In his talk at Ending Age-Related Diseases 2021, Steve Horvath discussed a cytosine-based, pan-mammalian epigenetic clock that works across a wide variety of species. He also discussed how several of the hallmarks of aging do not seem to have any effect on epigenetic alterations.
Hello, thank you for your interest. My name is Steve Horvath. I will talk to you today about recent developments surrounding epigenetic clocks.
I will start with an old observation: induced pluripotent stem cells are perfectly young. In other words, they have negative epigenetic age, according to the pan-tissue clock. So, further, when you culture iPS cells, and you passage them, the epigenetic age does not seem to increase, but these results were found in humans.
The question is, do iPS cells from non-human species also have a perfectly young epigenetic age? People from Vadim Gladyshev’s lab have shown that the answer is yes for mouse iPS cells, but what about other mammalian species? To answer this question, and many others, we developed a new methylation array, which we call the mammalian methylation array. This array measures up to 37,000 highly conserved cytosines. By design, the mammalian array can tolerate sequence variation between different species.
So here, you see that we want to target the orange cytosine-guanine pair in humans. Overall, this nucleotide sequence is highly preserved in chimps and many other species, but the yellow nucleotides show inconsistent alignment across species. So, Adriana Arneson and Jason Ernst have developed an algorithm for designing probes that target these highly conserved regions. We’ve applied this mammalian array to many species, as you will see during the talk.
I’ll start out with a pan tissue clock for the naked mole rat. In our experience, you can build an epigenetic clock for all tissues and cell types from the naked mole rat. These are data from Vera Gorbunova, Chris Fox, Andre Seluanov, and you see a correlation between methylation age and chronologic age, which is 0.95, so clearly, the naked mole rat ages on an epigenetic level.
However, coming to this question about iPS cells, I show you here that iPS cells from the naked mole rat also show a very young age, according to two clocks, the skin clock for the naked mole rat, but also the pan-tissue clock, so that aspect is conserved. This idea that iPS cells have perfectly young age is also explored in several papers by different groups, where epigenetic reprogramming is used in a transient way or interrupted to lower the epigenetic age and other hallmarks of aging in humans, and the hope is, of course, that this kind of intervention will help us to rejuvenate different organs.
David Sinclair’s talked about restoring vision, and Jay Sarkar and Vittorio Sebastiano have done a very comprehensive study where they have shown that endothelial cells and fibroblasts get rejuvenated, and, of course, Carlos Belmonte and Ocampo have earlier shown these results in mice. I want to briefly mention another type of clock, I call it a third-generation epigenetic clock. Why? Because this is a clock that applies to multiple species.
Here, I show you a clock that applies to two types of elephants, the so-called African elephant but also the Asian elephant, but importantly, the same clock also applies to humans. You can see that in the left panel, where the orange dots are human samples, the green dots are the Asian elephant, and the purple dots are the African elephant.
You see the age estimates on the Y axis have a very high correlation with relative age on the X axis. The relative age is the ratio between chronologic age and the maximum lifespan of the species. These third-generation clocks are truly remarkable because these three species diverged tens of millions of years ago, and you can still use a single regression model, a single mathematical formula for estimating ages.
There is, of course, an elephant in the room when it comes to epigenetic clocks. It is, what do epigenetic clocks measure? What are causes? What are the consequences? What’s the mechanism? More precisely, which other hallmarks of aging relate to epigenetic aging?
Here, I would like to draw your attention to a new article by Ken Raj, who carried out very detailed in vitro studies in fibroblasts, keratinocytes, and endothelial cells to answer these questions, and I will walk through this article in the following.
What Ken has shown is that several hallmarks of aging do not seem to relate to epigenetic clocks. For example, cellular senescence, telomere attrition, genomic instability do not seem to relate to epigenetic aging. Conversely, deregulated nutrient sensing, mitochondrial dysfunction, stem cell exhaustion, and altered cell-cell communications seem to relate to epigenetic aging.
I will show you some details of these experiments that corroborate these claims. I’ll start out with cellular senescence. There are many different ways of inducing cellular senescence; we probably looked at all of them. For example, when you expose cells to radiation for ras oncogene expression, you get senescence within a couple of weeks, but this does not have an effect on the methylation age in these keratinocytes.
The figure shows you on the Y axis methylation age, the X axis are different donors, different genetic backgrounds, and you see that the green dots and the orange dots do not differ from the blue dots in terms of methylation age.
The one difference you see are the red dots. These red dots come from another way of inducing senescence, which is replicative senescence, the famous Hayflick limit. You passage the cells, they stop dividing, replicative senescence, and there, you do see an increase in epigenetic age. However, it took up to six months to see replicative senescence.
The hypothesis is that the passage of methylation age reflects the passage of time in the dish, and that’s indeed the case. We know that by the following set of experiments. It turns out, when you immortalize cells by expressing hTERT, you do not prevent epigenetic aging. This can be seen here in the scatterplot, where on the X axis, we have cumulative population doubling levels, in other words, how often the cells have divided, and on the Y axis, we see methylation age, and what you notice is that the hTERT immortalized cells, the orange dots, continue to gain an epigenetic age well beyond the point when the control cells have already senesced.
So, hTERT immortalization does not prevent epigenetic aging, and that immediately shows that preventing telomere attrition also does not prevent epigenetic aging. From the last two results, we also arrive at the finding that the increased age of these replicative senescent cells is not due to the senescence per se but rather due to the aging of the cells as they were cultured for up to six months in the dish.
These results that were obtained, I should say, in many different donors, have been replicated over and over. What they really show is that epigenetic age is distinct from two of the best characterized hallmarks of aging, which is cellular senescence and telomere attrition.
I now come to another hallmark of aging, genomic instability, which is also not associated with epigenetic age, and this can be seen by comprehensive experiments from Ken Raj surrounding radiation, which induces genomic instability. I should mention several other authors have done these experiments.
Ken looked at primary human cells and exposed them to acute irradiation, 20 Gray, and there was no increase in epigenetic age, but also Ken looked at chronic low dose gamma irradiation, one milliGray per hour for 70 to 150 days, or higher doses, 20 milliGray, 50 milliGray. These studies were all done in humans, but Ken also repeated these experiments in mouse embryonic fibroblasts with pulsed radiation. Again, the lower panel shows you that the radiation-exposed cells in orange did not gain in methylation age compared to the controls.
In short, radiation, in any way you administer it, high-dose, low-dose, pulse, and so on, does not increase epigenetic age.
Now I want to come to hallmarks of aging that are linked to the epigenetic clock, and the first one would be stem cell exhaustion. So, one can look at skin cells or keratinocytes, and one can distinguish stem cells from stem cell-depleted regions. Here, we see that stem cells in orange have a lower methylation age than stem cell-depleted parts in blue. In short, that strongly suggests that stem cell exhaustion in skin relates to the epigenetic age of keratinocytes.
Another hallmark of aging is nutrient sensing, and that is really strongly related to the epigenetic clock, and we know that from several lines of evidence, first experiments by administering rapamycin in keratinocytes. Ken conducted extensive experiments, where he showed that rapamycin really slows the progression of epigenetic age as you culture cells, so the red dots are control keratinocytes. They gain methylation age with cell passage number for cumulative population doubling level; the orange dots show what happens if you administer rapamycin.
In short, rapamycin may slow epigenetic aging of skin tissues, but I should mention right away, we also looked at rapamycin effects in blood, and we did not see any rejuvenating effect.
Another line of evidence linking nutrient sensing to epigenetic aging was provided by many mouse studies. So, caloric restriction in mice is very much associated with slower epigenetic aging in mouse liver samples; there are several articles to that effect.
The third evidence comes from human studies. So human body mass index is actually strongly correlated with the epigenetic age of human liver samples, and to a lesser extent, there is a correlation between BMI and epigenetic age in blood tissue. So all these studies, they much implicate nutrient sensing.
What about mitochondrial function, there are ambiguous experiments, but I would still say there is a link to the epigenetic clock. One of the nicest set of experiments comes again from Ken, where he administered two types of compounds to cells growing in a dish. Here, the control dots are in red, and the blue dots are see CCCP-exposed carotene fibroblasts.
What you see is that CCCP, which dissipates the mitochondrial membrane potential, increased epigenetic aging. Conversely, bezafibrate, which increases mitochondrial biogenesis, was associated with a decreased epigenetic age. So these types of experiments link mitochondrial function to epigenetic aging.
I mentioned that they are also ambiguous results because we looked at a couple of knockout mice where mitochondrial function was disrupted, and sometimes we did not observe accelerated epigenetic aging. But in these in vitro studies, there seems to be a connection.
Here, I show you several interventions that extended the proliferative capacity for cellular lifespan of cell cultures, anything from rapamycin to bezafibrate, metformin, NR, NAD, and so on. So these interventions affect cellular lifespan. However, only a subset of these actually affected the epigenetic clock.
The conclusion seems to be, based on this limited set of interventions, that if your intervention affects the epigenetic clock, then it also affects proliferative capacity in the cells. But the converse is not necessarily the case: you can have an intervention such as the ROCK inhibitor, which affects proliferative capacity, but it does not seem to have an effect on the epigenetic age. The same statement also seems to pertain to metformin, NAD, and NR. All of these have benefits for proliferative capacity but don’t seem to have a strong effect on the epigenetic clock in these in vitro studies.
I want to end this part by talking about another important hallmark of aging, which is altered cell-cell communication, and that seems to be very strongly associated with epigenetic age. To demonstrate that, I show you some results from Harold Katcher and Akshay Sanghavi, who developed a young plasma treatment and administered that to rats. Interestingly, this young plasma treatment halved the epigenetic age of several tissues; it halved the epigenetic age of blood, liver, and heart. The effect in hypothalamus was less impressive.
Here, the red bars show you old control animals, old rats, and the orange bar, the plasma-treated rats, so you see dramatic age reversal in these different tissues. This article is available on bioRxiv, and we are currently conducting additional experiments to validate these original claims. Because if you make extraordinary claims, you need extraordinary evidence to convince yourself and others.
Now, the thing I didn’t mention in the previous slide is what clock did I apply. I applied a dual species clock, I call it a human rat clock, that, again, is very accurate in both species. I mentioned relative age before, but I want to give you a little math formula: the relative age of a species is really the ratio between its chronologic age and the maximum lifespan of the species.
We really need it because the maximum lifespan of rats is 3.8 years, and conversely, for humans, it’s 122 years. To put these two species on the same scale, you need to form some sort of a ratio or to carry out a nonlinear transformation, but you need to do something. In our experience, simply forming a ratio actually works; you can build very accurate dual species clocks.
I want to now talk about the Mammalian Methylation Consortium. This is a collaboration of over 160 investigators. We have profiled over 200 mammalian species, and this data allowed us to build universal methylation clocks, maybe a better term would be pan-mammalian clocks; these clocks apply to all mammals.
In panel A, I show you one of these clocks: you see correlation .98. Each.dot is a tissue sample from a different species; tissues are labeled by species. The lower left panel D shows you, again, this relationship between relative age of the animal and its methylation estimate. You see a correlation of .95, so it’s really remarkable how accurate these clocks can be by simply forming a ratio.
I want to emphasize that any scatterplot I show you in this presentation is based on cross validation or test set evaluation, so it’s always an unbiased evaluation. Panel B shows you how this clock performs in marsupials, in other words in kangaroos and possums and Tasmanian devils.
I highlight these species because in these species, the clocks, I expected to do worse, because the mammalian array does not contain as many cytosines for marsupials as it does for eutherians. So only about half of the cytosines on the mammalian array apply to marsupials. I think you still see its quite impressive performance.
For some species, the maximum lifespan is not known, and therefore, we have developed another universal clock called clock three that does not make use of maximum lifespan but rather makes use of age at sexual maturity, which is typically known for a species. Here, I show you how these universal clocks apply to individual species.
The upper left panel shows you the performance in humans, upper middle panel, mice, dolphins, dogs, elephants, and bats, and you’ll see that these clocks are very accurate. Although they apply to all species, they’re still very accurate in individual species.
Now, come to a different topic. Forget chronologic age; now we’re talking about maximum lifespan of a species, which is an important characteristic of a species. I mentioned our maximum lifespan is around 122, and what I want to show you here is that we can predict the maximum lifespan of a species quite accurately on the basis of cytosine methylation.
These ages are all on a log scale, where you see a correlation of close to 0.9. The methylation carries a lot of information for lifespan and other life history traits. You can also predict gestation time, age at sexual maturity, and so on. We have this maximum lifespan predictor, and we applied it to mouse studies and, interestingly, growth hormone receptor knockout of these dwarf mice have an increased epigenetic maximum lifespan, which matches what one knows about dwarf mice, dwarf mice live longer.
But there was also a surprising result, which may be unexpected to some of you, which is when you conduct caloric restriction, you do not seem to affect the maximum lifespan of the mouse species. That’s the right panel. CR did not perturb the maximum lifespan estimate of a mouse, and this really is a good illustration that there is a fundamental difference between average lifespan for mortality risk and maximum lifespan. Maximum lifespan is a species characteristic, and not everything that perturbs mortality risk also affects the species characteristic.
Another clock I want to briefly mention is our dual species clock for humans and sheep. I want to show it because we evaluated the effect of castration in male sheep, and there was an interesting finding. If you castrated male sheep, the blue dots, then the ear samples of these animals showed delayed epigenetic aging, but not the blood samples, I should say so. So that’s interesting. The green samples in the middle show you the epigenetic age acceleration in intact males, and I think you’ll see a difference.
We also evaluated the effect of castration in other animals, but only in blood, and in blood, I never saw a difference. So for example, in dogs or horses, or cats, I found no association between castration and epigenetic aging. But in these skin samples, ear samples, there was an effect in sheep.
We have a very accurate epigenetic clock for dogs. By now, we analyzed 742 dog samples, 93 different dog breeds. This is a collaboration with Elaine Ostrander, and we have a dual species clock for humans and dogs.
Similarly, we have a dual species clock for a primate that’s very interesting to aging researchers, the marmoset, it’s interesting because this is a short-lived primate. I want to mention we have clocks for all primates. You can think of rhesus macaques, baboons, and so on.
We have also what I call the human pig clock. So this, again, dual-species clock that applies to many pigs, also regular-sized domestic pigs, but also humans. And human cat clocks, for the cat lovers among you, so you give us a blood sample from a cat or yourself, we can estimate the ages using the same regression model, the same formula.
So, why do I mention all of these species? The reason is because I think they really usher in a new era of testing, which is to apply the same clock, for example, the same pan-mammalian clock, to several species at the same time. So, instead of wasting your time going through a mouse model, then a rat model, then a dog model, then a primate model, you can streamline it and apply the same clock to all animals at the same time.
This clock will have one massive advantage: it will also apply to humans. So if you have an intervention that slows epigenetic aging in six different species, the chance is, it will also work in humans.
I want to come to a challenge. There are really too many epigenetic clocks, and there are many epigenetic clocks for humans. There are many epigenetic clocks for mice. There is really this danger that we encounter the situation of the Tower of Babel, a total confusion of languages; how do we understand each other when we publish things?
Therein lies a challenge, which is we quickly need to characterize new epigenetic clock somebody develops a new clock, we run it through a benchmark data set and quickly understand the properties of the clock.
There is some good news, because the Mammalian Methylation Consortium has assembled powerful benchmark data in mice, we call this the Methyl Gauge Clock, and this project is spearheaded by Amin Haghani.
We really have a framework for characterizing mouse clocks but also any clock that is generated on the mammalian array in over 100 mouse experiments, anything that aging researchers could be interested in evaluating: high fat diet, caloric restriction, growth hormone receptor knockout, iPS cells, progeria models, rapamycin, Yamanaka factors, and so on. Many different perturbations, as I said, over 100 experiments, and also importantly, gene knockout studies, many gene knockout studies.
If you generate your data on the mammalian array, you can immediately tap into this database to characterize your own findings. Not just epigenetic clocks, also your most exciting cytosines. This database was, of course, a community effort. Here, I list some names, but there are over two dozen researchers who contributed to it.
I want to mention our best human epigenetic mortality risk predictor, GrimAge. I talked quite a lot about it in the past. I want you to know that there are many cross-sectional studies that have shown that GrimAge predicts lifespan in dozens of epidemiological studies. More recently, we published a paper that is a genetic study.
We have developed a polygenic risk score for GrimAge, and it is associated with parental longevity and also educational attainment, adiposity, and so on. So if you have GWAS data, SNP data, you can evaluate epigenetic clocks using that framework.
Some questions people always ask, is cytosine methylation causal? Or is it just a marker? And my answer is yes and no, because one can find plenty of examples to support both points of view.
The no camp will point out that many invertebrates do not have cytosine methylation. C. elegans, Drosophila, do not have cytosine methylation. So, if you believe that these model organisms very much mirror mammalian aging, then you will discard the cytosine methylation. Also, in mammals, many age-related cytosines do not seem to have an effect on gene expression levels.
However, on the yes side, there have been several exciting case studies that start to build evidence that cytosine methylation can be causal. There was a paper from David Sinclair’s group, where he showed that this OSK-induced epigenetic reprogramming for axon regeneration did require TET enzymes, that’s kind of a smoking gun that cytosine methylation is in some way involved in regeneration.
There was also another very nice paper, and that studied methylation in the regulatory region of the ELOVL2 chain. If you are in my field, which is methylation studies of aging, you will know this gene, it’s really one of the most robust cytosines correlated with age.
Here, this team, Daniel Chen, and Dorota Skowronska-Krawczyk and their team, published a paper that showed that reversal of ELOVL2 promoter hypermethylation in vivo leads to ELOVL2 expression and rescued age-related decline in visual functions. That was a nice illustration of the relevance of this one cytosine.
In conclusion, epigenetic clocks have been linked to four hallmarks of aging, but also they seem to be disconnected to several others. In particular, epigenetic age seems to be distinct from telomere length, attrition, and cellular senescence.
I also showed you some evidence that interventions that slow epigenetic clocks often extend the lifespan of cells in culture in cell cultures, but not vice versa. I showed you that castration slows epigenetic aging in ear samples of sheep but has no effect on the epigenetic age of blood.
Epigenetic clocks are already being used in clinical and preclinical studies. We published some articles on it in the past. I mentioned, for example, Greg Fahy with his thymus intervention. I also showed you that cytosine methylation allows one to build these pan-mammalian clocks. Cytosine methylation allows you to estimate the maximum lifespan and many other species’ characteristics for mammals.
I mentioned many people during my talk I mentioned Ake Lu, Amin Haghani, many collaborators, Ken Raj and others. Thank you so much.