 Welcome back to the second session and we have four very diverse talks again, so this is exciting. Just a reminder, when you write your questions in the chat, and thank you for getting some great questions, please write them to the panelists and attendees, not just to the panelists, so the attendees can also see the questions, and that's also so that if we get too many questions to be answered after the talk, then this speaker will type in the answer and then everyone can see the question and the answer. All right, great. So our first talk is from Mania Elmai, if you want to see if you can screen share. So Mania is from the Institute for Research on Cancer and Aging of Nees at Inserm in France, and he's going to talk about the switch from apoptosis to senescence, results from conflicting tissue, proliferative demands to cells unable to proliferate. So good afternoon everyone, I would like first to thank the organizer for this great opportunity for me to present that work. I'm Mania Elmai, and I'm a postdoc in Miguel Golinos Ferraraslav, and today I will present you our recent work showing that a switch from apoptosis to senescence results for conflicting tissue, proliferative demands to cells unable to proliferate. Chromosome ants also name as telomeres shortened during each cycle of cell division. This phenomenon can be counteracted by telomeres, but the expression of this enzyme is limited and is not sufficient to counteract telomere shortening with age. So telomere shortened during each cell, its cycle of cell division, until they reach a critical telomere length where they trigger DNA damage response leading eventually to either apoptosis or senescence. However, we currently don't know what determines the cell fate between apoptosis and senescence. So in order to address that question, we are using premature aging zebrafish model which are zebrafish line the tert mutants where telomere is deficiency lead to faster telomere shortening. Indeed, as you can see here in wild type fish, telomere shortened with age. But in tert mutants, this telomere shortening is anticipated in these fish leading to a 12 month old telomere length that is similar to those of wild type at 24 months. This lead to a premature aging phenotype as you can see here at 10 to 12 months in tert mutants that is similar to the phenotype of 24 to 36 months old wild type. And as a consequence, the survival of the tert mutant fish is reduced compared in red. I guess you can see in red compared to the wild type. So we first looked at those young fish trying to compare different tissues of comparing tert mutant and wild type. For this talk, I will focus on the guts but we also describe the similar phenomenon in other proliferative tissues. So as you can see here in this H&E staining, we couldn't see when we looked at young fish, we couldn't see any difference in term of morphology between the tert mutant and the wild type at young age. However, at that age telomere shortening already induced DNA damage response. As you can see here with this increased levels of Gamma H2AX and P53. However, we couldn't detect any difference in term of senescence looking at P16 protein level mRNA level or P21 mRNA. In fact, at that age the damage response and P53 induce higher levels of apoptosis in young tert mutant compared to a wild type. As you can see here with this terminal assay and quantify in this graph here. But we couldn't detect any sign of senescence looking at P16 IF or ACB regal. In contrast, when we looked at all tert mutants, the all tert mutants that I showed you earlier, we couldn't see any difference in term of apoptosis between the tert mutants and age match wild type here. However, we noticed an increase in P16 positive cells by IF and quantify in this graph here and an increase in ACB regal staining. So at that age in all tert mutants senescence become predominant. So we show that there is a transition between apoptosis to senescence with age of the tert mutant and at all age senescence becomes predominant. But what determines cell fate between apoptosis and senescence? Considering that P16 accumulation has been shown in the literature to be linked to mitochondrial dysfunction, we wonder whether mitochondria were affected in our all tert mutants. And as you can see here, when we looked at ros levels in young fish, we couldn't distinguish any difference between the tert mutant and the wild type. However, the ros level increased gradually and becomes higher in the tert mutant compared to the wild type at all age. This was associated with an increase of mitochondrial dysfunction, as you can see here in the all tert mutants, where we could see more swollen mitochondria, rounded mitochondria and membrane disrupted mitochondria. And this was also associated with a decrease in ATP levels. So why would mitochondria be dysfunctional in all tert mutants? Oxford's defenses guarantees the mitochondrial homostasis. So we went to look for that. And as you can see here, we noticed that SO2 levels, one of the major Oxford's defense protein, is reduced in all tert mutants compared to the wild type, which could explain the increase of ros levels and mitochondrial dysfunction. But why would SO2 be downregulated in these fish? SO2 gene can be transactivated by foxal protein. And foxal can be translocated outside of the nucleus and inactivated by phosphorylation. And interestingly, we noticed that in all tert mutants, there is a higher levels of the inactivated form of foxal protein. We also confirmed the translocation of foxal outside of the nucleus by co-immunofluorescence staining of foxal and P16. And we quantified the intensity of nuclear to cytoplasmic ratio intensity of those two factors and plotted in this graph. And as you can see here, in the in tert mutant in red and orange, the majority of the cells have high level of P16 and low nuclear P16, sorry, and low levels of nuclear foxal compared to the wild type ingrate. But why would foxal be inactivated in these fish? AKT is known to forcerate foxal and inactivated. And interestingly, we also noticed that in tert mutant, we have higher levels of the activated form of AKT, phosphoryl AKT. So we showed that in all tert mutants, we have inactivation of the mTOR-AKT pathway that will inactivate foxal, leading to a reduction of sodium expression, triggering higher loss or loss levels and mitochondrial dysfunction, and even eventually leading to senescence. But why would this property pathway will be activated in a context of cell proliferation block? When we look at when we compared young and old tert mutants, we noticed that in all that the old tert mutants exhibit tissue damage. So we propose that in early age of tert mutants, p53 is activated by tiramine shortening and trigger cell proliferation block and activate apoptosis. This will lead, this will impact dampen and impact the cellular replenishment of the tissue and lead to a gradual loss of cellularity and tissue damage. The tissue damage in turn would activate a compensatory proliferative pathway, the mTOR-AKT pathway, which in a context of cell proliferation block would induce senescence. So if that's true, if this model is true, by inhibiting p53, we expect to rescue senescence. So in order to test this model, we then inhibited p53 and use a tert p53 double mutant. And we showed that when we inhibit p53 in a tert mutant, we reduce the level of activated ROS as you can see here compared to the tert mutants and rescue salt to expression and senescence compared to the tert mutants. Conversely, by inhibiting mTOR pathway, we also expect to rescue senescence. And we also use then a model where we inhibited the mTOR pathway in this model here. And we showed that by inhibiting mTOR pathway in tert mutants, we reduce the activation of AKT, and therefore we reduce the levels of p16 and therefore senescence. So in conclusion, we showed that during age of tert mutants, we have a transition between apoptosis to senescence. That is due to the concomitant activation of two antagonistic pathway, the mTOR-AKT prop-properative pathway and the p53 anti-properative pathway. And this activation of these two pathways leads to senescence. So to finish, I would like to thank you all for your attention. I would like to thank also my team and my team members, sorry, the team members, and the co-authors of this paper, the different facilities and the funding agency that allowed for this work to be done. And if I may, I would like to add that we recently moved to Nice in France, and we are welcoming postdoc and student applications. Thank you all for your attention. Wonderful. And thank you for being on time. Thanks. Can I ask the first question? Sure. So, I mean, this is really cool. I don't know if you've had chance yet, or if you have any reason to think that this mechanism in zebrafish, and I think that's an amazing model, is this also conserved in mammalian cells, for example? Would you think? So we haven't, we didn't get the chance to test that yet. But we believe that this could be also a phenomenon that we could see in mammalians also. The thing is that in the mouse model, we will have to wait more generation of mutants of telomerase to see telomere shortening, and the effect of telomere shortening in these models. So, and the other thing that's using in vitro models, for instance, to study human cells. In vitro, we are using a lot of growth factors that would then induce the proper pathway, like the AKT pathway. And it will be trickier to disentangle that question in this model. But do you know if telomere shortening activates P53 in mammalian cells? For instance, in human cells, P53, telomere shortening activates P53. And also in mice in the late generation of telomerase deficient mice, you have activation of P53. So that brings me to Alex Chen's question. Which is very similar. How does P53 mechanistically sense the shortened telomeres? So actually, so when telomere shortens, they become deprotected. So telomeres are nucleoprotein, nucleoprotein complex, composed by the telomere repeats sequences, DNA sequences, and proteins that are called shelterings. And they protect the extremity of the chromosomes. And when they become extremely short, they become deprotected, and are recognized as DNA damage by the DNA damage response machinery. So it's ATM activation? ATM ITR activation that would then trigger P53. Okay, cool. So while anyone else is wanting to type a question, I just wanted to say that Miguel, your boss, if he's there, hi. When I was a new faculty member 21 years ago at the University of Colorado School of Medicine, Miguel was a postdoc there. He was working on telomeres in Pombi with Julie Cooper. And he used to come up to my office and we used to have these long conversations about DNA damage checkpoints. So watching his work evolve from telomeres in Pombi to now in his own lab working on zebrafish and aging. And I'm sure that's nice, Miguel. When we're allowed to travel. Thank you very much, Jess. It's a pleasure to see you. It's very nice to be here. And also doing dog sitting, which was very nice. Okay, if we don't have any more questions, we will move on. Wonderful stuff. Thank you. Okay, so we're gonna nice now have a talk from Morgan, Morgan Levine, sorry, at the Yale School of Medicine. And the talk is entitled a rat epigenetic clock recapitulates phenotypic aging and co localizes with heterochromatin. Okay, thank you so much. Thank you to the organizers for having me. I'm really excited to present this work, which is really, even though I'm presenting at a collaboration between my group and also Rafa de Cabo and Luigi Ferrucci at the NIA. So just to jump right in, a brief background on epigenetic clocks. So basically what these are is these are estimates of biological age based on measures of DNA methylation. So we think these are really important because DNA methylation is an epigenetic mark that plays a really important role in a number of different cellular processes. But what people have found is that there are striking changes in the patterns of DNA methylation across the genome as cells or individuals age. So what people have done is we've used advanced kind of machine learning techniques to estimate biological age, using these patterns of change across DNA methylation. And clocks have been developed using as few as one CPG location, too often up to a million CPG locations. So the first epigenetic clock was developed in humans actually using human twins in 2011. 2011. That was the Buckland at all paper. And since then there's been a plethora of other epigenetic clocks developed mostly in humans and some in mice. And actually some of you might know, so currently Steve Horvath is actually doing mammalian epigenetic clocks so across hundreds of different mammalian species. But the aim of this paper was to develop an epigenetic clock for rats, which at the time had not been published yet. So the other difference kind of that with our approach versus some of the epigenetic clock approaches that have been used before is that we really try to use what's called an unsupervised machine learning approach. So just to put that in context with what's been used. So most of the epigenetic clocks that have been developed used what's called the supervised approach. So this is you traditionally train a clock by using some measure of ground truth that you're trying to predict. Most often in the epigenetic clocks, this has been using chronological age. So what we we can call these clocks, chronological age predictors. However, there's a number of drawbacks for this. Number one is that the whole point of making a biological age estimate is to decouple biological aging for chronological aging. Because we know chronological age is an imperfect proxy of biological aging. There's also issues with mortality selection. And the fact that you actually might be picking up epigenetic changes that may not be actually causal or important in aging, but are just bystanders of changes with time. So so the idea with the supervised approach is that you need some ground truth. However, when we're trying to estimate biological aging, we think of this is what we call a latent variable is something not observable. So there's actually no real ground truth. And there's been some debate in the field on what people should actually use as this ground truth in which we're trying to predict. So for this study, we actually took a different approach, use an unsupervised approach. So this is not tethered to any variable. We basically are just looking at patterns of change in our data that were previously undetected. So we don't need to determine that we're trying to make an age predictor. And assuming you have a data set with sufficient kind of variability in this dimension that you're interested in. It actually might be ideal for uncovering some of these kind of latent or unobservable patterns. And this is the case for biological aging, that because it's latent, we actually want to understand these kind of multidimensional patterns of change rather than making an age predictor per se. So for this, again, I mentioned we're doing this in rats. So we had male Fisher rats between the ages of one and 27 months. And we had a really nice age distribution where we had six rats at each month of age. So one month, two months, etc. We also had data that we use for blood to measure DNA methylation. We also had information on facts. So we knew cell composition, and there's behavioral testing done. So we have information on rotor rod open field as well. So to develop our epigenetic clock, we split our data into what we call training sample, which was about 100 grats. And our testing or our validation sample, which was about 30 rats. And first, we actually tried a number of more sophisticated unsupervised approaches that considered kind of nonlinear facts. And really what came out was that actually a very simple one principle component analysis did just as well. So we ended up just going with that because it's better just to keep it simple if possible. So what we did is we ran principal component analysis. And really what we found is that the first component actually captured this aging effect that we were interested in. So this is just showing, this is called a scree plot. So the proportion of variance explained by each of these components. So this is the first one. And this is if we just plot age. So this is chronological age against this first principal component, which we converted into units of years in our validation sample. And what we find is that you actually get a correlation between age and this principal component of 0.93 in the validation sample. So the next thing is again, you know, there's some debate on how do you validate a biological age measure. So yes, it should track with chronological age. It's important to have a correlation, a high correlation with chronological age, but more importantly, is actually relating it to other aging outcomes of interest. So for this, we use composite measures of cell composition using the facts data, as well as phenotypic aging measure using the Rotorod. And that should say open field and open field data. And we put these in a multivariate model. And basically what we find is that when you adjust for age, we still see an association with our epigenetic age measure and this phenotypic age aging measure, which means even among rats of the same chronological age, having a higher epigenetic age is associated with higher phenotypic aging. And the same thing holds even when we adjust for cell composition. So this is not explained by differences in kind of cell proportions in the blood data. We also then actually took this equation and used previously developed RBS data in mice and actually applied this exact same equation to the mouse data. And what we find again, this is the rat data, is that we actually get a high correlation between age and this epigenetic aging measure in mice. And interestingly, we also had mice that were either calorie restricted versus controlled. And what we find is that the calorie restricted mice had lower epigenetic ages. And I'm not showing this figure, but we actually also showed that the longer the mice have been calorie restricted, the bigger the difference in their epigenetic ages. So it seems like it's this kind of cumulative thing, or that it actually might not just be lowering your epigenetic age, but actually slowing the rate of change in epigenetic age. But, you know, the one thing that we kind of struggle with with these epigenetic clocks are that they end up being very much a black box. So these are composite measures of a lot of different changes happening in the methylome that might not actually mechanistically map on to the same thing. So it's really important to kind of step back and maybe take a more reductionistic approach where we can decompose these signals, and then try to map them onto specific epigenetic changes that are happening with aging. So for this, what we did is we use a clustering approach, so a network analysis approach. And we're actually able to take the approximately 3000 CPGs that were in our measure, and actually cluster them and assign them to these different comethylation, what are called networks or modules. So these are groups of CPGs that seem to be operating similarly. They're changing the same across different samples. So we found four of these modules. And actually most of the CPGs in our measure were not assigned to a module. They were not strongly connected to a bunch of other CPGs that were changing. So that was about 95%. So then what we can do is say, let's look at our original measure and see which of these modules is playing a big role in the overall epigenetic age score. So this is showing the PC loading, so how strongly, so the absolute value of the how strong a role it's playing in that score. And when we find our CPGs in the screen module have a large influence on our epigenetic age score in both directions. So these are CPGs that are becoming both hypo and hypermethylated with age. Conversely, CPGs in the purple and blue modules tend to be ones that have a strong positive association. So these are ones that are becoming hypermethylated with age, whereas the pink ones are not influencing it that much. And the gray, these are unassigned or kind of random, but really would be kind of centered at zero. So not having a strong influence. So next, what we did is we actually made what are called sub module scores. So again, we split our epigenetic clock measure into a bunch of these modules. And then we can just make a measure a clock specific to a module and ask what are its associations with age with phenotypes, and with color restriction. So these plots at the top are just showing the age correlations for for these sub module scores. So you find that in both rat and mouse, the green module has a high age correlation, as does the purple and slightly weaker, but we also find it in the blue. And again, there's no age correlation in this kind of pink module. But again, going past that, we looked again at the association with our phenotypic age variable, after adjusting for chronological age, and it seems the green and the blue modules were associated there. I'm not showing the actual statistics, but they're in the paper. We also find that all four modules were reduced by color restriction. And then we also looked at reprogram fiberglass or reprogram into induced polypone stem cells. And in that case, only the green and the blue module seemed important. So for this reason, we started to focus in more on the green and blue module as they were associated with all four of these phenotypes. And so we looked more at kind of the genomic features of these two modules. So basically what we find is that both the green and the blue module tend to be associated with CPGs in these intergenic regions. They also tend tended to be in regions of high CPG density. So these might be what we call CPG islands. And we also were able to find what they co-located in. So there was enrichment for locations that co-locate with H3K9 trimethylation and H3K27 trimethylation. So in conclusion, what we did in this paper was we presented the first published epigenetic clock in rats. We found there's robust epigenetic age. Oh, we found the robust epigenetic age biomarkers can be developed even using unsupervised machine learning methods. So you don't have to use chronological age or some other proxy measure to train to. So again, in part one, we generated this composite epigenetic age measure was associated with phenotypic or functional aging conserved in mice. It was reduced by calorie restriction. And I didn't show this data, but we also have a figure in the paper showing that it can be reprogrammed via Yamanaka factors. In part two, we actually were able to decompose the signal and identify two really important what we might call modules or parts of the clock. They're enriched in intergenic regions and they overlap with H3K9 trimethylation H3K27 trimethylation, which really kind of points to this idea that it's actually capturing something about changes in heterochromatin structure with aging. And with that, I just want to acknowledge the team that participated on this group, especially Luigi and Rafa, who are actually together two corresponding authors on this paper, and also our funding and happy to take questions. Thank you, Morgan. That was fantastic. Okay, so we have a question from YL. The first PC is used as DNA MH. Here, the PCA is based on all the CPG methylation data from the microarray. So that's a question. Is it you based on all the CPG methylation data from the microarray? A second question. The first PC can explain the most variants in your study. Is it specific for your current data? I mean, the first PC may be not cannot capture the most variants in other data set. Yeah, so to start at the first question, so yes, we actually ended up using the first PC. The first PC here is based on we had RBS data, so it's not a microarray. So for humans, we use the arrays for mice. Actually, there are arrays that were just developed. But for this, we use RBS data, we started with a few million CPGs. But what we did here was just take the CPGs that are overlapped between our study and the mouse validation studies. So it was based on all the CPGs that are overlapped. It was, I think, only about 3000 though. So yes, the first PC is based on all of those. For the second question here, I agree. I think the reason the first PC worked so well was because if for people who know about PCA, most of our variants was in that age domain, right? We had a really wide age range. And these were all fairly homogeneous rats. So I think that's why the first PC ended up being this aging measure. We've done a lot with these unsupervised approaches in humans. And that first PC is not always the one that's of interest. And it actually doesn't really have to do with how much variance it explains. It's really, you know, it might be the fourth PC that's of interest. I don't want to go too much into the statistical details. But yeah, it's not a given that that first PC is going to be the one that we should focus on. Okay, Maria Ricchetti has a question that I too was wondering, how long are you doing the calorie restriction for? You mentioned there was if it was longer, you had a larger effect. And also what was the level of the CR? For example, was it 30% less than normal or ad libido? To tell you the truth, I don't know that off the top of my head. I wasn't the one that did the calorie restriction. But I believe it's in the paper. Yeah, I can't remember all the animals were started at the same time. And then we assessed methylation at different ages. So the one the older animals had been on calorie restriction longer. But yeah, I'd have to throughout the whole lifespan. It wasn't that long. It was. Yeah, I don't want to say because I might be off. But yeah, it wasn't. Okay, we have another great question from Joris Deeland. Have you also measured additional omic layers in these rats, such as proteomics, metabolomics and transcriptomics, to see if these can improve your clocks and to determine if these epigenetic changes represent changes in specific molecular pathways? So we have not. I believe that they that rough and Luigi do have some stored plasmids here from the rats. So you know, there is a possibility to do that. I'm actually working on a collaboration with them and also Betty and Gladyshev to do this in the SLAM cohort, which is a mouse cohort. But in that example, we are hoping to have multi omics and really actually be able to dig into what pathways are being affected. And to get a, you know, maybe more of these multi omic aging measures, you know, as opposed to just focusing on methylation as well. So I mean, I was kind of wondering, with the, you said this intergenic regions, are they in heterochromatin? Are they regions of known, you know, repetitive elements like line elements? Because we know they get activated during aging and also in cancer. And there's a lot of really cool recent work saying that they promote aging when they're expressed. Yeah, so it actually it was interesting when we looked at these regions, if you look at the mouse annotation, they did say that they were in repetitive regions. The rat annotation did not. So we didn't really go into it that much in the paper. But there is a hint that that actually might be what's going on here. Turnix would be really cool. Yeah, absolutely. Expression. All right. And we also have the link to Morgan's paper. Okay, so I think we're pretty much perfectly on time. Thank you, Morgan. Thank you so much. Okay, so our third talk in this session is from Dudley Lamming from the University of Wisconsin in Madison. Talk is that overreact to me uncouples lifespan for metabolic health and reveals a sex hormone dependent role of hepatic M top two in aging. And your slides are up. It's looking good. So take it away. Great. Thank you very much. So many of us are familiar with the idea that rapamycin extends lifespan. So rapamycin is a drug originally found on Easter Island or as a native call it Rapa Nui. And many different studies by the NIH's intervention testing program and others have shown robust extension in both sexes using this drug. And it's shown many beneficial effects in some other organisms as well. So why don't we all start taking rapamycin? Well, rapamycin is approved as an FDA approved immunosuppressant. And the number of studies that have been done so far on healthy people taking rapamycin is very limited. But overall, there are some studies that show that side effects in people taking relatively high doses, including increased infections, various dermatological events, and metabolic consequences as well. And this was something that I was really interested in when I was a postdoc starting to look at some of these metabolic effects of rapamycin. Typically, they're not what we would normally think of as associated with longevity in general. They include hyperlipidemia, decreased insulin sensitivity, glucose intolerant, and peers that there's an increased risk of new onset diabetes as well. And so when we began digging in to this, we found that we could reproduce this in mice, in multiple strains of mice. Here we show black six mice on the left, as well as het three mice used by the NIH's intervention testing program on the right. It applies in both males and females. It doesn't matter whether you deliver rapamycin to the IP or in the diet, you get very robust glucose intolerance. And so what is the molecular basis for this? We don't have time to go into the full story today, but using hyperinsulinamic euglycemic clamps, we found that there was an effect on hepatic insulin sensitivity. Rapamycin treated animals produce more glucose from their liver in the basal condition during the clamp. And when high levels of insulin are used to suppress that hepatic gluconeogenesis, vehicle-treated mice experience much more suppression than rapamycin treated animals. Now, rapamycin's molecular target is called mTOR complex one. It's a protein kinase that regulates many different longevity pathways, including S6K and 4BP. And it's sensitive to stresses including amino acids and glucose. And what we did was to knock out mTOR complex one the liver genetically. And interestingly, we found this had no effect on glucose tolerance or hepatic insulin resistance. And so we began to think about the possibility of rapamycin hitting a second target, mTOR complex two. mTOR complex two is primarily a factor of PI3K signaling. And it doesn't directly regulate the same set of targets, although probably it is functionally upstream of mTOR complex one. And while mTORK two is initially characterized as rapamycin resistance, some work in David Sabatini's lab by Doss Sabarus in 2006 had shown that in cell culture and to agree and at least one mouse tissue or two, mTOR complex two could be inhibited by chronic rapamycin treatment. And so we wanted to know whether this was true in the liver as well. And to do this, we treated mice with rapamycin. And we then performed immunoprecipitation experiments where we IP to mTOR and looked for the association of the mTOR protein kinase with either Richter subunits that's specific to mTORK two or Raptors units that's specific to mTOR complex one. You could see vehicle treated mice here on the left. Essentially, there's good association of mTOR into both complexes. Whereas in rapamycin treated animals there's good disruption of both complexes. So to complete these studies, we made a mouse model in which we deleted Richter specifically in the liver of specifically in the liver using an albumin Cree driver and call these mice the L dash RKO mice will show talk about them in the context of this E life paper shortly. And these mice we characterize as having glucose intolerance as well as pyruvate intolerance as well. And so we're able to reproduce the effects on hepatic insulin sensitivity simply by deleting Richter specifically in the liver. And using clamp studies, we were able to show that indeed mTOR complex two mediates these effects. So this led us to try and address a question that I actually started working on and thinking about about 13 years ago now, which was does rapamycin actually promote longevity via inhibition of mTOR complex one? Or does mTOR complex to play a role in this effect as well? And so to address this question, about six years ago, we published a paper looking at the lifespan of animals where we disrupted Richter genetically. Here on the top looking at the black versus the blue the blue are Richter heterozygous mice. They have one deletion of one copy of Richter this mTOR complex to subunit. We got about a 40% inhibition of longevity. And in males, we saw about a 30% decrease in longevity when we deleted Richter specifically in the liver. And so interestingly enough, this effect was sexually dimorphic. And so the question we wanted to try and address was, what is actually happening here? And why is mTOR complex two inhibition so deleterious in the case of males? And so we started thinking about this. And of course, sex hormones underlie many different sexually dimorphic phenotypes. And it's relatively straightforward to start addressing that question. And so we use a pre pubertal go and ectomy approach to try and understand whether sex harmony mediated the effect of hepatic Richter deletion on a lifespan as well as metabolism. And our initial hypothesis was that either castration would protect the Richter knockout males, essentially by feminizing them, or that to overectomy would sensitize the liver Richter knockout females by removing the protective effects of estrogen. So in this paper, what we did was we characterized the phenotypes in mice longitudinally, we looked at a number of different phenotypes. Of course, one of the most simple is weight and you can see here that the wild type males are actually the heaviest throughout the study, whereas castrated males as well as liver Richter knockout males have a bit of a lower weight. In terms of females, the overectomized mice weigh much more than the wild type or the LKRO sham surgery animals. And this actually fits pretty well with our expectation about what go and ectomy should do to animals. We know that testosterone in particular is associated with building lean mass. And we looked at the body composition of these animals using an echo MRI scanner. And what we found was that the castrated animals of both genotypes indeed had less lean mass than their intact counterparts. In the case of overectomy, overall, we know that estrogen in females is generally believed to be protective against metabolic dysfunction. And we saw that there's a lot of more adipose mass in early life in the overectomized animals than in the intact females. And you can see that those differences disappear around 16 to eight months of age. Mice don't go through menopause like humans do, but they sort of go through a reduction in estrogen around this time period in their life. And probably that's why the body compositions here begin to come together. Looking at metabolism, we found interestingly that castration didn't on its own have a particularly negative effect on metabolism in the animals. And that as we expected the liver Richter knockout males did indeed have glucose intolerance. And you could see that there's an overall effective genotype as well as an individual effect in both the sham and the castrated animals. In females, estrogen is believed to be protective against diabetes, as well as obesity. And so we expected that there would be a negative effect on glucose tolerance when we perform anectomy on these animals. Indeed, that's what we found. We found that there was an overall effect of overectomy in terms of impaired glucose tolerance in addition to an overall effect of Richter deletion. Now, one of the things that we wanted to do in this study was trying to get some insight into what the effects of liver Richter knockout as well as the good ectomies had on a cause of death and cause of death analysis in mice is very difficult. But one thing that is relatively easy to do is to try and understand whether the animals die with cancer. You can't necessarily say that they died from cancer, although I think we can draw that supposition in many cases if we see that the animal has a lot of cancer. But we necropsyed the animals and just scored whether or not they had cancer during of this gross necropsy. And so in males, interestingly enough, we found the mice that died with observable cancer. There wasn't really any differences between any of the genotypes or surgeries. So all the groups had very similar effects. In contrast, the mice that we didn't see cancer during necropsy, those are the animals that seem to be living shorter. In the case of the males, you can see that the liver Richter knockout males here in red have a reduced lifespan. And there's an overall increase in the hazard ratio when we delete Richter. You can see the castrated animals very similar overall effect. Castration did not protect the animals in this case. So if we put all the mice together, including the ones that where we were unable to score cause of death or cancer at death due to the fact that the animals had decayed too quickly, what we find is that, again, Richter knockout has a negative effect on lifespan. So that's reproducible. It was smaller in this particular study. All the animals in this case lived longer in the mouse facility we are in at UW Madison than in my former lab when I was in David Sabatini's lab at MIT. So the animals live significantly longer here, but there's still a negative effect of liver Richter deletion. And castration had no positive effects on those phenotypes. Now in the females, again, we found that when we found the mice that died with cancer, there wasn't any effect between any of the genotypes or gonectomy groups. But there was a really interesting effect in the females. So in the females, the sham surgery animals had a vastly reduced lifespan relative to all of the other groups. And so these are mice where we didn't find cancer and necropsy. Those animals were shorter. And this was very unexpected because in our initial study, we found that there was no effect of Richter loss on female longevity. And overectomy rescued these animals. So if you remember our original hypothesis, we thought that overectomy would actually be protective or rather negative for the animals because they'd be more like males which had this negative effect on lifespan. And we found that the opposite is true. So overectomy seems to protect those animals very dramatically. When we look at the overall effect, we see that there's no overall effect of Richter. And in fact, there's, again, still no change significant change in the overall lifespan when we delete Richter in female mice. Overall, there's about 6% decrease and it doesn't reach statistical significance by law-ranked tests. And overectomy seems overall to be protective, although that's driven by the dramatic effects on the liver Richter knockout mice. And so in conclusion, our conclusions are quite interesting and we're not what we expected. So first of all, the sex-specific effect of reduced hepatic m-torque complex 2 on lifespan that we initially found was not reversed by the pollution of sex hormones in either males or females. And we found really surprisingly that loss of hepatic m-torque complex 2 negatively impacted survival those females who did not die with observable cancer. And that doesn't mean that they didn't have cancer at all. We didn't do the type of detailed histology that would be necessary to make that conclusion. But certainly there was nothing observable during our gross necropsy. And overectomy was protective of those animals. So it protected mid-life survival overall of female mice lacking hepatic m-torque complex 2 and it does this by increasing the survival of those mice which do not develop cancer because it doesn't seem to have an effect on the mice that die with observable cancer. And finally, we found that pre-pupil overectomy really uncouples lifespan for metabolic health. So the overectomized mice typically lived longer and particularly in the case of the protective effect on the Richter female knockouts. Despite the fact that they were heavier, they had more adipose mass, they were glucose intolerant, they had some insulin resistance as well. So metabolically they weren't healthy but if anything they tended to live longer and there certainly wasn't a negative effect on lifespan. So I'd like to thank everyone who was involved in the study that was really driven by my former postdoc Sebastian Ariola Paolo who's now faculty at UW Madison. But many others in the lab contributed and I'd like to particularly thank AFAR which was primarily responsible for funding this as a junior investigator grant. Thank you. Thank you Dudley, great talk. Although you're making me feel really cold there. Are you not cold? Well, people were saying that my backyard couldn't possibly look like that anymore when I was showing the summer photo. So we just got a nice snowfall yesterday. Oh, that's that's from yesterday? Saturday rather. Wow. Okay, we got some good questions here. So from Alex Chen, how about ketosis and M talk to inhibition, i.e. rapamycin and ketosis? Oh, that's an interesting question. So we have been doing a project where we're trying to characterize some of the effects on metabolic pathways. We do see that there seems to be some additional ketones, I believe in the male rector knockout animals, but not the females, I think. And we don't exactly understand why that is, but certainly it makes sense that the animals probably through an inability to sense insulin probably think that they're in a bit more of a fasted state. And just to clarify, Alex added, i.e. are the effects of m talk to induced glucose intolerance, not detrimental in the context of a high m u f a keto low carb diet? That's an interesting question. We did I did do a study that that we never got around to publishing. It was a relatively small study where we tried feeding the animals a high fat diet to see if that would compensate for the fact that they have a lipogenesis defect as well. And there was no there was no effect of a high fat diet in terms of being able to rescue those animals. Now that wasn't exactly a ketogenic diet, but it was high fat rather in Western. So it was not not there wasn't elevated levels of sucrose there. So I'm, you know, I work on aging and yeast. They don't. They don't have livers. So are you saying that high doses of rapamyzing can predispose to diabetes? Well, it's been shown that in humans, people who are on immunosuppressants, and this is complicated by the fact that all immunosuppressants typically have some negative effect on metabolism. But people have shown that serolimus in particular increases the overall risk of developing type two diabetes. And so some people say that's like a 20% elevation risk. I think I've seen as much as five fold. But certainly there most studies agree that to the extent that can be parsed out because people are on multiple immunosuppressants, that does increase the risk of new onset diabetes. You know, something that's relevant, right, is that people who are already sick may not have the same exact profile. And a lot of people who take this for tubular sclerosis are taking very high doses. And so, you know, it may not be relevant to to a pro longevity phenotype. But that's definitely been seen in humans. Max, do you see that in the dogs? No. But I think again, there's there's a couple. I mean, we have first of all, our longest study in dogs has been six months, right, of treatment. So so I think there's a duration component and a dose component. So I think Dudley, the way Dudley explained it, I think is is is right that, you know, in people, these are sick people taking high doses of the drug in combination with other things. And there there clearly is an elevated risk for metabolic consequences. We just don't know in healthy people, you know, or dogs or mice taking lower doses, how how important this is. And since I'm speaking, I'll just I'll just give my sort of opinion on this. I think the question in my mind is, we know that the the mice that are long lived as Dudley and others have shown repeatedly on on the doses of rapamycin that extend lifespan have what would typically be thought of as glucoregulatory phenotypes, right there. They have a loss of what people typically would consider glucose homeostasis. What's unclear to me is whether that phenotype is bad, good or indifferent for longevity. I think we just don't know at this point. My gut feeling is that what it's really reflecting is a change in the way that mice on rapamycin are what their preferred metabolic state is. And it it may not be bad, but if you give them a non physiological glucose tolerance test, they don't perform well. And so I think this is still remains to be there's a lot of work that needs to be done to really understand what's going on. And I think Dudley's, you know, one of the leaders in figuring this out. Thank you. Yeah. And Alex suggested take metformin with rapamycin. No, that's certainly it's certainly a possibility. And mouse studies have shown that I believe that that's protective in one sex, but not the other. But I don't remember which one. OK, so Nainita asks, what is the effect of high animal protein diet on the mTOR? I mean, overall, it's been shown by Steve Simpson and Samantha Solon-Bignet that dietary protein levels correlate very well with mTOR activity in the liver. And we and Luigi Fontana have shown that in mice, generally speaking, protein restricted diet seem to lower mTOR activity in multiple tissues. So it seems to correlate very well, dietary protein mTOR. And Kai Zia asks, what to expect in your study if the mice were castrated when they were adults? Well, that's a very interesting point because the existing literature actually suggests that castration as a rather overectomy in post-puberty mice is actually detrimental for lifespan, whereas in our study, we see a positive effect. And so, you know, I think overall that studies in insects and C. elegans generally have supported the idea that overectomy is beneficial to lifespan. And I think maybe that's in young animals more started pre-puberty before reproduction begins, whereas this post-puberty overectomy definitely in rats and mice has been shown to be detrimental. We were a little bit surprised about the castration result because castration has also been shown to be associated with longevity in multiple species, including humans. But in those cases, there's an aggression factor that also needs to be taken into account. And so perhaps in our mice, which are relatively placid and happy and not aggressive, that wasn't really a factor. Or it might be a factor, the fact that we did castration prior to puberty. All right. Thank you very much, Dudley. Thank you very much. The talk of this session is from Amanda Kowalazik. I apologize if I mispronounce that, which I'm sure I did. Amanda is at the Carnegie Mellon University, the University of Pittsburgh. And she is a graduate student in the joint PhD program of computational biology between those two institutions. And Amanda is going to tell us about pan mammalian analysis of molecular constraints underlying extended lifespan. Great. Yeah, thank you so much. You guys can see my screen all right. It looks great. OK, awesome. So yes, I'm Amanda Kowalczik. Thanks to you. Thank you for the valiant effort of pronunciation. And I'm sort of going to zoom way out in my talk. Our goal was to use sort of the powers of convergent evolution to try to find the genetic underpinnings of longevity across all all mammal species. So since I study evolution, I like to start these types of talks by acknowledging the incredible diversity of life on planet Earth. Perhaps it isn't surprising that we observe this incredible diversity because the Earth itself is super diverse. And each of these species is uniquely adapted to survive and thrive in their respective environments. What may, in fact, be more surprising is when we see less diversity than we expect or when species that are not closely related or unrelated develop similar phenotypes. And when that when that happens, we call it convergent evolution. And the classic example is depicted over here. Birds and bats independently developed the flight phenotype. We call that convergent evolution. And when we see this type of phenotypic convergence, we may also expect to see genetic convergence. And by looking for the concordance between those two sort of signals, we can try to link genes to their associated phenotypes. So my goal was to do that, of course, for longevity. And to do that, longevity must be convergent, which fortunately it is that I can demonstrate that for you right here. So this is showing the 61 mammal species included in the UCSC 100 way alignment, for which I also have longevity data. This is the phylogenetic tree representing their evolutionary relationships. And here are their maximum longevity values from the ant age, senescence and longevity database. So I can pull out the species that are very long lived. You can see that they are pretty distantly related. So they are convergently long lived. And the same is true for short lived species. So overall, this indicates that, yes, longevity is a convergently evolving phenotype. Another thing that we can pick out just based on the species that I have selected is that when we consider longevity from this cross species perspective, it's very heavily confounded with body size. Species that are large also tend to be long lived and species that are small also tend to be short lived. We were interested in both of these phenotypes. So how a species evolves to be large and long lived as well as how a species evolves to be small and short lived. But we were interested in the unconfounded phenotype as well. So the sort of more classic independently long lived phenotype to get at both of those sort of contrasting phenotypes. We had to invent brand new phenotypes or I should say, calculate brand new phenotypes. We did that using principle component analysis. I'm not going to go through the steps for the sake of time. But basically, principle component analysis works by fitting lines like you can picture it graphically like fitting lines to the data and then projecting points down onto the lines in the distance along these sort of fit lines are the new principle component values. So if we look at this as body size versus lifespan and log scale, they're very strongly linearly related. And I can pull out the first and second principle components. PC one represents the agreement between body size and lifespan. So the extreme values are things like whales and elephants that are being in long lived and shrew mouse, small rodents that are small and short lived. PC two is the sort of independently long lived phenotype. This is lifespan corrected for body size. And so the extreme values are things like that and the naked mole rat that are exceptionally long lived given their size. So now we have this phenotype information and we want to connect it with some genetic information. And to do that, we have this method that we've been developing in the lab called RER converge that can link any genomic regions to a convergently evolving phenotype. So if you're interested in these types of analyses I'm about to talk about it's available on GitHub. Please check it out. Lots of people like to use it for their phenotypes. It's very fun, pretty easy to use. But so the genomic information that this method uses is relative evolutionary rates. These are calculated from plain old vanilla evolutionary rates. These are calculated over a phylogeny where the branch lengths represent the amount of evolutionary change that has happened along that branch. So just like the number of substitutions that is predicted to have happened in the sequence. They're converted into relative evolutionary rates through a bunch of very fancy statistical corrections to make the behavior of these data nicer. But the main one that you guys might be interested in is this average rate normalization. This is what makes them relative evolutionary rates. So each branch length is normalized for the average rate of evolution on that branch genome wide. This corrects for the fact that some species evolve faster than others. Now point out that in this data set we have about 20,000 genes. So this is we have one tree for every single gene. So there are about 20,000 trees with relative evolutionary rates for those proteins. We also have trait information. Again, we have the PC1 and PC2 phenotypes to represent. The long-lived large-bodied phenotype and the independently long-lived phenotype. We know what those values are for the extant species and we can predict what they were for the non-living species for these ancestral species. And then our branch lengths are just the difference in values between the nodes. So the branch lengths then represent the phenotype change along that branch. Then we put that information together to look for correlations between evolutionary rates and the trait values. Point out this is just a little sketch sort of of this, but each of these dots represents one branch. So basically each dot is a species, whether they're a living species or a non-living species. If we see a significant correlation between our trait values and our evolutionary rates that indicates that that gene is associated with the phenotype. I'm gonna focus on mostly the negative correlations because we saw very little signal in the other direction, positive correlations. We believe that negative correlations are indicating genes that are more important for the phenotype. So more important to evolving, to be a large and long-lived species because they're evolving slower, because they're being protected from mutations because their functions are very important. I also wanna point out a really important distinction between these sort of positive correlations which would indicate faster evolution in the long-lived large body and independently long-lived species versus these negative correlations. Sort of looking at this for a more conceptual level. Sometime in evolutionary history, we had the ancestral mammal that was probably small and short-lived. And then there was some period of a selective pressure shift during which time this small short-lived mammal evolved to be large and long-lived along some of the mammalian lineages. And at some point that stabilized and now we have the phenotypes that we observe today. During the time that this trait change was occurring, there were genes, there must have been genes that were evolving faster, that were under positive selection in order to generate this phenotype change. This must have happened. But at some point when the phenotype stabilized, these genes were relaxed from this sort of positive selection and went back to neutral evolutionary rates. So as a result, this is sort of explaining why we aren't seeing a lot of these positive correlations because in evolutionary time, this period of positive selection of accelerated evolution was very brief. So at the end of the day, we're actually capturing very small rate shifts, sort of an average out over this time period. But on the other hand, when these genes were under positive selection, there was a whole shift in the evolutionary rates landscape in many, many genes, including some that were very important to enable this evolution, but were not driving it. And those genes would have been under increased purifying selection because they were being protected from a cumulative mutations. And then when the trait was established, they continued to be under purifying selection to enable the trait to continue to persist. So as a result, we're able to capture these rates. This is why we're able to see these negative correlations more strongly than the positive correlations. But this also is a really important distinction between my work and what a lot of other people do in that what I'm looking at is genes that enable species to be long-lived rather than genes that cause them to be long-lived. So everything that I'm going to be talking about is sort of in this camp where these are functions, genes, pathways that are required to allow these species to continue to be long-lived. So I had all of my gene results. I did pathway enrichment to do some sort of more functional analysis of them. And this is a representation of those pathway enrichment results for the PC1 phenotype, the long-lived large-bodied phenotype. Each of these dots is a pathway that's significantly enriched and the lines between them are just representing the number of genes they have in common to help us sort of sort through the redundancy and pathway annotations. For PC1, we see lots of immune pathways, cell death pathways, a very specific set of DNA repair pathways and cell cycle control pathways that are really important to enable species to grow to be large and long-lived. And what's interesting about this is that these pathways all work together to prevent cancer. So if we think about how cancer happens, it starts as DNA damage that causes unchecked cell cycle progression and or inadequate cell death. Those cancerous cells that evade the immune system and in some cases, inflammation even promotes tumor growth and then we get downstage tumor development processes. So we're seeing a significant signal for all of these early stages of cancer prevention, which is a potential resolution to something known as PETO's paradox, which is the observation that large species do not get cancer as often as they should given how large they are. The prediction is just because large species have more cells going from left to right, they should have a higher probability of getting cancer. This line in theory, what should happen, but what we actually observe is more equal cancer rates across species. So one possible explanation for this is that as species get larger, they have lower mutation rates. And my work is suggesting that there's a whole host of cancer control mechanisms that are actually at work helping to prevent these large species from getting cancer. We look at the other side of things. This is the independently long-lived phenotype. We see the same sort of large scale story with immune function and a more diverse set of DNA repair pathways. But if we look under the hood a little bit more, we can see that for these DNA repair pathways, this is showing their enrichment and orange is PC2. So if we see this sort of uptick to the right, that's indicating enrichment. We're seeing a lot stronger signal for these DNA repair pathways for PC2 and orange versus PC1, which is sort of flatter or has a weaker signal in comparison. And if we look at the immune pathways, don't read these. These are just a bunch of immune pathways. So don't worry about what their actual names are. The bars are representing whether or not there's a significant enrichment in that pathway. And basically the takeaway from this is that when there is a bar for PC2, there is not for PC1. And vice versa, with the exception of these couple NF Kappa B pathways, which is indicating although immune functions are important, it's a distinct set of immune pathways for these two phenotypes. So there were really distinct evolutionary trajectories leading to long life and large body size versus long lifespan independent of body size. So with that, we'll thank everyone who has been involved with this and I'm happy to take any questions. Thank you Amanda, that's super cool. We got a question from Alex Chen. Did you calculate DN over DS values for each of the genes across all four categories of animals? So in our branch length calculations, those are DN. So our branch lengths are the number of non-synonymous mutations with corrections. What we did do branch slate models for positive selection on genes that were, like on all four categories basically, as you're saying, we saw very little signal for positive selection in any of the genes that we thought that we may see that type of signal for. Alex also asked, what about correlations between evolution of one gene and another, e.g. inflammation ones and DNA repair ones, like OGG1 or TNF alpha? Yeah, yeah, so I haven't looked at those, but that's something that we work on in the lab as well. That's called evolutionary rate co-variation or ERC and there are definitely some genes that sort of evolve together, both because of actual physical interactions between the protein products as well as just because they're under similar evolutionary pressures. So we're aware of that and there's actually a bunch of corrections in the stuff that I do that I didn't go over called permulations to try to correct for the fact that this sort of inter-correlation of genes exists. So the functional categories that you found for example, homologous recombination, I mean obviously that pathway exists in the small short-lived animals and the large long-lived animals. So are you saying that the pathway became more efficient? Yeah, so our argument is that that pathway is being protected for mutations and in evolutionary time scales, if it's being protected for mutations more, that will lead to slower evolutionary rates. So in the smaller species, since it's not as important, it's not as big of an issue, if it's not sort of working as well, if it gets kind of a little mutation and it's not as big of a deal. So that's the type of signal that we're picking up. Can alpha-fold 2 allow us to model the geometry of all these different genes? I.e. structure of short-lived DNA repair genes versus long-lived DNA repair genes. I mean, I suppose if that's protein modeling software, to be honest, I have no idea what that is. I think it sounds like protein modeling, you know, the actual structure. Yeah, yeah, I mean, sure. I don't do protein modeling, but if you're a protein modeler, sure, go for it. But the homologous recombination thing is of a lot of interest to me, because that's actually one of the things we study. We found that in old yeast, the homologous recombination efficiency goes down because the proteins aren't made. And actually when we overexpressed the homologous recombination proteins in the old yeast, it could then live longer, which to my knowledge is kind of the first example of showing that improved repair in an organism can make the organism live long. So I'm really glad you found HR. No, that's really interesting. It sounds like that's something like regulatory, maybe. I mean, maybe like the regulatory functionalities of those proteins aren't working as well. I don't know. I'm working on that now on the regulatory sort of side of the story. So that's really... Well, we actually found a defect in protein synthesis in old yeast, to less proteins. Interesting. Okay, I'd like to thank the speakers for really clear, exciting talks.