 I'm not sure what I can add to what we just heard. This is the title that I think I said myself. I'm going to take a look back and look forward and tell stories that will be familiar to many of you. But I thought my job here is to sort of set the tone. I'm going to show data that I think most of you have seen before and are familiar with, but in a context that sort of sets forth what we've learned and what we don't know, and the roadmap is down here. I think it is important to emphasize that we will never get anywhere unless we understand the underlying biology. We have to make discovery in order to implement discovery. And I think that's self-evident. I know we're here to implement. I'm not going to dwell on the discovery part, but I'm going to talk a little bit about it. Now, if you ask your mother whether every single person who gets a drug responds in the same way, the answer will be no. Everybody expects there to be variability in response to drugs. These are old data from the Pharmacogenomics Research Network. Julie will recognize this graph. Maybe it's a long, long time ago. And the notion is that no matter what metric of drug response you look at, there is always variability around a mean effect. What we're much more interested in and what catches people's attention is the rare adverse drug events. And some of those are pictured here. And for those of you who can't figure out what they are, they're shown here. So just to make the point, and I think Simone will probably make it later, that there is a lot of activity across NIH looking at basic mechanisms. These are web pages from the current Pharmacogenomics Research Network. And this is us, and this is the Stevens-Johnson project going on at Vanderbilt, supported through a P50 award. And the important project is the QT project that is highlighted here, because that's what I do. Adverse drug events are rare. And in the practice of an individual physician, come up so rarely that they may not be recognized as Mr. Anderson highlighted for us. That said, they are an enormous problem across the country. These are old data, venerable data. People love to quote these as adverse drug events being the fourth to sixth leading cause of death in hospital. Those are 1998 data. In 2010, the data were really not very different. 100,000 people a year would be the equivalent of one jumbo jet crashing and killing all the passengers aboard every single day. So this is not a small problem. This is an enormous problem. In the United Kingdom, Manir Parmohamed and his colleagues looked at hospitalizations in three years across 18,000 patients, and 6 and 1 half percent of them were due to an adverse, 6 and 1 half percent of admissions were due to an adverse drug reaction. These are the drugs I don't expect you to look at the list and memorize them. But the estimate is that somewhere around a third of these have a prominent genetic component to risk. For those of you who can't remember why drugs work and how drugs work, this is my simplified version of truth. That is, you give a dose of a drug and it's delivered to its molecular target. Lots of things happen in between. And those are the pharmacokinetic variants that people worry about. And then once it reaches a molecular target, good and bad things happen at the whole organ and the whole organism level, and that's the pharmacodynamic piece. And of course, all the genes that are responsible for transducing pharmacokinetics or pharmacodynamics are candidates for variable drug effects through pharmacogenomic mechanisms. This is a venerable slide from Mary and Bill in a science article in the last century listing major drug metabolizing pathways. I used to highlight which ones had polymorphisms in them, but I don't do that anymore because they all do. But I think this is the framework. That said, I think this is not a terribly, this is a catalog. And I think another good way of thinking about it is this concept of high-risk pharmacokinetics that Mike Steen and I actually wrote about 10 years after Mary. And I think it's a concept that's sort of self-evident to everybody in this room. And that is there are certain situations in which a polymorphism in one of those drug metabolizing pathways assumes particularly clinical importance. One setting is a pro-drug that is bioactivated by a pathway that either is absent for genetic reasons or because of phenocopying by administration of an inhibiting drug. The best example is enkinide, but that's no longer in the market. And I spent 10 years of my life with enkinide, so that's why I put it here. But I do put it in smaller type. And clopidogrel, tamoxifen are like this. Coding, of course, is a pro-drug that's bioactivated through this pathway, not bioinactivated through this pathway to morphine. And you all know the stories around that. The other high-risk pharmacokinetic setting is where a parent drug is bioinactivated through a single pathway, the same idea. That only becomes important if there's a narrow margin between efficacy and safety for these drugs. If a drug has a very wide margin, then inhibiting the pathway doesn't have as great of an effect. Examples of that are warfarin, urinary t-can, is a thioparin. And the other way in which this can happen is if you have failure of the excreter organ. That's not genetic. But I'll come back to that and give you a specific example of sodalol and renal failure. Somebody on sodalol develops renal failure. They will develop toxicity because of very elevated plasma concentrations unless the dose is reduced or the drug is stopped. I want to talk about warfarin. And I do this with very mixed emotions, because I think warfarin as a drug is going away. But warfarin has taught this community a lot. And I think the thing to take away is the lessons that we have learned good and bad over the years. So Alan Reddy described a paper, described warfarin metabolism, and described the fact that there was a specific high affinity pathway for warfarin metabolism in liver microsomes, again in the last century. That turns out to be CYP2C19. He actually described the Star II polymorphism as suggested that might be a mechanism for variable drug actions. And again, in the last century, people noticed that there were patients who received very low doses of warfarin and had a high burden of reduction or loss of functional leels in CYP2C9. And interestingly, they had an increased incidence of bleeding, an observation that's really interesting, and I'll come back to later. My throat is better, in fact, than it's been for the last two weeks. So if you ask somebody at the beginning of the 21st century, what's the pharmacogenetic story around warfarin? This is the story, S. warfarin bioinactivated by CYP2C9, and there are variants in CYP2C9. And the FDA was all ready to start a trial looking at the effect of preemptive genotyping for CYP2C9. Some of you will remember that. And around that time, there's this paper in Nature that describes variants in B-core C1 as a cause of warfarin resistance, and it becomes clear that there's another piece to this story that involves the pharmacogenomic piece, B-core C1, which is the target for warfarin action. Terry Klein and others organized the International Warfarin Pharmacogenomics Consortium that put together 5,000 people and their genetic data from across the world. And there are a couple of observations that we made that were important. Number one, though a steady-state warfarin dose varies by ancestry. Number two, a lot of that variability is attributable to changes to a common polymorphism in the promoter of B-core C1, as well as to rarer variants in CYP2C9. We catalog the variants in CYP2C9 in European populations, it turns out there are others in African populations that we just didn't look at in those early days. And then you could compare an algorithm based on pharmacogenetics, an algorithm based on clinical features alone, or an algorithm that said start with 5 milligrams a day and go from there across all the patients that we had in the consortium. And you could show that adding genetics makes a difference in people who end up on higher low doses. The fact is, most people end up in the middle. So this is one of the lessons that if we're going to implement pharmacogenomics, it only benefits a subset of the population. These people don't need genetics. Another piece of the puzzle that I think we haven't paid as much attention to as we should is this idea that there are rare coding variants, rare non-synonyms variants in the coding region of V-core C1. This is a really nice study looking at people who take very high doses of warfarin. And the reason they take very high doses of warfarin is basically because they're non-compliant. But when you start to do genetic testing in those patients, you find occasional patients who are compliant but who have variant genetics. And this is a nice example. This is D36Y. It's a rare variant, except if you happen to be running an adicoagulant clinic in Israel where it's quite a common variant, and it accounts for high doses. So that's the story in 2008. Fast forward to 2017. You can look up in Nomad as I did last night. There are 231 non-synonymous variants in V-core C1. So each one of those is a candidate for transducing high warfarin dose requirements in a rare patient. Carol, this sounds like GM9, doesn't it? I mean, the challenge to us is now to figure out the functional genomics of each one of these. And as, of course, the warfarin lesson has been that there have been randomized trials, I'm not going to get into the details. They are very important. The details that are important are what the ancestry is, what the genotyping is, what the endpoint is, and you know that the US trial showed no difference when genomics were added to a clinical algorithm. The European trial showed a difference when genomics were added to a standard warfarin dosing regimen. One of the things that all those trials were underfunded. They were underfunded and underpowered to do is look at important endpoints like bleeding. So this is the US, the COAG trial and the bleeding events. There were very few, but the trend, interestingly, is in the right direction. At Vanderbilt, we used our EMR to look at a very, to continuously ascertain a large number of patients, around 500 patients with bleeding events that led to admission to hospital and 500 controls. And what we found was that CYP2C9 star 3 is a predictor of that event, not a, not a, with an odds ratio of 1.7, so not a huge effect, but an effect nevertheless. And Dave Veenstra has very similar data, but he landed on CYP4F2 and Josh Peterson and Dave are going to put the data together, along with Mike Steen, to figure out, and many others in this room actually, to figure out what the right, what the right predictor is. And then GIFT was announced about two months ago. I asked, I asked Brian Gage for the slides and he said he would ask the steering committee and the steering committee is still deliberating. So this is what I know from the, from the American College of Cardiology website, 1650 patients who are randomized, who are going to get Warfarin randomized to a pharmacogenetic algorithm or a clinical algorithm with sub-randomization to various INR targets. There's a composite primary endpoint, which is listed here, which I think is very important, because genetics beat, beat the conventional therapy. This is the first time that genetics in Warfarin have actually shown this big effect in the United States. This is a largely Caucasian trial. The African-American alleles are not included. And there are sub, I think it was major bleeding in the INR greater than four, but Julie may know better. I can't remember, Julie, which, which are the components of the component composite endpoint, but you're not allowed to look at components anyway. So, but this is, this is calming. This is in review and, and I'm not sure how this will affect Warfarin use because Warfarin, I think, is going away, but I think that there's an important lesson here. This is a really interesting, there's a lesson in this study, and I think it's important to say it. So, azathioprine and TPMT, this is a study done in, largely in Europe, randomizing patients with IBD who are going to get inflammatory bowel disease, who are going to get azathioprine to a conventional therapy or an intervention arm. The intervention arm requires genotyping if you have, if you're intermediate or if you're homozygous, now the doses are reduced or reduced drastically and, and then followed. So, one question is, when you reduce the dose of azathioprine, do you actually change the outcome in inflammatory bowel disease? And they have lots and lots and lots of metrics in this study that show that there's no difference in the disease course when they adjust the dose. What's interesting is, when they look at the percentage of patients with hematologic adverse drug events, there's no difference. Whether you, whether you use genetic guided therapy or not, so you'd say, okay, I don't understand why that is, but there is no difference. But when you then look, not across all the patients, but only the ones with the TPMT variants, then there's an enormous difference. Then the number of people who get adverse hematologic effects is very, very small, almost zero in the, in the intervention arm and about 22% in the conventional arm. So, this again highlights the idea that if we're gonna do pharmacogenetics in this way, the benefits are there, but the benefits are there in the people who have the variants, and that may be a small subset. I don't wanna dwell on this. It's a prevent, it's a terrible adverse drug reaction. It's terrible and predictable. Simon Malal and Elizabeth Phillips showed over 10 years ago that there's this huge odds ratio for a back of your related SJS, and you know that there's a trial called Predict that looked at a genetically guided therapy versus conventional therapy. You're kidding. Conventional therapy in the initiation of a back of ear, and basically you could get rid of all a back of ear related skin rash, which is this line down here by doing genetic testing, and this has become part of therapy. The same story holds true for carbamazepine and HLA B-1502. This is the study in nature that described that in 2004, and 1502 is interesting because it's highly prevalent in Southeast Asia, and not so prevalent across the world. There are other alleles that are prevalent in other parts of the world that transduce carbamazepine induced Stevens-Johnson syndrome, but again, it depends on where you are and who you are. There is this interesting Hong Kong carbamazepine experience. They saw the data. They put in a policy that you had to do genetic testing. So what happens when you do that? New prescriptions for carbamazepine just fall. People just don't do the genetic testing. Instead what they do, and the incidents of carbamazepine induced SJS falls, instead what they do is they increase the use of phenitine, a drug that also has SJS associated with it, although we don't know exactly what the right mechanism is for that drug. So the total number of SJS cases stays stable. And the moral is not that you don't do genetic testing. The moral is you have to do genetic testing and an educational program and tell people what to do with the results, not just sort of say don't use carbamazepine because it's a pain in the rear so people stop. So I can't talk about implementing without showing this slide. I've been showing this slide for probably 17 years now. And I think it says a lot about the expectations of the public when we see cartoons like this in the New Yorker. And this is what Francis Collins said when he became director of NIH about pharmacogenetics. And many of you have seen this slide before. Basically, what Francis says is if everyone's DNA sequence is already in their medical record is simply a click in the mouse to find out all the information. So we all, I think many of us in this room have drunk this Kool-Aid and agree with this, but it's not simple as we're, and that's the reason we're here. And it's probably not as inexpensive as Francis thought it was going to be. So I told you I'd say something about Sotolol. This is an 82 year old male in Sotolol for atrial fibrillation, develops renal failure and no dose adjustment is made. And he develops this really interesting long QT related polymorphic tachycardia due to Sotolol toxicity. And if you look at the label for Sotolol, it says adjust the dose when renal failure occurs. And you would be an idiot if you were a practitioner and didn't follow that advice because what would happen is what was shown in the previous slide is there are the data that actually support that, the answer is no. But this is the way we practice medicine. So is this a case of genetic exceptionalism or a case of Sotolol exceptionalism? I don't know, but I think that we have to sort of think about what kind of evidence do we want, do we need to have before we implement? And that's one of the reasons we're here. Clopidogrel was approved in 1998, was known to be a pro drug. The bioactivation pathway was not defined, but it was approved anyway. In 2006, the pathway was defined. And what was interesting about this study is a small CRC related study. And they studied heterozygates for the loss of function allele CYP2C9-2, and what they show is that this is platelet function on the y-axis. At baseline, there's no genotype effect on platelet function. After a week of clopidogrel, you have reduced platelet function in people with a wild type allele, but with tremendous variability across the population, across this small CRC population. And in the star twos, there is a small effect, there's no effect on average, but there's an effect in some people. That translates into retrospective looks at a difference in the incidence of clopidogrel related events. If you carry a variant, you have less good clopidogrel outcomes with higher incidences of adverse events compared to. If you don't carry the common alleles, we don't know anything about the rare alleles. At Vanderbilt, we have the Predict program, and the Predict program has been genotyping. It's genotyped about 15,000 people. And what we find is that 2% of our population are homozygates, so those are high risk for not responding to clopidogrel. About 20% are heterozygates. High risk for not responding to ordinary doses may need higher doses. So, if you're a cardiologist, you're looking at a difference between 12% and 8%. You're gonna genotype a lot of people to chase down a little bit of an event, and you sort of say to yourself, why do I bother to do this? If you're a geneticist, you say, well, you're doing it to chase this group here and perhaps some of this group here. Sarah van Driesen, our group, looked at five drug gene pairs. Clopidogrel is one. There's a group at high risk, a group at moderate risk, and for each of the drug gene pairs, we define a group at high risk and moderate risk. The important point is that 5% of the population is at high risk for something, and these are data that have been repeated worldwide. And 91% in our hands, just for five drug gene pairs, are at moderate risk for something. You just don't know what the drug is, and you don't know whether they're going to get that drug. So that's the argument for the idea of large preemptive pharmacogenetics, not across a single drug, but across large numbers of drugs, and we're doing that in Predict. Mary's doing that at St. Jude. There's a program at Mayo. There are others in this room, I'm sure, and I apologize that I haven't included them. We did that as part of the Emerge PGX project that Laura will talk about, and there's a large project in Europe, the UPGX project that is also doing that, and there's interesting design challenges in those studies. So this is my last slide. So what do we need? We need to not lose sight of the fact that we don't know everything there is to know about what we might want to implement, and we have to stay grounded in the basic biology and the basic science of variable drug actions. We do need methods to identify, accumulate, and study those outliers, so I'm fond of saying, if you want to study a group of people where an event occurs 0.001% of the time, you have to start with a very, very large denominator. So that's the argument for doing these kinds of things within Emerge or within the All of Us cohort or within other very, very large cohorts to capture the small number of outliers that become interesting, whether that's through passive methods or through active methods, like the kinds of networks that Elizabeth Phillips is trying to put together remains to be seen. I don't think there's one answer. I think there are many answers. We need good genetic tests. CYP2D6 is not easy. There are people in this room who know a lot about that and it's not a simple test, and we will need functional genomics because there are tons and tons of variants that remain to be defined. I had this thing that we need guidelines on how to use drugs, but I put that in sort of gray letters just to acknowledge the fact that Mary and others in C-PIC have done an enormous amount of work to try to actually answer the question, what would you do with the result once you got it? And that is the go-to place right now. And I think that's still a need. I'm not trying to take Mary's funding away, God knows. But I think it is a big step in the right direction. We will need IT infrastructure to do this right. We will need data on what works and what doesn't work, whether it's panel testing or preemptive testing. I'm an advocate of preemptive testing if it can be done. We obviously need education. We need data on outcomes. There are people in this room who are desperate for outcomes data because if they don't have outcomes data, it's hard to go to their stockholders and stakeholders and say that we want to support this. And there are lots and lots of partners. There are patients that are partners. There are payers that are partners. There are users that are partners. And this is a worldwide effort. This is not something that is gonna happen only in the United States for all the reasons that I've talked about. So I hope that sort of sets the tone for what we will talk about. I hope that's what you wanted to hear about. And I'm happy to take questions if there's time. Great, thank you very much, Dan. Yeah, I think we do have a few minutes for questions. You ended early, which is. No, I ended on time. I ended on time. You ended on time. I thought these numbers being flashed in front of me. Well, thank you for doing so. Questions for Dan? Okay, Howard. Dan, I wonder if you'd come in on the technology, thinking back to Angela Anderson and obviously that's a very complex case, but the turnaround time for results, even if that hospital or that clinic, that Dock in the Box, I can't remember the name of the one you went to, but had the technology, it might have been two weeks before the results came back to say what her HLA status was in order to make a decision. Where are things going in terms of trying to change that because that's holding things up as much or more than some of the other aspects? I have my biases and I wish I had included this slide. Josh has seen this slide and a couple of you have, but not many. I used to show, so one idea is the woman has her genetic information. So first of all, the answer to the question is you have to have your genetic data done beforehand and you have to have it accessible. So is that on a piece of paper that a woman has to a pharmacist? No. Is it on a smart Christ to go around talking about this stuff and I'd have something called the international gene card and I have a picture of myself with a little dot of DNA and I think that was very cool, but I think the way it will work is that you'll have a QR code on your phone and you'll have a reader in the pharmacy or in any podunk hospital in the country or in the world. You put the QR code under that, it accesses a website that's secure and it says, oh, by the way, this person has X. And it's only if it's been done beforehand, done, validated, high quality data, stored against the day that it needs to be used. So I think that some way of sticking stuff for lack of a better term in the cloud and having it accessible on the fly to healthcare systems, big or small, rural or urban, is the way that it may play out. And I'd be interested, I mean, I'm not married to that concept but I think that that's something like that is the way it's going to work. To do proper HLA typing, to do proper CYP2D6 typing, I know there are people, some of them in this room who have experience with whole genome sequencing in neonates but whether they do whole genome sequencing in neonates that includes CYP2D6, which is a tough nut to crack or HLA, it's awkward that the two genes we're most interested in are the ones that are among the hardest to sort of annotate properly in the genome. So I think that has to be done preemptively. And I look forward to the day that people will have DNA sequence and have it accessible with the proper point of care decision support. Rex, oh, sorry, where was I? Somebody said, yeah. Question over here, good morning, Dr. Melissa Clark. I was curious in your predict trial and the preemptive pharmacogenomics, how are you accounting for in your design the ability to translate any positive results that you have to community-based physicians given that there's that large learning curve and the need for education that you mentioned before. So I'm not the right person to answer that question. I will let either Josh, Danny or Michael over there answer that question because our IGNITE project is trying to do exactly that, trying to sort of crack that nut and I don't think it's a simple nut to crack, no offense, Michael. So one of you should answer that question. Well, I think it's probably early for us to really answer that question, although I think we can point to other efforts that you can look at national efforts, for instance, in southeastern Asia countries that have looked at screening for SJS-TEN and efforts there have shown differences with what's happened by sort of national awareness campaigns. I think in our efforts, we saw differences, Vanderbilt's not a community hospital, but it is interesting that without a lot of ground education to providers, people do sort of seamlessly follow what the decision support recommends and alter their prescribing based on our advice and predict. And then community experiments are ongoing. Okay, so that would be clinical decision support integrated into their EMR. So we do want to leave some time after some bonus presentation for questions, so maybe we'll do one more from Rex and then we'll go on. So this may be not something you can answer in a short answer, but obviously the medical outcomes of even saving one life is really significant, but to think about the economic outcomes, if we actually calculated the elimination of SJS cases, we calculated the known cases that we have, do we have any estimates of what the economic benefit to the healthcare system might be? I'm not aware of that, there may be others in this room who are the, but I think it's an important calculus because putting in place a large program like Predict or like UPGX or like Mary's program at St. Jude is not cheap, but when you start to think about, the calculus of human life is, that's difficult, but when you start to think about the costs of taking care of critically ill patients for days and days or weeks and weeks, that's also incredibly not cheap. And so the calculus is in there somewhere, maybe John Wilson at Optum can figure that out for us, John, but I will hear from you later, I'm sure, but that's the, and Mark has an answer. Right, I always have an answer. So the- No comment. So the answer is that there hasn't been a comprehensive economic look at this at present, but there have been a number of cost effectiveness analyses that have been done on individual drug gene pair. So for the back of ear, for carbamazepine, for warfarin, clopidogrel, et cetera, we have those. Now my understanding is that Josh Peterson at Vanderbilt is actually working on a comprehensive pharmacogenetic economic model to try and get an answer to the question that you're doing, but to my knowledge, that is not, the results from that are not quite yet available, but it is being worked on because I think that is an important question. So Mark, I had a conversation with Josh about five years, Josh Peterson about five years ago, and I said, you know, it ought to be possible to build a Markov model and you attach, to every transition you attach a probability and then a cost, and then you stick it in an Excel spreadsheet and you just push a button and say, well, if the cost are this, then the outcome is that. And I said, it feels like an afternoon's worth of work, Josh, and five years later, and one very large NHLBI grant later coming soon. But I think that's an important point. Great, thank you very much, Dan, and we may have a few minutes after, afterward before the break for the discussion. So it's my pleasure to introduce my colleague, Simone Evolpi, who will be reviewing NIH-supported PGX research to sort of help us identify what gaps there are and where we need to go.