What do methylation clocks need to become biomarkers in aging trials?
Drugs and interventions come with clinical trials. And trials won't start unless we know what to measure. That is why, as with any disease, we need to come up with reliable ways to measure what aging is. Like many other diseases (and the worst case of them all), aging is no one thing. As we age, our body fails in an increasing number of ways, from the accumulation of misfolded proteins to chronic inflammation, from oxidative stress to reduced sensitivity of important signaling pathways. Some of the current aging trials, such as TAME, measure the progression of age-related diseases late in life. Others focus only on 1 of the modalities of aging. However, there is yet no set of biomarkers that, if fixed, will surely help us defeat aging. Of the schemes that try to explain what aging is, pillars and hallmarks offer a high-level glance at age-related deterioration and not an actionable mechanism. Thus, the next big step for the aging field is biomarker development.
If we want a biomarker, it is desirable for it to be conserved when we go from model organisms to humans (when it is not possible, such as with Alzheimer, we have to “engineer” the disease and create imperfect surrogates, which makes it harder to target relevant problem). We likely want something that by default would correlate with chronological age. Chronological age is still an imperfect proxy though: differences in lifestyles can lead to different age acceleration and, thus, different correlations with chronological age. It is much more helpful if a given biomarker correlates with aging phenotypes (e.g., protein aggregates or cholesterol levels [1] ). One suitable framework that meets these parameters is a methylation clock. First developed in 2011 [2], it has since been updated to cover its use for predicting both chronological age and age-related conditions. Current clocks are capable of predicting lung and breast cancer incidence, stroke, coronary disease, Parkinson’s disease, and dementia [3] .
Methylation is an epigenetic process where a methyl group is attached to either DNA itself or histone proteins around which DNA is wrapped. Attachment of methyl groups can both activate and repress gene transcription. Methylation also affects alternative splicing [4] - the way an mRNA transcript is cut to diversify proteins that can come from it. Putting it all together, methylation serves as a lever for changing the amount of protein made from a given gene, potentially also adding or by-passing some features (e.g., enzymatic activity, localization within the cell, and stability).
While methylation patterns affect all of our genome, clocks look exclusively at cytosine methylation at sites called CpG islands (cytosine phosphate guanine). The current analysis is limited to only a fraction of CpGs. By looking at the pattern of cytosine methylation, the original Horvath clock managed to establish a correlation coefficient of 0.98 for the prediction of chronological age. Today, we can even compare how methylation clocks in animals will translate to methylation clocks in humans, allowing us to connect methylation patterns in model organisms to methylation patterns of our bodies. They can also predict the epigenetic age of organisms with negligible senescence, such as naked mole-rat (using Pan tissue clock). Yet, as amazing as they are, right now there are several problems they do not address. These are the things that would be essential to consider before clocks can be used as biomarkers in clinical trials:
- The methylation clock is very much a statistical tool and is, thus, susceptible to statistical limitations. Correlation measurements by themself do not inform us about causality within the system. Does aging cause epigenetic drift or does epigenetic drift make us age? Should we restore the epigenetic pattern or find the upstream reason for its existence? These questions are part of the quest to find the so-called causal clocks. Currently, there is no clear evidence showing that changing methylation only can change age-related phenotypes. We know that, in mice, a number of interventions can slow down epigenetic aging and, thus, the clocks. The same is not yet obvious in humans, since we do not have any intervention that would reliably extend our lifespan. There is also evidence that demethylation is required for regeneration of eye axons - in other words, we have to affect clocks if we want to restore eye tissue (though demethylation alone does not restore vision[5] ).
If we target methylation in trials without knowing that it is causal to at least some aging phenotypes, we can end up with treatments that change the pattern without doing anything useful for life extension (similar to Alzheimer’s treatments that use only beta amyloids as their proxy). In practice, many complex diseases do not yet have “causal biomarkers” ( think of any disease that we haven’t cured yet - its biomarkers are very likely imperfect, such as in Parkinson’s or cancer or Alzheimer’s).
- Different organs age at different rates [6] . And epigenetic age acceleration in the blood is a bad predictor of epigenetic age acceleration in most of the other organs. With this, even if we find a drug that slows down methylation change in the blood, we won't know whether aging was affected in many other tissues (at least not with the current clocks). If we measured it directly by doing a biopsy of lung, kidney, heart, our longevity trials would be extremely expensive to set up (generally, the field is much more oriented towards finding non-invasive biomarkers).
Alternatively, we can just pick tissues that would be the best predictors of age-related morbidity. There is some work happening to make it possible, including a preprint from Morgan Levine's lab on something called "systems clocks" that can be used for different organs. The problem remains though. There is little data on methylation patterns across the whole body in humans. Getting methylation reads of the brain quite literally requires drilling a hole in the skull - this is unlikely to ever happen on a large scale. So, technically, right now clocks can not predict the aging of the most important organs using just blood or saliva.
In addition to different organs aging at different rates, cells within the same tissue can also exhibit distinct aging patterns - and the bulk clocks (most of the current clocks are bulk clocks) do not address this. Imagine if someone tried to determine your chronological age by taking the average ages of your neighbors. There is a preprint from Vadim Gladyshev to address the problem of distinct aging rates, this time with single-cell clocks. For now, It seems like it would be quite challenging to come up with a way to “ensemble” these individual predictions.
- Yes, methylation pattern changes over time and, given the signal researchers were able to get when correlating it with chronological age, it changes predictably. However, knowing that DNA methylation has diverse functions within the organism (transcription control or protein alteration) we need a way of untangling these contributions to the clock.
Epigenetic change starts from day 1 of our development. In fact, the Horvath clock changes the fastest during the development stage of the organism [7] . Still, we do not consider maturation to be a deterioration process. Because of this, it is important that we delineate epigenetic adaptation from deterioration (OR understand at which point quasi-programmed aging starts). Consider a simple example: a given protein can be downregulated because it is no longer needed at a particular age, but it can also be downregulated by accident (the so-called “epigenetic drift”). Developing a "mechanistic" understanding of methylation as a biomarker would be essential for distinguishing these two cases.
Having irrelevant biomarkers may lead us to irrelevant treatments. Right now, there are some directions that seem less important for further improvements of the DNAm as a biomarker, such as fitting better clocks for predicting chronological age (after all, age is already a known variable). There are others, like mechanistic understanding, causality, and systems clock, that appear to be prerequisites for trials.
All in all, it is unlikely that methylation alone will be used as a biomarker. Even diseases that are less complex than aging have more than 1 biomarker. Because of this, it is important to figure out the combination that works best for a particular condition. For methylation clocks specifically, most invertebrates (including E.Coli and Drosophila) do not have cytosine methylation. Yet they still age in a way independent of methylation, so we need to look for other mechanisms and biomarkers that can complement it there.
[1] pretty comprehensive list of molecular & phenotypic aging biomarkers with which we can try to correlate clocks: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5473407/
[2] https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0014821
[3] table with a more detailed breakdown of clocks+diseases: https://clinicalepigeneticsjournal.biomedcentral.com/articles/10.1186/s13148-019-0656-7/tables/2
[4] https://www.nature.com/articles/nrg3230
[5] https://www.nature.com/articles/s41586-020-2975-4
[6] https://www.nature.com/articles/s43587-020-00015-1
[7] https://genomebiology.biomedcentral.com/articles/10.1186/s13059-019-1824-y