Dr. Steve Horvath hardly needs an introduction, so we will be brief: he is the inventor of the epigenetic clock and, currently, principal investigator at Altos Labs. We talked about the recent developments in this immensely important field, including pan-mammalian clocks, two-species clocks, and single-cell clocks, along with the challenges the field faces.
You are a mathematician by training. How did you end up studying biology and the biology of aging in particular?
I was already interested in aging research as a teenager, but my first love was math. Over the years, I became more and more applied. After getting a PhD in mathematics, I retrained and got a second PhD in statistical genetics. While working at UCLA, I turned into a bioinformatician. These days, I view myself as a biologist and biogerontologist. My lab generates lots of data which we distribute via Gene Expression Omnibus to the public. As you see, every year, I become more applied.
Why? It’s a good question. I’ve always loved math, but you should know a lot about biology if you want to reverse aging. The ideal training would provide knowledge in molecular biology, genetics, medicine, computational biology, and computer science. The problem is all of this would probably require 20 years of training. Our lifespans are too short for that, so everybody has to find their own angle.
I have always wondered what people coming from mathematics or physics think of biology when they encounter it?
In the beginning, my angle was statistical genetics. When you think about it, Mendelian inheritance laws are quite mathematical and probabilistic. So, for me, it was a natural transition from biostatistics to genetics and ultimately epigenetics. But we need to use many additional data including gene expression, histone modifications, transcription factor binding information, proteomics, metabolomics.
My first reaction was that there is so much noise compared to physics. In physics, you have elegant formulas that encapsulate the laws of nature, whereas in biology, there’s so much noise. The large amount of noise probably explains why simple statistical models work well in biology. Only a few people attempt to use partial differential equations for modeling biological phenomena.
Let us get to what you are most famous for, which is epigenetic clocks. This field keeps developing at a stunning speed. I watch talks from a few months ago and I see that a lot of things have already changed since then. So, could you give us a quick update?
In my lab, we are interested in third-generation clocks. We are looking for clocks that apply to multiple species at the same time. For example, universal pan-mammalian clocks. Several groups, including mine, are working on single cell methylation clocks. Many researchers are building clocks that respond to lifestyle interventions, such as exercise. Moving away from methylation, it would be nice to build similar clocks for other ‘omics’ data. Many researchers build clocks on the basis of other omics data, such as for chromatin, proteomics, and gene expression.
Why are things like pan-mammalian clocks, two-species clocks, and single-cell clocks important? What can they give us?
Single-cell clocks are important for understanding the mechanism underlying epigenetic clocks.
These clocks lend themselves for assessing whether each cell from a given person has the same age. Let’s take blood cells. Do all the blood cells of a 50-year-old man have the same age, or are some of them a hundred years old and some two years old? By the way, I think it’s the latter. I think that each cell has a separate age. Most current epigenetic clocks measure age in bulk tissue, which represents an average over thousands of cells.
It is a very important question: does each cell have its own age? It also relates to the question of how methylation relates to gene expression. This is a very difficult question, and most people believe that you do need to study the relationship between methylation and gene expression in single cells to avoid confounding by different cell types. So, that’s the case for single-cell studies.
Why do we need third-generation clocks that apply to multiple species? It’s all about enhancing translation. Imagine that you find an intervention that rejuvenates a mouse. There’s no guarantee that it rejuvenates a human being. You can enhance the chances of success by looking at highly conserved DNA. If you have an intervention that rejuvenates a mouse, a rat, a dog, and a cat according to the same clock, then chances are high that it will also work in humans.
But there’s a trade-off in terms of how precise the clocks are, right?
Yes. In general, when you have a clock that applies to multiple species, it’s less accurate than a clock that only applies to one species.
On the other hand, if you have a rat and human or a mouse and human clock, it should be very useful in translation.
Yes. And we have such a clock. We’ve developed it and even applied it. For example, we applied our human-rat clocks to a study of young plasma in rats in collaboration with Harold Katcher and Rudy Goya.
Are we getting any closer to understanding the mechanisms behind the correlation between methylation and age?
Yes, definitely. Fortunately, many people are working on it, including my group. We already know quite a lot. There was a paper from Ken Raj in Nature Aging where he describes the relationship between the hallmarks of aging and epigenetic clocks in vitro.
We have learned a lot from genetic studies in mice and humans, such as studies about developmental disorders, such as progeria. Just to give you some highlights, when you look at cytosines that relate to methylation aging, they are often close to polycomb repressive complex 2 binding sites. We also know that stem cell biology relates: often, stem cells are younger. We also know that there’s a connection to cellular identity. Epigenetic clocks relate to the loss of cellular identity that comes with aging. We have also a good understanding about which stress factors accelerate epigenetic aging: for example, metabolic stress and viral stress from HIV.
Epigenetic clocks come from a machine learning analysis, which means you start with a black box. To characterize this black box, you need to characterize perturbations that affect epigenetic age. For a biologist, black box predictors are not entirely satisfactory. They will, say, enumerate enzymes and pathways that play a role. Naturally, DNA methyltransferases and TET enzymes play a role. We know several other enzymes that play a direct role. Having said that, although we know a lot, the research is ongoing, so we learn more and more.
Considering aging in other species, which seems to be a hot topic now, how do you interpret the finding by Vadim Gladyshev that the naked mole rat ages epigenetically, even though it doesn’t age demographically?
There are actually three papers on epigenetic clocks in the naked mole rat. We published a paper in Nature Aging a few months ago. We have several clocks that apply to the naked mole rat including dual species clocks that apply to both humans and naked mole rats.
How to interpret it? Initially, I was puzzled because the naked mole rat appears to exhibit negligible senescence. But then I thought about it and realized that there’s another species that seems to have negligible senescence, and that’s humans. Humans live remarkably long lives now. If you measure human aging with clinical biomarkers, you will perhaps come to the conclusion that humans have negligible senescence in the first 30-40 years of life.
The question of why you can build an epigenetic clock for the naked mole rat is equivalent to the question of why you can build an epigenetic clock for humans. Both are very long-lived species, and this comes down to the question, what do epigenetic clocks measure? When I published the pan-tissue clock, I proposed that epigenetic clocks relate to the action of the epigenomic maintenance system. If this turns out to be true, then it stands to reason that all species have an epigenomic maintenance system. Therefore, you can build epigenetic clocks for all species, especially for long-lived ones. Our mammalian methylation project has shown that this is true. If anything, it’s easier to build clocks for long-lived species than for short-lived ones.
Another interesting topic is the germline reset, or the embryonic reset. Many geroscientists think this holds the key to solving aging. What can epigenetic clocks tell us about this reset event?
First of all, embryonic stem cells (ESCs) are perfectly young, as you said, as well as induced pluripotent stem cells (iPSCs), which is an interesting insight. Also, passaging of those cells doesn’t affect their epigenetic aging. If you have embryonic stem cells, and they divide, and you culture them in a dish, it doesn’t accelerate clocks. I showed this back in 2013 in the pan-tissue clock.
Recently, Vadim Gladyshev had this paper where he revealed a fascinating early rejuvenation event. I really liked the paper. People debate whether this is valid, because epigenetic clocks are usually trained on adults, or postnatal tissues, and so, it’s risky to apply these biomarkers to tissues collected during early development. But Vadim’s team evaluated many different epigenetic clocks in several data sets. They all pointed to the same early rejuvenation event.
The question is, which biomarkers are usable just a few days after conception?
Exactly, and this highlights the great advantage of epigenetic clocks. They are “life course clocks”. I don’t know any other biomarkers of aging that applies to fetal tissues as well, because most other biomarkers measure organ dysfunction.
I think it’s really interesting that we know for sure that such a rejuvenation event does happen, it just cannot be any other way, and it’s exciting that this has been confirmed by epigenetic clocks.
Yes, I agree with you. Back in 2012, when I first saw that epigenetic clocks reveal that iPSCs and ESCs are perfectly young, I was quite pleased with this finding. It makes intuitive sense.
I’d like to go back to that paper about the hallmarks of aging and epigenetic clocks. It covers a lot of ground, so can you give us just the gist of it, things that you find most important or surprising?
The most surprising to me was that cellular senescence doesn’t seem to be related to epigenetic clocks. But everyone knows that cellular senescence is important in aging, right? Senescent cells clearly increase with chronologic age. Before seeing our results, I would have said that a senescent cell must be epigenetically older than a non-senescent cell. The fact that epigenetic clocks are disconnected from cellular senescence is puzzling. I think more work is needed here, it’s an opportunity to understand the dichotomy between these hallmarks of aging.
Some say that cellular senescence is not really well-defined. Could that be the problem?
Maybe. I really admire the experts in cellular senescence for their honest communication. They are the first to point out that senescent cells are very difficult to define. To begin with, it depends on the cell type. There’s no consensus on what cellular markers to use for senescence. I expect that cellular senescence is very cell-type-specific. However, if we go back to the paper from Ken Raj, replicative senescence is very well-defined, as well as oncogene-induced senescence. Ken used these gold standard interventions for inducing senescence and found no relationship with epigenetic clocks in vitro. So, though senescence is an ill-defined term, we can still confidently say that all these different ways of inducing senescence don’t seem to relate to epigenetic clocks in vitro.
This reminds me: many people are saying that the field needs a unified, or at least an established, theory of aging. Do you think the lack of such theory is a problem? How can we proceed without it?
We should proceed even without having a theory because we cannot wait. To reach consensus, we’d have to wait decades. But yes, the lack of a theory of aging has consequences. If we had a good theory of aging, it would help the regulatory process at the FDA.
In one of your talks, you called it “Catch 22”, which I think is a great metaphor. So, what can be done about it, and when do you think we will see epigenetic aging clocks used in human clinical trials?
The good news is that epigenetic clocks are already being used in clinical trials. There’s this company called Intervene Immune, founded by Greg Fahy and Bobby Brook, and they are using GrimAge and other epigenetic clocks in clinical trials. I could name several other groups who are using epigenetic clocks in clinical trials.
Right, the thymus rejuvenation study, TRIIM.
Yes, and they use epigenetic aging as their primary endpoint, if I am not mistaken. Most researchers will agree that we need to use multiple biomarkers of aging at the same time for clinical trials. It is a bit frustrating that we haven’t come up with a consensus view of which biomarkers should be used. But we have come a long way.
Initially, there were good reasons to be skeptical about epigenetic clocks. I remember submitting a grant proposal promising to develop an epigenetic clock for mice. The grant was rejected two times because the peer reviewers had good reasons to doubt the utility of clocks. Since then, the field has changed. My impression: most scientists who study age-related molecular changes will measure methylation. So, epigenetic clocks are increasingly being used. I think it would be interesting if more people would measure epigenetic age in clinical trials in humans, at least as a secondary outcome, because there’s always an opportunity to make a discovery.
What do you think about the TRIIM study, and did we really see rejuvenation there? The results were quite remarkable.
I certainly liked the study. As a statistician on the study, I saw the data firsthand and was impressed by the results. The major limitation was the small sample size, only nine participants. This does give me pause. Fortunately, Greg Fahy and Bobby Brooke are doing a Phase II clinical trial. I think they’ve already enrolled about 30 people. By the way, I’m one of the participants. But we haven’t generated any methylation data yet. Only after all samples are collected, we’ll carry out the methylation study.
We know that epigenetic aging can be transiently affected by factors such as stress. How serious are the implications for collecting methylation data?
I think it only has a minor effect. You only detect an effect if there’s massive, prolonged stress, such as PTSD, or the most severe forms of depression. Short term stress probably doesn’t affect our current methylation clocks.
Still, as Morgan Levine has discovered, when you take blood from the same person several times over a short period of time, there can be a pretty big variability in the results.
Everything is relative and depends on the biomarker. The original version of PhenoAge was highly variable. By contrast, GrimAge is very robust to technical noise. It’s about 4% of technical noise. If you compare GrimAge to other biomarkers, such as cholesterol or glucose levels, you will see similar noise levels there. Epigenetic clocks are remarkably robust compared to what else is used in the clinic. We just released GrimAge version 2, which has even less technical noise than the original version of GrimAge. I would say that the issue with technical noise in epigenetic clocks has really been solved. Select clocks are ready for prime time. We are already using them in clinical trials.
Both you and Morgan are currently with Altos Labs. I know it’s kind of a secretive company, but maybe you can tell me something about what’s going on there?
I don’t speak for Altos, but I would say that we are not a secretive company. Quite the contrary, I would call us an open science company. Altos is very comfortable with sharing and publishing results. Many of those who have joined Altos are academics who believe in open science. Altos doesn’t view itself as an anti-aging company. Rather, Altos aims to promote cellular health and resilience. Clearly, cellular health declines with aging, but many stress factors impact health. Therefore, Altos is taking a broad view. Altos aims to boost the resilience of cells to various stresses. Altos is also very interested in the rejuvenation paradigm that started with Yamanaka’s OSKM factors.
That’s one of the few things we actually know about Altos, but we don’t know whether it pursues any other avenues.
You need to remember that Altos is just a few months old. Altos Labs aims to promote cellular health and resilience using the paradigm of reprogramming. The founders, including Rick Klausner, recruited several academics. You can just look at what these academics have been working on to extrapolate what they will be working on in the future. Several people are working on the integrated stress response, like Peter Walter. Juan Carlos Izpisua-Belmonte pioneered partial reprogramming. Wolf Reik also worked on reprogramming and the connection between aging and development. But others pursue other exciting avenues.
What is your opinion of cellular reprogramming? After all, it’s one of the hottest topics in geroscience.
I clearly like this approach, but I think of it as one strategy among many others. I’m really glad that different companies and researchers pursue different avenues, since it diversifies our risk. If one of these approaches works, it will change the world. Take senolytics, for example. I thought senolytics were a beautiful idea. Or exosomes, young blood – those are cool approaches too. I like caloric restriction mimetics, exercise mimetics. It’s good that many researchers and companies pursue this idea of using Yamanaka factors, because it will take a lot of work and a bit of luck to optimize this intervention.
What do you think about commercially available epigenetic clocks?
I never did a comparison of different vendors, so I cannot quite comment on what is out there. I give people the benefit of the doubt. Several companies were started by superb scientists. I started the nonprofit Epigenetic Clock Development Foundation. The CEO Bobby Brook collaborates with many groups all over the world to promote rigorous epigenetic clock testing in humans and other mammals.
The companies that work in the field can be distinguished by what kind of samples they collect. Some companies use saliva, which is easier to collect but might have certain drawbacks. Other companies focus on blood. Companies also use different technologies for measuring methylation levels.
Do you maybe think that it’s just too early for such products to hit the shelves and that this can even cast a shadow on the whole field of geroscience?
I wouldn’t go that far. I think we should definitely educate the public not to misinterpret epigenetic clock results. This is my fear. Let’s say that a 50-year-old man gets a result that puts his epigenetic age at 60. I fear that such a person may misinterpret this finding. There’s actually a highly complex, non-linear relationship between methylation age and lifespan.
Should we forbid commercially available clocks because of this concern? My answer would be absolutely not because I believe in giving people access to information. Here’s an analogy: people can also buy a blood pressure monitor or a glucose monitor, and those parameters are also highly predictive of how long you will live. Or even weight! Should we forbid measuring all those parameters because people may misinterpret the measurements? Of course not.
Do you feel like there’s sort of an arms race in the field of epigenetic clocks?
No, I don’t think we’re in an arms race. Let me use an analogy again. Take mobile phones. The are many providers, there’s Android and iOS, et cetera, and every technology has slightly different features, but on another level, there’s a convergence. I feel this is a very good metaphor for epigenetic clocks. There are many different platforms, but they all attempt to measure the same thing: biological age. This is probably frustrating to the reader of scientific literature because papers use different clocks. It would perhaps be desirable to standardize things to enhance “reader friendliness”. However, it will probably be impossible to get consensus.
In this context, do you think that some researchers maybe go shopping for a clock that will give them the desired results?
Yes, this represents a moral hazard. It takes inner strength to be honest with oneself and acknowledge that a finding is weak or nonexistent.
The best way to write a paper is to use several clocks or to explain why a particular clock was chosen. If you have space limitations, you should give a rationale why you used one clock but not the others, and then report the results of other clocks in a supplement.
I also want to make another critical point: not all clocks are equal. Clocks have different properties, so it shouldn’t be a plurality vote. For example, when I analyze fibroblasts, I use our skin and blood clock because it was tailored to fibroblasts. On the other hand, when it comes to mortality risk, I use GrimAge, because it was tailored to predict human mortality risk based on blood methylation data. I have a pretty good sense of what clock to use in what situation. I don’t want to confuse the reader with ten different clocks.
