# Using nematode worms to solve aging - Dr. Jan Gruber

- Channel: [Edge City](https://streameth.org/edge_city)
- Date: 2024-10-19
- Duration: 31:58
- Watch: https://streameth.org/watch/67133ff98f864ede0365d63a
- Download: https://vod-cdn.lp-playback.studio/raw/jxf4iblf6wlsyor6526t4tcmtmqa/catalyst-vod-com/hls/a98feeqlb5powenf/720p0.mp4

## Transcript

I'm not giving secrets, I'm just saying I made it a little bit more in detail than it was originally because I recognize that, you know, we have a lot of people here in the field. So, I'm going to talk a little bit about worms and I'm going to keep the introduction relatively short. So these are these C. elegans I mentioned already yesterday. These are tiny nematode worms. They are as big as an eyelash, essentially. Small, 100 microns wide, about a millimeter long. big as an eyelash, essentially, small, 100 microns wide, about a millimeter long. And they are very widely used as models for developmental biology and aging research and some other areas, genetics. And they were introduced to science by Sidney Brenner in 1965, and he won a Nobel Prize for this. And there was recently the fourth Worm Nobel Prize this year for basic biologies discovered in C. elegans. So it's been an incredibly promiscuous, in terms of scientific insight, model. They are soil dwelling. In other words, they live around the root systems of plants and in decomposing plant matter. They're about a millimeter long, 100 micrometer wide. They're 99% self-fertilizing hemafrodites. They only make males when they need to do some gene exchanges because they're stressed. So, essentially, they don't normally need men. They produce 300 eggs per adult animal in about eight days. So, they have a massive reproductive rate. And that should already tell you from my discussion yesterday on evolution that you expect them to be very short-lived because if they all survived until they were reproducing adults themselves, these 300 offspring, then the world would essentially be a ball of C. elegans. So, obviously, they have a massive extrinsic mortality. Only very few of these 300 eggs will produce egg-laying adults. And therefore, they will have a very short lifespan because the evolution will have pushed them to the fastest development time and the short lifespan. So, that's exactly what happens. We know a lot about them. They have 952 cells and 302 neurons. So, we know a huge amount of very microscopic, very high-resolution data on them. They were the first organism that had their genome sequenced and so we have a huge amount of data on them. So, they're extremely powerful model system. And you can make beautiful videos. This is the nervous system of C. elegans in their essentially head. This is data from my lab. So, if you want to do neurodegeneration, you can literally in a live animal visualize at very high resolution their nervous system. This is what they look like when they age. They live about 30 days in the lab. On top here on day six, you have a young adults that are laying eggs. In the middle on day 16, this is also running. You might not believe it, but they're really sluggish and tired and fed up. Actually, it wasn't running. Now it's running. And at the bottom here, day 23, they're almost dead or dying, partially paralyzed. They have constipation. They have neurodegeneration. They have sarcopenia. They have all the things that old people have. I mean, a lot of the things. So they're kind of a neat model for aging. And remember, they don't live that long in the wild. They only live, you know, like this in the wild. They have about a 1.5-day median lifespan in the wild versus the 17 in my lab. And the important thing about C. elegans for aging research is that they were the first organism that actually had the single gene mutation that extends lifespan. And this is really something which I think we are so used to this idea today that we don't realize that not so long ago in 1993, this was really, really unexpected. So people thought evolution has basically made living things live as long as possible. There's entropy, there's random noise, there's lots of damage, and this damage accumulates. And it's happening across so many systems, and it's so untrackable that you can't do anything about it. There's nothing we can do about aging. And there might be some truth to this. But then in 93, or in the 90s, they basically found mutations where you change a single nucleotide in the genome or a single gene and you get a lifespan extension of 50 to 100 percent. So that means there are ways in which, you know, organisms can tune their aging rate really rapidly with just one signaling pathway. And this wasn't something that they expected. When they started looking for this, when people started doing these experiments, they were told that this is pointless because those mutations can't exist, right? There can't really be a master regulator of aging because it's just random noise that happens across the system and there's really no regulation. And when these were discovered, they really spawned a lot of excitement. Now, the reality might be somewhere in between the excitement that comes out of this and the original pessimism, but it's just realizing that model organisms like C. elegans literally kicked off the field. There wouldn't be, you know, the kind of aging research that we have today if we hadn't made these discoveries. Okay, so this paper in Nature says this suggests that there is regulated lifespan extension mechanism. So, just the fact that those mechanisms exist was groundbreaking science at the time. And that's what they look like. The top one here is wild type. The bottom one here are the slow aging strain. As you can see, the top one is a time lapse. It's moving around really quickly and then running out of steam and dying. And the same age ones at the bottom are still doing things and doing things and doing things. So, when Cynthia Kenyon, one of the lead authors of the original paper saw this, she said, or the lead author, she said, I viscerally wanted to be a dove to an age one. I forgot which one she talked about. But the point is, she viscerally wanted to be a slow aging worm, right? So, who wants to be a worm? But the point is, you're looking at this, you're realizing, you know, that this is possible to make massive differences to lifespan in animals. Okay. And this is data from my lab just to like a huge jump. Later on, lots of drugs were discovered that also extend the lifespan in C. elegans. These are all drugs that were not discovered in my lab. They're drugs we confirmed, but this is data from my lab. Allantoin, which you heard about yesterday already, metformin, rifamsin, rapamycin on the left here on the top, and a calcium channel inhibitor called PSORA. And so we did all these different lifespan experiments, and the cool thing is there's technology now to do this high throughput that I'll show you in a minute. And so overall, there are over 600 compounds that we know extend lifespan in at least one of these model organisms. So, it's quite a lot of library of things that affect lifespan. An interesting point is often when you talk about worms, people say, yeah, yeah, that's worms, don't worry about it. And again, there might be some wisdom to this, but I did a thing where I looked at the ITP, I looked at the intervention testing program which is a mouse lifespan project which has far fewer drugs, of course, because it's much more expensive. And then I looked at the equivalent worm experiments in my lab or in labs I believe in, and I basically scattered this. And what this shows is that all the things that we couldn't reproduce or that don't widely reproduce also didn't work in the ITP. So false positives in the literature don't work in what I would consider careful worm experiments and they don't work in the ITP. And then the best possible drug that really works extremely well, metformin-rupermysin combination works very well in worms and it works very well in mice. And I re-plotted the original data from my lab and from the ITP in the same sort of format so you can compare the curves. And it looks very similar. So, what this means is that, you know, there's actually quite a bit of conservation between the lifespan extending drugs, which I think is surprising because those are chemical compounds affecting, you know, the lifespan of animals, you know, that are very evolutionary distance. And so, therefore, we do these experiments. This is a robot in my lab that has a camera platform. And basically, it's the worm bot developed by Matt and his group, and also used by Aura Biomedical. And that system takes pictures of every well, 122 wells, 144 wells. And so you can take 144 drugs in one run instead of having graduate students do this. And what's also interesting, it takes like about 100 pictures per condition per day over 30 days. So it's a half a million pictures or so that you can get from an experiment like this. So if you have, you know, machine learning algorithm and you have a question about some phenotype other than lifespan, you can actually really research that sort of database really nicely. There's a lot you can learn. You're looking at all these different animals under the drug perturbations. So, I think the ability to hide throughput this has a huge value. And you have already seen pictures from this. This is actually worm bot picture. So, the time lapse I just showed you is essentially from a worm bot. So, what can you do with this? If you can do high throughput screening, you can actually look at combinations between things. So this is the thing you already saw, rap plus met, rapamycin plus metformin. And so it's, if you start treating with our best drug combination at adulthood, you get rapamycin plus riframcin plus allantoin. A combinatorial screening takes a lot of trials. You need to try the one drug and try it against all the other drugs, and then the next drug against all the other drugs. It's combinatorially, you know, combinatorial explosion. You get a large number of screening conditions, but you can actually get additive and maybe synergistic effects between drugs, and so that's really exciting. So, as you can see, you can take an adult animal and almost double the median lifespan by treating them with a drug combination of some of those drugs, which we individually already knew worked. And we tried the same combination in flies, and it's also quite a powerful lifespan-extending combination. Go on. I mean, there's this joke that says AI is not going to take your job, but somebody who uses AI is going to take your job. And I think that's absolutely true in this context. So, I don't think you're going to make a giant model, I mean, eventually, I don't know, AGI and they will like understand all biology and then tell us which drugs to use. But I can tell you that there's a huge amount of data that we can generate today that we analyze in a super shallow way, right? If, you know, everybody has done this, we do transcriptomics, we get like the whole genome, thousands of genes, tens of thousands of genes, and then we do essentially a hypothesis-driven group of things. We say, does it increase inflammation? And we do a very simple statistics and we say, oh, this proves that my mechanism that I like is involved in this. And then you do like a single epistasis experiment and you write the paper and push it out. And you've just essentially ignored 99.8% of the data that you collected, right? Because you can't make sense of it. And I think for that sort of data sets, specific AI tools, you know, like that are able to better extract patterns, for instance, and this delineate cause and effect will be super powerful. I'm going to talk a little bit about that in a minute. Here's a very shallow thing that I think is, however, also important. If I take millions of pictures and give them to a graduate student, they're going to find out when the worms died and compress that data into a single number. They will say, I've seen this worm a thousand times. It died on day 17 at this time. And that's all they care about, that the drug works. But you saw phenotypes of that animal in theory over a long period of time. So if you had the human data, you know, quantified me or your tracker, you would take all sorts of information, activity, distance, travel, body morphology, partial immobilization, all of this sort of stuff, maybe even behavioral and cognitive readouts. And we just completely ignore them because there's no way of having a student, like, visit every worm and ask them how they are, or even just measuring speed of movement for every one of these animals. But if you have a data set like Aura, where you have a million compounds times, you know, hundreds of individuals over the whole course of life, there's so much you could see if you could just run the pipeline over this and say, okay, you know, for instance, which of these affect the amount of movement I see with these animals, lifetime distance travel. Then I can find compounds that affect metabolism, you know, muscle function, possibly, you know, things that affect wakefulness. You can have, you know, day-night cycles, stuff like that. There's a lot of stuff that we just don't see because it's way too difficult if you don't have the hypothesis for a human to analyze this data in all the ways that are possible. So I think this will completely revolutionize what you can do with data. So that's why I would never want to delete this data and turn it into a spreadsheet of only a lifespan curve. I'm more skeptical about the idea that some sort of, you know, massive AI system that reads all of the literature will suddenly come up with the answers that we can't find any other way. But I think it's a huge force multiplier for the things we're already doing. Okay. All right. So just to make the point that this works in worms and flies. Now worms and flies are about a billion years separated in evolution, which is quite similar to us. So we are separated from worms and flies by about a billion years separated in evolution, which is quite similar to us. So we are separated from worms and flies by about a billion years, right? So they're as distant to each other, give or take, as we are to them. And the fact that it works in both worms and flies suggests that this is actually a really ancient mechanism that is hit by these drugs, right? It's there in animals that haven't shared an ancestor for a billion years. Okay. So that's my introduction to worms. So what do we know about aging? It's interesting. So 60 years ago, we basically said it's hopelessly intractable. You can read the evolutionary literature on this. It's like, you know, damage mutation accumulation theory, antagonistic pleiotropiedas, little we can do. Thirty years ago, we realized that we can actually modify aging really rapidly by single mutations in model organism, so there are such things as aging pathways. Twenty years ago, we figured out that that actually can be drugged. And about ten years ago, we figured out that those pathways are highly conserved in humans and models and that we can do the same drug interventions in mice. Now, whether or not we can do the same drug interventions in humans is, of course, not answered by this observation, but there's certainly a big difference in the way we view aging. So then the question is, you know, do we? Kamel, did you have a question? Okay. So do we actually understand it? So this is the thing I added after the discussion yesterday. Can I, am I done? Okay. I showed you this slide yesterday. Most of you were here and essentially it said there's a mortality increase that's exponential, right? Your risk of dying per year goes up from 1 in 10,000 when you're young to 1 in 20 when you're 80. And it's exponential and it's massive and it affects disease risk and intrinsic resilience. So the question was how do we get there? So a while back, and it's massive, and it affects disease risk and intrinsic resilience. So the question was, how do we get there? So a while back, and I'm giving you the napkin version here, we believed in the free radical theory of aging. We believed that aging is just due to oxidative damage and especially affects mitochondria. And so the idea was there are these things called reactive oxygen species, ROS, and they make some sort of damage, for instance DNA DNA damage and mutations. And that causes aging. And because having more damage makes the mitochondria make more ROS, I have a positive feedback loop where more damage gives more ROS and gives more damage and gives more ROS. And if you are a little bit mathematically inclined, you realize that that gives you, ignore the math, if you're not mathematically inclined, that's fine, gives you a differential equation that relates the rate of damage accumulation on the left to the damage that's already there. And that essentially leads to amplified growth, exponential growth. And so the solution to this is that you get an exponential increase in the amount of damage. It's a beautiful way to explain exponential mortality, right? I have a feedback mechanism, leads to exponential growth of damage, therefore the organism oxidizes itself to death really rapidly and therefore my death is directly related to my damage. And if I change the constant that connects the damage accumulation to the damage, then I can change aging rate. And so therefore the mortality rate doubling time should essentially be a function of that linkage. So that was the high, the paradigm I would say a lot of people were sort of looking at about 15, 20 years ago, me including. And, you know, I'm going to summarize a lot of stuff really quickly because exponential effects from exponential causes would be the way I described this is the thing we believed in and it makes a bunch of predictions that can be measured. And so one of them is that there should be an exponential increase in damage leading up to death. And if you look at ROS, at free radical level itself, this is data from my lab many or from my own work many years ago, and basically you get a linear increase in damage. There's more damage in old animals, but we were happy if we get statistical significance, right? P-value. We're like, yeah, we get statistically significant increase in damage. So, the theory must be right. But it's not exponential. It should be, you know, 600 times. Mortality goes up by a factor of 10,000, right? So, we're like, okay, well, maybe the damage accumulates. Maybe it's not the rate. Well, okay, so then we measure some forms of damage. I'm going to show protein oxidation here. But we did this with many things. Do this with the slop blood, western blood, you can do this with mass spec. And yeah, you got a linear increase of about a factor of two, it's statistically significant. You get the paper published, but it doesn't really explain an exponential increase in mortality in any direct way. So, then the question is, so how can you explain exponential mortality from linear change in state variables? And at that point, I'm going to talk about a thing called feature space. And again, if you are not into this sort of abstraction, then don't worry about it. But I think it's a really important way of thinking about it. So feature space is the thing that tells you where you are in physiological state. And it could be any level of abstraction. So in this case, it could be your blood pressure and your cholesterol. This would be the feature space your physician is concerned about when they're trying to evaluate your cardiovascular risk. If either of those two are high, you probably have a higher risk. If both of them are high, you have a much higher risk. Or it could be something like catabolism and inflammation. So what's your systemic inflammation levels like and what happens to your, you know, anabolic catabolic balance? Do you, are you wasting away? Are you becoming hypertrophic? Or it could be any other combination of parameters. So this is the state the organism is, it's in. It can also be at any level. It can be an individual cell that's described or the whole organism, right? It could be a cell that you're receiving some damage because you had a chest X-ray and activating DNA damage responses. It's basically how you describe the state of a system, including a living system. And so your physiological state at now, at this moment, at every level along any one of these axis you want to choose is a point. You're currently somewhere in this phase space, feature space. And so there's an area around, because we're all alive, so we all know we're in the circle, right? There's an area oval. It's an area that's compatible with survival. If you're pushing any parameter way outside of its normal function, you're leading to a pathological process and it leads to a disease. And you can almost define at least degenerative diseases as self-amplifying loops in this. In other words, if you push a system to a certain point where it starts doing the wrong thing over and over, that's a disease, right? Like you get enough mutation in a cell, it starts making a tumor, it grows, it doesn't stop and it hits other cells, it doesn't respond to regulatory signals from the body, that cancer cell is going to grow until it kills you. And so there are these attractors here in this space which are failure or disease zones. In other words, you want to stay within the circle, that's your normal function, and if you go far enough out of it, you lead to some self-amplifying disease process that will kill you. High degree of abstraction, but, you know, you can draw anything you want along the axis. And so then you can say, well, okay, so it's like a spring system, right? So the system's trying to stay there, and so you can think of these as these springs that push the point back when it moves out from this position. But of course, there are no springs in like face space in the body. How you describe this when you're a physicist is you say, okay, this is like a potential well. And then about this point moves out from the center, it goes up some, you know, energy barrier and then it falls back because the system relaxes to the lowest energy state. Okay, so that's a high level of abstraction just to make sure what that actually means. That's what that looks like, right? You had breakfast, you didn't have breakfast, you're hungover, you didn't sleep enough, and your system just oscillates around, you know, and eventually relaxes back to normal, especially when you're young, much faster than being old, I can tell you that. And so what do we mean when we talk about this potential? And what we mean is just everything in your physiology. There are hundreds of thousands of variables. I could measure millions, you know, many millions. And so, for instance, there's DNA methylation we're all very excited about. And this is a sort of abstracted network of some sort of metabolic pathways. And if I hit DNA methylation, I hit transcript levels because I transcribe genes more or less. And if I hit transcript level, I hit lipids and proteins. Some components are made and they move around in your body and they change. And that changes metabolism. And if I change metabolism, I'm, you know, will probably eventually change something I can measure using a clinical analyzer. This is a clinical, you know, an analysis pipeline for things like cholesterol or iron levels. And eventually, I'll affect physiological features like your facial shape or your blood pressure or something I can measure that's macroscopic. So, these things are all highly correlated on this network and what they collectively are is this barrier, right? So, you know, my physiology compatible with survival is somewhere and, you know, any perturbation gets pushed back. That's the sort of notion here. So, that's what that looks like. So, that's the same thing. Okay. So, when you're young, this is like, you know, 50 young people and they basically do all sorts of things, but none of them die because the well is deep enough that they do not fly over the barrier. But if you pick random damage to this network, you make a mutation to a gene, you break something, you make like plaques in the vasculature or any of these things, what this will do is it will slightly affect the way the system works and usually for the worse. The very few kinds of damage make a system better. So, the sort of barriers drop as the system becomes less capable as dealing with the perturbations until you reach an old age where the same amount of perturbations kick some of these individuals over the edge and they'll get a disease and they'll die. So, that's the model. It's a highly abstracted way of thinking about it, but it does a really, really neat thing, which is that if I look at the height of this barrier and I say linear amount of damage that I accumulate changes the height of this barrier in a linear sense. So, any one of these types of damage changes it only a little bit, but they add up and together they lead to an H-dependent linear decrease from blue to green to yellow to red. Then I get a decrease in the height of the barrier and that's a well-known physics problem called Langevin equation. You can also talk about Boltzmann distribution here. this will automatically lead to an exponential increase in the probability of falling out. So, and you can see this. This earlier I showed you is essentially when it's high, it's almost infinite, right? These bolts never come out the way I made this little toy model. When you make it flat, they will all immediately roll out, right? So, it goes from nothing escapes to everything escapes and what you actually get is an exponential increase in the probability of disease. And importantly, you get that from a linear input. So, it's a neat mechanism to turn a linear amount of cumulative damage into an exponential probability of failure and death. And that's a way to maybe bring together these two theories to say maybe oxidative damage does matter, but it doesn't matter because it's not the one type of damage that causes the death. It's one of the millions of types of damage that collectively degrade the performance of this network. And you can then do something neat. You can take data, for instance, of patients, of transcriptomics, of anything you want, and you can take it at different ages and you can do what's called a principal component analysis to figure out in which direction does the system distort as the well gets flatter. And if you do that, you get, this is actually done with the simulation data, you get a principal, first principal component that points straight at the failure attractor. So it says this is where you are when you're healthy and that's the direction in which you're going to fail. And then you can use that to predict, for instance, how well people are, you know, can make a clock with it, in other words. And this is where Peter comes in when he came to my office, I don't know, 2018, long time ago. He says, said something in a meeting, just sitting there talking. He said, the first principle component of any high-dimensional data set derived from aging will always be aging. And I had a lot of high dimensional data and a lot of experiments in the lab at the time and I was like, really? Is that what this is? And then I took a data set. This is fly transcriptomics. It shows the gene expression of flies as they age. And these are male and female flies. And in fact, the first principle component in this data set is sex. So, male and female flies are quite different. But the second principle component analysis is indeed aging. So, these are day five, day 20, day 30, day 40, and day 50. So, and this is like thousands of genes and it's just a PCA and it says this is the direction in which aging occurs. And you can get this very easily because you essentially, as you get flatter, you move that way. So, that's a really neat trick for analyzing aging data. And then you might say, fine, now I know how to cure aging. All I'm going to have to do is push back, right? See the guy going this way, I get the direction from this transcriptomic, I just push back and nobody ages. Okay, great. So let's try this. So if you do this, this is the direction of failure. You push back and you're going to get this. And this is exactly what we always talk about. If do something about cardiovascular disease risk you're just going to get cancer. If you fix those two you're going to get neurodegeneration. If you fix that you're going to get some new disease that doesn't yet kill many people that will just kill you. Because you can't do anything about it because it isn't the only cause of death. I know I'm over, I'm going to shut up in a second, but an interesting point is that cancer and cardiovascular disease kill about equal number of people. This is exponentially dependent on the height of the well. So that means they have to be exquisitely declining at the same rate and you need an evolutionary argument for this. Why is cancer as likely as cardiovascular disease in an 80-year-old? It's actually really unexpected. It means all of the aging in the system is highly synchronized. Why would two completely independent things happen exactly at the same rate given that the individual rates are exponential? Go on. So centenarians are really interesting. You have people who are centenarians, especially super centenarians are either escapers. In other words, they never had an H-dependent disease or survivors. They had it and they recovered from it. And it's unclear whether this is really genetic because there isn't that much evidence on genetic combination. It's almost certainly not lifestyle because, you know, the individual trajectories are completely different. Some of them smoke and drink and some of them are like, you know, the exact opposite. It's just a very rare probability, right? Like, in the system as well, you have some individuals that will not bounce out, and that's just exponentially unlikely. And if you know how many supercentenarians you have in the population, you can probably make a simulation and say, is this, has this actually changed over the years? And I think, if I'm not wrong, there's very little evidence that we're making a big difference to the number of supercentenarians. So, in other words, we're not making more supercentenarians. People who are that lucky just remain that lucky. So, going back to what you talked about, the standard type of policies. Yeah. Is it possible that there's actually heterogeneity in the population where... Of course, there will be heterogeneity in the population. Some of it will be lifestyle, some of them will be genetics, some of them will be pure stochasticity. But I think the overwhelming factor is randomness, right? There's even some paper that suggests that like something like, what was it, two-thirds of all cancers are not explainable by lifestyle or genetics. And so, you know, where you are through random events is actually might be the most important factor. Go on. This is more macro scale, but how does this relate to like homeostatic regulation? I mean, it's exactly what it is. What it would mean is that your homeostatic regulation will get worse as you get old and that's exactly how to measure this, right? So for instance, a way to measure this is to calculate something called an autocorrelation. You ask, how are you today and how were you yesterday and how were you last week? And if you're young, those are almost unrelated because whatever perturbation you encounter is completely declined back to zero. If you're old, then that might persist for longer periods of time or forever. And I think we all know this from exercise, right? When you're young, you hit the gym and the next day you have sore muscle and by the evening you're fine and by now it takes me almost a week to recover. So, usually if you shock the system like you know that, like just all this other stuff they kind of build on your static capacity. So, if you do this, the question is are you doing it over all the systems? Are you reducing the damage or are you just improving one axis? Right, if you just increase one type of failure axis, you have almost no effect on the system, right? If this is the axis affected by cold shock, then yes, that's going to prevent that particular thing, but you're just going to fall into another attractor. If it makes the well deeper or maintains it or cools the system down, then it might work, but we don't have clocks yet that measure this accurately, so we can't tell any of these interventions what do they actually do. I think I'm probably making Kamil upset by taking too much time, so I'm going to wrap up really quickly. Okay, so in other words, what we want to probably do is in some way is to understand the system and to figure out predictors of those changes, the amount of perturbation, the noise, and the height of the overall well, and those are related to damage. And it's important to realize that I made my life really easy by making this potential time dependent. So I put time in explicitly. I said if you're old, it's lower. When you're young, it's higher. The problem is in reality, that's of course not true. There isn't some magic one time factor. What there is is all the damage that occurred. Go on. One more real quick question. Just related to lifestyle and events that happen in a person's life like accidents and everything. Yeah. You know, there are, from a stochastic processes perspective, there are things like simulated annealing where, you know, as the temperature changes, the amount of change changes. Is there, like, an opposite effect of this as well? I mean, I think that's exactly the question, and that's, you know, Peter will probably be able to tell more, talk more about it when he talks about, like, the theory. But that's exactly the kind of thing that you'd be looking at. So, you know, is there a way to affect the temperature? And is there a way to affect the sort of cumulative damage in the system through those perturbations, right? The question is how do you do this? So, one of the things is here, you know, I just made a very high picture, high level picture showing you like metabolic maps and saying that somewhere in all of this biology are the things that pick up the damage that negatively impact this potential. And there's an argument that you can never learn this from measuring the microscopic states, right? If I measure just oxidative damage or I just measure DNA methylation, I can see the process because it affects all these variables, but I don't know necessarily how to identify even in theory the causative variables. So, in other words, I don't know how many to identify, even in theory, the causative variables. So, in other words, I don't know how many Ns there are here. And they could be millions. And they all have small additive effect on them. But the point of simulated annealing is you have these highly nonlinear equations that you're trying to, topologies that you're trying to solve. Yeah. So, you can definitely make a model to explore the behavior of a system like this. But in terms of mapping it to specific biology. Like, you want to make a methylation clock that picks up height of potential versus just movement along the failure axis. It's hard to do because those two are very convoluted in the data, right? But I absolutely think that's the right way to think about it. Yeah, exactly. As long as you have the perturbations and the data and the algorithm to separate them. Yeah. Yeah. So that's basically, yeah, there's a point here that if you have enough of these, they basically become a one number, right? You have millions of things that are different between everybody, but they collectively affect. They therefore become like a stochastic damage in their own right. Okay? And so you can make the case that it doesn't matter where those are, what eventually matters is how many have you picked up. And the determinants of this are extremely hard to measure from the data which is dominated by the movement in the face space. But it's a challenge that is addressable, you just have to do it. And so that's my discussion points. I guess we don't have time for discussion. I think I'm over already. So I'm going to let you do the moderation. I'm just going to say, oops, sorry. Thank you, everybody. This is Peter and Kirsten.