 So, thank you to Terry and Eric for organizing this and for inviting me. I was a late addition to the speaker list and I'm hopeful that I can contribute. I spent some time last night reorganizing as I heard in more detail what it is we're going for here. So, I'll tell you that the way I've thought about this is in thinking about genomic variants to guide treatment. I wanted to talk a little bit about what's been done, a little bit about what we're doing and then an idea of how I would think about this in terms of large cohorts. So, these are my financial disclosures, none of which influence my talk today. So, with the title of using genomic variants to guide treatment, I think if you look at the highlight list of what's out there over the last decade or so, certainly as was mentioned last night, PCSK9 story is one of the biggest on my list of important variants guiding treatment. I guess thinking of other highlights on my list, I think the APO L1 story, which has been associated with, as many of you know, with risk for end stage renal disease in African Americans and is showing itself to be an autosomal recessive risk in what, as a major driver in a complex disease, getting into relatively rare disease, the story that has been put forth by Haldit's most prominently and others of changing, are thinking about Marfan syndrome from a structural, purely structural disease to a chronic inflammatory disease through what they found with fibrillin and then with the TGF beta receptor genes. And then certainly the stories out of pharmacogenomics, which many of you are involved in and know much more than me, are all some of the highlights I'm sure other people could add enormously to this list, but those would be some of the things I'd be thinking about as variants guiding treatment and the highlights of the recent decade. So Terry sent this email to me and I'm sure to many saying that, you know, what we're really trying to accomplish here is to think about sequencing and complex disease to reduce disease and help patients. And the key questions, as you know, I spend most of my time involved in clinical genotype, phenotype correlation, so I'm a clinical geneticist and run a clinic at our hospital for adults, it's about half my time. And so like many of you and many around the world, we're excited about using whole genome sequencing in cohorts like this. So doing whole genome sequence to find genotype, phenotype correlations, we've been engaged in this and we've started a clinical service around that. So in that arena, this is a large cohort for us. And the other part of what we're excited about and we're doing as far as looking for variants to guide treatment is I'm involved in a U01 project led by Robert Green, funded by the NHGRI, where we have a 2x2 design in a small trial and of only 200. But we're looking at physicians and their patients together as the subjects of this study and looking at standard of care versus whole genome sequencing and looking at the support needed to actually use that within the clinical care model. So a little bit bigger cohort than our families, but not that much. Is that the last slide? No. I hope not. You shouldn't put all your sequence data on the next slide. That's all that animation. Just call for the next slide please. Okay. Next slide please. So based on some of the discussion last night, thinking about a uniform database to go to, I think one of the things around clinical medicine that I've always been impressed by is the amount of knowledge that's carried around by clinicians but never published or reported in any way. And I think that as we get a unified database that people can go to, I think creating opportunities for clinicians to annotate changes based on single cases should be something that we should think about. If we could set up the structures to align the motivations, it could be a powerful tool. And as I said right now, there's lots of information that doesn't get out into the published world. I don't think we should stand for that as we get into the genome error. So in thinking about this use of genome variants to drive treatment, I like all of you are excited about kind of the bottom row there where we're getting to ultimately. But I'm a little worried about this decade. And I think there's some threats to our use of genomic medicine which I wanted to just talk about a little bit. So one of the things I'm concerned about is that we're not really in control of the pace at which genomic medicine will be incorporated into clinical medicine. And I think there's a couple reasons for that. One, you've seen the slide last night and in many ways the other way to represent this that I use is to think about the cost of DNA sequencing like this in 1985 in or around when the Human Genome Project was proposed. It would cost you about $10 to get one nucleotide sequence, 1991, $1 per base, 2001, 2000 bases. And as of last year, I guess this is debatable by people who know more than me about this field, but about $100 million. So for the cost of what many of us spend at Starbucks buying our coffee, you can get an enormous amount of data. So this, as you know, has gotten the attention not only of scientists and clinicians but also of companies that are interested in providing DNA sequence sometimes directly to consumers. And this was mentioned last night too, this idea that people will go directly to companies outside of healthcare and get their genome data. But they will bring it back to healthcare and I think the enthusiasm for this will rightly be driven by important stories that are being told of the use of genomes to help people. This is a story about the Milwaukee case that you're all probably familiar with. And as genomes get introduced into medicine outside of our control and probably faster than many of us might choose, I think there's some threats to genomic medicine. I think the biggest is economic, but then there's possibly some political ones too. So the Affordable Care Act was discussed last night. I think we've all seen projections like this. I don't think any of these projections take into account the cost that will be driven off of genomes as we start to use them widely in clinical medicine. And I think in the next decade we could see a real ballooning of costs around genomics. And my worry is that it will draw the wrong attention and people will start to wonder what we're getting for genomic medicine when there's costs without clear benefits. Fears that will be acting on false positives. So now as people are ready to launch pre-natal screening of parents for autosomal recessive disease, I think it's instructive to remember that one of the best examples of how that can be done started out with a false positive on the list of variants that were screened. The I-148T was later removed from the standard screen. But people that may have made reproductive choices off that or may in the future make it could draw bad press and pay that attention to our efforts here. Many people have cited the potential problems associated with false positives around BRCA. There undoubtedly will be others. Sorry that's so small, but we can all rest assured that it will eventually come to this. This is a man being carted away by the police saying that his genome made him do it with a dead man in the background. So concerns about drawing the wrong attention have made me wonder by the end of this decade when I think some of that attention could be focused on us, what will we have to show as far as improving the effectiveness of healthcare? So Eric told us that we should focus back here. So in thinking about that and thinking about cohorts, I was going through a thought experiment of what the priorities should be and choosing targets for which genomic medicine can and should guide treatment. I think one of the or some of the recommendations would be that we might choose to focus on things that we're really bad at in clinical care as priorities. Things that are driving or will drive significant healthcare spending. Things where they're a moderate impact could lead to significant improvements in health. Priorities and targets chosen which affect a lot of people and obviously they should be things with a strong genomic basis. So in thinking about this I thought of smoking cessation which meets many of those priorities, but not all. And I guess I would wonder if we shouldn't be thinking about a grand vision of how to use cohorts to do a genome project for this decade. And so if we fill in the blanks I'm sure people come up with all different fill-ins here, but if we were to promise at this point that cohorts or one large cohort would set out to understand the genomic basis of a specific problem and that we would anticipate that this genomic understanding would contribute to the development of significant therapies to control that problem by the end of this decade. I think if we had a strategy that looked something like that it would help us to meet whatever criticisms might be in the near future of genomics. And so I would propose that one thing we think about is this problem. So this is the CDC's obesity trends in U.S. adults over the last two decades. And you can see here from 1990 when no state had more than 15% of the adults obese to 2000 and then to 2010 by 2020 this map will only get I guess a deeper brown or burnt orange there I imagine. So this problem cuts across all age groups, all ethnic groups. I think it probably cuts across all institutions of the NIH, obesity is associated with cancer, diabetes, renal failure, arthritis, changes in longevity. And obviously it's a worldwide problem, not just a problem here though. We are world leaders in it. And so as far as a grand vision for a large cohort or several cohorts I would propose that we think about obesity or other problems in this way. And with that I'll end. Thank you Mike. Any questions? Let's open the discussion. Please. With respect to the question of not trying to identify causes of disease but the question on this topic of how to identify differences in the effects of treatment I'd be interested in people's views on whether the emphasis should be on the efficacy or, and I think Eric was getting at this point earlier, on the safety of treatments because I think there's in the past been a huge emphasis on taking treatments that are known to work and then looking to see whether we can find out whether they work more or less in other people. Which I think is, my view is largely futile. I think if they work then they largely work in most people. I think a lot of the time the huge efforts into trying to find out whether they work differently is not the focus. Whereas I think there will be much more mileage to be gained in looking for those people who get side effects and where I think it's quite reasonable to anticipate that there will be genetic determinants that have big effects. We've had a number of false claims for differences in efficacy. If I take statins, GIF 6 is a really good example of false claims of differences in efficacy, whereas SLC 1B1 is a very clear genetic determinant of safety of statins and then turns out to be relevant to the safety of other drugs such as methotrexate. So I wonder whether there should be greater emphasis on safety in the next 10 years and less emphasis on efficacy in the direction we go on to use the treatment. As a note of caution, many studies looked at efficacy and as of today, there are no clinical recommendations. For some reason, I presume it's the complexity of the phenotype. This has not made it into clinical practice. While I think I completely agree with the worry, it seems to be a naturally low-hanging fruit almost, but it has not panned out so far. Actually, I was arguing that efficacy, I wouldn't put emphasis on. I think I put the emphasis on safety. My comment is actually for both. Neither efficacy, at least not in brain disorders, the area I can speak for. It may not have got into the clinical armamentarium, but I think if you actually look at the evidence, there's really good evidence for genetic determinants of safety of drugs. And it's actually been relatively poorly studied. And yet there are good examples, whereas there are, I think, far few good examples on efficacy. So this is, yeah, pharmacogenomics is something I think I know something about. And I say I think the algorithm, the determinants of efficacy obviously vary by drug and by drug class, but tend to be pretty mushy. They tend to be driven by co-administration of other drugs. They tend to be driven by the disease substrate you're treating. And the likelihood that they're going to be single genetic variants with large effect sizes is small. The flip side is that there are these exotic and sometimes not so rare adverse drug effects that really do sort of seem to come out of the blue in an unpredictable way. And that for me means that there may be a big genetic basis. And Rory's group has been a leader in defining the genetic basis of myopathy during statin therapy. And there is a big effect of a single SNP. And some people have actually moved that toward implementation. And there are many, many examples of adverse drug effects of the same kind of ilk. The things that I think of are the Stevens-Johnson syndrome and the hepatotoxicity with certain drugs that are well predicted by HLAB variants. So I think that an early focus on unusual and unanticipated drug effects is there's mileage in that. That said, there are examples of variable efficacy. Clopidogrel comes to mind. And maybe you can sort of say, well, the reason you have variable efficacy is because what you're really looking at is a predictor of failure of efficacy in some people which is predicted by a pharmacogenetic variant that influences its pharmacokinetics. So I think that you have to do it sort of drug by drug and take one step back and think to yourself, is there likelihood that variability in response to this drug is going to be driven by a small number of variants with large effect sizes or a large, large number of variants, some of which has to do with the disease, some of which has to do with the way the drug is handled. But I do think that if there's low-hanging fruit in this space, it's in the pharmacogenetic space. My question was PCSK9. What do you do once you know somebody has a PCSK9 variant? What's the clinical action? Or even fibrillin. I mean, so somebody has Marfan syndrome and they have a fibrillin mutation or a TGF-beta receptor mutation. How does that change therapy? You're as qualified as me to answer that. Or, you know, hypertrophy. It does. So in Marfan's or TGF-beta mutations, we would screen for aortic root diameter and prevent aortic. No, that's not my question. My question is, is somebody who you know has Marfan syndrome, how does knowing the genotype inform the therapy? Not making the diagnosis. Well, I think we're more focused on the people who you don't know have Marfan syndrome. I can buy that. If you know they have Marfan syndrome, we don't genotype them usually. Do you think the diagnosis is made by a genetic test? Or by a clinical phenotype? There's certainly a percentage of Marfan patients and Lois Dietz patients and other patients in this category who have the diagnosis unclear until there's genotype information. Yeah, we make this unsuspected diagnosis in adulthood frequently. Yeah, so, you know, when I'm not doing pharmacogenomics, I'm a cardiac arrhythmia guy, so I live this life all the time and I'm ambivalent about the role of genotyping in identifying patients who have variants that may or may not affect arrhythmia susceptibility and maybe Marfan and hypertrophic cardiomyopathy but it seems to me we run a great risk of telling people you have a variant of unknown significance that might make you drop dead. Have fun. Because there's not much you can do right now beyond agonize. And I think it's something that the whole field needs to deal with. I agree with basic concern and a fundamental tension in thinking about the applications of kind of data work talking about collecting really has to do with his personalization of care, which has received an enormous amount of hype and guiding the development of really new therapies. I think both of the examples that you cite are actually quite exciting in terms of their potential for guiding development of new therapies, especially drug development and I'm not a clinician but have been unimpressed with their potential in this sort of personalized medicine space and I actually would advocate a fairly major shift at least in rhetoric in the direction of what we're trying to do is improve the development of really new therapies. To this end, just coming back to some of the comments this morning these are very, very important individual questions but we're at that paradox of trying to figure this out from very large sample sets unless we knew already who we wanted the sequence for a very particular outcome per se designing the study to sequence those individuals who have a particular severe hypersensitivity to a drug or what have you. The issue I guess is when we turn this around and we have these very large cohorts and we have large scale sequencing which I think is on the table here is this something that we can have as a secondary or should this be a primary point of choice for what samples and what would be prioritized in terms of some portion of the space of sequencing be individuals who are selected particularly on some clinical outcomes that have been identified. There are very different ways to choose samples from many of these cohorts in clinical series and the like and I think we should explore that a little bit more about what kind of prioritization would it be the life threatening Stevens-Johnson syndrome or the drop dead cardiac complex response to a particular anti-inflammatory agent or whatever. I just have a comment, that was a really nice presentation Mike and really gives us the clinician medical geneticist perspective. One thing that's a little bit missing from the overall discussion there seems to be a lot of emphasis on drug treatments and not on preventive or prevention treatments because I mean the way I think about genetics is it's an exposure over a lifetime that leads to risk over a lifetime. Maybe it's a risk of single events like a cardiac arrhythmia maybe it's a long long-term risk like atherosclerosis developing over the lifespan and both for the treatments and maybe even behavioral modifications it would seem to me that we're thinking about the preventive space not the treatment of the disease at the end of its course. So I guess I would just advocate for having prevention somewhere in the general statement of what is being thought of for these cohorts that are being sequenced. I would agree with that but to the treatments wouldn't an interesting at least an extra thing to get is even if you're in a large cohort and the focus might be on etiology and so from our friends you would discover something like this but wouldn't it also be interesting for those conditions where therapy is so variable where half of the people do fabulously well and half of the people do terribly I'm not saying this would be the sole basis for choosing but that would certainly be a point in favor of additional information from your cohort on how people with that disease subsequently did that would suggest there is a findable thing which may distinguish those that do well radiation or chemo actually from those that don't so I'm not sure it would drive this selection but a cohort would be more attractive to me to sequence if it had the capacity to add on subsequent success for therapies that were very variably affected and therefore I was suspicious maybe if I only knew more genetics I would know why half of the people did well and half did poorly that would be largely a therapeutic question but embedded in a cohort chosen where the focus would be on incidents but you'd get more points if you could tell me something about therapy okay I guess Mike and Gail maybe is one of the things you get when sequencing large cohorts is a complete or fairly unbiased representation of the phenotype given the genotype or if you go to OMIM you see sort of genotype given phenotype and I'm wondering if you think in medicine we'll actually start to diagnose and treat disease based on genotype and give up not give up but down weight the role of phenotype because when we start to sequence large numbers of individuals that according to OMIM they should have four years and two heads and they seem to be perfectly fine so I think that cystic fibrosis is like the classic example of this right so cystic fibrosis was this horrible disease that killed you in childhood and I remember arguing with a pulmonologist 20 years ago that I thought there were adults who had mild forms of cystic fibrosis because they got two monosinfectants cystic fibrosis pediatrician pulmonologist who said like no that's not cystic fibrosis and maybe he's right it's not cystic fibrosis but it's the same gene it's just a different mutation and so we need to learn about that variability and I think we understand in clinical genetics that what we see is often the tip of the iceberg and the phenotype spectrum is often much broader and may include just infertility and no lung disease but that said there is specific cystic fibrosis treatment for one mutation not the common mutation but only for people with a specific not ridiculously rare but not the common mutation for cystic fibrosis so that the actual treatment what is broken in the gene affects your therapy not just that that gene is broken but what part of that gene is broken and so I think there's a lot that we're going to learn in that spectrum of what represents the disease but we have to be very careful when we see a mutation in a gene to say oh we know what a mutation in this gene does because we know what one mutation in that gene does doesn't mean we know what a different mutation in that gene does The last comment before we move on In a DNA first world we'll be doing we'll be using that as a signal to go back and re-phenotype people but in somebody with zero recognized prior probability that gets a variant we have to be careful not to not to make the diagnosis based solely on that One more comment I think Gail makes a really important point that for many of these severe disease mutations we don't actually know the full impact of these variants and one very powerful thing we could do with large cohorts would be to go back and genotype all of the variants that are present in these databases of severe disease mutations and look for phenotypes for instance in individuals who are homozygous for alleged recessive disease mutations who appear to be healthy indicating either errors in the database or variable penetrance also looking for phenotypes for carriers of recessive disease mutations I think all of us who work on clinical exome data know that there are a number of these mutations in the databases that are just wrong or where the penetrance is very poorly estimated and it would be fantastic to have that data One point I think we do need to make though is distinguishing between questions that can be addressed with targeted genotyping of large cohorts and questions that require large scale sequencing I think there might, there's a little bit of confusion because we don't necessarily need to sequence everyone in these big cohorts to get these answers there are many questions that we can address just by going back and doing much cheaper genotyping of these very large groups Just before we move on in terms of the outcomes I mean there are cohorts that we can address questions of efficacy and treatment and prevention just thinking about the large HMO cohorts that all have EMRs