 My job, really, is where are all these variants coming from and what's the clinician's attitude or what do we need to interpret them? And I just want to make a couple of disclosures because I'm going to talk about some guidelines for our medical professional group. And also, I've been on the External Advisory Board for the mouse genome informatics database and at the very end, trying to integrate these two, I think, is partly what we need to do. So currently, if we look at clinical utility of precision medicine, whatever we want to call it, I think it's really been in the diagnosis of real rare Mendelian disorders, which is what I deal with, as well as in cancer diagnostics and personalized therapeutics. We're beginning to make some headway and pharmacogenomics, and I think we all hope that eventually we will understand better multifactorial disorders. So for those who are not clinicians, clinical exome sequencing has really revolutionized the way I practice clinical genetics. It has a very high diagnostic yield higher than anything we've ever had in the past. At least in pediatrics, we know it's important to study trios. If we study a child in both parents, the yield is going to be much higher than just studying the child alone, and that's because a lot of the pathogenic variants we're finding turn out to be de novo mutations. What we're here to talk about today is how do we deal with variants of unknown significance, and there's a new class of them that we are dealing with in genomic sequencing, and that's secondary findings, which I want to just spend a couple of minutes on. So actionable secondary findings can be defined as damaging variants in disease genes that are unrelated to the reason that the testing was ordered, for which there is significant morbidity or mortality, and where early diagnosis can ameliorate or prevent the disease. And there's a unique opportunity when one does either exome or genome sequencing to identify these variants, which can have implications for the patient or the family. And so, in 2013, ACMG, as well as the President's Commission on Bioethics, came out with policy statements to search for and report selected of these variants. ACMG came out with a minimum list of 56 actionable genes. Some of the authors of that document are in the room here. They're mostly hereditary cancer genes, Marfan-related syndromes, cardiomyopathies, arrhythmias, familial hypercholesterolemia, and malignant hyperthermia. So high penetrance, and there is something one can do to prevent or ameliorate the disorder. The pathogenic variants, and I'll stress that these should be known pathogenic variants, should be reported regardless of the indication when a clinical exomer genome is analyzed. Some labs do additional genes. The minimal list should be reported regardless of patient age, i.e., those reported in children. And while originally our recommendation was to do this and not have an opt out, that position was updated due to a firestorm, if you want to call that, from everywhere, so that parents and patients can opt out at the time of consent, though, at least in our experience, maybe 5% do. Labs should seek and report only pathogenic or likely pathogenic variants, and this is because there's a very low likelihood of disease. It's different than when someone already has had a cardiac arrhythmia or has a strong family history of sudden death from arrhythmia. When you're looking in healthy people, you need to be sure. Labs are supposed to list the quality of the data. They don't have to cover every exon in every one of the genes listed, but they ought to report what their coverage is. And the clinician, whether it's a geneticist or a family practice doc, it has the responsibility to tell patients about this and to interpret the findings. Our organization, as well as the QuinGen project, are refining and updating this list, and I believe at the present time there is still no consensus about reporting these findings in the research setting. So if we look then at clinical exome sequencing, who are the best candidates? And again, I'm talking from the pediatric standpoint. So when you have a specific phenotype, you ought to order the most specific tests possible. So if a child has an abnormal sweat test, you sequence the cystic fibrosis gene. If you think a child has Noonan syndrome, there's now a very good gene panel that covers all of the known Noonan genes, 85 to 90 percent, and the chance you're going to find another Noonan gene in the next week is very low. They're probably a couple out there, but they're rare and they're not coming along all the time. Whereas when you look at intellectual disability and other similar kinds of conditions, they're new genes all the time. They're probably at least 500, and so exome sequencing, in my opinion, becomes the best way to cover the spectrum. There is a longer turnaround time, and right now, though we're not going to talk about it, there may be lower reimbursement than doing specific panels. Again in pediatrics, we almost always recommend a microarray before doing an exome. Not only will you find copy number variation that may be responsible, but you may uncover consanguinity that you didn't know was there, and that, as we've just heard, can help point you to specific regions where a causative gene may lie. And compared to before exome sequencing, we still do some relatively cheap screening tests, but I certainly move to exome sequencing much quicker than we would have in the past. Oops. Oh, my God. Now what did I do? I think it's important to talk about utility of a genetic diagnosis, and even if it's not going to change management, it can make a dramatic effect on a child and a family's life. First of all, it's going to prevent additional unnecessary testing, and kids and families go through this long odyssey, and both the financial but the emotional impact to have a diagnosis is really important. As we learn more about disorders that may help prevent or recognize future medical complications, it may help predict function as an adult. For sure it's going to provide better guidance concerning recurrence risks, and some insurers are actually beginning to accept this information as a cause for doing the test. Rarely it will prevent, permit specific medical therapies, although we always tell our families this is rare, at least in the pediatric field that I deal with, and they shouldn't get their hopes up. So what's the experience that we've had clinically? We're a large children's hospital, but we don't have a huge genome center, and so this is really a clinical perspective. So at our institution, all exome sequencing must be ordered or approved by a clinical geneticist. We feel that this is a bit of a gatekeeper. A lot of our physicians don't understand all the nuances involved. This can happen in different ways, referrals to genetics, which is the standard way. I've been involved in starting some case conferences with different disciplines, where, particularly with neurology, we've been doing this a year and a half now, they present cases they think are important for exome sequencing. We discuss them, and it's worked incredibly well. We're beginning to understand them, and they're understanding us. We move from is this case good to do an exome to how do I interpret this variant of unknown significance? And I think this is one way where we're talking with them, and we can present our feelings and use some of the basic model organism data to help sort out what might be actually disease causing. And then we are planning, or I am planning, a genomics clinic starting small, and then we hope to expand. So how have we done? Our exomes now go to Baylor, America. We're starting in-house, hopefully by the end of the summer. But through the end of August, on 160 clinical exomes done, we had a 44 percent positive yield, which I think is fantastic. It wavers a little. And it's all across the board, different kinds of disorders. And then we had 2.5 percent or four secondary findings that were positive, BRCA1, MEN2, BRCA2, and a calcium channel gene that we've had to deal with. I've gone through the records and analyzed the first 100 cases. This takes a fair amount of time. But for the positive cases there, 46 percent, with a positive result, had a change in management beyond the reproductive risk. Now most of these were changes in surveillance that you were worried about a disorder. You didn't find it. So you could stop surveillance, or in some cases you had to do surveillance. Three of them were specific for a medication or a diet change, including a patient that had a seizure movement disorder, and based on the gene found they could change to a drug that was actually helping the movement disorder much better. The other thing is that 20 out of the 41 were de novo, which results in a dramatic reduction in recurrence risk. So we just don't know it. We had assumed it was probably 25 percent. And now it's almost certainly less than 1 percent. And that is really important to families. And I think something that we need to push going forward. And there was some novel gene identification, like in many sites. So where do I think clinical sequencing is going? Well, it's rapidly expanding to carrier and population screening. Some of the companies that are now doing carrier screening prior to pregnancy will do panels of known mutations, but a couple hundred bucks more you can do a whole exome. And those need to be interpreted. And as we're moving into population screening, again, we're looking at healthy people, and when we find variants, the bar has to be set fairly high. This whole meeting is moving from gene identification to variant pathogenicity, and there's a need for tools to validate these, particularly the missense variants. And as research is moving to whole genome sequencing and clinical will follow, in one test you'll be able to look for chromosomery arrangements, which I think will be very powerful. But there will also be the complexity of assessing non-coding variants, which right now I don't even want to think of from the clinical side. So ACMG has put out standards and guidelines for a good while now on variant calling from our Laboratory Quality Assurance Committee. And in 2013, there was an update to this to try to assess how to call variants from whole exome and genome sequencing, to try to standardize the process, make it transparent, and to have similar results across labs. The work group consisted of lab directors and some clinicians from our organization, the Association of Molecular Pathology and the College of American Pathology. The classification terminology is complicated like CNVs, there's pathogenic, likely pathogenic, VUS, likely benign and benign, and there's a very complicated scheme of how labs are supposed to grade evidence. And I thought the best way would be to give an example, Howard gave a similar kind of thing. So this was actually a case that we talked about in our neurology conference, and I think one of the lab fellows for the slides. So this was an 18-month-old with severe progressive epilepsy, they did a seizure panel, there were no pathogenic variants, but they had a variant of unknown significance in a potassium channel gene. The gene was known to cause dominant seizure disorders originally benign neonatal seizures you have as a newborn and then they go away, but it's also now been found that it will cause later onset seizures as well. There were three publications with some functional data, and so how do you grade evidence for this? So looking at the population frequency, if you look in the exact and other databases, it looks like it's about 1 in 250, which is much too frequent for a disease that maybe is 1 in 50,000. A case control study, if you look at the bottom there, shows that it is enriched in disease. The variant in this region is highly conserved. And while the functional prediction programs are mostly that it would be deleterious, it's not 100% like that. And then you could look in ClinVar where variants are reported by laboratories with their interpretation. And again, there weren't many here, but one said it was benign, one said it was unknown significance, and one said it was pathogenic, which doesn't really help you very much. So you could plug this into the checklist that ACMG had, and our lab has modified that. Hopefully you can see this, oops, I got to go back one, I think. So there's strong and supporting evidence, and in this case, there was evidence that the allele frequency for the disorder was too high, and also that you saw it in healthy individuals, which would go against this being pathogenic. But you could look on the other side, and there was some established in vitro functional data saying it was damaging, and there was also tracking through families and effecteds that it associated with disease. And what we're looking for here is to look at these in vitro or functional data and how can that help us sort these out, these kinds of things. And so when you get this conflicting data, what the guideline says and what the lab says is, you need to call it a variant of uncertain significance. And I will just say here that I think the lab has to be very conservative, but this is where the clinician's input, I think, is valuable. And so while I wouldn't have been convinced near the beginning of Howard's example, by the end, I would have been willing to say to that family, I think this is involved, and I think that the clinician has to be informed and has a responsibility to take all of the data and put it in context. And so what can model organisms do? They can demonstrate a role for the protein in the biological process. I don't think they can demonstrate pathogenicity, but they certainly can help, which is what we're trying to do here. Hopefully in the future, we can use model organisms to demonstrate gene-gene interactions and eventually to test possible therapies. The model depends on the gene and the phenotype. This is too simplistic, but there are lots of models at the disposal that we have, and I think one of the challenges is understanding the advantages and the limitations of each model and that you're not going to have an autistic mouse, but if you see CNS expression and behavioral abnormalities, that may tell you that this gene or variant may be involved in a human disorder. So to just do a couple of examples, one can use yeast to study conserved metabolic pathways. And so for much of my career in the old days, I studied X-linked mouse models that turned out to be defects in cholesterol synthesis. And this was the major mouse I studied, the bear patches mouse, X-linked male lethal, and I spent years trying to clone this the old-fashioned way with genetic mapping, physical mapping, and eventually we found that it was a disorder of cholesterol synthesis. Now we did one thing smart through all of these years, and that was that there's another mouse model called Tattered, also X-linked male lethal, looked very much like bear patches. We mapped that to a completely different region of the X chromosome. But once we knew bear patches, we looked in that region and were able to find another gene that was the next step in cholesterol synthesis and very rapidly found out that that was the gene for Tattered and a human X-linked male lethal disorder. So I think this is one example where we used yeast way back when and it helped. Certain types of disorders I think are only going to be able to be studied in mammals. You can't study placental disorders in zebrafish. I think also as we get to higher neural function behavior, the mouse and other higher mammals may be very important. And so again, from some of our autism work, we found damaging de novo variants in novel genes in two of our human autism patients that I think are likely pathogenic based on behavioral phenotypes and knockout mice. Now those are total gene knockouts we're in our human patient. It's a de novo heterozygous variant. But again, to look at where it's expressed, when it's expressed in some relation to phenotype to me means it's probably pathogenic. I have not told these two families as yet that this is the likely cause, however I will say that. And finally, I just want to finish. I think that one of the best places to focus who we're going to direct understanding how models help is the clinical molecular geneticist who runs a diagnostic lab. And that slide's gone. But through my work with JAX and then the American College, I was able to facilitate a webinar that the outreach coordinator at JAX did on how to use the mouse genome database and how a clinical lab could try to look at model organism data to inform some of their deliberations about whether a gene or a variant was pathogenic. And I think it's the laboratory directors that need to review all this data and might have the skill and the time, more than the clinician, to try to integrate that into their report. So I'll stop there. Thank you. So while people think about their questions, we'll have five minutes for you and then we'll have an open free for all. I want to come back to the question of who is the interpreter of the laboratory data? Is it the director of the genetics lab or is it the cardiomyopathy expert in cardiomyopathy clinic? It's both. And when we came out with our new standard of practice for our profession, there was some somewhat heated discussion about what is the laboratory director's role in interpretation and what is the clinician's role. So I think the laboratory director is the one who has to go through and decide was the test done right. Is this pathogenic meeting some criteria? But then I think the clinician has to put it in context. If you have a pathogenic variant clearly, but it doesn't fit the clinical disease, that's not the answer to the child or the patient's problem. And so I feel very strongly that we take the report that the laboratory generates to the best of their ability with the clinical information provided. So if the clinician doesn't provide any information or it's just intellectual disability but you don't say they have a mole or two sign or other things, it's hard for the lab to do their job. But then I think the clinician has to take that and put it in context. So you know, you could say I'm hedging, but I think our role is to take what the lab says and interpret it in the context of the patient. So let's hear from the cardiomyopathy expert. Yeah. Sorry, I wasn't really going to talk about cardiomyopathies, but I do think that we have a historical precedence of laboratories that are providing information that inform clinical care, but are being run and organized and data is interpreted by someone who's never seen the patient. And in the imaging field, of which we make extensive use, there's a very common phrase that ends many reports. Clinical correlation is advised. And I think that's a really important thing to put in the context of genetics and genomics. If we knew that A caused B and phenotype would emerge and evolve, we'd be done with this conference and we could move on to therapy. But I think we have to recognize we don't understand that, and so clinical correlation is going to be, in my opinion, where the buck stops because if it doesn't make sense or it might make sense in the future, there can be an advisory, but I would have difficulty in terms of making a diagnosis. I don't think these are diagnostic tests. I think they are informative tests, the same way that an X-ray that shows a mass in the lungs is informative of lung cancer, but not diagnostic of lung cancer. My other question to you would be actually a bit of a surprise. I'm actually in favor of having a bit of a free-for-all of letting people do genetic testing. And I think the reason for that is because clinicians are scared. They don't understand a lot of the nuances. And I think it's a see one, do one, learn one kind of approach. And so I frankly would urge you to take off the buffer of the clinical geneticists. I mean, I think it's an expensive proposition from a healthcare dollar's research utilization, but I think we can make people learn a whole lot rather than sequestering this kind of information into the really subspecialists. So I guess I'll make one comment at that. So the neurologists who've come to our conference, we don't have it over the summer. When they email me a one sentence about why they want to do an exome, it's fine. Now we still have to have a genetic counselor who can sense and all of that, but I think there's some baseline education. And I think the clinicians who really don't know anything about this get scared. But the hope is that if we can interact, get some idea, and they get some idea, yeah. Then I think having them order it. And this is again partly driven by finances, but also really by what's in the best interest and the best test for the patient. So Les, and then I'll. Yeah, listening to this and a number of the other talks, I'm struck by the fact that I think here we're talking about a number of different assertions of truth. And they're qualitatively quite a bit different. And Daniel, I think sort of your nature paper from that workshop started that conversation. These assertions of relationships of causality between disruption of a gene and a phenotype, then assertions of the relationship of a variant to that phenotype. And then as you work toward the clinic, then assertions of the human relationship of a gene to a phenotype and the relationship of a variant in a person to a phenotype. And then finally, the molecular diagnosis of the patient. And all those are different assertions of different kinds of truth. And the nature and the habits of evidence that we have for those are a bit in flux. But we have some, I think, established approaches to these. And it strikes me, I'm glad you brought up, you didn't really use the word, but you started talking about Bayesian inference. And I'm sorry to see you shudder, because I think it relates to what Howard was saying. And I think that we have a well-established tradition, especially in basic science and common disease of setting up large controlled trials with the very frequentist Fisherian approach to analyzing significance. And I think in some ways we have a bit of a tyranny of that mindset in science and medicine today, which is probably why no one raised their hand at your meeting. And I think, though, as we move, what you're actually talking about is taking a mass of evidence and using Bayesian reasoning, prior probabilities and conditional probabilities, and using that to come to an assertion or in the final decision, a diagnosis about a patient. And I think what we have to do is begin to convince our colleagues when it is appropriate to stick with Fisherian, and when we have to start thinking in a Bayesian way. And that affects a lot, I think, about how we treat functional data. Because functional data are somewhat challenging to introduce into Bayesian modes of reasoning and decision making. But I think we will have to begin to do it, because some of the data Howard and others have presented really are conditional probabilities that modify the posterior that that is the cause of the patient's disease. And we have to be very careful to separate, and I'm glad, was it, who of you, so you both I think were saying you were arguing about the separation of the assertion of pathogenicity of the variant from the diagnosis of the patient. And the former goes on in the laboratory, the latter goes on in the clinic. And we really, if we muddle those, it'll be a bad mess. No, I would agree. I just, I can't do a chi-square, and so when we get to math, it really, that's why I started mapping in the mouse years ago, because you just did a cross and counted up recombinance in the old days. But no, I think how do you weight some evidence against other evidence? And I think we will have to come to that, hopefully again in some way that I can just look at it or plug it in. But I think right now it's trying to get evidence and what are the types of evidence and then it will be how to weight it. So no, I agree. So I was actually going to extend a little bit on less is common, because I think you're absolutely right, this distinction is very important. But one of the areas where it's blurred is actually in therapeutic trials. So we have lots of therapeutic trials where we are actually supplying the post-hoc condition. And we ignore causality. There are lots of situations where this sort of limited set of preconditioning sets plus one biomarker led to a therapeutic outcome that looked beneficial, therefore for time memorial thereafter, we will use that biomarker as a surrogate and go on and treat it. And this is natural history of how therapies are introduced in medicine. I think one of the things that may be extending a little bit on that and on Gail's talk is if we move beyond diagnosis to therapy more quickly, then I believe we'll actually close the gap much more rapidly. I think one of the things that we have focused on as geneticists is diagnosis and making most patients don't care if they have LQT 10 or 13 or 19. In fact, antibodies probably go LQT more than seven as being mislabeled. What they really want is something that will actually make them live longer or feel better. And so either we supply the conditions and do the tests or else we continue this sort of introspection until we have the perfect solution, which as you point out, I don't think will ever come. I would consider that was a comment not necessarily requiring a response. And I'll move on to Elizabeth and then Bruce and then we'll let you sit down. I think it's on the same vein. So one of the things that you had mentioned was that there were a number of cases where you found novel genes which were you diagnosed the patients with deleterious variants and novel genes. And in one of the examples, it looked like it was a animal model that had been used to make that leap. So I was curious what the other cases were. But this discussion is why I was curious because I think we have examples of where we're using lots of disparate data and we're making that leap into something that couldn't be diagnosed. We've got some sort of analytics. We make a diagnosis and I was just curious what those other cases were. But mainly to highlight the ability that we're already doing this, we just need to do it faster, better and more automated. Was that in the autism example, I guess? It was in your 41 whole genomes that you had done through Baylor. You said three of them were novel gene. Oh, OK. So that was Baylor coordinated. And in all of those cases, they waited until they found at least a second unrelated case before calling it. And so I can't say because they weren't my cases, but I know at least a couple of them. It was read out as negative. And then three, six months later, you get the call that, hey, we found another case. We think this is it. And by the way, can we write it up? So again, a clinical lab often, I can't say everyone, if you have a damaging variant in a novel gene and you think it might fit until you find the second case, they're not going to call it out on a clinical report. That's why we need to do this. We have the compute to do that now. I think another way to frame the point Les was making that doesn't require the use of mathematics is to remember that medicine, the practice of medicine is fundamentally dealing with ambiguity. Most of what we do, there aren't simple black and white guaranteed to be correct answers. So we're always integrating information, taking into account history, lab findings, and so on, to make clinical decisions. So we shouldn't actually really be that uncomfortable with the idea that we're faced with variants of unknown significance. It's what we do as clinicians all the time. However, what made me uncomfortable putting the data that Howard had talked about in the medical record was two things. One, that's probably the best way to bury the information. But actually on the other side, there is a danger that it will look more solid than it really is. And although you could put all the caveats you like and you can have the lab director be the sort of interpreter and put cautions up and down, there's a real likelihood that somebody somewhere won't read them or won't understand them and will misinterpret them. And I've certainly seen that, signing out lab reports and also seeing patients that were first seen by other clinicians who were the recipient of the lab report, then they find their way to me and you're kind of aghast that they misinterpreted what was in clear black and white. So I think that the problem here isn't dealing with ambiguity. It's figuring out ways to make it clear that many of these results, in fact, are ambiguous. And how do you put that into a context where the potential for misuse is minimized? Do you want to have a, you can gavel the last word for that. I never disagree with Bruce. Okay. No, I think he's right. Howard wants to follow up for that discussion. Yeah, so. You get to step. Yeah, no, you're safe, yeah. So Bruce, just as a follow up, so I agree with you. So I think part of the discussion here is, is that, so how do we move to that level of confidence, right? So I think that is the crux of the matter, is that what is the level of evidence for which we would be willing to, I'll say, take the risk to make something that's definitive? I'm not sure I know the answer to that, but what went through my mind as I was listening to your presentation was that the piece that was missing was, so what difference does this make? What am I going to do with it? And to me, that would make a big difference. Just cementing it in the medical record so it's on the problem list is not a terribly useful thing, I don't believe. So would I talk to the family based on the evidence you presented? I think I would, and I would try to give them a balanced understanding as best I knew about what this is. But until this becomes a finding that's been verified in lots of cases, which may never happen for many of the things we're talking about, I'm not sure I would say this is black and white end of discussion, this is what it is, but I think a lot depends on what they want to do with the information. If it's to offer testing to other people so they can have surveillance, I think there's a valid use and with the caveat that there are pitfalls, why not? Would you use it for prenatal diagnosis? The answer probably depends a lot on what the phenotype is and what the impact of all this is. So I think putting it into as a question in total isolation of the clinical usefulness of this makes it a very abstract conversation. When you put it into a clinical context, I don't think there's a black and white answer. My answer is I would talk to the family, tell them what I know and what I don't know, and engage in shared decision making.