Hi everyone, my name is Camille Poe. I'm from Mass General Hospital in Boston and it's my pleasure to introduce Dr. Anna Glowen today. Dr. Glowen is going to provide an overview of our current understanding of the genetic landscape for type 2 diabetes and outline progress in translating genetic association signals into clinical insights in diabetes pathogenesis. Dr. Glowen is a professor of pediatrics and by courtesy genetics at Stanford University and she is also the recent recipient of an Outstanding Scientific Achievement Award from the American Diabetes Association just last week. Thank you Dr. Glowen. Thank you so much for that invitation and good morning to everyone and I feel like I want you to all move down the front so that I I'm speaking to people but hopefully there are some people online as well. So I'm excited to be here at an in-person meeting and to be taking part in this symposium which I thank the organizers for inviting me to be part of. So what do we want to know when we design a drug for diabetes? Well we want to know if we perturb a protein will it work and will it be safe? So how can human genetics help us identify safe and effective drugs? Well if we have a DNA variant that alters the level of a protein or the activity of the protein we can use that information to understand whether or not it's a gain or a loss of function that is responsible for that effect and that's very important for pharma because they can use that information to decide whether or not they need to develop an agonist or an antagonist. Now if we study the effect of that change in activity or expression in a population we can look to see whether or not it's associated with a particular phenotype or trait and if it is and it alters risk of developing diabetes in humans that gives us really good evidence that perturbing that protein is likely to be effective at altering diabetes in humans. What's really valuable though with human studies is that we can study the effect of perturbing a particular protein over a period of time by looking at the consequence of perturbing that gene and the DNA level in a population and this gives us really important information about potential adverse effects of that particular protein as a therapeutic target. Now there's very strong evidence from human genetics that this is likely to be a very successful way of identifying effective and safe therapeutic targets and an example that's very close to my heart is work that I was involved in as a postdoc when I was at the University of Exeter working with Andrew Hattersley. During my time in Exeter I identified that heterozygous activating mutations in a key component of the machinery that couples glucose metabolism to insulin secretion were a major cause of neonatal diabetes. Now this was really exciting not just because we were able to understand what caused the beta cells to fail in response to glucose generated from metabolism but also because we could build on the work of others to show that patients with this variety of diabetes could be transferred from insulin injections onto oral medication. So this was an example of precision medicine where identifying the underlying genetic etiology allowed us to tailor treatment for individuals with this particular variety of diabetes but perhaps more importantly for type 2 diabetes it was also evidence that human genetics could point to safe and effective targets for the development of treatments for type 2 diabetes because after all the sulfonylurea class of drugs had already been used to close this KTP channel and elicit insulin secretion in type 2 diabetes. So my work is really focused on using human genetic discoveries whether these are rare mutations that are causal for monogenic varieties of diabetes or more recently common genetic variation that's increased associated with increasing your risk of developing type 2 diabetes and using these DNA variants as tools to uncover cellular and molecular mechanisms that underlie pancreatic beta cell dysfunction in diabetes. Ultimately I want to use this information though for clinical translation whether it be through identification of novel targets for therapeutic development or through patient stratification. So being able to identify groups of individuals based on their genetic makeup that may respond differently to a particular class of medication. So over the last 15 or so years human genetics has been in a very exciting phase through the use of genome-wide association studies and for type 2 diabetes we now have over 350 regions of our genome that we know robustly influence your risk of developing type 2 diabetes. But a few things have become apparent from this information. First of all we've got relatively few variants that sit in the coding regions of our genome. Most variation that's associated with type 2 diabetes risk is thought to have a regulatory role where it's mapping to non-coding regions of the genome. If we look at the most recent publication out just a few months ago we probably know now about 50% of the genetic risk for type 2 diabetes. For those signals are about a hundred and seventeen candidate genes that we think are influencing risk of type 2 diabetes. Many tissues are involved but from the genetics we certainly know that human pancreatic islet is a major player with a large proportion of the signals that we've identified thus far working through this tissue. So why is it so difficult for us to move from those 350 signals to the underlying biology? Well this comes down to the fact that the signals are not in the protein coding space they're in this intervening intragenic sequence and this gives us enormous uncertainty over the proteins that we should be studying in the laboratory. So if you think about it if we've got a non-coding variant we have to establish whether it's influencing expression and function of gene A, gene B or perhaps gene C. And because gene regulation is context-specific this means we have to ask these questions in the context of which cell type in the body this is occurring, when is this happening, is it during development or adult life and is a stimulus or perturbation necessary for us to see this effect on gene expression. And this complexity has given an enormous bottleneck to the field in our ability to really make the most of those genetic discoveries that have come about through genome-wide association studies. But my lab has now for some time been really interested in finding a shortcut, some way that we can leverage human genetic data without knowing necessarily how to map those regulatory variants to protein coding genes. And the way that we do this is by asking are there coding variants in our sequence that are independently associated with type 2 diabetes risk. Because if we have a coding variant that's a shortcut to taking us to the protein that we want to study. I was part of a collaboration that was headed by David Altshuler, Mike Benke and Mark McCarthy several years back now called the type 2 diabetes consortium. And the type 2 diabetes genes consortium performed exome sequencing on around 13,000 individuals about half of whom had type 2 diabetes. And these individuals were drawn from five different ancestries. Now the results of this study showed us that there were very few instances where there were coding variants that were associated with type 2 diabetes. But one of the exciting exceptions were two coding alleles that we identified in a gene called PAM that we were able to show influence your risk of developing type 2 diabetes. And through the work of Karen Mulkey and others we were able to show that these variants were associated with altered islet cell function. So what does this gene PAM do? Well the gene encodes for an enzyme called peptoglycine alpha-amidatine monooxygenase. And it's a really fascinating enzyme because it's the only enzyme that we have that is capable of amidating glycine extended peptides to increase their biological potency. PAM is expressed very widely in the body and you'll find it in neuroendocrine secretory granules. Now those two coding variants that we identified that alter your risk of developing diabetes actually are located in one of the catalytic domains of the protein which gives us a really good indication that they might be functional and altering activity of this critical enzyme for altering activity of neuroendocrine peptides. So we wanted to check if PAM was expressed in a human pancreatic islet. So we were able to establish that you can find PAM in both the alpha cells and the beta cells of the pancreas. And critically for our downstream work it's also expressed in the experimental model that we favor in our laboratory which is the human endo C beta H1 cell line. So we set about to ask the question those variants that we've identified in the PAM gene are they associated with a loss or a gain of PAM function? So do you need more or less of PAM to increase your risk of developing diabetes? So Anne Raimondo, a postdoc in the lab at the time, developed an assay so that she could measure PAM activity. And if you look at this graph you can see amidating activity of the PAM protein for the wild type variety of the enzyme in orange and then for our mutant PAM in green. And as you can see the mutation in the PAM gene led to reduced amidating activity. The second variant that we identified we weren't able to do this assay because the variant resulted in an unstable protein as you can see in the immunofluorescent on the right hand side of my slide. So both of these variants that are associated with altering your risk for diabetes result in a loss of function but they do so by two different mechanisms either by altering enzyme activity or by altering protein expression. So having established that it's a loss of PAM that increases your risk of developing diabetes the next logical question to ask is what does this PAM enzyme do in a human pancreatic beta cell? So to address this question Cern Thompson, a PhD student in my lab at the time, set about to use siRNAs to knock down the PAM gene in a human pancreatic beta cell and to see what happened to insulin secretion and insulin content. Cern was able to show that when you lose the PAM gene from a human pancreatic beta cell you see reduced insulin secretion and reduced insulin content. We looked at this at the level of the insulin secretory granule and were able to also show that there were fewer granules and fewer insulin content in the beta cells. Benoit Hastoy, an electrophysiologist in the team, was able to take this one step further and to demonstrate using patch clamping that there was delayed insulin exocytosis. So loss of PAM seemed to inhibit the ability of the insulin secretory granules to traffic from the cytoplasm to the plasma membrane. So why is an enzyme such as PAM that alters activity of neuropeptides leading to altered insulin secretion? Insulin itself isn't amidated, so what are the intermediate steps that result in this phenotype? Looking at the proteins that are expressed in a human beta cell, we noted that chromogranin A, which is a really important protein for packaging insulin in the secretory granules, is known to be amidated. So Shahanas and Gupta in the lab set about to look and see what happened when you knock down the PAM gene in terms of chromogranin A amidation. She was able to show that if you lose the PAM gene you get an accumulation of the non-amidated variety of chromogranin A. So what happens if you lose chromogranin A from the beta cell? Well CERN repeated his experiments using siRNAs to look at see what happened with loss of chromogranin A in the pancreatic beta cell and he was able to show that it mirrored the effects of PAM loss. So this tells us that loss of chromogranin A through defective amidation is part of the mechanism for how we affect insulin secretion and content. That was all in a human pancreatic beta cell. What about if we were to look at primary islets from people who carry those variants in the PAM gene? Through an existing collaboration with Patrick McDonald at the University of Alberta in Edmonton we've been systematically genotyping islets which he's been very carefully phenotyping in his lab and this enabled us to look at the relationship between donor genotype and insulin secretion in a primary islet. We were able to demonstrate that we could mirror those effects on insulin secretion and content. So if you have a type 2 diabetes risk allele in the PAM gene you have reduced insulin secretion and reduced insulin content. So if we were to put this information into a schematic to explain our understanding of what PAM does and how it alters pancreatic beta cell function we have direct evidence now that if you lose the PAM gene you have reduced granular insulin content and insulin secretion and altered kinetics of insulin release. But you'll remember at the start of my talk I told you that PAM is very widely expressed and it plays a role in amidating lots of different neuropeptides. So we wondered whether there were other reasons that we might have altered islet cell function perhaps through defective amidation of neuropeptides that are known to influence insulin secretion such as GLP-1 which of course we know is amidated. And could this alter the incretin response in carriers of this variant? So our next phase of experiments has really been about trying to understand whether or not there are indirect methods that lead to altered insulin secretion in carriers of these type 2 diabetes risk alleles. So the question that we've been asking in our physiology lab is do PAM type 2 diabetes risk allele carriers have a reduced incretin response due to reduced amidation of GLP-1? So Mahesh Ampathasayam who is a clinical fellow working in our lab set out to perform a recruit by genotype study. We were able to go to the Oxford Biobank and recruit individuals to come in for clinical investigation based on whether or not they had a type 2 diabetes risk allele in the PAM gene. The first thing we wanted to do was to prove that we could see a difference between the two groups of individuals. So Mahesh developed an assay to measure serum amidating activity and he was able to show that if you have one of these type 2 diabetes risk alleles you have about a 50% reduction in amidating activity in your serum. So this was really important because it told us that our study design was effective with this group of individuals. We were somewhat surprised then when we performed isoglycemic clamps that we saw no impact on the incretin response between non-carriers in blue and carriers in red. However in collaboration with Jens Holst in Copenhagen we measured circulating GLP-1 levels and we were surprised to see that carriers of the type 2 diabetes risk allele had elevated GLP-1 levels. But remember this doesn't result in an effect on the incretin response. So if we normalize the incretin response for the GLP-1 levels we can see actually if anything there is a reduced incretin response in carriers of the PAM allele. So we believe that these carriers are actually GLP-1 resistant. So given that carrying a variant in this PAM gene makes you GLP-1 resistant, does carrying an allele actually matter for how your doctor should manage your diabetes? Is this an opportunity for stratified medicine? So working in collaboration with colleagues at Dundee, Ewan Pearson and Adam Dord and Angus Jones at the University of Exeter, we were able to look in three trials that had looked at response to a GLP-1 receptor agonist. Using HbA1c reduction at six months as our marker of response to treatment we were able to show that if you carry a variant in the PAM gene you have a worse response to a GLP-1 receptor agonist compared to individuals who don't carry a variant in this gene. If you look at the PREBA study in particularly, depending on whether you have the rare 1% or more common 5% allele in the PAM gene, we found that carriers of these variants failed to reach their HbA1c target of less than 7% at six months, between 0% and 10% of the time, compared with 30% of individuals who didn't carry a variant in the PAM gene. So clearly this is clinically significant, particularly in countries such as the UK where the NICE guidelines instill that you can only stay on a GLP-1 receptor agonist if you meet your HbA1c reduction. So if we bring this new information into our schematic of how these variants influence a pancreatic beta cell function, we now know in addition to those direct effects on the beta cell where we have effects on content and secretion, we also have elevated GLP-1 levels, a reduced incretin response and GLP-1 resistance, which we believe alters your response to a GLP-1 receptor agonist. But where does that GLP-1 resistance come from? So we wondered first of all if it was at the level of the pancreatic beta cell. So we were able to look in our gene expression data from over 400 cadaveric donors and to look at whether or not those PAM alleles altered expression of the receptor. And we found no differences in pancreatic islets. So it's clearly happening somewhere else. In a collaboration with Markus Stoffel at the University of Zurich and Elisa Eraldi, a really talented postdoc working in his team, we were able to bring together our human data with some elegant mouse studies that she'd been performing. In mice that lack the PAM gene, she'd been able to show that they had an increased gastric emptying rate and they had reduced expression of the GLP-1 receptor in the pylorus cells of the gut. When she measured gastrin-releasing peptide, they also had elevated levels. So we believe that that GLP-1 resistance is not happening at the level of the beta cell, but rather at the level of the gut. So let me summarize the mechanistic insights that we've learned from just this one of 300 odd signals that we've identified for type 2 diabetes. Well first of all we know that loss of PAM in a pancreatic beta cell affects insulin content and secretion and alters the kinetics of insulin release. We also know that there are indirect effects of PAM loss, leading to elevated GLP-1 levels, a reduced incretin response and GLP-1 resistance, which alters your ability to respond to GLP-1 receptor agonists. We believe that this is due to increased gastric emptying, which leads to more rapid nutrient delivery to the duodenum and leads to a chronic elevation of GLP-1 levels and GLP-1 resistance. So I told you at the start that PAM is very widely expressed. So what else do these PAM alleles associate with? Well we know that they're associated with type 2 diabetes and type 2 diabetes adjusted for BMI, but we've also got a number of other anthropometric diabetes complications, ECG, hematological, immunological and hypothyroidism and sleep phenotypes that are associated with these same variants. Okay so let me give you some take-home messages. I hope I've given you insights this morning that most of the type 2 diabetes GWAS variants that we've identified to date are in non-coding sequence and are presumed to affect gene regulation. Coding variants provide us with natural perturbation assays which help us to understand gene function in diabetes relevant tissues. If we can continue to couple human genetics with cellular and physiological phenotyping, it will offer us opportunities to understand disease mechanisms. And hopefully through this one example out of the 300-odd that I could have picked, it shows you how genomics can start to aid patient stratification and help us with efforts for precision medicine. It leaves me to thank my team back in Stanford, the wonderful trainees and collaborators who've been instrumental to delivering the project that I have shared with you this morning, and my colleagues in the Accelerated Medicines Partnership, who I'm working with across the pharma-academic divide. Thank you for your attention. Thank you, Dr. Sloan, for a marvelous talk. We are able to take some questions now. This is also a great time to stand up if anybody wants a break, get some exercise, blood flowing. So questions? There you go. Hi. That was a wonderful talk. Thank you very much. Joanne Davis from University of Wisconsin-Madison. I was just really intrigued by the PAM-GLP-1 connection, and I was just curious, I think you said, well, I mean, number one, was the GLP-1 actually not amidated in the PAM knockouts? You mentioned that it seemed to function normally, but I guess I was wondering if some of the GLP-1 resistance might actually be due to some dysfunction of the GLP-1 molecule itself on the receptors. Yeah, no, that's a great question. I think talking to Jens as the expert on GLP-1, I think there is some controversy in terms of the role of the amidation of GLP-1 and what the significance of that is for activity. And it may differ depending on the cell type of where GLP-1 is mediating its effect. We certainly didn't see any difference in the circulating levels between total and amidated and unamidated. But I think that effect is very quick, and whether or not we would actually pick it up in the assays that we're using in serum, I'm not sure. But it's certainly a very reasonable hypothesis to think that that could also contribute to the effect that we're seeing. Thank you. Bernal, Washington University in St. Louis. The question that I have is in regards to the resistance to GLP-1 intervention in the trials. When they didn't decrease the A1C, do they respond to decreasing weight? Oh, that's an excellent question. I was wondering that myself. I don't know that they have captured that in all of the studies that we have had the opportunity to look in so far. But we're currently replicating our study in the Excel study. So I hope that that will be something that we can address, because clearly that would also be of clinical interest. Go ahead. Vincenzo Triscita from Rome University, Italy. Wonderful speech. I wonder whether you can speculate about the possibility that those people who are carriers of the variants associated with type 2 diabetes can be probably treated with higher doses of GLP-1 agonists. Do you think this is possible? I think it's possible, but potentially unlikely, given the resistance. If our hypothesis around the mechanism is correct, I think you would just be continuing to perpetuate the resistance. You don't think the resistance can be overcome? No. No. I think it's probably a feedback. I reckon it's happening at so many levels, I almost imagine it's a bit like a glucokinase situation where you can't intervene and change the inherent regulation. Vinay Simha from Mayo Clinic. I found this fascinating, and this reminded me of the earlier trials on GLP-1, where they showed that if you give erythromycin an increased GI emptying, then all the effects of GLP-1 analogues are completely lost. Do you think that that has a more overriding influence than on insulin secretion? Just want to check. Is the effect on motility more important than the effect on insulin secretion? Got it. I think that's what our data would certainly suggest. I'm really excited that Mahesh, my former student, is back in Adelaide now and actually working in a center where they do gastric emptying in humans. He's planning to replicate Elisa's study in the mice in humans, and I think this will be a phenomenal way of testing that further. We have time for a couple more questions, so I'll just ask, how close are we to being able to use genetic information in selection of pharmacotherapy for type 2 diabetes? You gave a beautiful example, but how close are we to actually implementing this in the clinic? I think pretty close, actually. There was a wonderful paper that came out in Nature Medicine two weeks ago from my colleague Ewan Pearson, where he had really looked at different ways of stratifying patients based on clinical features, but also using something called partitioned genetic risk scores. These are very powerful, I believe, and so do many others. This is a way of taking those 300 signals that we've identified and putting them into groups based on whether or not those same genetic variants altered lipid levels, whether they alter waist-hip ratio, whether they are associated with measures of beta cell function, whether they affect insulin processing. By putting these into discrete scores, you can start to understand whether or not an individual has a type of diabetes that is more heavily influenced by effects on the beta cell or by adiposity. I can see that having a very powerful role in helping a clinician decide a second-line treatment for type 2 diabetes. Thank you very much. Wonderful. Thank you. All right, nobody took me up for standing up and stretching. So I'm excited to introduce our second speaker, Dr. Rochelle Naylor. She is an assistant professor of pediatrics with a secondary appointment in medicine at the section of adult and pediatric endocrinology, diabetes, and metabolism at the University of Chicago. Her medical school training was at the Mayo Clinic and residency training at University of Chicago, where she stayed on faculty. And her research work focuses in genetic and atypical forms of diabetes. She's a co-investigator there for the large U.S. National Monogenic Diabetes Registry housed there, as well as an investigator of rare and atypical diabetes with the RADIAN study and continues to contribute to the ADA Precision Medicine Diabetes Initiative. So Rochelle, we look forward to your talk. Thank you for coming. Thank you so much for the introduction. So it is my pleasure to speak today on monogenic diabetes as an example of precision diagnosis and follow-up in treatment. I don't have any disclosures. And there is a QR code to interact with evaluations, et cetera. So I start with these groups just to give a sense of who I am. So I, as mentioned, I'm a pediatric endocrinologist. Clinically, I spend a lot of time taking care of children with diabetes, including monogenic forms. But I spend much more of my time thinking about diabetes as a researcher. And this started as a fellow when I joined Lou Phillips and Graham Bell and Siri Greeley and others doing work in monogenic diabetes. And then in recent years, this has expanded to the RADIAN study, which is for the rare and atypical diabetes network, where we grapple with what people have when they don't fit the label of type 1 or type 2 or monogenic diabetes, and then also being able to contribute to this Precision Medicine in Diabetes Initiative. And it's wonderful that Dr. Gloin gave the first lecture because she has demonstrated so eloquently the amount of work and effort that goes into figuring out how best to treat the patients. And then I love to take that in the context of precision medicine as being a great exemplar of our, excuse me, monogenic diabetes as being a great exemplar of how we can do precision medicine because it also informs us of what we do well and what we still need to resolve as we expand the reach of precision medicine. So this session is about actually individualizing patient care. So I have to start with a disclaimer that precision medicine is not the same thing as personalized medicine, but it is an imperative step toward personalized medicine. And so precision medicine or stratified medicine, as Dr. Gloin used that term, is medical care designed to optimize efficiency or the therapeutic benefits for particular groups of patients. And we especially mean this when we think about using genetic or molecular profiling. We do know though, once we've started with stratified medicine or precision medicine, we do have to tailor it to the individual, to their circumstances, to their desires around diabetes treatment for it to be effective. The reason that personal, excuse me, precision medicine is such an important starting point to individualizing care is that we all know that diabetes itself is a very heterogeneous disease and the unifying feature is sustained hyperglycemia, but the reasons for that sustained hyperglycemia vary by individuals. And the idea is that if we can be better at classifying the reason someone has diabetes, we'll be able to be, we'll be better able to decide on treatment. So the idea is nicely captured in this graphic. This is a little bit weird to use and apparently I'm failing at it, so I'm not going to use it. How do I get rid of it? Oh, there we go. Just not going to bother with that. You can see the clusters. You take this group of people and you subgroup them to say, oh, you all seem to have absolute beta cell failure. We need to treat you with insulin. This other group, you seem to have a lot of resistance to insulin, but good insulin production. So metformin is a reasonable place to start with you. And then again, there's been a wonderful example where GLP-1 receptor agonist is not going to be probably the next right augmentative therapy. So this idea is better classification will lead us to precision medicine. And precision medicine coupled with personalization holds the promise to improve outcomes. And right now, monogenic diabetes is our best exemplar of precision medicine because it's already proven feasible and it's already practiced, although not with fidelity. So as a reminder, monogenic forms of diabetes are due to highly penetrant pathogenic variants in genes that are important to beta cell development and beta cell function, including the insulin gene itself. It's about a half a percent of all diabetes. But when we're thinking of populations where diabetes onset was under age 30, it represents as much as 3.5%. There are two main clinical phenotypes. There's neonatal diabetes with typically, we classically say diabetes onset under six months. But if you expand up to nine months, you're really going to capture just about everybody. And then there's MODY, maturity onset diabetes of the young, which is an autosomal dominant form of diabetes. And there are syndromic forms and there's overlap with the genes, but this kind of covers the main subgroups. This is a table from Indotex, and it's just listing out the common causes of neonatal diabetes and the common causes of MODY. And in this talk, I'm going to talk specifically about KCNJ11 and ABCC8, as well as GCK, HNF1-alpha and HNF4-alpha MODY. Here as you can see in here with under the rare causes, there were a couple that were asterisked when I wrote this chapter with Dr. Philipson in 2020. There were a couple of genes that had limited evidence for MODY. There's since been a publication really refuting them as MODY genes. So let's talk a little bit about precision diagnosis. So this graph is from the first consensus report of the Precision Medicine in Diabetes Initiative from the ADA and the EASD. And the idea here is that through precision diagnostics, we can refine the characterization of diabetes to optimize therapies or to use it to prognosticate what's going to happen with this particular individual with diabetes, what other medical conditions are they going to be at risk for. And again, this is all in the context of a person's unique biology, as well as their environment and other contextual factors. Of course, we always start where we always start. You get a good history, in particular in monogenic diabetes. You want to get a good family history, which you expect to identify a unilineal family history of three or more generations of young-onset diabetes. And you also want to get a good physical exam. But then it's important to add to that biomarkers. And we have specific biomarkers that help us to distinguish monogenic forms of diabetes from the more typical type 1 and type 2 diabetes. There's also clinical prediction tools, including a MODY calculator. This was developed by the Exeter group, and it is handy as an app that is on my phone. And also the website, you can access the MODY calculator to give you a sense of the probability that your patient has MODY by putting in some basic clinical information, and it allows you to then make decisions on whether or not to pursue genetic testing. I mentioned biomarkers, and these are just a list. This is not exhaustive by any means. But these are the ones that are either the most widely used or the most clinically relevant and usually easily available. I would say if there were only one biomarker class that you were going to use, I would say autoantibodies are very important. From distinguishing type 1, you cannot distinguish type 1 from type 2 on any clinical features anymore, because obesity is quite prevalent. And you can't often reliably distinguish type 1 from monogenic forms of diabetes without first assessing for autoantibodies. And if they're negative, you want to expand your diagnostic considerations. There have been a number of publications that demonstrate the utility of using biomarkers to select people highly likely to have monogenic diabetes, and then going on to do genetic testing in this population. And so this is a nice approach to identify monogenic diabetes. You start, you know, obviously anybody diagnosed under six months of age, I've already mentioned I usually extend that up to nine months of age, they absolutely warrant genetic testing. If you have someone who is diagnosed older than that, but under 30 years of age, then you want to start to think about the other monogenic forms. If they've got deafness or other neurologic features, you want to think about mitochondrial forms. I will tell you anecdotally from our registry, our earlier panels didn't test for mitochondrial forms of diabetes. And recently, when we go back to patients where we've as yet to discover a cause of their diabetes and we've ran them on our newer panels, we're picking up a number of people with mitochondrial diabetes who have less overt features of deafness or other neurologic features than we would expect. And so it's important to, you know, to keep these forms in mind. And then you're going to use, again, the clinical features we just talked about. We're going to use the biomarkers we just talked about to determine who should have modigenetic testing. And again, using this biomarker approach is not only proven to be effective, but it's also proven to be cost-effective. And so I'm going to switch gears now and just mention a couple, two cost-effectiveness studies regarding precision medicine for monogenic diabetes. So I'm going to mention a study from my colleague, Dr. Gree Lee, in neonatal diabetes, and mention a study that I did along with Dr. Goodsmith, who is now a medicine pediatrics resident at the University of Chicago, Dr. Wong and some colleagues. So first starting with neonatal diabetes. The nice thing about neonatal diabetes is that it has a very clear clinical feature, being the age of diabetes onset, that helps you identify who needs to be tested. And so my colleague, Dr. Gree Lee, did a cost-effectiveness analysis of this study. There's a couple things that I have to say so you can, in case you aren't familiar with how cost-effectiveness analyses are done. So when you're doing cost-effectiveness analyses in diabetes, it's typically using simulation modeling. So you take a population, in this case, you took our registry population, and you used their characteristics, but then you use a simulation model to look at what would happen if you did the genetic testing. And that means, in this case, we were looking at the causes of neonatal monogenic diabetes related to KCNJ11 and ABCC8 mutations. These allow for a change from insulin to sulfonylureas, and we'll talk about that a little bit more in subsequent slides, versus the sort of counterfactual of never picking up this diagnosis and continuing insulin treatment. The other important thing to know about cost-effectiveness analysis is that, generally speaking, we actually think it's okay if we're spending money to improve quality of life. And so we usually measure that as quality-adjusted life years, or QALYs. And so in the U.S., we say, if we're spending about $100,000 to increase quality of life, we consider that to be cost-effective. If you spend money and worsen someone's quality of life, we don't like to do that. That's when the scenario's dominating. And then there are times that we like to, where we can actually spend money, excuse me, we can actually save money while improving quality of life. That doesn't happen often, but immunizations are a good example of this, and it rarely happens, though, in other forms of medicine. So those are the important terms to know. And so then this schematic, and again, I'm not going to try to use the mouse anymore, the pointer, but essentially, again, the top line tells you what would happen if you're testing for monogenic diabetes, find these mutations, and are able to convert people to sulfonylureas, versus leaving them on insulin for the rest of their life, including the episodes of hypoglycemia that happen with this, as well as all of the monitoring that happens with insulin use. And the main thing to know here is this word, dominant, that I told you. Every once in a while, you have a health care advancement or intervention where deploying that intervention actually saves the health care system money while improving quality of life, and that is the case when we do genetic testing for neonatal diabetes. What about matured onset diabetes of the young? I will tell you, this is more difficult. There's a lot more clinical overlap between type 1 and type 2 diabetes and modiforms. And so it's not as easy to figure out who ought to be tested. So you end up testing many more people who aren't going to have the mutation than occurs in the case of neonatal diabetes. And so in this case, this is where the biomarker pathway approach, utilizing our known biomarkers to distinguish who is likely to have monogenic forms of diabetes is really necessary. So in this study, and this is much busier because, again, it's more complicated to figure out who you ought to test, but this study is based on the search for diabetes in youth population. And so we use that U.S. epidemiologic population of children with diabetes, and we apply the biomarker approach saying everybody in the cohort gets tested to see whether or not they have autoantibodies, and to see if they don't have autoantibodies, whether or not they still make C-peptide. Those people who have negative autoantibodies and positive C-peptide then go on to get the genetic testing. By doing this, you whittle down your testing population substantially, and then you're able to test for, in this case, we did just the three common forms of MODY because they're clinically actionable. We're going to talk about that, you know, in subsequent slides. But this is HNF1-alpha and 4-alpha, also amenable to treatment with sulfonylureas, and then this is GCK-MODY that doesn't need therapy at all. And so, again, we count, we model in the top portion this scenario of doing the genetic testing and switching people to precision medicine. And in the bottom scenario, we model them never getting this diagnosis. The strength of this study is because it was modeled on the search population, we didn't say, oh, well, every kid is going to be on insulin because most kids have type 1 diabetes. We actually modeled them being on the therapy that they actually were on in the search study prior to getting a genetic diagnosis of monogenic diabetes. And once again, the thing that I want to highlight here is this word dominant. So this is another example of being able to apply genetic testing to the correct population, improving quality of life, and saving money. Again, a rare, rare feat, particularly in the U.S. healthcare system. So I will drive home the point that it is cost-effective to make these diagnoses, so when the insurance company tells you they will not cover the genetic testing, you should push back. Okay. And then we're going to take one of our two last turns to now focus on examples of precision medicine for the common forms of monogenic diabetes. I've alluded to them, and we're just going to talk about them in a little bit more detail. So Dr. Goyne also showed this slide when discussing neonatal diabetes, and so when we think about the KTP-related forms of neonatal diabetes owing to mutations in the KCNJ11 or the ABCC8 gene, the issue here is these activating genes hold the ATP-sensitive potassium channel open. So the cell can't depolarize, and you cannot get insulin secretion. But as already mentioned, these sulfonylureas can bind this channel, excuse me, bind this receptor, close the channel, and you can get insulin secretion. And so in this way, sulfonylureas are precision medicine for KTP-related forms of neonatal diabetes. And what that looks like is you take a population, and I'm actually gonna be brave and try one more time. Oh, nope, that didn't work. I'm, nope, so I'm, I tried, I failed. So you take that population and you see on the left that before the introduction of sulfonylureas, when all of these patients were on insulin, you see a widespread of hemoglobin A1Cs, a number of which are not at goal. And remember, these are infants. These are people who will have diabetes for their whole life. So you really do have to have good control the whole time. And you then switch them to sulfonylureas, and you see this switch, this mean A1C decreases from 8.1 to 5.9. That is the power of precision medicine. When we think about HNF1-alpha and 4-alpha, the two, the first and third most common cause of MODY, both of these are also responsive to sulfonylureas. Both of these result in a progressive defect in glucose-dependent insulin secretion, result in a young-onset diabetes. And like all other forms of monogenic and polygenic diabetes, excluding GCK-MODY that we'll talk about, they are at risk for microvascular and macrovascular complications related to their control. So good control is important. Again, this is young-onset, often in adolescence or young adulthood. So again, you have to, they're going to have a long exposure to diabetes. In this case, sulfonylureas, again, bypass a number of issues that are perturbed by defects in the gene. Okay, thank you. Got a tutorial there. He gave me one previously. I just forgot what he told me. So there we go. So you see you bypass a number of the defects in the two genes when you use sulfonylureas. And in fact, these populations are hypersensitive to sulfonylureas, so that you're often using low doses of this medication, quarter of a tab, a half a tab. In fact, some people are so hypersensitive that they can't tolerate the sulfonylureas. They're actually hypoglycemic on the lowest doses. And then when that happens, the meglitinides that bind in the same spot, but not as potently, are a nice alternative. Similar to the story in KTP-related neonatal diabetes, when you transition people, even after decades of the wrong therapy, sometimes people can transition, you will often find a hemoglobin A1C benefit. And this is a cohort from Ireland where they're showing you their, theoretically I had mastered it, and I've somehow not mastered it again, their A1C before transition, and then they follow up A1C after transition. And then I'll lastly end on GCK-MODY. So GCK encodes the glucokinase enzyme, which is the body's glucose sensor. People who have a heterozygous inactivating mutation in the GCK gene have a higher set point for their body to say, hey, our blood sugar is rising, we should secrete some insulin. However, the rest of their glucose metabolism is intact. And so what this results in is a stable, mild fasting hyperglycemia, and A1Cs in this population that usually fall within the range of 5.6 to 7.6%. What this graph depicts is the population in yellow are unaffected relatives of people with GCK-MODY. The population in, we'll call it green, are those with GCK-MODY. You see these lines are really parallel, but just shifted up for the population with GCK-MODY. And then you see in both populations, everybody has that age-related decline and glycemia that we all experience. Unlike all other forms of diabetes in GCK, those microvascular complications and macrovascular complications that we are so worried about are exceedingly rare. In a study of a cohort with a mean age of around 50, which means 50 years of elevated blood sugar because that elevation occurs at the time, like in utero and at birth, the only significant microvascular complication that was found was all background retinopathy that hadn't needed any intervention. The other thing, and Dr. Goyan alluded to this, is that this is a type of diabetes, if we want to call it that, certainly a form of MODY that doesn't need therapy. Precision therapy is no therapy. And giving therapy doesn't alter treatment. So you see in the population, where I'm probably just not pressing in the right spot sometimes, before diagnosis, you see people who had oral medications, you see people who had insulin. They then got the correct diagnosis, switched off of all therapy, and you see A1C is not altered. There is an exception to pregnancy, and we'll just mention that. In the last few minutes, we're gonna talk about, after these compelling examples of precision medicine, what are the remaining barriers? And I'm ending with this session because this is one of the most important things to think about. Precision medicine is really actually quite straightforward in these monogenic forms of diabetes. And so whatever is complicated here is going to be that much more complicated when we try to apply precision medicine to polygenic forms of diabetes. So what are some of the barriers? One huge barrier is the delay in precision diagnosis, so you can get to the precision therapeutics. For this, I'm gonna highlight just some data from the Monogenic Diabetes Registry. So this registry has been in place since 2008, and we looked at our data from 2008 and 2020 in a recent publication. And the one piece of data, the point that I wanna pull out is how long people go from the time they are diagnosed with diabetes to actually getting a molecular diagnosis of monogenic diabetes. That is in excess of 12 years. And that is usually because there are delays in considering monogenic diabetes and then there's even delays between considering monogenic diabetes and actually being able to get the genetic testing. And this is very often because of insurance barriers, and that is why I mentioned this is cost-effective. So you should push back when the insurance company does not want to cover this genetic testing. In fact, within the registry, about two-thirds of all individuals with a molecular diagnosis have it on the basis of research testing rather than commercial testing. What is the impact of delay? This is one example. Going back again to KTP-related neonatal diabetes, what these graphics show is that the older someone was at the age of SU initiation, the higher the dose they need. And I should point out that unlike HNF1-alpha and 4-alpha, which are quite sensitive to sulfonylureas, neonatal diabetes requires really relatively large doses compared to what someone would take if they had type 2 diabetes. And so you needed even larger doses, which is a dosage burden for patients when you delayed SU initiation. The other thing that you see is that the older they were, the more likely they were going to need augmentative therapies. One, I like to tell you bad news and then leave you with a little ray of hope. So one thing that we can tell with the registry is we took those 12 years that we looked at, cut them right in half. We saw that the amount of people who needed research-based testing was declining over time from 71% to 59%. And that was because they were being able to access the commercial-based testing a little bit more easily. Although sometimes this was just because they were able to self-pay because a lot of the commercial labs have nice robust payment protection plans for patients. There's also failed implementation of precision medicine. So I mentioned for GCK-Modi, precision therapy is no therapy at all. The only exception is in pregnancy where a mother with GCK-Modi is carrying a fetus that does not have the mutation. In that case, the baby is not gonna like mother's increased blood sugars and you may need to treat with insulin. However, if baby does have the mutation, you shouldn't treat mother or the baby. This algorithm is generally the recommendations for how to treat. So often we don't know the fetal phenotype and so we infer it based on growth in the second trimester ultrasounds. This is not without controversy. There are groups that advocate for insulin right away, but I will tell you, when you don't apply precision therapy, you do do harm. So looking again at the registry, looking at data for women in the registry with GCK-Modi who'd had pregnancies, we specifically focused for a portion of the study on women who knew they had GCK-Modi when they were pregnant to say, well, what happened with your therapy? And what you can see here is a large number of them were treated, including more than half treated with insulin therapy. And the time of initiating insulin ranged from as little as five weeks to 32 weeks and seems in no way tied to the antenatal scan as recommended in the slide I showed you before. And the thing I wanna point out is this was harmful. Almost a quarter of women had severe hypoglycemia, meaning they needed another human being to assist them. There was one registry participant who actually quit her job because she couldn't function because of her hypoglycemia. The infants who were positive for the variant and their mothers received insulin also had growth restriction. So they were smaller than those infants who weren't treated. So when we fail to apply precision medicine, even when precision medicine is no medicine at all, we do harm. Now, the silver lining to this is that using cell-free fetal DNA, there's been proof of concept studies to say that we would be able to use this to identify the genotype of fetuses and so we can hopefully more rationally apply precision medicine in pregnancies affected by TCK-MODY. And then the last thing I'll leave you with is that precision medicine can actually still be hard and we need to remember this and this is why the individualization becomes so important. This is a study of patients in our registry with KTP-related diabetes who've been transitioned to sulfonylureas who actually have a lot of difficulty taking the large number of pills or the multiple doses a day. So almost a quarter are frequently missing doses. 90% are often taking their doses differently than prescribed, sometimes intentionally, but other times unintentionally resulting in harm including hyperglycemia and even DKA. And lastly, this paper recently came out looking at psychiatric comorbidities in patients with GCK-MODY where I usually think, hey, you've got no diabetes at all, you don't need treatment, life is great, and yet this study showed that in children with GCK-MODY, while their quality of life was better than patients with type 1 diabetes, they had a higher incidence of anxiety-related psychiatric comorbidities and that's something that's fully unexplored. So maybe GCK-MODY diagnosis is not exactly a get-out-of-jail-free card for these families and that needs to be explored more. Nevertheless, despite these unresolved questions and some of these difficulties, precision medicine coupled with personalization still holds the promise to improve outcomes. And I will leave you and remind you of the three compelling examples including the robust response of sulfonylureas in K to P-related neonatal diabetes, the robust response of sulfonylureas in HNF1-alpha as well as 4-alpha-MODY, and no therapy at all in GCK-MODY. And with that, thank you for your attention. Oh, let me thank all the people I work with, some who've passed on in the great sense, some who have moved on but have influenced my career, my funding sources from the NIDDK. Thank you so much, Dr. Naylor. We do have a couple minutes for questions if anyone wants to come up to the microphone. Yes, sir. First of all, those are very nice. Al Powers from Vanderbilt. I was, your comments about mitochondrial defects in your monogenic restory, that surprised me, right? But I thought that was interesting. So has that, have mitochondrial defects been looked in widespread run-of-the-mill diabetes? I mean, the way monogenic forms have done that, have people looked in large populations for how frequent those are? I'm actually not aware of whether or not people have looked in large populations. I know, as I alluded to in our registry where we have people coming with atypical forms. Unfortunately, it didn't, mitochondrial, the common variants didn't make it onto our panel until in the last several years. We're up to version five of our panel, and that's finally the panel that we are able to pick up mitochondrial mutations. As we're going back, we've uncovered a handful of cases now. So I think they're more common than we were suspecting. And again, I think part of it is that the phenotype is more variable, and the expression of all of the neurologic findings that we're expecting to find are not always there. And we've seen this in other cases when we are able to look with a panel, we realize that there are more mild forms of syndromic forms of diabetes. But I'm not sure if it's been looked at in the large population. One of your slides mentioned HSCRP as a way of differentiating between MODY and type two. Can you please elaborate? Yeah, so the high sensitivity CRP has been studied, and it actually differentiates HNF1-alpha-MODY from other forms of MODY, as well as from type one and type two, just because HSCRP is in part an inflammatory marker, and so it's higher in type two and type one than it is in MODY, and then again, higher in the other forms of MODY than HNF1-alpha-MODY. The problem is it's not as easily obtained. I think I could order it in my clinic, but it's just not as robustly used. But there are a number of studies that prove its utility in HNF1-alpha-MODY, but really, probably by the time you've applied antibodies and proven that there's endogenous insulin production, that's enough to really, you were suspecting monogenic diabetes in the first place, that's probably enough where the HSCRP is not adding on a ton of value in the clinical setting. I'll sneak in one more quick question. It was so exciting to hear. I think maybe endocrinologists are the best at understanding that a test doesn't always give you a black and white answer, and you alluded to the difference between research sequencing and commercially available sequencing. Can you talk a little bit in your work about the burden between clinical sequencing, research sequencing, and how we could actually extrapolate this to serve lots of people? Yeah, so I'm not sure if I'm gonna answer quite the question you intended, but I will say, so on a research basis, we have this huge panel. We're trying to discover new things. We come up with things we don't necessarily know the significance of, and we can go back to family members and do all sorts of testing. Most commercial panels only have the proven genes, although a number still have genes that were recently refuted, and so when you're ordering this in the clinical setting, it still is really important to be mindful that the genetic testing report is understood. We have seen instances where people say, I thought that they had monogenic diabetes. There's a variant in the monogenic diabetes gene, and therefore, they have this form of MODY, and unfortunately, that's not true. We all know we all have mutations all over our genome, and most of them aren't doing anything, and so certainly in the clinical context, it's very important that, of course, you don't kind of go fishing for MODY if there's not clinical suspicion, and that even when you find something that seems like it may confirm a MODY diagnosis, that you're careful about the diagnosis, and certainly reach out to people who have expertise if there's any questions. I'm not sure if that quite got at it. It did. It's too long for our discussion today, but I'm sure I'll talk with you more about it soon, so thank you again, Dr. Mason. Thank you all. So, I'm happy to be able to introduce our final speaker for this symposium, Professor Vincenzo Truscita, who is a professor of endocrinology at Sapienza University in Rome, Italy. He has expertise in the epidemiology and genetics of type 2 diabetes and its cardiovascular complications, and as such, he is an expert in the field of endocrinology and diabetes. And as such, he is an author of more than 240 publications in this space. He is going to be speaking on the prediction of all-cause mortality in type 2 diabetes today. Thank you. Thank you for the presentation. Dear colleagues, first of all, let me thank the Endocrine Society for inviting me to deliver this speech. I'm very happy to be here. And, okay, I have no, I have no financial relationship to disclose. I think I have to stay with this slide for a short while. Okay. And so, today we have already heard about some precision medicine approaches. We have heard about precision, I think I better get rid of this. We have heard about precision intervention treatment, precision diagnosis, and I'm going to talk about precision prediction, which very simply means the aim to identify subgroups of individual who are at different risk of developing a given disease or a complication or some clinical outcomes that you are interested on. And this slide reminds us what we know about the risk of all-cause mortality in patients with type 2 diabetes. As you see, the incidence rate is decreased over the years, but, I mean, this is exactly what happened with matched control, so at the very end of the day, the gap between people with diabetes and those without diabetes is still there, if not, probably it's also a little bit increased over the years, so there is something wrong that we cannot do, we are not able to do, and part of the answer to this gap is, I think, reported in this slide here. In this slide, you see the proportion of people with diabetes who are a target, A, for glycemic control or for blood pressure or for cholesterol or those who do not smoke, but the frustrating data are down here because then if you look at the proportion of individuals who are a target for all four very well-known risk factors, the numbers are really frustrating, about 20, 25%, and this proportion is not increasing over the years, so we are not doing very well, and this, to me, makes very clear that what we need are better and probably more expensive model or care that should be tailored to the individual risk profile, limiting them to patients at higher risk, otherwise, they will not be doable, and also, we need to maximize the effectiveness and minimize costs. Let me give you an example of what I mean with better and probably more expensive model or cares. I live in Italy. We have four million people with diabetes, and we have 700 diabetes care units, which actually do a very good job as an average, but unfortunately, only one-third of people with diabetes can attend this diabetes care unit, so two-thirds cannot attend, and the choice about who is going to attend or who is not going to attend is not based on any rational basis. It's just because where they live or how good is their GP or how good is their own empowerment and so on, and this is a pity because it's very well known that attending a diabetes care unit increase the likelihood to be a target for these four and also many other risk factors, and eventually, reduce the risk of all-cause mortality, so basically, we got to the point, and all this was to say and to share with you the idea that to customize different types of care for different people, what we need are well-performing risk prediction model, and of course, I mean, they better be parsimonious, especially if they are going to be used in context with limited resources, okay? So do we have models, prediction models, for all-cause mortality in patients with type 2 diabetes? The answer is yes. We have at least two well-established, validated prediction model I will show you in a second, but to make closer the dream of precision medicine in type 2 diabetes as far as mortality risk is concerned, these two models, both models have to be improved, and what we have done in the last four or five years is to try improve these two models, okay? So let's give a look to the two model. One, we set up one of the two model. We set up the first version in 2013, then improved it a little bit in 2019. I mean, these are the main features of this model. I like to remind you that it's very parsimonious. We only need nine very common variables to run the model and it's freely available on the web. And there is a second well-performing prediction model, which is even better than ours, which has been set up by Basu, John Yudkin and co-workers. It has been validated, as in our case, in several independent populations. It is probably not so parsimonious. You need 14 variables. But what is important about this model is that the model works also with some missing data and that if you change the information that you give to the model, the model helps you also to predict all chronic complications in people with type 2 diabetes. So an excellent model. As I said, we spent the last four or five years to try and improve the performances of both models and now I'm going to show you some of the data, the very recent ones that we got working with metabolomics. But let me just still have a general slide for sharing with you what you do to improve models. Well, the most simple thing that you want to do is to put novel or biomarkers on top of the model and see whether or not you are able to improve them. And when you choose the biomarkers that you want to put on top of the model, you can either use established markers and we did that with some inflammatory markers. And this was a proof of concept that if you had inflammatory markers on top of both models, you improve the models. As a matter of fact, if you want to use our model and you go online, you will find two different versions, one with CRP and the other one without CRP, if you don't have CRP. Or you can go out for a fishing expedition, try to discover novel markers. This, of course, is a hypothesis-free approach. You can do that, as you already heard, with genomics, proteomics. And we have done that with metabolomics markers. Okay. And I'm going to show you some data, some very recent data. Some of them are just unpublished. If I have time, and I think I'm going to have time, I will show you also two slides to share with you what you can do when you discover novel markers, not only about prediction, but also to address novel pathogenic pathways. Okay. So one more slide, which is a sort of tedious one, but I ask you to stay with this slide for one minute because it is important. These are the three indexes that we are going to talk about now in the next few slides. These are the three indexes that you want to use if you want to see whether the new model, I mean the model that you are trying to improve, is really improved as compared to the standard model. And so we have the SIS statistic. I'm going to read it because I want to say it well, which tells you the probability that the risk predicted at baseline to an individual who subsequently will develop the event is greater than that predicted to a counterpart will not develop the event. Of course you know that because you are at the end of the study, so you have this data, and then you can make this calculation. And of course the Delta SIS statistic compares the new model with the new markers that you want to test with the whole model, the standard model. And then we have also the IDI, which is the integrated discrimination improvement. This time you are not comparing single individuals, but you are comparing the mean risk probabilities predicted at baseline in individuals who will develop and those who will not develop. And again, of course, IDI compares the new model with the new markers and the whole model, the standard model. And finally you have the net reclassification improvement, which tells you about the ability of the new model to correctly reclassify the wrong prediction, which is to turn false positive into true negative and false negative into true positive. So now you are going to see these three indexes, and now we know what they are. So let's talk about the data that we got using a couple of different metabolomic platforms. Okay, in the first study we wanted to discover metabolomic markers, metabolites if you prefer, that were associated with all-cause mortality in people with type 2 diabetes, and then we decided to use two different samples. GMS stays for Gargano mortality study. Gargano is a nice area in the Apulia region, south of Italy, and GMS2 is Gargano mortality study 2, and we decided to use the first study as a discovery sample and the second study as a validation sample. Okay, the clinical features that you see here are very typical of people with type 2 diabetes, nothing very interesting to talk about. And by using the Biocratest platform, which assesses 182 metabolites, we found actually that after Bonferroni correction, 49 out of these 182 metabolites were significantly associated with all-cause mortality in the first sample that we used. And you here find the families to which these 49 metabolites belong. However, since we know that all-cause mortality is influenced by several clinical factors, and also we expected what we found, which is that the 49 metabolites were very much intercorrelated, we ran an additional analysis to be conservative enough, where not only we applied stepwise approach, but also we adjusted for anything we could take, we could adjust for, age of recruitment, sex, smoking habit, BMI, A1C, GFR, diabetes duration, and all ongoing treatments, and only six, these six metabolites of the previous 49 metabolites survived the association. So we went down from 182 to 49 and then to six. But to be even more conservative, as I said at the beginning, we wanted to validate this data, and then we tried to replicate these six metabolites in the second independent sample. And actually only three metabolites, tryptophan, quinuranine, which is a degradation product of tryptophan, and exonoyl carnitine, remained, survived this further step. So at the very end of the day, we found three metabolites that were associated in an independent manner, with all-cause mortality in type 2 diabetes, and that have been validated in an independent sample. So let's see what these three metabolites, how we can use these three metabolites for our purposes, to improve prediction model. So the three metabolites altogether have a percent of C-statistic of 71, not that bad, but not that good, which is of course worse than what we have with mFORCE by itself or RECODE. But an interesting thing is that when you add the three metabolites on top of mFORCE and on top of RECODE, you get an improvement of the delta C-statistic, the percent IDI, this is discrimination. I want to remind you that the American Heart Association and the American College of Cardiology suggest to add new markers on top of their prediction models only if percent IDI is below the threshold of 6, and we are here well below the threshold of 6, okay? Sorry, above the threshold of 6, okay? And most importantly, there are a great proportion of individuals that are reclassified in both models when you add these three metabolites on top of the two models, and most of them, although not exclusively, are non-events that have been correctly reclassified. So we do find metabolites that are associated with all-cause mortality and then can improve the prediction model that we have. But as I said at the very beginning, when you find novel markers, you probably can also address some novel pathogenic pathways, and the next two slides are to show that this is actually the case. There are some data in the literature saying that at least in vitro tryptophan is linked to some inflammatory pathways. Data are very not clear, but I mean, we decided to pursue this possibility, and we looked for the association between tryptophan in our samples and several cytokines we were able to measure, and actually five of them, interleukin 6, A13, interferon gamma, and TNF alpha, were strongly associated with tryptophan. Then we asked the question of whether or not these five cytokines were able to mediate the protective effect of tryptophan on all-cause mortality, and the data shown in the next slide, this is the effect of tryptophan on all-cause mortality as a ratio of 0.79, and as you can see, while interleukin 18 or interferon gamma doesn't change the association, this is not the case for interleukin 6, 13, TNF alpha, because the association is reduced when you take into account this cytokine, and this is the proportion of the protective effect of tryptophan on all-cause mortality that is explained by each single cytokine, and when you use them all together, you explain approximately 60% of the protective effect of tryptophan on all-cause mortality. So we started to find or to look for metabolites in order to improve prediction models, and we have here a good hypothesis to work on about a new pathway that can shape the risk of all-cause mortality in people with type 2 diabetes. But let's go back to prediction with the last two slides. An interesting question, a very important question, is this one. Can general population markers be used in type 2 diabetes, or if you prefer, do I really need to look for novel markers in people with type 2 diabetes? Can I use in people with type 2 diabetes markers that have popped up from studies in the general population, which are, of course, more common and, generally speaking, much larger? And the answer is, unfortunately, sometimes yes and sometimes no. To answer this question, we used as a base to work on a beautiful paper published a few years ago by Joris Dolin and co-workers on natural communication showing that using a different platform, a metabolite platform, is the Nightingale platform. They went out with a 14 metabolite score with a very, very good AUC. AUC is equivalent to the CIS statistic that we were talking before, so approximately 84% and much better than conventional risk factor. And what about these 14 metabolites in people with type 2 diabetes? To these purposes, we use another sample that we have in our lab. It's the summer study. Even here, these are typical features of people with type 2 diabetes. But the interesting results are here, because in type 2 diabetes, the same 14 metabolite score, sure, the score was strongly associated with all-cause death, 10 to the minus 16 or something like that, 1.76 for one standard deviation of the score. But unfortunately, it's a poor predictor of all-cause death as compared to the general population. To understand why this was the case, we gave a look to each individual metabolite that is comprised in the score. And basically, we found two totally different results. We found four metabolites which were significantly associated in our hands with all-cause mortality in patients with type 2 diabetes, very much at the same level of the general population. And okay, so these four metabolites are validated. But we found also three metabolites here, two lipids and isoleucine, which are not associated with all-cause mortality in type 2 diabetes, where of course, by definition, they were associated with all-cause mortality in the general population. And they were not associated with all-cause mortality in type 2 diabetes, despite the fact that we have enough power to find an association similar to that reported in the general population. And most importantly for me, the results that we got are totally different, significantly different than those reported in the general population. So these three metabolites seem to be really specific for the general population and not applicable to people with type 2 diabetes. We don't have power enough for the other seven, and I'm running also out of time. So basically, let me summarize with this slide. We have discovered or validated several metabolites that are strongly and independently associated with mortality rate in patients with type 2 diabetes. And this is important. Data from the general population are not always usable in people with type 2 diabetes. So if you want to do that, you have first to validate the markers that have been used in the general population to see whether they can also be used in people with type 2 diabetes. This is not always the case. So on the prediction side, these metabolites improve the performance, as I showed you, of established prediction model of all-cause mortality in type 2 diabetes. On the mechanistic side, there is much to do, but we know that if you go out for a fishing expedition and you discover novel markers, this can also help address novel pathogenic pathways. And before to conclude, let me thank the people that have done all this work. Claudia Menzaghi, who is an independent scientist in our institution. Maria Giovanna Scarale, who now moves to Milan. And our quantitative person, excellent quantitative person, who was not there the day we took this picture, because, as usual, he was skiing, is Massimiliano Copetti. Okay, thank you very much for your attention. Thank you. Thank you so much, Dr. Tuscuta. We have time for some questions. Shane Hamlin from Melbourne, Australia. Thank you very much. This is very thought-provoking. I'm a bit confused, though, with the tryptophan. It appeared to be helpful or protective, but the metabolite of tryptophan appeared to be deleterious. Did I get that right? I can't understand the mechanism why that would be. No, no, no. No, no, it's right. I mean, you can have different hypotheses, but tryptophan is degraded in not only, but also in canurin. And the idea is that the enzyme, if the enzyme works too much, you have low tryptophan level and too much canurin, which is not good for your health. And in this case, for the risk of people with type 2 diabetes to die or not. So, basically, yes. And, in fact, what you can do and what we actually did, I didn't show the data. You can use the ratio. You can use the canurin to tryptophan ratio. And the higher the ratio, the higher the risk to die. So, I mean, you got the right. Again, thank you for a wonderful lecture. My question is, do you have data or hypotheses? Will these markers be also predictive on the personal effect of medications on a patient? Is that useful? That would be great, but, unfortunately, the answer is not. Al Powers, Vanderbilt. Thank you for the talk. So, you said that the markers that are useful in the general population are not always useful in the diabetic. How about the opposite? Are there insights that the diabetic samples gave you that might help in the general population? I don't have data on that. And, as far as I know, there are no data in the literature. But, I mean, it's a good question. And I would not be surprised if some of them would be extendable or has people that knows about. Right, because you could think that maybe the pathogenic mechanisms are accelerated or enhanced in diabetes. And they might not be as visible in the general population. So, they might provide insight. Yeah, on one side can be accelerated. On the other side can just be different. So, in the first example, I mean, I would not be surprised if metabolites that are associated to all-cause mortality in people tied to diabetes are also associated in the general population. Of course, I mean, if the metabolite is, you know, sort of assigning a specific pathway, which is specific for people with type 2 diabetes and not in the general population, we expect not to find the same metabolite in the general population. But, I mean, I don't have the answer, but I think it's a good question. Hi, thank you again. That was a really wonderful presentation. Dawn Davis from University of Wisconsin-Madison. My question is a little bit more just in general about using metabolites as markers here. So, are there studies or have you guys looked at, you know, following a single population and then retesting for metabolites over time? So, you know, genetics is something that doesn't necessarily change, but metabolites certainly can change minute to minute within an individual. So, I'm just wondering how robust these are in terms of if you test for metabolites in the same individuals at two different time points, do your predictive models hold up? Yes, I mean, I understand the point. I don't think we should decide to do either one or the other one. I think we should do as much as we can, both metabolites and genetics. But, I mean, let me spend, I mean, some words if I have time, yeah, about why to choose metabolites. I'm not saying that we should choose metabolites. From a biological point of view, the distance between the genes and the phenotype is enormous. From the same biological point of view, the distance between the metabolites and the phenotype can be much, much smaller. It's not for sure, but it can be much smaller. So, although you are right, metabolites change and genetic tests do not change, there is probably some reason to prefer metabolites, okay? The real stuff is that genetics is perfect because you can start testing at the very beginning of the life. Genetics is the real strong point for genetics, which, of course, you cannot do with metabolites because they keep changing according to basically probably everything, also treatments. Thank you. Okay, we have one more question from online. And the question is, are models validated against insulin resistance? And I'll add a little bit to that. I think they're looking, that we understand within T2D, there's still a lot of heterogeneity. How do you think metabolites can provide insight to the heterogeneity we know exists within T2D? And are there ways that metabolites actually cluster or reverse predict subphenotypes within T2D? Well, I don't have data. When you say that a model has been validated, you mean very simply that you set up a model in a specific population and then, of course, those are by definition data-driven, and then you move to additional population two, three, four times and see whether the model performs well or not. This means validation. But, of course, it would be interesting to have the possibility to apply the model in subgroups of individuals. I understand the point. I think it would be great. Probably the performance in some of these subgroups would be even much higher than what we have in the general sample. And, of course, in other subgroups, it would be much lower. But I don't have data on that. And as far as I know, I mean, there is no one that has been trying to do this kind of analysis. Thank you again, Dr. Truskita, for a great presentation and to all the speakers for a fabulous symposium. Thank you.

precision medicine

monogenic diabetes

diabetes type

treatment strategies

biomarkers

genetic testing

cost-effectiveness

patient outcomes

prediction models

all-cause mortality

type 2 diabetes

metabolomic markers