A new methylation clock works in 128 different mammal species, using the same methylation signals. This is the latest evidence that at least some of the mechanisms of aging have been conserved by evolution—strong evidence that aging has a useful function in ecology, so that natural selection actually prefers a finite, defined lifespan.

Einstein taught us that time is relative. Indeed, there are rodents that live less than a year, and Bowhead whales that live more than 200 years. Some of this is just about size and has a basis in physics; but it is well-known that size is only part of the story. Bats and mice are the same size, but bats live ten times longer. Humans are much smaller than horses, but live three times as long.

The first time I met Cynthia Kenyon was circa 1998. She offered me a one-line proof that aging is programmed: the enormous range in lifespans found in nature defies any theory about damage accumulation, because no conceivable process of chemical damage could vary so widely in its fundamental rate. (Think mayflies and sequoia trees.) My own one-line proof is that yeast and mammals share in common some genetic mechanisms that regulate aging, though the last common ancestor of yeast and mammals is more than half a billion years old. These mechanisms include sirtuins and the insulin metabolism.

These intuitions about aging rate and evolutionary conservation have recently come to the world of big data. In this new BioRxiv manuscript, Steve Horvath collaborates with an all-star cast of biologists the world over to compile evidence that there is a universal mechanism underlying development and aging in all mammals, and it is a pan-tissue epigenetic program, not a process of chemical damage.

Brief background on methylation: It is increasingly clear that aging has a basis in gene expression. The whole body has the same DNA, and it doesn’t change over time. However, different genes are turned on and off in different times and places. Turning genes on and off is called “epigenetics”, and evolution has devoted enormous resource to this process. One of many epigenetic mechanisms is the presence or absence of a methyl group on Cytosine, which is one of the 4 building blocks of DNA (A, C, T, G). There are over 20 million regulatory sites in human DNA where methyls can appear or not. Of these, several thousand have been found to consistently correlate with age. The correlation is so strong that the most accurate measures of biological age are now based on methylation. There is (IMO) a developing consensus in the community that methylation changes are an upstream cause of aging, and there remains strong resistance to this idea on theoretical grounds. More background here

The team assembled tissue samples from 59 organs across 128 species of mammals, and looked for commonalities in the progression of methylation that were independent of species and independent of tissue type. They found thousands of methylation sites that fit the bill, attesting to an evolutionarily-conserved mechanism “connected to” aging. It is a short leap to imagine that “connected to” implies a root cause.

How did the authors map age for a mouse onto age of a whale? Just as I might say, “I’m only 10 years old, in dog years,” a year for a whale might be a hundred “mouse years”. The authors took three different approaches. (1) Just ignore it, mapping chronological time directly. (2) Adjust time for the different species based on the maximum lifetime for that species. (3) Adjust time for the different species based on the time to maturity for that species.

Predictably, (1) produced paradoxes; (2) and (3) were similar, but (3) produced the best results. What they didn’t do — but might in follow-on work — was to optimize the age-scaling factor individually for each species to target the best fit with all the other species. Even better would be to choose two independent scaling factors to optimize the fit of each species. Ever since the original 2013 clock, Horvath has divided the lifespan into two regimes, development and aging: In development, time is logarithmic, moving very fast at the beginning and slowing down at the end of development. In the aging regime, time is linear. So it would be natural (optimum, in my opinion) to choose two separate scaling factors that best map each species’s life history course onto all the others. Mathematically, this is (roughly) as simple as matching the slopes of two lines. Horvath has told me he is interested in pursuing this strategy but for some species the existing data doe not cover the lifespan sufficiently to support it.

“Cytosines that become increasingly methylated with age (i.e., positively correlated) were found to be more highly conserved (Fig. 1a)  …Interestingly, although there were 3,617 enrichments of hypermethylated age-related CpGs [i.e., increased methylation with age] across all tissues, only 12 were found for hypomethylated [the opposite] ones.”

Interpretation: with age, we (and other mammals) tend to lose methylation, i.e., to turn on genes that shouldn’t be turned on. There are more sites that demethylate with age than that methylate with age. But the sites that gain methylation tend to be more highly conserved between species. I presume a lot of demethylation is stochastic. It’s easy for a methyl group to “fall off”, but attaching one in the right place requires a specialized enzyme (methyl transferase). What we are seeing here is stronger genetic determinism for the process that requires active intervention.

Question: Would it be useful to develop a methylation clock based solely on sites that gain methylation? What we would thereby avoid is the situation where the age algorithm combines a great many large positive numbers with a great many large negative numbers to make a small difference. This characteristic makes the algorithm overly sensitive to bad data from one or a few particular sites. We can see from the figure above that (red) sites from the top half of the plot have stronger evidence behind them than the (blue) sites from the bottom. What we would lose would be diversity in the basis of the measurement. If retaining that diversity is desirable, it would be possible to design a clock algorithm with both red and blue sites in such a way that all coefficients are relatively small, and no one site contributes inordinately to the age calculation, even if data for that site is completely missing.

Speculation for statistics geeks: I think the methodology that has become standard for developing methylation clocks is not optimal. The standard method is to identify N sites (typically a few hundred) where methylation is well-correlated with age, then derive N coefficients such that you can multiply each coefficient by the corresponding methylation, add up the products, and you get an age estimate*. The way I would do it is with a more complicated calculation, from a methodology called “maximum likelihood”. The idea is to choose the age that minimizes the difference between the expected methylation and measured methylation for the collection of the N sites. To be more specific, minimize the sum of the squares of the z scores for each site, where z is the number of standard deviations by which the measured methylation is different from the expected methylation.It may sound like a complicated calculation to find the age at which this number is a minimum, but it is not. Yes, it’s a guessing game; but the algorithm called “Newton’s method” allows you to make smart guesses so you home in on the best (min Σz2) age within four or five guesses. The calculation is more complicated to program, but it would still execute in a tiny fraction of a second. My proposed method requires maybe 10 or 20 times as many fixed parameters within the algorithm; but the data submitted from each sample is the same.
Caveat – This is all theoretical on my part. I don’t know how much performance would be improved in practice.
————————
*Two footnotes: (1) A constant is also added. (2) In case the subject is young, below the age of sexual maturity, what you get is a logarithm of age, not age itself.

“Importantly, age-related methylation changes in young animals concur strongly with those observed in middle-aged or old animals, excluding the likelihood that the changes are those involved purely in the process of organismal development.”

These plots are adduced as evidence that aging and development are one continuous process under epigenetic control. They come from EWAS=epigenome-wide association studies. Start by asking which sites on the methylome are most closely correlated with age, across many different animals and different tissues in those animals. Start with just the young animals (different ages, but all before or close to sexual maturity. Arrange all the different sites according to how they change methylation with age (increasing or decreasing), just in this age range. Then repeat the process, re-ordering the sites according to how they change with age during middle age.

The left plot above includes a dot for each methylation site, ordered along the X axis according to how they change during youth, and along the Y axis according to how they change during middle age. The point of the exercise is that it is largely the same sites that increase (or decrease) methylation in youth and in middle age.

The middle plot shows the corresponding correlation between middle age (X axis) and old age (Y axis). The right-hand plot shows the correlation between young (X axis) and old age (Y axis). (I believe the labeling of the figure on the right is a misprint.)

This evidence points to a conceptual framework that views development and aging as one continuous process. Development is a lot more complicated than aging. Consequently, most of the sites in the clock are developmental.  Maybe a clock could be optimized for aging only, and it would be more useful for those of us who are using the clocks to assess anti-aging interventions.

“The cytosines that were negatively associated with age in brain and cortex, but not skin, blood, and liver, are enriched in the circadian rhythm pathway”

Here we see again the intriguing connection between the brain’s daily timekeeping apparatus and the epigenetic changes that drive development and aging.

“The implication of multiple genes related to mitochondrial function supports the long-argued importance of this organelle in the aging process. It is also important to note that many of the identified genes are implicated in a host of age-related pathologies and conditions, bolstering the likelihood of their active participation in, as opposed to passive association with, the aging process.”

Another theme in the set of age-correlated genes that the team discovered is mitochondrial function. Mitochondria have an ancient association with cell death, and a long, conserved history with respect to aging. The simple damage themes associated with the free radical theory have yielded to a more complex picture, in which free radicals can be signals for apoptosis or inflammation or enhanced protective adaptations.

The big picture

“Therefore, methylation regulation of the genes involved in development (during and after the developmental period) may constitute a key mechanism linking growth and aging. The universal epigenetic clocks demonstrate that aging and development are coupled and share important mechanistic processes that operate over the entire lifespan of an organism.”

This is cautiously worded, presumably to represent a consensus among several dozen authors, or perhaps to appease the evolutionary biologists looking over our shoulders. The statement is akin to what Blagosklonny has for years called “quasi-programmed aging”, to wit, there are processes that are essential to development that fail to turn off on time, and cause damage as the organism gets older. In the version put forward in this present ms, it is not the gene expression itself but the direction of change of gene expression that carries momentum and cannot be turned off.

Evolutionary theory

Modern evolutionary theory began with Peter Medawar, a Nobel laureate and giant of mid-century biological understanding. (He was 6 foot 5.) Medawar’s 1952 monograph contains the insight that launched all modern theories for evolution of aging. His fundamental idea was that it’s a dog-eat-dog world in which very few few animals live long enough for aging to be a factor in their death. The three main branches of evolutionary theory in response to Medawar are called Mutation Accumulation, Disposable Soma, and Antagonistic Pleiotropy. According to Medawar’s thought (and all three theories that followed) old age exists in a “selection shadow” so random processes are at work in old age. It follows that we would expect the aging of a bat and a bowhead whale to be subject to very different random processes. If it is a burden of recently acquired mutations that natural selection has not yet had time to weed out, these should be different for different species. Or if it is about tradeoffs (pleiotropy) between needs of the young animal and the old animal, we would not expect the bat and the whale to be subject to the same tradeoffs.

The Medawar paradigm and its three popular sub-theories all predict that there should be little overlap between the genetic factors involved in aging of species that are adapted so differently. Therefore, the present work documenting a common epigenetic basis of aging is a challenge to the established evolutionary theories of aging.

As I see it, the expression of genes is exquisitely timed for many purposes, so we must view gene expression as subject to tight bodily control. “Accidents” or “mistakes” or “evolutionary neglect” are implausible. For some genes, methylation changes from minute to minute in a way that is adaptive and responsive. Blagosklonny’s idea that there are genes turned on for development and then the body forgets to turn them off doesn’t feel right. Equally, the idea that certain genes are being turned on (or off) progressively through development and then, after development has ended, the process has a momentum of its own so the body can’t stop further turning on (or off) of these same genes is equally implausible. I assume the body is adapted to do exactly what it wants with gene expression, and if the body expresses a combination of genes that causes aging, it’s because that’s what natural selection has designed the body to do. Of course, this looks to be a paradox, as aging is completely maladaptive according to the notion of Darwinian fitness that became accepted in the first half of the 20th century; but evolutionary biologists have broadened the notion of fitness since then, and I’ve written volumes concerning this paradox.

The bottom line

For personal application to individuals who want to know how well they are doing and their future life expectancy, I recommend Horvath’s Grim Age clock as the best available. (Elysium has done a lot of work on their Index product, and it may be as good or better, but it’s impossible to evaluate unless they release their proprietary methodology.) For application to studies of anti-aging interventions (including my own project, DataBETA), the choice of clocks is not clear, because it depends not just on statistics but on theory. We want a clock that is not only accurate, but that is based on epigenetic causes of aging, not epigenetic responses to aging. The multi-species clock is a welcome contribution, precisely because epigenetic processes that are conserved across species are more likely to be linked to the root cause of aging. For the future, I’ve made suggestions above for ways the multi-species clock might be made even better.