One of the biggest breakthroughs in biology in the last few decades has been the discovery of epigenetics. Rather than changing the genes themselves, epigenetics change how genes are expressed, allowing our cells to differentiate between their various types.
However, the epigenetics of our cells change over time. There is some debate over how much epigenetic alterations are a cause or a consequence of other age-related damage, but they are one of the primary hallmarks of aging.
Multiple “epigenetic clocks” have been developed over the last decade. These clocks are now displaying an uncanny ability to determine biological age, and Steve Horvath’s GrimAge can predict, with limited accuracy, how much longer a human has to live!
Professor Morgan Levine was a postdoc with Horvath before she became a professor at Yale. She has extended Horvath’s and others’ work on epigenetic clocks, as we discuss below. I met Levine at the Longevity Therapeutics conference in San Francisco in early 2019. We conducted this interview by email.
Why do epigenetic changes matter for longevity?
We are finding that age-related epigenetic changes are associated with mortality risk and, perhaps more importantly, with disease incidence. For instance, we have different algorithms that represent levels of DNA methylation that we expect to see for someone of a given age. Individuals who have methylation profiles indicative of someone older than they are have increased risk of morbidity and mortality. For instance, if you compare two 40-year olds and one has the methylation profile of someone who is 45, while the other has the methylation profile of someone who is 35, the former will, on average, live for fewer years and develop disease earlier.
What is the theory/mechanism behind the various epigenetic clocks now available?
This is ongoing work that we are actively pursuing. There are about a dozen epigenetic clocks in the literature—perhaps the most famous being the Horvath clock (although it wasn’t the first). However, even though these clocks are all intended to capture the same latent concept (biological aging), they differ in their predictions of age and age-adjusted death and disease risk. Using transcriptomic and proteomic data from both blood and brain, we have found that accelerated aging measured using the most widely known epigenetic clocks (i.e. Hannum, Levine, Lin, and the two by Horvath (pan-tissue and skin/blood)), seem to relate to mitochondrial dysfunction, PI3K/Akt signaling, and immunosenesence.
There is also some evidence coming out that they may reflect cellular senescence to some degree. That being said, our theory is that the clocks—because they are composites of hundreds of CpGs (cytosine separated from a guanine by one phosphate) —represent a grab bag of mechanisms. We are currently working on decomposing the various clocks and are finding that they differ in their proportions of various “types” of methylation changes, each of which may have their own distinct mechanisms. Our hypothesis that breaking the clocks down into constituent parts may facilitate our understanding of the underlying biology that is either driving these age-related changes, and/or the functional implications of such changes.
How does your version of the epigenetic clock differ from previous efforts?
Most of the early epigenetic clocks were trained to be age predictors. This means that the authors considered hundreds of thousands of CpG sites and used machine learning to develop the best combination to predict chronological age. By design, these methods are meant to remove the variance among same-aged people—the best outcome of these analyses would be that everyone is predicted to be exactly the same age as they are chronologically. However, we know that people don’t all age at the same rate, and we should want some degree of variance in the epigenetic age among those with exactly the same chronological age.
However, instead of this being noise, we (in our version of the epigenetic clock) want it to represent aging biology. Therefore, we chose to develop an epigenetic clock that was a predictor of an intermediate variable (which we called phenotypic age) rather than chronological age. This phenotypic age variable was based on clinical chemistry measures that are routinely ordered by a physician (kidney/liver panel, CBC, lipids, glucose, etc.) and, together, are robust predictors of death and disease. Ultimately, we found that using phenotypic age, rather than chronological age, to develop a methylation predictor, produced an epigenetic clock that was a more robust healthspan and lifespan predictor. Finally, even though we used DNA methylation data from whole blood to develop the new clock, it tracks aging in nearly every tissue or cell type we have tested.
How much better is your epigenetic clock at predicting healthspan and lifespan? Are these predictions being tested in humans or only in model organisms at this point?
Our new clock significantly predicts many disease outcomes and mortality better than previous epigenetic clocks (Horvath, Hannum, etc.) when estimated in blood. All of these studies were done in humans, and we are working on developing a similar clock for mice and/or rats. That being said, the new GrimAge clock is a better predictor in blood than ours. However, the difference is that the GrimAge clock seems to be specific to aging in the blood (leukocytes), whereas the PhenoAge clock also tracks age in most other tissues and is being shown to differentiate cancer versus normal tissue and is associated with Alzheimer’s disease in brain tissue and IPF in lung tissue.
How far are we from understanding epigenetic changes well enough to be able to turn back the clock by changing the epigenetics in individual organisms?
I think we are quite a long way from that. The epigenome is a complex system, and we are not at the point where we can model these changes very well—there is a lot of room for improvement when it comes to the clocks. That being said, we are even further away from understanding what these changes represent or if they are even causal.
I hypothesize that many of these changes are actually compensatory—reactions to something going wrong in some system. Thus, altering DNA methylation directly will not be beneficial and could, in fact, be harmful if this isn’t accompanied by changes to the system/extracellular environment.
Furthermore, people are thinking only about the clock CpGs specifically, but these CpGs are acting as part of a larger system that encompasses many more CpGs—most of which are not necessarily part of the clocks because they represent redundant information, but that are equally important to the phenotype. Nevertheless, I absolutely think we are at a point where we need to be identifying aging therapeutics that target other aging pathways, but also exhibit changes in epigenetic age, or age-related methylation more generally.
What are the next big steps to be completed in understanding epigenetic changes well enough to be able to turn back the clock?
The next big step is to relate specific epigenetic changes to other hallmarks, pathways, and/or aging processes. This will probably not happen at the whole clock level. In doing so, we will be able to use these measures as potential endpoints in clinical trials targeting one or more hallmarks. The other critical thing is to distinguish DNAm that is related to chronological time (less potential for intervention) versus those related to biological aging (potentially modifiable).
If we understand epigenetic changes well enough at this point, do we have the tools to roll back specific changes to mimic youthful epigenomes?
Again, as of now, I don’t think there is evidence to suggest that altering the epigenome directly will be beneficial. If some of these changes are effects (read-outs) of aging, then they are not the correct points of intervention. Further, many of the changes may be compensatory, and thus making an old cell epigenetically young but leaving it in an aged organismal environment could be detrimental and possibly contribute to neoplastic transformations.
Does partial cellular reprogramming using enhanced Yamanaka factors, as discussed by Ocampo, et al. 2016, and Sarkar et al. 2019, among others, work via epigenome reprogramming? Or is something else going on?
Likely, but I don’t think aging researchers should necessarily look to this as a promising way to intervene in aging. This may be the case, but as of now, I think jumping to such a conclusion could be dangerous. Aging doesn’t just happen at the individual cell level—the entire organismal system declines. Thus, reversing aging in a more reductionistic manner, versus a systemic manner (similar to what happens with dietary restriction (DR)), may not produce the outcomes we are hoping for.
We thank Prof. Levine for agreeing to this interview and for speaking at our upcoming conference, Ending Age-Related Diseases 2019, on July 11-12. If you’d like to attend, click here or the graphic below.