# Aging Clocks, Entropy, and the Limits of Age-Reversal: Peter Fedichev at EARD 2023

## Метаданные

- **Канал:** Lifespan Research Institute
- **YouTube:** https://www.youtube.com/watch?v=rzerGOJ2FdU
- **Источник:** https://ekstraktznaniy.ru/video/41888

## Транскрипт

### Segment 1 (00:00 - 05:00) []

yeah thank you very much for the introduction and for the possibility to talk I think I will talk from here because I need to switch the slides um so let me talk about what we are doing uh jir and what we have learned uh about aging with our friends jir and what we think we should be doing to actually produce a meaningful intervention against aging um we have just had a panel which I wish we could do in the other way so first uh tell what we're doing and then discuss well aging uh there are multiple definitions to aging uh for me the most let's say uh interesting one is the uh exponential amplification of your chances to die of all causes as we age so for fun because normally people do lock plots like you have this straight lines for fun I just plotted the probability of death in humans as a function of H and natural coordinates just for you to understand how how bad the problem is so the chances to die of all causes double every 8 years in humans this uh counts also for uh death from accidents for example so um you know a busy street is not dangerous for for a teenager but it's dangerous for an elder individual because of neurophysiological deficits by the way uh we have animals that presumably Almost Do not age uh we have animals like naked more Reds where the same graph no lock plot here just the plain mortality as a function of H stays more or less constant as a function of H uh for thousands of years of course it's not doesn't prove that these guys do not age but these guys are aging very slowly and uh the comparison between these two graphs actually brought me into the field because I want to be like that guy not like those two guys so my background is physics and whenever I see an exponential disintegration of something I of know by heart uh from textbooks that this is exponential instability so our first idea was that uh there should be kind of two kinds of animals there should be animals that amplify errors and presumably do not so basically we started thinking that uh all living beings could be classified into two situations just by observation right because we have those naked moates and we have the humans or mice that exponential agers we were thinking what would happen if you produce any kind of error into a biological system for example I take out a molecule out of an organism so that molecule was there for a reason most probably it was doing some job so now it's not doing any job which means that it will produce or not produce other molecules and there will be kind of expanding whole in the whole organism of missing or extra molecules and then in some species uh this uh disturbance uh can be cleaned out so if repair systems are working very well then on average you would not even if you introduce an error into the system you wouldn't produce more errors on average or there could be systems where one error amplifies errors and those simplify more errors and then you start exponentially disintegrating so the closest analogy here is the infection you could have a situation where the infection is controlled when you have currane immune system and everything else if you put a virus into the system then the system just you know eats it and uh nothing happens or you could have a situation where the virus produces more viruses and so on so we can talk about error amplification uh coefficients if the system is unstable error amplification coefficient is larger than one you disintegrate exponentially if it's below one you just leave okay not forever but at least you don't have aging so we decided to uh sort these things out using uh lots of data so this is a fully data-driven approach it's like weather prediction or Market predictions as I said when I'm coming from physical and Engineering Sciences which means that all these arguments can be actually backed up by calculations so we started from uh mice because when we started this mice were kind of most studied animals on Earth but uh things changed and over let's say last let's say five years I firmly believe that humans are now the most studied animal on earth we have tens or even hundreds Millions electronic medical records in digitalized form we have uh tens of millions of uh genomes in humans we have lots of molecular data so in fact we are now in a good position we can study AG diseases in humans uh first so this approach I would say gave some fruits we published a bunch of papers we got a collaboration with fiser precisely on this uh Pharma is actually very interested what people are doing in the Aging field they interested they're not impressed I would say the problem with Pharma is that uh in terms of effect size and say our bravest aspirations it's still a little bit below what they expect from a drug but that's very interesting because I will tell you in a moment how these kind of models can actually in form Pharma how

### Segment 2 (05:00 - 10:00) [5:00]

to separate aging from diseases how to help Pharma make drugs against aging not aging I'm sorry against diseases but not Aging for the God's sake not make a drug against aging and at the same time study uh study aging so as I said we use the data and we are using longitudinal data means that you study the same animal multiple times this is very different from uh most of the work that is done in biology where you have huge biab Banks and you have just one individual measured only once you don't know how this person ended up here at this point so you have the outcomes but you don't know the history even if you have long-term outcomes like 30 years survival this is cool but this is not sufficient because you don't know how the phenotypic molecular changes are translated in mortality and diseases so this is just one graph in mice we were actually able to to put this theory in practice we actually measured the error uh amplification coefficient on mice we Pro that if you have a disturbance of if you have two mice one a mouse a and mouse B which are different by some parameters you can actually see in the data that the distance between these two mice in parameter space will get exponentially larger so two animals become less and less similar to each other so all errors are exponentially Amplified in mice so this is the actual graph from one of our papers so this think that we can call the total damage accumulated so far or the biological age is actually an exponential function of H and I like on this graph those red dots the red dots are the biological age measurements of mice that are scheduled to be killed in the laboratory because they are too sick you know in a good laboratory you kill animals because they are sick so they are not allowed to die from old age and you can see that the animals are sacrificed when they reach kind of sailing in biological age right so in mice things are very simple the biological age shoots up exponentially the system is unstable and when it hits a certain level of toxic death much the animal is dead so this is really this exponential amplification so it sounds scary I mean exponential amplification of errors as a cause of death this scary but it also has a very interesting uh consequence if your system is unstable it doesn't know what is the equilibrium position there is no homeostatic equilibrium position in mice uh which means that if you find an intervention that reduces biological age at a certain point in time the system wouldn't know how to get back to the control the system will stay younger forever and that what that is what we see in infinite number of experiments in mice in different groups just in our hands it works like this on the vertical axis I have the biological h on the horizontal axis I have time and then um here in between the first two points I apply only once the anti-aging intervention right so just one shot of a drug and then you can see that since the system is unstable this is the precise definition of instability since the system is unstable it never gets back to the control so the mice are younger forever so I would tell that if you have any intervention that taks against Aging in mice if you do it once the mice will remember that forever it's very easy to rejuvenate uh a mouse and I think we have lots of evidence that this is happening just maybe to show that this is these are not only computations this is just one of our examples which is already particularly old uh we worked with uh Brian Kennedy in the National University of Singapore these are survival curves uh on the vertical axis you have the number of animals alive to a certain point and then along the horizontal axis you have the time since the beginning of the experiment these are geriatric mice so they are 100 weeks old in the beginning of the experiment you can see that in the control they are dying of Aging so you can see how these guys are dying of aging and then we have one shot of experimental intervention that was predicted by our calculation and you can see that even though there was just one intervention remember the instability there is a mortality DeLay So these guys are younger and two months after the treatment you can measure essentially any hmark of Aging there for example the amount of sent cells in the immune system of these guys and you would find that there is no drug in the circulation anymore but these guys are younger um we used the same approach in humans and we were totally surprised I mean that's not something that I was expecting because I started with the fact that humans are exponential agers so our chances to die actually go up exponentially as we age so I was expecting that if I build a system that would predict a biological age in humans I would find that this biological age would also go up exponentially as it does in mice it does not of course I don't have enough data to plot the biological age trajectory along the whole trajectory of uh life in humans but schematically it looks like this all humans are born in a state where there is a solid homeostatic equilibrium which

### Segment 3 (10:00 - 15:00) [10:00]

means that your biological age oscillates all the time that's what Vadim by the way is showing that if you apply stress to a biological system the biological age is going up and down and then it tries to recover in mice it never recovers at the end but in humans it thought it totally recovers I think that's what we know from experience if you are young if you go to a party on Friday you pretend to be working on Monday and that's okay Elder people also want to go to party but they cannot pretend to work on Monday because it takes longer time for them to recover from the intervention so Aging in humans is not a matter of level of certain things like biological age the biological age is kind of irrelevant in humans because the biological age oscillates all the time you didn't sleep enough the biological age is going up you eat too much is going up your exercise is going down but it it's getting back to the equilibrium position what happens and uh for those of us who care to measure their weight along our life trajectory we know that once in a while we have kind of earthquakes in our bodies when the equilibrium position changes so once it happened your equilibrium weight is now a different thing and no matter how many exercises you do you would never get back to this I mean you can reduce your weight but when you stop exercising it will go back to a new level so what happens is that and mean you can never see it in the cross-sectional data sets but there is only one measurement if you have longitudinal data and we have lots of that right now you would see that instead of those smooth trajectories like you know all the biology papers are telling you about in humans you have fluctuations jump fluctuations and then on average you have those smooth curves so what happens with humans is that we have those jumps and very late in life we experience a very large jump that produces the instability of the biological age that becomes you know Frailty and accumulation of chronic diseases so in order to to prove that these things do occur like this we we concentrated on measures of Aging that are not the measures of levels of certain things like biological age we started the fluctuations of biological age from blood analysis from varable devices from mutilation doesn't matter from anything so what we observed is that when you have sufficiently long trajectories you can actually as I said your life is a source of perturbations it always perturbs you in such a way that your biological age is flu to 18 so you wouldn't call it even a biological age if it's fluctuating right it's more like a measure of stress it's really the measure of quality of your lifestyle your biological AG is going up when you're living toxic lifestyle it's going down if you're living a healthy lifestyle the challenge here is that if you stop doing what you have been doing is going back to the same level right so you don't care so the interesting part is that if you measure how quickly you go back to the equilibrium level you will find that the Elder you are on average the longer it takes to get back to the equilibrium level so on the horiz on the vertical axis is the inverse time the rate at which you approach the equilibrium level the horizontal line is the age and you can see that your resilience your ability to come back to the norm to recover from perturbations is going down to zero and would disappear at about 100 years old so this graph essentially is telling you that if you are lucky you don't have lots of diseases and you survive to till 100 years old your resilience is so small that even a minor you know stress event will kill you so this is actually Aging in humans so interestingly that contrary to Aging in mice Aging in humans is not related to biological age it's related to a dynamic property of the biological age I mean most importantly the recovery rate of the biological age and that's not just in a mathematical exercise so on the graph on the right I plotted the same I plotted the inverse Hospital stay as a function of H right I mean this is you have a life-threatening event so they put you in into the hospital it's a inhuman way of measuring uh human resilience right this is too much of a hustle to do it at scale but if you have B A Bank you can plot inverse Hospital stay as a function of Ag and you would you could see that exactly as we see in bi markers this thing is going down linearly with h and approaches zero which means that the recovery that the hospital stays infinite you are dead uh at about 120 years old so with that I want to conclude that in humans contrary to mice aging is not a property of the biological age levels this only happens when you are very old aging humans is a property of fluctuations of the biological age level and the human life history I can schematically I mean this is a schematic representation but all the parameters of these visualizations are actually computed from the data I can visualize like this so when you are born okay let let this guy Dead die okay he's dead now we have a new human being so we can see that the organism state is stable we are born into a stable state so life is hitting us you know ber us but we are we

### Segment 4 (15:00 - 20:00) [15:00]

are getting back but then as we age our recovery potential disappears and then a small shock throws us away and then our biological Edge becomes exponential like in mice then we hit the wall and we're dead so once again this is not just a metaphor because we have computed all the parameters of these things and this tells you that once again biological age is not relevant variable in humans for aging but the parameters of these regulatory interactions are the I mean the change of the shape of that regulatory interaction is actually agent so now you will tell me where are my biological clocks what are you doing I mean what you are doing so different from what other people are doing and as I said you have biological age and you have some other Aging in humans and this actually what you can see in any kind of biological signal you can take DNA mutilation you can take metabolomics promic doesn't matter so if you do clustering analysis of trajectories of biological Vari in humans you would find that in any aging data set in humans there would be one feature that is exactly linear with ag and then increases variance this is a Hallmark of a stochastic process by the way and then there are multiple features that depend on H in a nonlinear way and one of them would go to Infinity at exactly 120 years old so this is the weakest link feature that is disintegrating first so there are two kinds of agent clocks and we know that so there are agent clocks that predict from ological age and there are aging clocks that predict mortality and you can produce infinitely many other biological clocks that produce you know risks of specific diseases the most important part for us that there is one that is linear and all the others are actually driven by the linear biological age because they produce these nice quadratic curves which means that they that there is a small nonlinear coupling to the first one is Aging in humans the thing that is changing the regulatory interactions and produces death and the worst thing about this Aging in humans is that this is not a single biological process all the other guys are single biological processes and this one is not so this thing I mean the Aging in humans the linear feature in humans is a result of uncorrelated bad things that happen to you over life like I think what is telling this is the damage is called is through nonlinear coupling produces change into processes all the other things are processes and this is the damage why is the damage because it's linearly going up with time so this is just uncorrelated events that I get accumulated and why it's damaged because it's stochastic that's why the variance is going up exactly linearly with time so this is a poonian stochastic noise for those who care so this means that uh we are now in a very uh interesting I would say situation most of the things that we have learned in practice about aging is from mice if you a mouse you are a happy thing because we can rejuvenate you at least once or maybe a few times I think Vadim can do it we soon do it a few times in a row right I mean you can make kind of a leather of the biological H reduction in mice nicely once again short interventions produces lifelong Rejuvenation wonderful the problem with that is that this phenotype is not relevant for humans it's relevant only in the last 10 years of life once you lose the dynamic stability of course you can apply drugs late in life they will produce effects just one week ago I saw another results of clinical trials of antiaging drug in humans and everywhere this is the same in humans if you do it for old people there is a large effect if you do it in young people or relatively young people the effect is small once you stop the treatment the effect disappears right so which means that there are two kinds of Aging there are two kinds of anti-aging intervention one of them Works in mice rejuvenates a mouse works late in life in humans will produce infinitely many uh you know billions of dollars in sales because it will produce you know Improvement in quality of life of very uh old people but this thing will not affect your cognitive decline if you don't have Alzheimer this thing will not affect your V2 Marx decline so this thing does not change this linear aging that occurs all the time in humans so most of the industry is now working on this we call this phenotype Frailty Associated phenotype most of the industry is working here why because we're testing our drugs and mice and then there is another kind of Aging which is the stochastic damage accumulation which also occurs in mice but mice are de within so short time that it doesn't matter I mean it doesn't have time to contribute to their lifespan which means that the drugs that affect this true aging phenotype would not have large effect in mice they will have some small effect but not large effect so with that I would like to make maybe a few prospective statements like uh we know that if we do experiments in mice um most whole marks of Aging in are correlated if you have a drug that acts

### Segment 5 (20:00 - 25:00) [20:00]

against any H related feature in mice it reduces all other H related features in mice as well it's very easy to produce lasting effects there will be infinitely many papers about Rejuvenation and mice uh the very same drugs and I think this will be unfolding in a few years to come we will get the clinical trials now we have alterate there will be I know other papers published soon where will see that the same drugs that rejuvenate mice do not rejuvenate humans they produce small transient effects you stop the treatment the effect disappears which means that if uh we want to do drugs that produce multiple Improvement in human lifespan we should be doing something else and once again I would like to refer to Vadim since he came here Vadim made all those signatures of aging and animals uh of interventions and across different uh animals like mamals with different lifespan I think the message from that slide is the same the things that we are doing with interventions that are kind of primed to my uh studies are not the same that are required to produce multiple fold extension of human lifespan these are two different worlds of interventions and well um I think the there is a practical way to approach that so this is just for intimidation purposes so how to actually approach this what to do about that so to solve this problem we need first to study aging not in mice but in humans probably we can study it in other animals but we don't know what kind of animals to study so to understand how aging Works in other animals it's a lifelong project so I think we should better study Aging in humans just for practical consideration so what we do is that we take very large data sets of electronic medical records uh we're working with a data set of 10 Millions electronic medical records right now you can get 100 Millions easily there is another 60 millions in the UK so I think you can actually beef up this calculations by a lot so what we do is that we take electronic medical records and we use machine learning in order to Cluster diseases and to cluster of diseases with similar biology and we look at the electronic medical records longitudinally along the life history of patients and we try from the dynamic properties of these trajectories to predict what we don't know and is the total amount of damage in the system that is contributing into the probability of transitions between health and disease so this system infers biological age automatically and this biological age is the target variable for our studies so how does it work so we are feeding hundreds of diseases tens of millions of electronic medical records into these systems and the system outputs the biological H is a function of H by the way we don't fit well okay forget about that we don't Feit biological H to chronological H that's not a good idea but also on top of that the system tries to predict which systems in the body are broken and those systems are associated with related diseases so for example on the right graph you see the probability of certain systems which are predicted by Machine learning broken and one of them is associated with cardiometabolic health and this is by the way the probability of failure in this system is affected by biological age and it is exponential and doubles every eight years so you can see that the system produces the estimate for aging the biological age and it produces the estimates for the health of subsystems that are related to diseases and I think that the system recovers what actually the medicine knows without gero science so medicine has been knowing forever that we have a weak Link in our body this is a cardiovascular health uh this disintegrates first so this is the weakest link so that's why you have the highest occupation of that cluster and this is H related because the damage consists of infinitely many parts and those things are uncorrelated each of them is benign unless they cause cancer but that's another story and uh they accumulate over time and they produce linear stress on systems and let and and force them breaking so the nice thing is that this system is totally unsupervised we didn't ask the system to find cardiovascular diseases or other diseases we only ask it to find the biological age and some clusters of diseases and the system fights the biological age and diseases so what you can do with this is this you can now have if you have half a million people with genetics you have phenotypes you have this clusters of diseases you have biological age and now you can start looking for genes that are controlling the rate of Aging because you have the biological age and you can find that are controlling risks of diseases okay this is uh for demonstration not for intimidation we actually produced uh three genetic studies here just to give you a flavor uh of what's going on so uh here we have uh on the bottom graph we have lots of genetic variance in your body and whenever you have a vertical line it means that this genetic factor correlates to a disease on the bottom line we have the parental longevity Jus as you may know unfortunately genotyping

### Segment 6 (25:00 - 28:00) [25:00]

is a I'm kidding here unfortunately or fortunately unfortunately genotyping is a young technology which means that most people that contributed their genomes to banks are still alive which means unfortunately that we cannot use them for genetic studies of longevity but fortunately unfortunately uh two3 of their parents are dead and since some of the genes of parents are in the kids you can still correlate genes in kids to the longevity of parents so that's why it's called parental longevity so you can see that there is a bunch of genes that are associated vertical lines on the bottom graph that are associated with parental longevity and then we do the same genetic associations with cardiometabolic cluster from the model and from the biological age and you can see nicely that uh if you have any heat in Parental longevity almost all Heats are either in cardiovascular health or in aging but it's never like there is one heit in the cardiovascular health and aging so it's nicely that the system separates genetic factors that are risk factors of accelerated aging and separately uh genetic factors of diseases and this is the point where Pharma starts liking you because Pharma knows that aging is not reversible I mean technically they need to do phase two clinical trial within a year or two right if you cannot reverse something within two years it's irreversible and they want to regress this out from the calculation so we ended up ourselves in the business of regressing out Aging for farma I mean they for the God's sake don't make it against aging because that would be too long to try right so the system outputs aging and disease risks we regress out aging from our calculations we help uh we inform Fara about human evidence confirmed targets against diseases and we have still aging as a phenotype that we can associate uh with genetic factors and everything else so that brings us to the point where I mean first we learned how to rejuvenate the mouse separate aging from diseases in the data we learned how to partner on this knowledge from humans by the way not from animals but from humans with those guys who know how to do drugs and still we have aging phenotype as a separate entity that we can study on our own and uh make experiments that are now squarely aimed only at search for factors that are controlling the rate of human aging so with this I would conclude Alex is here we have wonderful team we have a bunch of people like Brian like we have u a number of Publications with wiim we have andreaa who is helping with the genetic studies so we are you know looking for more brains and hands and uh computers as usual and data uh to make this all possible um I think that uh as I said uh the industry is I believe in is on a very firm footing to solve the Frailty phenotype we have now thousands of companies entering clinical trials we see the data from clinical trials I mean when we get old and sick there will be a solution that would help us a little bit I think now it's time to raise Stakes substantially and go after the Aging phenotype in humans that is not Frailty in order to you know help most of the people sitting here not to get to the point or where the frity sets in as uh late as possible so that's about it thank you very much and if you have questions