 Welcome to today's lecture, the 10th in the current topic series. As I said last week, most of the lectures are focused primarily on the genomic analysis of mammalian systems. Last week, we took a departure from that topic and talked about how to analyze the microbes that live on and inside of us and how their composition may differ both between people and between healthy and diseased states. Today, we're going to explore another way that people differ from each other, namely in their response to medicine. Our speaker this morning is Howard McLeod from the University of North Carolina. Dr. McLeod's field of expertise is in integrating genetics principles into clinical practice to advance personalized medicine. Dr. McLeod is a Fred N. Eshelman Distinguished Professor and the Director of the UNC Institute for Pharmacogenomics and Individualized Therapy. He's also a principal investigator for the Create Pharmacogenetic Research Network, a network of labs evaluating pathways regulating drug activity. He's also a member of the FDA Committee on Clinical Pharmacology, and he directs the Pharmacogenomics for Every Nation Initiative, which aims to help developing countries use genetic information to improve national drug formulary decisions. Please welcome me in ... Please join me in welcoming Dr. McLeod to give this morning's lecture on pharmacogenomics. Thank you very much. It's a pleasure to be here and to talk to this audience. There's many of the aspects of genomics that you know much more about than I do, and so I'm going to try to tell you a bit about how we're trying to use the genome for good and hopefully not evil, although I'll hit on some of the evil part as well as we're going forward. But the area of pharmacogenomics is one that has been hyped up quite a lot in terms of trying to use genetic information for making decisions about medicines. And so I'll hopefully talk through some of the realities of where we are currently, including some of the discovery pieces that need to be done. And so it won't just be on late clinical application, but we'll start off on some of the discovery aspects there. Now I'd like to start with this quote here, a surgeon who uses the wrong side of the scalpel cuts her own fingers and not the patient. If the same applied to drugs, they would have been investigated very carefully a long time ago. Now this quote is supposedly from 1849, I don't read this particular journal, but it's very true today that of the FDA approved drugs that we have available in this country, there are none of them that we truly know the mechanism of action. Now if you're a student or a fellow or something like that and you have to take an exam, there's of course a right answer of how a drug works and you have to pretend that that's the truth. But the reality is we don't know how drugs work. We know that they hit certain targets and we often label them by that. But COX-2 inhibitors have activity in COX-2 knockout mice. And topoisomerase one inhibitors have activity independent of topo one levels. And the list goes on. And so we know something about medicines, but not enough about medicines to really dial in the right one for each individual person. And so if you take a group of people with your favorite disease, treat them with your favorite therapy, in retrospect people will stand out. And they'll stand out for two important reasons, a lack of therapeutic benefit and unacceptable levels of toxicity. And so the goal is how do we go and take this technology about the patient? And try to make it so that these folks here get an alternate therapy rather than the usual therapy that we might give. Now the clinical problem that we have in the therapeutic side is really a wonderful problem. There's quite a lot of young looking people in the audience. Not everyone, of course. But there's quite a lot of young looking people in the audience. And so you may not realize that it wasn't that long ago that there weren't a lot of choices for most common diseases. If you take the area of cancer where I spent a lot of my energy in the treatment of colon cancer, up until about 10 years ago, the big question was, do I give five flora uracil by bolus or by infusion? So basically from 1958 until 1996, there was only one drug available for that common disease. Now there are four cytotoxics, three biologics, and more coming, just for the treatment of that one disease. And so real choices need to be made. The unfortunate truth around the modern therapies, though, is that with the exception of bone and mineral disease and bacterial infection, most therapies will only work in about half of the patients in which they're tried. And so you give a patient a drug, half of a group of patients, will need to switch to a second drug, either sequentially or additively. And some will need a third drug or a fourth drug. And you get to the point where people switch over to homeopathy and all sorts of other things because they lose faith in the allopathic approach because we don't seem to know what we're doing. We've tried five different drugs. Well, after that, why not go to quackery? Because it's just as good as what we were up to. There's also unpredictable toxicity. There's Dan Rodin from Vanderbilt, likes to say, can we take the idio out of idiopathic, trying to make it so that we can predict some of these toxicities. But toxicities are a big deal. They happen to the patient and not the prescriber. And so they're not given the first priority. But they are a very big deal in terms of whether patients take their medicines and in terms of whether they're able to sustain their therapy. Much of the modern therapy for colon cancer is incredibly active, but also harms the peripheral nerves. And so we have many situations where a patient's tumor is responding nicely, but we have to stop the therapy because we're frying their nerves. They can't button their shirt. They can't feel their fingers to do anything. And so that situation where we cannot optimize control of the disease because of toxicity is a big deal. In the area of cancer, they grade toxicity from zero, meaning none, up to five, meaning you actually killed the patient with the toxicity. And so when I go to the data center to look at trials that I'm running, I just say, well, just give me the severe toxicity, the grade three, grade four stuff. Well, if I had grade one diarrhea right now, I'd be talking to you from a little room outside there that says ME in on the door and not from up here. And so even a small amount of toxicity can be debilitating to a patient. And yet it's not even on my radar because it's just a silly little problem, literally in that case. And why bother? And so trying to make sure we try to optimize both control of disease and acceptable levels of toxicity is an important aspect. And then there's the element that those of us on the academic side just don't want to face. And that is the fact that modern therapies for most diseases, certainly including cancer, cost real money. And in a formerly rich country like ours, we can't afford the modern therapy for all people. And it's timely that we're talking about this today, but having spent a fair amount of time in Europe, I was there in Scotland for eight years, I've seen what it means to have a limited therapy budget. And when you get done with that budget, there is no additional budget for therapy. And if you ever wonder why clinical trials are so successful in other parts of the world, well, part of it is there aren't a lot of other options. If you run out of your drug budget, clinical trials are often the only option for the treatment of many diseases. And so can we offer a patient therapy that will optimize the control of their disease, minimize the toxicities, but in the context of our health system and the limited expenses? And I don't mean that we should all become health services researchers, please do not do that. But rather, we need to stop thinking about things in such a narrow view and realize that it's the compilation of these areas that drive healthcare. And if we only do research, trying to do one of these three things and ignore the rest of it, we will never succeed to the fullness that we want to. Now, I showed this slide, it looks like someone from a review from Elliot Vacelle looks like the font didn't come through to give him full credit. And what this is showing is that genetics is at the hub of the wheel in terms of variation in drug effect. And that's great, you wouldn't have come to this talk if you didn't think genetics and genomics meant something, unless you're really desperate for continued education credits. But the genome is certainly a variable that is important. Looking around the room here, a lot of variability and outward appearance, much of it regulated by genetics in terms of skin pigment, hair color, some aspects of body habitus. The few of you that have met my children know they look more like me than they do you. That's a good thing for you, bad thing for them. But the facts are that genetics can regulate things like outward appearance. And so it's no surprise that absorption of a drug, metabolism, breakdown of a drug, the amount of target for a drug in a tissue can be regulated at the genetic level. But shown on the outside of the circle are a number of factors that can be very important clinically and have nothing to do with genetics in many cases. So a drug-drug interaction can be very important clinically, often has no genetic basis whatsoever. Poor kidney and liver function can be very important clinically, often has no genetic basis in terms of that. If your patients are from the coasts of North Carolina and are surfers, marijuana intake can influence metabolism, may have no genetic influence, but can it be important in terms of the therapeutic side? And so just a reminder that the genome is a tool for understanding intrapatient variability, but it is not the tool. And certainly the genome project has, during its initial years, came off saying get out of the way so we can cure your patient, as opposed to here's a useful tool that might benefit you in understanding your patient in a much better way. And so we're just now getting around to the point where people are willing to strongly consider the genome as a tool because of some of these initial issues. And so I'll try to highlight where the genome is useful, but also the fact that we need to ask the question, is the genome useful? And we'll come back to that point in a couple of slides. Now, I'm not gonna go through all of the current examples that are out there, but I put this slide to show you that we are not at the beginning of pharmacogenomics. This is a list of examples where the Food and Drug Administration has changed the prescribing recommendations in the dosage and administration section of the package insert. This package insert is the aspect of the, this section of the package insert is the aspect that all clinicians are supposed to read, and we know that doesn't happen. All insurance companies do read, in terms of their decisions, and of course they're heavily read by litigators. And so we're now seeing examples, not just in the clinical pharmacology section or somewhere buried deep in the package insert, but up in the drug administration section, the one that people actually have to pay attention to, for a number of different drugs. Some of these are somatic mutations, and Philadelphia chromosome is older than most of you, in terms of its finding. Her too is an old story. But other examples, in terms of some of the HLAs for severe hypersensitivity reaction. And then most recently, a week ago, Thursday, the FDA put a black box for genetic information in terms of lack of efficacy for clopidogrel, which if you're not familiar with that drug for acute coronary syndrome, it's also known as plavix. And if you've watched TV, you've seen a plavix ad. It's almost as common as Viagra, in terms of its ads. So we're talking about now drugs that you've even heard about that have genetic information. In that case, a large percentage of the population, about 20% or so, cannot activate this drug and therefore will not get clinical benefit from this therapy. And you don't know about it until it's too late. The person has restinosis of their stent or worse. We're also talking about a number of drugs that are commonly used across disease areas. So the five HD3 antagonists are used for nausea and vomiting and cancer, but they're also used for anesthesia prior to general anesthesia to prevent the nausea and vomiting, mainly the vomiting, in that case. And so while there is a lot of cancer in this country, there's a lot of surgery going on. And so we're talking about examples where the thousands of patients per medical center are now eligible for genetic guided therapy for common diseases. So it's no longer a boutique incident to some rare disease where one person in the US might care about it, but something which is a very common issue. I mean, I'm gonna focus in on a couple of these examples as we go along through this, but we're not talking about a news story, but something that is now really rolling along at a fairly rapid pace. Now I'm gonna hit on three fundamental issues as we go through these slides. First of all is trying to find the right biomarker. So looking at more of the discovery side of what's being done and what can be done to be smarter about that. Secondly, a little bit about validation in terms of having proper data sets to validate as opposed to some of the approaches that have been taken in the past. And then lastly, actually applying this information to real people for real treatment of real diseases and where things are in that case. Now, the reason why discovery is still needed comes back to that initial slide about understanding drugs. We think we're quite smart. We can go and take a drug like, this is a rhenotecan, a cancer drug. It goes into this cell. It can be pumped out by active transport. It can be inactivated by liver P450 enzymes. It can be activated by carboxyl esterases to this metabolite, which itself can be pumped out, inactivated, hit a cellular target. There's mediators of cell death, and we are smart. And if I had a good graphic artist, I'd be even smarter. So we look at this sort of thing and we think we're geniuses, except here's the real pathway. Especially in the pharmacodynamic aspect, we really practice yogi-bara pharmacology. We know what we know, but we don't know what we don't know. We know that these genes have the potential of being important because someone's knocked them out or overexpressed them or done something to substantiate that, either in the laboratory, in a mouse, in man. But we have not stepped back and asked the biology to tell us which genes should we really care about. And so often we have these pathways and they look really slick. If you go to the Pharmacogenetics Knowledge Base, farmgkb.org, you can see these beautiful graphic artists depicted. I think that might be me, I'll throw it over here. Beautiful graphic artist depictions of these drug pathways that really look like we understand something, but we have a long way to go. And so there is a lot of, even in 2010, there's a lot of need for stepping back and asking questions about whether we know the right genes. And so whether it's in mouse, in families, in large population studies, we need to really ask that question. So one example that we've done is to try to use families, use the family structure and all the decades of knowledge about how to conduct family genetic studies in the context of chemotherapy drugs. Now we can't do true family studies with chemotherapy drugs. You can't bring an extended family into the clinical center here and give them all a cytotoxic chemotherapy and see who gets severe neutropenia. It's slightly unethical. You don't want to be saying, sorry about grandma. That's not really something you want to be saying with your studies. And so because you can't do those kinds of studies, you can't do twin studies for the same reason, what we've done is taken this into the in vitro setting. And one of the first questions we wanted to ask is, is cytotoxic chemotherapy response a genetic problem? Is it likely there's a genetic solution to this? And it's shocking to me after we finally came around to thinking this way, it was shocking to look at the way we've been approaching pharmacogenomics over the last decade or so. We've assumed that genetics was relevant. And that's fine. Those of you that are in genetics, you kind of want it to be relevant. It's your life. Well, there's the old saying, if you have a hammer, all the world looks like a nail. Well, if you have a sequencer or a next-gen machine or whatever it might be, every problem looks like a genetic problem. And too often what we're doing is spending a large amount of money to genotype or sequence or in some other way look at the genome. Well, we don't even have a clue if genetics is involved at all. And it really is pretty stupid when you think about it. I know some of you are involved in funding decisions and I'd like you to please forget that. But we really need to ask the question, should we even be trying to look at genomics? Because too often we're doing it because we can, not because we should. And so we ask a question using some of these families. And so most of you are familiar with the CEPC cell lines. They've been an important part of the human genome project and the HapMap project and many other projects. And you can get these cell lines from the Corial Institute where there's either large two generation or more commonly three generation families and you can take these cell lines, grow them as a culture and do cytotoxicity tests or other types of testing depending on your assay. And so this is an example of a 96 well plate. We use mainly three to four well plates now but it's prettier on the 96 well plate. Where there's two different drugs, assay and quadruplicate on this plate. And then there's an increasing amount of concentration of each of these drugs so that you see different levels of killing based on the dye indicator. In this case, alum or blue being used as the dye. And so what you can do is from this sort of data, by using multiple replicates, you can get kill curves. And so in this case there are, each of these lines is a separate replicate for a specific cell line from one of these families. And with increasing concentration of the anti-cancer drug, you can see a rather steep killing that's going on here. And each one of the replicates itself has 12 repeats within it. So it's 12 replicates times three for each of these points. And then other cell lines with the same number of replicates have a much more shallow kill curve. I don't know where the, here we go. Sorry the mouse is, there we are. Much more shallow kill curve for this particular cell line. And so you can look at the killing of these cells and ask the question, is this a heritable trait? How heritable is cytotoxicity? And so we have one project that since these are the CEPH cell lines, we call it the CEPH-DA project where we've taken a panel of these cell lines from a number of families and taken all of the FDA approved anti-cancer drugs, including some of the kinase inhibitors and demethylation drugs, et cetera. And ask the question about heritability. We did not use some of the drugs that needed activation. So there are many drugs that are pro-drugs. You swallow them, they have to be activated in the liver before they can work in the body. That's difficult to do in cell culture. So we didn't attempt that. But by taking these, we can generate this kind of data. Now let me walk you through it briefly. What this is showing is a series of anti-cancer drugs, most of them fairly commonly prescribed anti-cancer drugs, either corrected for growth, for growth rate or uncorrected for growth rate. On this axis is heritability of the maximum heritability across several concentrations for each drug. Over on the left side here are the controls. So 1% DMSO, water, or media with serum depending on, as the various controls for this experiment. And so what you can see is that there is a degree of heritability seen across the different anti-cancer drugs, such that some of them, such as Epirubicin or Timozolamide, have a very high heritability up in the 60% range. And then some of them are really no different from control in terms of heritability. Now this is cell lines. These are lymphoblastoid cell lines. There's a number of assumptions and limitations with this system. But it is one of the few ways that one can ask the question, which drugs are more likely to have a high level of genetic influence going on? Now heritability is only one measure of trying to look at whether genetics is relevant. And we're well aware of that. But what we are seeing is that we have some drugs with a very high level of heritability and are much more likely to be the source of genomic discovery. Other drugs at the low end, we don't know exactly why they're low. It's not a lack of variability in cytotoxicity. It's just a lack of heritability. And so we think that, and it is not EBV transformation. We've looked into that part of it. But we're still trying to figure out does this mean these drugs should never have genetic studies? Or does it mean that this acid just isn't good enough in terms of trying to prioritize drugs? But it gives us an ability to put some context into whether a genomic discovery approach is likely to have high utility or whether it's gonna be a relatively low yield exercise. And this sort of data really should be generated for every aspect of pharmacogenetics. Not by us. I'm not trying to plug that we should do it all. Please, I do not want that. But I do think that a question that needs to be asked is really is genetics likely to give us the answer we want? And can we prioritize this? Now, using these families, one can do good old fashioned linkage analysis. And thankfully we have some old timers around UNC that remember how to do that sort of thing. It's gotten so trendy to do a genome wide association, et cetera, that a lot of the young folks don't know about that except for their history of genetics class. But we can go and look across the genome. And more importantly, we can look within families of anti-cancer drugs. So take a family of drugs where there is a multiple members that are used clinically that are within the same chemical class. And so in this case there's two of the fluoro-primidines that are out there. And there are examples, such as this little box here, of peaks that are where there seems to be some level of association, in this case not a very high level of association, between cytotoxicity and this region of the genome. And then there are other areas where there's a rather high level peak, at least for this cell system, in both members of this family. And so we have examples, for example, with the Camp de Thessons, where we were able to go and pull out peaks that were common to the initial six, clinically used Camp de Thessons that we have and then go in and validate those findings with additional experimental Camp de Thessons, ones that have not gone into humans, to try to look at replication of these peaks and are finding peaks that are not at a place where we have a known target, but seem to replicate across the family for this area. And so one can look under these quantitative trait loci and start to ask the question, well, why? Is there an association between this region of the genome and cytotoxicity for this family of drugs? Now, of course, we have the problem that we always have, whether it's expression arrays or linkage analysis or genome-wide association analysis. When you put together your list of genes under a particular region of the genome, they all have the perfect story for why they are the right one. So this is the start of a lot of work, not the end of a lot of work. And thankfully we have a SHRNA library where we have five clones for every known human gene we're able to scan through using some cell-based systems to try to identify which genes under the QTL are important. But this concept of narrowing the genome down from a lot of genes down to a focused number of genes to try to do new discoveries in terms of mechanism of action is one way of trying to revisit this important area. And this is no surprise, the approach I laid out is one that has been used for years with disease genetics. But pharmacogenetics has really derived out of clinical pharmacology. There are very few geneticists who are applying strong genetic techniques to pharmacogenetics. And so people like me who really know more about pharmacos than they do genomics are gradually learning as our genetics and genomics colleagues come through and teach us a little bit more about how to apply this in the proper way. The other aspect that I want to briefly touch on is that there's been some growth in the mouse model systems that really are starting to open up the mouse as a strong genetic tool. And there's always been utility in the mouse, not just in terms of knockout and transgenic, but in terms of the variability within strains of mice. But more recently, the collaborative cross, which you may be well familiar with, has come forward. Initially led by David Threadgill, now by a number of folks at UNC in Chapel Hill, where several of the more disparate strains, inbred strains of mice, have been brought together for a massive breeding exercise in which there are now almost 1,000 new unique strains of mice that are now available, that are at or greater than the amount of genetic variability seen across the human populations across the globe. And so instead of having 32 strains of it, 32 inbred strains of mouse in which to do studies, we now have the equivalent of 1,000 person study in which to be able to do discovery. These mice are very reproducible because they are inbred and fixed in terms of their genetics. There is a current effort to complete the entire genomes of all these mice with the Sanger as well as US-based efforts so that these mouse strains will have the entire genome laid out for you. You can go in, do your phenotyping, and then be able to ask questions. And so for example, one can take a high density imaging system in which you can look at various phenotypes and this is a slide from Tim Vulture and it's his fancy graphics, but you can use a high content imaging system to look at a number of different cellular features. And so in this case, we're looking at the effect of a drug on cell loss, nuclear size, total nuclear intensity, permeability, membrane potential, cytochrome C, a number of other measures can be done simultaneously. And therefore, you can take these strains of mice, either in this case using the mouse imbrennial fibroblasts as an in vitro assay or using more classical in vivo assays and try to do discovery in that case. And so for example, we have a region of the genome that came out from the taxanes, dosataxyl and patlitaxyl from our Ceph experiment and a region of the genome from the mouse imbrennial fibroblast experiment that overlaps in terms of sentinine. And so two independent mammalian systems both pointing to a very narrow region of the genome in which there's a single gene that had not been implicated in the activity of this drug but does have biologic plausibility. And so we're now doing the additional experiments to try to nail down whether this gene is indeed a regulator of its effect. But the idea of doing discovery in multiple mammalian systems in a relatively high throughput manner and try to look at where there's commonality across those systems is really, I think has a lot of potential in terms of its discovery. Sir. So lines and mice, live mice as well. So in this experiment, sorry, I was going too fast. In this experiment, well, you could make as many as you want. In this case, we make a one set of mouse imbrennial fibroblasts from this. So I don't know the answer to that. I know locally we have cell lines that are derived. As I understand it, these mice will be available to the community. This was paid for by NIH, DOD, and DOE money. And these mice, I believe, will be widely available as a resource from them. We didn't do the vivo experiment for two reasons. One, many of these phenotypes are hard to measure in that and usually you can only measure one phenotype. Secondly, cost. So with these cell lines, we can quickly go through and using 3D4-well plates, very rapidly assay all 1,000 cell lines in a very rapid manner. This data is only showing 32, by the way. I don't want to oversell this particular slide, but one can look at a large number in a very rapid manner. And so with the robotics that we have and the 3D4-well plating systems, we haven't used 1536, but one certainly could. You can scan through 1,000 cell lines in a pretty quick manner. Now, when I was a postdoc, if I had five cell lines to work with, I'd freak out. Now we have these technicians who just don't know any better. The key thing is to hire someone who doesn't know any better. Because then they think, oh, 1,000 cell lines, well, that doesn't seem like very many. Oh, yes, yes. So the nice thing about these mouse strains is because of the genomics is so well characterized in terms of variance and copy number, we now know which mice we can go back into for those initial experiments. As I'll show you in a couple of slides, we also have large clinical trial cohorts. So if we credential a gene to be of relevance with our cell-based system, we can quickly take the variance that are a functional consequence to those studies and then take it into our clinical material. Sorry, say any question again? These are inbred strains. Yes, they weren't done quickly. The question was, how can we make them inbred quickly? And the answer is, you cannot make them inbred quickly. No, they're out at about 25 generations now. No, there's no quick way of doing inbred mouse strains. Nothing's changed there. You didn't miss anything. It's still as slow as it always is. It's just that it's been done now. So there's a lot that can't be done in terms of discovery. Those of you that are on the discovery side, and maybe you're getting tired of working on disease discovery, come on over to pharmacogenomics. The water's fine, come on in. We need a lot of help. We have a lot of drugs that are out there and a bunch of drugs that are coming into development and we don't have a clue how to use them. And so there's a lot of work still to be done using even approaches that are old, tried and true, but will hopefully help us go forward. Now the second issue I wanted to hit on was validation in robust data sets. And too often what happens is you do a really cool discovery in the laboratory. And then you go to your favorite clinician that you have T with on Thursdays and you say, hey, you got any DNA? And they say, oh yeah, we have 42 samples that we've consented to the patient and we have a toxicity phenotype. And so you genotype your favorite SNP and your favorite gene in all 42 samples and both heterozygotes had toxicity and no one else did and therefore everyone should be tested for this gene the end. And there are steaming piles of literature along this same line, including some from ourselves where we've gone and done these types of experiments. And they were useful early on to demonstrate that one could do these studies but they are completely useless when it comes to changing practice. Really the only thing they're useful for nowadays is to try to expand one CV and potentially get promotion. It really has no usefulness at all for helping a fellow man. So trying to do the right studies though is very hard. The reason you do the 42 patient studies, the reason we did the 42 patient studies is because it was convenient. We had the tumors in the tumor bank. We had the clinical trial samples sitting there. But it doesn't mean that it's actually going to be useful. And so what we've done in the context of the NCI cooperative groups, mainly the Cancer Leukemia Group B or CALGB or CALGB depending how you like to call it, is to go in and start integrating blood sampling and when possible tumor sampling into the prospective clinical trials. And so rather than having 46 breast cancer patients from Chapel Hill, we have 4,600 breast cancer patients all on a prospective clinical trial with auditing of the evaluation of toxicity and efficacy, independent imaging evaluation. These studies are being done at over 200 centers across the US and Canada. And so we have the sort of normal variability in terms of the various centers that you would want in terms of trying to prove something being useful. And having these numbers, they're not all large numbers, but having this sort of data from prospective studies even if we're doing a retrospective look at the evaluation of the markers, allows us to have a much more robust conclusion about whether a marker is likely to be of use or whether it's just something that's unique to Chapel Hill. We've certainly seen in the past many markers that seem to be relevant in one particular center, but just do not replicate when you take them out across different places. And often it's a protein biomarker and on further evaluation what you find out is that that particular investigator only has his clinic on Tuesday mornings. And it's the same lab techs that are running everything and the same nurses and there's a level of homogeneity that ends up influencing things in ways that we don't understand, but lead to a lack of replication. And so by having these data sets across the country, we can hopefully get a better conclusion on whether a marker is truly useful or truly useless. And too often the pharmacogenetic literature has been neither. It's been promising. You know the biggest cop-out in science is a paper where the last conclusion is further research is needed. Complete cop-out. What it means is we didn't do the right study. Now in some cases you can't do the right study. The clinical data sets not available or the mice aren't available or the cells aren't available. But in most cases where that conclusion is there, it's because we really didn't do the right study. And so what we're trying to do and not succeeding in every case is get into the position where we can do definitive studies that will either put something to rest or cause us to go forward into patient directed therapy using these sort of markers. And so one of the examples I have is for a ovarian cancer study. It's called the Scott Rock study. This one was actually conducted in Scotland, not in the US, but it's an example of a paper we've just are sending to a prestigious Boston-based journal. It may not get accepted there, but we'll start there. And in this study, what was found is that the outcomes, sorry, this study looked at the common therapies for ovarian cancer. Carboplatinum, the platinum agent was common to both arms. And then one of the taxane twins, either dosataxyl or paclitaxyl were used. And so these drugs are very similar to each other and both have activity in ovarian cancer. But the question was, are these different? And the bottom line was clinically they're not different. Their outcomes in terms of progression-free survival or overall survival were not statistically different. The toxicity profiles were pretty similar. There was a slight difference in toxicity profile. So what one can then do, sorry, but the main toxicity that was seen in both arms was toxicity to the nerves, peripheral neuropathy. As I mentioned to you before, many of the platinum-containing regimens will cause a peripheral neuropathy. And it's a big deal. It affects patients' quality of life quite dramatically. Any of you who have ever been to an ovarian cancer clinic or have had a friend or loved one with ovarian cancer have been there, you'll see a number of women with walkers. And these aren't necessarily old women, but they're using a walker to try to get around. The reason why is that the drugs have fried their nerves. They can't even walk normally because of these drugs. And it might be controlling their disease, but it's also really affecting their nerves in a negative way. And so we wanted to ask a question of why? Can we predict this in a better manner? And there really was not known. So we did, at that point, a custom chip, taking genes involved with nerve function, genes involved with nerve function, genes involved with inherited neuropathies, realizing that the variants found in chocomeric tooth or whatever will not be those variants, because those are extremely rare, but maybe some more common variants might have some functional consequence. And then genes known to be involved with the pharmacology of these drugs realizing the caveats that I said earlier when I was ridiculing my lack of knowledge about drugs. And so looking at these variants, using, in this case, an alumna golden gate SNP assay, to try to look at predictors of neurotoxicity in this patient population. And so we started off out of the 1536 SNPs that were available. There were 1261 that met our quality assessment. Many of them fell out because there were monomorphic. There were some of them that were too far out of Hardy-Weinberg to be believable. There were others where there were controlled issues in terms of the genotypes. So of these 1261 steps, we initially looked at a cohort of 500 patients. And there were 69 that came out to be positive after multiple comparison correction for this initial cohort. We then took these 69 SNPs into a second cohort of 500 women on this same therapy. And there were five of them, five SNPs, that were still statistically significant after multiple comparison correction. Four of these that were in a consistent direction. So there was one of them that was a positive predictor in the first data set and a negative predictor in the second data set. And two wrongs do not make a right. That variant got tossed out. So we ended up with four variants that looked of interest. And when you look at these genes, they're variants in BCL2 or some of the other, that gene you can kind of make a story in terms of cell death. Some of these other genes are, there's not an obvious story, except for their involvement in nerve function or in peripheral neuropathy syndromes. But the odds ratios for each of these were between 2.2 and 4. So the type of odds ratios that you'd pay attention to clinically, none of these odds ratios of 1.15 that we don't have a clue what to do with, but odds ratios that you would actually care about in terms of individual markers. When you look at the population attributable risk, when you put them all together, they had almost 85% population attributable risk. So a very substantial amount of risk for in terms of potential clinical utility. And then when we do the simple thing that people do these days, when you basically make a score where you add up the number of bad variants that a person has, in this case it went from zero to four and this is showing a mapping of the odds ratio. I realize there's a very steep rise here. The curve makes it look much smoother, but that's what our statistician gave us. But there was an additive or in this case, a super additive effect depending on the number of variants you had associating with the risk of peripheral neuropathy such that the folks with four of these variants had an odds ratio of over 40 in this particular study in terms of their risk of having peripheral neuropathy. We also asked the question, do these risk scores associate with outcome? So the old teaching for cancer chemotherapy was that you needed to nearly kill the patient in order to kill the tumor. And back when the drugs were intense cytotoxics, nitrogen, mustards, et cetera, that was probably true. In modern therapy that is not true. There's study after study demonstrating that severity of toxicity has no relationship with the degree of efficacy for whether it's a kinase inhibitor or one of the cytotoxics such as was used here. And so we asked that, but we still wanted to ask that question. And what we saw was there was no relationship between their neuropathy score with these genes and either their progression free survival or their overall survival in this study. And so what we ended up with was four SNPs and four genes that validated in a separate 500 patient cohort with biology that we would not have associated directly with these drugs. The markers were associated with risk of peripheral neuropathy but not with the survival or efficacy outcome measurements. And so we now have an opportunity to both try to prospectively see whether these markers do predict neuropathy and that study is ongoing, as well as take these genes and put them through a screen for inhibitors. And so we have one of the NIGMS screening centers at UNC. And so they are screening a large library of compounds against these genes. They're actually not doing BCL too, that one's been kind of done to death. But using the other three genes, trying to identify inhibitors to see whether they might be potential adjuvants to be used with this chemotherapy. Because based on this data, the degree of neuropathy and the efficacy are not linked. And so in theory, we can inhibit the neuropathy without influencing the efficacy. At least there's the promise of that being true. And it would be a great day if we were able to minimize the neuropathy because it is so debilitating to these patients. Often we're not trying to cure these women. We're trying to give them a better quality of life for the time they have left on this earth. And to screw up their nerves so they can't play the piano or they can't button their blouse or they can't hold their grand kid is really not our goal. And so hopefully genetics, this sort of data will lead us towards some approaches where we can at the least know that we're gonna get into trouble early at the best have some interventions that can make things even better. The last piece on in terms of robust data sets is we're using these data sets now initially for genome-wide association studies. And we've completed four genome-wide association studies to date. The first one was in a pancreas cancer study that paper has just been submitted to the Journal of Clinical Oncology where looking at genetic predictors of both outcome as well as toxicity to the drugs, neutropenia as well as the hypertension that you get from some of the vascular endothelial growth factor antagonists. We have two genome-wide studies in this area, one for neuropathy, one for neutropenia. And then we have a genome-wide study that's just been completed in prostate cancer that we're now starting the analysis on. And if, depending whether Francis smiles our way with the most recent director's grant that was submitted on the 15th of March, we have a next-gen proposal in this same data set to go in and use some of the next-generation sequencing, or in this case a whole lexome approach to try to do discovery in the context of a prospective clinical trial, which is something that has not been done, at least on the cancer side of things, to date. So we are starting, we're slow. The disease, go to the NHGRI website, you can see hundreds of GWASs for disease, very few on the drug side, and partly because we didn't have the data sets in which to go in and do discovery. In clinical pharmacology, 50 patients is a huge study. In GWAS, it's not even worth doing it. We did it with 50 patients. So we're now getting to the point where we have the numbers where it's worth trying to do a GWAS, and we'll see what we get in terms of new discovery. I know on the neuropathy and a neutropenia side for this study, we are finding genes that we would not expect to come forward. Unfortunately, one of the genes that looks interesting for neutropenia is associated, unfortunately, with a syndrome called McLeod syndrome, and that has been the source of terrible ridicule by my colleagues as they make fun of me and my disease that I didn't even know about until now. And so this sort of area is certainly one area we're trying to do discovery in the context of this sort of data sets. So the last piece is really about applying this sort of work, and too often we find something cool in the laboratory. We maybe do a local assessment of whether this biomarker is gonna work. We even do a large validation study. And the next step is nothing. Oh, sure, we publish the study in New England Journal of Medicine and we get an NIH grant and we get a free trip to Bethesda. But what do we really do with these studies? A couple years ago, my mother was on the internet, hopefully looking for recipes or something. And she came across some PDFs of some of my scientific papers. And so she pulled them down and she sent them to my 97-year-old grandmother. And my grandmother lives in an assisted living facility where a few years ago they had a bunch of 14-year-olds come in and teach the inhabitants how to use computers and learn some of the lingo and stuff. And she's not blogging or anything, don't get me wrong. But she's able to actually do some stuff. Pretty good, better than my parents. And so she sent them to my, sent the papers to my grandmother. And so I was talking to her and she said, oh, your mother sent me some of your magazines. Like, oh no. But turned out it was a science stuff, so it was okay. So I said, well, what'd you think, grandma? Well, my grandmother has a third grade education. She's a, you know, she's an active lady, but not a real savvy science type. And I said, what'd you think, grandma? And she said, beautiful font. That was, and I was impressed she knew the word font. But that was all it did for her. It did nothing for her, I mean, she was happy to see my name on there and everything, but it did nothing for her. So are we doing the kind of science that is gonna help her grandson get promoted and do lots of cool things? Or are we doing the kind of science that's gonna help grandma? And too often, we're doing stuff that has no handoff. You know, if you look at industries, when there's a group that's doing engineering for some part of a car, they have a plan for when they're done, handing it off to the next people who are gonna do the next part and to go on the next part, the next part. In science, often our plan is osmosis. We do our science and we publish it out there and we pray that someone will accidentally come across it and think that, hey, maybe they can use it in their area. We have no plan at all for a handoff of where it's going forward. And we need to have that. And one of the things we're trying to build with my institute was pulling together all these people to have a handoff. And I'm not saying we're succeeding, but it's really something that we need to be doing as a country is making sure that some of the phenomenal basic science and fundamental science is fast-tracked forward. And the reason it's not now is mainly because the groups in the various sectors don't interact, don't see each other and don't read the same journals. And so it's no surprise that there's amazing things tucked away that are not helping people because we didn't plan for them to help someone. So do... Well, no, yeah, well, I think that there, I mean, certainly we are helping people, but too often it's on accident, where we know this problem. So part of what we need to do is we work very much forward, do a basic discovery and then build on that, make it more translational and go forward. Too often we don't work backwards at all. And yeah, oh, yeah, everybody in this room knows that there's been progress made and certainly don't want you to give them away saying that we're all a bunch of losers and should go down to Burger King and apply for a job. They do give you fries with that. But rather, I think that we can do better by planning our science in a more careful way. And those of you that are focusing in an individual area, I don't mean that you should change and become something else. Please keep doing what you're doing, but just remember there's a handoff and having that handoff, you know, when you drop a baton in a relay race, you're the idiot and you get kicked. And you know, we drop the baton a lot in science because we don't plan for the handoff. So doing biomarker driven studies in this context is often not the step. You know, there's been a lot of new in the journal papers that have had nothing done after that because the investigators thought that they succeeded. That was the goal and nothing done further. And so one example where we're trying to take this forward is in the breast cancer drug Tamoxifen. And most of you are familiar with Tamoxifen. It's a drug that's in the news quite a lot. Back when I trained, life was a simpler time. Tamoxifen, which is a partial antagonist of estrogen, was converted by a number of different enzymes to four hydroxy Tamoxifen, which is a potent antagonist of the receptor, of estrogen receptor. And that was it. That was how the drug worked. And a couple of years ago, oncologist named Varid Stearns over at Georgetown University and her colleague David Flock, a clinical pharmacologist who was there at the time. Varid had a patient with breast cancer who was taking Tamoxifen, was getting the hot flashes, the Perimenopausal syndrome that you get from this drug. And that's expected and normal. But she also had clinical depression. And so was treated with one of the serotonin reuptake inhibitors, SSRIs, for her depression. And within a week, the hot flashes went away. Now, if it had been me, I would have been jumping for joy. Anybody with Scottish blood loves two for the price of one. And so, hot flashes go away and the depression will eventually go away, two for the price of one. Gotta love it. Well, Varid's a very smart person. What she did is she realized, she said, wait a minute, that's wrong. That the hot flashes went away so quickly, there's something different going on. And so pushed it and analyzed it further. And what she identified is that there's a metabolite, now called endoxifen, that is formed. It's predominantly formed on this pathway here where there is some redundancy on the first step. But then P450 2D6 is the second step. Most of the antidepressants that were used at the time are inhibitors of this enzyme. And so what was happening was that the antidepressant, yeah, it was making the hot flashes go away, but it was making the go away because it was neutralizing the drug. There was no active metabolite being formed. And so your hot flashes can go away if you don't take the drug, but you also get no clinical benefit. And so this sort of data led to a dramatic change in which antidepressants are used in breast cancer patients. And there are some in the vaccine in particular that do not inhibit this enzyme, and therefore have become the drug of choice because of that data. Well, as many of you know, somewhere between five and 10% of the people in this room are deficient for this enzyme. You either have a deletion or a non-synonymous SNP in this gene that causes you to not have any function of this gene. And you probably don't know it because it doesn't cause any outward effect. The ones of you that do know it have done a phenotyping assay. And that is you went to the dentist, you had some dental pain, he gave you Tylenol with codeine, you took it and it didn't help you at all. And you went back to the dentist and he told you you were a wimp. Well, five to 10% of you have a genetic reason why you're a wimp. The rest of you are just wimps. The five to 10% cannot activate codeine to its active metabolite, cannot activate oxycodone to its active metabolite and cannot activate tamoxifen in this case. It's the same gene for that case. And so the studies have been done to look at the genetic variation. And what's shown on this axis here, the y-axis, is the concentration of active metabolite. And there's a stepwise relationship between whether a person has two normal copies, one normal copy or no normal copies, so-called poor metabolizers of this gene. Now, there's variability within each of these. So this is not the only contributor to this gene's function. But it does influence this gene's function. What's shown on the right-hand side are people with two normal copies or one normal copy, but are on one of the inhibitors. And so it's reminding you that drug interactions can cause a similar effect to an intermediate or to a heterozygous or a homozygous variant genetic condition. And so either drug interactions or genetics can both influence the active metabolite levels in this particular case. Now, there have been a large number of studies that have shown an association between the genotype of CYP2D6 and recurrence of breast cancer in patients on tamoxifen. There have also been three studies, two from the same group in Sweden, one from Arkansas, that did not find an association with tamoxifen. Now, you can try to explain back and forth between these. There's differences in how many variants they looked at because there are over 75 different variants in this gene if one wanted to be complete. Also, most of these studies here are larger and have patients that are just on tamoxifen, whereas some of the studies here have adjuvant chemotherapy also mixed in. But we've got to the point where it's really beyond a publication bias in terms of the relationship. And these are the sorts of curves one sees. This happens to be from one of the initial studies where the folks with a poor metabolizer genotype have a very poor outcome, relatively speaking. Intermediates have an inferior outcome but not quite as dramatically different. And then the extensive metabolizers have this outcome shown here. Now, what this shows you is that the genotype does influence outcome at some level. But even in the good group, the green group here, there's still patients having recurrence of their breast cancer. And that could be tumor biology, it could be some other germline effect, but there's never going to be one genetic solution to a complex problem. And so one shouldn't expect there to be one gene or such that's gonna be the answer to this. It's a complex disease, it's gonna have a complex selection of solutions. But we do see this sort of effect. Now, some of the insurance companies are now picked up on this and are starting to offer testing, in many cases, offering testing for free. So MedCoHealth Solutions has worked with the various companies that they represent and they're a drug benefits management company. And they now offer testing for Tamaxvin and also for Warfarin and for Clopidogrill for free. So it completely takes out the occasion of the discussion about is it worth the cost of the testing because they've factored it into their relationship with IBM and General Motors and other companies that they represent. And so that every patient who gets a prescription for any of those drugs immediately, literally immediately in the electronic sense, gets offered a genetic test. If they're willing for it, their prescriber is offered the test and then they have clinical support staff that help them through that. But you could see how this sort of thing could be used also for evil. So you identify the people who are gonna not benefit from Tamaxvin and you offer them an alternate therapy. Okay, that seems like a noble thing. But Tamaxvin is much cheaper than the alternate therapy. So you could also see a situation where you go and you assay all the people on the alternate therapy and only the five to seven percent that will not benefit from Tamaxvin will you leave on that therapy. The other 95%, you say, well, you can stay on that therapy if you wanna pay for it, but we will only pay for your Tamaxvin. Now Tamaxvin is not a bad drug. It saved more women's lives than any other therapy that's out there. But that patient and clinician autonomy will now be put into question because of this sort of relationship. And so there are ethical issues and legal and social issues to try to work through in terms of how this is implemented in a way that is good for patients but not driven solely by cost savings and issues like that. The other thing is, for those people who have a especially bad outcome, they need a different therapy. For these folks here, what do we do? Now, the few of you that are interested in oncology might know that Tamaxvin is normally given as a 20 milligram dose in this country but is FDA approved from 10 to 40 milligrams. And so the most common call we were getting from community oncologists is what do we do with these women? This is like 40% of the population. What do we do? They know what to do here. They know what to do here. They don't have a clue what to do with these 40%. And so we initiated a very simple study where we took patients that had been on Tamaxvin for at least four months. We measured active metabolite levels and we see a statistical difference. I don't have the proper version yet because our statisticians don't like to generate pretty graphics but so this is a cartoon version of that. But the active metabolite levels are significantly different between those that are extensive metabolizers and those that are intermediate metabolizers. So these folks have differences in their active metabolite levels in our local study. The additional study was 120 patients. We're now finishing up a 500 patient study to do the same finding. So we wanted to ask the question if we leave these people, the green people, on their 20 milligram dose, are they still having similar blood levels? And they do. There's no statistical difference between the start of our study and should be four months later for this case. We then took the patients who have the low blood levels. We took the patients who are genetically heterozygous. We didn't measure their blood levels in real time. And we doubled their dose. Gave them the 40 milligram dose. So it's nothing sexy. No kinase inhibitor or anything like that. We just took the same old boring drug and we gave them double the dose. Still an FDA approved dose, but double the dose. And what we found was we were able to normalize the blood levels. They went from a statistical difference here to no statistical difference here. And there's a dramatic increase in terms of looking within each patient. And so by doing something as simple and boring as doubling the dose, we're now able to completely normalize blood levels. Now, as I showed you from this slide here, even with the extensive metabolizer genotype, and therefore, in this case, the higher blood levels, as we've shown, you still have patients who have recurrence. So we're not curing everyone. But we believe that we've closed this gap, at least at the pharmacokinetic level, we've closed this gap. We now have to do the further studies in terms of outcomes. But we've been able to demonstrate that by doing something simple, that could be done at any practice across the U.S. And using an already FDA approved dose, one can take this biomarker in terms of metabolites and normalize it for what is functionally 95% of the women with breast cancer. So this sort of story needs now to go on to the survival type data, but using a short-term biomarker, able to show that genetics can cause this sort of effect. Now, the ideal is we get to a point where we can have toxicity evaluation, disease evaluation, infection risk, supportive care, of all done prior to ever prescribing a drug. And that day may come at some point in time, but it's still a ways off. Especially, it's especially a ways off if we look at from a global standpoint. If we look around the world, there's a lot of countries that have a intact health system, but don't have the money to do genotyping in every single patient. And so the last few minutes, I'm gonna mention briefly some of our global health efforts, where we've tried to make the genome useful to the developing world. And with the human genome promise, there's lots of things that it was supposed to do in terms of better diagnosis and better selection of therapy. And genome type guided therapy is now starting to happen in rich countries, mainly in the West. But what about the rest of the world? You know, there seem to be always 30 years behind, and that gap never closes. And so we tried to come up with a way to make the genome useful now to the rest of the world. You know, the genome, when you talk to the folks in many developing countries, the genome is completely a Western toy. The human genome project has been as useless to them as your favorite Disney Princess. I mean, it's really something that is meaningless because they know it'll be 30 years before they can use it. And so we've tried to move forward and say, well, if we can't do individual patient genotyping, can we do something that's useful? So the best option is that each person understands their makeup genetically and otherwise and then makes decisions on their therapy based on that. That's the ideal. The worst situation is that you infer data from one population onto the rest of the world. And that's what we do now. Most of the world uses either the EMEA, the European Medicines Agency, or the FDA safety data for their starting point for their drugs. They either can't afford or just don't want to do separate studies there. And so what happens is that most of the initial studies in the U.S. and most initial studies with new drugs are conducted in the U.S. or in Western Europe. Even the Eastern Europe studies that are being done now still represent a very white population. So phase one studies, where a lot of the initial dosing done is in 18 to 40 year old white males. Oh sure, there's a few people that are not white. But over 80% based on our FDA analysis are white. They may now be whites from Ukraine instead of whites from Australia or North Carolina. But they're still not very representative of the world population. And so it's no surprise that when drugs get into the rest of the world, they find differences in terms of the toxicity profile, both the incidence and the types of toxicity, the dosing that is needed as they go forward. And so an intermediate step is to try to understand the genetic risks of toxicity and efficacy in individual populations within a country. And so very much going on the Voltaire, the best is the enemy of good. The best is the ultimate goal. Perfection is our ultimate goal. And in this case, it's being able to look at the individual patient in every person that needs therapy. But in the meantime, we can still do good. And too often we wait for perfection and therefore don't do good. That's not very good English, but you know what I mean. And so can we do something now while we're waiting to do the ideal? And so we've taken the WHO essential medicines list. I've done a lot of work on identifying of these 206 systemic drugs, what genes influence the metabolism transport or our targets for these drugs and looking at the variants that influence these drugs. And we don't mean variants that are non-synonymous. We mean, and we don't mean variants that cause a change in blood levels. We mean variants that have been shown in at least two populations to cause a change at least a doubling of the risk of toxicity or altered efficacy in real and tech people. And right now in the US, you have to have some pretty amazing studies. We're always talking about we need prospective randomized trials to change which therapy you might give an individual person. But if you're buying drugs for a national formulary, the national drug menu for a country, if you have two equal therapies, if you have two drugs that are equally effective, have equal cost, equal accessibility, all you need is a feather of data to shift one versus the other. You don't need to have a huge amount of data because you have to choose one of these. You know, in this country, if you don't like the drug that is on your formulary for insurance company, you fill out a little form, you get a sign barrier doctor, and usually you can get a waiver so you can use some other therapy. In most of the world, they have one set of drugs for the treatment of most diseases. And if you want anything else, well, you can go fly to Switzerland because there are no other choices. It's a high stakes deal about which drugs are available because they can only afford one or two options for each common disease. And so by using this data, one can then try to make these choices. So as an example, the thiopurines are on this list for pharmacogenetics. These drugs are used for rheumatoid arthritis, inflammatory bowel disease, some dermatologic disorders. They're also used for the treatment of childhood leukemia. And so if you look at the genome, there are variants in the thiopurine methyltransferase gene that are associated with severe neutropenia. In some cases, fatal neutropenia and these are the main variants that have been shown across the world to be important. And depending on what genotype status you are, there's different dosing requirements that are already well known. And so one can take that data and certainly here in this country, we use that data to try to identify risk and to adjust the dose of the drugs. Now we've created these world maps and so let me walk you through them very briefly. If you see gray, that means there's no data. If you see green, that means the data is very similar to the US white population and I picked the US white population not because I'm a US white but because that's where drugs are developed and we don't have to like it but that's the current state of affairs. Light blue means the risk is one half or less. Yellow means the risk is double or more. And so if we take a country like Ghana which we started working in back in 1994 and look at the frequency of the variants for this gene, they have, it's over 10% of the population have a high risk variant for severe toxicity from these drugs and that's been backed up by a high risk, high incidence of toxicity in the pharmacovigilant studies that they've done. So one can go in and look at the data and looking across the most common tribes within Ghana, you can see that they're all pretty much the same and the fontae are a little bit higher but really there's no statistical difference among these particular groups within the country. But what's shown here on this slide is the Yoruban population from the Hap map. Now if you're geographically inclined, you'll know there's Ghana and then there's Togo and Benin, two little tiny countries and there's Nigeria in West Africa. The both former English colonies, a lot of similarities. Ghana has a much better soccer team than half any of you are from Nigeria, I'm sorry, but Ghana just, you know, the Black Stars rule. But if you look at, you'd expect that they'd be the same. Matter of fact, the Yorubans have been pitched as the West African population for the Hap map project. Well for this particular clinically actionable genotype, the frequency in the Yoruban population is one half that seen in nearby Ghana and the Yorubans are identical in frequency to the U.S. and U.K. Caucasians. So when we're talking about these population differences and wondering why we would need to look at each country and each group within the country, this type of data, where you're talking about a clinically actionable variant, when you start looking at that level of resolution, there are dramatic differences, in this case a doubling of the incidence between countries that you'd expect to be identical. And so this sort of information we're using as we pull this forward. So we have data on the surveillance of patients where you go in and you'd, if there's a heightened risk, you would monitor patients more carefully. So something like isonizid for tuberculosis. If you have a heightened risk of liver toxicity based on your genetics, you're not gonna stop using isonizid. It's too important for tuberculosis. And so you'd monitor the patients more often. So we have several programs where we're monitoring patients every six months, sorry, every three months instead of every six months, using cheap little dipstick type of evaluations that are affordable in those countries. You also can prioritize available therapies. So if we're looking at malaria, the WHO has four different drug sets that they propose as being any of these four are the right ones for malaria. In many countries, these two have too much resistance and so are not used. And so you're really stuck with these two on the outside. And if you look at Amadiaquin, one of these particular drugs, it's metabolized by a P450 to this metabolite. Now, either the parent or the metabolite are equally effective. So metabolism doesn't influence efficacy at all. But if you have a genetic variant in the metabolism, you have more active drug, sorry, more parent drug that shifts down to form this toxic metabolite. And so genetic variation influences the amount of toxicity that's seen in these different populations. This was initially found in Zanzibar off of the coast of Tanzania when they were using this therapy. And so when you look at the incidence of the variant in Zanzibar, it's 2%. Now, those of you that do population genetics, if you have a 2% of the population being at heightened risk, I bet you don't even care. You can completely blow it off because it's so rare, I mean, who cares? Well, if you then put in the burden of disease, 2% of the Zanzibar population will get severe hepatotoxicity from this drug. That equals 30,000 people per year on one tiny island. So the burden of disease combined with the level of incidence suddenly heightens the, you care. So if I came to you and said, you could identify 30,000 people who are gonna get severe liver toxicity from this antimalarial, you'd care. And so bringing in the burden of disease along with the genetic variants is really important. If you go to Bolivia, almost double the frequency of this variant, but very little malaria, only 64 people per year are gonna get harmed by this drug. Go to Malaysia, half the frequency, a fair amount of malaria, about 8,000 patients. So bringing in this sort of data allows us to work with the ministries of health and help guide them on which therapy one can look at. And so we don't blink at all when we look at HIV virus or malaria or TB in terms of sensitivity of the bug, but we also have to look at the patients and their germline, the host genome in terms of making these decisions. And that's been completely neglected with the current global health plans and hopefully we'll be able to push this forward. So in conclusion, we are getting pretty good at discovering things and we discover something and we look at the outcome and we feel pretty good about it. Except there's a lot more that needs to be done before things can become routine practice. And so as you're going forward, I would encourage you to start thinking about who are the other people that need to get involved? You do not want to become a clinical pathologist or a medical informaticist or a health systems integrator, whatever that is, and you certainly don't want to become a health economist. But they may be the key to your science helping someone. And so planning the types of teams that can really drive this forward into routine practice are key. And until we do that, we're going to create great science that will eventually help someone, but not at the pace that it could. So the Pharmacogenetic Research Network gives us a bunch of cash to do the studies that we do and we thank them for it. And then my local institute pulls together the various types of people that help us to do this work. And I'll stop right there. Thank you very much. Could you use the aisle microphones for the questions, please? Oh, thank you. Okay, here is a question. I think it's a nice idea. I'm talking about global health issues which we've raised. It is clear this is a very large effort. And we are, as you pointed out, we are just beginning in this effort. What is your estimation? How many more years, 50, 100, 200 years, will we take when we reach the level of sophistication which you want to achieve? So it's hard to predict because the changes that have happened over the last year have happened much faster than I would have anticipated. So with companies that are developing new drugs now having large data sets with genetic information and actually doing something with those data sets, it's really changed the game. With the NIH now having genetics blood sampling as part of their clinical trials infrastructure and enforcing it in many cases, it is changing the game. So now that we have data sets for many of the large, the therapeutic areas that are large enough to actually draw conclusion, I now think that we're looking at probably a 15 year horizon before half of the drugs that are prescribed will have clear data. I don't think we'll ever get to the point where all drugs that are prescribed will have this data. Partly because there's many drugs that are so safe that they don't need it. They don't need to be individualized. But the ones that are being focused on now are either unsafe in a substantial amount of the population are super expensive, such as some of the rheumatoid arthritis drugs, or both. And so we're seeing that sort of emphasis. And so therapeutic areas that I never thought would have markers now have some really nice data. Interferon for hepatitis, I would have bet you good money that we would never have a marker for predicting who's gonna respond to interferon. We do. We have markers that's been shown in multiple populations now. The ribovirin data that just came out in science. Really, drugs that I thought were so far down the list we'd never have any data, and here we go. So I have become cautiously optimistic now that we're starting to see some real traction with drugs that we actually care about. Yeah, it's good to hear that. Thank you. Another question. So in thinking about the chemotherapy in the neurotoxicity associated with another way of approaching the mechanism would be more at the cellular level. So if people looked at demyelination or anything like that, they're exploring those possibilities. Is it directly on the nerves or on the Schwann cells? Have anybody used, looked at that level? It, they have. Mainly in rats to date. Sure. Of the humans and stuff they're gonna do there. And the dorsal root ganglion seems to be a major site for toxicity. And so we have a follow-up study in the collaborative cross with all these mouse strains to try to look at that. The poor technician has to dig all those dorsal root ganglia out of each mouse, needs a medal. But we're gonna try to look and see are there genetic predictors of who's getting toxicity at that local site? Cause you're right, there's so many different factors that could be interesting. It appears that that's where the toxicity is occurring rather than myelination issues or central issues. Yeah, cause if memory serves me correctly, I think SOX-10 is a marker for no crest development. And you picked that up in your screens. Right. And so we- You can go further with that too. Yes, yeah. And so we have, so we've been, actually maybe you're even the one we got it from, there is a, there's collaborators that have given us knock out mice for each of those four. Yeah, sure. And we're going through, including someone here at the NIH. And I'm not doing those mouse experiments, but and so we're looking in the mouse to see if we can replicate this and then are taken into the collaborative cross. So we have the knock out mice and then we have the inbred strains as well. Thanks. It was a great talk. Thank you. So the question was, is there a good review article for the markers on drugs? There are a few out there. There's not a lot, there's not one within the last six months that I can think of that is especially good. Would, is there a website for this? Like if I sent you some, some people, let me go pull the actual citing for a couple of those and put it up on the, on the website. Yeah. The last, so New England had a review, but that's not, you know, it's 2003 when they last had that review. It's, it's, it's far out of date. There's a couple from 2008 that are pretty good. A lot of them are very specific. So there's, there's, you know, some great review on Tamoxin Pharmacogenetics. Well, that's great, but you know, what about everything else? I just want, I probably don't need this, but for your study in Ghana versus, in the Yerubin versus Ghanian ethnic groups, what were your sample sizes? Because they probably weren't as big as the ones here at NIH that you're pointing out. Well, in, in Ghana, each tribe had 500 individuals. So it ended up being 2,500 people for the Ghanaian data. The Yerubin data was the HAPMAP data. So that was much smaller. That's only 30 trios if I remember correctly for that. So that's, and that, you know, that could be influencing it. Certainly the Yerubins, you know, those individuals don't even necessarily represent all Yerubins, much less than they certainly don't represent all Nigerians. So it could be, from, from the Ghana standpoint, we have good numbers, but from, we use 500 for each population based on our, we want to be able, so our statisticians went through and asked the question, how big a sample size do we need to be able to detect some of the rarer variants with at least a doubling of incidents? And so that's why we're doing 500 per population. And you would see that much variation more in Africa than elsewhere? We see it, we see it all over the place. And so I don't know if you, you probably noticed on that particular slide, there was, you know, there was a very high instance in Bulgaria, a high instance in Peru. We don't have a clue why, and probably we never will have a clue why, but certain populations just seem to have these, even populations that if you looked at a million snip chip would be nearly identical. When you look at the, because we're talking about an individual decision based on one or a very small number of variants. And so it, you know, all of the data that helps someone be similar goes away. So what does that, is that what you use? Yes, so, so we currently use the, the Afro-Metrix DMET plus chip, which has more of the variants that, that we need than some of the other products that are out there. We need to supplement it for the HLA markers and for some of the pharmacodynamic genes that are out there. We, we are having a debate in our circle about the utility of HAPMAP cell panel, lymphoblastoid cell panel. Now, is that panel being largely used to identify the targets, or is it being used to find new drugs? Is it a drug discovery or target validation? So I, I'm not a big fan of using the HAPMAP cell lines for discovery. I was a big fan until our statisticians did power analysis. And there's, there's just, you know, there's a huge amount of data, you know, back to the Scottish thing. I thought I'd be able to go in, phenotype these cell lines, get the genome for free and be great. Well, you do get the genome for free, but there's so few samples that you, because you put them all together, it's 270 cell lines, whatever it is. It's still too few to do discovery for a lot of the markers that we're talking about. So we've moved away from that. We now have collections of, of unrelated individuals. So we have a thousand, well, just over a thousand European Americans, a thousand African Americans, and a thousand Taiwanese, where we have immortalized EBV, EBV transformed mononuclear cells. So we do discovery in that way because we had enough statistical power. But, you know, several of my colleagues do use the HapMap cell lines on a regular basis, and, you know, they believe what they find, and they know my opinion. And more power to them. I, you know, they, they certainly are doing good quality work. I just don't happen to believe their results because of that. In general, should we be concerned with EBV transformed cells? I mean, these are all EBV transformed cells. Yes. Which is not the nicest thing to do, or, or make trans, you know, transversal. Should we be concerned about that? Well, it's, it's one of those necessary evil things. I mean, until there's a better way of doing it, it's about the only option we have. We, we look, so we look at both copy number, as well as, as expression of some of the EBV, what is it? There's a certain phase that we, lytic phase, the lytic, lytic phase gene. So we use that and put it into our statistical models. And so far, I've not seen a big effect of that. But David Altschuler did a study where he looked at copy number, I believe it was, and did see an effect of, of, he used me to look at cytostatics, not cytotoxics. So, you know, I think that the key thing is that you look. And there are some people that aren't even bothering to look, bother to look. And, you know, their results are really hard to interpret. You have a question or are you? Yeah. So you had chosen your, your 5FU data was the toxicity, right? You did a GWAS study against the toxicity. How come DP-YD didn't show up? Do you think that these candidate SNPs are really that important? Or do you think that the GWAS approach is going to be the best way to go? So cell lines, so the advantage of our, of our use of cell lines was that we could actually do a family study and, you know, we could do discovery and less cell lines than we would have had to if we did it unrelated. The, there was many downsides to using cell lines. First of all, they only represent one compartment of the many types of cells in the body. We say there are, it's a cell autonomous approach. It's our, our cop out. But there, there's a lot of genes that down-regulate immediately when you go into culture and DP-YD is one of them. And so we've shown in, in mononuclear cells from people that are not transformed that within a couple of hours of, of culture there is less than a 10th of the enzyme activity still left. And so it's, they're, they're not just going down in terms of expression. There's active, active degradation going on of the protein. And so in, in that case and the case of a lot of the P450 enzymes, there's, there's, they're not well represented. And we really look at these as more of a pharmacodynamic discovery approach. The pharmacokinetic aspect are, are not going to be well represented in these cell lines. And, but pharmacokinetics, you can do those studies in humans much easier, not in families, but in certainly in patients. So we kind of thought that was a worthwhile trade-off. And can you just comment? I mean, we, we had shown that peak like a protein polymorphism related to dose-taxal and tachylotaxal neuropathy development, for example. And you just showed us that probably not so important. Do you think these candidate gene, do you think that candidate gene approaches are good? I mean, they're hypothesis-driven or do you think that these GWAS studies, do you think they're going to be sort of the way? I think, so if you have good biologic possibility behind your candidate, then the ultimate goal is to ask the question, is this variant real? And having enough samples to ask that question, you know, prospectively plan it, even if it's a retrospective dataset, I think is the key thing. In terms of GWAS, especially with the pharmacology, the pharmacokinetic part, the P450s are either absent or very poorly tagged with the two available million-snip chips. And so while you have million-snips around the genome, you have deserts around a lot of the P450s that we want. And the data, the snips that are in some of those, I just don't believe at all. And it's because of the gene homology issues. And it's very hard for both Illumina and AFI with the current technology approach to convert those assays. The only reason that DMET plus chip works so well is they use the inversion probe technology, which is able to lay down nicely. But even within AFI's products, their DMET plus chip works really well. For P450s, their million-snip chip is completely useless. I mean, you've had experience with that. Yeah, well, I mean, there is an instance where there are statins, for example, where they identified a SNP in SLCL-1BF-1. So there clearly is some representation of these. But yeah, it's not as well-covered as something like that. Yeah, that was a solute transporter. And they got lucky. That's great. Luck is good. Well, I mean, it was kind of funny because it was actually confirmatory of some candidate gene approaches from, we already knew that this was related to the AUC of these statins. So anyway, yeah, thanks a lot. I appreciate you. So the question... I don't know if there's people actually at... Frederic's still there, so I'll repeat it anyway. The question about complementary techniques for picking up new SNPs. The... Oh, other than SNPs. Oh, gotcha, okay. Because we are looking at other techniques to find other SNPs as a discovery approach. We currently are not... So, we currently are not for our large population studies, mainly because we're stuck with the samples that we have. Now, Jason Lee, but at our place, has some techniques for looking at some of the, I guess the, it's not methylation, it's more chromatin, yeah, epigenetic issues. And he's able to, he's been showing some really nice data on the Ceph cell lines. And so we're now, he's now gonna use some of our patient samples and see whether that technique can actually get some results that are robust. And so that would open it up for us there. But in our studies where we have tumor, and we're able to look at methylation differences between tumor and normal, or between tumors from responders and non-responders, we certainly believe that genetics, that DNA is only one part of a very complex issue. And so as the techniques get better for the late translation stuff, we're eager to apply them. Yes, we know that DNA will not be enough. Or at least genotyping snips in DNA will not be enough.