Underlying Features of Epigenetic Aging Clocks | Morgan Levine

11:09AM Nov 12, 2020

Speakers:

Keywords:

clocks

age

epigenetic

cells

capturing

question

methylation

modules

senescence

biomarker

tissue

people

dna methylation

measure

compared

horvath

signal

chronological age

interventions

genes

Morgan, thank you so, so, so much for joining, it's been really, really quite fantastic. And to to add to Well, I'm very proud of myself for having been able to get you onto this call. And because your name has definitely far more than more than twice I think in this group, we've been talking a lot about biomarkers lately. In fact, I think maybe even Robbie's, as on this call, who has been kind of circling spinning off a side project on biomarkers, and which is a separate group that has no meaning. But it's been of great interest to this group and on a variety of different levels. I think they can tell you best I won't, I won't try to. And I won't try to be the soundboard for this. But thank you so so much for joining, I discovered and your research we can find as a coincidence as I was kind of like scouring the web for different ways of how one may be able to do biomarker standardization. And I found your printout of the fantastic paper that he co authored, which I'm going to pause here as well on which you co authored really quite recently. And it was on underlying features of epigenetic aging clocks. I'm going to post all the links to your bio, into the paper and in this chat, but now I want to I want to give the stage all up to you. I think you're going to talk for about 30 minutes. And I'm going to try to collect questions in the group. And I'm already going to ask folks to collect questions mark with a cue in the group and then we can maybe streamline them in for discussion afterwards. Thank you so much for joining. I can't wait to hear what you are about to share. I am going to make you Cobos right away so that you can share the screen. All right, Morgan, take it away. It's really a pleasure to have you here.

All right, thanks so much, Allison. And I'm excited to be here, I see a lot of familiar faces. So this will be good. And yeah, I'm excited to share this paper that I think came out earlier this year, I believe. Um, so its underlying features of that genetic clocks. That tool is in vivo. And in vitro, I can put that on the title. Um, so my lap I usually when I'm not giving an aging, talk to aging researcher, I start with the idea. You know, aging is the biggest risk factor for death and disease. And we think this is actually causal. So the things that are changing with aging are actually causal in disease pathogenesis. And for this reason, we really want to quantify aging, beyond just a proxy of using chronological age. So aging is actually something very easy to perceive, you can my five year old can distinguish an older individual from a younger individual, um, hold on, let's see if they'll go for it. There you go. And that's because the impacts nearly every seller system in our body. Um, and and this explains why the consequences of aging are so dire, right? It impacts or at least drives a bunch of different diverse diseases. But this also is an issue in that it's not exactly clear how we would quantify aging. So I almost think of this as, like, you go to a Starbucks, and there's almost too many options. So it's not not exactly clear where to start. So people have started to use, um, even you could use images of someone's hand to actually make an aging marker, you could use facial images, essentially anything. But the hard thing is actually determine which information is the most important if the long term determination is figuring out who's at risk for death and disease. And also if then intervening and actually seeing a change in that biomarker, we believe should impact differences in healthspan and lifespan. So my lab likes to approach this kind of from a systems biology perspective. In that, we start to think that aging probably starts at the molecular level, and this then manifests upward to higher levels of biological organization. So, you know, you have changes at the molecular level, then the macro molecular organ, now cell tissue organ system, and eventually the organismal level. And so failure at each of these lower levels would then impact aging phenotypes at the levels above. So we, we study aging at that kind of molecular macro molecular level. And specifically, we work on epigenetic clocks, which I'm sure a lot of people in this meeting are familiar with. So but if you're not, this is mainly even though we call them epigenetic clocks, we really are focusing specifically on DNA methylation. So DNA methylation is actually the most abundant epigenetic modification in eukaryotic cells. And it really influences kind of the chromatin structure. So whether you have heterochromatin versus you chromatin, and kind of how it's packed.

Um, and DNA methylation is really important because it determines cell identity, it determines cell phenotype, and and basically how every cell is responding to its environment. So I like to use the analogy, I know there's a lot of analogies for epigenetics, or DNA methylation, but I like to use the analogy of a recipe. So basically, your genes are all the ingredients that you need to make anything. And your epigenetics or DNA methylation is the recipe on how each cell is going to make whatever it's going to go about doing. And essentially, what we know is that with age, this recipe gets a little messed up. So maybe you add too much of something, or you completely don't add a different ingredient. And this really is going to change the overall phenotype of the cell. So when we're looking at this data, to actually describe what this looks like, is we measure DNA methylation at what are called CPG sites. So a CPG dinucleotides. And so these cytosines can become methylated. And we measure this essentially, as a number between zero or one. So in your population of cells, what proportion of this exact cytosine are methylated, verse unmethylated. And we see really interesting changes of very specific CPG sites with age. So for instance, perhaps in this side of the scene, if I looked at a sample from a 20 year old, perhaps you would have 90% of cytosines, at this location would be methylated. First, if I looked at an 80 year old, maybe that drops to 60%. There are other regions that you get increased methylation with age, so this is in a single direction. So some, some CPG sites increase with age some decrease. So for this example, at the bottom and a 20 year old, perhaps you would have 5% of cells methylated, at this location versus an eight year old, you get 45%. Um, these two represent what we would call drift. So they're moving towards 50% from the two extremes, but you also get some interesting ones, which actually are also moving away. So you start perhaps as a 20 year old with 20%. And then by the time you look at an eight year old, only 1%. So because these changes are actually so precise, and actually, interestingly, even precise, across different tissue and cell types, people have developed what are called these epigenetic clocks. So basically, what we can do is we can take a blood sample, a tissue sample, or even cells and culture. And we'll see Graham's DNA methylation and get information on between 20,000 to oftentimes millions of CPG sites across the genome. And then people have applied different supervised machine learning approaches to actually train predictors of various aging outcomes. Usually, we're training predictors of chronological age, although the newer clocks have been training predictors of what we might call age correlates or phenotypes of aging. And then what we get is your predicted age based on some subset of these CPG sites, and you can compare those to the observed age of the individual or the sample and ask if kind of the discrepancy between that is predictive of, for instance, mortality risk, morbidity, risk, or any other outcome of aging that you're interested in. Um, so dozens of these epigenetic clocks have been developed. So the first one was developed in 2011, by Buckland at all. But since then, there's been a ton of these and actually, I don't even have all of them on, on this timeline, because there's so many, and I can't even really keep track of them, a new one seems to be coming out every month. Um, but, uh, so kind of these ones in the early that were developed earlier, what we might consider these first generation clocks. So again, these were trained to predict chronological age, whether in blood or in multiple tissues. However, some newer ones, like the one we did in 2018, the grim age clock, the belski piece of aging clock, or what we might consider second generation, so these weren't trained does chronological age predictors, but rather predictors of things like mortality, or some phenotypic age or other measures related to that. And today also, at the very end, I'll talk about the medic clock, which is a new clock that's part of this paper.

So one interesting thing, you know, as you know, we started to look at these epigenetic clocks is we started to ask, you know, they're all attempting to approximate the same signal like we're all we're all trying to capture what's happening to the aging methylome and understand how that either the mechanisms that feed into that or also downstream consequences. So I like to think of this as there's, you know, a tutorial to draw a hand. And even though these are all meant to actually be approximations of this hand, they obviously look a little different. So some are a little bit better than others. Some are a little worse and the genetic clocks are, you can kind of think of them along the same lines, they're all attempting to capture the same signal, but they're not actually getting the same exact thing. So. So yeah, they're trying to approximate the same signal. But you actually get fairly different results when you compare the different clocks. And that's because they differ either in terms of the training sample characteristics. So who was included in that training sample, when they developed it, the outcomes, they were trying to predict whether it was chronological age, or some of these age, or lips. Um, some of them were trained in whole, actually, a variety of a majority of them were trained in whole blood, but some also incorporated other tissue types. And some of them actually have this pre selection criteria. So they didn't look at all the cpgs that were sequenced, but really focused in on certain types of cpgs. So you end up with slightly different clocks. And what we were asking is, what do the different clocks capture? Are there some shared signals across them? Which ones are better at approximating specific processes. So this is the paper that I'm going to guess today, I'm the first author was one of my former postdocs, Zeon, who's now moved on and has a faculty position of his own. And basically, what we did is we had this kind of hypothesis development, where we compared 11 of these epigenetic clocks in terms of their tissue prediction, their association with gene expression, and also differences in terms of tumor versus normal tissue, and also in vitro aging hallmarks, then what we did is we actually deconstructed the clock. So we tried to determine if you imagine a Venn diagram, not in terms of the cpgs in them, but if you think of it more abstractly, like a signal, are there some signals shared that are core across the clocks, and some that are different. And by knowing that can you actually put them together to make an even better clock than any of the original score. So for the first part, I'm just comparing the age predictions across tissues, what you can see is that different clocks have very different age predictions. So this is the kind of famous Horvath clock, the original can tissue clock, you can see it has, it has the highest age correlation. And that's because it was trained as a multi tissue age predictor. But the other interesting thing is actually a lot of these blood specific clocks or training blocks that are trained in blood actually show very good age prediction, or at least age correlation within other tissues. So there's some kind of signal that is not blood specific that they're capturing, but also tracks across whether you're looking at fiberglass, or whether you're looking at samples from brain. Um, but also, when you look at these blood trend clocks, you'll notice that they seem to suggest that teachers are actually aging at different rates. And we think that actually, the the Horvath findings that show all tissues are aging at the same rates is actually an artifact of the method that was used to develop it. So that's actually, if anything adjusting out tissue differences so that it gets consistent age predictions across. And actually what we would expect would be that tissues aged at different rates, we know they have different kind of, like, different environments, they have different replication rates. So it's not surprising. Um, and also, what's pretty consistent across these other clocks, is that it seems that the brain is aging at a slower rate than most of the other tissue or cell types.

Um, so the next thing that we did was we asked, we didn't look so a lot of people have taken the genes in which these CPG sites are co located in and run on either pathway and return analysis. So what we did is we actually just looked at differential expression associated with the epigenetic clocks. And we didn't want to capture something that was specific, again to one tissue or cell type, because the epigenetic clocks are fairly universal across tissues and cells. So what we did is we took gene expression data from terrified monocytes and also bulk cells from dorsal lateral prefrontal cortex and we asked what They're conserved transcriptional changes that were associated with either accelerated or decelerated epigenetic aging. And we use a network analysis. To do this. I won't go into the methods, but I'm happy to answer questions if people are interested. But first, we just asked whether the clocks seem to have similar associations with transcripts. And basically, this shows that at least this core set of clocks so Yang, and I'm Lynn Levine and the two Horvath clocks have very similar associations with gene expression. This has been monocytes and also in brain so they seem to share transcriptional signals. This also shows the strength of those associations. So when you're This is relative to the original Horvath clock, when the slope is greater than one, that means that clock actually is stronger, but similar signals. So for blood, the hanham clock, which is not surprising, it's it, it's a blood specific clock, and the strongest transcriptional signals. And in brain actually, this Horvath two, which is the skin and blood clock actually had really strong signals compared to the original format. So she thought. So when we actually looked at what types of genes were either associated with accelerated or decelerated epigenetic aging, again, we use a network analysis. So we aren't looking at specific transcripts. But instead groups of transcripts, what we find is that higher epigenetic age was associated with a decrease in expression for genes associated with things like cellular respiration, mitochondrial translation, oxidative phosphorylation, and also a decrease again, in kind of the mitochondrial gene expression for this, what's called this green module and this turquoise module. Um, however, we also have found that they're associated with an increase in genes associated with kind of chromatin modifications, histone modifications, and also kind of cell cycle checkpoints, and potentially on top of G down in this red module. So next, we compare the weather so a question I get a lot of times is, is cancer tumors actually accelerate in their epigenetic age versus normal? And? And the answer is actually depends on which clock you're looking at. So you get fairly different answers. So at least we found for our recent, the fino age clock, we get significantly increased epogen age in tumor versus a normal set of normal tissue across four tissue types. So breast colon, lung, pancreas, same was true for the Yang clock, which is actually a my todich clock. So it was supposed to capture a kind of my topic history. But for the other clocks, we actually find no significant difference between tumor or normal. And in fact, in the original horbat clock, we actually find a slight decrease in tumor versus normal, which again, we think is a function of how statistically it's adjusting out some of this signal for replication.

We also looked at data from cells and culture. So for this, we had data in where we actually induce cellular senescence, so we do it using two different induction methods. So one is serial passaging, where we they're passive cells to become repetitively senescent, which is shown in blue, or what we what you would call near senescent where they're still proliferative, but they're expressing high senescence associated beta gal, and then also looking at cells induced via oncogene. So a tress retrovirus, and we can compare these to the early passage cells. And basically what you can see here is, again, the fino clock, you know, age clock does show an increase for that near senescent and the replicative senescence compared to the early passage and a slight increase for the oncogene induced senescence cells compared to that early passage. Um, same thing for the Yang clock again, although it was not significant for the near senescence and then, interestingly, the hammam clock does not pick up replicative senescence, but it did actually have the strongest effect for the oncogene induced senescence and the limb clock picked up only replicative senescence, but not opportunities in essence, and neither the two Horvath clocks significantly showed or showed any significant change as a function of senescence. We also had a A perhaps imperfect model of mitochondrial dysfunction on basically where we have depleted mitochondrial DNA. So these rows, zero cells, and again, we show that the living the fino age clock, the link clock in the clock show an increase in epigenetic age in these red zero cells compared to the controls. There's a slight increase for the two core clocks. However, it wasn't significant. I believe there's only three samples in each of these categories. So the next question is, you know, what we find is that the epigenetic clocks are actually behaving slightly differently, depending on if we're looking at age correlations across tissues, whether we're looking at cells and culture. And I didn't, we didn't show it in this paper. But a lot of people have also published differences in terms of their predictions of different diseases or things like mortality. And what we actually think is that the epigenetic clocks are composites of a lot of things that are changing in the methylome, with age, and that these different parts are actually mapping on to different aging processes. And so each clock is going to either prioritize certain parts or, or be enriched for certain parts. And by actually decomposing the clocks into these specific pieces, we can determine more mechanistically what the different clocks are capturing why they behave slightly differently in terms of their predictions. And also, which of these pieces actually are most important for outcomes of aging that we care about. And finally, I think you know, where the field is going is trying to determine are all of these parts modifiable? So are all the epigenetic age changes modifiable? Or is it only certain ones that we can target? Ideally, what we'd want to find is that they're they are modifiable. And the ones that are modifiable are the ones that are actually predicting disease or mortality outcomes. But I think you can't do that when you look at the clocks as a whole, because again, there's these composite measures. So again, we use

what's called a consensus network. network analysis, I again, won't go too much into this. We've actually redone it since that paper, so I'm showing our, our improved version that my postdoc, Albert Higgins, Chen actually took over. And so what we do is we look at all the cpgs across all the clocks. So we've now expanded this to include additional clocks. And we can then put the cpgs into what we call either different modules or different networks. So the format clock has 353 cpgs. And we'll assign each of those cpgs to a different module based on co methylation across a variety of different tissues. And then what we can see. So this up here is signifying which clock is found in so if you look, this is the original format clock. And you can see it has some cpgs that are assigned to this yellow module, some in the green some of the turquoise. And down below are the age correlations for these cpgs in different tissues. So the ones with this kind of blue box are ones that tend to decrease methylation with age, and the ones in the red box are ones that increase methylation with age. And you can see that there's some distribution across clocks, although there are certain clocks that tend to really capture very specific processes. So I'll go a little bit more into kind of what those are. So again, as I mentioned, every CBG in each clock gets assigned to each of these modules. So then we can say, are some clocks enriched for certain modules? And is that why they behave slightly differently? Um, so well, you can see here, so this is not all the clocks, but a majority of them. And they do actually have fairly different, um, distributions of these modules. So you know, the Horvath one clock has a big, big enrichment for this turquoise module, whereas scrimmage essentially has very little turquoise module, but a much more a much bigger enrichment for gray, or blue. And then we can, we can also ask the question, even if they all have the blue module are because every clock will wait, the CPG is slightly differently on the blue modules actually capturing the same thing. Um, so when, when we look at the turquoise, we find very high correlation. So they're all capturing the same signal from that set of turquoise cpgs. So even though they don't have the same cpgs in the clocks in each of the clocks, they're capturing that signal by probably using clocks that are cpgs that are highly correlated. But when you look at the blue module, they actually pretty A lot of discordance in the signal that they're capturing from this. So the next question is, which ones are actually capturing the correct one, and which of these modules are the most important for outcomes that we care about. So what we did is we actually use these different sub clocks to train a mortality predictor. And then we compare that to, at least, their original fino age clock, and also the john clock, which is mortality clock. And we use the Framingham data for this. And what we can find is this new what we call meta clock, which takes pieces from each of the different epigenetic clocks to create what we think is a better clock, um, actually has a much better predictor of mortality than either of these two. So the loving clock for every one standard deviation increase in your epigenetic age relative to chronological age, you're about a 2.6 fold increase risk in mortality. Whereas for the meta cloth, this actually ends up being a 5.6 fold increase risk in mortality for every one standard deviation, increase in the clock. We also show that even though this is trained as a mortality predictor, it actually tracks really strongly with age in a variety of different tissues. So we looked at whole blood, bras dorsolateral, prefrontal cortex, or brain and dermis and epidermis. And again, we find that this is tracking pretty strongly again, it's not trained to be an age predictor, but still captures age. And I'm not showing this here, but you actually, um, in epidermis and dermis, you get slight differences, but it also captures whether it's taken from a sun exposed person on some exposed region.

Again, it captures the kind of mitochondrial, that mitochondrial DNA depletion and cellular senescence for both oncogene induced and replicative senescence. And then finally, um, I want to quickly show this, this is where we're moving with this. So it's not actually in the paper, but I'll just show it really quickly, um, is basically, you can now look at these different module clocks and ask which ones are tracking which mechanisms. So I'm not, I'm not going to show we're doing some stuff with replication, some stuff with senescence, some stuff with mitochondrial dysfunction. And you can see some of these modules track better with age. But something that we're really interested in is kind of this epigenetic reprogramming. And if all of the clocks are equally responsive to epigenetic reprogramming. So this is a time course, basically, where you start with your somatic cells, I believe this is fibroblasts here, and then you using Yamanaka factors, I believe if this was five factors that I look back and see, if people are actually interested, you can track changes in these specific modules. And what we would expect to see would be them to go down, right you want this is as a, this isn't an absolute value of age, but you want it to decrease in its relative epigenetic age. Um, so there's, at this stage, these cells are partially reprogrammed. So for people who are following kind of this partial reprogramming, this is kind of the sweet spot of where people think that we want to be hitting, you want to do kind of this pulse, reprogramming and, and you don't want to D differentiate the cells, but you actually want to try and reduce the age signature, and then you get incomplete and then finally, fully differentiated induced pluripotent stem cells here. And again, what you can see is not all the modules, or not all the types of epigenetic age measures are responding to this. So there are a few, for instance, the blue module, salmon, brown and red that seem to actually show a stronger effect for reprogramming However, there's some that actually increase their quote unquote, epigenetic age as a function of reprogramming. And then the one other thing is we actually went back and looked at some data on reprogramming that had been published. So this is from the original paper that was published, where they use the Horvath clock and showed that this partial reprogramming they can get a decrease in genetic age in fibroblasts and endothelial cells. And this is the exact same data just plotted on a spiderweb plot. So basically, the ones in blue are the reprogram cells, and the ones in red are the fiberglass controls. And the closer the data is to the outside the older it is So you want your reprogrammed cells to all be around the middle. So this is showing the exact same thing. But when we actually parse it out by modules, we find that it's completely captured by just a piece of the Horvath clock, the whole clock is actually not being reprogrammed. It's just this one module. And so the question then becomes, you know, is that module what, what mechanistically is that capturing, and is that one that we actually care about in terms of predicting future more mortality or health measures, and it's still early, and we're looking to this, but actually, that does seem to be the most predictive of things like all cause mortality, and I'm not showing that data because we're just starting to look at it. Um, but just in conclusion, basically, epigenetic clocks have some shared transcriptional associations that suggests there might be some links between aging and things like metabolism, immunity, and toxicity. Only two of the clocks Levine and LAN reflected both the hallmarks of aging we looked at so cellular senescence, either from replicative or oncogene induce and also mitochondrial dysfunction. Using network analysis and ensemble learning, we're able to identify unique signals that are underlying some of these epigenetic clocks. And in the future, hopefully look mechanistically at what those are capturing. And we can also use these modules is input to train potentially better epigenetic clocks going forward. So you might combine certain pieces if your goal is to predict cancer risk, or maybe a different piece, if your goal is to predict Alzheimer's disease, or whatever it may be. But actually, by using these different pieces, and understanding how they map to different outcomes, we think that we can predict better clocks. So this meta clock, which we showed in this paper, is a better mortality predictor than some of the prior existing epigenetic clocks, but also was applicable to non blood samples and seem to also capture things in vitro.

So with that, I just want to thank the people in my lab, this is an outdated pre COVID picture. So a lot of people aren't pictured, and also some of my collaborators and my funding from NIH. And Glenn, I'm happy to take questions and discuss.

Thank you so much. Fantastic. And I'll set like this. I can see a few hands already up. Okay, so we have a bunch of questions already connected. Thank you. So so much. And new data, and this new interpretation. It's very helpful. And okay, I'm trying to go chronologically, I think that's to everyone was in the chat. If you want to upload questions, and then feel free to do so otherwise, I'm gonna go chronologically out the way to upload them early. So you don't go chronologically you have to go biologically. I'm

sorry, you didn't learn the lesson here?

Oh, my God, we needed we need a good clock for them. So a good clock for it for the chat would then be the uploading and downloading, then. Let's do that as a standardization. I think we had the first one there. And you can moderate the questions of as you as you so choose Korean, go for it. Okay,

my mind might be a little too general. And so maybe get down moderated. But what I want to know is, um, can you maybe upfront, just summarize what the purpose of clock research is? Is it to I mean, obviously, not to predict someone's chronological age, or even the chronological age of their tissues, which, but so what is, what is the underlying purposes of this research? As you see it?

Yeah. So that's an important question. And I imagine that different people have different answers to that. So actually, some people might say that it is to predict chronological age. So I know there are some of these applications for forensic research or, or even I, you know, studying animals in the wild, and you can actually tell someone, this is done whatever age animal So, so it isn't necessarily not to do that. So in some applications, it is, um, at least in terms of the way I think that it's going to be the most useful is, number one is I actually think it will inform some some understanding of mechanism, right? If we can link molecular changes to kind of later health outcomes, and then also link those to different mechanisms. That might be an easy way to do it, too, I actually think they'll be potentially very useful for intervention trials or clinical trials that don't have, you know, 10 to 20 or 30 years to wait to see if the intervention group dies later than then the controls and Third, I think, you know, I actually think they'll be very useful for kind of preventative medicine. So secondary primary prevention to actually understand individuals risk of death and disease much earlier in the process, and potentially even in form kind of, I'm not saying necessarily health choices, but actually, once we get further along with this potentially in forming individual health choices, and and I'm also kind of medical care as well.

Thanks a lot. Let's see before.

Lovely, thank you. Okay, next one up, we have Tom, Tom, can you say whether your initial questions were to Morgan, or whether they were just for the group at large? And maybe Morgan not have answered them anyways?

Yeah, Morgan,

one question I had is, is whether there is data.

That could be from cohort studies, or biobanks. That would be useful to the aging research community that you don't currently have access to that you would like to have access to.

So to clarify, is that that there are samples or existing data? Um, both? Okay, I would say samples. I mean, I think the fact that UK Biobank doesn't have methylation is very disappointing, at least to me, and I think a lot of other people because they have a wealth of other data that would be great. To link to this. I think a lot of the cohort studies are generating methylation data. So Framingham Women's Health Initiative, the health and retirement study, I heard and Haynes is getting methylation data soon. But I think the other thing that's lacking is longitudinal data. So we there's very few studies that have collected this longitudinally, so we can actually understand within individuals how these patterns are changing. And ideally, what I think in the future you would want is not just a general measure for everyone, but I, I would want for me, compared to me, 10 years ago, how have I diverge and in which way? And what does that mean for my future health? So I think the only way to do that is to have longitudinal data where you can reference you can put an individual reference to themselves, rather than that kind of population as a whole.

Alright, thank you. Did that answer the question? I think sometimes, yes. Okay. Okay, lovely. Next time we have. Let's

go. Hey, um, so I guess my question is,

can you make a clock

that uses a blood sample to detect your liver failure, or your Alzheimer's disease or whatever, which is clearly something that you're getting at? Right. But if we're going to use it in trials, we're not going to do brain biopsies?

So is there any?

You know, I'm sure some people would have published that if it existed. But are people pushing towards that, I guess, is the main question. And maybe also, what do you see as the feasibility or main challenges there?

Yeah, so so we're actually working on that exact question. And obviously, some of these are going to be easier. So it does seem like you can develop a clock that's going to capture liver, I mean, maybe not predict liver failure, but it captures something about liver aging, um, brain seems to be a little trickier. We obviously we can make really good clocks, using brain samples. But as you said, You're not going to take brain by biopsy. So we are we are actually working on whether you can use blood saliva or buccal, or some other easily accessible sample to get an estimate of kind of brain aging. is not the signal is not as straightforward. And the question is whether you're going to need to use cell free DNA or actually, if you can capture it using kind of leukocytes, so yeah, to be determined. But yeah, I think that's an important thing that we agree. So it's

not clear yet

whether like, let's say the liver works better is that because liver secretes a bunch of stuff. And so the blood cells are more exposed to the results of the livers dysfunction versus cell free, you know, you're picking up actual liver DNA?

No, I think it is that you can actually capture it using blood cells. So leukocyte DNA methylation, because I think of this exposure to liver proteins from liver and yeah, metabolites and lipids and stuff. Yeah.

All right. Thanks.

All right, Rob. Rob. Ally. I guess you're next. Hi, yeah,

I'm just wondering basically the biggest risk factor for concert development is a But not all cancers are aging related on some types, like breast cancer actually decrease after specific ages. So I'm just wondering how these clocks might diverged based on the type of cancer.

So we've looked at a few cancers. So breath we have looked at breast cancer they do they are predictive of future incidence of breast cancer when measured in blood. And actually, probably more so when measured in breast, we've looked at normal tissue from women who have had cancer versus those who have not, and the clock is the clocks are accelerated, even though we're looking at normal tissue. I would argue that, you know, yes, not every cancer is age associated. But even some of these that show a decline later, I think is more of a selection. So people who are at, you know, near can probably answer this, there is this resilient subgroup of individuals who seem to be at much less risk. And when they're the only ones left in the population, it's going to look like an artificial decrease in the risk of some of these diseases. But I think it's a selection process. So I actually think that might not map on to whether the clocks are gonna pick up breast cancer or other cancer risks. We're really interested in how my todich rates in different tissue types, maps on tap, change aging, and then how that that relates to cancer risk in those different tissues. But that's kind of it's fairly preliminary. We have some data, but not enough to make a strong comment on that yet. I hope that answered the question.

Thank you. Next up with Ravi and Ravi, maybe you can say a word or two in case mom doesn't know you just about who you are your background, right? The question is relevant, then she has some context.

And yeah, Robbie Pinder work in Microsoft Research in partnership with adaptive biotech on immuno immuno immuno sequencing and trying to understand that as well as the bekkering genome in cancer genomics. So my question is, I'm also I'm sort of helping growing needs and discussion within this group around biomarkers to kind of figure out if there's an interesting project that we can pull together related to biomarkers. And so this is really relevant. There's one question I had was sort of looking at these different modules. And you sort of gotten some sense of if they're functionally different. And what those functions are that affect those different modules? And then, you know, are they causal, or the upstream or downstream of the things that are she has a phenotypic effect? And how do they will really relate to different interventions?

Yeah, so so that's what we're working on. Right now I didn't, we have a little bit of data on that I didn't show it for the sake of time, and because it's very preliminary, but the idea is to map the different modules, functional mechanisms, so whether some of them are picking up. So it seems like some of them are changes that are happening as a function of cellular replication and understanding of why and what those are and what that means. Um, some of them are probably picking up things like solar senescence, so like a change in the cell state. There could be a number of other things play potency, or stemness of the cell. So yeah, the goal is to understand to map these two different mechanisms, which we think you can't do looking at the whole clock, you have to kind of piece it apart and say this mechanism is mapping to this part. And once we have that map, say, which things are causal, or upstream or downstream of different things in which things are driving different kind of phenotypic manifestations of aging that we see. So that's kind of the long term goal. And we're a little bit early doing that, but they do seem to map start to map to different mechanisms. When we're looking at them. We're doing a lot of working culture.

Lovely. Okay, next one up, we have Lee and then I'll be with a question. I think that's directly relevant to this. So maybe we'll stay on topic, Lee and also one or two sentences about who you are will be awesome.

Well, I specialize in the quantification of health, wellness and aging. I think I'll leave it there that I have a business coming out in the next two months, which will be based upon Morgan, I would like to ask you how how do you what how would you describe the the gen two o'clock in terms of being responsive to interventions, like lifestyle, like nicotine and dry besides supplements, diet and so forth?

Um, so I think that's the outstanding questions. We don't know what whether these are modifiable or what interventions would modify them. I think we do know I mean, We have some indication that they're probably modifiable, right? Because you can look and say, Well, we've looked at G wasps or, or twin studies, and it only explains a very small proportion of the differences between individuals and their epigenetic aging. And you can also look at observational studies at behaviors and say, you know, all the things that your mother told you to do seem to be associated with better or lower epigenetic age. So we do think that lifestyle things I'm not saying supplements, but our behaviors, whether it's exercise, nutrition, sleep stress, do change that genetic aging, um, we don't know the degree and I would also argue it's probably going to be different for every person, right? So, you know, my, if I ran 30 minutes a day, every day, that might have a different effect than if you did, and I think that's where we don't quite understand this kind of personalized.

But do you think you'll be able to pick it up and clocks?

life? Yeah, I think we will be able to pick up some, again, based on the observational studies. Um, but again, I'm not, I would not suggest that epigenetic clocks are the only biomarker of aging we should be measuring or the only indicator of health they're going to capture, I think, alone capture a fairly good part of the signal, but they're not the whole picture.

Your only thoughts on transcript anomic or proteomic clothes?

Um, yes, so the I'm really excited about the proteomic clocks, and we're actually doing work some some stuff with near some stuff with people like Luigi Ferrucci. I'm trying to understand integration there. Um, the transcriptional clocks, I think, are a little bit noisier, but I don't think that means that they're not useful or informative. The one thing that makes me excited about epigenetic clocks in relative the other types is that they seem to be universal across most tissue, and cell types, whereas the transcriptional proteomic clocks seem to be a little more tissue or cell specific. So that's just interested me from a basic science perspective, not necessarily in terms of application, just understanding what it is about this kind of universal aging change that happens, regardless of cell or tissue type that you look at. But yeah, I think that's a long way of saying I think the transcriptional and proteomic clocks are also useful in employment. I might not say the same thing about telomere length. But yeah, there's definitely other useful clocks out there.

Thank you.

Lovely. rb, your question?

Oh, yes, good. Um, that was a lovely presentation. And thank you for joining this group. So I think you were quite adequate in mentioning that there are diverse in, you know, reasons for interest in touch clocks, things like forensics. But I think it's fair to say that the broadest interest is in the prediction of interventions to postpone chronic problems of life. So, of course, the big, big question is, how can we make a clock that is really good protector, given that what we're interested in are new interventions, which by definition, the clock has not been trained on. So I'm wondering what the where the thinking is on this right now, in particular, I'm interested in the possibility that we might take existing interventions, ones that both work and ones that don't work, and do like measurements of epigenetic age or whatever. Before and after, and try to, you know, to get as broad a correlation as possible. But I haven't seen much discussion of that. Making Sense.

Yes, no, I mean, this is very much what we're thinking is, you know, I would not say that any of the existing epigenetic clock measures are the ones that probably should in the future be used for these interventions, I think we need to integrate some of the data on interventions to actually improve our clocks, right. So one way to do this would be if you did have a huge human trial, like team or some other trial, where you have methylation measured, you know, before and in the middle, and then you have these outcomes, that you can actually train that to figure out what methylation patterns are capturing, assuming you know, Metformin, or whatever it was actually does decrease the incidence of disease, you can actually then train what methylation patterns are capturing that difference in incidence and disease in the treated versus control. But the other thing, so I have a grant with budding gladyshev. That's looking at least in mice at interventions to try and understand kind of the change in the methylation patterns as a function of the interventions which I think is Little bit more what you are getting at. Um, the other thing I'm interested in. Related to that is, you know, calorie restriction has kind of different effects on lifespan and different studies and in different animals. So can you actually figure out their methylation pattern that's predicting the effect or kind of the benefit of caloric restriction. And I actually think those measures are going to be more kind of informative going forward, whether you're looking for a calorie restriction medic in the future, or something related to that. But I think we need to use all the intervention data that we have to actually improve the measure. So almost work backwards in some regard.

Allison said, I'm up next. So but before I get to my actual question, that you know, for the month nice thing about mice is you can actually then also follow them up till they die. And, you know, so I mean, that would be great, right, is do more of this stuff systematically. and nice. I'd also want to see an analysis like your papers that looked at the differences in terms of training models differently on which you would need to do in mice training models differently in terms of time till death, at different sort of durations, right different time, you know, how, you know, what's different between a clock that's predictive x timeout versus predicted to x timeout? Yeah, that was really interesting to look at? Well, my actual question is, how are you involved at all with Steve Horvath efforts to try and get any kind of official FDA blessing on some version of a methylone clock? for some purpose. Last week in this group, we talked a little about the huge gap between a biomarker which is useful for research and a full on surrogate endpoint for a clinical trial. And I think there's sort of broad agreement that we're not at the surrogate endpoint level. And I suggested that there's an in between point here, because, you know, many indications that people run clinical trials for our for conditions that where age is a risk factor, and therefore, a normal process of running the clinical trial is that you try to balance the treatment and control by age groups. And an inevitable part of the analysis is to stratify by age, when analyzing advocacy. Do you think that methylome plots are already at the point where they would actually be better than using Chronicle chronological age in terms of efficacy analysis? You know, in terms of what, what what age version you're using for stratified analysis?

Yeah, I mean, I think that is another important application that they could probably even sooner than as a as an endpoint be used in some of these clinical trials, are also seeing in some of these smaller clinical trials, that you can actually, in some ways, predict who's going to respond based on their baseline epigenetic age. So yeah, again, like these can be used to actually to pre select your right population to do this on. I'm involved in a little bit, and so much as I've been on panels and stuff with Steven, other people. And I know, you know, Nero has also worked a lot trying to get the FDA not necessarily to look at epigenetic clocks, but to have aging surrogates as endpoints. So I've been part of the discussion. But

yeah, it seems like the work you described could be really useful, because one of the big stumbling blocks to getting a clock used in a trial and just in terms of stratified analysis of efficacy, is the expense of running the the thing and if, for example, one of the modules is really the main predictor, for example of epigenetic reprogramming, maybe it's, you know, maybe there's a way to derive cheaper, but still effective ones that could be pushed for this purpose anyway.

Yeah, that's the other thing is what if we can actually map some of these modules or whatever they are to different outcomes or different mechanisms, you might not have to measure the whole clock and we're working, I'm working with someone else to do a more targeted approaches that hopefully, you could actually get clocks that are $25 instead of $250. But they'd be more that singularly focused ones rather than kind of use more global ones. But yeah, totally agree.

All right. Thank you, Larry, you are up next.

By the way, Morgan has another meeting in three minutes. Oh, yeah. Okay. To go.

What I was trying to get in Morgan said it's really difficult sometimes to, you know, separate out what the effect of methylation is. And when you're looking at chromosomes, is it forming condensation, is it shutting down a lot of other things in the cycle? You're looking at you really know what the effect of ethnic Russian usually methylation leads to less gene expression and, and things like that. And are these genes like if you look at senescent T cells, it's mostly, you know, they express PD one, and that makes them senescent for the most part, and so, you know, I mean, is, there are ways to correlate this sort of stuff with thumb, you know, with the actual functionality that you know, this site, and you know, what, how do you look at aging? Is there a certain rate of methylation on a specific site that happens over a lifetime? Or is, is that what you're really looking for? I mean, most of the time, when, when you're looking at, you know, if progenitor cells and things like that there's not as much methylation, so are you just looking at sort of a site that has no function, but just gets sort of cumulatively? methylated?

Um, so So the interesting thing that we're actually finding is not, it doesn't seem to be about specific sites or specific genes. It's, it's more of these global patterns. And I think, yes, that's harder to put into, harder to contextualize in terms of mechanistically, what is capturing, I mean, it would have been great if it was just, you got methylation at a promoter region and this gene, which is decreasing transcription, and then that makes sense. Um, it doesn't seem to be that and actually, um, I could sell out a CPG in a completely different gene, or completely different part of the genome that captures the exact same signal as this other one. So it's not, it doesn't seem to be something about the genes that they're controlling transcription for. Um, so the answer is, I don't know, we're capturing and I think that that's kind of the motivation for this module. Thing is, if we can find cpgs that are tracking together, or at least seem to have these emergent properties together, then somehow we need to backtrack and, and map that onto mechanism. But yeah, it's, it's very preliminary. And I think the answer is we just don't know right now what it is.

We have one minute left, and you have to hop off, Craig, I think Keith had a comment more than a question. Do you just want to shoot it out there? And

then wait till the done super quick follow up? The

basically what Carl was saying, is there being any thought applied to if a certain, you know, based on which tissue samples you have available, how much compared to the whole McGill you can get? So for example, if if you can get like 90% accuracy from something easy to collect, maybe we can get a longitudinal sale, a sample much easier. So is there analysis, ranking? Things like that?

Basically?

Yes, um, I don't think there is analysis, systematically ranking things like that, but that will be important going forward, right, like, you have to weigh kind of how easy it is to get the sample or the other thing is to weigh how expensive right so even with the original fino age clock, even though we didn't publish this, it had 513 cpgs, you can get almost the exact same measure using half of that. So this will be important going forward if he developed these cheaper targeted ones that

I'd like to see, you know, yeah.

We should shift our thinking there. That's really important. Morgan,

thank you so so much. We're now one minute about your next meeting. So we don't want to hold you out. But thank you so so so much for joining, it was really, really fantastic. I know that at the end of this, there's always more and more and more questions clogging. Thank you so much for joining. I hope we can have you back in the near future when you're going to when you're hopefully going to have follow up plans to share. Thank you very much from all of us. And I'll be following up and

pray questions to

buy by bargain if people want to stay on and then you're welcome to do so. And I can open up a break room in case you want to have follow up discussion on this, which I'm going to do right now. So I will open up one big breakout room for those who still want to continue and keep talking and breakout room should be open now. And see. All right. You should now be invited to the breakout room. Feel free to join there and perhaps we can discuss a few of the follow up questions in there. I'll see you next week. Bye bye.