Biological Age & Stress

hannah_went:
In today’s episode, we talk to Dr. Jesse Poganek. Welcome to the Everything Epigenetics podcast, Jesse.

jesse:
Thanks, happy to be here.

hannah_went:
Yeah, I’m extremely excited to speak with you today. I’m personally very interested in stress and how it actually affects the biological aging process. I tested my methylation markers quite some time ago and had some accelerated aging. I definitely think it is probably due to me just being a naturally stressed out person. So I’m personally super interested in the subject and your work in the field. So before we hop right into it, I’d love to give you the opportunity just to introduce yourself our listeners and tell them about yourself.

jesse:
Sure, so I’m a research fellow in Vadim Gladyshev’s lab at Brigham and Women’s Hospital and Harvard Medical School. Got into the aging space about two years ago now and been mainly working on what we’re going to talk about today, this connection between very severe stress and aging using these aging clocks, which I know you’re very interested in.

hannah_went:
Yeah, definitely. I appreciate that introduction. So your paper called biological age is increased by stress and restored upon recovery. That’s really going to be our main subject today. And you examine aging and surgery, pregnancy and COVID-19. I know biological age can be defined in a lot of different ways as well. So we’ll tighten that question as we go through here. And you do that in humans and some in mice as well. So I wanna break these down and really dive deep into each of those categories of your paper. So let’s first start with that surgery arm of the study. Can you just explain what you looked at, what you found, and take us through that?

jesse:
Sure, so actually if we go back one step from the surgery,

hannah_went:
Yes.

jesse:
the reason we got into this was because we started with this very classic model in aging called heterochronic parabiosis. And this is not particularly relevant to humans and it’s a very, very kind of experimental, artificial kind of thing where you connect surgically two mice together. And we use that to demonstrate. just conceptually that you can have these situations where biological age, as read out by these methylation biomarkers, is transiently increased.

So what we found was that when we exposed young mice to old blood, their age increased, and then when we separated the mice again and allowed the young mice to recover, their age decreased. So that proved that you can have these situations where age is transiently increased. And I just wanted to highlight that because the… we talked briefly already about this question about, are we measuring biological age? And this is a very important question from the outset with regard to the study and with regard to methylation biomarkers in general.

The parabiosis study is interesting because this is not in the pre-print, but it’s going to be in the published version, which I hope will be published very soon. But we were able to actually substantiate our methylation results with. results also at the transcriptome level, so the level of gene expression, and at the metabolome level. And we find exactly the same effects. The age is transiently increased upon exposure to old circulation and restored upon recovery. So this is not directly related to stress, but it proves this concept that you can have these transient increases. So that’s what got us into this space.

Now let’s talk about surgery. So from that… starting point we asked this question, are these transient increases a general feature of biological aging? And part of, you know, a component of the parabiosis, not the main component because the main component is this age circulation exposure, but we hypothesized that a component of the parabiosis effect may be due to very severe stress because of course, you know, for mice it’s very stressful to be connected to another mouse for three months, right?

hannah_went:
Great.

jesse:
we asked the question, what situations happen to humans that we could assess that represent a very severe physiologic stress? We started with this surgery data set, like you said. And so this was a publicly available published data set of elderly humans who were undergoing surgeries. And we were able to analyze three different kinds of surgeries. So the first was these traumatic hip fractures that had to be repaired with emergency surgery. So these people, these were elderly people again, undergoing a very severe and unexpected surgery. And what we found very interestingly was that before their surgery and after their surgery, in the space of less than 24 hours, there was this very significant increase in biological age, as read out by multiple methylation biomarkers.

And what was even more interesting was by the time they were discharged from the hospital, about seven days later on average, they had returned back to the baseline. So that was the first human example we had of a situation where you could have a transiently increased biological age. And that data set was interesting because there were two elective surgeries that we also analyzed. So the contrast is that the trauma hip surgeries were unexpected and, you know, emergent.

These other surgeries were elective, the people knew they were happening, they were selected as good surgical candidates, they did the pre-op prep, et cetera, et cetera, et cetera. And we didn’t find any effect on the biological age. Actually, in one of the data sets, the elective hip surgery, we actually found that they had slightly decreased biological age overall before their surgery,

hannah_went:
Thank you.

jesse:
and we think that that’s probably reflective of the pre-op preparation. But it was instructive because it made us think that really the severity of the stress, you know, the physiologic stress could drive the effect. And that’s what led us to the other models, which I know you want to ask about.

hannah_went:
Yeah, no, I really appreciate that answer. So that’s really interesting because the elective surgeries, you know, you said they even saw positive changes in their biological age, right? Those who knew about the surgery, they were going that for the pre-op procedure, right? So I find that very interesting in how that can have application in the real world one day, right? If people understand what they’re going through, then it could have almost that… counterintuitive effect where they’re actually gaining something from that rather than going under the stressful situation.

jesse:
Yeah, but just to be clear, so first of all, it’s educated speculation, right? So we don’t know what happened with these people, right? But nevertheless, the age was decreased at the first time point. So possibly it reflects that or not. The reason why I think it fits in though was what we found was actually, so they were decelerated at the first time point, then they came up to what we would expect their age to be after the surgery.

hannah_went:
Mm-hmm.

jesse:
So that could fit into the model because it could be that they’d still experience the physiologic stress of the surgery, but because they had decreased the age before the surgery, then they were kind of set up for success. And that can be directly relevant to future applications.

hannah_went:
Gotcha. Yeah, definitely. Very, very exciting there. So let’s move on to one I’m, I think, probably most interested in as well. Can you explain what you looked at and what you found in the pregnancy arm of the study? And I think it’s really interesting too, maybe you can touch on before we go further into these stressors. When you talk about stress, I love how you looked at the three models because it’s like when we’re undergoing this stress and going through this mechanism of action, our body doesn’t know if we’re having a surgery or we’re going through a pregnancy or have COVID-19, which will be the last model that we go through. So can you define stress the best you can in terms of how you’re

jesse:
Mm-hmm.

hannah_went:
measuring that?

jesse:
Yeah, so this is not a consensus or general definition of stress, but the way we’re using it in the paper is to represent these very severe physiologic and metabolic demands. Like something you would undergo with a major surgery, like pregnancy, which of course many people undergo, but it’s extremely physiologically taxing. You have to grow a whole other human inside of you.

hannah_went:
Yeah.

jesse:
And looking ahead to the COVID-19, we analyzed very severe COVID-19, and we can talk about that more later. So, yeah,

hannah_went:
sure.

jesse:
we’re not necessarily talking about, oh, I had a hard day at work and I feel very stressed, but we’re talking about very severe physiologic stress,

hannah_went:
Absolutely,

jesse:
so things that

hannah_went:
thank

jesse:
the body

hannah_went:
you for that.

jesse:
is subjected to, yeah.

hannah_went:
Yeah, thank you for that clarification. So yeah, let’s hop right in. What do we know about pregnancy and aging in general? And then how did you look into that further in your study?

jesse:
Yeah, so we know that, of course, as I said, aging is a very physiologically demanding process, and there are a lot of interesting connections to aging. So some of the work before our work showed, for instance, that the number of pregnancies over the lifetime is associated with accelerated aging. So more pregnancies over the lifetime, more accelerated aging on average. And it’s even been suggested as a model of aging. So for instance, just jumping back to the parabiosis, in that system you have a young mouse connected to an old mouse. Some people have even suggested that pregnancy can be a model of parabiosis because you have,

hannah_went:
Mm.

jesse:
you know, a very young human, a fetus, inside an average middle-aged, or 30-year-old anyway,

hannah_went:
Right.

jesse:
person. So the connections are very interesting. We were interested again, just because of the physiologic demands that’s involved in aging. And we hypothesized

hannah_went:
sure.

jesse:
that over the course of aging, you could have one of these transient accelerations. And we specifically expected that the peak age acceleration would happen near the time of delivery. Because of course, that’s the most stressful point. And another interesting connection that I should point out is that if you look at the risk of diseases that people experience during pregnancy, things like gestational diabetes and preeclampsia, which is a disease of high blood pressure, hypertension, these are kind of classically aging-associated diseases, diabetes

hannah_went:
Mm-hmm.

jesse:
and hypertension. And the peak risk for those conditions is also near the time of delivery. So it all kind of fits together. And what we found, both in mice and humans, was indeed exactly what we expected. This effect of increased age over the course of the pregnancy, peaking near the time of delivery, and then resolving postpartum. And those results were very nicely consistent across both the mouse and the human systems.

hannah_went:
Definitely. No, thank you for that, that insight. work with a lot of people who test their epigenetic methylation markers. Interestingly enough, one week I was reviewing some of those outcomes with two women. And one of the clocks that you include, or algorithms, outputs, outcomes that you include in your paper is that Dunnington pace, that pace of aging, how quickly you’re aging biologically for every one chronological year.

And I was honestly a little nervous to go over these results with these women. They had a Dunnington pace. And for those listening, the scale is zero. 0.6 to 1.4 of about 1.3. And did not have much medical background information on them, come to find out they were both pregnant. So I thought that was very interesting, took a look in some of the literature, right?

We know that when women are pregnant, we typically see this advanced epigenetic age acceleration. And then funny enough, we did end up retesting. One of the women, six months postpartum, and saw a huge, statistically significant reversal in that Dunedin pace as well. It went back to about.9. So I thought that was very interesting and follows that same trend that proves and shows that we can actually reverse these stressors.

jesse:
Mm-hmm.

hannah_went:
So the third part of your study talking about COVID-19, obviously that’s been huge in these past couple of years, right, you see everything that’s going on, even with COVID-19 in China right now, which is devastating. So can you explain what you looked at in that arm of the study and what you found?

jesse:
Yeah, so, you know, again, going back to this idea of very severe physiologic stress, we were thinking, what else could we examine? And we hypothesized that a very severe infectious disease would be a good model to examine. And because we started this study in 2020, you know, we were interested to look at COVID-19 because, of course, it was at the forefront of everybody’s attention at

hannah_went:
Thank

jesse:
that

hannah_went:
you.

jesse:
time and still is to some extent. But we selected a cohort of patients who were very, very sick. So we’re based at Brigham and Women’s Hospital. And at that time in Boston, people were being sent to the Brigham because of how ill they were, because we have the ICU capacity and expertise, et cetera. So the less severe patients were being funneled away and the more severe patients were being sent to us. So… For the purposes of our study, that was a nice opportunity because it allowed us to select these very, very sick patients. So all of the patients in our cohort came to the hospital because they needed intensive care. So they were all in the ICU for approximately two to three weeks on average. They were all mechanically ventilated and they were all very, very ill. But yet they all survived

hannah_went:
Mm-hmm.

jesse:
because we were interested in looking at this recovery effect, right? So

hannah_went:
Yeah.

jesse:
what we found was that because these people were already ill and were coming here, because they needed intensive care, their age was already accelerated by the first sample we could analyze.

hannah_went:
Mm-hmm.

jesse:
So for instance, going back to that Dunedin-Pace scale, if I remember correctly, the first time point that we were able to analyze, they were already at about 1.25

hannah_went:
Oh wow.

jesse:
on that scale. We know now that the… the outcomes for men and women were very different. So we separated males and females in our analysis. What we found was that for females anyway, there was a partial recovery by the time they were discharged from the hospital. And for males, the same was generally true, although the effect was a little more heterogeneous. So if you looked at the individual trajectories over time, for males, there was much more kind of… stochastic behavior in the biomarkers, whereas for women it was kind of more of a defined trajectory where they were accelerated, accelerated, accelerated, and then partial recovery near the end.

And for males it was more heterogeneous. What we also found with males were that the individual trajectories were more heterogeneous. And so that’s something that would be nice. although we’re not powered to do it just because of the way our data was collected. It would be nice to analyze, for instance, you know, do people who have poorer recovery, do they have worse clinical outcomes, right? And I think that’s one of the things we’re interested in looking at in the future with these results. But, you know, what we analyzed was the recovery when they were discharged, and again, we found at least partial recovery. cases. And just to clarify again, these were for very severe COVID-19 patients.

hannah_went:
Mm-hmm.

jesse:
So it’s an outstanding question I would say for people who like me have been vaccinated multiple times and were infected with COVID-19, had a very mild course, you know, just felt run down for a few days but basically recovered very quickly. I think it’s an open question whether anything would happen in that situation.

hannah_went:
Mm-hmm.

jesse:
But nevertheless, for the very, very severe case, it seems to fit this model where the age is transiently accelerated and then at least a partial recovery happens.

hannah_went:
Yeah, no thank you for that. And I’m the same way, you know, was vaccinated, got all the vaccines I could, still ended up having COVID, but was very, very mild case. But I tell you the one day that I really felt it, oh man, I felt it. I was down pretty bad. So it would be interesting to have that. the data on that type of population with COVID-19 diagnosis too. One of the questions that I always get after we discuss anything regarding COVID-19, we actually published a longitudinal COVID-19 diagnosis with Cornell and their group and even with vaccines too. So that was a really interesting paper. We can always discuss later, but

jesse:
Mm-hmm.

hannah_went:
what about long haul COVID? Have you had any questions about that or looked at that? the question people ask next, you know, they’re still feeling those effects.

jesse:
Yeah, yeah, and it’s an excellent question, because like you say, it’s something that’s, you know, for a lot of people has become the reality, unfortunately, and it can obviously be very devastating. So I think it’s important, you know, it’s a very important question to do more work on. But as kind of a… you know, quirk, if you will, of our data set, it’s not something we were geared to analyze. So again, because of the way the resources were being allocated at the time these samples were collected during the very early part of the pandemic, the idea was that the patients would become very severe and be transferred to the Brigham for intensive care and then discharged to rehab facilities or lower tier hospitals for

hannah_went:
Mm-hmm.

jesse:
ongoing care. So… the point at which we were able to collect the samples was only during the most severe part of the disease and

hannah_went:
Thank

jesse:
early

hannah_went:
you.

jesse:
into their recovery. So it’s not something we were able to analyze, as nice as that would be. I strongly suspect that those data will come out because a lot of groups and a lot of the leading groups, I would say, in this field of methylation

hannah_went:
Thanks

jesse:
biomarkers

hannah_went:
for watching!

jesse:
are very interested in this. So I know, like you just mentioned, your… company with Cornell has a study and there are many other studies from other labs looking at you know for instance epigenetic age years prior to COVID-19 and the association

hannah_went:
Right.

jesse:
with the risk of infection so that was one paper that came out from Morgan Levine’s lab and it’s it’s just you know this example and your paper just a few examples out of many many out there so and you know there’s a huge push now to study along COVID.

hannah_went:
Mm-hmm.

jesse:
So I think those data are forthcoming, but it’s not something we were able to analyze in the paper we’re talking about right now.

hannah_went:
Understood. Yeah, I’m excited for that data to come out. It’ll be a long time coming.

jesse:
Mm-hmm.

hannah_went:
So did you know you were going to look at all three of these stressors or did they really just, did you just have the assets because you know you’re at a great university, you have a great position and have all of these resources. So did the paper start with one stressor and then turn into all three?

jesse:
Yeah, so I mean, first of all, we can’t discount the, how fortunate we are to be based where we are, especially being at the Brigham, because it gives us this opportunity to get these human patient samples. And so that’s really a wonderful resource. And the other thing I should say right now is we have so many wonderful collaborators who contributed data to this study for us to analyze. So very, very, very lucky for

hannah_went:
Absolutely.

jesse:
those. But no, the story of this paper is kind of interesting actually. So I mean, you may notice and depending on how versed your listeners are in reading scientific papers, it’s kind of a weird paper, right? We kind of jump around to a lot of things, right? Parabiosis, severe surgery, COVID-19. And actually, even in the peer review process, some people were less happy about that than others. They were saying it’s disjointed

hannah_went:
I’m sure.

jesse:
and strange. But I think actually the innovation is the fact that we look across these different models, right? Because most papers would perhaps choose one of these models

hannah_went:
Mm-hmm.

jesse:
to examine in depth, and that’s nice too. But I would say we still go in depth to all of these models, but there’s this interesting through line that came out by the end of this study, which is this idea of severe stress. So, you know. Like most things in science, I would say this study didn’t start with the idea of saying we’re going to look at very severe stress and see what happens. We actually started because we were interested in the question of what is the shortest time scale over which one of these methylation biomarkers can change

hannah_went:
Mm-hmm.

jesse:
by some kind of input, either a lifespan extending intervention or dietary intervention or an age accelerating intervention. And so, you know, we were starting with the parabiosis model actually because we were interested in lifespan extending effects of exposure to young circulation, which is

hannah_went:
Mm-hmm.

jesse:
the inverse of what we talked about at the beginning. And indeed we found that it’s the subject of another manuscript from our lab, and we separated the two because actually we find very different effects. So for the old mice, their age is… decelerated upon exposure to the young circulation, so

hannah_went:
Mm-hmm.

jesse:
they get biologically younger. And I say that confidently because we can actually also track them and they live longer. So I do think that’s a bona fide example of an age… a biological age reduction. But

hannah_went:
Mm-hmm.

jesse:
what’s interesting is that when you… so it’s the same setup, three months of parabiosis and then two months of separation. When the old mice are exposed to the young circulation, the age is decreased, and then after the separation, the age remains lower. So the effect persists longer. And that’s in contrast

hannah_went:
Mm-hmm.

jesse:
to what we’ve talked about, which is the effects on the young animals where the age is increased, and then it rapidly goes back down to the baseline once they have a chance to recover. So very different effects. But, you know, so that was the seed, right? This idea that… the age can be transiently increased. And then separately, we were looking into these questions about what are these short-term interventions that can modulate biological age? And it was just this nice, you know, confluence that we found that actually these other interventions that we were separately looking at were very reminiscent of the parabiosis effect, which was more of just a

hannah_went:
Okay.

jesse:
proof of concept. These other things that we examined, surgery and pregnancy and COVID-19. these were showing a very similar effect, where you have a transient increase that then resolves. And so by connecting these, I think we shed light on this very interesting and previously unknown facet of biological aging.

hannah_went:
Yeah, I know. I’m always interested to understand how these questions come about or how these papers become put together because they’re so intricate, they’re so detailed. It’s like when you have an answer to one question, you have a million other questions, right?

jesse:
Uh huh.

hannah_went:
We don’t know what we don’t know. So thank you for that background. And I personally think the paper, I have all of it right here, I think it’s put together. very, very well. I like how you look at three different aspects of stress and dive into each of them. I really like it, even though some reviewers may have said it’s choppy or goes from one place to the other.

So you mentioned, Jesse, a really good point when you started asking yourself some of these questions. So one thing that makes the epigenetic clock so exciting is that they can change, right? But there’s this huge criticism out always been a concern, you know, when’s that retest time, what’s the variability between that. So you did retest pretty soon after that initial test in some of these arms. So how do you respond to that type of criticism since you’re retesting so frequently?

jesse:
Yeah, so one of the interesting technical things we found, because we applied, at least for the human data sets, a suite of epigenetic clocks. So we’ve talked about the Juneteen Pace, but we’ve also used Horvath’s classic clock and other things your listeners might be familiar with, like Grimmage and Phenoage. So the standard set of clocks. And you can broadly divide these clocks into what some people call first generation clocks and second generation clocks. This is not a universally agreed upon division, but the idea, at least in the way I’m using the term, the first generation clocks, when you make the models, all the models are given is the chronological age and the methylation profiles. And the second generation clocks, they’re given, sometimes they’re given chronological age, sometimes not. but they’re given multiple parameters on which to train the model.

hannah_went:
Mm-hmm.

jesse:
So for instance, the Dunedin PACE is based on this really elegant study out of Dunedin, New Zealand, where this cohort of people have been tracked since birth, since the 70s, I think, when they were born. And they have these really in-depth phenotypic data sets, so the people will come in for like a whole day. and they’ll just do a battery of biological and physiological and medical tests on them. And it’s, again, it’s, I can’t say enough good things about the Dunedin

hannah_went:
Yeah.

jesse:
study because it’s just such a wonderful data set. I think it’s exactly what we need and need more of in the aging field, these longitudinal data sets. Because it goes to this idea, right, of how does biological age change over the life course? And we don’t know, in part because we just don’t have… good data to analyze at this point. But I digress. So

hannah_went:
Yeah. No worries.

jesse:
we applied… we applied this standard set of clocks, right? One of the things we found was that the first generation clocks, in general, were insensitive to these effects of severe stress. And the second generation clocks did pick up the effects pretty consistently.

And this… you know, again, I think that the part of the innovation of the study was… using these very different on their face models, right, of severe stress, sorry, severe surgery, COVID-19 pregnancy, very, very different models, but actually very consistent effects produced in the clocks, namely that the first generation clocks were basically flat line across, changing at least with the age as you would expect, although over relatively short timeframes, not a big magnitude of change, right? But the second generation clocks were uniquely sensitive to these effects. So, how do I respond to the criticism? I would say two things. Number one, we don’t know, which is a little unsatisfying,

hannah_went:
I’m

jesse:
but

hannah_went:
sorry.

jesse:
it’s just because the field is relatively new and it’s just the subject of active work. It’s not that people are not concerning themselves with these things. A lot of people are. You know, our

hannah_went:
Mm-hmm.

jesse:
lab is coming up with approaches to answer these questions directly and many other leading labs in the field, particularly I would highlight Morgan Levine’s lab

hannah_went:
Mm-hmm.

jesse:
and also of course, Steve Horvath, who is one of the founders of the field, right? Very, very concerned with these questions of how do these clocks work? What are they specifically measuring? And then what facets of aging are different clocks measuring to different degrees?

hannah_went:
Mm-hmm.

jesse:
because the underlying assumption, and one that I agree with, is that these different clocks are measuring different facets of aging.

hannah_went:
and

jesse:
Right, and probably different combinations of the different facets to more or less of a degree. So, yeah, one question is, sorry, one response is that we need to do more work to understand, and people are doing

hannah_went:
Absolutely.

jesse:
it. But the other response I would say is that What our study offers is the idea that you can have different clocks for different purposes, right? So if you’re

hannah_went:
Mm-hmm.

jesse:
interested in just testing the chronological age of a sample, you know, people talk about this in the use of forensic science, for instance. So if you need to identify how old a murder victim, I guess, is

hannah_went:
I’m

jesse:
or something.

hannah_went:
going to go to bed. Bye.

jesse:
you know, then you would use, but actually that’s kind of a, that’s kind of even a far-fetched example, maybe a murder victim, but actually I’ve talked to one scientist who said, well, the problem is that, the problem with your data is that if you have a biomarker of age, it shouldn’t be affected by stressors. It should just predict the age. And

hannah_went:
Hmm

jesse:
in some ways that’s a very fair critique, right? But what our data indicate is that, you can still have this. If you use the first generation clock,

hannah_went:
Mm-hmm.

jesse:
they don’t care about the severe stressors. And if you’re more interested in modulating the stressors or maybe fine tuning aging, if that’s what it turns out to be, that the second generation clocks are measuring more precisely, then use a second generation clock. So, again, this is educated speculation, right? Because the data have to prove these explicitly, and we’re working hard on that. But I think

hannah_went:
Yeah.

jesse:
that as a proof of principle, that’s what our data have to offer.

hannah_went:
Absolutely. I could not agree more with that answer. That was absolutely great. I think if you ask anyone in this space, you know, their first answer should match up exactly with yours is we don’t know, right? We need more data. We need more information. And I like how you broke it down in the first generation and the second generation clocks because, you know, sometimes the first or the second may receive more criticism than the other, but I think they’re both very useful for those exact reasons that you just stated, right? Another example where the first generation clocks were really used in the beginning is to date refugees, to see if they were able to seek asylum,

jesse:
Yeah.

hannah_went:
right? If they were above a particular age. So I think they’re very useful in their own different purposes. We just need to understand how to use them in the actual application, which I think we’re still discovering. And that is why I just appreciate your work so much is because you do, you look at every clock, right?

jesse:
Oh,

hannah_went:
value

jesse:
not every

hannah_went:
going…

jesse:
clock.

hannah_went:
like…

jesse:
I mean,

hannah_went:
you’re…

jesse:
you know, there are more clocks than you could really look at, but we look at a standard set and one that I think is representative

hannah_went:
Thank

jesse:
of what’s

hannah_went:
you,

jesse:
out

hannah_went:
yes,

jesse:
there.

hannah_went:
thank you.

jesse:
Yeah.

hannah_went:
A good representation.

jesse:
Just have to give credit where credit is due, right? Because a lot of people develop clocks and there’s a lot

hannah_went:
Exactly.

jesse:
out there.

hannah_went:
exactly multitude of different

jesse:
Mm-hmm.

hannah_went:
clocks. So you do a, you go through a great grouping of those clocks. And I think something that could be very valuable is even if we go back and look at a lot of these interventional trials that were published or even the epidemiological trials that have been published, and we apply newer clocks to that older data set. So I think that’d be really interesting just so we can unlock more information and again say maybe there’s, the stressors don’t have an effect on the generation one, but definitely a generation two. So super exciting work to be done there. And that leads me into this next question. So we’re gonna talk a little bit further about these clocks. You have your first generation, your second generation, and then there’s this component added to those clocks, I guess. You’ll do a better job of explaining

jesse:
Ha ha.

hannah_went:
this, but what is a principal component analysis? What does that mean when you have a clock and its principal component analysis corrected? Can you talk about that?

jesse:
Yeah, so you know it just goes back to this idea that we’re talking about, right, that you have if you want to, if you take an individual person, right, and you want to understand what happens with their epigenetic age over time, what

hannah_went:
Mm-hmm.

jesse:
you don’t want is a high degree of noise. And by noise what I mean is just kind of random measurement errors in the methylation profiling that would give you an artifact in the result. What you want is that I mean, what you specifically want, and what the paper that came up with these PCA clocks is, what they did, is to say, if I take a blood sample, a single blood sample, the same blood sample, and I split it in half, and I measure the epigenetic age of the same sample twice, it should be exactly the same. Right? So, this really innovative paper by Albert Higgins Chen, who’s now a PI himself at Yale, at the time he was a postdoc in Morgan Levine’s lab… at Yale before she moved to Altos, they did exactly this experiment.

They took a data set where there were a series of blood samples that they split in half, and for each blood sample they measured twice, so measured the methylation twice, which is called a technical replicate. So the key is that it’s exactly the same blood sample. The only difference is just the measurement. So again, if you would, in a perfect world, you should get the same answer both times. But what they found was that if you look from clock to clock, different clocks performed to different degrees of precision. So in other words, some clocks, you got two completely different answers and then other clocks, you closed the gap and got better. But what they were able to do in this paper, which is now published, I’m pretty sure in Nature Aging,

hannah_went:
Mm-hmm.

jesse:
was to develop these new algorithms, new models, where they use principal component analysis, which I don’t think we need to get into the weeds about

hannah_went:
I’m

jesse:
PCA,

hannah_went:
going to go.

jesse:
but it’s kind of a way to reduce a very complex data set like a methylation profile into single or, you know, a small number of points.

hannah_went:
Right.

jesse:
They were able to use this approach to build these new clocks, which they call PC clocks, principal component clocks. And the idea is that… they’re still reflective of the original clocks, but they’re denoised. So in other

hannah_went:
Mm-hmm.

jesse:
words, if you have a methylation profile that looks kind of very jagged and it’s going up and down, what this algorithm will do is it will collapse that. And

hannah_went:
Mm-hmm.

jesse:
then, you know, if it’s a regular aging profile, it would

hannah_went:
Yeah

jesse:
go up, right? But the point is just to remove this technical noise, because we don’t want technical noise. We want true signal.

hannah_went:
Right.

jesse:
And

hannah_went:
Yeah.

jesse:
so… So again, it’s really just a wonderful study, I think one of the most important for the clock field in recent

hannah_went:
Yeah.

jesse:
memory. And so we were able to apply those clocks in our study. And they do exactly what they say. They take the noise out of the data and they make the trends much more clear. One thing that’s interesting just to highlight. And it doesn’t take away from the utility of these PC clocks, but it’s just an interesting observation. What we found, this dichotomy between the first and second generation clocks, even after the principal component correction, actually it remained to be true. So again, only the second generation clocks picked up the effects of the stress. So that’s again why I was saying in the previous topic that I think it seems to be the case that you can have different clocks for different purposes. And so we need to understand when to apply these various different clocks.

hannah_went:
Sure, sure. Do you think moving forward in any types of studies where they’re trying to measure anti-aging interventions or look at something just like you did, do you think they, of course we want outcomes with a multitude of those different clocks, but do you think people should give more weight or more attention into those principal component analyses corrected clocks?

jesse:
Um, you know, what’s the best answer for this? I think

hannah_went:
Yeah, you know, we still don’t know.

jesse:
the unscientific, yeah, or the not fully scientific answer is probably, I think.

hannah_went:
Okay.

jesse:
But that’s a personal suspicion. I mean, you know, what would be nice, and I think what most people are doing, certainly what we’re doing is, you know, you analyze the same data set with both the original version of the clock and the PC version

hannah_went:
Mm-hmm.

jesse:
of the clock, and you compare the results, and at least

hannah_went:
Mm-hmm.

jesse:
in our… at this point relatively limited experience, it seems to be that the results generally agree, but you do indeed reduce the noise, which

hannah_went:
Mm-hmm.

jesse:
is important,

hannah_went:
Right.

jesse:
especially if you’re looking at individual trajectories. But part of the beauty of Albert’s paper is that he uses exactly the same CPG sites

hannah_went:
Mm-hmm.

jesse:
that are used for calculating the original clocks to use the PC clocks. So in principle… Anytime you’re calculating any of these clocks, you should be able to simply apply their algorithms and produce the result for the PC version of the clock and check your results, or compare the results between the two. And I think more people, I’m sure, are going to start doing that, and indeed have started doing it,

hannah_went:
Right,

jesse:
including

hannah_went:
just so we can

jesse:
us.

hannah_went:
compare the two and yeah, and look at the trends and see what it’s actually telling us. Well, no, very nice, very nice explanation there. I think one that is a little bit harder for people to conceptualize or understand that really it’s just reducing that noise. And I like that example that you gave with your hands as well. So what is next for you? What’s next for DNA methylation clocks in general? Or what are you interested in?

jesse:
I’m interested in a lot of things,

hannah_went:
Hahaha

jesse:
but what’s next for clocks? So the biggest thing is what we’ve come back to many times now is that we need to understand what clocks are measuring what.

hannah_went:
Mm-hmm.

jesse:
And if we can get very slightly philosophical, part of this ties into what is aging

hannah_went:
Mm-hmm.

jesse:
biologically. I think actually this has become my biggest… the biggest thing I’m interested in. And it’s kind of too big of a thing to be specifically interested

hannah_went:
Hehehehe

jesse:
in because it’s not something you just wake up one day and say, I’m gonna

hannah_went:
Ha

jesse:
define

hannah_went:
ha!

jesse:
aging today. Because we, you know, of course the aging space is growing, there’s a lot of excitement, both in the scientific community, of course in companies like yours and in the general public, and that’s great. But, you know. the reality of the situation is biologically we don’t really have a good idea of what aging actually is. And so part of the motivation for us doing this study was that our thinking is that if you can understand these kind of fundamental aging principles, so how does biological age vary as a function of time, then maybe you can understand what is the nature of biological aging.

hannah_went:
Mm-hmm.

jesse:
And so, yeah, I would say that the long-term goal is to try to chip away at that question. But it relates to what’s next to the clocks, because of course, you know, if you want to measure biological age, as the clocks claim to do, and, you know, again, it’s an educated hypothesis, right, that these

hannah_went:
Mm-hmm.

jesse:
epigenetic clocks are measuring biological age. Strictly speaking, it’s not unequivocally proven. And of course, there’s a… There’s a kind of vocal community within the aging, biology community,

hannah_went:
Mm-hmm.

jesse:
who very strongly believe that these are not the best way to measure biological aging. And some of their critiques are quite valid,

hannah_went:
Mm-hmm.

jesse:
as long as they’re engaging in good faith, of course.

hannah_went:
Hahaha

jesse:
But anyway, the point is, if you want to, in an ideal world, what you should be measuring with a biological aging clock. is biological age. And to get there,

hannah_went:
Mm-hmm.

jesse:
you need to understand what biological age is, and you need to understand how the clocks are working. So you know, there’s a degree of, people call it the black box, right, with these clocks. They’re made with these machine learning approaches. The machine learning approaches are highly agnostic to the biology that’s underlying the changes in the methylation sites. So you know, biologically, it’s fascinating, right, that you can build a model. just based on methylation that can so accurately predict even chronological age, you know, leaving aside the idea of biological age.

That in itself is a fascinating biological phenomenon. But what we need to know is which clocks are actually measuring biological aging. And so that’s, I think, the big next frontier for the clocks. And again, like we mentioned briefly already, it’s something that we and others are working on very actively. So the idea is that you would… build a clock that selects methylation sites that are truly causal to aging.

hannah_went:
Mm-hmm.

jesse:
Because you can have changes, of course, over the life course, which may be causal. In other words, they may be driving aging phenotypes. You can have sites that are changing because they represent adaptations to aging, and

hannah_went:
Mm-hmm.

jesse:
that can be relevant to stress too, because if you have chronic stress, the body has to adapt. You can do changes to do that. and you can have changes that just change for some other reason and are completely irrelevant to aging. And the way

hannah_went:
Right.

jesse:
the models are built now, they don’t discriminate between these and whatever other classes we haven’t even covered

hannah_went:
Mm-hmm.

jesse:
of methylation changes. So getting a really solid biological understanding of what changes are causal and relevant to aging, I think is gonna be the next big step and then building models that are. that are based on that. And I just plug, one of my colleagues has a preprint actually,

hannah_went:
Yes.

jesse:
on Bioarchive looking at exactly this question and he was able to build what he calls causal clocks, which separate these different effects using an approach called Mendelian randomization to identify the specific CPG sites that are related to causal aging and adaptations and so on and so on. So. I would suggest people check that out. That’s I think a very nice example of where we’re going to be heading and I know other people in the field are working on it too. So I think that’s our next frontier. Stay tuned.

hannah_went:
Yeah, that’s a great, great answer there. Super excited about the work in that paper. I think that’s a great paper and one that I need to dive into and understand a little bit further. But I’m extremely excited about all of the work that’s been done, right, when those first generation clocks came out in 2011 and 2013. Fast forward, now we have all of these second generation clocks and we’re actually looking at them. But like you said, understanding is this actually tracking biological changes and what is biological aging in the first place are very important questions. So we’ve come to the end of our interview, Jessie. This is the last question I end on. It’s a curve ball question. If you could be any animal in the world, what would you be and why?

jesse:
Oh, I would definitely be a cat so I could sleep all day. I’m so

hannah_went:
Oh,

jesse:
jealous

hannah_went:
there you

jesse:
of

hannah_went:
go.

jesse:
my cats.

hannah_went:
How many cats

jesse:
I

hannah_went:
do you

jesse:
just,

hannah_went:
have?

jesse:
I, too. My wife is a veterinarian,

hannah_went:
Oh.

jesse:
so we have a lot of pets.

hannah_went:
OK, gotcha, gotcha.

jesse:
But yeah,

hannah_went:
That’s great. That was.

jesse:
I know, I see them sleeping on the bed all day and I’m just like, oh, what a life.

hannah_went:
I wish I had that life. That was a quick answer. You already knew the answer. So thank you so much. We’ve come to the end of this amazing podcast. So for anyone who wants to reach out to you or connect with you, where can they find you?

jesse:
Yeah, so I just created a website recently, Poganic.com, my last name, and it’s basically just a link to my academic work.

hannah_went:
Perfect.

jesse:
But it’s not difficult to find me if you just Google my name. And

hannah_went:
Perfect.

jesse:
yeah, I would encourage people to reach out if they’re interested.

hannah_went:
Well, thank you so much,

jesse:
And please

hannah_went:
Dr.

jesse:
check

hannah_went:
Poganic.

jesse:
out the preprint.

hannah_went:
Yes, yes, everyone please do. And I’ll post

jesse:
Yeah.

hannah_went:
it in

jesse:
Yeah.

hannah_went:
the show notes as well. So

jesse:
And fingers

hannah_went:
thank

jesse:
crossed

hannah_went:
you,

jesse:
it will be

hannah_went:
yeah.

jesse:
published soon.

hannah_went:
Yes, we’re definitely hoping. So thank you for joining us at the Everything Epigenetic podcast. And remember you have control over your epigenetics. So tune in next time to learn how. Thank you so much, Jessie.

jesse:
Thank you, Hannah, it was a pleasure.