The Longevity & Aging Series is a collaboration brought to you by Aging (Aging-US) and FOXO Technologies. This limited series of video interviews invited Aging researchers to speak with researcher and host Dr. Brian Chen. Dr. Chen is an adjunct faculty member at the University of California San Diego.
—
Below is a transcription of the third installment of the Longevity & Aging Series, where Dr. Steve Horvath and Dr. Brian Chen discuss the evolution of aging research and epigenetic clocks. Visit the special collection of Dr. Steve Horvath’s research published by Aging (Aging-US).
Hi, I’m Brian Chen from FOXO Technologies. Welcome to our series with the journal, Aging. Today we have a previously recorded interview that I did with my former mentor and long-time collaborator, Dr. Steve Horvath reflecting back on some of the work that he’s done on epigenetic clocks and epigenetic aging. So, hope you enjoy the interview …
I think back to your first publication, back in 2011 on epigenetics aging and age, and now we’re in 2022. It’s been over 10 years and a lot has happened in that time. So can you reflect now – I’m giving you this moment to think back – it’s been a whirlwind. I know. How are you feeling about it? Which directions are going that excite you?
Yeah, I feel good about it. I mean maybe to tell everybody the very first epigenetic clock in the modern definition of the word, the term epigenetic clock has been used in previous publications, but my personal definition is it’s a multivariate predictor of age. So a regression model is used. So the very first, according to that definition was published in 2011 as an accident. So this was a study of sexual orientation and where we had saliva samples from discord and twin pairs. And there was no signal whatsoever for sexual orientation in these men. But then I was very interested in aging effects. And then I discovered that one can build these very accurate estimators of age. At the time it was an acute finding. You can spit in a cup, you can measure age, we all loved it.
I should mention there was a doctoral student Sven Bocklandt and my collaborator, Eric Vilain who studied sexual orientation. And anyway, it was a curious finding, but it was largely ignored. But we always hoped that one could develop a saliva based test. You take a buckle swap, you spit in a cup and it tells you a lot. We were hoping that it’s not just calendar age or chronologic age, we were hoping that it would predict mortality risk, that it would be useful in the clinic. And by now there are many companies that commercialize that idea and the companies differ a little bit, some collect saliva, others collect blood, but that idea has made so much progress through many people because the devil lies in the detail. We had this prototype, we had the first study that showed feasibility, but the devil is in the detail.
How do you reduce technical noise? How do you lower the costs? I really have to be very grateful to the field that took on that burden. By now, people can do Google searches, find a company, measure their epigenetic age. Now I do want to say a bit as a caveat there are many clocks by now, and the clocks have different properties, they really do. I think you noticed that during my presentation, I mean, some clocks do relate to immunosenescence. Other clocks do not. Some clocks are wonderful predictors of mortality risk, GrimAge, for example, others like my pan tissue, not really. And so the challenge right now that we face in the field is to create benchmark data so that people can evaluate these clocks. It’s also a communication problem. Imagine I publish a finding with clock A, another person publishes a finding with clock B, and we need to kind of be able to translate the results from one study to the other.
We, of course, make our software available. It’s all publicly available. I think it’s good practice to evaluate many clocks at the same time. So in my early papers, we only had one clock that applied to different organs, one or two. But lately I really evaluate four or five clocks. Sometimes we get discordant results. We find condition A relates to clock one, but not clock two. For the average reader, this may be very frustrating. But I have by now a pretty good intuition what I expect. For example, I know that my pan tissue clock and the skin and blood clock relate to stem cell properties. Whereas other clocks GrimAge, PhenoAge, and EEA relate to immunosenescence and many other stress factors.
So I have a pretty good understanding why some clocks relate and others don’t, but the average biologist cannot spend all that time learning the nuances of different clocks, so we really need to educate people about it. A drastic measure would be to not allow all of these clocks. One would say, take only the following five clocks for all future publications to enhance the correspondence of findings. But I don’t want to do that because the field is so active, there’s so much creativity. I mean, people have wonderful ideas to come up with ever more powerful epigenetic clocks, new statistical models, new training sets, new conditions for filtering cadences. So I just want that the field has this creative moment, where many clocks will be developed.
I think people that don’t work with you, and we’ve been long time collaborators for a while now, you are the kind of the open source version of research where you’ve been publishing code long before it was kind of the recent fad so that people can verify and improve upon your work. So that’s good to you.
I’ve also been the biggest benefactor of this open science movement. When I published the pan tissue clock, it was based on publicly available data. I didn’t have a dollar in research funding. It was all downloaded data, and so I want to give back. Whenever I can, I deposit data in the public domain and gene expression omnibus. But in general, these initiatives have really led to, I don’t know how many millions or trillions of data points. So it’s really wonderful.
So you mentioned your last comment about the poor biologist or some new novice coming into this field, there’s so many clocks out there and you are the world’s expert. So you understand which one they might look for, for their application. But just as a Cliff Notes, what is your go-to panel that you look for? Then is there a resource that you know out there that kind of guides people as to which one they might want to choose?
I mean, fortunately people have gone through the exercise to write reviews. I can tell you in a couple of sentences, what I do, it may help. So first of all, when it comes to mortality risk, mobility risk, and if I have blood or saliva, I would use GrimAge. That’s my primary clock. Now for the sake of sensitivity analysis, I also then evaluate the PhenoAge clock and also EEA, the Hannum clock. So I often publish that in supplements and also there’s a primary clock and secondary clocks just to convince myself that the results are semi robust, because in a best case scenario, you have, for example, an anti-aging intervention, let’s say vegetable intake. And in a best case scenario, it shows a beneficial effect, according to all of these clocks, which by the way, is the case, vegetable intake does relate in the way or so anyway.
However, in a different application, for example, HIV infection. HIV infection has an effect on blood cell composition and in certain ways it’s a trivial finding to say, HIV affects blood cell composition. Or take COVID. COVID affects blood cell composition. Then I actually would not use GrimAge or PhenoAge. Why? Because they’re confounded by blood cell types. Rather, I would go back to my pan tissue clock or the skin and blood clock, which measure more properties of hematopoietic stem cells. And there’s another application related to precursors of leukemia. So the somatic mutation and TET enzymes and DMTs lead to clonal hematopoiesis of indeterminate potential.
But again, it inflicts hematopoietic stem cell. Again, I would use the pan tissue clock. So when it comes to in vitro studies, there’s a clear winner: skin and blood clock. I would use that as primary readout. And that’s already it. In my papers, some people will say, well, Novartis should use clock X, Y, Z. And I agree with that. There are wonderful new clocks that have been developed, but I just want to tell you what I use in my papers at this point, without value judgment. I wouldn’t say they’re better than others that are out there.
So I think the bottom line is: Evaluate all clocks out there, but there’s going to be hundreds and thousands eventually, and you got to pick.
I always say it’s not a democracy, not all clocks are equal. I use another metaphor, vehicles. The fields of epigenetic clocks produces all these vehicles. You have the Ferrari for racing, you have the truck, you have the tractor, you have a Jeep, these vehicles serve very different purposes. And so don’t apply the wrong clock to the wrong setting. Having said this, they all, all vehicles are a mode of transportation one way or another. So there is a commonality.
I think that also focusing on clocks that other people have published on that are already somewhat more established. Those would be what I would prioritize because there’s already a lot of information that you can compare to. So those are my two cents. Then I think the other point to make is you do have an online calculator that’s free and the data is not stored anywhere. People upload it, it calculates it, it destroys the data, but it gives them emails them a report that has all of those clocks. This is a nice way for when you’re publishing a paper on your method section to have something that’s standardized and reducible.
That’s correct with the code is available. Absolutely. I want to mention for the future lately, both you and I, we’ve worked on animal clocks, and UK is here for a wonderful array for mouse studies. But they are really a lot of clocks for mice. I need to tell you, different mouse tissues. We have pan tissue clocks, but also clocks that are specific to maybe ear samples or blood samples or liver samples. But then there’s a new generation of clocks. We call it third generation clocks and these clocks apply to multiple mamalian species at the same time. The purpose is to enhance the translatability of research findings. So imagine you do have a mouse study. You do show that an intervention rejuvenates the mouse, according to an epigenetic clock, you would like to increase the odds that this finding translates to humans.
One way to do it is to look at cytosines that are highly conserved between all mamalian species and more than that, to use the same formula, the same regression model to measure age, both in mice and in humans at the same time. And so this is what I call a third generation clock and they have recently become available. So we have dual species clocks, human mice, human dog, human cat, human pig, human, anything. So look out for that. So there are already a couple of studies that have used these multi-species clocks, but I hope that they will turn out to be useful. So I’m kind of sitting here with a baited breadth and see whether this third generation of clock fulfills their promise.
So you alluded to the multi species clocks, so that’s one direction that you and your group is moving towards. What are some other kind of new, interesting, exciting areas that you see yourself or others or other trainees have gone in?
Yes. So there’s many people work on this idea. I want to call it single cell methylation clock, but one doesn’t have to be too rigorous. Let’s just say clocks that can deal with a very low amount of input DNA. So typically all the clocks I’ve shown you use, let’s say 250 nanogram of DNA. So roughly speaking, maybe a hundred thousand cells or so. But can we lower the requirement on DNA so that we can measure methylation age, maybe in a thousand cells, or as an end point in a single cell? And I remember Vadim Gladyshev … as a single cell methylation clock, but this is a very active area. The whole field wants to have that. Let’s see. Then people develop very precise clocks for forensic applications. Similar, as I said, in forensic applications, you often have only very little DNA available. Now a forensic clock does not want to estimate mortality risk or morbidity risk, not interesting.
Rather, for forensic application, you want very accurate measure of chronologic age. So it’s a slightly different task. But the other very big area of research is really to understand the biology of the clocks. All the clocks that I describe were developed using machine learning methods. And there’s so much work to be done to understand which pathways, which transcription factors relate. I think we have really managed to gain quite a lot of insights into the biology of these clocks. I mentioned various progeria, but there are many other conditions that have been associated there. There are genetic studies of epigenetic clocks, which SNPs relate to epigenetic age acceleration. We published a paper where we described 137 novel SNPs, genome wide significant for various epigenetic clocks. So just more biology, more genetic studies, and of course the most exciting thing for me are human clinical trials. I want that these clocks will be used as surrogate endpoints in human clinical trials of anti-aging interventions.
So in those clinical trials, which clock do you think is the most important to look at?
I mean, if you force me, I would go with GrimAge, at this point. Now the great question is: whether GrimAge shows reversal for gold standard interventions? So imagine we have a … gold standard means, it truly prolongs human life. And so it would be nice to demonstrate that such an intervention actually has the expected effect on GrimAge after one or two years of this gold standard treatment. GrimAge is reversed by, let’s say five years, but there’s a problem. We don’t have a single gold standard intervention for humans that at least that affect the methylone.
I mean, a bit ridiculous. We obviously have medications for hypertension. We have medications for lowering cholesterol, all of these, or various diabetes drugs. Yes. They prolong lifespan. But some of them may not affect the methylone. But these are very important studies I would say, to show whether interventions that both prolong human lifespan and have an effect on the methylone, that they have the expected behavior on the expected effect on these clocks.
So certainly we don’t know what those gold standard interventions are. Usually, there’s just like your lifetime, eating rights, sleeping well, all those things, but there’s no clinical trials to prove- –
To prove lifestyle factors have all been linked to GrimAge. No question. So it is smoking and exercise all of that.
Then for mice, I think in some of the epigenetic clocks with mice, they’ve looked at, but there are gold standard interventions that extend life known in laboratory animals.
In mice, I want to say we are really, really close to nailing it. I know there’s no question. We have clocks, mouse clocks that show the expected behavior with respect to caloric restriction, growth hormone receptor knockout, leading dwarf mice. Even parabiosis this idea of young plasma. So in mice, there are very exciting results. Questions, whether these results for mice then translate to… Whether these insights allow you to develop human clocks that have that expected behavior.
Right. Because you now understand the biology, and so you don’t have to rely on the machine learning so much too.
That’s kind of the idea. Understand the biology in a mouse or other animal model, and then use that insight to build what I call causal clocks. Causal clocks means if you rejuvenate age, according to a causal clock, there’s a guaranteed benefit for your health span, for human health span. That’s my definition of a causal clock.
Well, certainly thanks for this discussion today. It was great to go down through memory lane with you. I know we intersected at multiple points there and then super exciting time for humans, also for mice even more so, because now we’re really close to understanding the biology of their aging using these clocks. Congratulations on at least a decade’s worth of work. I know firsthand you’re juggling a lot, but I think that you’ve advanced this field pretty fast relative to other fields that I’ve seen.
Oh, thanks so much. Above all, I was very, very lucky at the right time. Timing is everything. Thank you so much Brian, I’ll talk to you soon. Thanks to the audience for listening.
Okay. Thanks Steve. Bye.
Click here to read the full Special Collection on Steve Horvath’s Publications in Aging (Aging-US).
Longevity & Aging Series Episode 2: Dr. Steve Horvath’s Special Collection in Aging
Longevity & Aging Series Episode 1: Drs. Alex Zhavoronkov and Frank Pun
AGING (AGING-US) VIDEOS: YouTube | LabTube | Aging-US.com
—
Aging (Aging-US) is an open-access journal that publishes research papers bi-monthly in all fields of aging research and other topics. These papers are available to read at no cost to readers on Aging-us.com. Open-access journals offer information that has the potential to benefit our societies from the inside out and may be shared with friends, neighbors, colleagues, and other researchers, far and wide.
For media inquiries, please contact [email protected].