Pharmacogenomics - Howard mcleod - 2012

Dr. Tyra Wolfsberg: Good morning everyone and welcome to today's lecture, the ninth in this Current Topics Series. Our speaker this morning is Dr. Howard McLeod, who's the director of the Institute for Pharmacogenomics and Individualized Therapy at the University of North Carolina. Dr. McLeod is a leader in the field of pharmacogenomics, a relatively new discipline that explores how genetic information influences our response to drugs. Dr. McLeod's institute is working to create targeted therapies and treatment options for individual patients suffering from a wide range of conditions, initially focusing on cancer therapy but later branching out into other medical conditions as well. His research has already had several effects on FDA policies. For example, he and others have shown that SNPs and other genetic variants play a role in patients' responses to Warfarin, a blood thinner that's prescribed to more than two million patients in the U.S. every year. Based on these analyses, the FDA has issued new dosing guidelines based on the genotyping of two different genes. During this morning's lecture, Dr. McLeod will be expanding on this story and also telling us about other developments in the field of pharmacogenomics. Please join me in welcoming Dr. McLeod to NIH. [applause] Dr. Howard McLeod: Well, thank you. It's a pleasure to be back here and to talk about this subject. It's also great to see many of the folks in the audience who know a lot about this subject and could easily have given this same lecture, except things sound better when it's someone outside your own institution. So, some of the experts in the room will be following this. If you would like to look at their work, they're doing excellent stuff as well. So, I want to talk a little bit about pharmacogenomics but framed not so much in a genetics and genomics context, but in a clinical practice context. And we're going to dig in to some, I was going to say basic research, some fundamental research, but with the eye on where this is all going. And at the very end I'll come back to some of the practicalities that we're completely ignoring, or almost completely ignoring, that are keeping us from driving things into practice. You know, often we say, "Well, why hasn't this science led to a changes in practice in a more rapid fashion?" Often, it's because we haven't planned for it to do so, and if you don't do the kind of studies that will drive things forward in practice, why in the world would you expect it to ever be useful? And so we need to turn the glass around and look at ourselves a little more closely, speaking to myself, in terms of the kinds of science we're doing; making sure we do good fundamental science, but making sure that it can lead to some improvements in health. Now, I like to start off almost every lecture I give with the, well I guess with the disclosures. Two consulting disclosures to disclose, nothing directly related to the content of the talk but in the general field of personalized medicine. I like to start off with this quote, and that is "A surgeon who uses the wrong side of the scalpel cuts their own fingers and not the patient; if the same applied to drugs, they would have been investigated very carefully a long time ago." This quote is supposedly from 1849, I don't read this particular journal. But it's very true today that none of the drugs that are approved by the FDA have a known mechanism of action. Now, you remember when you took exams that there was a mechanism of action for every one of these drugs. You know, ***-2 inhibitors hit ***-2. Topoisomerase one inhibitors hit topoisomerase one. But what else do they hit? ***-2 inhibitors have activity in ***-2 knock out mice. It's kind of a little bit of a clue that there may be something else going on. And so, we know something about drugs. We know enough about drugs to make them useful for some people, but we really don't know a lot about drugs. We don't approach drugs in a way that allows us to understand completely how they work. And so it's no surprise that when you give a group of people a drug, your favorite drug for your favorite disease, that there will be some folks that will get great benefit, some that will get no benefit, some that will get no benefit and severe toxicity and whatever other iteration you want to put in there. And so we have a lot to learn in terms of drugs. We've kind of ignored them because of the convenient labels we've put on them and there's a lot of work still, going forward. Now the clinical problem in 2012 is a wonderful problem to have. I realize this gets put out on YouTube and lots of different folks watch it. But in the audience here, there's a lot of folks here that are rather young in age, or at least you look like you're young. Take that as a compliment. And you won't remember that it didn't used to be the way it is now in terms of drug therapy for most common diseases. We have a wonderful problem, and that is there are multiple active therapies for most diseases. So cancer, where I spend a lot of my energy, it used to be that, for about 40 years, the treatment for colon cancer was 5-fluorouracil. The big question was bolus versus infusion administration. Now there are six set of toxics, two biologics, a lot more coming, in terms of the treatment of just that one disease. It wasn't that long ago that treatment of kidney cancer was limited to one therapy, placebo, also called IL-2. Yeah I know, some of you in the audience work on that. But now there are many different tyrosine-kinase inhibitors with active influence on survival in that terrible disease. And so many diseases, not just in cancer, in any areas, there are a lot of different therapies. If you take a common illness like high blood pressure, hypertension, there are over 100 FDA-approved drugs and drug combinations for the treatment of that one area. So what do you pick? If a patient's sitting in front of you and they have high blood pressure, what do you pick to treat them? You pick the one you know how to spell. It's almost like that, it's not quite, there's a little bit more that goes into it, but the concept that we treat with what we're familiar with or what we think might work instead of what we know might work, is one that we've come to accept and we need to start weaning ourselves away as we try to pick the best therapy. The medical decision most often nowadays is choosing from amongst equal options, not awesome therapy and terrible therapy. And so, amongst equals, you need just a little bit of data to influence those choices because either way it's a good choice. And so we need to start reframing the way we think about the clinical problem because it's not an all or none approach. It's a buffet approach. If you go to whatever your favorite local buffet is, there's 20 great entrees, and if one of them is missing that day, there are 19 great entrees. There are a lot of options for the treatment of most diseases, but yet we still approach it in a very old-fashioned way. The variation in response is also something we've come to expect. With the exception of bone and mineral disease and bacterial infection, the average response to the therapies that we use across all the other therapeutic areas is around 50 percent. Now I know you can find some exception somewhere to that, but looking across the general areas. And that means that the first time we get it right half the time. The second time we might get it right half the time, so now we're up to 75 percent of patients getting options. The third time we get another chunk of that. And so often we have to have iterations of the first therapy, second therapy, third therapy, to try to treat the disease of interest. And that variation of response is something that we expect, we anticipate, and we do nothing about except use another iteration. So trying to again reframe how we think about treating disease is important. In some areas, like cancer and many viral diseases, getting it wrong the first time means you have a very poor chance the second time. And so, second line therapy -- first line therapy for advanced colon cancer you get about a 50 percent response rate, second line therapy, it's about a 10 percent response rate. It's not just a, you know, oops I'll try it again; you lose something in that process. And so we need to be more aggressive in terms of not tolerating variation in response, but rather trying to push forward. And if you look at the amount of science that's devoted to this problem, it's a very small amount compared to the amount of science that's devoted to the basic biology of disease, et cetera. And so we need to at least consider that. And I'm not saying let's all get more grants. It's a clinical problem we need to push on. The other issue is unpredictable toxicity. Toxicity is also something, you know, from that slide I showed you about the surgeon cutting themselves and not the patient, toxicity is something that happens to the patient and not the prescriber. If you look at toxicity, when I go to the data centers for the studies that I'm sharing in the NCI Cooperative Group, when I go to the data center I just ask them for the extreme toxicities. Toxicity in the NCI system is rated from zero, meaning none, up to five, meaning the patient was killed by the toxicity. And usually, thankfully, there's not a lot of grade five. We'll ask for just the grade three or grade four toxicity; just the severe stuff. Don't bother me with the trivial grade one-two toxicities. Well, I can tell you, if I was receiving chemotherapy and had grade one diarrhea right now, I'd be talking to you from the little room across the hall there that says "M-E-N" on the door. Hopefully by audio and not video. It is a -- grade one toxicity means nothing to investigators in terms of its analysis, but it means a lot to the patients. My father was treated for cancer about a year ago and I learned a lot of lessons from that experience. One of them is, grade zero toxicity can matter. And I just told you grade zero means no toxicity. But he wasn't getting out to walking in the mall and he wasn't going and playing with his grandkids because he was concerned he might get some diarrhea from the chemotherapy. He was told it was possible. He didn't want to be caught out. And so the idea of toxicity was keeping him from activities of daily living, from trying to do things that would be good for his health, both physically and mentally. And so, toxicity matters a lot to the patient. It is the thing that keeps them from following our direction. It is the thing that influences their outcome more than most other aspects. And so we have to start reframing the way we look at that and we're seeing that now. The NCI recently issued some studies on -- basic fundamental studies around peripheral neuropathy from chemotherapy, and we'll come back to that point a few slides. But the idea of trying to look at animal models and other mechanistic studies around nerve toxicity from chemo is an important step towards understanding what is a very debilitating problem in this context. So toxicity does matter even if us academics have overlooked it for decades. The other issue is something we don't like to think about in academics, and that is that these drugs actually cost money. I mean, who knew? We just give them and see what happens. Well, it starts becoming a really important issue, not just because of the economy in its current state, but we're now getting therapies that are highly active in a very small fraction of patients. And so we look at therapies that, if you give them to all patients with non-small cell lung cancer, you'll get about a 10 percent response rate from those patients. And spending money on a therapy that costs $20,000 a month for a one in 10 return is not a great return on investment. And you say, "Well, I've read about studies that looked at tumor markers and tried to use those to guide who gets the individual therapy." And that is a great step forward. It enriches it to the point where we now get 30 to 40 percent response rates in most of the real world studies. In clinical trials we were getting higher than that. We look at the real data; it's not quite as high. And so we're still getting to the point where even with enrichment with tumor markers, just the ones we looked at so far, we're still spending a lot of money for less than a coin flip's worth of odds of data of achieving a successful response. And so we need to be doing better there, partly because we have a lot of patients in this country that are choosing not to get modern therapy because of the low return on investment. If you want someone to mortgage their house because they have good insurance but not great insurance, then they better be getting more than just a coin flip of return of benefit. And that's often the situation patients are placed in; that they have to spend a lot of money in order to get modern therapy with not a lot of direct evidence that they will benefit from this therapy. And so this is an area that we've, with exception to some AHRQ work, that we really try to completely ignore. I want to find out what shrinks tumors and what prevents nerves from frying. I don't really want to dabble in all this economic mumbo jumbo, but if we don't, then we will not be helping patients. All we will do is get great screen savers from our PET scans and a few less people using walkers, which is a very important thing, but we will not be able to have the wholesale change if we ignore that part of it. And I know that's a terrible thing to say because that means I have to start doing it myself. But the idea that we can just approach personalized medicine from the drug target or the drug metabolism gene is insufficient. Now, I don't mean that all of you have to suddenly stop doing phenomenal biology and extraordinary pharmacology and become health economists. But rather what I mean is, as we push this forward through the trial system, we need to be taking that into account so we can look at studies that are powered to demonstrate good return on investment. You know, we talked about risk-benefit ratio, but all we look at is benefit. We look at tumor shrinkage and that's it. We don't really look at a true risk-benefit ratio, even from a toxicity standpoint, much less an economic standpoint. So there's a lot of work that we're going to do and I know that that's painful because if you ever spent much time with health economists, it is an hour lost. But we have to learn how to converse with these other elements and I'll come back to that point shortly. Now, when we talk about pharmacogenetics or pharmacogenomics, I have -- this slide happens to say pharmacogenetics, I should have gone in there and changed it, but I use the terms interchangeably, and that is really the interaction between the genome and drug therapy. And pharmacogenetics was the term that had been used for a long time. It was coined back in the 1950s and really has been used more for small, more focused studies. So studies where you might look at one single nucleotide polymorphism and its interaction with therapy as opposed to looking at the entire genome. So pharmacogenomics is more the entire genome type of approach. Some have said that pharmacogenetics are for those that are over 40 and pharmacogenomics for those who are under 40 years of age. There are other ways of cutting it. But the bottom line is that either way you say it, it's trying to look at the influence of genomics on drug therapy. And we see, if you look at the NIH portfolio of studies that are currently funded, we see a lot of work in the discovery aspects. And so you look at patients or just subjects and sequence them, look at the frequency of variants in genes known to be important for drug therapy and find that there might be a certain polymorphism, this happens to be a pyrogram from a pyrosequencer showing some sort of variant that's there. And so there's a lot of discovery type work that is just cataloging the variants that are out there. We have studies looking at phenotype, this might be blood level, or some other pharmacodynamic effect, or looking at the variation that's there, trying to explain that variation with a genetic analysis. Other studies are trying to find out why that one-in-100 patient is having some extraordinary event. So maybe you treat your group of patients, someone gets Stevens-Johnson syndrome, or some other severe toxicity, you go and try and figure out, well why did that person get that when everyone else did not? And so a lot of work, very exciting work happening in this space. We see a lot now in the pharmaceutical industry and some in academia of studies that are using pharmacogenetics as an inclusion or exclusion trial. Most of the early pharma trials are now doing this. It's starting to get into academics a little bit where we might do a study initially only in the extensive metabolizers, excluding the poor metabolizers, see where the drug works in this context where it's optimized for benefit. If it does work here, then look at comers and try and see what it would be in a general population. So it may end up being a drug label that would be only available to a certain group of patients defined by genetics, or it may be just some initial data for proof of principle, you know, why invest $100 million on a clinical trial when you can do -- spend several million to do a smaller focused study and see what your odds are of benefit. And so we're seeing that happen a lot in drug trials, mainly those sponsored by the pharmaceutical industry. And then clinical practice is something that is now being pushed on, and we'll come back to that towards the end. But we're starting to see a lot of efforts where hospitals, health systems are starting to integrate pharmacogenetics into their routine practice setting. We're starting to see groups like NHGRI put out requests for applications or other funding mechanisms around these kind of genomic medicine approaches. And so the idea that one can start driving this into practice and look at the process of doing that is now becoming activity. And there are other areas, as we'll come to, but a lot of different aspects come under the term pharmacogenetics or pharmacogenomics. Now when you look at the examples that are out there, and this is relatively up to date. There's at least one drug that's missing from this, but it's relatively up to date. This is an example of places where the FDA have changed the package insert labels in the dosing and administration section to reflect pharmacogenetic information. Now, the reason that I worded it that way is that if you look throughout the drug label, the prescribing recommendations that are -- that come -- you know, when you get a medicine, it's a little pad of paper you throw away. But in that, if you ever were to open that, I know that's crazy talk, but if you were ever to actually open that and read it, what happens is there's a bunch of different sections there, including dosing and administration section, clinical pharmacology section, all sorts of different sections. Now, if you look at through the entire package insert, there are just over 150 drugs now that have pharmacogenetics somewhere in the label, mainly the clinical pharmacology section. But there are smaller lists, the one I show you here, where the dosing and administration section has included pharmacogenetic data. And the reason that's important is that's the section that is supposed to be read by prescribers, and it is the one that's read by the folks that created the PDA programs that most clinicians use to help them, their peripheral brain. It is also read by the insurance companies in terms of reimbursement activities. And then of course it's also read by litigators. And so -- unfortunately, it's read by them. Who knew they could read, but there we go. And so we do see activities in those three areas of trying to push this forward. And so you see some examples where it's a cancer marker, a mutation or copy number change in the tumor itself and not in the normal tissue. And then a lot of examples of germ-line changes in areas like cancer, blood thinners, *** drugs, carbamazepines for seizures, manic depression, chronic pain, copidogrel for heart disease, interferon for hepatitis c is the main change, but also used in other areas. And then a number of different drugs used to prevent nausea and vomiting, anti-depressants, ADHD drugs, et cetera. So there are examples. It's just one slide, it's still a pretty big font, it's not exactly a huge list but we do have examples where genetics have made it into the package insert. Now, of these, there's only one example where it's malpractice not to do the test, and that's the example for the abacavir. Anybody that's managing *** that's using the drug abacavir and doesn't look at the HLA-B*5701 is just begging to be shut down. And so, that's the one area -- there's been prospective intervention studies published in New England Journal. If you're in the *** area and you miss that, you really should consider retiring. And so, that is the one example where it is malpractice. The rest of them are not mandatory. Now, the IL28B genotype for using interferon has become a very common test in a very short amount of time mainly because the new protease inhibitors use that data in terms of selecting the length of therapy that one would get. And so we see some tests like IL28B that were first published in -- just a few years ago and then rapidly made it into the package insert and into routine practice. The lifecycle from publication to practice was less than five years, so it's an extraordinary example. But these various examples we start seeing them in practice, but none of them is required with the exception, as I mentioned, of abacavir. But we are seeing examples. And we also are seeing some fascinating phenomena in terms of adoption. I'll come back to that point. Because we are now seeing centers, for example with clopidogrel used in acute coronary syndrome and other areas, we see some centers where every single patient gets genetically tested when they are on their way into the cath lab, and there are other centers that think it's not ready for primetime and don't do it on any patients. And so you see this really interesting phenomena at the start of this discipline where some places have just, you know, found the gospel and are believers, and other places are still, are not there yet, and so there's a lot of differences out there in practice in terms of what's happening. And the irony, because I didn't put in a slide for this. We've done some surveys of community practice versus academic practice. And we've seen that the diffusion of this, many of these, within two years after the data is out, we see right around 40 percent of community sites are using the testing, whereas it's about three percent of the academic sites. And so, don't focus so much on the number, but the idea that the academic sites are the slow adopters for personalized medicine has been a phenomenon that we've seen now with multiple different examples. And it's just been fascinating; it could be that the community folks are blindly adopting what should not be adopted. Or it could be that we're so interested in that next trial that we haven't bothered to thing about whether something should be applied and applied now. And so there's some work that needs to be done to make sure that academia and the community practice where most people are seen are more in sync, and certainly from an academic standpoint, I hate the idea that we're following and not leading, and we need to be doing better work in that area. Now, if you look at the applications of pharmacogenetics practically, they come down to a few areas. One is the explanation of a untoward event. So you have examples of genes where someone gets the drug and falls down, and you want to know why, and you can do a test to figure out why did that person have extreme toxicity from 5-fluorouracil for colon cancer and do a test to see whether this was the explanation for that. There are some which are required for insurance coverage. So if you are going to treat a patient with certain classes of drugs, epidermal growth factor receptor antagonist for example, you need to know the KRAS or EGFR depending on the tumor type before the insurance company will reimburse that drug. And so it's a required test that way. You have some where it'll identify low utility. So if you have colon cancer and you have a mutation in KRAS, and it's the right mutation, your chance of benefit from an epidermal growth factor receptor antagonist is really zero, you know, zero or less than one percent anyway. And so it's a very low utility and rules out that therapy for the patients. Dose selections, you can use genetics to dose Warfarin, to dose clopidogrel, as well as select a different drug in terms of therapy selection in that context, and so it's being applied in that way. And then in terms of preemptive prediction, so I mentioned the abacavir example. If you did this test -- so for example, at our *** center at University of North Carolina, every patient that comes in, every new patient with *** gets this test as a part of a panel of tests that are done from the start. Not because they are going to prescribe abacavir today, but they want the data preloaded so that when they are ready to prescribe it, they know whether the patient is going to have a risk of severe hypersensitivity reaction or whether they're going to be fine. And as most of you are aware, the New England Journal paper that looked at this found that by doing this testing, you can completely eliminate the risk of hypersensitivity reaction, not just reduce it. And so it's a way of making that drug either one that's useful, or one that should be avoided. In the same way one might approach a penicillin allergy or some other status like that. So those are some of the ways that we're seeing application happening out in practice. Now I'm going to walk through three different areas, one more fundamental nature in terms of some discovery approaches that need to be done or are being one. One around the types of validation that is needed in terms of making sure that we're finding something real. And then, lastly, the application. With application, not only talk about some of the ways we would apply things here in the U.S., but also talk a little bit about some of the global health efforts where we're trying to use genetics as a useful way of managing therapy in countries that don't even have electricity 24 hours a day, and may not even have clean water. Something that could be considered quite a frivolous exercise; hopefully I'll convince you it's not. So when we look at drug therapy -- okay, the arrows aren't appearing on my screen but they're up there. When we look at drugs, often we create these kinds of pictures and we look at it, and this is an [unintelligible] cancer drug; had this slide handy. You know, here's this drug and it goes into the cell and it's pumped out by active transport, and it's inactivated by these B450s in the liver and it's activated by these [unintelligible], so this metabolite, which is pumped out, which is inactivated, which hits this other target, kills down these death pathways. We're geniuses. I mean, look at that. And if I had a good graphic artist, I'd be even smarter. I mean, you see that kind of stuff and you're like, "Wow, we are so smart." Well here's the real plot. Especially in the area of pharmacodynamics, but even in the area of pharmacokinetics, we do Yogi Berra pharmacology. We know what we know, but we don't know what we don't know. It's one of those things where someone has looked at this protein because it's one of their favorites, but really hasn't stood back and said, "Well, what are the right genes?" in terms of the influence on this drug. And so what we see is a need to really step back and look at animal models, family studies, I'll talk a little bit about some of this in a slide or two, large population studies, and we'll come back to this point. Trying to understand what are the genes that are important, because as I mentioned at the start, we know something about drugs because they were designed to hit a certain target and indeed usually do hit that target. But then there's all sorts of other approaches that are out there. I mean, we look at the example of Sorafenib, a drug that was developed at a RAF inhibitor, hence the Sorafenib. Well it turns out it does hit RAF, but it's really a VEGF inhibitor, a vascular endothelial growth factor inhibitor, in terms of its mechanism of action. And so it's unfortunate that they found that out after they chose the name, because, you know, they would have called it something else than Sorafenib. But the concept that we know something but not everything is something we quickly forget as the drugs start being applied. We see approaches like the Collaborative Cross that some of you are familiar with where hundreds of new inbred strains have been developed and can now be tested for all sorts of different biology reasons, but also pharmacologic phenotypes, and so we see some exciting data coming out of that, people looking at drug effects in a mouse system but it now has a as much or a greater amount of genetic heterogeneity as is seen in people. And so this concept that one can go and phenotype a group of a large number of inbred strain where the genetics is already done and do that analysis rather rapidly is a very attractive approach. We're seeing some exciting data come from that work. Now, when we look at the -- sorry, I put another slide in here, but it ended up being so large I couldn't email it out. And this was the NHGRI GWAS page. For some reason the way that I saved it ended up being too big of a jpeg or too big of an image. But, if you look at the NHGRI's GWAS catalog, you know, it has all those different, at least as of a few days ago, there were 1,196 genome-wide association studies that were listed on that website. And so if you go through all of the different example that are there, among the almost 1,200, actually by today it probably is 1,200, but as of that day it was just under 1,200 different examples. We find is that there were 50 that were a drug-related phenotype. Some of them were hypersensitivity reactions, some of them were dose, some of them were toxicity from a drug. But looking, casting a wide net going through each of those examples, we found that there are -- about 4 percent are genome wide association studies. So the good news is there are some discovery approaches that are being done, but there are more than 50 studies in diabetes alone. There are way more than that. There are more studies than that for height. I mean, there are a lot of phenotypes that are interesting phenotypes, important phenotypes, where there have been more studies done than looking at the entire catalog of drug-related genome-wide association studies. So 50 is a good start, but one of the things I noticed, and you wonder, "Well, why don't we have more examples?" Well, we haven't tried. You know, at some point in time, we need to try. Now, only 10 of the 50 had more than 500 patients in the case category. And so if you know anything about genome-wide association studies, which if you showed up last week you do, because another Tarheel talked last week about that approach. We know that size matters. We need large studies in order to be able to have power to detect robust signals for a given phenotype. And yet only a small number, only 10 studies had a large enough -- have even 500 cases; and I didn't do a statistical analysis to see whether 500 was even enough to detect a phenotype. So the point is that very few attempts have been made. Fifteen of these found no significant hits at all, so whether that's under power or there just aren't any genome hits, just you remember, genetics is not the answer to everything. I didn't put this slide in this time, but I always like to remind myself that, you know, there's the old saying that if you have a hammer, everything looks like a nail. Well, we fall into that trap with genomics big time. We have a next-gen sequencer, therefore the answer is DNA. And I don't care whether all of the data, all of the evidence says that it's environment, diet, exercise, or whatever, I'm going to sequence the heck out of these patients and find all the variants and find all the answers. And I think, you know, part of the strategy needs to be, you know, what is known clinically, how do we add that in, is genetics likely to be the case and be useful? And we'll come back to that point in a few slides. Twenty-nine of the 50 studies had a replication cohort. So again, relatively few had what is now a normal part of doing genomic-wide analysis. And so I think what we're seeing, and I'm not trying to be an apologist for the field, but I think what we are seeing is some pretty weak attempts at doing discovery in the context of pharmacogenetics and we need to go and do some proper studies. You know, it's no surprise that we're not finding that much when we haven't tried, and when we have tried, we've done a relatively poor job because of using underpowered studies. And so it's time, and you'll hear a little bit about this, to do some properly-powered studies in that approach. Now, out of these 50 examples, eight of them got -- created data that contributed to changes in the FDA package insert. I mean, you could argue that, you know, eight out of 10, but all of them weren't in this category, but -- so from a hit rate standpoint in terms of implementation, changing regulatory evidence, it's been a phenomenal success because eight out of the 50 examples, and I can tell you that that is not the hit rate, the success rate that other disease areas have had. And part of that is that some of these studies have found the, what you'd term the low-hanging fruit. When you look at some of the hypersensitivity reactions, Stevens-Johnson Syndrome, which is an immune reaction to the drug which basically the cells, the immune cells start eating the drug, but also start eating the skin. And these patients have been admitted to the burn unit, there's about a 30 percent mortality, it's a bad thing to have. And so when you look at those, many of those studies have had odds ratios of over 1,000. One of them had an odds ratio 2,500 for the HOA marker that was associated with the effect. And they did bad calculations. They only needed nine cases to have statistical power to find that because it was so strong. And so some of these are cheating because they're just so powerful, the markers are so powerful that you almost couldn't do anything but trip over them and find them. But there are other examples where, like the IL28B example, where a genome-wide scan was done in the context of interferon therapy. I was in the room as a part of the analysis group, I was not an author on the paper but I was one of the external folks when that data was released, and there was nobody in that room that expected IL28B to be a hit in terms of interferon therapy. Now, everybody that left the room knew that of course that was going to be the hit and had a great explanation for it. But going into the room, no one had that as the hit. And so some of these examples have been true discovery that has led to changes in practice in a very rapid time. So there's a lot to be done. And I think pharmacogenomics offers a promising area in that the phenotypes are a little bit less complex in at least some the cases that have been found compared to something like height, or diabetes, or many of the other diseases that are such a mixture of gene environment polygenic effects, et cetera. Now one of the things that we started off doing, I'm not going to spend a lot of time with this because we published a lot already on this, is trying to ask the question, "Is a pharmacogenetic end point even heritable?" And it seems kind of crazy, but no one had really asked that question. So if you're going to spend a couple million dollars on a genome scan, wouldn't you like to know whether the odds are high or low that genetics is even involved at all? It kind of seems obvious now in retrospect, but the time, you know, we had the machine, we had the chips, we had to use them before they expired, we had the clinical cohort, we had the DNA from that clinical cohort, why would we not do a genome-wide association study or a next-gen sequencing study, or some other approach? Well the reason why is that it might be stupid. And we need to ask the question, "Is it likely that genomics is an important factor?" Now, I'm sure that I have some genetic influence with the adipose that I have around there. But I can tell you that genetics is not the main factor to why it's there. It's the lack of exercise and the high intake that are the problem. And so I can scan my genome all I want, but my answer is going to be in my hand as I put it into my mouth, and in my shoes, not in my genes. And so we need to be looking a little more carefully about when do we do this. And so we took the example of the cell lines that many of you are familiar with from the Human Genome Project, from other gene mapping studies that have been done. These are part of the HapMap project, part of the 1000 Genomes project, and so you can get these multi-generation families that, where cell lines are available. And one can go and do high throughput studies. I show you here a 96 well plate, you have increasing drug concentrations, you have two different drugs on this plate, we do this in three to four well plates but it's a whole lot prettier on a 96 well plate, so I stuck that picture in for your visual pleasure. And with increasing killing, you get less pink, and you put this into a highly accurate florescence analysis and come up with kill curves. And so these are two separate cell lines where we have increasing drug concentration for a chemotherapy drug, Docetaxel, and you have viability on this axis here. So you see, and these are three separate replicates. So, each of these lines is in quadruplicate and then there are three separate experiments performed that are shown here. So each color is a different experiment, each line represents four replicates. And so you can see that some cells are killed very rapidly, and others, even in high concentration, don't even get down to a 50 percent kill rate. And so you can see a lot of variation amongst these cells. And so one can go in and take, for example, some of these families and look at a bunch of the FDA-approved drugs and come up with these sorts of plots where I'm showing you on the Y axis the heritability, so basically what is the inheritance of the cytotoxicity in these families, and a whole bunch of different drugs. And there's a publication you can go to if you actually want to read this X axis. What we have is corrected or uncorrected for cell growth. It didn't make a big difference in this context. And what we can see is that some of the drugs had about 60 percent heritability, very high degree of heritability, and there were others that were, I don't go down to very low, there were others that were basically at similar level as the controls. And so some of these drugs, it's not a really great idea, at least based on our data, to do a genome-wide association study, or next-gen study, or whatever, at least in this context, in cell line context, because we're able to show that the heritability is very similar to just the drug vehicle. There's not a lot of heritability there and I know that you have plenty of phenomena that have low heritability and yet have a genomic basis. There are genes that influence whether you have one arm or two. And yet, there's not a lot of variability among people. You know, most people, not everyone, but most people have two arms or two legs, even though genomics was, genetics was a part of influencing that. So heritability is not everything, but it is something. And so you can get this sort of data that gives you some hints at where we should go. Now the reason we did this in cell lines is that you can't do it in people. And so there are a lot of examples out there where you can't do the study, the drugs are a little bit too toxic to look at normal volunteers, the patients, even if you have familial cancer, or familial some other disease, it's not that everyone gets it on March 21. It's rather that everyone gets it age 30, or whatever it might be. And so you don't tend to have families receiving the same therapy at the same time. And so trying to look at heritability and those aspects is something you really can't do in vivo for drugs that are toxic, some of the antivirals, some of the chemo drugs. And so you need to do that. The other thing and I'll get to in a couple of slides, is that replication is critical in this field. And when a clinical trial is performed, especially in the area of cancer, typically one trial is performed. Now, if that trial is positive, it goes on to have another, you know, the winner from that trial then competes with the next best idea. And the winner from that competes with the next -- and that's the way the trial program goes. So you often don't have an easily available replication set. If you happen to use a clinical trial cohort for discovery, it may be five to seven years before you'll have a replication set that's available to you. And so we've been stepping back into the laboratory and looking at ex vivo systems like these cell line systems to try to help us do discovery. And what's shown also on this paper, but I didn't put the slides in there is that we can do discovery in terms of finding quantitative trade loci across the genome that are associated with the toxicity of this drug, including loci that are found in all of the family members, the chemical family members of these different chemo drugs. And so we can find some very brief produce able hits that are found in all of the anthracyclines or all the fluoropyrimidines, et cetera, and use those as a way of taking discovery from the laboratory into a clinical trial setting. One of the things we've also done, and this is a paper that's just been submitted, and I know this will go on YouTube. Hopefully reviewers will still be kind to us, I took a risk. But this paper has been submitted. We then went into an in vitro genome-wide association study. So we took cell lines from 563 unrelated individuals and we repeated that experiment. Now, if -- you can't read this but the top hit here, you can see over here, Temozolomide, a drug for brain -- for glioblastoma multiformity -- is the top hit in terms of heritability. So we looked at that drug in a collection of 563 unrelated individuals, in this case they were all Caucasians. We're also doing this study in a group from Taiwan and also in an African-American cohort that all have -- the Taiwanese set has 9,000 cell lines, the African-American set has just about 1,500 cell lines, and yes, the technicians that run this are extraordinary in their ability to complete this stuff in a reproduce able fashion. Thank goodness for robotics and bar coding is all I can say. So, when we did a genome-wide association study for Temozolomide, when you scan across the genome, I think you are probably familiar with this, you scan across the various chromosomes. Even without these green lines you can see a big peak that's scanning out here. There's only one peak that goes up above, and here's a blowup of that area here. Only one peak that goes above the 10 to the minus 8 statistical size, and you could argue that 563 is a small number, and it is, but there aren't a lot of large cell line collections out there, and the ones that are out there are almost EBB transformed B cells, lymphocytes. And so there is a restriction in terms of this approach. So we see this hit that's there. Well it turns out that this particular hit, the SNPs in here, are in a MGMT that had already been known from its biology to be involved in the repair of the alkylated DNA from Temozolomide. So using this system, the bad news is we've found something that was already known. The good news is we've found something that was already known and now have genetic variation there that was associated at least in vitro with cytotoxicity, something that had not been done before. These SNPs also were not only associated with viability, but also associated with gene expression in these same cell lines. And so we can now take this data and look at these SNPs in the context of clinical trial material to ask the question, are these variants predictive, these in vitro discovered variants predictive of either toxicity or advocacy in the context of patients with glioblastoma multiformity. Now, these drugs are one of the few active drugs for that disease. There are not a lot of alternates. But these drugs are not extraordinarily active. They're just active in that disease. And so if you can identify a patient who's going to have extreme toxicity, and you may want to put them on a promising new clinical trial as opposed to give them a drug that's going to decrease their quality of life over the remaining time that they have. And so there are decisions that can be made even though there aren't a lot of alternate therapies out there in this context, but the idea of using in vitro discovery approach is one that we're working on to try to understand what can we get out of the lab-based systems that will help us ask smarter questions in the clinical material because there's so little of it and it's so precious in that way. This sort of thing is now being completed for the other drugs that we've been working on, the replication's going on. We'll see what happens in terms of the follow up from that. There are a lot of other interesting discovery approaches also happening. I just mentioned that one because it's a little bit different from the usual approaches that have been taken for other diseases, for example, because of the differences in being able to administer drugs to only certain groups of patients. Now the second area I want to talk about is the validation of robust data sets, and if you look at cancer as an example, and I'm not trying to be exclusive with cancer, just have the better examples in this area. Here are the common types of cancer in terms of incident cases, and what we've done is integrated blood sampling and possible tumor sampling throughout the cooperative groups. Now the group that used to be called Cancer and Leukemia Group B, there's now been a merger and it's called the alliance. That group has been the most active. We started doing this back in 2002. But, if you've been involved in this setting, what happens is it takes two to three years to develop a clinical trial. It often can take five to seven years to conduct the clinical trial, and then three to five years to wait for follow-up and analyze the clinical trial. So you don't have to be a math major to figure out that you start these studies as an assistant professor and finishing them up as an emeritus professor, yeah almost that bad. It's something that you, you almost set them up for your kids because it takes so long. Some of these studies were started before my kids were born and are just being finished. So, while things are started a while ago, they just now start maturing. And so over the last few years, we're now getting studies where we have over 4,000 breast cancer patients from a prospective study looking at two different types of chemotherapy where the toxicity and efficacy were all collected in a uniform matter prospectively where the auditing was done both on imaging as well as the clinical data, so you have a robust phenotype with large numbers and can do discovery. And so I'll come back to that specific example in a few slides. But the idea of one doing studies based on what's in your institution's tumor bank is really something we need to get away from. And we've been a big culprit of this. There are studies where you have your favorite oncologist you have tea with on Thursdays and you say, you know, "Hey, you got any samples?" And they have 46 breast cancer samples and you do your favorite SNP and your favorite gene and those 46 samples, and both heterozygotes had toxicity and no one else did. Therefore everyone should be tested, the end. And there are steaming piles of literature along this line, including some of our own, where people have gone and done this sort of study. And it's great in terms of starting the field, getting people interested, seeing that there might be something worth chasing. But what happens is that very few of those studies have then gone on to do a well-powered study much less replication. And so the idea that you have 4,600 patients from 280 centers from across the U.S. and Canada where you have real-world variability built in and even though it's the context of a clinical trial, it gives you a much more powerful way of doing both discovery and validation. So in these trials, the good news is we now have over 40,000 samples worth of clinical trial material and growing by several hundred every month as we go forward. Now obviously that's not one study, that's all a different types of studies, many of them that are show on here. The bad news is that replication is very difficult. And so often what we can do is we can split a dataset and do self replication, you know, look at half the patients and try to replicate the other half, but that's not real replication. I mean, it is replication, but it's not a real validation. And so the idea that we have a separate dataset in which to discover in is a big problem. And so, there are none of the studies here where there is an easy replication set because this was the study looking at this approach. And if it's positive, we have another 12-year cycle of developing the study, conducting the study, waiting for follow-up and analysis to go on through. And so, you can see why we've gone to some of the in vitro and other model system type work. Now with these clinical trial samples, it's a very powerful way of trying to ask questions. So one example is in ovarian cancer. This is a study that was completed almost a decade ago and now has long follow-up in which we can then go and do discovery. And so, this is for the treatment of advanced ovarian cancer, the patients receive Carboplatin, it is a platinum drug and was identical in both arms. And then one of the two taxane cousins, either Docetaxil or Paclitaxel. And so not a big difference between these two arms, and indeed when you look at survival, pressure free survival or overall survival, no statistical difference. So this was a practice changing slash practice confirming study in that it really established that either of these two arms were the equal therapy and are the first line therapy for the treatment of ovarian cancer. So that's great, but if any of you have been to an ovarian cancer clinic, what you see is there's a lot of youngish women using walkers. And you think, "What a terrible bummer." Here is a disabled person who got ovarian cancer. I mean, what bad luck. Well unfortunately, most of these women were fine when they walked in but the chemotherapies fried their nerves. It's caused them so they can't feel their feet. They can't feel their fingers, they're pins and needles now. They can't button their blouse, they can't do a lot of things. And so the therapy is nearly killing the patient in order to kill the tumor. And so this concept that we have to nearly kill the patient to kill the tumor is one that many of us were taught during our training. The younger oncologists are taught much better now because they know differently with some of the new drugs. But the idea that that has to happen and that toxicity is just part of the price of doing business is something that we really need to approach. If we're going to do genetic discovery, we need to be tackling those kind of issues where it's just frankly not confirmed to be true. And so we ask the question, in this case, we did some initial broader discovery and then honed down into some candidates that fit in to informatic buckets; biology nerve function buckets, inherited neuropathy buckets, or drug action buckets, and did a separate -- so this was data from the literature, from other screens, from animal screens, some early clinical screens. And so we ended up with a custom Illumina chip based on variants from these areas. And so when we looked at it, out of the 1,261 SNPs that started off being valuable in terms of, you know, they weren't monomorphic, it wasn't failures in some other way, we looked at half of the patients, about 500 patients, and 69 of those SNPs came out as being positive. We then looked at these 69 SNPs in the remaining half of the patients and found that five of them were positive, but one was in the opposite direction. Only four were confirmed to be true. And when you look at these four SNPs, they're in some genes that make sense, you know, BCL2, oh cell death -- so it doesn't matter what you're working on. If you find something in a cell pathway it makes sense. And so I remember hearing a story from one of the Stanford guys during their initial discovery, during the study of using microarrays. And the way I remember the story is they came in, they did their analysis of breast cancer versus normal breast tissue, they come up with a group of genes that were differentially expressed, and then they spent all the afternoon going through why that gene made sense, and that gene made sense, and that gene. And then the next morning, the statisticians came by and said, "Actually, there was an error in some of the coding, here's the list of genes." So basically every gene in the genome makes sense in terms of why it's the right gene. And so I think what we should do is eliminate names. We should not be allowed to call genes by their names because it fools us into thinking there might actually be something real and we need to prove that they're real as opposed to saying, "Well, BCL2, cell deaths, causing the death of a nerve." So these genes can make some biologic sense, I mean, we selected genes that made biologic sense, so they're all going to make sense. And the odds ratios are somewhere between two and four, okay that's enough. And when you add them up, the population triple the risk is about 85 percent, okay that sounds good. And you add them up, you get higher and higher risk. And so that's great, the accumulation of these events causes a patient to be at higher risk of nerve toxicity, you can see how you might apply that. Except, no one is going to be willing to take less therapy or a different therapy if their chance of benefit is altered. No one's going to allow modulation of these genes if it's going to affect their chance of benefit. And so we ask the simple question, looking at the accumulation of these genetic events, do the people who have lots of these genetic lesions, as in three to four of them, have a different survival then those who do not? And so this is showing the people with zero to two of these variants or three to four of these variants. So it's basically these two groups versus the rest of these groups. And what we've found is there is no difference in either progressive free survival or overall survival. So what this suggests for really for the first time, is that the toxicity, the nerve toxicity that is being experienced by these women is not directly related to the survival benefit that they might get from this therapy. And it at least supports the notion that one may be able to modulate these targets or in some other way apply them to try to optimize therapy. Now there are alternate therapies from this that one could use. It's usually reserved for second line therapy. If someone either didn't want the toxicity or had, I mean, there are patients that have a platinum allergy, not very many thankfully, but it could be somebody that has that hypersensitivity reaction, you would go on to a different therapy. But, so there are alternate therapies, but the idea that one can now look at either using different dosing or different therapies based on genetic risk data is something we're now exploring. Now, this data is a first attempt, and we are now looking to validate it in a separate dataset. I'm not saying that this is the answer and these genes are important, but the concept of trying to dissect out toxicity and efficacy; trying to find markers that will predict extreme toxicity to have a real risk-benefit discussion is where a lot of pharmacogenomics is going right now. Using genetics to give quantitative decision making as opposed to looking back and saying, "Why did that person crash and burn when everyone else didn't?" Is really a lot of the effort that we're seeing happen in this particular area. Also, these datasets allow for discovery. And so we've completed now a number of genome-wide association studies where we had a pretty sizeable power of patients. Some of them a little bit on the smaller side, others quite large. We've done one for peripheral neuropathy and one for neutropenia in the context of breast cancer patients. We also have an initial next-generation sequencing study being done in the context of a prostate cancer chemotherapy trial. There's colorectal cancer, pancreas cancer, and many more being planned. And so the idea that one can go and have these data sets and do discovery is also an attractive approach. The other thing is that there is way more genomic capacity than there are quality phenotype datasets. And so in this example, some of these genome-wide association studies were hatched in the bar at Cold Spring Harbor where Yasuki Nakamura from the Rican, soon to be University of Chicago, was sitting there drinking something, and myself and Mark Ratain were having our cokes and he was complaining he had all this genomic capacity and no good phenotypes and we were complaining that we had all these good phenotypes and no genomic capacity. And so we've done a collaboration now where many of these genome scans have been done using the barter system. So they've done what would have cost us around $25,000,000 worth of genome-wide association studies as a collaboration. Some of the other approaches are being done with U.S. based genome centers. And so, there is a way of getting this done even -- some of these are funded to do. Some of these are being done without any funding at all. Because there's such an interest in trying to actually try to do these sorts of studies as opposed to just the small candidate approaches that have been done to date. So there's a lot more to be done, but the idea that one can get the datasets to do pharmacogenomic discovery is, it takes a lot of planning but we are at a time where this can start happening. Now I want to spend a little time on the last piece, that's actually applying the stuff. And I'm going to put -- first I did a little more controversial study but try to show some key points about application and then talk about some the ways we've applied across the globe. And application is something we all put in our grants, and we put in our papers, we put everywhere. But it's rare that we actually do it. And so there have been a number of different studies that have found interesting genetic markers and even replicated them. And then you see nothing else happen. And so we decided, well, let's go for it. And so one example is here with Tamoxifen and you know, when I trained, Tamoxifen was activated by these enzymes for 4-hydroxytamoxifen, which is an active blocker of the estrogen receptor and therefore is useful in breast cancer, estrogen-receptive positive breast cancer, the end. Well, a couple of years ago, Vered Stearns, who was at Georgetown at the time, had a patient that received Tamoxifen, was given the peri-menopausal syndrome that you get from that drug because you're blocking estrogen. She then needed some anti-depressants for depression, not related to the breast cancer, and gave the anti-depressants, and the hot flashes went away in just a day or two. Now, if it was me, I would have been ecstatic, because the hot flashes went away and in four to six weeks the depression's probably going to go away, and anybody with Scottish blood loves two for the price of one. [laughter] So, for a moment, well Vered's a little bit smarter than that. And so she said, "Wait a minute." And she and Dave Flockhart and some others dug into it and what they eventually found was the anti-depressants that were given were blocking the step. And so there were similar levels to here -- of this metabolite, but by blocking this step, the formation of this metabolite here, this endoxifen they called it, was not happening. It was much reduced, and at least in their studies, by giving the anti-depressants or not, they were able to block hot flashes and show very decreased levels of this metabolite. And so what's happened is the field of oncology has just stopped using certain anti-depressants in conjunction with Tamoxifen. Medco showed some great data where the prescribing of these drugs was relatively high, and then after the presentation in June , I think it was 2008 or 2009, something like that, where the data was presented at the American Society of Clinical Oncology meeting, just plummeted. The drugs are just not used together because of this worry you're blocking the activation, or at least one of the activation pathways of this drug. And then Matt Getz [spelled phonetically] and others showed this kind of data, where the extensive metabolizers for -- so I should mention one point. So, this enzyme here, many of you know that this enzyme is polymorphic. It has genetic polymorphism. About 10 percent of you in the room, a little bit less than 10 percent of you are completely missing this gene, either through deletion or through some major genetic variation. And some of you might already know that, either literally or phenotypically. So, those of you who have gone to the dentist and had a procedure, got some Tylenol with codeine or some hydrocodone for that, and it didn't work, you still had pain, it didn't work at all. You went back to the dentist and they called you a wimp. Well that drug's activated, those two drugs are activated by this enzyme. And so 10 percent of you cannot activate that drug at all and have a genetic explanation for why you're a wimp. The rest of you are just wimps and need to get over it. [laughter] But 10 percent of people just can't activate that, and it's the same with activating Tamoxifen at least in this data. And so you have the extensive metabolizers, who can activate fully, those who are missing the step, and the intermediate folks. And this is looking at years after start of a clinical trial, and this is a relapse-free survival on the Y axis. So every time you see a little bump here, somebody's breast cancer came back. And you can see that you don't have to have any different color to see that this group, it came back way more often than this group, or in this group. Now the extensive metabolizers, those that have, at least from a genetic basis, the ability to form this metabolite. There are still folks that have breast cancer coming back. So this isn't the answer to curing breast cancer, but there does seem to be, at least in these studies, some differences that are there. Now if you look at the literature, and this isn't every study, it was every one that I could find, you see a whole bunch of studies that found this same phenomena, that genetic variation in CYP2D6 was associated with a reoccurrence of breast cancer in patients treated with Tamoxifen. But there also are studies that did not find this basis, including two that came out, were released a couple weeks ago, early release, but came out in yesterday's Journal of National Cancer Institute. So, there are studies that, including, you know, Getz looked at a cohort in prevention that didn't find any relationship, even though in his advanced disease patients there was a relationship. And so when you look at all these together it appears that the patients that are on a monotherapy, on Tamoxifen as their only therapy, this is a more important phenomenon. For those that are getting the 20 milligram dose, not a higher dose, 10 to 40 is the FDA-approved dosing, although 20 is the normal U.S. dose. A higher dose, like was seen in the Wegman studies, reduced the effect. And the administration of additional chemotherapy seemed to change that so, as you can imagine, mopping up some of the cells that are not going to be sensitive to Tamoxifen. But we still don't know the exact answer for this. There's definitely a publication bias. It's a whole lot easier to publish a positive study than a negative study, and so hopefully we'll see more papers come out, because all I care about is whether it's useful or not. And if it ends up being not useful, that's fine. I'm not super happy about it, but it's fine, as long as we are definitive. And that's what been missing often, is these studies are decently powered, but not definitive in nature. But we see that sort of thing -- so, oncologists started testing and if you have an extensive metabolizer genotype, if you have a normal gene, you keep on the normal dose of Tamoxifen. If you have -- are a poor metabolizer, you go to some other type of therapy and in premenopausal women that's hard, because there aren't a lot of other therapies, post-menopausal there are aromatase inhibitors, et cetera. But this is -- this white line here, this intermediate group, is about 40 percent of patients. And so what we started doing, we started getting phone calls from oncologists saying "Hey, what do I do? You know, I know based on this type of data they have a worse outcome, but what do I do?" And we didn't have a thing to tell them. You know, we told them "Hey, I could make stuff up if you want, but I don't have a clue." And after a few calls you kind of get tired of saying you don't have a clue and so we decided to have a clue and do an intervention study. And so what we did is something very simple. We did -- we had it powered for an intermediate biomarker, active metabolite levels, and then we also collected, and are collecting, survival data. So this is a paper that came out late last year, the initial one, and we have -- we've now completed a larger study. I'll get to that in just a second. So the people that have the extensive metabolizer genotype, we kept them on the 20 milligram dose. Now, what we saw at the start of the study is that there was a statistical difference in active metabolite levels between the intermediate metabolizers and the extensive metabolizers. And then over time there was a slight drop, which was really irritating, because it makes the picture look bad, but that's what we found, and it could be adherence. It could be -- who knows what it is, but there's no statistical difference in the purple folks between now and then, but these were kept on the same dose of drug. These folks that had the intermediate genotype, we did something really simple. We did a blood sample, of course, and got the genotype. We did that with all people. We asked them to take two pills instead of one. So no major surgery. No major this. No major that. Just take two pills instead of one. Still within the FDA-approved dosing, because remember it's 10 to 40, so instead of 20 milligrams they got 40, so we didn't have to file a new IND or anything, and we just did that simple intervention study. And then four months later, which is two -- which is, because of the half-life of the drug, well after the study state time. What we found is by doing that simple intervention we were able to normalize blood levels. There was now no statistical difference between the two groups in terms of active metabolite levels. Now, this doesn't prove that normalizing blood levels equals better survival. I'm not telling you that this group here now has this outcome because of our intervention, but the concept of normalizing variability at the blood level is the concept that we use for every aspect of our FDA drug dosing. All of the changes that are recommended for organ dysfunction, for drug interactions, for age, for whatever your favorite phenotype is, all of those are based on blood levels, and trying to normalize blood levels, trying to reduce variability in blood levels. And so at the level of what we practice with, we've now achieved that. Whether this ends up being --improving survival, we shall see, but at least we've got to that far with this. Now, we ended up -- we thought we'd enroll 20 patients per year as a pilot study, and we had it powered to need just over 100 patients. And so we enrolled 119 patients in just a couple of months and we had people calling us and trying to get involved with the study. The very first patient we enrolled, talked to her, "Do you want to be on the study?" She was like, "Yeah, you know, why wouldn't I want to be on the study?" And so we go and do that, and we were in a pilot phase. We weren't telling anybody about it, we were just trying to make sure we had all the paperwork right, et cetera, in terms of data collection and all that. So the next patient comes in and before we could say anything she says, "I want to be on the DNA study." And I was really irritated, because that's the thing that I say, so she stole my thing. And I said, "Well, how did you know about it, because we're not advertising it or anything." She said, "Oh, a lady just came out and went around and told everybody in the waiting room about the study." Patients want personalized medicine. As a matter of fact, they think we do it now. This is kind of the embarrassing part. But they want to know that their characteristics are part of the decision in terms of how therapy is managed and they're more than willing. We ended up enrolling 500 patients in less than a year, and we enrolled them in 64 of the 100 North Carolina counties. So we can do all we want at our egghead academic centers, but you know the patients that get to us are often extraordinary: they're young, they're more fit; they're not your normal patients. But we went out and involved the community sites. You can see some of the stars of where they're located. We're now affecting patients from all around, in this case, the state of North Carolina. Our data is data that can be replicated in community-based sites, because it was derived from community-based sites. And so it was really encouraging us. We're doing a lot of this now. We're doing a lot of our straightforward intervention studies out in the community. There's no reason someone needs to drive all the way across the state to Chapel Hill just to do something that we could easily do in their own practice. Now it means simplifying the toxicity measurements, making it so they're more amenable to practice, but most of the data we collect in clinical trials is useless anyway. We take the NCI clinical trials entire catalog and do it on every single patient. In a drug that we have 30 years of experience, we kind of know what's going to happen and can hone it down to the things that are a little bit more relevant. And so this idea of going out and doing genetic intervention studies out in the community is one that I would encourage you to get involved with, because, you know, you have it here. You know, people that come to this building, come to the data -- the clinical trial center, are extraordinary in many ways, but they are not normal in terms of what the average patients look like for any of the diseases out in the community. And so there's a lot that needs to be done, and I think we've -- we've really -- I would say, what's a nice way of saying lazy? We've been very convenient in the way we've recruited. We've recruited patients that come to us because it's easier. If we go out and get patients in situ it's a lot harder. I can tell you, it's a lot harder to go down to these various sites. This isn't to scale. It's a lot more than -- a long drive to go to these places. And so we go, and we have to teach them. At one of the sites we had to buy them a centrifuge, because they didn't have one. You know, there's -- there are things you have to do. They're a lot more hassle, but the end result is we have examples that we can translate. Now, our ideal is we get to a point where a patient before they ever get therapy, we understand at least some of their toxicity risk, disease risk, infection risk, supportive care issues prior to ever giving therapy. And we're a long way from that, but that's the ultimate goal of where we're heading, and it's not so that we can use new science. It's so that we can do management in a more efficient manner. And I think a lot of us get excited about new science, but at some point in time we need to turn it and say does this new science matter and can we make it so it's boring? Because when it comes down to it, the things that we use on a regular basis that are known to benefit people are no longer exciting. They're boring. And our goal should be that all of our research should eventually become boring, because it is routine. It is normal. It is something that is used all the time and therefore is boring. And I think there's a lot of ways that we can try to do that. Now in the last few minutes I want to mention a little bit about some of the efforts on the public health level. And we can go and try to include public health as part of the work. And, you know, I think one of the things that we realized as we were working in different countries is that one of the highest stakes endeavors that a ministry of health undertakes is the selection of their national drug formulary. The cost of people is not that high in most countries, but it's the cost of medicines that is one of their biggest expenses. And so when you're picking medicines for your country, again, how do you pick them? You mainly go to the WHO essential medicines list, which is a wonderful robust list of data across all the commonly -- or all the known disease areas, but it's almost exclusively white data. And so if you're making a selection of therapy for any part of the world, and use WHO data, you're really using data from a population that is not your own. And when you're looking at a patient population that's different from the clinical trials, it's functionally experimental therapy. And so that's sort of -- because it's such a high stakes undertaking, because it's such a big part of their health care expense, it's critical we do it right. Now, if -- I don't know about your health plan, but in my health plan we have a whole list of drugs that are on formulary. And, you know, if you don't -- there's literally, I think there's 60-something anti-hypertensives that are on formulary. But if I want one of the drugs that is not on formulary, I have to write a little note on the website and it goes to my doctor who writes a little note and it gets adjudicated somewhere in New Jersey and comes back saying all right, but you have to pay a higher co-pay. So I can get it, it's just not on there. In most countries, they can afford one or two drugs for every indication and there's no other options, because they can't afford anything else. So if you got cash, you go to Switzerland, but otherwise you're stuck with whatever the national formulary is. And so it's a high stakes endeavor. Much more -- if we waste a dollar over here, we got another dollar, you know, we just borrow another dollar from Beijing. But in terms of many of these countries, they can't do that, and so they're stuck. And so it's a much bigger deal for them. The best option is we know your risk and my risk and make decisions based on that. The worst option is that we get data from one part of the world and infer it to the rest of the world in the hopes that it might accidentally work. And so it's no surprise that, you know, whatever country you're in or you're from, you can get plenty of anecdotes or we can't use that dose. We can't use that drug because of that. And so at intermediate is trying to look at groups within a population. And the Voltaire saying is definitely part of it. We'd like to be best, but can we at least be good on our way to being best? Now, this is some data from a chip. I'm not advertising this chip. AlphaMetrix has a drug metabolism and transport chip. I don't work for them or any of that sort of stuff. But if you -- it has just under 2,000 functional variants in drugs that metabolize or transport -- I'm sorry, genes that metabolize or transport drugs. So this is the sort of thing -- many of you know Doug Figg and Doug Price in the group there at the NCI. They're doing this DMET plus chip on a lot of the patients at the clinical center, at least that's what I -- or they're going to be, anyway. So, when you look at these sort of work, these are only -- these are not tags. These are the actual functional variants that cause a change. So when you look at this data and you look at the HapMap populations, even this little data separated out patients, separated out groups, rather, based on geography. So here's the Japanese and the Chinese HapMap samples. Here is the Caucasian Europeans from Utah HapMap samples, and the group in these green circles, you can kind of see them there, are the Arubans from West Africa that are there. Now, we've also layered in some of our other African populations. There's two different Ghanaian populations, and there's also a Kenyan population, and they all clump here in the Africa group. So it looks easy. We do one study in Africa and then whatever we find will translate across all of Africa. Perfect. Because look, if they're all clumped together, so, you know, one size fits all in terms -- at the continent level. So that sounds good. So we looked a little further. This is an example with thiopurine methyltransferase. TPMT is the gene name. It degrades. It helps get rid of azathioprine and mercaptopurine, so drugs used for rheumatoid arthritis, inflammatory bowel disease, pemphigus, pemphigoid syndromes, acute childhood leukemia, et cetera. And so there are genetic variants that have been found across the world to be -- to knock out the elimination of this drug and give you a high risk of severe neutropenia. And so you can look at the frequency -- I'm sorry, there's dosing as well. So you look into frequency and the way we do these maps, green means your frequency is the same as U.S. whites and the reason is not because I'm from the U.S. and white, but rather because most of the initial dosing data is from normal volunteer studies in U.S.-based white males. Now, there's some U.K.-based white males and some Australian-based white males also used. So, we have diversity. But it's -- the main dosing is from those populations. And you can see -- and if it's gray that means there's no data. If it's light blue that means the risk, the frequency of the toxicity risk gene is half or less, and you can see that yellow is here in Bulgaria and Ghana and Peru. That means the frequency is double or more for this risk genotype. And I have no clue what those three countries have in common, but that's there. So if you look in Ghana, you focus down in there, you can pop it out here. Here are five of the most common tribes in Ghana and they all have around the 10 percent risk. So it's not like one of the tribes was really driving this thing in terms of this specific example, but rather [unintelligible]. Now here's the Aruban data from Nigeria. Now, if you're geographically challenged, you've got West Africa, you've got Ghana, you've got Burkina Faso above, Cote d'Ivoire, or Ivory Coast on the one side. You have Togo and Benin, two small little countries, and then you got Nigeria, a big old country. And so if you look at the Arubans from just one part of Nigeria, they're right next to each other. They're both former U.K. colonies. They have a lot in common, except Ghana's soccer team is way better than Nigeria's. And if any of you are from Nigeria, you can take that to the bank. We'll talk about it later. But, so you look -- so the frequency of the risk allele in the Aruban population is half that seen in the Ghanaians. As the matter of fact, it's almost identical to the U.K. Caucasians. And so, based on geography, we would've got this wrong in a big way, because the risk is very high. Now we've gone in pharmacovigilance studies and identified that indeed this is a very -- this genetic risk also happens clinically, in terms of being clinical risk. We can't do that for a lot of our examples, but we could for this one. And so, even though that all the genes together clumped people together in Africa, when you look at specific actionable events that would change formulary decisions or change individual clinical decisions, we see differences based on geography. Now this is really tough to read, but in this box here -- so this is CYP2C19, two different alleles, star two and star three. This is the gene that activates clopidogrel, Plavix, as one of its genes. Also, many of the proton pump inhibitors and other drugs. So within this box, so these are different countries across the world. Here's the Asians, because they're the few groups that have the star three allele, but in this box are four different African populations. And you can see one of these is quite high and one of these is quite low. These are two different tribes within Ghana. So even within a country, within groups that are somewhat similar in appearance and they're different in language and other things, they're similar in geography within the country, but have a very different frequency as a population. And so as we're looking at this, in terms of identifying risk, we can now use this data in order to try to make decisions in terms of what drugs are going to be best and for which part of the country, in terms of its implementation. Now, the ideal is that we know the individual patient's risk. That is the ideal. The worst is that we just ignore that there's an issue at all. And so it's a really tough thing, because we don't want to be propagating this idea that ethnicity and race matters, but on the other hand, if there are self-identified groups that have a differential risk, we can't ignore that. We wouldn't ignore patients that are old and age as a factor for toxicity in many cases. We don't ignore age just because it's not, you know, we don't want to be ageist, so we're going to ignore toxicity risk in old people. I mean, that's just not ethical. So, the same thing is in terms of how we push this forward. I'm going to skip through this in the interest of time to get to the last piece, and that is I showed you some of the ways there applied, but there's also other aspects that are coming. And all of these are boring. They're things like the bundling of care. You know, before if you got your hip replaced you had a separate bill from the surgeon, from the device company, from the hospital, from the anesthesiologist, from the physical therapist, et cetera. Now it's all bundled into one cost. This is the -- the hospital or the health system is paid one amount of money and then they can sort out the details. And so what we're finding now is that toxicity is expensive. Bounce-backs. Anybody who is readmitted within 30 days after discharge from the hospital for the same indication, that second admission is not paid for in many cases, and so it's free. You broke it, you buy it. And so this idea that suddenly, I mean I'm over-simplifying things, but that's where we're headed. This idea that expense is just something that's a pass through is starting to go away. It used to be that adverse drug reactions were a cash cow for the hospital. You know, Stevens-Johnson syndrome costs about $50,000 in our health system to manage. That was $50,000 we got. Now it's $50,000 we lose or is heading toward that. And so this idea that toxicity matters is becoming really important because it now not only matters to the patient and to the clinicians, but now it matters to the accountants and the others that are involved there. And so we're seeing a real push for genetics in the context of minimizing these different elements, trying to make things more efficient and move forward, not just decreasing bleeding in the head types of things that we thought we would. The other element is we're great at discovering things and decent at validating them, but there's a whole lot more to this. We need to be able to look at all these different elements. And so, one of my cousins is a family practitioner in New Mexico, uses a lot of Warfarin. You know, sees this in the FDA package insert, reads my New England Journal papers, and says, "Hey, I'm going to order a test." Orders the test, calls me up and says, is CYP2C9*3 good or bad? I mean, he doesn't know anything about it. He slept through genetics just like the rest of us. And so what does it mean? So by -- and there's always other factors. So by Brian Gage developing a website and by us developing our iWarfarin iPhone app, it's free of charge, go ahead and download it, hours of fun. And you now have a way to translate that stuff into a dose and to apply it in a way that's a little bit more straightforward. And so there's a lot of work that has to be done with these types of things to really push this forward. So in closing, we need to still do good discovery and great validation, but we also have to be working on this part of it. And many of these aspects can be boring. We need to be realizing that the patient part of it can be quite boring. It's much less sexy than the early discovery piece, but if we get all the way to here and stop, we've done nothing. I don't know if any of you have ever seen half of a bridge and thought it was a beautiful thing and useful, it's not. We need the whole bridge. And right now we're building and building and building great first halves and need to do a lot more in taking it forward. Otherwise, the genome will be a tool for us eggheads and not necessarily something else. So I'm going to stop at this point in time. I want to thank a whole bunch of people. These are groups from around the world in my backyard having a barbecue as part of our PGENI, or Pharmacogenetics for Every Nation Initiative, or PGENIUSES as we call them, and I want to thank the folks at UNC Institute for Pharmacogenomics and Individualized Therapy. Thank you very much. [applause]