Metabolomic sweet spot clock predicts mortality and age-related diseases in the CLSA

Okay, well, welcome everyone. My name is Jennifer Boyko. I'm the senior manager of scientific operations with the Canadian Longitudinal Study on Aging. Thank you for joining the CLSA webinar today that is entitled metabolomic sweet spot clock predicts mortality and age-related diseases in the Canadian Longitudinal Study on Aging. Before we begin, I would like to acknowledge that the CLSA National Coordinating Center and McMaster University are both located on the traditional territories of the Mississaugas and Hodena Shani Nations and within the lands protected by the dish with one spoon wampum agreement. The BC Cancer Center in Surrey is located on the unceded traditional and ancestral shared territories of the Katzie, Kwantlen, Kwikwetlem, and Semiahmoo and Tsawwassen Tsawwassen, sorry about that, First Nations. Simon Fraser University respectfully acknowledges the Musqueam, Squamish, Tsleil-Waututh, Katzie, Kwikwetlem, the Keightley, the Kwantlen, the Semiahmoo, and the Tsawwassen people and those whose unceded traditional territories our three campuses reside. As attendees of the webinar today, I encourage you to continue your learning following the webinar and to acknowledge the original inhabitants of the lands where we currently have the privilege to do research, live, and work, wherever that may be for you. Now, let's review a couple of housekeeping points. Everyone but the presenters will be muted throughout the webinar. If you need to change or test your audio during the webinar, you can click on audio settings that's on the bottom of the Zoom window. At the end of today's session, there will be a question and answer period. If you have a question for the presenter during the webinar, you can post it in the Q&A box, which is located in the bottom toolbar. All the questions will be addressed at the very end of the webinar, and um questions will be visible to all attendees. If you have any technical trouble concerning the webinar, please use the chat box to communicate with our webinar team. Uh a feedback Finally, a feedback survey will be launched at the end of the webinar, and we do invite you to complete it after exiting the Zoom session today. Uh the brief evaluation does provide us with important feedback that we can use to plan future webinars. Now for today's webinar, again entitled Metabolomic Sweet Spot Plot Predicts Mortality and Age-Related Diseases in the Canadian Longitudinal Study on Aging, presented by Olga Vishnyakova, who is a postdoctoral fellow at the BC Cancer Research Institute and the Department of Biomedical Physiology and Kinesiology at Simon Fraser University. Dr. Vishnyakova holds a Bachelor Science and Master of Science degrees in Applied Mathematics, as well as a Master of Science in Computing Science from Radboud University and a PhD in Computational Biology from Simon Fraser University. Uh her research focuses on healthy aging using multiomic approaches to study heterogeneity in aging and also the biological mechanisms underlying resilience. So, welcome to our presenter as well as all of you today, and I look forward to a great session. Over to you. >> Thank you, Paula, for the introduction and uh for having me here today. Um And also for your interest in our work. I'll walk you through the challenges in the field that might related this research. How we use untargeted metabolomic data from CLSA uh to overcome this challenges and introduce the sweet spot clock, a molecular predictor of aging to investigate how we age. We know that age is a universal process, yet each individual experiences differently. Although chronological age is a major risk factor for many age-related diseases, cancer, diabetes, cardiovascular disease, still it does not consistently predict disease susceptibility, treatment response, or variability in aging trajectories. The search in aging investigate this variability on multiple scales. It may look at macro level at the scale of a whole body, and here we may uh we may look at frailty, cognitive, or physical function. And we use instruments such as frailty index, osteological load, uh or comorbidity index. Next, we may focus at a specific system and investigate variation in cardiovascular health, neurological um function, or skeletal health. Molecular biology can look even deeper on a several resolution and look at uh cardiomyocyte senescence, microglial activation, or specific skeleton uh cell dysfunction. But, going even deeper into a single-cell resolution, we can look at difference in and health at the level of a specific components of a cell organelles and determine the reason of for mitochondrial dysfunction, energy plant of cells, or lysosomal impairment compromising um waste center of a cell. Finally, we may go go as deep as uh the level of the micro level of omics and try to find the root of these differences in each of patterns at the molecular level starting from a genetics or blueprint and further down the road uh through DNA modification, gene expression, uh translation into protein and their degradation. And that allow us to investigate the genetic predisposition for specific disease or identify disregulated pathway. So, my my token fiber search focuses on this bottom layer uh and on the high resolution scale. So, what is aging? It's often described as time-dependent decline in function that results in accumulation of cellular uh damage over time. And it might manifest as through, for instance, mitochondrial dysfunction or genetic instability and that contribute to cancer. But on the bright side, for some people, it's not the case. Some people experience an early onset of age-related diseases, while others remain remarkably healthy well into old age. And this heterogeneity of aging and actually the main question, why do some people stay biologically younger? That's exactly what drives this research. So, what is biological age? It might be described broadly as a level of age-dependent biological changes, such as molecular or cellular damage accumulation. But in aging research, we are dealing DNA way more simple uh calculation and it's treated as a numerical value, which corresponds to the chronological age at which an average individual in a reference population displays similar age-related biological changes. So, if we age faster than an average uh person in a population, uh we are saying that this person is a faster aging and vice versa. And if we are age slower than an average one, so we are talking about slower aging. And the question is why we even bother of measuring this biological age. There are multiple reasons. One is uh is really straightforward. We can quantify individual deviation from some normative aging patterns. How different is a specific individual from an average person in a population? But they also can identify factors including modifiable risk factors. Some of them are known and others still unknown associated with accelerated or decelerated aging. And other way we also can identify some life course events that drive the divergence and aging trajectories. Ideally, when this biomarkers of biological age, we can assess the effect of longevity interventions. Uh just because in contrast to some animal models, we cannot wait for 10, 15, 20 years in order to assess the quality of this intervention. So we do need some proxy for mortality in order to see how good we are in terms of uh specific uh clinical intervention. And the idea of measuring biological age on a molecular level has been gaining traction for over a decade. And the early attempt to build this biomarkers uh using uh telomere length uh until in 2013 Dr. Horvath introduced his well-known Pan Tissue Horvath's uh clock. And actually he coined the term biological clock. And since then uh many other um biological clocks using different omics data appeared uh using data such as um gene expression uh or proteomic or uh metabolomic. Basically, how this models work. So, we are using some type of data in the data, in a protein abundance or metabolomic abundance or gene expression data. Feed them into some complex or less complex model starting from multinomial regression uh tree-based model or then some neural networks in order to predict uh biological age. And even though the many models exist already, we still do have some challenges with the quality of this models. First of all, as all data-driven models suffer, uh we have a limited specificity. And we cannot adapt kind of ask answer question, do these molecular biomarkers reflect true drivers of aging? Or they also uh have some consequences of aging in it or some unrelated uh signals just because they correlate with some true drivers. We also some models, especially tree-based model or neural network-based model, uh have some problems with interpretability just because we cannot clearly answer what individual feature represents. The other problem and then the other challenge is the choice of a train targets. We do have a pure generation clock um trained on on chronological age. We do have a second generation called uh trained on mortality. So, the question is, what should serve as a primary training target to predict not just mortality, but a broad range of age-related outcome? And we do have two other challenges in the field such as non-linearity, especially when the multinomial regression models, how can we account for complex relationship to capture more realistic pattern uh such as non-linear patterns while preserving interpretability of the model? And finally, we know that aging is multi-dimensional. So, the question is, can a single omics layer, uh let's say epigenetics, capture it all? Or do we need many different uh omics in order to make a prediction? So, in response to these challenges, uh we started to work on omics-based aging biomarkers with an ambitious uh aim to make it interpretable and focus on relevant features, account for non-linearity, choose a project training in such a way that we can predict not just mortality, but multiple age-related diseases. We also want to discover some disease-related pathways, and ideally use for these predictions. So, to build this biomarker, uh we're we use data from Canadian Longitudinal Study on Aging, which allow which provides really rich uh participant profiling with uh genetic data, metabolomic data, DNA methylation data, and also uh rich phenotypic phenotypes. We used uh obviously a comprehensive cohort with almost data available at the baseline. So, first question that we asked was can we identify measures relevant to health and healthy aging? And if we can, can we determine the optimal ranges so we can further account for non-linearity? And when we talk about health and healthy aging, we first should define what is it health. And there is no simple answer to that question. Otherwise, what we are doing here? So, we used five different instruments to assess health that try to capture different aspects of health. First, we calculated the deficit accumulated frailty index, which is uh we used 51 health deficit to represent overall health. After that, we focused on multi-m- morbidities, and we uh split it into two, and I'll explain uh further why we did it. So, we first calculated as a a normalized number of five major diseases, uh cancer, cardiovascular diseases, major pulmonary diseases, dementia, diabetes, and complemented with 28 self-reported clinical diagnostic condition as the instruments uh number three. Also, we calculated cognitive and physical function scores to capture uh again different aspects of health. We normalized to to unit interval. So, we have a scores from zero to one in order to make possible to compare between this. And if you're interested, so here is our correlation between our instruments. So, once we have this, we know that some uh person is more or less healthy than other one. Uh we started to to investigate how exactly we can identify this health-related phenotypes. And we used our observation from the SuperAger study. This study was launched in our lab more than 20 years ago. SuperAger phenotype was defined as being 85 years and older and never been diagnosed with a five major diseases uh that comprise our instrument number two: cancer, cardiovascular disease, diabetes, dementia, and uh major pulmonary disease. That was uh this uh almost 15 or even more years of recruitment resulted in a 700 SuperAgers each 85 to 110 recruited in BC. And these are physically and cognitive uh high-functioning group. So, here are just the sum um means for uh common test for elderly people such as time up and go or mini mental state exam. What we observed in this remarkable group that our initial hypothesis was that they should have a telomere length this uh protective cap uh on a chromosome longer than an average uh population. However, no difference in mean were observed between two groups. But what we did see in and the reduced variation in telomere length compared to midlife controls. Also, similar patterns were observed for other um hematology data such red blood cell count, hemoglobin, or uh red cell distribution width. No mean difference, but difference in variance. So, we used this idea uh in order to generate what a simple parameter's one hypothesis that measures that are important for health tend to be closer to some optimal values or sweet spots among the healthiest individuals. What is important here is that by measuring the distance from the sweet spot, we can quantify the level of dysregulation and inform this regarded processes. And this deviation from sweet spots also can serve as non-linear transformation uh when we are talking about some complex model down the road. This idea of a was formalized into statistical two-step statistical approach where we first screening uh phenotypes of interest where we are uh variance uh heterogeneity across health extremes meaning that we are comparing variance between most healthy and least healthy groups. And after that, we estimate optimal phenotypic values uh where deviation up from this values linked to poor health. And we are doing this using segmented regression with a three knot points. Applying this framework uh to phenotypic data and available in CLSA, we were able to identify more than a hundred of phenotypes with significantly higher variance among the less healthy. What we started from phenotypes but not from uh complex more complex middle moment data because we first of all we wanted to uh validate uh this idea and understand whether statistic can help us to identify related features where we do not introduce any prior biological knowledge. So, our results suggest that indeed we can identify health-related phenotypes that make sense just because we know that they make sense. When the mhm hemoglobin A1C an indicator of an average blood sugar uh that showed the strongest evidence of difference in variance across both sexes and all age groups. Also, majority of these phenotypes were appeared to be uh, related to health in both sexes. Uh, some of them were unique for either females or males. Another key finding that uh, what I derived from this from the first from the first step of this uh, study in that uh, that 40% of our of the invest of the phenotypes of interest exhibit the sweet spots. And here just one example for pre-adoption how we can uh, can determine that a specific phenotype has a sweet spot. And another uh, important observation that uh, confidence intervals were uh, identified sweet spots appeared to be a narrower than a standard clinical reference interval. Here is just one example uh, you have the interval confidence interval for uh, for thyroxine identified with the thyroid index as a uh, health instrument compared to a reference for the reference intervals. What we did next, we cross-checked identified optimal ranges uh, with the uh, clinical reference intervals used in a life lab or Sinai Memorial Hospital lab. And for majority of blood chemistry biomarkers, uh, we did have a match with an exception to cholesterol and HDL. One of the reason for that is that we probably should be more Now, we do know that there are many South Asian participants on study. Uh probably that stopped the analysis and identification as we did not see a clear picture of the true distribution of this uh phenotypes. So, with that, we're moving forward and uh we ask ourselves, how does the metabolome reflect variation in aging across individuals? All right, first couple of words, why metabolomics is so important and what are they? Metabolome is a collection of all metabolites that which are small molecules in body and metabolome is the end product of metabolism. And it is the end result of all environmental and biological processes. From starting from protein degradation, medication intake, our nutrition patterns, how we exercise, whether we do have any uh exposure from some chemicals or even some exposure, whether we exposed to some level of stress. What are our sleeping patterns? Do you sleep well? All of these impact our metabolome. What is also important that about 30% of identified genetic disorders involve defects in small molecular metabolism. And this metabolite, these small molecules, act as cofactors and signal molecules for cell events of proteins. Given this important importance, uh we did start from uh metabolomics data available in CLSA to calculate this uh metabolomic biomarker. We were at two valid CLSA, so we do have uh a data for almost 10,000 participants at the baseline. We specifically focused at endogenous endogenous metabolites produced by our body that might be a kind of categorized uh metabolites might be categorized uh based on the super pathways. The majority of being liquids, uh lipids, and some uh smaller groups such as amino acids, cofactors, and vitamins, uh nucleotides, and so on. Um we restricted our analysis to metabolites uh with less than 20% missing rate. And that resulted in this nice number 88 uh metabolites for our analysis. So, we applied uh the statistical framework I was talking uh previously to this complex metabolomic data, and we were able to identify 178 or 20% that 20% of metabolites showed significantly uh higher variance in healthy individuals. Uh so, we looked at the two biomarkers, and they completely make sense. For instance, uh N6,N6,N6-tri uh methyllysine as a the the the derivative of amino acid lysine and it's a component of histone protein. Or the the second hit was uh glucose. Obviously, we all know how glucose is like a fuel important to your body. And this health-related metabolites, all 178 of them, we use as a predictor of of metabolomic age. Next, we wanted to account for non-linearity, so we calculated sweet spots. So, we determined optimal level. We're able to do that for 74 metabolites out of these 178. What uh The remarkable result here that more than a half of them 95% confidence intervals for identified sweet spots did not include population mean. So, we assume that that might be partly because the all the Canadian population might and it might be a quite a strong claim not healthy enough in order to maintain an optimal um value of a specific optimal abundance for a specific metabolite. What we did with this many many uh small molecules and their abundance uh again, they were used as an input to a to a model for those metabolites where that exhibit the sweet spots, we performed the transformation. Specifically, we calculated the Euclidean distance uh between a sweet spot and uh measured metabolomic abundance in a specific person. And here the transformed or untransformed whether where we do not see any non-linear patterns where after that fit to a plastic net which is a double penalized regression multi multinomial regression model in order to predict health index. Why health here? Again, we didn't want to use a materiality as an outcome cuz we did did want to see uh changes not directly linked to the later in life changes. We want our biomarker to be predicted to multiple uh conditions not just mortality. So, we choose some um composed uh representation of health status. And here what were our results are So, we basically compared this biomarkers against uh some uh baselines. So, we were interested whether this transformation makes sense. We also wanted to know whether uh choice of target is right and we obviously compared with a frailty index itself. Uh so uh And obviously it's all this uh The results are on a train so we split in a train and test set made all the uh training on a train set and assessed the quality of the model on a test set. It's It's pretty small so it's 1,200 of participants. So, that's why probably you won't see any uh um extremely uh strong uh association with mortality, but nevertheless, uh, the model and how it was built definitely makes sense. What we uh, did after that, we calculated this mm, we assessed biological age, and here is a regression uh, so we regressed on the age. Above the regression line uh, line, we're talking about partici- participants healthy partici- age faster than average, and below uh, are slower agers. What we also observed is that uh, metabolic age deviation, there's differences uh, with higher amount females than males. Next, we are measured association with ultimate challenge and age-related disease between this molecular metabo- metabolomic molecular biomarker, and and uh, uh, diabetes, cardiovascular uh, so represented by strokes, uh, kidney failure, Alzheimer, cancer, and pulmonary diseases. We did observe uh, strong association with mortality even on our smaller mm, uh, test set. Uh, mm, but not for Alzheimer's and cancer. And again, there is a reason for that. Uh, dementia was an exclusion criteria uh, and the time of recruitment, and um, the metabolomic data were uh, available only at the baseline. So, we did not expect to have uh, to see much for Alzheimer's. Same for cancer, too heterogeneous, so we had too uh, many uh, too too We just had a few for each type in a on a in a death set so that's pretty predictable. So in order to understand what that is means so what does this mean to have a hazard ratio of 1.08 for all cause mortality so what we did just to get an intuition behind that was put participants into a four parts for equal parts and compare the survival probability for each of them for slower stagers and for faster stagers and now it's kind of now we feel what that is mean and after that we went a little further and we looked at two extremes uh, fastest and slower agers and here the difference in a survival uh, even more pronounced. So obviously we measured association with the health lifestyle and social economic factors and we observed expected association in expected direction obviously on nutrition smoking alcohol consumption all of this contributes to a uh, faster aging while uh, income and level of education reduce uh, uh, how about biological age. Uh, so what we also wanted to know whether our metabolites that were we used as a predictors as it's a data driven model so it's um, made some uh, feature selection uh, during the training process. So, we will wonder whether this uh, features we mean after selection make sense, biological sense. So, what we did, uh, we used 50 top predictive metabolites selected by the model and looked at association between them and 14 known disease biomarker chronic condition and diagnostic measurements or all of what we can found. Uh, find at that point and all of these metabolites showed strong association with at least five distinct phenotypes after multiple testing direction. Here is just uh, piece of the results. Uh, not suggesting to look over, but just to explain that indeed, each of this metabolite makes total sense. Um, uh, it's associated with uh, multiple, uh, diagnostic condition. We also uh, performed the validation of this result in super seniors. And for that, we uh, performed the meta- uh, metabolomic profiling for super seniors and age-matched control exactly the same way as it was done for CLSA data. Uh, the metabolome uh, had the same uh, control samples just to to be as close for validation as possible. And that resulted in a um, positive results, obviously. Uh, so, we did observe that uh, um, biomarker trained on CLSA data and applied to uh, uh, Sebastian data set uh was able to catch a difference between uh healthy and uh age-matched controls, the extremely healthy individuals in terms of our phenotype and age-matched controls and was associated with significantly associated with mortality. Uh what we also wanted to look uh is where our metabolomic biomarker stands compared to uh well-known epigenetic, highly promoted epigenetic biomarkers. Uh Hannum and Horvath clocks are available at the baseline and um um with the part of comprehensive uh cohort data set, we also calculated state-of-the-art uh GrimAge boosted version and PhenoAge using uh BioLearn Python library generally provided by uh Biomarker Aging Consortium. And what we observed and that our Swiss Port uh clock based on uh metabolomic age deviation outperformed uh epigenetic biomarkers and was stronger stronger associated with mortality. I obviously acknowledge that uh GrimAge, PhenoAge, and Hannum and Horvath clock were trained on a a different data set uh that I believe here years best participants, while Swiss Port clock was trained on even on a separate uh training set, but still on a CLAC, so we expected to uh or the this metabolomic biomarker to perform better. Yes, there is definitely some bias, but nevertheless, we might be better, which is good. What we wanted to you know, uh and uh I already expressed that uh we want to know um that whether multiple layers capture uh unique bi- whether each layer capture unique biology or not. So, for that, we actually uh built a uh multinomial model and used our biomarker and pre-image as a predictor. And that actually improved uh uh the performance of the model measured by a concordance index. So, indeed, different layers capture different biology. So, we obviously ask ourselves whether first of all, whether we can do better than pre-image in terms of uh developing biomarkers. And whether these two uh epigenetic-based and metabolic-based share something common. Uh couple of words about uh epigenetics, if we still have time. So, it's a heritable changes in gene expression without altering DNA sequence. And if we In simple words, how we can think about epigenetic, we do have a multiple cell types. We do have a muscle cells. We do have a uh immune cells. All of them cells, but they express different uh it's but but each of them have a different gene expression. And that is with the help of this methylation pattern that either uh silence uh some gene expression or allow them to be expressed. Um, and the promise that epigenetic achieved by environmental development and aging, and that also might be impacted by our habits such as smoking. So, the question is whether we can use this DNA methylation pattern in order to build this biomarkers. And again, we looked at two different level of changes such as different differentially methylated uh uh position upon the genome, and differentially variable position upon the genome, and uh aggregated into regions, and that resulted in uh epigenetic biomarkers. What we also observe in that there is a almost a little overlap within these two. And using this information, we were able to build a with a quite a fancy model mm epigenetic biomarkers where we first predicted uh health status, and after that, we weighted this prediction using mortality data. Uh It makes sense. So, we we showed that uh this uh model makes sense where you compete against uh uh benchmark models. What is uh relevant to today's talk in that uh there is a just a moderate correlation with metabolic and epigenetic biomarkers. Each of them have a So, genetic component. Uh each of them might be partly explained by immune uh lifestyle, socioeconomic factors. However, each of them still have this large portion of unknown which is still need to be investigated. So, the main takeaway in that we proposed methods and that uh allowed to identify measures related to health and health the healthy aging and by measuring deviation uh from them, we can quantify the level of this correlation. Uh we also showed that our omics biomarkers were predictive of mortality and onset of age-related diseases. Uh talking about utility uh and the major clinical implication is that using this or similar models, we can create individualized health report cards that pinpoint this relation pathways and able to predict future health risk complementing uh genetic risk score. Now, we also uh with help of these biomarkers that might help guide lifestyle and uh prevention efforts. And going forward, we can investigate how lifestyle or pharmacological intervention can modulate biological age deviation and integrated multiple omics and a single uh by may maybe assembling single all omics prediction, which is the easiest way we can improve uh uh the prediction ability of such biomarkers. And that would wrap up my presentation with this nice comparison. I love to think about healthy aging as a magic cocktail and unfortunately, we just don't know its recipe. So, I want to believe that our study uh made us tiny bit closer to the secret of this recipe. We just need to I don't know. Dig deeper. And with that, I want to acknowledge uh both of my labs, my main and supervisors, lab members, collaborators, and obviously our participants. Without them, uh this research would not be possible. And thank you for for listening. >> Okay. Well, thank you very much, Dr. Vishnyakova. Um we will go on to some questions now. Um in the Q&A. Some of the I know the first question I pre-read it is is long, so I'm going to read it out loud and maybe you can also read along and then reply to Walter. Uh really interesting research. Thanks for sharing. In the association of metabolomic age deviation with with age-related disease and all-cause mortality on slide 34, the odds of the individual disease outcomes, such as diabetes, stroke, etc., were relatively similar. Um would I interpret this to mean that Oh. Sorry. Would I interpret this to mean that metabolomic age deviation is a precursor of disease and death, but the specificity regarding risks for specific conditions is still unknown? >> The I'm just double-checking that. It's still And it's not exactly the same, but similar. Although, well, partly because we incorporated uh metabolites related to each of these diseases. Just because we trained on healthy index, where and healthy index already include uh markers, uh not markers actually, but uh a health deficit, such as uh whether person was diagnosed with diabetes or cardiovascular at all and it was treated equally in quality index. So, we included metabolites related to each of these diseases as a predictor of metabolomic age. Uh Yes, we're In terms of answer to this question, it's not specific to a single disease. But we're still able to catch this regulation leading to each of these diseases. I hope that makes sense. >> And the next question, uh fascinating work. If I interpreted interpreted correctly, you observed a sex difference in the distribution of the sweet spot clock. Can you comment on what may be responsible for this? >> Honestly, so uh indeed we uh performed all this analysis stratified by sex. So, we acknowledge that adjusting uh for sex uh would not result in a proper estimation. Uh The biology work differently. And we expected to have a hormone level for metabolites that are precursors to uh sex hormones to have different uh optimal levels between sexes. So, that's totally uh makes sense. Yeah, but that's an answer. >> Um Okay, and uh the if there's any more questions, just a reminder if anybody does have any more questions to type it into the Q&A box. Uh but for now we have this last one. Could you please recap the calculation of the sweet spot. >> Sure. So, it's a two-step statistical framework. So, first we looked at difference in variance using Brown-Forsythe test for heteroscedasticity. We looked at two extremes and uh statistically tested the difference in variance. And after that, the uh metabolites that passed multiple test correction after the first step, on them we tried to uh to apply the segmented regression model with a single knot point and looked whether we observed any change of trend. And if we did, we looked at the direction of changes if uh direction was consistent with our expectation. We treated this knot point as a sweet spot. >> Okay, well, I think we will slowly start to wrap up. Then, I'm sure if anybody has any other questions, they can reach out to you directly. Uh but, for now I'd just like to thank you very much and I also would like to do an extra special thanks because I know you jumped in at the last minute to do this webinar. Um so, much appreciated. Uh but, I'd now like to remind everyone that the next deadline for data access applications, if you're interested in doing your own research, is July 8th, 2026. Please visit the data access section of our website to review what data is available uh as well as additional details about the application process. Also, a reminder for all CLSA data users to keep us updated on any changes that may affect your data access agreement. Um have you moved institutions or has your research team changed or has a trainee graduated or moved to a new position? Um you can always reach us out to us at access clsa with any questions or updates. I'd also like to remind everyone to complete your anonymous survey upon exiting the session today. The CLSA has introduced an updated data access fee structure effective October 2nd, 2025. The first adjustment since we actually launched the platform more than a decade ago. This update supports a sustainable cost recovery model that will help us maintain exceptional quality and stewardship of CLSA data for years to come. And if you'd like more information about this, please visit our data access webpage on our website or you can click on the link that's provided in the chat. And lastly about our upcoming webinar. The details of our next webinar are still being finalized, but you can visit our website under webinars to stay up to date on the next webinar. If you or a colleague are interested in presenting a CLSA web webinar, please reach out to our to us on our webinar team. We would be glad to hear from you. And the recording from today's webinar and the slides will be available in the coming days on our website. Finally, you're invited to fill out the exit survey that will appear right at the end of this webinar. And thank you all for your attendance today. And then again, thank you to Dr. Vishnu Kolla. Bye everyone.