed a method that could properly model the longitudinal change of methylation variability with age in simulated data. The method included a linear model to regress methylation on age, followed by a random intercept model to regress the absolute residuals on age. Next, we applied the method in an EWAS and identified 570 age-varying CpGs. The inter-individual variance of most CpGs increased with age longitudinally.

In Study III, we comprehensively studied epigenetic mutations, which are extreme outliers in the distribution of methylation level. The number of epigenetic mutations significantly increased with age in our longitudinal data. We also identified other factors associated with epigenetic mutations, including sex, B cell, sample quality, cancer diagnosis and first genetic principal component. Further, we classified CpGs into frequent mutated CpGs, highly methylated outliers (HMO) and lowly methylated outliers (LMO), and found frequent HMOs were more related to biological factors. In the end, we validated epigenetic mutations using bisulfite pyrosequencing and proved that epigenetic mutations were persist and could accumulate in aging. In Study IV, we performed an EWAS to analyze methylation levels, methylation variability and epigenetic mutations in association with mortality. We observed age-varying effect of methylation level on all-cause mortality which may explain the poor replication in previous studies. We also identified CpGs of cancer genes related to death from cancer. In the end, we provided evidence that methylation variability could predict all-cause mortality.