In recent years, researchers have established that machine learning approaches can be used to produce any number of clocks from biological data that shifts with age, finding patterns that match chronological or biological age to a great enough accuracy to suggest that they can be useful assays for the assessment of potential age-slowing and rejuvenating therapies.
It remains an open question as to whether and how exactly the assessed patterns correlate to the specific forms of molecular damage that cause aging, or to any of the specific downstream consequences of that damage. Scientists here raise the possibility that much of the epigenetic change of aging may not in fact be as useful as a basis for measurement as thought, and suggest that more fundamental research is required in order to robustly connect clocks with specific processes of aging.
Image credit: Pixabay (Free Pixabay license)
Our meta-analysis of the largest available age-annotated methylation dataset to date found: 1) as much as one fifth of the measured cytosines contains age-predictive methylation patterns; 2) tissues show largely similar aging patterns despite having methylated regions that define their identity; 3) epigenetic clock sites are enriched in intergenic regions, gene enhancers, and sites near expression quantitative trait loci (eQTLs) and 4) are depleted in the regions generally thought to have the largest direct impact upon gene expression (e.g., CpG Islands and gene promoters); 5) patients with age-correlated diseases did not appear significantly age-accelerated according to the chronological epigenetic clock.
The fact that many different sites can be used to create an epigenetic clock with minimal impact on predictive performance argues against the idea that methylation changes are either programmed or individually important. Yet, because the clock is robustly predictive and age-related methylation changes are mostly similar between tissues, this argues against entropy as a driving force. This could be reconciled by hypothesizing some genomic regions and/or features receive less methylation maintenance than others.
Perhaps the changes occur in regions of the genome where they have no consequence, and instead, vary with absolute time such as in determining speciation time using pseudogene mutation rates. This “pseudomethylation” would be problematic for modeling aging biology, as they would likely not respond to aging intervention. Methylation maintenance mechanisms (e.g., DNMT1) serve as a counterbalance against entropy. However, if some genomic regions are less maintained than others, then we would expect the probability of a methylation state change with age to be correlated with the degree to which it is subject to methylation surveillance and maintenance. Because maintenance costs energy, it is reasonable to hypothesize the degree of maintenance correlates with the adverse impact an unregulated change in methylation would cause. If so, the probability a site's methylation will vary with age would inversely correlate with its impact on an organism's survival.
Given that methylation changes with age are robust across tissues, yet small in magnitude, leads the field to question whether the “ticking” that drives them is due to changes in cell population composition, such as a reduction of pluripotent stem cells or an increase in senescent cells within every tissue, or possibly high magnitude effects in rare cell populations (e.g., immune cells in the central nervous system compared to astrocytes or neurons). In either case, it is not clear whether the phenomenon driving ticking clock sites is due to healthy compensatory changes or deleterious drift toward age-related fragility.
In summary, the predictive power of the epigenetic clock is robust, but such a large fraction of the genome can be used to predict, the magnitude of the changes is small, and these regions tend to be depleted near genes. This leads us to hypothesize that the pan-tissue predictive loci are more likely to be molecularly “silent” methylation changes that accrue outside of strong regulatory regions due to entropy in methylation maintenance, which must be explored in the future studies. Furthermore, if current models inconsistently annotate patients with age-related diseases as “age-accelerated” and the confidence by which one can declare a sample age-accelerated is small, this argues against the idea that epigenetic clocks can disentangle biological age from chronological age.
Link: https://doi.org/10.1111/acel.13492
Source: Fight Aging!