 So, I'm a mouthpiece for a large number of people who are interested in the question of applying genome sequencing to clinical care, and so everything that I say that you agree with is mine. Everything that I say that you disagree with belongs to somebody else who helped create this talk. I don't need to tell anybody in this room what this was all about. My perception is that engaging the electronic medical record falls somewhere in here and understanding genomic predictors of disease somewhere in here, and then implementing genomic medicine we've already said belongs in this space here. And so I'm going to outline a number of efforts in NHGRI that are going on to actually move those two little squares into the forefront of what we do. So most of what I'm going to talk about has to do with CSER, which you've already heard about, but I wanted to mention other networks that are working on this within the genomic medicine section of NHGRI. So the two that I'm not going to say anything more about are the Undiagnosed Disease Network, which has just been created and follows really the Mendelian disease discussion that you've just heard, and Insight, which is around the use of genome sequencing in the neonatal setting, and raises a lot of interesting ethical and operational issues that are just now being addressed. So the other doodads on this slide I'll say something about individually. Ignite is a series of demonstration projects looking to incorporate genomic information into EMRs with decision support. The particular projects are focused on family history, on hypertension, on pharmacogenetic variant implementation. There is one project from the University of Maryland that is sequencing around 40 diabetes and related genes in specific patients to ask the question whether targeting patients, for example, with extreme obesity or with lipodystrophy, using sequencing can help in their care. The Clinical Center at NIH has launched an exome project that basically will provide exome sequencing for clinical center patients, and they prefer to look at germline rather than somatic sequencing. It gives me the opportunity to say that most of, almost all of, what I'm going to talk about is in the germline space, except for some of the CSER projects, which I'll mention a little bit later, and obviously somatic sequencing is a whole different topic, as Eric already pointed out. And then there's a mechanism in place to support the parts of implementation in sequence data that are going to be important, as I'll outline for you, in terms of consent, in terms of how to implement, and what to do with incidental findings. The Emerge Network, which is something that I have been a part of since its inception in 2007, currently has around 350,000 subjects with DNA samples and electronic medical records across around 10 sites. I know how many sites there are, but for bureaucratic purposes it's hard to count them. Electronic phenotyping is what we do. So just because we have an electronic medical record doesn't mean we can extract any useful information out of it. We spent the last seven or eight years showing that we can find, with pretty high accuracy, people with common diseases, there are some phenotypes that are easy to do, there are some phenotypes that are much harder to do. The hope is that we'll be able to generate data on, for example, the tempo of disease progression, subsets of disease that are particularly interesting in terms of drug response or in terms of developing complications, or in terms of not developing complications, which diabetics never get retinopathy, for example. Those are interesting questions, and we think the electronic record should be suitable to that, and that's an ongoing process. We've also pioneered the use of this idea of going from genetic variant to asking with what human phenotype is that particular genetic variant associated. This idea of pleotropy or phenotropy, I couldn't find phenotropy on Google either, but the idea of phenome-wise scanning to understand pleotropic genetic effects is something we've spent a lot of time on, and then most recently we've started to use targeted sequencing across a set of 84 genes that the Pharmacogenetics Research Network has identified as important in drug action. Those are metabolism genes, drug transport genes, and drug target genes. So the idea, I'm not going to dwell on this, the idea is to find patients within the electronic medical record system that would be suitable for this kind of analysis, identify the set of actionable variants within that set of 84 genes. I could say the set of actionable variants right now is about 10, but then sequence all 84 genes, create a repository of all the variants that we find that are not actionable, and for the small number of actionable genes, implement implementation within the electronic medical record system with clinical decision support and looking at outcomes. One of the things that we have found is a variant of uncertain significance problem that will come up over and over again as we start to sequence individuals either chosen because they have an electronic record or chosen because they have a particular disease, but we're looking at their whole genome. So two of the genes that are on this 84 gene platform are ion channel genes where rare variants can cause something called the congenital lung QT syndrome and arrhythmia susceptibility. So the first 2,000 subjects that we analyzed there are variants in 128 and most of those are, as you might expect, rare. So we asked two companies that make a living doing this and a research lab that makes a living doing this to look at that list and say which ones would you tell the physician and the patient are likely pathogenic? The English literature around this is interesting because they don't say pathogenic. They say likely pathogenic, possibly pathogenic, maybe pathogenic. A lot of sort of qualifiers. So one lab called 16 pathogenics, another 17, another 24. The overlap among the 3 was 4. So there's a problem in terms of understanding the relationship between rare variants and disease at that level. The electronic medical record allows us to look back and actually ask the question whether any of these people, any of the 48 people who had one, at least one of those variants called by one of those 3 sites, had a variant, had anything in their electronic record that might indicate a variant phenotype and there was one patient with atrial fibrillation, a very common arrhythmia. There were 31 who had ECGs and one with a lung QT interval in that ECG. So the questions are, you know, which results do you return? What do you do with patients who don't have ECGs, should you tell them to get ECGs? How extensively do you screen the families? And what happens when the interpretation of these data changes next year or the year after? How do you recontact the patients and their families to deal with that? So NHGRI has funded ClinGen, which is a large clinical genomic resource, which has many mandates that are listed in the bottom of the slide, standardizing sharing and developing methods for annotation, interpretation, assessment of actionability. So hopefully that will, that focus on this particular problem will help as we move forward in the implementation space, but this is a large problem. So CSER, which is where most of the efforts in terms of thinking about implementation have gone in NHGRI, currently has three and a half thousand patients, or the target is three and a half thousand patients, 10 projects. And the idea is to select individuals who have specific phenotypes and they will then be studied by sequencing mostly targeted regions of the genome. The hope is to develop standardized exome and genome sequencing and reporting, and LC has been part of this since the very beginning, and I'll say more about that in a moment. So these are the phenotypes selected for study. Most of them are focused on the germline genome. There are two or three projects that focus on childhood cancers and adult cancers. The susceptibility to colon cancer in polyps is a germline project. And the questions are, what are the, what is it about a patient that says that genome sequencing or whole or exome sequencing or some variant thereof will be useful in the care of their patients? How do you analyze those large data sets in a clinical environment? There's this whole issue around what happens. The focus here of course is on Caucasian populations and the question is how do you extend the breadth and the reach of a project like this into other populations for all the reasons that we've heard about and that we all understand in terms of not just being politically correct but in terms of actually informing the biology. And then the management of non-targeted data that I've already alluded to and continues to be a problem. So the progress report from CSER is that as of March 2014 and now I'm really being a mouthpiece, 1500 subjects studied mostly Caucasian and the total number sequenced over a thousand, mostly germline but with some tumor as well. I'll skip this slide. Actually, let me just say one thing about this slide and that is that the rate of incidental findings in this population is around 3% using the ACMG list of 56 important genes as the benchmark and if you look at it a different way it's more like 6%. These are data that were updated this weekend and so these are germline analyses only and the important part is this last column, the yield. The initial data looked like there were higher yields in some diseases compared to others and there may still be a higher yield in retinal disease than the others but with the exception of these cancer cases and again these are unselected cancer cases looking at germline genome susceptibility. Looks like the yield is around 40 to 50%. In these selected subjects, these are data from the Baylor Center looking at pediatric solid tumor exomes. These are relatively unselected so the number of patients who have actual recognized category 1 mutations is small but there's still a large chunk with category 2 mutations and then this idea only over here of category 4 mutations no one knows what to do with. So as I said there has been a major focus on LC in CSER since the beginning and that includes issues around informed consent trying to figure out how informed consent is delivered across the sites and trying to develop recommendations for best practices and then this business of incidental findings which I've already alluded to the CSER sites have actually had input into the way ACMG is revising their recommendations with respect to those incidental findings and then trying to figure out what it is that patients need to know or want to know that if you ask the patients what they want to know the answer is everything but if you then ask them well you'll need to make a phone call to find that out it turns out that even small barriers like that create a disincentive for people to actually want to know everything and then collecting data on what it is that patients feel after they've gotten these data is an important part of this effort and has been embedded in CSER since the beginning. The issues around sequencing findings I've already alluded to but sort of how the question of if you find an ion channel variant for example how many patients have had any kind of work up alluding to the ion channel variant what do you do to create that work up and then issues around cost effectiveness how fundings are transmitted to patient and family members and then this whole question of ongoing reinterpretation of the genome a variant that is of uncertain significance today maybe pathogenic tomorrow or maybe benign tomorrow how do you transmit that information in an ongoing fashion to the patient and to the family members who are also variant carriers so across the initiatives in NHGRI in the genomic medicine space there are common issues around integration of information into the electronic medical record which I haven't talked about very much but represents a continuing and interesting challenge the return of results the question of what is an actionable variant and how do you how do you share data when people are nervous about sharing their genomic information and then this this longitudinal issue that I've already alluded to a couple of times I am part of a group that advises the genomic medicine initiatives at NHGRI that we call ourselves the genomic medicine working group and one of the things that we are trying to do is create mechanisms to engage other stakeholders within the community so the other stakeholders might be payers other academic centers other non-academic centers stakeholders within the federal government and international stakeholders the last meeting that we had was for international stakeholders we had a meeting in in cold snowy DC in early January and surprisingly there was a there was a huge amount of interest from across the world US tax dollars did not go to support any travel for any of these individuals and yet there were individuals from 25 countries and 50 participants who showed up from the from the countries that you can you can see and the focus is there focus there was on evidence generation on varying levels of health information technology from across the world on issues around education and workforce development that's an important issue not only for us but across the world and that's not only genomics but other supporting I call them supporting disciplines like bioinformatics and then other health professionals and then the public there's a big interest in implementing in the pharmacogenomics space first and and then there are questions around policy that obviously vary from country to country and if we're going to share data across across countries then those are the things that are going to have to be addressed. This is a snapshot of what it is that individuals at that meeting felt that they had today and felt that they really wanted going forward so what people really want going forward our systematic family history genetic counselors and electronic medical record that provides clinical decision support not so interested in interestingly in in in the sequencing part or but they think that implementation in the pharmacogenomics space might might be interesting so so just to summarize the challenges in implementing genomic information are that we don't know the function of most of the variants that we identify and as the speakers who came before me emphasized the way in which we're going to get at rare variant function is to develop very very large data sets and I call this the paradox of personalized medicine if you want to personalize medicine for an individual with a straight face you have to be able to draw from a denominator you can't do it because one person is different from the rest the other 999 you can do it because a thousand people are different from the other million people that you study so you have to have the large denominators so this is one one crude attempt to sort of give you a sense of the where the large denominators are around the country and around the world. I'm tempted to say here that you know there are places on this on this map that are contemplating sequencing every single person that they have in their data bank and the problem there is going to be the sequence and they will have very little idea of what to do with the data. So whether sequencing every single person in the UK Biobank or in Qatar or Estonia or Iceland is a is a great idea for implementation or not remains to be seen for discovery it might be an interesting opportunity. So the challenges are the quality of the data and the analysis. What why was somebody sequenced? Were they sequenced because they're a worried well person or were they sequenced because they have a particular disease if they were sequenced because of the particular disease. What are the results of that sequencing for that disease and then what do you do about the incidental findings and they're going to be a huge number of incidental findings. And then and then there is this idea that that we all know that certainly in the iron channel space which is where I live partly there is an emerging understanding of modifier genes. And if we're going to do large scale sequencing we have to be able to take advantage not only of the variants that we find but think of ways in which the variants that we find plus the modifiers could be implemented clinically. That's a huge challenge. I don't even know what an actionable variant is at the actionable variants. It's a great word to use but nobody actually has a great definition of it and it may be for the indication they may be different actionable variants for incidental findings. And then I'm you know you could read the rest of the slide engaging the patients figuring out what they want figuring out how to best to deliver this within the constraints of perceived and real privacy and consent issues. And the real potential that if we do this wrong that all we'll do is we'll have a huge number of people with variants that they don't understand or with extra medical care that they didn't need. And so there's this tremendous opportunity for NHGRI to really screw up if we don't do this correctly because I can certainly imagine situations in which everybody who gets a rare variant in a desmosomal gene will end up with a cardiac MRI. And that's probably not what we want what we want. So the challenges are figuring out outcomes clearly engaging multiple ancestries across the country and across the world. Training is a huge problem. We need to figure out a way of expanding the scope of implementation as it moves forward from not from academic centers to non-academic centers to non-informatics rich centers. And that's going on a little bit in the Ignite projects figuring out how to implement in diverse electronic medical record systems. There are a number of stakeholders that we've become aware of. For example, almost all laboratory data that are delivered in most hospital centers are delivered to the clinician or to the electronic record or to the written record through the pathology department. And so interacting with the College of American Pathologists and other pathology entities to figure out how best to deliver that information is an important part of this. I've said it before I'll say it again and other speakers have said it before me. We really need very, very large datasets linking genotype and phenotype how to do that. I don't know. And then there's a question of interacting with regulators. There's some of this some of this sequencing is viewed as as investigational by regulators like the FDA and of course figuring out how to interact with payers to get this right. So I was asked to close with a slide that said if NHGRI doesn't take courted action, what will happen if NHGRI doesn't do this? So my initial take was that the promise of genomic medicine, which we think we recognize in terms of identifying variants that are important for disease, identifying variants that are important for drug action, identifying variants that may be clues to drug targets will be delayed. But at the end of the day, the real imperative is to figure out how best to roll this out that maximizes patient benefit and minimizes risk that we'll do this wrong and just confer lots of costs and develop a lot of cynicism around the country and the community. So which patients and which targets, which patients and which genomic targets are the right ones to focus on the well-patient or the sick patient. And then the other point that I'll just sort of say as you read the slide is to work out the realities of implementation. So it's easy to say but hard to do in terms of consent, in terms of EHR integration, in terms of educating patients and providers, providing the clinical decision support, figuring out what to do in followup, and then figuring out what any of this has anything to do with economic outcomes and healthcare outcomes. And I think NHGRI, my own bias is that NHGRI needs to play a role in that. I'm not sure it's our only role in that. There are many other stakeholders who are interested in the economic and healthcare outcomes part, but NHGRI needs to be at the table. So with that I'll close and take questions. You have five minutes for questions. Okay, nobody from the Broad, remember? No, in this session you can't. That was great. I wanted to pick up on the thing you said. The last time you asked me a question, you said that the electronic record was useless. So I'm glad to hear that it's great now. No, I think I said it was not interoperable. And that's still the case, but we need to fix that. I was really struck by your last point about regulators, because I think this community has a special role to play in providing guidance to the regulators. Right now, we're either in a situation where people are just randomly making declarations about the importance of a SNP or a mutation, or on the other hand requiring a 510K to be filed on each nucleotide in the human genome. Neither of those are really good solutions, and it's gonna take somehow a genomics community coming together and saying here are things that are appropriate for FDA approval or appropriate for reimbursement. And there are other fields like preventative services where there are external groups that provide meaningful insight and meaningful guidance and even in legislation, they're entitled to provide that. And I think that one of the things we might think about at a large scale is how do we provide some kind of quality filter on the, well, actionability of relevance of, choose your word of these variants. So I was really glad you put that there, and I was wondering if somehow we might include that in our thinking. That's a comment rather than a question, but I agree, and the idea of getting a 510K, is that what it is? Yeah, for every single nucleotide is something that we've sort of discovered in the last eight months. It's something the FDA is interested in doing, and I think that's pretty lunatic as well. But if you put yourself into their shoes, they don't want this to sort of be the Wild West. They have to figure out a way of making sure that tests that are deployed widely through commercial pressures actually have some rational basis. And so I agree that if this community isn't going to provide that advice to FDA, I'm not sure who will. Tim? When you said delayed on that slide, I assume you meant delayed in the US. It's not going to be delayed in the UK. Well, I'm glad we're going to follow in your wake. I don't know how else to respond to that. As you stress, I think the bottleneck is really the follow-up and getting the assays to go. And I think that's where NHGRI, in my opinion, could take a real lead in marrying these results and thinking more systematically about how to do these sorts of follow-up assays in a high trooper fashion. Because quite frankly, everybody who's in these disease states really wants, even if it's far-fetched, they want to have the follow-up assays done on them on these variants to figure out which ones are relevant. Very expensive, but this is what NHGRI is good at, I think, actually trying to high-trooper things in ways that could go very quickly and less expensive ways. Well, I guess I'd argue with that a little bit. And I'd say that, and now I'm being a mouthpiece for me, I'd say that you don't want to be in the position of even insisting that every single person who has an ion-channel variant gets an ECG. There ought to be a better way of filtering that because even that is gonna be a huge extra cost when it turns out, but there are other genetic variants that occur in genes that have been associated with cardiomyopathies, for example. So as soon as you find one of those, does that mean every single person gets an MRI? You're gonna bankrupt the healthcare system in an instant by doing that. So we have to figure, the NHGRI imperative has to be to figure out the filters. Sure, and I think stem cells, for example, setting up surrogates that way could be done in a less expensive fashion in which you could. So thinking creatively about how to do this, and again, in ways that are not as expensive as an MRI but other ways to actually get up some of these phenotypes that, and obviously you have to prioritize them if they look more severe for whatever reasons, a nonsense mutation, obviously you probably would do an MRI and for maybe not so clear ones, maybe you wouldn't. I think that's a really interesting topic for debate. Yes. So I thought you made a very good point here about the marrying between the cohort scale with the electronic healthcare records that will allow you to, because you know the genetics doesn't change, you can measure cohorts, you can look at if I had made this decision to inform these people for the subsets that were tested, what would have I seen coming back out again? And then turn that round for individual level feedback. So I do think the large population cohorts are quite important here in this space that are linked to electronic health records so you can then, from current practice, work out what would have been an effective use of genetic information for those people. I agree, and I think that one point that needs to be made is that while I'm enamored of the electronic record because I think it's the real world sort of nitty gritty of the delivery of healthcare, it can't answer every single question. And there are some questions that are better answered by prospective cohorts or whatever. So I think the idea of putting all your eggs into a single basket is probably not what I would think is the right recommendation. That said, the end game has to be to figure out how to use the, not only deliver, but use the electronic resources that we have, the electronic medical record to answer questions like the one you've just posed. Yeah, I guess I want to, and this is also an answer from the sangha, you know, we're thinking about the same lines, and I think ideally you want to develop resources that are deeply phenotyped with a reconsent for recalling people back. Ideally an IPSC or other type of resources where you can do genome engineering experiments before this extensive recalling people. So I think that there are many kinds of resources that could be used, and that's one of them. Yeah. Just to follow up on Ewan's comment about the genome not changing, one thing we haven't really covered at all is what is the impact of somatic, I'm not talking cancer here, but somatic mutation changes over the age of the patient impacting adult onset disorders, which would obviously require a different scale and type of sequencing. So I'm a heart doctor when I'm not doing this, and I've never had a patient say, yeah, sure, have a piece of my heart. Well, but heart is probably uniquely difficult to biopsy. Well, I think brain is probably just as hard. No, but you know, I mean if you spend your time in hospitals as do I, and we biopsy all types of tissues all the time. I understand. I mean, no, no, I agree with you. I would just try to sort of highlight the, you know, what I'm fond of saying and other people have heard it said before. You know, the oncologist goes to the patient and asks for a piece of the tumor. The answer is a piece. Take the whole thing, please. Yeah, so I've never had anybody say that. But I agree, I mean, it's a great unknown right now. I mean, the extent to which somatic mutation occurs and contributes to all the diseases that we wanna study. I think we have to go to the break now. If everyone could be back here in about 15, 20 minutes, we could go on with the final presentation.