Blood Test Liver

andy baxevanis:please join me in welcoming today's speaker, dr. howard mcleod. [applause] howard mcleod:thank you. appreciate it. and i have the conflict of interest shown here for completion. so,it's great to be back here, and to have a chance to catch up with what's happening hereon campus. and also, to update you a little bit on what's going on in the field of pharmacogenomics.and the issue around -- they were trying to solve with pharmacogenomics hasn't changed,and never will change -- and that is that we now have many different active therapiesfor the treatment of most disease. and the

changes that will occur will be that therewill be even more therapeutic options for these diseases. and i'm sure you have yourfavorite disease that does not currently have a therapy, so it's most diseases. but formany of the common diseases, we have a lot of different choices to make. if you takesomething common like high blood pressure, there are over 100 fda approved drugs or drugcombinations for the treatment of high blood pressure. and so, how do you sit down andpick one of them for the initial treatment of a given patient? and too often, what we do is pick somethingthat we're familiar with, or that we know how to spell, or that we have some sort ofaffinity to that would cause us to start with

that medicine. and if it doesn't work, we'lltry something else in a different class or a different family of some sort. and so, thatreally speaks to the need for something more precise in how we choose from amongst thevarious medicines that are out there. often, we're taught medicine as if it's a michelinthree star restaurant. and if you had the good fortune to be at one of those restaurants,you go there and you -- there's no menu. there's a list of what you're going to receive becausethe chef has chosen it, and there's a bunch of things that are very small and have a frenchname. it tastes amazing, and you wonder what that was, but you don't have a choice. it's,here is this the menu. really, we're much more like golden corral, where you go in there,and there's 20 different entrees, and if you

don't like one of the entrees, you pick oneof the other 19 entrees, and go with that. and too often, you pick more than one of theother 19 entrees. and the idea that we have all these choicesto make is a very different way of thinking than we're normally taught. really, we needto be thinking about, how do we best choose from amongst these options, and how do wechoose what the next therapy would be, should that not work? and it's a different philosophythat we need to start building on. and the reason why we need to do that is that thereis so much variation in the response to most every type of therapy. if you work in bonedisease, bisphosphonates are quite effective, and there's not much variability there. withsome bacterial infections, not a lot of variability

and response. but in most other areas, importantareas like psychiatric disease, cancer, viral diseases -- hiv, hepatitis, the -- many ofthe other diseases that you work on, there's variation. someone will respond to the mostcommon therapy. about half the people won't. they need a different therapy. half of thosepeople will respond -- eventually, you might find a therapy for everyone. but it won'tbe the same therapy, in the way that we normally think about it. and so, understanding that variability ispart of the goal. and just to be clear, the goal is not to be perfect right away. if wecould choose from amongst those 100 fda approved anti-hypertensive drugs, and narrow it downto 25 drugs, that would be an advance. we

wanted to go down to the point where we knowexactly the medicine that's the right one for you. but even in advance of filteringdown to a smaller group to choose from would be an important advance -- ruling out someof the other options that are there. and so, this -- you know, enemy -- good is the enemyof -- great is the enemy of perfect, or good is the enemy of great, or whatever you alllike to say it. the idea that we could be smarter than we are now is really the goal.toxicities also remain unpredictable. and when you look at the reason why medicinesdon't work, sometimes it's because of the biology. sometimes, it's because the patientsdon't take the medicine. i know from one of the -- one of the big advancesthat has been made over the last couple decades,

that the statin therapies for cholesterol-- these medicines, if taken, will have a significant public health impact. but at theend of the first year of prescription, only one third of patients are still taking themedication as prescribed. many of them have stopped altogether. and i know, from my ownparents that, when taking a statin, they had a family reunion coming up, or they were goingon a cruise -- or there was something happening. they didn't really like the muscle pain. theyjust stopped it for, you know, just for a week or two, just to feel better, and thenforgot to start it back up. and so, many of the advances that we have with medicines willonly happen if the patients take the medicines. and too often, the adverse events -- the unpredictabletoxicities are part of that.

and then, of course, there's the issue thatnone of us want to acknowledge, and that is that these medicines actually cost money.and who knew [laughs]? i mean, the europeans knew. but the -- we really have a situationwhere we can no longer ignore that factor. if you're well insured, you will likely havea 10 percent co-pay on your insurance. and if you're taking one of the new, amazing anti-cancerdrugs, they can be as much as $20,000 per month. so, if you're spending $20,000 of yourown money -- your co-pay for that month on something, typically, when you spend $20,000per month, you've chosen what color it was, whether it had leather seats, satellite radio,other things like that -- usually a trident at the front, maserati in the back. you know,these sorts of things. and instead, you're

spending them on medicine that may or maynot work. and so, we have some changes that have occurred,in terms of the way we think about things, even for the well-insured, and this idea ofvalue is something that has to be part of our equation. and i know -- at least, i canonly speak for myself, i hate the idea of having to even think about that. but if we'regoing to take our science and make it real, we have to have some of these realities atleast be part of the late translational phase for this. now, when we approach these sortsof problems, typically, we think about it as a yes/no type of -- you're going to respond,you're not going to respond. or you're going to get this side effect, you're not goingto get this side effect.

in reality, probability, or probabilisticdata isn't enough. you know, i mentioned the statin medicines. you don't take statins outbecause someone just had a heart attack. i mean, actually, you do, for secondary prevention,but you try to take it early to decrease the probability of having a coronary event. samewith anti-diabetic drugs. you don't wait until someone has lost a kidney or gone blind tostart treating their diabetes. when you see uncontrolled glucose, you try to control itnow to decrease the probability that that sort of -- those kind of in-organ effectswill occur. same with antihypertensives. you don't wait for a stroke. you treat it nowto decrease the probability of a stroke. and most of our preventative medicine is thatsame way. here's an increased engage on the

x axis, increasing number of cases on they axis for colon cancer. and you see, right about at this age here, there's an inflectionpoint that a number of cases go up. that's when you start offering colonoscopy to tryto detect early cases -- polyps or early cases of cancer, to try to deal with that in thatvery treatable stage. and so, the concepts are there. but yet, when it comes to medicines,we don't really approach in that same way. and it's time to start thinking about whatis the science behind this? what is the epidemiology behind this? and can we act on it? now, thereason to focus on medicines is that they're an important part of our treatment for mostdiseases, but they're also an important cause of morbidity and mortality in this country.they're a -- at least, the most recent reports

have the adverse drug reactions as the fifthleading cause of death in the united states. there was a report from last night, from thebritish medical journal showing that -- it was a study from johns hopkins showing thatadverse events in general, not just the drug events, were the third leading cause of mortalityin the united states. so, it's a -- you know, medical mistakes, and including those aroundmedicines, are an important issue that we need to deal with. and you know, if you lookat other industries, the way they deal with errors is much more systematic, and must -- muchmore objective than we currently approach things in the biomedical sciences. and so,there's some opportunities to go there. adverse events are heavily litigated. manyof these things are predictable. and so, there's

an opportunity. when we look at -- not onlyare the large number of cases of death, but a lot of emergency room visits that can occur.and if you look at a place like -- in psychiatric areas, you see almost 90,000 patients visitthe emergency department each year due to an drug induced adverse event, just in thepsychiatric area alone. so, the problem is quite a substantial one. and every -- if youhappen to go to an emergency department and you wonder why there's so many people there,some of them will be because of the medicines. now, sometimes it's the patient not doingwhat they're supposed to do. so, if you take a full dose of insulin and then you forgetto eat, you know, you've contributed to the reason why you need to go to the emergencydepartment. if you take too many of your pain

meds, you know, you may have needed them,but there is a patient component of that. but often, the difficulties are things thatare not predictable and not related to dose. and in that case, there are, in many cases,genomic reasons we could act on that. i mentioned before that these therapies areexpensive. and so, there's an opportunity for value. it could be that $2,000 per monthfor that anti-cancer drug i mentioned before is a fantastic value because you have thequality divided by cost equation for value. or it could be that it's a complete wasteof money, and you need to know that and move onto something else. and so, the idea thatwe can be looking at what is the end game, and moving towards the more basic elementsis really important as you look at these end

points. and you know, some of these thingswill require human studies. it's kind of difficult to some of these studies in mice. they don'thave a -- they're not insured. they don't have a health budget for each year. but theidea of doing of these mechanistic studies to drive towards these end points is veryimportant. now, the things i've been talking about, interms of framing the clinical problem for medications -- the solutions have had a numberof different names, over the decades. they've been called personalized medicines, stratifiedmedicine, individualized therapy, patient-driven medicine, tailored therapy, if you're in theu.k. as i'll show in the next -- one of the following slides, pharmacogenetics, pharmacogenomics.but last year, when the president called it

precision medicine, we now all call what wedo precision medicine. and the reason why we call it precision medicine is really simple.the chances of getting funded are much higher if we call it precision medicine. the realityis we're not very precise. i don't like using the word precision medicine outside of a grantapplication because i don't think we're very precise. precision medicine is more of anaspirational term. someday, we may be precise in the way that we prescribe medicines. and-- as opposed to what is reality now. and so, it is a goal. it is something we needto work towards, but it is definitely just an opportunity. now, the circles here -- back in the late'50s, the term pharmacogenetics was coined.

first came in least in english print at thattime, and that was referring to some of the interactions between the genome in twins,and response to medications, and showing that there was a heritable component. and -- atleast, inferring that there was some genetics involved in that. the genes weren't known,but the term was applied. with the human genome project, we had to convert to genomics becausewe're looking at more than one gene. we're looking at multiple genes. it's a genome ofopportunity to try to understand the drugs, and therefore, it was -- it was made forward.as we started taking this into practice, it became personalized medicine, in terms oftrying to take patient factors and understand how to combine them with things like genomicdata to individualize the choice of therapy,

the choice of dose, or the choice of monitoringthat you might do for a given patient. and then, as mentioned, precision medicineis now taking this on in a much broader sense, trying to really look at multiple factorsof patient data. much more than one can think about using your own brain to try to reallypull this together to guide patient care. and i won't spend a lot of time on that exceptat the very end, come back to it. but i wanted to put that framework because you know, there'sso much buzz around precision medicine right now. but in reality, it's just part of a naturalprogression of trying to take things like the genome and understand, how to become morerealistic? how do we add some of the layers of complexity that are there, that are real,and put them forward? the idea that one gene

will drive most any disease is just not true.take a great example, like cystic fibrosis. there are -- there's a lot variability amongstpatients who have the same genetic variants. and some of that is due to other genetic modifiers.some of that is due to how aggressiveness they've been with their pulmonary hygieneand other aspects. and it all comes together to focus in on a patient of their lung function. and the same is true around many of theseother areas, including medicine. we need to know other aspects of the patient to reallypull it together. the other thing is that personalized medicine, precision medicine,pharmacogenomics, whatever you want to call it -- has really always been part of the waymedicine is practiced. there's always been

a component where a doctor has looked at thepatient, had some attribute about their lifestyle or about their size, or about some other feature,and used it. i -- during part of my training, i went to spend one year in scotland, andended up staying there in faculty for eight years. and part of that time was up at aberdeen,where they were having their 500 year anniversary as a medical school; 1495 is when they werefounded, and so they had a great library. and so, you could go in and -- had to putthe funny gloves on, cotton gloves, and you could look at some of the old texts. and onceyou got to know that an f was actually an s, and all these other copper print type ofthings, you could read some of these old texts. and i read an old translation of an even oldergreek book that talked about how they had

the patient pee on the ground. and if certaintypes of worms or flied appeared at that spot, they would give him this type of bark. andif a different type of bug or worm or fly came up at that spot, they would give himanother type of bark. so, the idea of personalizing medicine is not new. the idea of using biomarkersto drive therapy is not new. now, that was not a clear, pre-approved assay. there weresome, i'm sure, some regulatory issues around where you peed on the ground for that. butthe concept of taking individual patient data and selecting a therapy has been around fora long time. it's just that the level of precision, i guess, that we're applying has changed andimproved as we go forward. but medicine will never be truly personalized.i know your mom told you were special, but

you're amongst a group of special people.you're not the only one that's special. i have a little mug that says “world's greatestdad.” and it was such a disappointment when one of my colleagues had his own mug thatsaid “world's greatest dad.” i don't know what my kids were thinking, giving him thatmug, but the idea that, you know, we think we're special and we need our medicine -- well,we'll never -- for practical reasons. for cost. for other elements, regulatory elements.but we can make it so that from amongst the medicines that are out there, the ones thatare most likely -- that have the highest probability of working, will be there. now, some of this is all -- moving to thenext slide, is towards issues like greatest

customer accountability. and those of us onthe more clinical side just hate terms like that. but for some of the reasons i alreadymentioned, in terms of the amount of contribution patients are making now to the cost of care,that's becoming a feature. and then, that's changing really -- the practical expectations,the process. and if you don't -- if you're a health system and don't have an app forsomeone to be able to pull it out and look at their chart and order a refill of medicineor whatever, you know, they're going to try to move onto a health system that does. andso, there are practical things that are out there that are really driving this forward. now, the technology is -- has been a big driverfor pharmacogenomics, personalized medicine,

precision medicine. especially -- even thelast five years, the amount that one can do is just astounding. and if you think back10 years, which is -- i know it's hard to do, it's amazing what -- you know, we havesummer students that come in and can generate more data and interpret that data better thana whole lab full of phds could 10 years before that. and because things are not hand-craftedas they used to be, you can now do things in massively parallel approaches. and so,the technology is there. we can look at many different genes in very rapid fashion. thereis -- there's even some technologies where you can take something that basically plugsinto your iphone, spit on it, wait 12 minutes, and get up to five snp genotypes back. andso, in some places like the intensive care

unit, you can get a genotype result back fasterthan you can get a blood gas back and have some results on which to act on. infectiousdiseases, changing dramatically because of the technology where the idea that one wouldwait for a culture to come back is becoming a foreign part of -- or a part of the historyof medicine because one can get more rapid turnaround, especially in cns infection, rapidturnaround results on whether this is virus, fungus, whatever it might be, and act on itaccordingly. i mentioned about the patient burden. if you'repaying more, you want to have more say, and that's natural. and we definitely see that,out on the clinical site. there's less personalized care. if you are lucky, i have a doctor thati go to; it's kind of an old fashioned notion

these days. many people go to a clinic, andmight have a doctor of record, but if they're not available, will see one of the other doctorsthat are there for that practice. and that's become more common, and certainly, there'selements there. and then, the issues of even countries like the united states can't reallyafford to treat 100 percent of people with an expensive medicine that will only benefitabout one-fifth of the patients. and certainly, in the cancer area, it's relatively commonthat drugs will work in a 20 percent subset of patients, yet we don't often know who thoseare until after treatment. and so, that has certainly been a big driver of how thingsgo forward. now, one of these last backgrounds is -- iput up there just because it's often really

easy to get lost in your favorite technology.you know, this is a genomic series, so there's a lot of focus on dna and, you know, couldbe comp [spelled phonetically] the number of dna, or methylation of dna, but dna -- there's,you know, you could also expand this out to protein or rna, functional imaging, bloodlevels, circulating tumor cells in the case of cancer, other measures like that. lotsof differently ways of trying to measure a patient and understand what is it about themthat would cause you to do one type of treatment versus another type of treatment. and i mentionall these, and not that it is an exhaustive list, but it shows there's more than one wayof trying to do this. and you know, too often, it's the old, you know, if you have a hammer,everything looks like a nail type situation

where, you know, we do dna genome science,therefore the answer -- doesn't matter what the question is, the answer is dna genomescience. and really, the answers are going to be as we combine these things. now, i'm going to use a lot of dna examplesin the remaining time, but part of that is because of the practicality that dna technologyis ready. it's being used. it's great for discoveries, has application on the clinicalside; whereas some of the other technologies are not quite that far along. dna is alsoquite easy to obtain. you know, dna is very stable. it -- you know, you got dna from kingtut. you know, you get dna from a lot of different things. dna is also a -- you can really getit almost anywhere in the world, and it will

be stable enough to get to a laboratory. youcan get dna from blood, from knuckle scrapings, from hair follicles. i like to joke that youcan get dna from everything except oj's glove. so, it's basically -- there's lots of differentways of getting dna that can be valuable to the patient. and so, i'll talk about dna.but really, as we're thinking about, how do we take this forward, it's taking whateveris robust in helping us look at the patient and try to drive that into practice. last little background piece is just a reminderthat, as we get towards the patient side of things, we have to deal with the complexitiesthat patients have more than one set of genomes. in this example, there is a tumor genome andthe normal genomes, and you know, this might

be toxicity and efficacy, if you want to keepit in a more simplistic type of model. but the same is true with viral disease, and for-- there's some data that even things like heart failure have a somatic component interms of changes that have occurred over time. and so, the idea that we have this simplisticmodel where we can measure a genome and it -- a genome will be important for a patientis not true. and we need to be ready in thinking about, all right, how do we deal with multiplegenomes and the way they're applied? now, pharmacogenomics is not new, as i mentioned,nor is it something that will start happening in the future. it's something where thereare changes already. now, there are over 160 drugs where the fda has put genomic informationsomewhere in the package insert. so, know

when you pick up your prescription and youtake out the bottle, and there's that wad of paper at the bottom? i'm not recommendingyou read it, but i would like you to recycle it. so, as you pull it out, if you happento read it, there's a bunch of different sections that talk about dosing administration or toxicitiesor this or that. and somewhere within that, around 160 drugs have genomic information.a smaller list, and it's -- most of them are shown here on this page have genomics in thedosing and the administrations section. and that's the section that is supposed to beread by prescribers, and it is read by the app makers that prescribers use, at least.it's read by insurance companies, and it's also read by litigators. so, there's the kindof the three main audiences of that section.

and these are examples where, in some caseshere -- from here up, changes in tumor. some of these where the fda has approved the drug-- you know, this drug for only patients who have this translocation or this translocation,or this -- mutations in this gene for this drug's case. and the list goes on. and so,you have the examples where it's very specific, and you have others where it's really moreof general prognostic terms. you have also germline examples. normal dna examples whereit could be metabolism, it could be the immune system, in terms of hypersensitivity reaction.it could be a response to certain medicines, in the case of cystic fibrosis. but the ideathat genomic information is already out there for use is something that's been around fora while and is growing each year. the fda

has a nice site that has these examples thatone can easily pull up. now, very few of these examples are such thatit would be malpractice to manage a patient without testing. certainly, in the germlineside, i would say, really only the abacavir example would be a malpractice type of situation.but rather, there's information saying, “if you have this genetic information, here iswhat you can do with it,” in terms of starting with a lowest dose, or starting with a higherdose, or whatever it might be for that. and so, it's the type of information that is valuableif you have it already, but unless you're already a doctor, you're not going to necessarilyorder that test. and so, you see some areas, like hiv with abacavir that pretty much everyhiv patient is evaluated for this particular

hla molecule -- if i get that arrow back,to identify risk of hypersensitivity reaction, whereas if you take some of the other examples,and it's much less common for the genomics to be applied. even for the same gene -- thisgene here is important for nausea and vomiting, for anti-depressants, pain control. you seethe psychiatrist do a lot more testing than the oncologist for -- even though they -- youknow, the oncologist might use a lot more of certain medicines. and so, some of it is cultural, if you will,in terms of early adopting personalities versus otherwise. the other thing is even withinan area like cancer, where there's a lot of activity around tumor sequencing, there'sstill a lot of elements where the germline

-- the normal genome is important for medicinesthat are there. so, pain control, nausea and vomiting control, anti-depressants, stimulants,blood thinners, all have a genomic component in the dosing administration section. andat least some centers are now using more of this information as it's applied. there areguidelines that are now being produced -- so, there's something called cpic, the clinicalpharmacogenetics implementation consortium. there's around 140 institutions in 23, i thinkit is, countries that are involved in putting together a large number of different guidelinesthat are available on the pharmgkb. -- pharmacogenomicsknowledgebase.org website. and these guidelines are now workingtheir way into the national registry and other places where guidelines are produced. andso, the field is maturing, from that standpoint.

the way things are being applied in pharmacogenomicsare really around the testing for avoidance in many cases, and that is identifying a patientwho would have a higher risk of a hypersensitivity reaction or a bad reaction try to avoid usingthat medicine. testing for inclusion -- so, some of the cancer examples. if you have aparticular variant, you are eligible to have that medicine. if you don't have it, you arenot eligible type of thing. stratification, in terms of whether someone is high risk orlow needs to have a different type of therapy. and then, testing for explanation. that'smy term for that. it's basically, someone takes a medicine, something really bad happensto them, and you want to figure out why. and so, often, you'll see testing done to say,“well, did that patient crash and burn because

of this genetic problem?” and if so, i willnow avoid a whole class of medicines, or was it something else of which now the whole therapiesare available to be used. and so, that idea is being applied. and so, you see it for areaslike pain control. now, there is no prospective randomized trialthat i'm aware of that has done genotype guided pain control versus not genotype guided paincontrol. but there are studies showing, for example, with codeine that if you have theextra copies of the gene, you'll convert it to morphine more quickly, and there have beena number of fatalities with neonates and in the pediatric situation to the point wheremany pediatric centers no longer use codeine at all. you know, they don't genotype thepatient. they just don't use the drug at all

because of these situations. it also reallyhighlights the issue: if you have an awesome therapy and a not so awesome therapy, it wouldtake an amazing bit of data for you to not use the awesome therapy. that's a genericname, not a brand name. and so, awesome is going to rule every time. but if you havetwo equal options, and you have to pick one for the patient, i mean, even a feather ofdata would cause you to shift one way or another, in terms of choosing that medicine. and so, really, much of medicine is tiebreakertype decisions, where you've got a number of equal options. you've got a number of differentways of treating pain. and if you knew that oxycodone will work much of the time but notall the time in someone who is deficient for

this particular gene, you would shift to somethingthat doesn't require that particular medicine. they’re not all shown here, but for example,hydromorphone, or some of the other drugs that don't require metabolic activation. theidea that 10 percent of patients cannot activate a pain med, and that somewhere between 3 and5 percent hyper activate a pain med means that somewhere between 10 and 15 percent ofall your patients are going to have trouble with those medicines. and that's a big percentage,just based on one gene and one class of drugs. and so, the idea that we could start usingthis information, not to choose whether someone gets lumanissa [spelled phonetically] or notblindly but whether shifting the scale to one versus another is the way we're seeingthis starting to play out. same with anti-medics.

again, there's a number of different medicines-- if you're an anesthesiologist, you're used to using ondansetron, and you don't want touse another medicine. ondansetron is the one that's worked. anesthesiologists are a veryfine-tuned group of individual -- they -- individuals. they have very clear protocols, in terms ofhow they do things. they are the most objective -- or at least, i think the most objectivepractitioners of medicine among all the different specialties. and they need a good reason tonot do what they normally do. and so, we're seeing even our own anesthesiologists nowusing cyp-2d6 to tip the scale from using ondansetron to one of these other medicines.i suggested they just stop using ondansetron, but no, that's what they're used to using.and then i said, well, it's only three percent

of patients that have extra copies of cyp-2d6,so do you really care? and they looked at me funny and say, “we worry about a onein 1,000 event. three percent is huge, in our world.” and so, again, it comes back to the context.if you ask an oncologist if he's -- or she is worried about three percent of the patientshaving nausea and vomiting, they would say, “we would be delighted if it was only 3percent.” if you ask anesthesiologists, they're terrified of 3 percent. so, again,context really matters as we see this start to be applied. and then, the last little backgroundpiece is around -- is there enough of -- are there enough patients with these abnormalitiesto really make it worthwhile? and so, we've

seen places like vanderbilt and moffitt andothers start to do pre-emptive type genomics. and so, this is a paper from almost two yearsago now, came out of vanderbilt, and they looked at this example for clopidogrel inthe cardiac situation, one of the statins of -- this genotype affects muscle pain forthe statins. these genes influence the dose of warfarin, the blood thinner. this geneinfluences whether you're going to get severe neutropenia from this medicine, for arthritis,and dermatologic disorders, g.i. disorders, and leukemia. and here's -- in the solar organtransplant setting, this particular immunosuppressant. and if you look in some of them, there are-- there's very common -- yellow is at least one actionable variant. some of these, it'squite common to have an actual variant. others,

it's much less common. and then, it's veryrare for there to be extreme risk type of variant. but when you add it all together,86.5 percent of patients had at least one actionable variant, and an additional -- almostfive percent, had a high risk variant. and so, it ends up being 90 percent of patients,just with these five examples, over 90 percent of patients had something that was actionable,in terms of pre-emptively preparing for the choice of medicine. now, if you are choosing a statin, and youdon't have genomic information, you might want to just start with one now. but if youhad that information already, it would cause you to shift from one versus another. justlike you may not really want to know that

someone's organ -- or you know, renal dysfunction.but if you know they have a creatinine clearance of 40, you're going to adjust that renal excreteddrug, even if there's not a randomized clinical trial showing you exactly what you shoulddo. and so, what we're seeing now is this -- with these pre-emptive strategies, moregenomics being done when the patient arrives at the health system so that it's pre-loadedinto the electronic medical record, and it can be acted on as we go forward. but these-- and the follow on studies have shown that it's not the minority of patients that havesomething to think about. it's the majority that have something to -- that -- to preparefor, as you change practice. now, there's still -- shifting gears to themore basic discovery elements, there's still

a lot of discovery that's needed. there arevery few precision medicine, pharmacogenomics, whatever you want to call it, genome-wideassociation studies, and even fewer where next generation sequencing has been applied.they're starting to come out. but if you look in the nhgri gwas catalog, it's somewherearound 4 percent of the phenotypes are drug-related in some way. and that's any type of medicine-- psychiatric meds, cancer meds, et cetera. a very small number of investigations havebeen done. and if you look at the number of patients and people across the country thatget a prescription medicine, and you look at what those prescription medicines are,very little has been done to really understand, how do genes influence the choice of medicine,the choice of dose, the choice of monitoring

for a given patient? and so, there's a lotof work still to be done for that. part of the issue as well is that replicationdatasets are difficult to obtain. so, one example that wasn't planed, but that cameonline this morning was a study that we did out of the nci's clinical trials cooperativegroup. in this case, they're -- one of the groups called the alliance did this particulartrial. used to be cancer/leukemia group b, was the historic name for that particularstudy. and so, this was in prostate cancer, people who had advanced prostate cancer, andthey were given this particular chemotherapy drug, plus a placebo, or this chemotherapydrug plus an anti-vascular agent. and the bottom was that the addition of the anti-vascularagent did not change overall survival for

the patients. it did change progression free[spelled phonetically] survival, but overall, survival was the primary end point. but wecan go in, then, and look at the different toxicities that occurred for these patients. and so, it's just over 1,000 patients, ofwhich 864 are registered for the pharmacogenomic study. and then, when you filter out someof the filters for population stratification, it ends up with just over 600 patients onwhich we could study for the initial phase. and there's a number of end points one couldat for these kinds of studies. so, the chemotherapy drug causes neutropenia and causes neuropathy.the anti-vascular agent causes high blood pressure, protein in the urine, clotting,and bleeding. and so, we can look at these

different features and look at, what are thegenome predictors of these particular end points. i'll skip that in interest of time.so, one can go in and go a model -- for example, for neuropathy, which i'll show you some datafrom right now, that was the paper that came out this morning. and for neuropathy, onecould go in and include, as well, these variant co-factors. so, if someone already has longstanding diabetes,they're going to either have or be at high risk for neuropathy. one needs to know that,as we're looking at drug-induced effects. what is the baseline that's there? so, themodels we would use in the past were relatively simple. here's a group of patients. treatit with this group. this drug causes neuropathy

as one of its side effects. we look at whoexperienced neuropathy. maybe we can normalize it per dose so we can take into account doseadjustments that had occurred. and who completed the two years without getting neuropathy?and so, that's very much a yes/no type of model. but unfortunately, we have all thesecompeting risks that occur. there are patients that stop the drug because their disease progressed.there are others that died while on therapy, some that had some other sever toxicity thatcaused them to stop taking the medicine, others still that just got fed up with being on thetrial and decided to go do something else. and you know, they're allowed to do that,and we have it into account. and so, by taking a more complex model, acompeting risk analysis model, one can then

go in and functionally do censoring aroundthese different events. because if someone withdrew from study after the second cycleof therapy, they didn't have enough time to get neuropathy even -- they might've gottenit, particularly to include them in the no-neuropathy group would really be a false thing. and partof it is because the -- what's shown here are a number of different end points. in blackis death or disease progression. green, it's other treatment associated adverse events.and you see early on, it's these other events that are occurring. and so, there's a lotof other events, non-neuropathy events -- neuropathy is shown in red -- that are competing here.and so, these patients that have these events never had a chance to really get neuropathy.and so, we need to model them accordingly.

and the reason for belaboring this point isthat there are so few pharmacogenomic studies, i would hate for us to go and do some anduse inappropriate statistical models. we -- every one of the studies that are done are precious.we don't have enough. you know, if you're doing studies in diabetes,there are so many diabetes studies that you can have a few crappy ones because they'llbe overwhelmed by some great ones. but for pharmacogenomics, we don't have enough studies.we need to do them well. and so, going in with some kind of, you know, hey, “i haveexcel; i can do statistics” type of mindset really needs to be overcome. and as we'rethinking about these approaches, and especially in places where it's hard to find good statisticians,we need to be doing more collaboration with

the people who know how to do these more complexmodels. i don't know how to do them. i could go and buy a stats package and stick it onmy pc and error apple and do something with it. but by having folks that really know howto do this stuff, one can really start to get the truth here. now, as we look at neuropathy as the end point,we get these genome-wide associations, these so-called manhattan plots. now, this particularmanhattan plot has a couple features to it. first of all, you can tell a real statisticiandid this analysis because, if it was me, each of the chromosomes would be a different color,and it would be -- each dot would be a little bit bigger, and we'd actually be able to seeeverything. second, you know, i had to draw

circles around the dots so you could actuallysee it because of that. second thing, it resembles manhattan, kansas much more than manhattan,new york. and so, we have a few hits, if you will, nothing up into the 10-8 range, butsomething to go forward. and then, as we put in more of the adjustment factors, the clinicalfeatures, we do start to get some of the genome-wide significance with some of these -- some ofthese features. so, what's shown here is a list of genes thathad some level of genome significance, and as we looked at the functional data, had somesort of biologic plausibility. and i show this for two reasons: one, because we didit, and the follow on slides are featuring this particular example, the vac14 gene, butalso because this is not a very smart way

to do it. we did this analysis this way, butreally, what it does is it opens us up to be fooled by naming. so, if anything has theword cell death in it, automatically, it's the best candidate gene you've ever foundfor whatever the phenotype is that you're looking at because you can make a great storyfor why it is the gene. whereas if it's kiaa7439, it doesn't really have a ring to do it -- you'retrying to -- actually, it probably does have a ring to it, but the idea that you're tryingto use that and come up with some kind of plausibility -- no one knows what those genesare. we don't have a clue. and yet, they might be the right gene, in terms of how they'regoing forward. and so, at -- you know, as we're doing genomewide studies, we need to thinking about are

we being fooled by the naming, or are there-- and are there more things that are going to be important for us. so, taking that caveatin mind, we then looked at this variant in vac14 that was one of the -- one of the candidategenes with functional stability. this stabilizes one of the proteins that's involved in charcot-marie-tooth,an inherited peripheral neuropathy syndrome. and so, looking at vac14, the variant -- therewas an increase in the risk of neuropathy. there was a gene dose effect, if you will,that occurred, and so, that was encouraging. we then went in and did some studies withips cells, using some cells that have been differentiated into sensory neurons. and i'll point down to one particular circleso you don't have to get lost in what is figure

three of the paper. what's shown in this particularpanel right here is increasing dose of the drug, and looking at the relatively branchingunits. so, a measure of neural outgrowth in this particular study. and you see, there'sa difference between those that have had the vac14 gene knocked down compared to thosethat had had a scrambled control in the particular experiment. there was a series of additionalstudies that are shown in the paper to show this. and so, that's interesting, from a biologicplausibility standpoint that we've now shown in an in vitro system that we can see a differencebetween the cells expressing and not expressing -- at least, not over expressing or highlyexpressing the vac14 gene. and then, one can go into vac14 knock out mice -- heterozygous;the homozygous aren't viable.

and what one can see is that the -- there'sa difference between -- sorry, this is a ugly looking slide. the journal hasn't redrawnit for us yet, so this is a little bit more raw than you're used to seeing. but the bluex are the mice that are heterozygous for this gene, and we're traded with a vehicle. andthe red x are the same mice treated with the drug. and you see a real drop in the hindpaw withdraw threshold, one of the many measures of neuropathy that was performed in theseparticular mouse studies. but you can also see that even in the -- even in the wild typemouse, there was a drop in neuropathy and in differences that had occurred. so, we havethis in vitro data that -- or ex vivo data that shows that we have a difference in thesegenes. but it's all very unsatisfactory. and

part of the reason why is that replicationdatasets for pharmacogenomics are very difficult to obtain. the clinical trial that i justshowed you that we used for this discovery, we're not going to go and do the trial again.you do the trial once, and then the winner, if you will, of the trial goes on and facesthe next opponent in the next clinical trial. you don't do a trial twice just so you havea dataset on which to do replication. it's not considered ethical, in many cases. so, you have a situation where you're reallystuck with a dataset for discovery and some level of biologic plausibility, hopefullywith some mouse and cell line data to help back it up. and then have to really go forwardinto something more prospective, in terms

of intervention. and so, there's a lot ofopportunities to think of creative ways of generating the kind of data we need to reallybe able to do more robust replication at the same level that someone working on diabetesor many other diseases have, in terms of these large cohorts. the other thing is that -- justwant to go back to the title slide here. there's a bunch of different people that have to beinvolved in these studies, and this is just a small list. this isn't even the completeauthor list. but the idea that a number of folks that are running the clinical trialthat are experts in the different types of phenotyping in the clinical trial, that areexperts in the ips cells and in the knock out mouse models that are involved in highlevel statistical modeling, all these folks

have to be involved, and preferably from thestart, and not involved in the usual way. so, often, what will happen in statisticsis that there's something that is referred to fondly by statisticians as the statisticalautopsy. they brought a dataset, usually in pieces, and he says, you know, “can youfix this?” as opposed to, “hey, we're going to do a clinical trial. can you helpus design a really robust trial that will help us answer a question?” and the ideathat these folks get involved from the start, it makes a world of difference. it's a painin the butt because, first of all, they don't speak any language that i've ever seen. it'sno language known to man. it's all a version of greek that has a lot of numbers in it.they think in 12 dimensions, not the ones

that we're involved in. but they can makeus so much smarter as we go forward. and so, you know, team science -- not teamas in tennis, where you have a team of physical therapists and psychologists and hitting coachesand whatever to help you, but team as in hockey, i guess that's what's on right now, or basketball,where you take one of them out and you're in trouble. and so, we need really more ofthat to be happening. now, one of the things we also can do is once we have enough literatureis not do the discovery component, but go straight to a replication type component.so, this is the study that was -- came out a couple months ago from our group, just as-- you know, i had the slides ready so it was easy to use. this is a study of ovariancancer. it's a really quite a boring study.

it was docetaxel or paclitaxel, two chemicalcousins of each other. and then, combined with this one dose of platinum drug -- normalizedto blood level, and that was the randomization. and the end result was there was no differencein survival or -- overall survival or progression free survival. and this has become the standardof care. so, over a decade later, after this trial, the standard of care for the treatmentof ovarian cancer is still this particular regimen. now, if you're involved with ovarian canceror know someone who's had it, or you might've gone to the ovarian cancer clinic, you'llhave noticed a number of women using walkers. and you think, how unfortunate. this disabledwomen got ovarian cancer. you know, that's

-- how unlucky. typically, what has happenedis that she walked in, started her therapy, but the therapy is so toxic to the nerves,peripheral nerves in particular, that she no long can feel her hands, and has troublebuttoning her blouse and whatever else, can't play the piano very well anymore, things likethat, and has trouble feeling her feet and -- to the point where she's stumbling andgoing to fall and hit her head, and so is now using a walker to be able to keep forward.and the way i was taught, and the people that were taught around my era, and really, upuntil recently, were taught that in cancer, you need to almost kill the patient in orderto kill the cancer. and that was the mindset that was done, and certainly, it was appropriatewhen you take drugs like the alkylating drugs

from back in the '50s and '60s where, youknow, the more you use, the more killing you get of the tumor. and so, you need to surfthat careful line. and so, is a woman really willing to not haveas much therapy to control her cancer in order to avoid the nerve toxicity? and that's sortof -- has kept things back. and so, we were looking at this. we have this large dataset.it had been maturing now for a number of years. we had robust toxicities that had been audited.so, we decided that we were going to go and do a discovery study. but then, we realizedthat there had been many discovery studies already done. there were also studies lookingat the, you know, nerve function biology. there were inherited neuropathies. there was-- some of the end points that had come out

of these studies where pharmacogenetic orpharmacodynamic genes. and there was a lot of underpowered data sitting out there alreadythat really was cluttering the literature. and so, why not go in and just say, all right,any variant that has been shown in some study -- no matter how small, to be associated withneuropathy, let's evaluate it in this context. and so, it was kind of a bummer because wewere looking forward to doing, you know, a million snps. instead, we ended up doing 1,261,which seemed so old school to do such a small number. but those were the ones that had somerobust association to go forward. and we took the 1,000 women from the study,our statisticians pulled out -- using a randomization algorithm, half of the patients, and evaluatedthe 1,261 snps, and looked at the association

with grade of peripheral neuropathy in thosepatients. sixty-nine snps were -- met our threshold, statistical threshold. and so,these 69 variants went into the next 500 women on this study, and it was some directionalcorrection. there were four variants that came out to be important. and one thing fromthese four variants and these four separate genes -- firstly, there was about a doublingof risk. so, each one of the variants contributed something, but a doubling of risk is not enoughto change practice. it's enough to be interested, but it really needed more than in time, interms -- to change practice. usually, somewhere around four times odds ratio of four willchange practice. and so, that was -- that was interesting. but if you -- if you lookedat the accumulation of these variants, the

-- when you look at all of the variants, thepopulation attributable risk was about 85 percent. so, a lot of variability could beexplained by these particular variants. and one could go in and look at a -- kind of a,i guess, variant dose effect. the more variants you had, the higher your risk of neuropathythat was seen in this particular study. well, that's interesting. it was a replication,not a discovery dataset. so, that's interesting. but what about the original problem of needingto really induce toxicity in order to cause these problem? to cause control of the disease?we looked at these variants, especially those in the highest toxicity risk. we looked ata bunch of different cut points, but this is the one we put in the paper. and basically,we could not find a way of separating out

-- in terms of progression free survival oroverall survival based on these predictors of neuropathy. so, what this does is opensup the idea that neuropathy is not required in order to control the cancer. and that thegenes -- at least, from our study, appear to be regulating or been associated with,i should say, neuropathy, are not predictors of outcome in these women. and so, the ideathat we can use this data to now do some additional prospective studies, and we are, but alsoto do start something some drug development. now, some of these examples, for example,bcl2, there's already an anti bcl2 therapy that was approved earlier this year for atype of lymphoma that is on the market. and then, there's at least the possibility ofdoing -- of intervention with some of these

other types of genes. and so, it opens upsome drug development opportunities in addition to the possibility that it could predict,prior to the start of therapy, what the level of toxicity a woman might have. we have nowgone further, and using next generation sequencing technologies, identified that there are someadditional levels of risk that are out there. and so, in this particular study, we're takingpatients that had very severe response to these medicines. so, these anti-cancer drugs.so, you give them one or two doses, and their nerves just melt away. and in that case, thepatients seemed to have an underlying charcot-marie-tooth, a peripheral neuropathy syndrome. now, ifyou go to a neurologist, they would probably pick it up every time. but if you go to anoncologist, they don't really notice if someone

shuffles in, as opposed to walks in, or ifsomeone has a bit of a limp. or they certainly don't look at someone's, you know, reflexesin most cases. they don't take people's shoes off, and look and see if they have high arches,which is often associated with these syndromes. they're -- it's just not something that ispart of normal practice. and so, it's basically that iceberg underthe water. just because you didn't see it doesn't mean it won't sink the ship. and so,we're now finding that -- and others are now, too, that these inherited neuropathy syndromes,the moderate penetrants versions are out there. they're really not noticeable clinically verymuch. and they're really waiting to cause problems. and so, these types of discoverypieces are now allowing us to look at our

preemptive strategies, to the point wherewe're now looking for charcot-marie-tooth syndrome genes, cardiomyopathy genes, a numberof -opathy genes, prospectively in all of our patients to identify those few that aresuper high risk of these very severe toxicities. now, part of it also is trying to come backto those dollar signs that i mentioned earlier. and that is -- we can really use the costelement in our favor as we try to take pharmacogenomics from being the discovery of science, includingsome of the stuff i just showed you, into implementation type science. and so, one example is with this anti-fungaldrug called voriconazole. it's used to treat fungal infection. but we use it, in contextof a cancer center, to -- as prophylaxis for

myeloid leukemia patients. so, without prophylaxis,there's a high incidence -- at least, high in our mind -- could be as high as 30 percentincidence of death by fungal infection. but prophylaxis has reduced quite dramatically.well, i -- as some of you will know -- yeah, i took that out, as some of you will know,this particular drug can be inactivated by a liver enzyme called the cyp2c19. and a substantialamount of the population has an overactive ability to get rid of the drug. so, it's rightaround 28 percent of moffitt cancer center patients, similar across centers in the unitedstates have a high ability to get rid of this drug. and so, a normal dose for everyone elseis very inadequate, and you can't really measure blood levels. you're never getting to thepoint where you're getting prophylaxis. you

might as well not be using the medicine atall, or you need to know to switch to a different medicine for this particular case. so, there's a number of clinical trials -- clinicalstudies, rather, that have been performed on this. we had the clinical case done nicely.but if you go to your administrators and say, “hey, we have this strong, clinical case.we would really like to implement this in our practice setting,” what you'll get is“that is a very -- that's very interesting. we -- you know, patient safety is paramount,so we would like you to go and do another clinical trial, and we would like you to dosome cost effectiveness studies.” but if you go and do some of those economic analysesusing actual data from your center, you get

a very different response. so, what's shownis some analysis that was done by neil mason at our place around the cost of fungal infectionsand the treatment of and the genotyping for fungal infections, in the context of myeloidleukemia. and basically, what it came down to is that the myeloid leukemia patients thathave a fungal infection cost us just under $30,000 more to manage in the first year thanthe patients who do not get a fungal infection. and what happens is if you can prevent onecase, you can pay for testing in spades. and this sort of situation -- when we went toour administration and said, “hey, we would like to do this. here's the clinical casefor doing this. here's the economic” using our actual data, not data from kaiser or someother place, but our actual data, there was

nowhere to go but yes. and so, one of thethings that we're doing now a lot is we try to take our pharmacogenomic examples and bringthem all the way across the line to patient care is working with these economic folks,the folks in finance, the people who never thought we'd eve have coffee with, much lesscollaborate with, to make the case. because if we can go in and say, “here is the clinicalcase” because you wouldn't go forward without a good, solid clinical case, “and here isthe economics,” you can get to yes much quicker. and i drop that in there not so muchfor the nih folks because it's a different model. but for the folks that are going towatch us later and see -- as they think about their health system, if you can make the dollarswork, almost anything will happen. and so,

the idea that we only work on the -- finetuning our next gen sequencing machines is only going to get us so far. we need to havethe rest of that, and we don't need to do that. we don't need to be those people. butwe need to be collaborating with those people to really make this forward. all right, so we're going to come back -- swingback to the way genomics is starting to change practice, especially in the area of cancer.so, as i mentioned before, we have the germline genome for same things like peripheral neuropathy.cancer genomics is really becoming a normal part of care. and often, it's for the selectionform amongst equals. so, if you have a new diagnosis of lung cancer, there are a bunchof fda approved options that might be the

right ones for you if you have a particulargenotype. and so, having a small focus panel that allows you to make that decision is certainlynormal practice in many, if not all, of the united states space cancer centers. but afteryou run out of the first and second line options, which is where most of the randomized clinicaltrial data is, you have a patient that's fit, wants to do more. you often don't know whatto do next. i mean, you can pick a medicine. but what is really something that's actuallygoing to help that person? and the idea that genomic information will influence is becominga reality, coming back to this tiebreaker type medicine. so, you have these two clinicaltrials, or this clinical trial and this off-label use of a medicine, you know, what do you pick?and the genomic data is often driving that.

and it's really changing the change practiceis happening. so, it wasn't that long ago that a tumor in the colon would be calleda colon cancer. or maybe it has glandular formation under a hematoxylin and eosin stain,so it's an adenocarcinoma of the colon. or maybe there was a kras mutation. mutationof the kras gene. and so, it's a kras mutant adenocarcinoma of the colon. you can kindof see the theme that's building up on that except that's just been blown on its head.now, this tumor is a p53 mutant, ep300 deleted ddx3x lost cancer with all these variantsof unknown significance. with the handwriting being one of our clinicians who, before gettingthis report, was a world expert on this disease, and then after reading this report is a babblingidiot, trying to figure out what in the world

they're going to do with their life becausethey don't understand cancer anymore. what we found is that, you know, there's theold information, data, knowledge, wisdom. we have -- we're generating lots of information,lots of data and information. we have very little knowledge, and no wisdom. we are justso data rich that it's paralyzing. and so, the idea that we can take this and try tomove it forward is clinical. and what we're seeing, you know, we have initiatives nowat the nih that are helping drive this, but there is such a lack informatically informedpeople involved on the clinical end -- you have people that are involved in health it,electronic medical records. but people involved in trying to take information about a patientand help turn it into a decision are very

few and far in between. and so, there's abig opportunity and a big problem sitting there as we sequence all these patients. and,you know, sequencing cancer patients is not something you might do every once in a while.at large center like ours -- we're the third largest cancer center in the nation -- wherewe're sequencing about 120 patients a week. so, it's a lot. you know, it's not somethingthat we might do. it's something that we do. and so, it becomes a real problem very quicklyas we go forward. but it also opens up opportunities. so, this is a patient with leiomyosarcoma.very few fda approved drugs for this disease. they -- she ended up with metastasis to thelung. she received this doublet of therapy for three months, and it didn't work. thisdoublet for four months, and then it didn't

work. this kinase inhibitor, which is -- wasfda approved, finally, is for three months, and it no longer worked. she was fit. shewanted to do more. but she didn't know what to do. and so, we could just put her on whatevertrial we happened to have available, or we could try to look further. and so, we sequencedher tumor. and it's not so important, all the details or i would’ve made it larger.but the findings there came back and there wasn't anything she -- that her oncologist,who is an international expert on treatment of sarcomas, thought was actionable. but becauseof all this problem, we generated a couple changes in the way we practice. first of all, genomics -- pharmacogenomicsin particular, now has turned into the equivalent

of radiology at our institution. you mightlike to read your own ct scans, but a radiologist reads it anyway. and the same thing is nowhappening with tumor sequencing, and germline, for that matter, sequencing. that every singlecase is read by the first slide's medicine clinical service. and so, they all get seen-- a report is put in the electronic medical record. about a third of the time, the reportis “we have nothing to add to the pathology report. call us if you need us.” wordedit a little nicer than that. about a third of the time, there's a few changes. you know,this trial is no longer open or no longer accruing. you know, think about this optionas well. and then, another third of the time, there's some real complex work that has tobe done.

we've also generated something called theclinical genomics action committee. now, it's a molecular tumor board. but the reason wedidn't call it a molecular tumor board is that too often, molecular tumor boards areacademic freak shows. you go there and you say, “whoa!” this particular patient we'retalking about, she had a jack two amplification. so, you go in this group and you say, “shehas a jack two amplification.” and everyone around the room would go, “whoa, that'scrazy! that has been seen very often. next?” as opposed to, what do we do with this, andis it actionable? and how do we act on it? and so, we have a large group of differentdisciplines that are all in the one room, trying to work through what happens. and i'llcome back to a point about that, in terms

of the way our basic science colleagues arecontributing. but in this particular case, mahilda druda[spelled phonetically], our sarcoma medical oncologist whose patient this was, was youknow, presented the case. and a couple of the folks here in particular -- a couple ofthe leukemia/lymphoma/myeloma folks, they said, “oh, oh! jack two amplification iswell known in our area.” and at least, with in vitro data and some of the initial clinicalstudies, makes a patient more responsive to pd1 and pd01 inhibitors. you know, immunotherapy.well, immunotherapy doesn't work in sarcoma except in some of these patients that havethese particular features that make them more relevant. and so, based on the, you know,report worked up, she ended up going on a

pd01 inhibitor trial in august of 2014 andis on it. the previous therapies, the longest one worked for four months. this one's workedfor a lot longer than that. it will not cure her. she still has viable disease that isstable, sitting there, not decreasing, not really resectable. so, it will, at some pointin time, come and get her. but it has bought her way more time than before. and you know, this is an anecdote. this shouldnot drive your practice any way, shape, or form. but as we build up more and more ofthis, we start learning some of the rules whereby we can now objectively demonstrate-- and that's future tense -- that this genomic sequencing really makes a difference in termsof outcomes. now, whether it influences survival

-- i would certainly like it to. but eventime-on therapy is important. you know, those of you in the oncology area, the patientsthat haunt you and certainly haunt me are the ones that died three weeks before an fdaapproved therapy or a clinical trial, even, of a therapy became available that switchedto a more curative type of environment. you know, those gastrointestinal stromal celltumors where, now you know, once imatinib became available, they became a -- if notcurable, at least lifelong controllable disease. and if i had had the drug three weeks earlier,that patient would still be alive. and so, the idea that we're trying to cure peopleis definitely true. but the idea that we're trying to bridge them, have them living aquality life -- a quality life long enough

is also an important part of this. and weneed to be thinking about those kinds of end points as well. last little piece on this is we have severalof our cell biology and molecular oncology phd colleagues that now come to every oneof these meetings. and the reason they came initially was because we were -- we had examplesof patients with variants where there was no human data, no clinical trial data, nothingreally. but there was some cell line data. and so -- and i will -- you know, in the landof the blind, the one-eyed man is king. if we had some data, we could try to devise amore objective way of treating this patient than just picking a trial we happened to haveopen. and so, we looked at -- and had a couple

of the papers. well, it turns out, the paperswere from our own institution. there's a guy down the hall that was doing this work, andso we said, “well, could you come and actually, like, tell us the story behind the story?”you know, when he came, he said, “oh, yeah. this variant, you know, it -- you know, atleast in cell lines, it doesn't respond to sorafenib, but it does to some other drug.”and they're very -- they're terrified that someone is going to mistake them for a clinicaldoctor. you know, “listen, i'm not a clinical doctor. i'm” blah, blah, blah. “but ido know this about cells. i do know this about mice.” and often, we're a situation wherewe're trying to choose from amongst a couple of equal therapies. and even some cell lineor mouse data might be enough to say, “well,

let's go with this versus this.” becausewe're not talking about therapy and no therapy. we're talking about what we use first. andthen if it doesn't work, what we use second. so, the idea, you know, even when you lookat clinical trial in your favorite journal, you'll see this, you know, survival curves,and they're usually wide enough that you can see a difference between the two. and youthink, “oh, there's winner and loser.” no, there's first line therapy and secondline therapy. and you know, we often will say, “oh, the winner of that trial was x.”but then, if we go to a patient after the winner stops working, we don't say, “well,we're going to try the loser therapy on you now.” it, rather, is the next best option.and so, our whole mindset, in terms of how

we do trials, how we interpret them, how wego forward, needs to be changed as we try to apply things like pharmacogenomics intopractice. so, i'm going to skip that, in interest oftime, and hit the list little bit. you know, one of the things, just a reminder, thereare a lot of genomes that patients have. you know, this -- it's not just the germline andthe somatic. as a matter of fact, within the tumors, there are a bunch of different genomes.it's not like there's lung cancer. it's a bunch of different types of flavors of lungcancer within that particular mass. and so, we don't really have good models for thinkingabout in vivo, or ex vivo setting. how do we deal with all these different populations?you know, the population -- the folks dealing

with population evolution or whatever youwant to call it, really haven't jumped into the practicalities of what we're trying tothink about, and how do we evolve new ways of therapy? and i think as pharmacogenomicsmatures, in many of the diseases we're working on, we're going to be seeing more and moreof those elements. the last little piece is as we go into people,we really need to have more diversity. so, when we talk about america being a meltingpot, but really, in some cases, it's really more like a carton of eggs. it's a bunch ofpeople living in the same place, but not necessarily in -- with the same diversity. and the reasonthis matters is that this is a study that is not yet published but hopefully be willsoon, where we looked at 127,000 patients

over these three disease areas over thesethree different time periods. todd knepper, myself, and two of our fda colleagues. andbasically, what we saw is that over time, there was a doubling of the number of countriesthat were involved in pivotal clinical trials -- the approval trials for the fda. so, globalizationhas occurred. no surprise, but we just quantitated it in a little bit bigger fashion. and youcan see, you know, over time, here's the different countries that got involved and you know,that's great. but we found is that in 1997, just over 90percent of pivotal clinical trial patients were self-described as caucasian. in 2012-- and we now have 2014 data, it had gone all the way down to only 82 percent of theclinical trial participants being caucasian.

basically, a lot of the growth that had occurredmeant that there were more white people on trials, except they were from the ukraineand argentina, as opposed to actual diversity happening with globalization. and that matters.and look around the room. look around wherever you're going to be tonight. look around everyplace. there's a bunch of people from a bunch of places that are needing help, just evenwithin this country. and the idea that we need to be doing these trials is importantand -- at the pharmacogenomic level, it matters. what's shown here is a number of differentgenetic predictors for pharmacogenomics. and what's on the x axis is different countriesseparated by continent. so, here's africa. here's asia. here's europe. here's the middleeast. here's central and south america -- sorry,

here's central america, and here's south america.and using the data from the new england journal papers, here is the predicted average weeklydose of warfarin. very low dose in many of the asian countries except india, which ismuch more similar to europe in terms of its metabolism. much higher dose needed in africa,et cetera. here is a risk of gi toxicity from one of the anti-malaria drugs. again, a lotof diversity. risk of muscle pain from simvastatin. again, a lot of diversity. and then, if youlook within countries, taking the new england journal study, here's the average dose fora u.s. population. very similar in mexico. the nigerian and ghana populations need amuch higher dose. the chinese and japanese population's a much lower dose. but there'sa lot of variability even within that geographical

label. and the reason for bringing this pointup is that as we extend pharmacogenomics into the clinic, we need to be making sure thatwe capture some of this diversity. we're doing a project now with china. theyhave 56 ethnic groups in china. so, there's the han chinese that everybody knows about,and then there's a whole bunch of other groups. and you know, 1 percent of the chinese populationis a lot of people. so, these groups are not very common, except it's, you know, millionsof people that are in each of these groups. and so, the idea of trying to understand thediversity. you know, we heard about -- from a couple lectures ago, population genetics.we need to know this sort of data, not just so we could understand where we came from,but to understand where we're going, in terms

of trying to make good health policy and individualizedtherapy for our patients. so, i'm going to stop with this particularslide on the precision medicine initiative. and the way i view the precision medicineinitiative is it's a 3d printer. it's going to be a massive million-man or whatever personcohort, the majority of the money spent for that. but it is a 3d printer. the 3d printeris pretty cool. but what's even cooler is what you do with it. and so, this cohort thatis being built is the start of some amazing science. and so, i would encourage all ofyou, no matter what area you're working in, to not be thinking like we often do, whata blank waste of money, but rather, how can this be a strategic advantage for helpinganswer some of the questions that we've been

working on? and i'm not part of the pmi. idon't even have a t-shirt with the pmi label on it. i'm an outsider, looking at it. but after i got over the “what a waste ofmoney this is going to be” part, i looked and said, “wait a minute, this could bereally powerful if we get involved and help shape it now.” because 3d printers can makesome really stupid little plastic toys, or it can be a new valve for a heart. and it'sup to us what we use our 3d printer for. but i think there's a great opportunity to takepharmacogenomics and really push it to the point where it's known to be useful, and areaswhere it's known to not be useful, as opposed to where we are now, where we think we havesome hits. we think we have some misses. but

we really don't know what it's all about.so, i'll stop at that point. little bit of time for questions. and thank you very muchfor your attention. andy baxevanis:thank you, howard. we’ll take questions at the podium, and we’ll see you all againnext week. thank you for coming. that was great. [end of transcript]