I'll make a note of it. I'll probably... I remember last time. We've been there once. I know, but... Oh, yeah. Yeah, no, I'd love to. Yeah, let's just... Let's just... Yeah. That also... That way... I'll make a note of it. Thank you for watching. Thank you very much. Thank you for watching. Yeah, I'm sorry. I'm also sorry. All right. I was on here. She gives this to all of our professors, all of our students. I hired her to come talk to our high school. And then it was awesome. And I take it on with my students. And they usually know that. So that's the kind of, you know, I'm not really the same person as you, so I really appreciate it. I'll tell you my lab site. It started to stop. Why is this up here? I don't know. I honestly have no idea. This? Oh, no, no, no. Who said that you're supposed to do this? I think. I'm not sure. I'm pretty sure the moderator. That's why this is it. I think they were. Thank you. Thank you. and it's a lot harder. You know, during your... It's nice to be out in the city, you're probably too busy to go back to the village. Yeah, we're gonna help you. Hi everybody, welcome to this short course, we're going to get started because it's 1230. So my name is Heather Mefford, I'm a physician scientist from St. Jude Children's Research Hospital just down the road in Memphis. I'm a pediatric geneticist and neurogeneticist and I'm going to be your moderator today. So I'm going to give you a quick overview of kind of our goals for the day and give a brief introduction of who our six speakers will be. We'll have two kind of sessions, so we're going to have a short introduction, three speakers with questions and answers after each speaker and then a panel Q&A after those three speakers. We'll have a break and then we'll do the same thing with three more speakers, Q&A for each speaker and questions after that in a panel format as well. Welcome, I'm so glad to see so many faces in this room, we're so excited, there are so many attendees and I know that we have many attendees online as well, so I can't see you at the moment but we're going to get set up so that I can hopefully see who's here. If you're online, you should be able to enter questions into the Q&A and I will try to monitor that so that we can ask questions if there's time of either each speaker or the panel. If you're here, you can enter questions online if you're logged in or you can use the microphone to ask questions when it's time. If you do use the microphone, we ask that you wear a mask when you come up to ask your question and it would be great if you could introduce yourself as well. So today's short course is on, I think, an exciting emerging topic and that's the clinical application of long read sequencing and we're going to talk specifically about how to use this to end the diagnostic odyssey and increase diagnostic yield. So we all know, right, the diagnostic odyssey, we've all been through this if you're a clinician or even on the laboratory side or in the research lab. The diagnostic odyssey really refers to the journey that a patient goes on on the way to hopefully getting a genetic diagnosis to explain their findings. I'm a pediatric geneticist so often this is a patient who has either multiple congenital anomalies or developmental delays or pediatric epilepsy. The diagnostic odyssey we'll focus on today is on the genetic testing but of course this can also include non-genetic tests and often starts this way with things like imaging, so MRI, echocardiogram, EEG, et cetera, to define the phenotype. And then on the way to diagnosis, there are many different testing options and the ones in this room that we're most familiar with, of course, are things like the chromosome microarray or the old-fashioned karyotype which still comes in handy sometimes. Next generation sequencing has obviously changed how we diagnose patients but there are also many, many testing options ranging from small gene panels to large gene panels to whole exome, whole genome sequencing in some cases. The diagnostic odyssey is almost never a direct route so you can run into detours and here I've given the example of, of course, the variance of uncertain significance that any of these tests can give us whether it's non-genetic tests and imaging findings or variance of uncertain significance from your chromosome array, your gene panel, or your exome sequencing that require further workup of some sort. And then more recently there have been new testing options including methylation testing that have come online and are clinically available as well. And all of this, of course, is in hopes of ending up with a genetic diagnosis which we all know is not the case every time. Diagnostic odysseys also don't always come in the same shape and form, right, so you can take many different routes starting with different tests, ending with different tests, taking different detours and often it feels like you're going around in circles. This can take weeks to months to years. I think this overall, this has shortened over time and even in my career has shortened greatly with the advent of next generation sequencing but it can still take a long time and it's frustrating for both the patient and the physician. And we are still pretty much level out at about a 50 percent diagnostic rate across the board for patients that undergo even the full suite of testing that's available today. So our goal is to shorten this timeline and increase the diagnostic rate and, of course, we can do, you know, think about approaching this in many ways but what we're focusing on today is how to use long read sequencing to accomplish this. It won't be the final answer and it won't get us that, probably won't get us that entire 50 percent that we're missing but we all believe here in this room, that's why we're here, that it is going to help with both of these important issues. So I'm going to give you one quick example. This is a diagnostic odyssey that actually happened in the research lab but it offers many of the challenges I think that we face and reasons that we miss diagnoses and why we need better tests. So this example is a disorder called Baratella-Scott syndrome. So these are two children. The little boy standing up there is actually nine years old. The girl is five. So they have severe short stature and developmental delays. And one of my colleagues when I was in Seattle actually recognized that these two patients looked a lot alike and both had a deletion of 16P13. But this deletion did not explain their features. Most patients don't have especially the severe short stature so we said, well, maybe we should exome them or let's do something. And then we looked at the deletion and realized that one of the genes in the region was XYLT1. And it had been recently shown that recessive mutations in XYLT1 cause a severe short stature syndrome called Baratella-Scott syndrome. So being the wise PI, I said to my research scientists, oh, just sequence the gene. This will be a cute story. Deletion, mutation, disease, you know, two cases, it will be great. So we tried. We used Sanger sequencing first to sequence the gene. We used targeted next-generation sequencing, captured next-generation sequencing to sequence the gene. We used targeted next-gen sequencing. We used exome. We used whole genome, both low coverage and high coverage. We used chromosome array to look for deletions in the gene. And long story short, we could never sequence the first exon and kind of promoter region of this gene. And we did not know why. So long story short, after banging our heads against a wall, trying many different approaches, approaching collaborators who were working on the same syndrome, we figured out along the way that in the 5' UTR of this gene, there were 238 base pairs when we could sequence that region. They actually weren't in the reference genome. And within those 238 base pairs, there was a CGG repeat, also not annotated in the reference genome. After a lot of work, we figured out that all of the CPG sites of the first exon and that upstream region that we could sequence were methylated. And this actually was our collaborators who figured this out. And when we went back to our patients, we said, well, yeah, look at that. The remaining allele in this whole region is entirely methylated. We still couldn't sequence the whole thing. But we hypothesized that that CGG repeat was actually expanded in our patients, very similar to Fragile X syndrome, for example, and that the reason we couldn't amplify the region and sequence it is because there was a large GC-rich repeat and that it was hypermethylated. And lo and behold, we went back to the old-fashioned southern blot and showed that this was actually the case. So this was true for both of our patients. And together with collaborators, we actually collected 10 affected patients where 50% of the alleles were missing, so we couldn't actually find a sequence variant in those alleles. And when we went back and did a targeted test to look for hypermethylation and repeat expansion, every single one of those missing alleles was just like these patients here. So from the time that I said, just sequence the gene, to the time that we actually figured out what the missing mutation was, was well over two years. It was probably four years, actually, until we finally published the story. So it was hard. These alleles were difficult to find. So we worked forward a year. We collaborated with Danny Miller, who's going to speak today, and Evan Eichler, who had heard the story and said, well, what if we tried long-read sequencing on this patient? Turns out that that four years of work in one single sequencing experiment gave us all the answers. So we were able to use long-read sequencing, and this is showing you reads from the patient carrier, from the proband and from the mother, who's a heterozygous carrier. We were able to find the unmapped, the sequence that was not mapped to the human genome. We were able to see the repeat. We were able to see the expanded repeat. And in addition to that, using Oxford Nanopore sequencing, we were actually able to call methylated alleles and show that the repeat expansion was hypermethylated. So one experiment gave us all the answers. So this is our hope, right, in that long-read sequencing can be used in this way to find those hard-to-find alleles to give us new answers, shortening the timeline and increasing the diagnostic rate along this odyssey. So today you're going to hear from six speakers, and I'm just going to briefly give you kind of their names and topics here, and then we'll have each of them come speak, but we're going to start with Dr. Emily Farrow from Mercy Children's Hospital in Kansas City, who's going to discuss using this technology applied to diagnosing repeat expansion disorders, which I just introduced, and genetic counseling issues related to that, as Dr. Farrow is also a genetic counselor. Dr. Susan Hyatt from Greg Cooper's lab at Hudson Alpha Institute is going to talk about using long-read sequencing for diagnosis of early-onset congenital disorders. And then Dr. Danny Miller from Seattle Children's Hospital and University of Washington is going to talk about applications of long-read sequencing for finding those missing alleles. We'll have a short break after those three speakers and a panel Q&A session, and then in the second half of the course we're going to have Dr. Winston Timp, who's from Johns Hopkins University, talking about using long-read sequencing for some applications you might not think regularly about, such as the epigenome and transcriptome analysis. Similarly, Dr. Andrew Sergacis from the University of Washington is going to talk about using it for finding gene regulatory impacts of sequence changes in some very creative ways. And then at the very end, we're going to kind of wrap it all up with not necessarily applications of the sequencing itself, but applications of how do we then use the answers that we find to perhaps be able to start to think about treating some of these disorders in a very targeted and precise way. So Dr. Frank Bennett is from Ionis Pharmaceuticals and is going to talk about antisense oligonucleotide therapies. So we're going to start with Dr. Emily Farrow. And as I said, after Dr. Farrow's talk, if there's time, we'll have one or two questions for her specifically, move on to our next speaker, and then after all speakers, we'll bring everyone up again for a panel discussion. Thank you. Thank you very much. It's really my pleasure. I'm very excited to be here and be part of this short course and be part of our panel. And it's good to see people again because it's been a long two years. So today I'm going to talk about the challenges and genetic counseling considerations of systematic testing of expansion disorders. And as a disclosure, I'm part of an advisory committee member for Illumina on genome sequencing. So before we start, and to make sure everyone's awake and ready to engage for the first in-person meeting in probably a while, we have a polling question. And how would you describe long-read sequencing in one word? And so I think you can, this is the first one, but if you scan the QR code, then you could submit. So give me one word about the first thing that comes to your mind when we say long-read sequencing. I'm loving it. These are all terms I've used. These are great, expensive, comprehensive, I saw confusing, complex, error prone. So I think these are all things that I have exciting for sure, difficult. These are all things that I have, again, personally thought I'm glad that we have a mix in the room because I think these are all really appropriate. Wow. I hope that by the end of our session today we can kind of convince and so that we all can think these terms or we can all at least view it in a better way and at least address some of the confusing and difficult regions so that more of us have the exciting with the exclamation point for that, a magical. That is a great one to end on. So let's stop there. Okay. So I just want to give a little background because I suspect that they have very different levels of people who have been able to engage with long-range sequencing as we just saw. So just a little to set the stage for the afternoon. So our center, like many centers, is wholly embedded within a pediatric institution and that's because we all know that rare disease is not really that rare collectively, right? So there are about 7,700 genetic or inherited diseases that are impacting 400 million people globally. That means one in 10 Americans are touched by rare disease and about 50% of them are children. So molecular diagnostic yields by symptom-driven clinical exome in patients with disease can range from 30 to 40%. I think when we see higher than 50%, I tend to think those are on specific patient populations where we know we're doing better with our diagnostic. But regardless of how you look at it, if our best case scenario still means that half of our children aren't getting a genetic diagnosis. So we have room to grow. Also job security. But what we've also seen is that the addition of short-read genome sequencing is only incrementally increasing our diagnostic yield, right? And part of that's because our exomes have gotten better, but still there's more for us to learn. And so in addition to the novel discovery in rare disease where we've seen an exponential increase in the identification of new genes, new disease genes, new gene-disease association, the other thing that we're beginning to appreciate, at least me personally, more and more is the complexity of the human genome, right? So we know that our single nucleotide variants, which are accurately and efficiently diagnosed with our exome and genome sequencing using short-read technology, are only one component of the piece and probably why we're still missing diagnostics, right? So we all recognize in this that we have copy number variation, we have expansion disorders that I'll talk more about, we have methylation and imprinting disorders, we have structural variation that's difficult to detect sometimes, and we have mitochondrial variants. All of these are contributing to our rare disease and are something that we need to be mindful of when we're doing our testing. So with our exponential and comprehensive sequencing approaches, why are we still looking at long-raiser? Why are we so excited? And that's because we still have many challenges that still exist. So we all know that trio analyses are powerful, but the truth is that parents aren't always available. And so even when we don't have parents, obviously we can't detect a novel variation, but autosomal recessive disease is still a significant contributor to our rare disease, right? So we have singleton analyses, we have coverage considerations, which still exist even with PCR-free genomes, we have repetitive regions in the genome, we have pseudogenes, expansions, and we have structural variant detection, such as deletions, duplications, inversions, which can be really difficult, and translocations. And so if we put on our clinical hat and kind of step away from the research realm, what this really equates to is that we're running multiple assays, as we just discussed, which can be costly, time-consuming, they're limited to knowns in the clinical world, really, and we have an assay donut hole, where we know we still don't quite overlap all of our technologies. And so, again, if we take another step and think about all the different testing algorithms that may be ordered when it comes to our clinical patients, if we have a patient that presents in the NICU, and we suspect an aneuploidy, then it's obviously very simple for us to order a karyotype, and we usually get our answer. But when we have a case such as congenital hypotonia, suddenly our testing algorithms get much more complex, and knowing where to go first with our testing. And really what this ends up, then, is that we have multiple testing assays again, and that's equaling time and money, which is important for the laboratory and more important for the patients. Right? So if we think about our sequential testing, because we're probably not going to order them all at once, we start getting several weeks, which can stretch into years if we don't get an answer. So the addition of copy number variant detection to exome and genome sequencing, short read exome and genome sequencing can begin to condense these down, and we're seeing that more and more in your assays that you're ordering. But really, when we start to look at our long read genome sequencing, we can really start to convince multiple tests down into one, that we could then just have one single test that we could order, and hopefully be much more efficient in our detection. So what's the fundamental difference? Like, what's that foundation that makes long read sequencing so exciting, regardless of what platform you're using? And really, there's a fundamental start that all next generation sequencing technology requires the fragmentation of intact DNA. So we're going to tell you from the laboratory that we need more DNA always, right, and we always want it to be nice, pure DNA, and we want to have a high quantity. And then the first thing we're going to do when we get it is we're going to bust it up into pieces. So here what we're seeing on the left is that our genomic DNA, this is a gel showing our genomic DNA in our laboratory, which is really, this is about 100,000 base pairs. It pegs out the system that we get. And then in our short read technology, we're going to fragment that into about 425 base pairs, and that's what we're going to take in through sequencing. So in contrast, when we're using in our laboratory our HiFi genome sequencing, we're fragmenting the DNA down to 12,000 base pairs. So that's a significant increase, right? And in the end, this has a big impact on our alignment. And so if we think of our human reference genome, those 3 billion base pairs, and we're going to align our reads back to that, again, regardless of the technology, and our short reads, we're going to align those paired in sequencing, these 150 base pair reads back. And you can see here we're doing really good in the alignment in what I'm showing you because, of course, I'm going to show you the easy part. The contrast of what it might look like or what it thinks in the informatic world, when we take the long read and we can get one read that spans all of that variation. And so that's why these long reads are providing better alignment over repetitive regions, particularly with pseudogenes or expansions. Obviously, though, we both are doing well with single nucleotide variant detection. So here we can see our made-up variant highlighted in red. So I'm a visual learner. When I think of this and piece it together, when we look at our short read sequencing, we're really piecing the genome back together, two by 150 base pairs at a time. And we do pretty good when we think about our sensitivity and specificity and finding out where those puzzle pieces are, where those individual reads go, right? So we can fit that puzzle back together and be really happy. But when we take a step back and we start looking at it in 12,000 base paired chunks or even larger for some laboratories, what we actually see is that the puzzle goes together a little bit differently. And we can start to see that variation that we weren't able to piece together with the short reads. And so this is how I think of the difference between our short and long read sequencing. So how are we using this technology at Mercy? So the Children's Mercy Research Institute and the Genomic Medicine Center at Children's Mercy have created a pediatric data repository to facilitate the search for answers and novel treatments for pediatric genetic conditions. Under this umbrella, 30,000 children and their families are going to be enrolled. So up to 100,000 families will be enrolled and sequenced over seven years. And through this GA4K, we have comprehensive sequencing and analytics strategies including short read exome and genome sequencing, long read genome sequencing, whole genome bisulfite sequencing and single cell RNA for our functional assays, just as an example. And to look a little further specifically at our high reads, we have a systematic reflex to HiFi genomes after a negative exome analysis. So just keep in mind that all of the samples that are going into our long read sequencing currently have already had a negative exome analysis. And we have a goal of reaching 1,000 genomes within a year, which we're quickly coming up on, which we're excited about. In order to do that for our sample preparation, we've automated that to support kind of our production sequencing. So we use a robot, the Comagic 360, that gets us our genomic DNA that has 100,000 base pair fragments. We then are, again, as I mentioned, sharing them to 12,000 KB. The one important consideration for long read sequencing is that it does still require significantly more input of DNA than short read. So we're taking five micrograms of DNA, which is quite a bit more than what you would take in for a short read. And that can be a limiting factor for some cases. And then we've also automated our library preparation. So we use a cyclone system that allows for us to process up to 96 samples at a time to feed those sequencers. So just to give you an idea of our analytic strategy of what we're doing, all of our HiFi genomes are aligned to build 38, and we're moving to lift our clinical testing over to build 38 as well. For variant detection, we use multiple tools, including Deep Variant, WhatsApp, and PBSV for structural variants. We have a novel tool set to annotate our structural variants, and then we're running a tandem genotyper that annotates a known set of 52 loci for expansions, and that's what I'll focus on in just a minute. And then, of course, we deploy QC metrics to make sure that we have high sensitivity and specificity. And, in fact, what we find when we look at our high coverage genomes is that our single nucleotide variants and indels are highly concordant with short-read genome sequencing. In general, in our HiFi sequencing, we see 450,000 single nucleotide variants and indels and about 10,000 structural variants per genome that were previously undetected with the short-read genomes. Now, admittedly, most of these are in tandem duplications or common, so they may not have been contributing to the clinical phenotype. So if we apply a simple minor allele frequency filter of 0.01, then we have about 150 rare variants per each case for us to investigate, and that's a manageable number. So, as I mentioned, we're approaching finishing our first 1,000 HiFi genomes of probands and parents with rare disease. And so here is just showing you a snapshot of our first 80 cases that had both short-read genome and long-read genome to show the increased number of calls that we're bringing on structural variants. And notice that we only are seeing the inversions with the long-read. And then here on the right, we're comparing the sensitivity, and in general, with our PacBio at 30x coverage, we're seeing higher sensitivity and specificity than our short-read genome sequencing. All of our samples under the research umbrella receive an Infiniium global screening microwave, and that becomes our true set that we use for our comparisons. I think the caveat to point out here is that you look at the scale, and both of these sensitivities are well over 98%, so I think we're splitting hairs a little bit. I think the importance here is that we're getting high sensitivity with long-read sequencing, which wasn't previously accomplished. So now if we move to the systematic detection and how, let's take a closer look at how this is impacting our expansion disorders and some of the counseling conditions that we think of that. And so we always like to start with knowns and looking at established disorders and how that's been our experience there. So again, as I mentioned in the tandem genotyping tool, it's completed for each sample, and we currently are looking at 52 loci. One of the exciting and important things that we're doing is that we're creating a database of normal variation, which was previously unavailable. We've seen excellent concordance for positive controls. And so now I've told you all the good, and there is a caveat, of course, and that with some of the expansions, we do have some difficulty in reading through with the HiFi reads. So the GAA repeat and Friedreich's ataxia and the CTT and myotonic dystrophy, we have observed that we do have some allele dropout of the expanded alleles. What's been really exciting recently and what we've been able to do with our HiFi genomes now is that we're able to detect the native methylation using the HiFi long-read genomes. And so we can see the methylation of expanded alleles, which is serving as a functional validation of those expansions that we're seeing. And so even though we might see an allele dropout, we can see that that allele is methylated. So here again, we start with our knowns and our positive control. Here's an example of an FMR1. So CMH3372 was known to have a full expansion of about 200 repeats. And so we ran this as a test and a positive control. And they had clinical testing. And so the HiFi genome detected, of course, the 200 repeats, but also detected instability of the expanded allele. So here we can see in our IGV it's showing you 12, an insert of 1,200 base pairs. Here we have some at 1,000 and some we have at 600. So just remember to divide by three for this one. And here's an example of what our tandem genotyping tool looks like for the expansions. It will normalize back to account for the reference expansion. So here we can see most of our alleles are at 200, but you can see that it did detect and report out in the genotyping tool that we have instability. So this was exciting because the clinical testing will accurately report that it's greater than 200, but that's all with the triplet repeat, primed repeat testing that it will report out. So it's nice to start with a win. Similarly we had a myotonic case, a case of a patient with myotonic dystrophy that had been diagnosed clinically. And so here we're showing you the same thing on the right where I have the tandem genotyping tool showing us that one allele has an expansion of about five repeats, which would be a normal, and one is at 1,600. And here we can see that we phased those two alleles, so we have our allele 1 and our allele 2. These purple right here inserts are showing you where it's calling the expanded insertion, and you can see that we did see some allele dropout. This one has very high coverage. But importantly, what we're seeing in the same assay on the PacBio HiFi genome is that we're also capturing the methylation. So here you can see that there's a hypermethylated region, which is what you would expect. So even though we have an allele dropout, we could see the functional validation. And this was, again, completely concordant. So this is exciting for us, because going forward, we could do this without agnostically, right? We could just run it for our patients. So I promised you this was not planned at all, but we also had a really interesting case of Baratella-Scott syndrome. So again, as Heather mentioned, this has now been shown by her lab and their beautiful work, that there is an expansion in the promoter that has been associated with Baratella-Scott. So we also had a patient that had been on a diagnostic odyssey and clinically was given a diagnosis of this, and we couldn't find the answer. And so we reflexed them to HiFi genomes, thinking that we'd find the expansion. And when it first came off, we were very excited, because we saw, indeed, a homozygous expansion in the promoter. And we thought, this is it. We have the answer. And so we also then came to the same conclusion and found that, wait a minute, build 38 does not account for 238 base pairs. And so it looks like your samples are expanded when, in fact, they just are really matching what should be the reference. And so what we were able to do then is that we were able to correct for this in the tandem genotyper. And when we went and looked at a larger distribution, again, as we build up our warehouse, we were able to see quite nicely that our patient did not have an expansion. In fact, they fall square within the normal region, and this was not their diagnosis. We were able to find, with long-range sequencing for this patient, that they had a variant in another gene, B3GALT6, at the start codon, which is also difficult to detect. And so we were able to use the long reads. But this was just showing to us the importance of starting to build that normal variation so that you can see what looks rare, as far as the expansion, in fact, wasn't. In fact, it was normal when we accounted for the error. This is not too dissimilar from when we started with exome and genome sequencing at the beginning, when we didn't have these beautiful public warehouses that showed you allele frequencies. So hopefully I've convinced you that we're able to do this well on our known systems. But what does this look like when we start to look at things that maybe aren't as well-defined? And how are we going to challenge ourselves there? So if we look at novel disorders or intermediate alleles, the BUSs of expansions, right? The things to think about when we're starting to systematically or we're beginning to look at systematically our expansion detection is that expansion disorders often result in later adult onset conditions. And we can have anticipation where the parents might be affected in adults, but we'd see much earlier onset in children. One of the cabinets for this is that the natural history for many expansion disorders is not well-known. So as we're just beginning to recognize the expansion disorders, we don't have the same variant warehouses that show us what the range is. And so we don't know the natural history the same way that we do for others. And I think we all need to be aware that there's a significant ascertainment bias due to testing limitations. So in the case of baroteloscopic, it's not a clinical assay that you can order, right? We have to use other mechanisms to begin to look at it. So how are we beginning to approach these expansions beyond the current clinical diagnostic testing? So here's a case study, another one of our case studies, CMH1541, that was referred for our research program. And this was a very large family. I'm just showing you a snippet of the pedigree that had multiple affected individuals with dystonia, seizures, and ataxia with variable degrees of severity. And really, they came to the clinic because the children shown here, this was the first one that came to clinic, presented with infantile onset. So they were quite severely affected. But once we started looking at the children and talking to the parents, it was realized that the parents actually have adult onset dystonia, that they really hadn't noticed until they were older and may not have even noticed if it hadn't been for the children and various student clinicians working with them. And of course, as you might imagine with this family, they'd had a significant negative clinical testing of a microarray, an exon array, panel testing, exome sequencing, and clinical genome sequencing, all of which had been negative. They were referred and initially received a research exome and research genome, which were negative, which is good when they agree, I guess, with clinical. And so on the research side, we enrolled nine family members in total. And so given the phenotype of dystonia and extensive negative testing and really our growing suspicion that maybe we had an expansion disorder given that the parents were affected as adults, we referred them or sent them through our HiFi sequencing. And we completed this actually on four family members. So initially, our tandem genotyping tool did not identify any expansions and no known disease genes. And this was partially us because it was beginning just combing through and looking at these individually. So then we took a candidate gene approach. And as it turned out, the mantra that three hours in the library will save you three months in the lab continues to be true. And that not long before this family had been sequenced, a publication had come out about a new repeat expansion, a pentamer expansion in this case in star D7, that had been associated with familial adult myoclonic epilepsy linked to chromosome 2. And so this was really a kind of a novel or emerging fame disorder. And so sure enough, when we went and looked at our family and our affected individuals, we saw that they all have an expansion of about 200 repeats right in the same region of the intron that had been published. So here I'm showing you two individuals. And here's the genotyping tool, again, showing you one allele within a normal range and one that has an allele expansion of a little bit over 200. So we began to work our way through the pedigree. And we found that all affected family members actually have this repeat insertion. And we then confirmed this repeat because this was a novel disorder and not really well known. We confirmed this with a repeat primed PCR. And I'm showing you a snapshot of one of those patients down below of the expansion. This repeat pattern is absent from all of our sequence samples. And a control set of 40 healthy adult individuals. I think the caveat is, though, even though I'm telling you that our warehouse is growing, we're still only approaching 1,000, which pales in comparison to the amount of short read data that we have when we look at variant frequencies. And so do we have a novel disease gene for this family? And I think in this case, we've worked quite closely with the clinician and with the family in that we think we do have a novel presentation or an expanded phenotype of this disorder. Because again, this is the only family that we personally have seen this expansion in. And we've confirmed it with repeat primed PCR. And it fits with the phenotype and fits with the pedigree in this family. So we think this is a novel disease. But this isn't a slam dunk. And it really has led us to have quite a bit of conversations about how to report this and how to work with this family. So of course, so that was great for that family. And we, they're not always so straightforward, even when we have novel disease presentations. And so we have a second proband, unrelated, that was referred for dystonia. And part of this is that we are preferentially pulling our patients that have dystonia or neurological disorders and sending them for HiFi sequencing when possible. And so our second patient, CMH1592, was also reflex to sequencing. And here we identified an insertion of 80 repeats. So we haven't seen a repeat expansion of similar size in our internal CMH HiFi samples. And so now we're left with this conversation, is this normal variation? Is this an expanded allele? Is this an intermediate allele? What do we do with this finding? And here again, I'm showing you our tandem genotyping report. So we did the phone a friend option that I didn't give to you just a second ago. When we went back and looked again at another healthy adult, we actually have found one other person that has an expansion allele. So again, this is a healthy adult individual with the same insertion size. So we have one other person that we've now identified that has an 80, about an 80 repeat expansion. And so we're still faced with whether this is normal variation, whether this is an intermediate allele, or what could this be? And that's because, again, we don't have the natural history yet defined for this disease. So because this is under a research umbrella, results have not been returned, because really it's unclear what this allele means. So again, whether this is an intermediate or a pre-mutation allele. And under the research protocol, diagnostic results are returned. So what does, if we put on our genetic counseling hats and our clinical hats, what do we, what are our guidelines telling us about returning results? Because again, the research umbrella is different, a little bit different than our clinical. And so the NSGC in 2018 has this statement on genetic testing of minors for adult onset conditions. And it encourages deferring predictive genetic testing of minors when results will not impact childhood medical management or significantly benefit the child. And so predictive testing should optimally be deferred until the individual has the capacity to weigh the associated risk and to preserve his or her autonomy and a right to an open future. So I think most of us are used to that, that we shouldn't do predictive testing in children. And this is pretty easy and straightforward for many of our loci and disease, and known disease genes. So if you look at any repeat detection paper, all of them from a research standpoint always include Huntington's disease. Even though I think all of us in the room would realize that we're not going to report or do systematic screening for Huntington's disease. The reason that's done from a laboratory perspective is that it's relatively easy to detect as an expansion. And then there are well-established ranges. So we have quite clearly what is our normal, intermediate, pre-mutation, and affected. And it's also very rare. Right, so this one's easy. We're not going to report it. We're not going to look for it. We can wipe that one off, I think. But other loci are less clear. So what are we going to do with intermediate alleles and myotonic dystrophy if you have a child that maybe is a little floppy and you happen to find it? Or what about our pre-mutation or intermediate alleles in Fragile X in females? Those become a little bit less clear. So the American College, the ACMG, released a statement in 2021, a revision on the technical standard of laboratory testing for Fragile X. So again, if we go back to our guidelines, and they say population carrier screening and newborn screening for Fragile X are somewhat controversial and not recommended at this time. It's important that clinicians and laboratorians refer to the guidelines referred to make clinically relevant decisions. And then again, if we go back to our NSGC guidelines and see what we're seeing on the counseling side is that voluntary screening for FMR1 mutations appears to be acceptable to women in the general population and to parents of affected children. Genetic counseling and education are essential, and that newborn screening of Fragile X is also under consideration, but further research is needed to determine its acceptability and efficacy. So again, if symptomatic, very clear, but we're still working out what we should do as far as systematic testing in general. So we have another case study of CMH2215, and this child was referred for testing for constipation, dysphagia, reflux, global developmental delay, obesity, and sleep disturbance. And in fact, clinical microarray had already revealed a duplication that had been interpreted as a VUS, and that's because the orientation of this duplication was not known. And it contained three genes. So they subsequently had clinical exome sequencing to try to find either an alternative diagnosis or refine this duplication. And in fact, exome sequencing was able to tell us that this duplication is in fact an inverted duplication. And because it's an inverted duplication, it's predicted to lead to loss of function of the gene FRMPD4, which is associated, and loss of function is associated with X-linked intellectual disability with variable features, including seizures, developmental delay, and autism. And so this was found clinically to be a very good fit for this family, and so then that VUS was recategorized. But they were also enrolled in our research study, because again, this is for us to not just find diagnoses, but also understand genetic variation. And we completed a HiFi genome sequencing to really look at the duplication event and use this as a control. And so again, what we found with our HiFi genome, again, we were able to phase the alleles. Here you can see the bump in coverage where the duplication starts. And here, again, we're able to confirm what we had already seen in our short read, that again, this duplication was in tandem, which is important for its interpretation. So this part is all pretty straightforward and great. But once you're enrolled in research, that means research can happen, right? On your samples, that's part of our program. And so there's continuing analyses as we continue to develop our tandem genotyping and look at it, which includes, again, building our variant warehouse of expansions. And so continued analyses detected an expansion of FMR1 in this patient of about 117 repeats. So this is a female that had been referred for developmental delay, and we now have a likely pathogenic variant that was felt to explain her phenotype. So the research team came back to us, and at first, they were really excited because we found this. Our genotyping tool is great, and we should confirm this and report it back. But what should you do? You as the genetic counselors and research team, what are we going to do with this variant? So now I'm going to ask you guys to make sure we're all still awake. What would you do with this FMR1 finding? So your options are to clinically confirm and report the finding back to the family, not pursue confirmatory testing and not report, discuss the findings with the GC, or secretly be really thankful that it's under a research protocol. So again, we have our QR code, and we can give it a minute to see where people It looks like we have consensus. I don't disagree as well. So we have clinically confirm and report the finding is our leading answer here. We're still thankful it's under a research protocol, good with that. So actually, in this case, the clinical laboratory does not report pre-mutation alleles in children, particularly in females. And so in light of the diagnostic finding in the laboratory policy, so again, under a research policy, only diagnostic findings are returned back to the patients and we have a diagnostic finding. The finding has not been reported back to them. But really, this got us thinking about how we're going to approach these and maybe it's not so clear cut as it is with fMRRM, which as we saw, actually is really not as clear cut because we all would have done it a little bit differently. And so this precedent was set for additional findings, which I think may need to be reevaluated in the future because we can look at our guidelines and what they're telling us now. But I think those of us who are able to work with patients in a genetic counseling or clinical perspective, patients are getting savvier about their genetics and their genomics. They're wanting to know more. We're adding more and more findings to our secondary findings that we're reporting back. And so I think this area of not reporting back findings that we have doesn't always feel comfortable. It may not always be the best thing, even though our guidelines may be telling us another thing. So it really got us into a good discussion on something that's ongoing, at least within our center of how to handle these going forward. And I think we're not the only ones that are feeling this. And so again, if we go back and look at that 2021 from the ACMG, this is taken straight from them, that single-molecule, real-time long-read sequencing is able to sequence through a full mutation of repeats in fMRR1 and may be used to distinguish the number and location of interruptions, which is true. And although long-read technology is not yet widely available for clinical use, this may change as error rates and costs decrease and more bioinformatic tools become available. And so currently, TP, PCR, and SILRBLOT remain the gold standards for the identification of expanded fMRR1 alleles and repeat quantification, but I think our data and even the ACMG recognizes that this is short-lived and that we're still going to be adding more assays to this statement. So again, I think as we look forward, continued long-read sequencing is going to result, it will continue to result in more diagnoses. We're going to continue to refine our disease alleles. I think we're going to identify novel expansion disorders, and we're going to further define normal variation because I think there are some repeats in some areas that we have been very rare. They've been difficult to ascertain, but when we start to look at those in our warehouse at least, we can see that there's quite a few people that have an expansion. And so the story of XYLT1 where we have missing reference material, difficult to expand, but they're looking like there's an expansion, I think we're going to have that story play out for more loci as we go forward. And we just haven't been able to see them systematically before. I think the evaluation of guidelines and policy statements will likely need to be revisited, especially as detection of these expansion disorders is improved. And I think looking even further, the future of widespread implementation of de novo assembly, so I didn't touch on that in my talk, but that's going to continue to reduce the bias and lead to more accurate detection of variants. So in summary for my part, I think hopefully I've shown you that long-read sequencing is providing accurate single nucleotide variant detection with superior detection of structural variation and repeats. The systematic sequencing of a rare disease cohort has led to the discovery of a novel expansion in addition to detection of knowns, and that laboratories should evaluate interpretation and return of result policies as testing increased, is increased to balance future autonomy with diagnostics. And I would just like to quickly acknowledge our team, especially the director of our center, Dr. Tomi Pastan, and all of our collaborators within Genomic Answers for Kids, and of course, most importantly, the patients and families for their participation. Thank you very much. We have time for one or two questions. I didn't see an online yet, but if you can just keep an eye on that for me, Susan. Great. I think we have one question here. Thank you, Bimal Choudhury, Columbus, Ohio. Great talk. Since your subjects all undergo short-read sequencing before long-read, do you have head-to-head data on repeat expansion detection using computational methods like Expansion Hunter and other things for short-read versus what you can actually get, which is presumably better for the long-read? Right. So we do have a small subset of patients that did have PCR-free short-read genome sequencing where we were able to use Expansion Hunter. For those known, such as Fragile X, the FMR1, it was able to detect the expansion and give it to us. So I think for the known well-defined loci, we did have success with the Expansion Hunter. I have one quick question. For your FAME case, the intronic pentonucleotide repeat, I may be remembering wrong, but I thought that the repeat sequence actually changed when it expanded, so were you able to see that, and were you able to use that, for example, for that 80 base, or whatever it was, 80 repeat expansion, to determine whether it's pathogenic or not? Right. So we were able to read through and see the expansion and read through all of it to see how it changed with it, and it did help us a little bit with that intermediate allele. We were still, I think, maybe it was a bit of a red herring because it was so unique and not seen in other patients, and we were really hoping to find a diagnostic. You're correct, and the sequence does matter. All right. We'll bring Dr. Farrow back up at the end, too, for panel discussion. So thank you. Our next speaker is Dr. Hyatt from Hudson Alpha. That's not it. And can you start the timer back there, or do I need to start it up here? It's okay. There we go. And that includes questions. Okay. Sounds good. Awesome. Thank you so much. Yes, that was a great first talk, and it is so nice to be here with so many people. This is so great. It has been a long two years, but this is what we have been working on for the last two years, and a little bit longer. So I'm excited to tell you about our experience with clinical application of long-read sequencing to neurodevelopmental disorders and congenital anomalies. And so, again, I'm Susan Hyatt. I am a senior scientist at Hudson Alpha, just a few more hours down the road in Huntsville, Alabama. In case you want to come for a visit sometime, we have a very large double helix walking path. So if that helps get you there, we'd love to have you visit. And I have nothing to disclose. So in our lab, we are interested in the genetics of neurodevelopmental disorders, or NDDs. So these collectively affect about 1% to 3% of children. And although they're collectively common, the individual genetic disorders are often rare. So, again, as both Heather and Emily mentioned, a lot of times these programs go through a diagnostic odyssey. Often it starts with chromosomal microarray, karyotype, fragile X testing, and many of those still don't get answers there, and so this often leads to gene testing, either individual genes or panels, or exomes or genomes. But even after that, and I apologize, we tried to fix this several times, but even after that, over half of rare disease probands sequenced with genome sequencing still don't receive primary findings. So they don't receive things that explain their rare disease. And so these are our results in the lab over the last few years. We've had about 1,800 probands with rare disease. Most of these are neurodevelopmental disorders. Some are congenital anomalies, and some are adults with rare disease as well. So, again, we still get that about 56% of probands who received nothing, not even a VUS that could be interesting and followed up on. And so we think about this 50% to 60% a lot and how we can shift those probands from the no-returnable section to the pathogenic, likely pathogenic section. And so we've approached this in a few different ways. We have really worked on reanalysis, so going back and just looking at the genomes or exomes again to look for new disease genes maybe that have been published. We've submitted a lot of genes to GeneMatcher, which has been really fruitful for us. We've published about 20. We've been part of about 20 publications over the last few years. But, again, we were thinking, what about our sequencing technology? We know that there are regions of the genome that are not sequenced well or covered well in short-read sequencing. So is our sequencing technology not part of the problem necessarily, but is it an option? Is it a place for improvement? And so we wanted to ask the question, can long-read sequencing identify missed variants of interest? And so the probands that I'll tell you about today come from two different studies. So one of our first studies is the CSER I study at Hudson Alpha. And this was an NHGRI-funded study where we looked at almost 600 affected individuals with developmental delay or intellectual disability. And all of these probands were at least two years old at time of enrollment. So they were already going through a diagnostic odyssey in a lot of cases. And so as that project wrapped up, we wanted to try and get that diagnosis to probands even earlier. And so in our next study, our CellSeq study, we sequenced neonates with congenital anomalies. And so you can see in both of these, we still had that 50% to 60% no-returnable rate. So we started a pilot with this CSER I study. And we chose six proband parent trios with neurodevelopmental disorders who had negative genome sequencing in our study. And then we generated PacBio HiFi reads using CCS technology for all of these 18 individuals. We were shooting for a 30x coverage in probands and 15x in each parent. And so I'm not going to go through this whole table, but these are the six probands that we selected for our first study. And so the major phenotypic features here you can see often include seizures, ID, hypotonia, some other features. And again, a lot of these probands have gone through a lot of genetic testing and still have had no answer. And again, we did either genome sequencing or both exome and genome sequencing in our case through our study. So since we started this pilot, we have published this data. And there have been plenty of other papers and presentations showing this, that long read data does allow more accurate alignment and variant calling. So in our paper, we found more biologically relevant counts of de novo SNVs, more biologically relevant counts of de novo aloo insertions, and better mappability in low complexity and low mappability regions of the genome. All of those better than in short reads. Again, just longer reads will allow better alignment and better variant calling. So when we first got our long read data, one of the places we were excited to look were in these repeat expansions that Emily told us all about. And so a lot of our probands have developmental delay. So we wanted to, and some of them had had Fragile X testing, but most hadn't. And so we wanted to look at FMR1. And so this is the 5' UTR of FMR1. You see that CGG repeat expansion region in the orange bar. And then you can't probably see the numbers from here, but on the top, I'm showing the proband CCS reads. They're long reads on the top. And they are clearly and easily split into two different haplotypes that are really evident. And there are insertion sizes of either, in this case, 69 or 9 nucleotide insertions. The insertions are very easy to see. At the bottom you can see where our short read sequencing is aligned. And there are no variant calls in there. You really can't see it in this particular case. So we did not. We are also looking at those 52 loci. And Emily, we need to talk about Star D7 as well. We probably have another red herring. We haven't found any pathogenic variants in these yet, but we are still looking. So I am going to try and do another one of these Slido word jumbles. So this is one of the first cases that I want to tell you about is this event in Proband 6. And so if you have any familiarity with IGV, you may be able to tell what's going on here. But I would like to know, just by looking at this in a word or two, can you tell what this is? And so I think, yes, there we go. I think this will work the same way as before. And if we get the word confusing, that is totally fine. Maybe I should have let you look at it a little bit longer. Yep, that sound that's perfect Yes, so this one is actually kind of tricky It is sort of multiple things and I will show you as we go along to how it got called differently with different tools So in the most Yes, there is some soft clipping going on there Okay, so in the most basic sense proband 6 has a de novo insertion in cdkl5 and so cdkl5 is on the p arm of the X chromosome and Variation in this gene is associated with X linked dominant epileptic encephalopathy and this proband that we sequence does have seizures So as you can see here, there's a 7kb insertion just after X or yes Just after exon 3 and so when we looked at this insertion a little bit closer again, it was about a 7kb insertion and Inside that insertion we found some repetitive sequence We found l1 HS sequence, but also l1 pa5 some alu sequence some GA rich region And then at the far end of our 3 prime end of an insertion We found another exon 3 of cdkl5 and so at first pass this looks like a L1 mobile and mobile element insertion, so it's got some l1 sequence inserted it has that classic target site duplication or TSD, which is in the yellow box and Exon 3 of cdkl5 just happens to be in that target site duplication However, this is more than just l1 sequence that is inserted so it's not It doesn't look like just a plain old l1 insertion. So when we looked a little bit closer again There's that other l1 pa5 sequence some alu repeats and some GA rich region And when we looked at this that 3 prime end again It is actually identical to intronic sequence that is in the PPEF1 gene just downstream It's about 230 KB downstream And so this is a chimeric l1 insertion this type of thing has been described once or twice before But it's a very complicated complex insertion again something that we could Possibly see with short read sequencing, but we would have no idea what's in the middle of that whole insertion sequence So when we go back to this image this might be a little bit clearer But it's still it's still hard to see this just from looking at IGV for the first time so again in the middle we have that target site duplication in the In the yellow box and if you look at the very top at the depth track you can You can see a bump here where the depth is increased in that target site duplication so you can kind of see that duplication So in the top we have the proband CCS reads and then mom and dad CCS reads I've only shown a subset for mom and dad But their reads align well throughout their line well to the reference throughout this whole region But the proband has something going on there so if you look at the top reads you see reads that align to the reference of the left go through the target site duplication and Then have unaligned regions on there on the reads which Which indicate that insertion and those diamonds are hard clip sequence that just got cut off there as well so that the top part represents the left end of the insertion the Next part down represents the right end of the insertion where reads map to reference Again, go through the target site duplication and then show the insertion So altogether we know that this is one event that is that insertion of about 7 KB So we we wanted to know too is this this this variant looks de novo compared to the parents and so we wanted to know if it was on the maternal or paternal a little allele because often with structural variants and SNVs de novos are on the paternal allele and so here's a zoomed out view of proband dad and mom's reads With some SNVs surrounding that insertion, which is at the black arrow. And so I flagged a couple of Paternal SNVs with blue triangles and the one maternal SNV with a pink arrow and so at the top for the proband you can see that those reads clearly split into both paternal reads and maternal reads and The insertion you can only see on the paternal reads So it is so that insertion de novo insertion is on the paternal allele However, you also notice that there are some paternal reads that do not have the insertion and so when we looked at both sides of this at snips on both sides of this insertion, we actually found that Only about half of the paternal reads had the insertion. So it appears to be mosaic in this individual So what so they have an insertion in an intron Of a disease gene the the interesting part here that is unique is that we have that exon 3 that's duplicated So we wanted to know if that's duplicated in the genomic sequence. Does it actually get spliced into the RNA? And so we performed our TPCR with RNA from the proband dad and mom and We amplified from exon 2 to exon 4 and look to see what size those bands were And so in the bottom right corner, you can see that the proband even though it's very faint There is an asterisk there that there were two bands for the proband and we We isolated each of those bands and sequenced and we did find the sequence as expected. So that upper band did did represent an RNA with two exon threes spliced in and That is predicted to lead to a frame shift and premature stop so It does look like a loss of function allele Which makes sense for this proband and her phenotype So again for this proband to summarize this probands findings She had a 7kb insertion in CDKL5 that was L1 mediated. It was a chimeric very unique insertion de novo and mosaic on the paternal allele Again, it's likely loss of function due to the inclusion of that second exon 3 and based on all of this We consider this to be likely pathogenic So this is one of the cases where it was really clear and obvious Long resequencing really found that answer for us. And so I have one other interesting case to To show you this as well. So Proband 4 in this initial pilot study had several large de novo structural variants So with long read sequencing we identified at least 15 De novo breakpoints that affected six protein coding genes across three chromosomes And so I will walk you through each of these And as Emily briefly mentioned we may we also make de novo assemblies for each of these probands So when we do our long read sequencing We can make a de novo assembly for each of those probands. And so on the left Along the left side you will see four contigs listed. And so those are this probands paternal contigs that together represent chromosome 6 So there are four contigs that span the entire thing And if we did not if we were not able to make that de novo assembly There is no way we would have figured all of this out So on the left side again We have the probands paternal contigs along the left side and reference along the top And so when we compare those two if there was no difference and no large structural variance there You would see a diagonal line from the top left to the bottom right But you can see in the middle that part of that is inverted In addition to that up in this top right hand corner there are quite a few other lines that are not behaving like we would expect them to and so that is Blown up on the right hand side. And so again, we have the probands contigs Going down on the left and reference across the top and so you can see the pericentric inversion breakpoints here But then in addition to that there are at least eight other fragments where if you look at the sequence directly It looks like they are chopped up in random places and just thrown back together in the wrong order and orientation Which is indicative of chromothripsis So in addition so just as a easier way to look at this so for this proband she had a maternal chromosome 6 that looks wild type in the pink box there but her paternal chromosome 6 in the blue box has a pericentric inversion and Then the region where it should be a through H is in a jumbled mess With what's shown there? So in addition to this this proband also had insertional translocations that were also de novo on chromosome 7 and 9 So again on the left side are contigs from this proband Aligned against chromosome 7 and 9 across the top and so if you just look at the top Contig 57 that is the probands chromosome 7 sequence aligns well to chromosome 7 and then it aligns a little bit to 9 and then comes back to 7 and Then below that the opposite happens. So for the program probands chromosome 9 it aligns well to reference chromosome 9 Switches to 7 and then comes back to 9 and of course with a little bit of an inversion at one end because it can't Be that simple So again just a different way to picture this There are multiple breakpoints on 7 and 9 and in this proband she had Wild type chromosome 7 and 9 from her mom But on the dad's allele we had that piece of 9 going into 7 piece of 7 going into 9 Again all of these together We never could have figured out just with short read sequencing The being able to make de novo assemblies allowed us to really clarify clarify all of this And so again, she has extent and all of these breakpoints are de novo Again, she has de novo structural variation with at least 15 breakpoints affecting six protein coding genes None of the protein coding genes are actually associated with disease As of yet But the combination of all of these large structural variants we have considered to be likely pathogenic Okay, so those are out of our first six that we did those were the two that very obviously were solved with long reads that Really benefited from that technology Since then we have sequenced another 46 probands along with some parents and so we have about 52 total And again, so those first two were ones that were very obvious Very not necessarily easy to understand or resolve, but they were clearly resolved with long read sequencing But in that and in the ones that we have done since we've had complications and challenges There are just there are challenges that still remain So I'm going to try and do one more one more poll here, and I think it will show up on the next slide, okay How many long read analysis tools can you name oh? And they show up in different orders, okay, I Think yes, yes, that makes sense Please someone say so many, but I'm working on a new one and please come and talk to me afterwards Oh awesome Someone someone is working on a new one Great okay, so a lot of you Yeah with zero and zero to five so lots of five or less, so that's awesome. That is perfectly fine So what I wanted to point out is that there are so many tools Even so these are the tools that we have used so far and Most of our pipeline. This is not exhaustive But these are the ones that we have sort of worked through as we've been working on this in the last few years So we have multiple aligners Multiple things that we've used for SNV calling or SV calling or repeat expansion calling different tools for visualization and analysis of those variants, whether they're structural variants or SNVs, some QC, different assemblers that we've even used, and some different tools, excuse me, for genotyping comparisons, especially between long-read data and short-read data. And some of these, so we've mostly settled on some of these, but we've tried a lot because there are a lot. And don't forget, once you've done all of this with HG38, you might want to also try it with CHM13, the T-to-T assembly, so double all of what you've done. So while all of these are being, and all of these have multiple versions, all of them are continuing to be developed, which is wonderful because they're getting better as we continue. But there are just a lot to keep up with. And I just thought this was interesting, the longreadtools.org. They have a catalog apparently of over 587 tools for long-read sequencing. This is across all different platforms. But this just gives you an idea of how much is out there. So it's quickly, it's a new field, and it's all still moving very quickly, which is, like I said, which is great because we're getting better tools, but it's also a lot to keep up with. And we have not all settled on exactly how things should be done necessarily, which is not a bad thing. So I wanted to show you an example of the implications of this. So the CDKL5 variant that we found, we found pretty early on. And so as we've used different tools, different updates, we've been able to use that one as a good positive control. And so in one version of PBSB, which is the structural variant color that we use, this was originally called as two reciprocal break-ins. And so this was called as a translocation, which it isn't, but it was enough to alert us that something's going on here. And that's what it was called as the first time that we found it and figured that out. With an updated version, and I think also there was a difference in alignment a little bit, this was next called as that same event was called as a duplication, which it sort of is because there's that target site duplication in there. And it was also called as insertion of an L1 sequence, which again, it kind of is. So this isn't necessarily the best example because it's not just one clear structural variant either. There is a duplication in there and there is insertion of an L1, but it's also a little more complicated than that. But again, just something to keep in mind as we're analyzing all of these, that things can change with different versions. Okay. One other thing I wanted to mention, which again, Emily mentioned as well, that as we're finding new things, we don't know the true allele frequencies of these novel structural variants that we're finding. And so again, with CDKL5, we thought, oh, of course this is novel, surely this is rare, but how do we really know? So we ended up having to just go back and look at all of the long-range sequencing that we've done to compare. Other than that, there's no database to compare. And so we're approaching this, I think like most other people are, in amassing our own internal database. And that's what we've been doing. Again when you have a structural variant database, there's always some complexities in how to merge calls across multiple individuals, different ways to do that, and different complications with those tools. So we are doing that, however, but we've also thought about this in terms of an element affecting structural variant database. And the way that we have thought about this is that we have collapsed transcripts for a gene and then we identify elements or genes that are affected by structural variants that are called in our data. And so that element might be the gene itself. It might be exon 2. It might be intron 4 of a certain gene, things like that. But that way we can look for an element that is affected by a structural variant in only one individual and that lets us find rare structural variants. One other thing that I just wanted to mention that we've been able to do because of that is we've been able to look at that element affecting database and compare the frequency of gene affecting structural variants to ARVIS. So ARVIS is a SNV-based intolerance score of genes. And when we look at that in terms of structural variants, there is a correlation. So as expected, genes that are intolerant to SNV variation are also intolerant to structural variation that affects exons in that gene. Okay. Since our session is all about diagnostic yield and ending that diagnostic odyssey, I wanted to talk about a few different sources in long-read sequencing that are contributing to changes in diagnostic yield. And so I have two more examples to show you. This is a proband that we sequenced with long-read sequencing. We sequenced just the proband, but not the parents. But we did have short-read sequencing data on the whole trio. And so in this proband, the boy in the middle, we found an SNV that caused a premature stop in CSNK2B. We also... So this variant was found in long-read data. It was also found in short-read data. It was also found in the father. It was paternally inherited. And it probably explains some of the phenotypes of the father and this son, this proband who was sequenced. Unfortunately, the first time that this genome was analyzed with short-read data, CSNK2B was not a disease gene. Since that was done, it became one. And so this is a case where reanalysis is really important. We all know that. We all strive to do reanalysis as much as we can. But this will still continue to happen. There may be someone we do long-read sequencing on today who just had reanalysis yesterday and still in five years, they may end up finding an SNV in a gene that is a disease gene. One other interesting case that we had where we found a diagnosis is this one. So we, again, I think this one, we just sequenced the proband with long-read sequencing, but we had short-read on the whole trio. And so in CCS, in long-read sequencing, we found a one nucleotide deletion in SHNK3 that led to a frame shift. And this is a pathogenic SNV. It makes sense with the proband's phenotype. It was called in long-read data. It was not called in the short-read data for the proband, and it was not called in either one of the parents. And since, we have confirmed that it is de novo in this proband. What happened here is it looks like we just got a little unlucky. So we looked back at this region, and it does not appear to be a region that is systematically low covered, has systematic low coverage in short-read genomes. It looks like we just got a little unlucky in this particular case. So the mean coverage for the short-read genome for this proband was about 31X. The coverage here is about 13X. If you look at other probands, their mean coverage and their coverage here are about the same. So if we would have done short-read sequencing again, we might have gotten lucky and found it. We might not have. It's kind of hard to tell in this case. Okay. So like I said, we have at this point done long-read sequencing on 52 probands. And we have found interesting variation in about seven. So I have shown you the two pathogenic-likely pathogenic structural variants and the two pathogenic-likely pathogenic SNVs that we found. We also have three other VUSs that we're really interested in. These probably, at least two of them, lean likely path, but we just don't have the data to support that yet. And so we're hovering right around that 10% diagnostic yield. Despite this, or in addition to this, we are still investigating all sorts of interesting structural variation that we couldn't see before and single nucleotide variation. So in particular, in this 52, we do have seven rare single-hit pathogenic-likely pathogenic structural variants in recessive disease genes that have unrelated phenotypes. They don't necessarily mean anything for this proband. It's more of a carrier status. But they're very obvious, very clearly path-likely path, single exon or several exon deletions in recessive disease genes that are just easier to see, again, potentially shortening that time or showing evidence that we could shorten that time to diagnosis for other genes. And then there's a lot of other structural variants that we think are rare that still appear to be rare based on our own internal database that include things like retrotranspose pseudogenes or just retrotranspose transcripts that you can see the whole way through, but we couldn't see that with short-read sequencing. Or we've seen evidence of nuclear mitochondrial DNA where parts of mitochondrial sequence were inserted into the nuclear genome. And again, we can see over the length of that entire insertion. Okay. And I will close with this slide of the pros and cons. I think you will hear from all of us today about all of these pros and cons. It seems very obvious that all of the pros are very much pros. They are our improved SNV calls, improved structural variant calls. We're able to look at repeat expansions and clearly see those. And we have these phased assemblies that can really help in that really complex structural variation that we're seeing. However, it is still expensive to do long-read sequencing. And we do have that analysis burden of how do we find structural variant frequencies and which tools should we use at what time and when should we update and things like that. But we clearly in moving from short-read to long-read, we are not losing anything. We are only gaining things. So I think there's a lot of promise. So with that, I will stop and just acknowledge thanks to all the families who contributed and that there were a lot of people involved at Hudson Alpha to do this project. Thank you. All right. Again, we have time for one or two questions. If anyone has one, they'd like to come to the mic. We did have a couple online. One, I think you already answered, but the question was for the CDKL5 complex exon duplication. Had the patient had short-read? And I think the implication is, could you see it there? They did have short-read. We actually did short-read. We did see it, but we didn't know to look for it until it was there, right? Until we saw it in long-read. So when it was, when, and this is another kind of complicated thing with the structural variance. When we actually saw it, well, it was called as a duplication, a 230 KB duplication in short-read sequencing. When we looked at it, you can, if you just graph out that whole region, there's no clear increase in depth. It doesn't look like a duplication at all. So we just kind of discarded it and moved on. So, but it is, yes, it is visible there. Yeah, when you know what you're looking for. Exactly. Exactly. And then I guess the patient with the chromotherpsis, interesting patient. I have two questions. One came from online. I think it was referring to this patient. Could you see it on karyotype? I assume you could. So we tried to get the patient back to do that and have not been able to do that just yet. Got it. So they haven't had one. They haven't had, no, they haven't had one. And then, so all of those happened in dad's genome. Is there anything interesting about that, that family or about? Not that we know of. Not that we know of. We started asking ourselves all sorts of questions after that too. Nothing in particular that we know of. But yeah. All right. If there are no other questions right now, again, we'll have all speakers back up here after our next speaker. So thank you. Okay. So our next speaker is Dr. Danny Miller, who is just about to finish up his pediatric genetics combined residency program in Seattle. And I have to give him credit for actually putting together this session and bringing all the speakers together. So thank you, Danny. It's an honor to moderate, but I did not put this together, just so you all know. So he's going to talk to us today about using targeted long-range sequencing to find the missing variants. Okay. Thank you, everybody, for, well, first, I guess, for coming. It's really great to be here in person with everyone. And then, specifically, thank you to the folks on the, that are presenting today. It's been fun to kind of watch this come together, and I think we get a really nice variety of experiences here. So my name's Danny Miller, and I am a combined Peds and genetics resident at UW and Seattle Children's. And I also work in the lab of Evan Eichler in the Department of Genome Sciences. And then my disclosures are that I am a paid consultant for Myome, and that I have received travel support and have a research collaboration with Oxford Nanopore, and I'm going to talk about their technology today. So my interest, as many of you probably know, and as, you know, Heather, Susan, and Emily have kind of all said, the idea is, you know, we put patients through this diagnostic odyssey, right? We end up with one test. If it's negative, we go to another test, and et cetera. And at the end of that, we end up with basically a 50% diagnostic rate, which seems ridiculously low. So what I've always wanted to know is, could I use a single technology, in this case, long-read sequencing, to shorten that rate of diagnosis and also increase the number of diagnoses that we make? I used to say days to weeks, and I've changed that to hours to weeks lately because of, you know, the paper from Stanford that came out a few months ago showing the eight-hour genome, and I'm pretty sure we can get to a sub-four-hour genome with ONT technology today, honestly, and then see where that goes from there. And you know, I think long reads will have an impact really everywhere, right? We'll see it pick up in prenatal testing, looking at cell-free DNA, see if we can determine tissue of origin using methylation. Cancer genomics, I think that there's going to be a lot of potential applications there. You know, anything from, you know, you get blood tests before you visit your oncologist in the morning, and by the afternoon, you know if there's, you know, an interesting variant that's popped up in your cell-free DNA. And then where I think is really interesting, and I don't work in this field, but I enjoy watching it, is infectious disease. Kid comes into the EDU as a central line, you know, we throw vancomycin and really broad spectrum antibiotics on him, but do we need to? Could we draw some blood and figure out if they actually have MRSA, for example? Could we spare them that? Could we do that analysis in hours instead of waiting days for that result to come back? So I think there's a lot of interesting uses there. And also, of course, in viral surveillance, COVID, et cetera. So I just want to start with a really brief overview of what is long-read sequencing, the way I think about it. And really I think about it as a way to cheat, because it really simplifies the analysis of some of the complex regions of the genome that we look at. And a simple example is, imagine you have a region of the genome where you have a repeat, let's say five 1KB repeats. When you align with short reads, or when you sequence the short reads, you can uniquely align the reads here in orange on the edges, right? And you know that some reads are going to span the gaps between the repeats, but these green reads right here, they could go to any one of five places, right? You don't know where they fall. So you don't know if there's been an expansion or contraction of this repeater, for example, or if it's a transposable element, you know, what does the structure of it look like? Long-reads solve the problem by simply, in most cases, being longer than the length of the repeat that you're interested in, which makes it simpler to do the analysis. There's differences in the technology, of course, as most people know. So with short reads, the reads are anywhere from 100 to 300 base pairs long. They're very accurate. Quality scores are greater than 30, 99.9% per read accuracy. And it's inexpensive, right? It's artificially high, but it's going to come down over time, so it might cost you $1,000 or so to sequence a human to 30X or so coverage. Now long-reads are different. The reads could be anywhere from a KB to two, three, four megabases. I don't know what the record is on Twitter these days, but I think it's four and a half megabases or something for Nanopore. And the accuracy can be quite variable depending on the technology you use. So differences between PacBio and ONT, anywhere from Q10 to Q30, so 90, 99% read accuracy. And the costs are very different as well. So with Nanopore, that cost could be as low as $1,000 for a human to 40, 50, 60X coverage. But for PacBio, that could be in the $6,000, $7,000, $10,000 range or so. But these reads are really information-rich, right? There's a ton of data in the long-read sequencing data, and that's what I want to try to explore a little bit in my talk. And I know Winston and Andrew are going to get to that as well in their talks. There's really two commercial technologies out there, as everyone knows. I'm going to ignore the synthetic long-read stuff for now and hopefully for the rest of my career. But really, there's the platform provided by PacBio and the other one by Oxford Nanopore. I always got to make an aluminum dig, so it's just hard not to. As most people probably know, PacBio works by sequencing a circular piece of DNA. There's two modes, CLR, which can sequence molecules up to, I think, 80 or 90 KB, maybe a little bit higher. And then there's the CCS mode, which is what most people are familiar with, branded HiFi, of course. So you read that same DNA molecule multiple times. When you do that, any errors that are there, so these little red dots, for example, they get washed out because you've read the same molecule six, seven, eight times. So you get this really high-quality read. The data is beautiful. Susan and Emily both showed examples of it. But you're limited in read lengths up to, I think it's 18 or so KB is the longest fragments you can do today. Nanopore works a little bit differently. What you're doing is you're measuring changes in current as a single-stranded piece of DNA or RNA passes through a protein pore. These current changes are characteristic, so you can work backwards to figure out what molecule is in the pore at any given time. Now, what's nice is the pore doesn't care how long the DNA molecule is, right? This could be a two or three or four megabase long molecule, and the pore is just going to sequence it and then grab a new one and sequence it. The downside is that it's challenging to work backwards from the current to what nucleotides were actually in the pore. So this is a higher error rate technology at this point. And there are some tweaks I'll talk about here in a second. You can see the error rate when you look at the data. So here on top, I've got nanopore data. This is on an R9 flow cell called with Guppy 5 using the superior model, which is the newest base calling model. And on the bottom is HiFi data. I will say these are not from the same individual, but the polymorphisms are the same. So what you can see is that while the nanopore data is noisier, so you see the single nucleotide changes, right? The different colors here. You see the insertions, and you see the deletions, right? I would argue that the essential information is here. The individual's homozygous for this polymorphism, and they're also homozygous for this deletion right here. But the data just doesn't look as nice, right? And that's, I think, really been challenging for folks, especially on the clinical side for picking up this data and using it. Now nanopore data, the quality has improved over time. And what you can see here, this is data that the same data set in blue, if you will. This is base called with an older version of the base caller. And then the red here is called with the new superior model. And you can see that the quality's improved. I will say nanopore loves to talk about the modal error rate. I find the mean to be much more meaningful, no pun intended. But you can see that the mean moves from about 92% or so to about 95%, just with the change in the base caller, right? So you can take all your old data, you can rebase call it, and you get higher quality data, which is really nice. There are changes that are coming or that are available that improve the error rate. So the R9-POR is a POR in which there's a single reader that reads the molecules that passes through the POR. The R10-POR is a POR that has two readers. So you get a little bit of a different error profile with that. And I imagine maybe you can watch any of the public nanopore talks, and you can see all kinds of the fun things that they're doing with these PORs. So I think there's a lot of potential here. A lot of people have asked, there's a new sequencing mode called Q20 with nanopore. So this uses that dual-head, that dual-reader POR. And what I'm showing you here on top, so this is the same individual. And this was sequenced actually, we made the DNA library, did the sequencing in parallel to compare the R9 and the R10-PORs. And what you can see is, you know, on top, I guess at a macro level, you know, you see a lot of insertions, deletions here on the top. And there's fewer down here on the bottom. You still get these really wonky reads that come out of nanopore, which, like, what does that even do in there? But you can see that overall, the quality looks better. But the question is, does it dramatically improve your interpretation of the data, right? Does it uncover SNVs and indels that weren't there before? Now, we're not sure yet, because we're just starting to work with the data. But here's an example, so it's the same individual. The R9 data on top, the R10 data on the bottom, so Q20 data here on the bottom. If you look at this homopolymer, these five Gs in a row, you can see that in the R9 data, in a half or so of the reads, there's a deletion here, which isn't real, and it's gone in the R10 data. Now, I want you to keep this slide in mind for one of the examples I'm going to show you later of a case that we didn't solve, but figured out why later. And it has to do with homopolymer error rate. So you know, we wanted to know, how could we use long-read sequencing data to solve some of the challenging cases that we had that remained unsolved? And we wanted to know what cases would benefit from long-read sequencing, and could we use that data to really evaluate what the long-term uses of long-read sequencing might be? And one of the first things that we needed to figure out when we started the project was, you know, how do we do the sequencing given the high cost of long-read sequencing? That's one of the concerns that popped up, I think, at the beginning when Emily did the word cloud. What's the best method for isolating and preparing DNA for long-read sequencing? And then, how do you analyze the data, right? Because it's really challenging. It's a lot of data to work with, and you know, you need powerful computers to do it. So for us, when we actually started working on this, we were going to use the Cas9 approach. So there's a couple different ways that you could target specific regions of the genome for sequencing. The first is you could just PCR the regions that you're interested in, right? The challenge there being that PCR really only works for fragments that are 10, 15 or so kb long. And if you do that, you get really unreliable amplification of the target regions. But when you do that, you're going to lose epigenetic information, right? So that wasn't ideal. The next approach we thought about was the Cas9 approach, which I know Winston's going to talk about. And I think this figure is from your paper, I believe. Yep. Okay. And it's a really beautiful approach, I think. I will be honest, I've never gotten it to work really well in my hands. I don't have the best lab hands, though. But you know, the challenge here for us is I think it works really well when you're going to target the same region over and over again. But for us, there was a different region we were going to target in every case that we had. So it kind of became not practical to use this approach. So in this case, I actually, you know, it's always better to be lucky than good. In this case, we got really lucky. When we were going to start our project, that's when the Reedfish preprint from Matt Luce's group came out, which is what was built into Minnow eventually and is now known as adaptive sampling. And what adaptive sampling lets you do is it lets you computationally target specific regions for sequencing. So in this, oops. So in this video, which no longer wants to play. So in this video, what you're going to see is imagine there's a single region of the genome that's denoted by the yellow squiggles, and that's the region you want to target. So as the molecule goes through the pore, it's read in real time and base called in real time. You can then align it to the genome. So after about 400, 500 base pairs, you align it and you say, is this molecule in the region of the genome that I'm interested in? If not, you reverse the current on the pore, it kicks the molecule out, and then it grabs a new one for sequencing. So in this case, the second one, this is in the region that you're interested in. So it will continue sequencing the molecule. This is a bit of a funny video. It's older. It shows both strands of DNA being sequenced. That's not a thing right now, but I think it will be again soon. But the idea is that by rejecting molecules before you sequence them or, sorry, rejecting molecules shortly after you grab them, you increase your coverage because you're more likely to grab DNA molecules in the region that you're interested in. So we tried this. We used Reedfish. And the first time we tried it, it worked really well. This is data actually from my very first Reedfish experiment. We find that it works best when you shear the DNA to about 10 or so KB. We can target up to about 150 or so megabases of the genome on a single flow cell, although if you target less, you get better coverage. The sweet spot's probably around 10, 20 megabases of the human genome. That target, that sweet spot is really a function of how large the genome is you're working with. We get about 20 to 40X coverage of those target regions using a single release 941 flow cell. And that's a $500 flow cell, and then the library's about $100. So $600 or so in materials cost. And then you run it through a pipeline where you align it, you call variants, and then you do your analysis. And you can just see for these two examples here, the coverage is in gray in our target regions are in blue, and we get increased coverage over those regions that we targeted. So really nice result, and really flexible. Regarding DNA extraction, we've done some optimization, and it's probably of no surprise to anybody. It works best when you get DNA from blood or fibroblasts, especially if you use a high molecular weight extraction kit. Things like, I don't think you need to do phenylchloroform. I think something like a Monarch kit is good. We have less good results when we do saliva or buccal swabs. They seem to work best if we do size selection after we isolate DNA from those. They do work, and we've solved cases from those. And then of course preserved tissue is not going to give you great DNA fragment length. We do a lot of QC. I think it's important. I don't know how you transition that to the lab just because it takes time, but we do do a femto run for every sample to look at our fragment distribution, and then we do Qubit for DNA concentrations. And then shearing and size selection for targeted, again, we shear it to about 10 KB using a G-tube. And then for whole genome, we'll typically do size selection depending on what our DNA looks like going in. And what do I mean by that? Well, this is a femto run for two different libraries, or for two different DNA extractions. And just to show you, so our average length is somewhere around like 25 KB or so for this one. And what you see is under 10 KB, there's a big shoulder here. And what's going to happen when you sequence this is you're actually going to pick up mostly molecules from under this 10 KB window. So we would size select this into something that looks a bit more like this, where that 10 KB cutoff, where that area under this part of the curve is much lower. And we get much better results when we do whole genome sequencing using a profile like this. So as Susan pointed out, and I 100% agree with, the pipelines and the tools to analyze any long-range sequencing data are really complicated. And they're kind of like ever-changing, which is really tough to keep up with. From the ONT perspective, the first thing you have to do after you get your data is you have to base call it. You can base call on the machine, but if you do something like whole genome sequencing on a Promethean, the Promethean can't keep up with the output of the machine, so it has to be base called off the machine. That's a GPU-intensive process, and that can take days to run, depending on how many GPUs you can throw at it. It depends on the model you use, things like that. Alignment, of course, is going to be a CPU-intensive process. There's options. We mostly use Minimap. You can use NGMLR. There's probably other options out there that I just don't know off the top of my head. And then you want to call SNVs and indels. This changes over time. It used to be that Madoka was the SNV caller of choice, for example, and now it's either Clare or Deep Variant, but they both have drawbacks, depending on how long you want to spend letting them run. And then indels can be challenging to call, of course. Structural variants, right now we use Sniffle, Svim, and QtSV. We use all three because all three of them miss things, so even when you intersect that, there's still going to be things that you miss in this data. And I think the biggest challenge with structural variants is we don't have population-level structural variant data, so all of us, we look at the 20, 30, 40 cases that we've sequenced to compare our dataset to, but what does our SV look like in a population of 1,000 or 2,000 individuals? So we really need to do something like sequence the 1,000 genomes dataset using both Nanopore and PacBio, but we're just not there yet. And then methylation. So we have found that with ONT data, we like using megalanon to call methylation data. I think eventually we'll use Remora whenever it's built into Guppy, which is the base caller. We used to use Nanopolish. We don't use it anymore, and that's because we found it doesn't convert every read. We get much better read conversion with megalanon. The big downside is that megalanon might take three or four days to run for a whole human genome, whereas Nanopolish would run within a day or so, so it was much quicker. So you just have to decide what trade-offs you're willing to make. And again, we come back to we do need population-level data for methylation, so if anyone is interested in supporting sequencing of the 1,000 genomes dataset, please let me know. So what are some clinical cases that we've done targeted long-read sequencing of? I want to cover these quickly because I want to talk about some new stuff that we haven't published, both of targeted sequencing and whole genome sequencing. And this is a case that we reported in our AGHG paper. I think just a really nice, straightforward proof of concept. We had this kiddo with campyloid dysplasia. We knew he had a translocation between chromosomes 12 and 17. We just didn't know where they were or how close they were to SOX9, which is a gene that's associated with that. And we targeted about 70, 80 megabases, I think. We basically targeted the bands that were called out on the karyotype. And we got really nice coverage and we were able to see that the translocation breakpoint was a couple hundred KB downstream of SOX9, which is a well-described mechanism of disease for campyloid dysplasia. So very straightforward, right? We knew where to look, we targeted a large region, and we found the results. This is a case of a kiddo who we knew actually had Koffin-Syre syndrome because he had an array that told us that. But we suspected there might be more going on just because of how complex the array called out of three different deletions. And we were wondering if there was really more going on. So just really briefly, we targeted about a 15 megabase region, so fairly small. We got about 20X or so coverage, so not a great experiment, but still was enough to tell us that the three deletions that were identified on the array, we saw those. We also saw two additional deletions, which are D and L here. This was in a transposon-rich region that just had like one probe, I think, on the array. And then D didn't have any probes just by chance. But what was really nice is, what we find is there's actually a lot more going on here than just the deletions. There was a lot of rearrangements as well. We call this a subway plot. It's how we kind of make it easier to put the order of the segments back together. And what it lets you do is you just go from A, you can see it goes to E, and then E to C, and this tells you that C is inverted, and then C to I. And then things like this H inversion, you never would have seen this with the array, right, because there was no deletion associated with it. It was just a straight-up inversion. So what we can do, oh, I didn't include it, I'm sorry, is it then lets us reconstruct the region and then see the order of the genes. So we can start to think about things like, is there a TAD there that might have been disrupted by this rearrangement, things like that. So a lot of additional data here, as well as we can actually look at the sequence of the genes in these deleted regions and ask, was there any pathogenic variants in here that are now uncovered as a result of the deletions? And then we had cases that were true missing variant cases. So these were cases where we knew the individual had a phenotype associated with a particular disease, and then typical clinical testing had identified variants in those genes, but didn't find a second hit. And just as a simple first example, this was a kiddo who had nephron ophthysis, had an exome, revealed a single stop variant in NPHP4, my mouse is gone, and a single stop in NPHP4, but no second variant. So we did targeted sequencing, we found this intronic variant, we phased it, so we were able to not only see the two variants, but also phase them into haplotypes. We predicted it would affect splicing, and we confirmed it with qPCR in this individual. For B here, this was an individual who had all the features of Hermansky-Publiac, but no second hit. So we knew she had this paternally inherited stop, but we couldn't find a second hit after both a microarray and after an exome. We did targeted long reads and found that she had a 1900 or so base period deletion of exon three, and then clinically confirmed it with the exon level array. I always like this story, a kiddo with Ahlstrom syndrome, she had a paternally inherited stop, apparently there's a thing about paternally inherited stops at our institution, who had a, her second hit ended up being an alloinsertion in exon 20. She had had an array that didn't identify this insertion, but once we told them about it, the clinical testing company, they went back and looked and said, yeah, it looks real. And then they verified it, I think, with Sanger in PCR. And then this individual, young man had all the features of Lesch-Nyhan syndrome, including the biochemical features, but no second hit, or no, initial hit, I'm sorry. We did target a long read, and we found this is 187 base period deletion. And the way the reads were oriented, it looked like it actually was a big inversion. We had a karyotype on him. We went and looked at the karyotype for a long time, we couldn't see the inversion. So we designed fish probes to prove that it was indeed real. So, you know, a nice, kind of showing how, even though, you know, I always think karyotype's the one thing that long reads won't completely replace, but, you know, even then we couldn't see it with, see the inversion with the karyotype. Okay, now I wanna get into some newer stuff, and just kind of new ways that we're thinking about using this data. So segmental duplications are challenging to sequence with short reads, right? Because they exist in multiple locations in the genome, and short reads are gonna align with some similarity to those regions. And this is a good example of how long reads can help clarify questions when you have a variant in those regions. This is a kiddo I saw clinically in our NICU. She had been transferred for ECMO because she had respiratory failure at birth. We did a duo exome, the father wasn't available, and that identified a likely pathogenic two base pair deletion in a gene called Hayden. Hayden is a gene that when you inherit, when you have a recessive, when we have biallelic variants in the gene, you have a ciliopathy, a primary ciliary dyskinesia, and that fits really well for someone who has respiratory failure, right? Your cilia aren't working, you can't clear out stuff from your lungs, so maybe you have respiratory failure from that. So we thought it was really interesting that this was found, but the problem is when you look at Hayden, it's about a 400 kb or so gene, and it contains a 380 kb segmental duplication. So that two base pair deletion was somewhere over here in one of these exomes. So really, this whole region here, there probably wasn't a lot of very good visibility from the clinical testing company. And indeed, if you look at short read data, aligned to Hayden, this is just a random region of that segmental duplication. Short read data from NA12878, you can see that over these two exomes right here, you don't get very much coverage at all. So if there was a pathogenic variant in that region, you wouldn't see it, right? So what we wanted to know is could we use long reads to look for a second hit in this particular case to figure out if the patient actually had a ciliopathy or not. And the long read data here is on the top, and the short read data is here on the bottom. And with the long read data, you can see that you get nice, even coverage of the gene. And what I really like is you can see the individual actually has a heterozygous variant in this exon that really would have been challenging to recall with short read data. So in this particular case, actually, we didn't find a second variant in this gene that would explain the phenotype, and we actually sent the cells to a research group in Germany without telling them what our result was, and they said that the individual does not have a phenotype consistent with Hayden. So the two pieces really aligned quite nicely. Another example here, this is actually a paper that we just submitted in collaboration with Harold Junepepper at Mass General. This is a example of how we can use long reads to identify both structural variants and differences in methylation in really challenging regions of the genome. And I'm gonna be 100% honest, I am not gonna try to totally explain the GNAS region or the phenotype associated with PHP1B and all those things. I just wanna point out that on the paternal haplotype, you have different methylation patterns than you do on the maternal haplotype, that affects which transcripts you make, especially in your kidney. And the one that I wanna focus on is at the AB region right here, this AB exon. The paternal haplotype, it's gonna be unmethylated, but the maternal haplotype should be methylated. And Harold had a case where they knew the individual had PHP1B, but they couldn't find the, and they knew they had a difference in methylation at this AB locus, but they couldn't actually find the variant that caused that. So we did targeted long read sequencing. And this is that region, and what we found was a 1,900 or so base pair insertion. This ended up being an SVA. And then we also, I'm sorry, we ended up finding 1,900 base pair insertion, which ended up being an SVA. And then what we can do is we can look at the methylation data simultaneously. And we can see in this individual, so here blue is unmethylated and red is methylated. And we can see at the AB locus, where in our controls, they're heterozygous, so they have half, half the reads are methylated, and half are unmethylated. In our individual here, we have complete, there's no methylation. Now, oddly enough, there was a group actually who, similar to the story Heather described, had a case like this, and they had spent years trying to find a second variant. They ended up finding a transposable element insertion somewhere nearby where ours is that had the exact same phenotypes. We actually submitted the papers together. But we did this in about a week, and it took them years to go through that process. Now, the thing about nanopore data especially is that it is more error-prone than the HiFi data. And you need to think about that when you're doing variant calling. And this is always a hard story to, it's always hard to tell a story about when you were wrong, but I'm gonna tell you a story about when I was wrong and just published it in the last year. So this is an individual from our 2021 paper. This is a young boy who has a phenotype of Duchenne muscular dystrophy, but had no variant identified after clinical testing. So we sequenced him with long reads, and we didn't find anything in the coding regions. We didn't find any splice variants, but we did find this 300 base pair GAA expansion that we thought was very interesting. We weren't sure what it was doing, but you can see it here. So it's in an intron. You can see it's in this GAA-rich region. And you can see that there are these, there's the insertions here, right? These 300 or so base pair insertions or expansions. So we weren't sure what to make of this because we didn't have population level structural variant data, right? We didn't have other nanopore cases that compare this to. So we went to the Simons collection, about 9,000 or so genomes. We used Expansion Hunter, and we asked how many times do you see an expansion of this repeat in this dataset? And we were pretty shocked with what we saw. So in this dataset, again, about 9,000 or so genomes, we found that, you know, our proband had about 107 or so copies. We found 72 instances where that repeat was expanded. 71 of those 72 were in females, which is pretty striking, right? That's a huge violation of Hardy-Weinberg. So we thought this was a really interesting variant. We correctly called it a variant of uncertain significance. As genetics goes, it turns out our proband had a cousin that he didn't know who developed symptoms of Duchenne muscular dystrophy and got clinical testing. And that clinical testing identified a pathogenic single nucleotide insertion in a coding exon that we didn't see in the proband and that wasn't identified on clinical testing initially in our proband. So what did we miss and how did we miss it? So I'm showing you the data from our proband, and I'm showing you the position of the insertion. So this is one of the DMD exons. You can see there's five Cs right here. And that insertion is an extra C right here. Now, when we wrote this paper, we used Guppy361 and we used what was called the high-accuracy model. And at this position, you can see that some of the reads have an insertion, so this I, whereas others have a deletion, right? And when you use the newer version of Guppy, the superior model, with the superior model, you can see, again, there's some reads with deletions and others with insertions, but it's not a slam dunk, right? Would anyone look at this and say, well, there's clearly an insertion of a C at this position? And Medaka and Clare, they didn't call it either. So we missed it. But why did the clinical lab miss it initially? When he got testing four or five years ago? Well, the reason was they used ion-torrent sequencing when they sequenced him, which has essentially the same error profile as, I think Winston was laughing at me, with me. Yeah, laughing with me. Yes, yes, yes, thank you. They used ion-torrent sequencing, which I think is interesting to spend just a minute on. So the way ion-torrent works is you take a single-stranded piece of DNA and you add nucleotides one at a time. What you're measuring is the change in pH as one of those nucleotides is added. So it's really easy to detect a change of one molecule or two or three hydrogens, right? But the problem comes when you're trying to detect when you add five or six molecules at the same time, how sensitive is measuring the change between five hydrogens and six hydrogens? It gets harder as that expansion gets longer, as that homopolymer gets longer. You know, and that's kind of what we ran into here, right? It gets harder to detect that change as the homopolymer gets longer. Now, what I have not done, but I do plan to do, is I haven't run this individual on Q20. I do suspect that we will see the insertion with Q20, right? But what do you do with seven or eight nucleotides, right? A homopolymer run of that, like how much do you trust that? So it's just something that you have to keep in mind when you're working with this dataset, right? Okay, in my last four minutes, I just wanna talk about two examples of how we've thought about using whole-genome long-read sequencing to simplify clinical testing. And the first example I'm gonna give is from some work we've been doing with Bloodworks Northwest, with Jill Johnson and her group there. And we've been looking at unsolved cases of hemophilia. We first used targeted long-read sequencing, and we found a lot of interesting results that way. But Jill called me up one day, and she said, hey, I've got a female who has hemophilia A, and no one can figure out why. Are you interested in it? And I said, what time can I come pick up the DNA? Because what I wanna know, this is like the ideal experiment for me, right? Because with the two-hit model, if a female who has hemophilia A, and you can't find a variant, then maybe there's something wrong with methylation in that case. So we wanted to know if we could use long-read sequencing to figure out why she was affected. This is a schematic of the factor VIII gene for hemophilia A. What's interesting about it is there's two segmental duplications. There's a smaller, 1KB or so, segmental duplication in intron one. And then there's a large 10KB or so duplication in intron 22. And you do get a lot of rearrangements between these. So you get individuals who have inversions that lead to hemophilia. So I actually suspected what they did was they missed an inversion within intron 22 with the workup they had done for this individual. So to make a really long story short, we sequenced her with whole genome long-read on the nanobore platform. We got about 30 or so X coverage. This was one of the earlier cases we did, so we weren't really great at making the libraries, getting data, et cetera, at that point. But what I'm showing you here, and this is actually phased data, and we phased it, we called methylation with nanopolish. And I'll show you some of the reads that don't get converted. But essentially what I've done is I focused in on a 23KB or so region of the genome. And what you can see within the segmental duplication, we do get nice coverage of long reads over this region, which is not what you get with short reads. So again, data from NA12878, we do lose a lot of coverage here because the reads just don't align within the segmental duplication. And that sits on top of the CPG island, by the way, which is useful in a case where you have a female who you suspect has skewed X inactivation. So we did our typical analysis pipeline, and we found a pathogenic, T to C, that affects the splice site, which was the known variant. I didn't know it at the time, but it was the same variant that they knew about. But they didn't know what the second variant was. And it turns out actually the second variant is she has completely skewed inactivation of her X chromosome. So again, this is nanopolish converted data. And when it's red, it's methylated. And when it's blue, it's unmethylated. And you can see that she has completely unmethylated the same haplotype that that pathogenic variant lives on. But the one that has a working copy, if you will, is completely methylated. We were able to de novo assemble this using fly, actually. And we got about a one megabase or so de novo assembly of the region surrounding factor VIII. We had RNA-seq data. And there were eight or nine SNPs in the RNA-seq data, all of which mapped to haplotype two. So we have orthogonal confirmation that she has skewed X inactivation. So I think a nice example of how you can use long read data as a single source to really answer challenging cases or to solve challenging cases like this. Okay, and then I just wanna end on something we're working on now that I'm pretty excited about. And that's using long read sequencing to simplify testing for imprinting disorders, things like Prader-Willi and Angelman. You know, it can be expensive, time-assuming, and it's kind of a workflow that you need to go through when you evaluate these cases. So what we wanna know is, could we use whole genome long read sequencing as a single test for these imprinting disorders? I'm just gonna show you one example. This is a individual known to have Prader-Willi syndrome. And what you can see is, we can map the coverage of this individual, and we can see that they actually have a dip in coverage within this critical region, right? So there must be a deletion there. We can look and see, we can look at polymorphisms, and we can see that they're homozygous for polymorphisms kind of across, as you would expect, but they lose heterozygous polymorphisms here. So a nice check that there is actually a deletion here. We have no heterozygous polymorphisms. What's nice is that we can go a step further with the same data set, and we can look at what the methylation profile looks like. And this is just for selected CPG sites within the region. We look at all of them, but we find the informative ones are, there's a CPG island within SNRF and SNRPN, which in our individual is completely methylated, but in our control data set is about 50% methylated, which is what you expect for Prader-Willi. So again, as a single data source, we can see the deletion, we can confirm that with polymorphisms, and then we can look at the methylation profile as well. So everything with one test is really where we wanna get to, and I think we're gonna build a nice cohort of these individuals that we can use as that example. This is a tool that we're gonna release, so I was one of the people that answered Susan's question as, I can name a lot, and we have a new one coming. So we'll have a tool that can actually identify differential methylation in nanopore data like this. Okay, so in summary, I do think long-range sequencing, as everyone here knows, that's why you're here, can be used to solve challenging cases and simplify clinical testing as a single data source, I think. I think that targeted long-read on the Nanopore platform is a really effective method for analyzing specific regions that you're interested in, but you do need to think about the error rate, right? So the DMD case that I showed you. But I do think that in certain cases, things like Cas9 would be useful. If you're doing something, I'd think the same region over and over again. These computational pipelines are very complex, they change, so that's something you have to think about as you're setting this up in your own lab or in your own, however you wanna work with this data. And then we really do need the population level data to interpret things like structural variance, right? How do we filter out those structural variance? That's gonna be key, so. With that, I'll say thank you for your attention. And I, of course, wanna thank my funding source, which is the Brotman Beatty Institute at the University of Washington. And then Evan Eichler and everyone in the lab, especially Miranda, who's a technician in the lab who I work with, who's been really fantastic. And yeah, with that, I will take any questions. Or maybe we should, I'm over time, so we should probably just go to the panel, I think. speakers from this morning come up, that would be great. So we have about 15 minutes before the break. We can always break early if we need to, but this is a great time to ask those burning questions, generate some discussion based on what you've heard so far today. And I can start with a couple from online. So I think Danny touched on it a little bit, and thank you for doing that, but there are some questions about cost, so the cost of probably whole genome sequencing with long reads versus short reads. Does somebody wanna just quickly address that? It's significantly more. It's more. And the other thing that we didn't talk about is that the throughput is not the same. So it takes a lot longer to run a long read genome on the HiFi. Anyway, Oxford Nanopore can push and go faster, as Danny talked about in publish, but they are not equivalent to short read. There's just no way around it right now. And so I think a related question. Someone recognized, Susan, that in your trios, you had higher coverage in the per band compared to the parents, and there's a question about was that really primarily related to cost or was there another reason? And then secondary to that was have you thought about combining higher coverage short read with lower coverage long read to kind of even out the cost or decrease the cost? So, and anyone can answer that second question. Yes, sorry, I was trying to go back to the first question already. Yeah, we definitely sequence parents at a lower coverage to save money. And like you said, we do have short read sequencing on almost all of these probands and their parents when available. And so I think there's definitely, we were talking about this earlier, there's definitely a balance between maybe doing long read on the parents and since we already have short read, or sorry, long read on the probands and since we already have short read on the parents, can we use that to our advantage? Because a lot of times for the structural variants that we found other than repeat expansions, you can see them, there's evidence of them once you know what you're looking for, right? So can we go back and genotype parents with short read? So we're definitely considering that. Does anybody have experience, I think there may have been a couple of examples given, but there's a question about experience detecting mosaic structural or other variants using long read sequencing. And again, I think this is probably a little bit related to cost, but you guys can address that. It's really challenging, right? Especially if you think your variant's gonna be present in 10% of your reads, for Nanopore at least, that's close to your, you know, is that error or is that a true variant that you're seeing? Thank you. So I think, yeah, for mosaics, it's gonna be really challenging, unless you do something like selection with Cas9 or adaptive sampling where, you know, you can pick up thousands of reads of the region that you suspect something in. But just to point out, that's not different than when we're sequencing with short read genomes, because we're typically doing those from 30 to 40X coverage. So your limit of detection from mosaicism is always tied to your coverage, really regardless of what platform you're using. And, you know, for us, that's where exome sequencing really still has a large role. It's efficient, we can do a lot, you can get really deep coverage to try to get at those mosaics. Thank you for some really interesting talks. I'm Stephen Wicks from the University of Colorado. I had a few questions, so... Start with the first, we'll start there. One was that these are all kind of pilots in research development. What's your pipeline for clinical validation? Is that something you're considering and how are you going about that? Great question. I'll start by saying we are completely research. Some of the, at Hudson Alpha, the CSL is starting to work through some of that process, but everything that we find, we do clinically validate through those methods, but all of ours is research right now. So we're sitting in the research realm too, but our lab does clinical testing and research testing, and I very much personally have a goal of making at least a clinical genome available now. We've really, as I mentioned, we've standardized our prep, because it was changing so often that there was no way we could get to clinical, and we've been on the same prep for several months now, which is why we were able to incorporate the automation, which is also really big for us. We've standardized, for the most part, I'd say a lot of the algorithms we're using, we've kind of settled on, although it's been really recently that we've been able to turn on the methylation detection from HiFi, so it does still change for us. I think the final piece that is difficult is finding, if you look at all your short-read exomes and genomes, and if you look at the emails you've been getting, there's numerous software solutions that help you pull all that data together, and that's definitely a component that we're working on, that there's not something off the shelf that will give you all of those answers, and so that's one part we're looking at, but we're definitely marching that way ourselves to hopefully get there. Yeah, everything we've done is in the research space as well, but certainly with an eye to what can we do clinically to reduce costs, to shorten the amount of time for testing, things like that. Is there anyone offering clinical long-read genomes? Do you know? So, I mean, others may know, but I've seen reports from Invitae that for FMR1 testing, where they validate the length of the repeat with PacBio, but that's the only one I know of off the top of my head. Okay. Yeah, go for it. You're generating all this data, like mapping and alignment. Are you making that available? What's your plans for that data? So ours is available. We recently just had our publication accepted, or it's in print now, finally. And so we do have a mechanism where our research data can be accessed through that. So there's a couple steps to go through, just to make sure. People are using it for the right reasons, but ours is available. Okay, great. Our first six, I think, are finally available. They've been submitted through dbGaP. Yes, I think it's still dbGaP. And we're working on the ones we publish, making them available. And the ones that we are sequencing, thinking about future. There is one interesting question here about, I guess, whether it's possible to, or you've thought about, multiplexing patients for sequencing. Is there any role for that in reducing costs? And I don't know the… It's probably hard on the bio side, just because, correct me if I'm wrong, a single HiFi flow cell will give you about 10x coverage of a human. So I don't know that there'd be a huge advantage in multiplexing. On the ONT side, we can squeeze 40 to 60x coverage or so out of a single flow cell. So you could think about multiplexing parents, for example, you know, 20, 30x coverage of each parent. You used to lose a lot with multiplexing with ONT. It's gotten a lot better, so you don't lose as much, so you'll get more of what you put in out, if you will. So we've certainly thought about that, but for now we're not multiplexing. The other thing about ONT, the washes are really nice, so you can run a flow cell for 48, 72 hours. You can run it for 24 hours, wash it, and then load a new sample without multiplexing it, and you basically, you get very little carryover between those washes. So that's another way to think about multiplexing. Okay, just a reminder to everybody to be close to your mic so the people online can hear. There's a question about using this for prenatal testing as it seems like the ideal case because of the rapidity. What are the challenges besides the usual interpretation issues given the limited clinical info in using this for prenatal testing? Who wants to take that one? Okay. So my first... Jay, counselor card. So I think when you mean rapidity that you can combine tests down into one instead of multiple tests, there is that advantage that you could in theory do. I think one limitation that immediately comes to mind is I'm not sure that the DNA quality that you would get from an AMU is going to be really conducive for HIFI sequencing currently both in the quality of DNA and the amount of DNA. And so if you then have to go culture your cells, you're no longer probably faster. I think anytime you're doing prenatal testing, we just have to be really mindful of what we're looking for and what we're returning and what we're not returning during the test. And we have to be very clear up front what we're considered with and have discussions both with their care team and make sure everyone's on the same page. I think if you could address those, then yes, you could do it. I think it's a very active area of research that, you know, I think it's quite a ways away from probably being clinically useful. So there's interesting things I think you can do there. There's a question here about DNA methylation signatures. So there's recently been numerous publications about disease-specific methylation signatures detected by methylation array. Question is related to, could you pick those up using long reads and methylation analysis? Yeah, you're talking about the EpiSign pipeline. We have an N of one that I've sent to them and they correctly identified the disease. So we're working on making that a more standard practice for us, but it's certainly a resource that we don't want to lose. What's really interesting I think is, you know, those methylation chips, they have 700, 800,000 or so sites on them. So if you think about like a CPG within a snrp snrpin, for example, you might only pull out three CPGs from that entire region, right? So you throw so much of that data away. So it's useful, but I think long-term we'll want to try to capture much more of that information. questions from the audience, there's plenty online. Is the audience putting the questions in online? That's what I'm assuming. Yeah, if you want. Also stand up and, you know. I think you kind of addressed this a little bit already, Emily, but someone asked a question about whether FFPE samples work for these, and I guess I wonder about DNA quality. We wouldn't be able to get FFPE to work, both in the amount of DNA you get, I can't imagine having enough scrolls, also that tends to be really degraded short DNA, and so then it's really don't have much point for submitting it for long read sequencing. Who wants to take a question about insurance reimbursement? I'm guessing we're probably not there yet, given that this is not clinically available, but if anyone has thoughts on that, and you know, what kind of data do we need to, you know, to convince insurance companies that this is reimbursable and et cetera? So I think that's tough. So we were talking earlier, and this is separate but related, and that our copy number detection from exomes has significantly increased, and yet we're still ordering a lot of arrays, and then we're reflexing to our exomes. And the reason, and so we've really worked, especially actually in our prenatal team, we've really, we've reached out to them, and we've talked to them, and they pointed out that the thing is is that microarrays are almost always covered. And so even if it's not the best for the patient, if we're ordering an exome that's not gonna be covered, and we do an array, and that found the answer, then that's better for the patient. And so I think we're still back looking at how do we convince insurance that we need to start looking at exomes plus copy number, or genomes plus copy number, let alone a brand new technology that we're just now validating, to get them over that hump. I'd like to hope that if we can convince them that genomes in general should be reimbursed, that the jump from a genome to a different, a long-read genome, maybe would be a smaller stretch. But yeah, I imagine it's gonna be a bit of a journey. I've often wondered if the argument could be made that you reduce clinic visits, which are probably more expensive than the testing itself. So instead of five visits to the geneticist and other specialists, you have one visit, you get one test, you get an answer. If that might sway them, I don't know. But it's just a different angle on how to convince payers. That's an interesting angle. Okay, two last quick questions, and then I think we can break. There's a question, any of you can answer, what is the recommended depth of coverage? I think there's probably nuances to that. And then the last question. It's complicated, but we, yeah, for our first set, again, we had 30X for probands, and then 15X for parents, and we definitely missed, there were a lot of things that were called de novo because parents weren't covered deep enough. But in the last ones, we've actually only been doing 20X for probands, and we've definitely done a lot of cases where it's proband only, but again, we already have short read sequencing for parents as well, so it's not exactly a fair comparison. But of the things that we've found, we've probably found half of them in sort of a singleton long read versus trio long read, and so it's hard to tell. We've gone from 20 to 30. I would not go below 20. Okay, great. For HIFI, for nanopore, I would say 50, 40 to 60X coverage is probably ideal for whole genome. Okay, so I lied. Two questions. One, someone asked specifically, Danny, would you trust SNV calling from Oxford Nanopore sequencing? So you can start with that. And then once you answer that, I'm just going to have each of you maybe speculate very quickly on when do you think this is going to be available clinically, and what will it look like? That's tough. I would say I don't trust SNV calling. You know, I have to say, I'm not sure I would sign a clinical report today that says I didn't see something, right? But I think when you see the variant and it's clear, I think that you do trust that, but you just have to be thoughtful about those homopolymers, right, and indels. So it's a very fine balance there. I think with improvements like Q20 and, you know, whatever the next ends up being, whether or not it's duplexer or whatnot, I think we'll get to a point where it's... All right, really quickly down the line, when do you think we'll see long-read sequencing in the clinic? And will that be targeted or genome? So I think you'll see genome first. It's easier, the same as it was for short-read genome. I think within the next two years you'll see it as an option. It's just how widespread will that be and what patients will be able to access it. Agreed. Yeah, I think the next few years for sure. I think it'll be specialized. So like repeat expansions, there was a really nice paper that came out in Science Advances showing you could use ONT for, I forget, 20 or 30 or so loci. I actually think something like that will be the first test we see, and I wouldn't be shocked to see it within a year or so. Not doing it, I'm just saying I wouldn't be shocked if it was, yeah. Great. Okay, I think we have a half-hour break, and then we'll come back for three more speakers and a panel discussion. So thank you all. We'll see you at a quarter after. That's so funny. Yeah, when you did that, I was like, we went on such a huge chase with that excel. Oh, really? Well, that's really interesting. I hadn't thought about that. Like, we called it an expansion. It doesn't seem like that. Yeah, yeah. So we corrected it after that case. And then it turns out that they have. I'm trying to figure out how to confirm it, and I'm like, never mind. I mean, seriously, I gave everybody such a hard time in my last interview. That southern plot. I can't tell you how much I sat around doing it. Thank you. I don't know. You've got to have food. You know what I'm saying? First of all, I'm used to this. You're welcome. I'll make you barbecue. You're welcome. I appreciate it. I was keeping it all in Slack, because I didn't want my Slack to be in my lap. It's really nice, though. Hey, guys. It's great with the black screen. You don't have to nod to everything. I mean, I have chores. I mean, when I started, my lap was full. Why is he, why is he such a strong man? Oh, okay. Winston, are you first? I'll go next. I think you're first. No, Winston's first. All right. I'm pretty sure. Right? Yeah. OK. Hi, everybody. We're going to get started so we don't go too far over into dinner time. So if you can come back to your seats. I think that was a great first session and some, I think, really good insight into how long-read sequencing can help us with diagnosis. I think the speakers this afternoon are going to talk about some applications that we got hints of in some of the first sessions, like looking at epigenetics and gene regulation and other interesting applications using long-read sequencing. And as I said, we'll end with a talk about ASO therapies for rare disease. So our first speaker of the afternoon is Dr. Winston Timp from Johns Hopkins University. Thanks for the kind introduction, and thanks for inviting me to come speak here. This is really exciting for me. This is my first meeting back after COVID, since February of 2020, so I'm really pleased to be here and to talk to you about some of the work that we have done using long-read sequencing. Specifically, I'll be talking a lot about nanopore sequencing today to interrogate the genome, epigenome, and transcriptome. So I do like to list disclosures, which I do have two patents, which have been licensed to Oxford Nanopore. And as I said, I'm going to talk a lot about nanopore sequencing. So to orient you, Danny and others have already talked a little bit about this, but nanopore sequencing works differently than sort of what you may be used to from so-called next generation or illuminance sequencing. Instead of sequencing by synthesis, it works like a coulter counter. You have basically a hole, you put a molecule, could be any biopolymer actually, but in this case it's DNA or RNA, through the hole, and you're measuring the electrolytic current, the ionic current that's passing through. That ionic current is influenced by a certain number of nucleotides in the central constriction of the passage. It's actually a biological pore from E. coli, CSGG is the 9.4 pore, some mutation of it that's proprietary. And you can measure out the current, and the current can be then base called into sequence. We find that in our hands we generate from the minion, which is a smaller sequencer, we can generate between 5 to 20 gigabases of sequencing data per nanopore flow cell. The reason why we're interested in doing sort of long read sequencing, especially for genomics, there's a couple. One of them has been well introduced, so I won't spend too much time on it, but structural variations are a big thing that is actually relatively hard to capture with short read sequencing. One of the things I like to hit upon is that they're pervasive in cancer, but poorly detected sometimes. Sixty percent of pancreatic ductile adenocarcinomas have SVs. This is something we think about a lot at Hopkins where we do a lot of pancreatic cancer work there, and it's common in tumor suppressor genes. Long sequencing reads can resolve structural variants much better than text structural variants, but I do want to note that even with the coverage I was noting from the minion, targeted sequencing can still be beneficial because whole genome sequencing can give you relatively low yield, and when you're talking about clinical cancer samples instead of looking for sort of germline variants, you may have a very heterogeneous mixture of mutations, and so you really want, in those cases, you're really going to want high coverage. Here's an example of some of this from work that we did with Mike Schatz, who's also at Johns Hopkins, and the figure of merit I want to point your attention to from this somewhat complicated Venn diagram is that Illumina sequencing detects, you know, on the order of 5,500 structural variants in this specific sample, but Oxford Nanopore and PacBio detect an additional sort of 18,000 together and, you know, on the order of 1,000 or so unique to each of those platforms that aren't detected by Illumina at all, so Illumina's missing, we think, a lot of this, and pictured over here in IGV is a region of the sort of world-renowned BRCA1 gene where you see the ONT structural variant and PacBio sees a structural variant, but Illumina sequencing doesn't really see it at all. Another big advantage, and this has also been touched on a little bit, of long-read sequencing is that with long-reads, we're way more likely to encounter a heterozygous SNP. What this means is that it means that almost every read can be phased. This is from a dataset that one of my grad students, Isaac Lee, generated for one of our papers on GM12878, which is a lymphoblast cell line, which is often, it's a control cell line often used for testing out methods, and we found that even not going after super long-reads, you know, targeting something like 10 kilobase long-reads, we can phase 65% of the reads, covering 86% of the genome to at least 10X coverage. This is pretty good. This means that we can start to access questions about really allele-specific mutations, allele-specific methylation, allele-specific epigenetics, as I'll talk about a little bit later. And to define epigenetics, because I don't want to confine ourselves to just thinking about methylation. I know there's been some talk about methylation already, but the definition of epigenetics I like to use is either, you know, cytosine methylation, which is modifications typically in humans at CG dinucleotides on the genomic sequence itself, the chromatin architecture, so whether the chromatin is closely packed or heterochromatinized or more loosely packed and more accessible or euchromatic, and where DNA actually is in the nucleus, whether it's out at the nuclear periphery or in the center of the nucleus, all of these things contribute to the epigenome. All of these things help to contribute to controlling gene regulation. And we found a couple years back now that we published a paper where we showed we could use the electrolytic current passing through the nanopore to detect methylation specifically. So to refresh your memory, if it's been a while since biochemistry, this is what T looks like, this is what C looks like, and this is what 5-methyl-C looks like. And what you can see here is that T has this methyl group and has a carboxyl here. C has an amine group and no methyl, and 5-methyl-C is sort of maybe you could consider it as halfway between because it has the methyl and it has the amine, not the carboxyl. And when you measure the current for three different k-mers, your GCT-TGA, GCT-CGA, or GCT-methyl-CGA, when these k-mers, when these, you know, combinations of different bases are passing through the pore, we get different current distributions. Remember, it's like a coulter counter, so we're measuring pico, but it's really small, so we're measuring picoamperes of current, and we can see that we can tell the difference between T, C, and methyl-C. So working with Jared Simpson at the Ontario Institute for Cancer Research, we developed this sort of software pipeline that Danielle referred to called Nanopolish to call methylation in these areas. We found it was very accurate, that, you know, it was sort of 94% accurate at the 77% of sites that we could call. We're relatively conservative in how we set the cutoff. You could be, you know, as you can see from this ROC curve, you can be more permissive and let more, and make more calls, but then your accuracy drops as a result. Or you can be more aggressive and make fewer calls, and your accuracy will go further up. But using the settings that we sort of recommend, we can get a correlation with whole genome bisulfite, which is considered the so-called gold standard for calling methylation, of, you know, 0.895 or above. More recently we've seen correlations at like 0.95, et cetera, from more recent whole genome bisulfite datasets. This is actually quite an old GM12878 dataset generated in the early days of ENCODE, I think actually at Hudson Alpha. And more recent data correlated on the same cell line from the same split number shows higher correlations. So we think that 0.895 is as bad as it gets, actually. We think we can do better. And importantly, you can't really map really well with bisulfite because oftentimes bisulfite, it's a three-based genome. You're mapping using just A, you know, G, and T because you don't want to make a call about C, so you reduce the genome typically to A, G, and T for mapping purposes. And that means that your mapping is even worse than it is for normal, in repetitive areas, for normal whole genome sequencing. And what you see here is an example of that, where in this MBPF region, which is a segmentally duplicated gene that has multiple gene copies with high homology to each other, we see really uneven coverage from whole genome bisulfite, whereas with long-read nail bore sequencing, we can see nice, even coverage because the reads are so long that they can always anchor to a unique region. They can always know where they should correctly map. These short reads just get really confused because it could map in lots of different locations in the genome because this is a segmentally duplicated region. And what you can see here is sort of the unmethylated promoter region in blue, and then the gene body's relatively methylated, suggesting that this copy of MBPF is probably expressed. This is also a gene that's important in neural development and implicated in glioblastoma, so it's good if we could profile it on a regular basis. So long reads are useful for profiling the epigenetics of this read and this gene family. We've also been working more recently on so-called ultra-long sequencing. So what do I mean by that? Well, I consider, you know, 1 kb, 10 kb reads to not be so long anymore because I've been playing in this field for a while, and now we're trying to push it further. We're trying to get reads that are longer than 100 kilobases long, and working with Circulomics, which is now a part of PacBio, and with Oxford Nanopore's sample preps themselves, we've been able to achieve reads in excess of 100 kb sort of standardly. We can get, from running this on the larger sequencing instrument, the Promethean, we can generate, you know, something on the order of 163 gigabases of sequencing data from two Promethean flow cells, so this is the ultra-long runs. So it's about 80 gigabases per flow cell, and an N50 of 110 kb, in that case, N50 means that half of the data is coming from reads that are longer than 110 kb, so that means that of that 163 gig, 81.5 gig is from reads that are longer than 110 kilobases long. We generated the same from a couple different cell lines here, so you see GM12878, MCF10A, which is a breast cell line, HTT116, which is a colorectal cancer cell line, and then we also did it from some clinical blood samples because we wanted to prove that, you know, this isn't just like a cell line thing where we're playing in tissue culture and we have sort of the perfect scenario. From a relatively fresh whole blood draw, we were also able to achieve nice, long, very long reads, you know, where the N50 here, half of the data was coming from reads that were longer than 106 kilobases long. And we see here reads that are 2.7 megabases, a single read that's that long. This is relatively, I don't want to say easily achievable, you have to pipette very adiabatically, very slowly, very carefully, so you don't rip the DNA, because it's actually, you know, very long pieces of DNA, but you can achieve this, and you can get very long reads and then use these to go through repetitive areas. Using these kind of ultra-long data, we're looking at the blood regions, we decided to look at a region here, focusing on these two genes, LTA and TNF, and this area actually has differential methylation between lymphoid and myeloid lineages. Now, so remember, we took just a whole blood sample, and we sequenced it, and when we actually are able to split it out and phase it, we can see reads that belong to the myeloid cluster versus reads that belong to the lymphoid cluster, and what I want to emphasize about this is that each of these long, single reads, each of these 100 kb long reads, is from a single cell. We're not doing any PCR here, so that means that, you know, I'm not trying to claim that this is the new single cell sequencing, they're coming at the problem of heterogeneity of samples from two different angles. One is, I have a very long single molecule, and I can correlate things along that molecule, but I don't know, from this location, what's happening on, like, chromosome 5 or something else, I don't know, I can't link those two pieces of information together. Single cell sequencing gives you lots of information that's not, you have dropouts and stuff, and you get lots of little pieces of information that are linked for a single cell, but you don't necessarily get that long, continuous stretch. Different aspects of this are going to answer different questions, and that's a theme I'm going to return to, which is, there's always a best tool for the job, and don't always be drawn to the shiniest object. I think that long read sequencing is super cool, but there's some things that it's still not great at. So closing this parenthetical, we can deconvolve ultra-long reads to look at the heterogeneity of a sample based purely on the epigenetics here. We also tried to look at other facets of the epigenome. So we, I'm going to tell you very quickly about two things that we wanted to look at. One was chromatin accessibility, and one was looking at proximity to nuclear lamina, where DNA is in the nucleus, whether it's out at the periphery or whether it's in the center. So chromatin accessibility, we're interrogating using a methyl transferase, which will decorate accessible areas of chromatin and leave the inaccessible areas alone. For lamina, we have a lamina dam, which is methyl transferase fusion, which will then decorate DNA, which comes out to the periphery. So when we look at chromatin accessibility with this tool, we can phase, from the same single reads, we can phase both sort of methylation, as shown here on the left, where you can see an unmethylated copy of PIM2 and a methylated copy of PIM2, so we see reads that belong to two clearly distinct sort of haplotypes. Then on the chromatin accessibility, we also see that the promoter here of PIM2 is also very accessible. We have this other accessible region with what looks like an inaccessible footprint in the middle, and then this other accessible region with another inaccessible footprint in the middle on this haplotype, and we don't see that kind of accessibility behavior at all on this other haplotype. So I was mentioning how we see these big accessible chunks, and then we see a little inaccessible area in the middle. What we think is happening there is that you have this big sort of open area of chromatin, and then you have a smaller protein that's sitting down right in the middle of it, and by observing the size of the inaccessible regions, we're trying to interrogate whether we can actually distinguish which protein, whether it's a nucleosome, which protein it is there, and then we can even try and think about being quantitative about that, because if we look at regions that have sort of ChIP-seq peaks, so ChIP-seq chromatin immunoprecipitation, where you're pulling down regions that have a protein bound, we see that in a region that supposedly has a ChIP-seq peak, we see like, for example, in this location, we see 85 percent of reads seem to have that little footprint in the middle of an accessible region. So again, blue is accessible, the sort of peach is inaccessible, and you see a big accessible chunk with that little 30 base pair footprint that we consider to be CTCF sitting there, bound there, and then we see, you know, some reads that don't really have that signature. And then in a region that doesn't have a ChIP-seq peak, that has low CTCF binding, we see the opposite, where you see maybe one or two reads that seem to have it bound, have that footprint, and the majority do not. But what's interesting about this is that when we look at all the regions that don't have a ChIP-seq peak, we generally don't see a lot of R-projection bound peaks. But when you look at reads that do have a ChIP-seq peak, you see a big distribution of the fraction of reads that are bound. So ChIP-seq is telling you, yeah, this location has CTCF bound at some fraction of the time, and the peak is calling that it's bound here. But we think that this kind of method, because we're not doing APCR, because we're just measuring sort of the accessibility along the location, we think that this method could try and get us at the quantitation of how frequently there's actually a protein bound there. And looking at this distribution of different frequency, we think that we might be able to interrogate this to get better information about the dynamics of protein-DNA interaction. We've even tried to use this deep in the centromere. So the centromere is the sort of mysterious region of the genome because it's extremely big and extremely highly repetitive. And we wanted to try and look at this and figure out, well, can we look at the epigenetics in the centromere with our long reads? And so working as part of the T-to-T consortium that was earlier referenced, we were trying to interrogate centromeric chromatin. And so what's plotted here is this region of the centromeric region where there's these regions called higher-order repeats, or HORs, as shown here in tan. And what we observe is that we see methylation that's relatively consistently high, except for this distinct centromere dip region, as we call it, or CDR. We see a distinct hypomethylated region. Over here on the right is a zoom-in of that same region. When we look at the chromatin accessibility for this same region, we see that it's largely inaccessible in the centromere. This is sort of expected. The facultative error chromatin is expected in the centromere. But we see a distinct, even lower accessibility, or the least accessible region of almost the whole chromosome is right here at the CDR region. We also could see, from cut-and-run, which is a method for measuring protein-DNA interaction sequencing, we could see, for reads that we could map, a lot of reads we can't map, but if we look for sort of unique k-mers in these centromeric regions, we could see what looked like peaks for the cut-and-run data for these centromere-specific proteins, SemPA and SemPB, at the same locations as the CDR. So we think that we can actually explore centromeric chromatin and centromeric methylation, centromeric epigenetics, using these long reads, being able to anchor through. Now this does require a high-quality assembly of the centromeres, which was produced as part of this telomere-to-telomere consortium, but we think that this is a first step towards trying to be able to interrogate directly the epigenetics of these regions. Turning our attention to the lamina-associated domains, these are thought to be pretty important. Mouse and human cells have something on the order of 1,000 to 1,500 lamina-associated domains that are really big. They range from 10 kilobases to 10 megabases in size, and they've been identified historically with this DAM-ID technology, where you have a protein fused to a methyltransferase, and then you use restriction enzymes that are sensitive to the methylation to be able to enrich for those locations, and then either sequence or run it on a microarray, depending. And what we're finding, what has been found is that genes that are out at the nuclear periphery are typically off. BLADs typically have low gene density, and the gene expression in those is low. How they're organized, they fall into the B, high C compartment. They're typically replicated late. There's a whole thing that are associated with this, but these can also get moved around during inflammation, during other big gene expression changes, during different transcriptional programs being activated. So understanding this aspect of the epigenome, like the physical location of where your DNA is in the nucleus, could be powerful. So we tried to work on this using nanopore sequencing, because we figured, oh, it's another methylation mark. We should be able to see this. It turns out that it's not so easy as that, because for nanopore, 5-methyl-C is pretty easy to see, as you can see here from, again, this is the electrical current, if we have like this C being methylated. But when you have this A being methylated, GATC is the recognition site of DAM, and so this A is the one that could be methylated. You see that the electrical shift is smaller. So our ability to resolve 6-methyl-A is not necessarily as strong as it is for 5-methyl-C. So it's going to be harder to pull out accurate calls for 6-methyl-A than it is for 5-methyl-C. This is with the 9.4-pore. Now newer developments and newer bioinformatics may make this more possible. I know that there are new models, both that we're working on and others are working on, to try and improve this. But just suffice to say, it's harder, and not all modifications are created equal in terms of how you can measure them using the ionic current through the pore. That said, using the ionic current through the pore and trying to call amethylation with—this is using a model that we trained with Taiyaki, which is one of Oxford Nanopore's tools—we find that at places where we expected LADs, based on sort of canonical definitions of LADs colored here in this sort of orangish color, we do see clear enrichment of the blue signal, which is our amethylation signal, our DAM signal. And we also see, because we can measure these at the same time again, we're getting multiple layers of information on each molecule, on each read. We can also measure a dip in methylation at the same places where we're seeing an increase in amethylation, because the LADs are typically hypomethylated, typically have lower DNA cytosine methylation. And we're seeing that the regions which we expect to be LADs also have that sort of nice blue signal. So we think that this is going to be possible to layer multiple types of information. I showed you we can layer—we can measure DNA methylation. We're getting the genomics data. We can measure DNA methylation. And now we can potentially also layer on where DNA is in the nucleus, the lamina association of it. So we can put multiple layers on distinct single molecules to sort of multicolor measurement of individual long molecules encoded and read out with the nanopore. But coming back to the earlier point, I don't know that whole GM sequencing, even with the kind of yield that we can achieve, and we can achieve 200 gigabases from our Promethean run now, which costs about $2,000, I don't know that that's enough. Because again, if we're looking at heterogeneous samples, we want really deep coverage at times. And so we work—working with—one of my PhD students, Tim Gilpatrick, developed this method working with Oxford Nanopore that sort of he calls NCATS, this Cas9 Enrichment Method. I'll walk you briefly through how it works. Danny had already referred to it briefly. You start with some region of interest that you want, and you've isolated genomic DNA. And the genomic DNA is a little fragmented because of how you isolate it, and you have these phosphorylated sites on the ends. First thing we do is we dephosphorylate all the ends with just like some CYP from NEB. Then we use Cas9. This is all in vitro. This isn't like trying to introduce like some sort of knockout or something with Cas9. This is just using Cas9 sort of as a restriction enzyme, but as a targeted restriction enzyme to cut, flanking your region of interest. After you cut, the fresh cut sites now have fresh phosphorylation sites, which means that we can then do ligation that's going to be specific to the locations that we care about here. And then we'll only sequence, or we'll preferentially sequence, the regions of interest. This works pretty well. So shown here over here on the right is MCF10A, where we're getting pretty high coverage of a region of interest that's 18 kb long. This is reads that are going in the negative strand from the negative strand. These are reads from the positive strand. We found that you can do this from lots of different cell lines. And even from the, I don't name these things, from the flongo, which is the $100 flow cell from Oxford Nanopore, we could still achieve 30, 40x coverage. So that's kind of nice because that means that you can make this, you can affordably target a region of interest and still get nice long reads. You might say, well, why don't I just PCR that? Long-range PCR, PCR beyond 20 kb starts to get pretty hard. We found that we could use this targeted sequencing and get out to 100 KB without too much trouble as long as we're purifying high quality, high molecular weight DNA. Moreover, we found that we could generate a whole panel of these. So Tim made a panel of about 10 of these regions and doing them simultaneously on a single flow cell, we could get in excess of hundreds of X coverage for the regions that we were interested in. He'd found that when you design multiple cut sites on either edge, that that really improves your yield and your coverage at these locations. And this is from about three micrograms of input DNA and we can get really high coverage. So this is important, again, if we're looking for samples that we expect to be very heterogeneous in their mutations or in their epigenetic state, that getting really deep coverage at a couple of regions is good. If instead we wanna get broad coverage, this isn't like, if you wanna cover, you know, a whole exome worth of sequencing, that'd be a lot of guides. I don't know if you really wanna do that. We are trying to see how deeply we can multiplex things and if we can tile the guides, but this is excellent for aiming at, you know, on the order of 10, 20, maybe even 50 regions. But if you start to get too many, you may wanna move to a different approach like either adaptive sequencing or some other methodology. Much has been made and correctly about the error rate and concerns with Oxford Nanopore sequencing. And I agree with pretty much everything that's been said. Shown here is, this is from the paper, which was I think two years ago now. So variant calling has been updated somewhat, but I do, it may have, it should have improved somewhat, I should say. But I wanna emphasize that with enough coverage, we can call a lot of the variants accurately. You know, because we can throw 300, 400x coverage at something, oftentimes all those random errors will fall away. Now there are still systematic errors at locations of homopolymers. At some of these locations, there are systematic errors. And importantly, what we've found is that there are some false positives. So look here on the bottom right. We see that there's these variants which are appearing on the negative strand, that's sort of colored in blue here, that aren't appearing on the positive strand. And we, you gotta remember it because Nanopore has a model of a k-mer. As you're going through the positive strand, you're getting your base calling based on a different sequence than the reverse strand. So you can make errors on the negative strand that you won't make on the positive strand and vice versa. So when we apply a filter to make sure that the variants are appearing on both strands, we actually improved our sort of accuracy, as you can sort of see here in terms of our sensitivity, significantly where we were getting only, you know, we were missing or getting a false positive of only one, especially from using either Nanopolish or Modaka. And again, I know tools have improved. DeepVariant is, DeepVariant Pepper, I think is current state of the art. But this is what we found at the time. And I just want to emphasize the point that Nanopore can make strand-specific errors. In 2020 and 2021, I was fighting with this a lot with COVID sequencing that we were doing because, you know, even from amplicons and from the high coverage you get from sequencing SARS-CoV-2, you'd see some strand-specific errors. So you want to introduce a filter there too. We found that this can work even on clinical samples. So we did this on both normal breast tissue and repaired cancer. And interestingly enough, we can see, we can split out the alleles pretty easily here. We can see variants and mutations that are specific, as shown here by the arrow, to the cancer sample. And moreover, we can see variants that are specific to allele two that aren't on allele one. And we see an imbalance between allele one and allele two, suggesting that there may be a copying of a variation here on our old friend TBPV3. And then Tim finally tried to look at BRCA1. BRCA1 is infamous. And what some may not realize, I'm sure this room realizes, but some others I speak to don't realize, is that this region is chock full of repetitive areas. It's full of signs, it's full of line repeats that are often hard to resolve. We were inspired to look at this. Mary Claire King came to give a talk at Hopkins, and we were inspired to look at it because she talked about her travails in trying to actually sequence this region in the first place and how hard it was because of all these repeats and designing and going after it. Tim was able to design guides flanking BRCA1 and capture basically the entire region, the 83 kilobases, as you can see here. And in GM12878, he was able to sort of reassemble some of these areas because the repeats weren't quite right in HG38 versus what GM12878's repeats actually were. So in hard-to-assemble complex regions, you can target out in regions that you're sure about, and you can get sequencing through them. Again, I probably wouldn't recommend going beyond 100 kb. It starts to get pretty hard because you have to have very intact DNA in excess of 100 kb to put into the method. But it can work pretty easily as long as you're staying below 100 kb. And then sort of another point about this targeted method is if you're looking at cancer samples, you can observe the entire region. You can phase the epigenetics with the mutation. Danny hit this point, I think, pretty well. But let me say here that we have a normal thyroid sample, an early metastasis here, and a late metastasis. And what you can observe in the epigenetics here is that in a normal thyroid, you have this unmethylated CpG island promoter of human telomerase. In the early metastasis, you can see the promoter sort of eroding. Each individual read, it's not, you can see that methylation is starting to sweep its way into the CpG island. It's, you know, it's like climate change, right, that the ocean is flooding up, the island is going under. And then the late metastasis, we see two separate alleles, one that has a mutation where the island is still sticking itself a little bit above water here to take the analogy too far, and we have this unmethylated region. But in the so-called wild type or the reference allele, we see methylation flooded through the entire region. So we can look for and test that epigenetic second hypothesis directly because we have the allele that has mutation, and we can phase that with the allele that doesn't have the mutation and look at the epigenetics of them individually. We've also been using this more recently for mitochondrial enrichment. We have a recent preprint that we put out about this with Amy Vandiver, who's a, I think she's a resident or fellow at UCLA. So she's worked on trying to target mitochondrial DNA. You know, it's a circular genome. It's only about 16.5 kilobases long, so it's short from our perspective. And what we did was we wanted to design the guides in a region where we didn't expect to have a lot of mutations, and we wanted to look to see what we could see from those mutations there. So what we did was we designed the guide sort of up here to cut just once, and then we started to try and sequence. The first sample we sequenced was a sample we got from Dr. Hilary Vernon at Hopkins, where it's a patient with mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes, or MELAS. We wanted to see, she had two mutations which were suspected from short-read sequencing, one that was sort of 8%, and one that was sort of 6% at 16.42 and at 13.513, but she wasn't sure if they were in phase. And so what we did was we sequenced these, and we found that we had all four possible phenotypes. We had some that had both, some mitochondrial genomes that had both mutations, some that had only one, some that had only the other, and, of course, some that had neither. That wasn't what we expected, because we expected to either find they were always in phase or they were never in phase. So we thought that this was sort of interesting, but this also speaks to the power of the methodology, that we can actually be able to split out and look at the full read, and knowing, especially when you already know where the mutation is, you can go and interrogate that SMV and see what's happening. Working with John Wannegut at UCLA, we were also looking to see big structural variants, big deletions that were occurring, because these can happen sporadically and heterogeneously. So not every mitochondria is the same. Some mitochondria have these structural variations, and from younger samples, from like 20-year, 26-year-old, or 31-year-olds, we had a relatively smaller number of structural variants of deletions. The deletions are indicated by these sort of arcs, which show the chunk that was deleted. In older individuals, in muscle samples, we found that there was a much, much larger number of large deletions in the mitochondrial samples, as you can sort of pretty plainly see from these figures. And these deletions, we're detecting them from individual reads, and you might say, oh, but Winston, I don't know if I believe your individual read data. Sure, but it begs, it's not necessarily credulous that we would see such a difference in the number of deletions here from older to younger samples. So we're looking at the distribution of the deletions on individual reads, because there's still a lot of reads that don't have any deletions, but there's a lot of single reads that do, and the deletions aren't always at the same point, so you can't get those nice break points where you're calling the structural variant precisely. They're variable across the entire mitochondria, but there's more of them in the older samples. I'm not gonna make a big point of this. Danny had already talked about adaptive sequencing. We developed an adaptive sequencing method at about the same time as MATS came out called Uncalled. I like MATS a lot, and Readfish works really well. Adaptive sequencing is great when you just want to program in. You don't wanna have to design guides, and you wanna go over broad areas of the genome. As I was saying, the Cas9 method is better if you're looking at a fewer number of locations, and you wanna go very deep because you're worried about heterogeneity of your sample, and so you need extremely deep coverage. So use the right tool for the right job. Sometimes adaptive sequencing is gonna be better. Oftentimes, if you're looking for germline variation, adaptive sequencing's way easier. You don't bother doing Cas9. If you're looking at a tissue sample where you expect a heterogeneity or maybe somatic variations to be occurring, then you would probably wanna use the Cas9 method. As the last point, because I'm almost out of time, I'm gonna run out of time for questions, aren't I? I wanted to briefly touch on the idea of doing long-read RNA or cDNA sequencing. My lab worked as part of a consortium to sequence direct RNA sequencing. Nanopore sequencing, as I mentioned at the top, you're just sequencing a biopolymer. So you can put any polymer you want in there, and we put also RNA in there, and we were able to sequence RNA. This means you can get isoforms. You can look at transcripts. You can look at splicing. Looking at, for example, this location here, P16, you can see that depending on whether you have this exon or this, you get a complete, even though this is the same RNA sequence in exon two and three, because it's the same exon, you get a completely different protein sequence depending on your splicing variant. So being able to determine, it's not enough just to sequence over here and say, oh, I see these, the RNAs. You wanna be able to see what your full isoform looks like. You wanna be able to see what the combination of exons is. You also wanna be able to see if you have allele-specific expressions. So I was just sort of showing here that in IFIH1, there's SNPs down here in these red boxes, heterozygous SNPs, but there's an exon way over here in the green box, which isn't in short-read range of these SNPs, but you have an exon missing for the maternal haplotype and is expressed only in the paternal haplotype. So we're having allele-specific isoforms that we can observe with long-read RNA or long-read cDNA sequencing. So to summarize, nanopore technology's full potential, but always use the right tool for the right job. Depending on what you wanna do, maybe you're doing straight-up single-cell sequencing, maybe you're doing just Illumina sequencing, maybe you're doing PacBio sequencing. There's some things that we can do here that are powerful and should be worth consideration. We can measure modifications to their primary DNA sequence. PacBio can do it too. PacBio's better at M6A. PacBio's starting to do 5MC as well. By labeling encoding information into the DNA molecule itself, we can measure not just the genome, not even just the methylome, but we can measure like chromatin state as well. With targeted sequencing, we can go deep. Adaptive sequencing lets us go deep. Cas9 sequencing lets us go really deep at a smaller number of regions. We can do quantitative measurements, I think, of protein DNA binding, and with long cDNA or RNA reads, we can even look at allele-specific expression or allele-specific isoforms. That, of course, I wanna acknowledge my lab and my funding sources, and I think I've ran out of time. Are there any burning questions from the audience? We'll take one if there's time. There was one online question about, for the BRCA1 targeted sequencing, if there was just a single pair of guides used. No, we use three, oh, I'm sorry. We use three pairs oftentimes on either side because we just find it improves the yield substantially. That's just what we've been doing so far. I think there's always room for optimization. If you're like PCR, you put in a region, the primer three, and then you order five different sets of primers, you figure out what works best. We found that Cas9 guides are somewhat the same way in that you gotta really guess and test a couple times to optimize it, even with CutCut and some of the other tools out there. Thank you, we'll have you back up at the end. Okay, our second speaker for the afternoon is Dr. Andrew Sturgacius from the University of Washington. All right, he's gonna talk about gene regulation. Andrew. Thank you. Well, thank you very much for this opportunity to talk with you guys today and also for the wonderful talks before this. So I'm gonna be sort of talking actually a little bit in terms of what Winston was just talking about in terms of how can you leverage this technology, long-read sequencing, to better understand not just the genome but actually some of these chromatin or other epigenetic aspects and really just taking advantage of the fact that you're already doing the sequencing, what additional features can you actually get essentially for free alongside the sequencing when you're thinking about it from an actual sequencing cost perspective. And I'm a medical geneticist as well as an assistant professor in the Department of Genome Sciences and Medical Genetics at the University of Washington. So in terms of chromatin architecture, Winston did a great job sort of introducing us already, but when you're thinking about how the genome is actually positioned within the nucleus, you're typically thinking about it actually in the context of chromatin, i.e. the individual proteins that are bound along the DNA, be it histones, transcription factors, or other proteins that are directly interacting with the DNA. However, traditionally when you're thinking about doing genomic sequencing, we're actually looking at DNA like this, which is essentially you just kind of strip all these information off of it and you're just looking at the core that is the DNA. So how do you actually preserve or how do you actually understand these aspects of gene regulation which are really essential for how genes are turned on and off and are essential for how exactly the DNA turns into a functioning human being. So traditionally the mechanisms that you would do this are approaches where you would measure chromatin accessibility using essentially tools that cut the genome at sites of interest. For example, you can use tools such as a TaqSeq which uses the transposase, TN5 or DNase 1-Seq or MNase-Seq, Cut-and-Run, ChIP-Seq, all these different methods. Essentially what they do is they cut the genome or selectively cleave the genome at a site of interest and then you isolate those small DNA fragments and sequence them using Illumina sequencers. As you can imagine, the fact that you're chopping up the DNA into millions of pieces is not necessarily the most conducive thing for long read sequencing. So when you're sort of thinking about how can you adapt these types of technologies, you need to essentially change the tables and go away from a cleavage-based method for actually mapping chromatin to something that's a non-cleavage-based method. So really what we wanted to do is can we actually stencil, rather than cutting, but stencil the chromatin architecture onto its underlying DNA in a way that actually preserves the underlying DNA. And really this question was born out of over two decades of research which has essentially shown that you can leverage methyltransferases, one of which the GPC methyltransferase, to sort of essentially mark areas of chromatin accessibility or other features onto the underlying DNA. And really the goal of this is that these are not endogenous marks that you're making, but actually exogenous marks that you've specifically made that aren't necessarily there endogenously within the cell. So to accomplish this, we actually ended up going with a class of enzymes that really at the time we were developing this was not that well understood, which is these nonspecific adenine methyltransferases. And the reason why we chose this class of enzymes was for a couple of reasons. One is the fact that since they're nonspecific, they enable us to go after pretty much every single AT base pair in the genome, which really gives us single nucleotide resolution or near single nucleotide resolution in terms of our chromatin architectures. Also, M6A is not an endogenous mark in humans, so we don't have to worry about any endogenous methylation that could conflict with any subsequent interpretation. And also importantly, M6A does not actually alter the DNA code itself. So when it comes to sequencing, we're not necessarily worried about impacts that this would have on our ability to actually do normal genetic variant identification when it actually comes to the point of analyzing the data. So the workflow that we ended up publishing and sort of working on was essentially taking nuclei from cells, treating them with these methyltransferases or these M6A methyltransferases. And we ended up showing that we could do this with around six different nonspecific methyltransferases. And really what these are gonna do is just methylate any accessible piece of DNA inside of the nucleus. So you essentially have footprints where nucleosomes are bound or where transcription factors are bound. So essentially what was described recently by Dr. Temp in terms of using nanonome as well. And then we would just strip the DNA of the proteins at this point, share them to around 15 to 20 KB, which is sort of what we're using for PacBio sequencing, and then put these through a PacBio sequencer. And we would do standard PacBio sequencing, CCS HiFi, which you heard a lot about this morning in terms of the accuracy. And really the reason why we were doing this is for a couple of reasons. One is the fact that for PacBio sequencing, actually M6A, as I'll go over a little bit, it's actually quite easy to detect, but it also enabled us to get sort of that improved sequencing accuracy, which is described earlier. And similarly, it's PCR-free. And really this enables us, as we've really heard a lot about this morning, to get these haplotype-resolved chromatin architectures, which really can be quite beneficial, as I'll go over. And then in terms of the M6A identification, so PacBio and how methylations are identified is actually a little bit differently than in terms of how methylations are identified with nanopore. So with nanopore, as you heard about, sort of methylations are identified based on the change in the signal current that's actually happening, which is the same way that nanopore is measuring DNA. So the exact same sort of thread that DNA is measured is the same way that methylations are measured, and those sort of, you have to control for both of those at the same time. In contrast with the PacBio sequencer, the DNA is still measured the exact same way in terms of its sequence content, which is based on the fluorescent signal that a given base has as it's actually going in, as the polymerase is going over and actually sequencing that base. However, how methylation is read is because once the polymerase, say, hits a methylation, or a methylated base, that can actually cause the polymerase kinetics, or in terms of how fast or slow it's sequencing that, to actually change. For example, here in this diagram, you can see that as it hits an unmethylated adenine, it moves quite quickly to the next base, versus once it hits a methylated adenine, it actually pauses a little bit. And these kinetic aspects of the sequencing can be directly read out, independent of the actual sequence quality itself for the calling of ATGCs, and they can be used to identify methylation marks at a given base. Okay, so what does this actual data look like? So this, I'm actually showing you some data for FibroSeq. This is using this methodology of single-molecule chromatin-fiber sequencing. And just sort of to highlight, I think this has gone over a lot already, but each individual sort of gray bar here is an individual DNA molecule within our sample. And all these purple dashes in here represent sites of M6A methylation. So these are sites where we caused methylation to be there. And you can essentially see that what you're pulling out are these sort of larger footprints right here, which correspond to mononucleosome or dinucleosome footprints with internucleosomal linker regions sort of marked in purple. And then you can also pull out the individual chromatin accessibility sites. And I think that what you can sort of appreciate just looking at from this view is that chromatin accessibility as measured on a bulk assay is quite different than what you see on a single-molecule basis. And what I mean by that is that you can see that on a single-molecule basis, each of these sites or each of these molecules has a very difference, or quite varied in terms of how big and sort of the content, the DNA sequence content of the actual regulatory element in this situation, the promoter for these two genes. Furthermore, by zooming in on each individual site as sort of highlighted earlier with Dr. Tim's stuff, you can not only just pull out the fact that regulatory elements exist but also you can actually pull out individual transcription factor binding events within those regulatory elements. This is kind of where one of the benefits of having the sort of single nucleotide resolution comes in is because essentially you can quite readily identify fibers that do not have occupancy of this transcription factor because they end up having not just a single, say, GPC that was methylated in it, but actually usually on the order of five or 10 or multiple AT bases, giving you a much higher accuracy in terms of actually calling the on or off state in terms of transcription factors, but also enabling us to actually go after transcription factors with much smaller transcription factor footprints that may not be sort of as big as sort of the CTCF, which gives a quite large footprint, which is why it's often used for a lot of these figures such as this one. Okay, so just to sort of show an example of how you can sort of leverage both the genetics plus this chromatin architecture, this is sort of an example where essentially you have a variant, you've identified with Pacbiosequencing. You would do your sort of standard computational pipelines, run it through it, identify it falls within a regulatory element. You'd be quite excited at that point. Could this actually be pathogenic? And just by simply layering the chromatin information on top of these same reads, you can now readily appreciate that, yes, one of these, you know, variants actually, or one of the two alleles actually has chromatin accessibility at it versus the other one. Essentially, you're just seeing a nucleosome that is being sort of heterogeneously placed at that site. In this situation, we were able to validate this particular finding using DNA's orthogonal approach to show that truly this G haplotype is associated with chromatin accessibility and the A haplotype is really quite inactive. This is just another example of that, which sort of just shows the fact that, you know, having this, you know, tight sort of single nucleotide, near single nucleotide resolution would be quite beneficial because, as you can readily appreciate here, this regulatory element shows clear allele-specific activity, whereas the regulatory element right neighboring it actually looks like it's active on both of the alleles. So you're really sort of highlighting the fact that you can zoom in at a very fine resolution in terms of how regulatory variants can actually impact the overlying chromatin architecture in terms of the accessibility of DNA. Furthermore, as we've sort of talked about earlier, you cannot just leverage this for looking at allele-specific stuff. You can also leverage this for looking at imprinting stuff, for example, X chromosome inactivation, which is sort of one of the canonical examples of that. So this is just data from the Xist promoter, which is sort of one of the genes really critical for X chromosome inactivation. And it's, what's interesting is that, yes, you can pull out that, you know, these promoter elements right here appear to be largely turned off or almost entirely turned off on the inactive X, which you would expect because that's really the entire function of Xist is to only be active on the inactive one. But actually, or sorry, but essentially what you have here is that in this situation, this gene regulatory element up here is actually as occupied or presently active on both the inactive as well as the active copy of Xist. Okay. So now that I've given you a little bit of background in terms of, I want to go over sort of three vignettes in terms of stories that we have sort of leveraging our FibroSeq data in this sort of pipeline. One is sort of going after, you know, the sort of complete human epigenome. And then sort of the other ones is I'm looking at actually chromatin architectures at telomeres on single molecule and single nucleotide resolution. And finally, you know, hitting on some of the points from earlier today, I, you know, couldn't help but also go over repeat expansion since that just seems to be sort of one of the consistent themes of the talk today. So in terms of the complete epigenome, so this is really I think nicely covered by multiple individuals during this session, but essentially, you know, our understanding of the genome, you know, is oftentimes limited to sort of areas that we can map quite nicely with short read sequencing, which leaves around 10% of the genome, which is really uncharted from not just a genomic, but also from a chromatin perspective. And really, you know, some of the challenges that exist within this, you know, one is that, you know, if we're going to be looking at these portions of the genome, we actually need to have an accurate reference to map against, and also we need a methodologies that can actually enable us to map reads to these accurate references. So I hope that, you know, in the past couple slides, I've shown you that, you know, these technologies, you know, long read sequencing really enables us to go after this, and, you know, as Winston Tip was touching on, you know, with the advent of the telomere-to-telomere consortium and the CHM13 complete genome, we now actually have an accurate reference genome that we can actually start to look at and start to understand how does, or how is chromatin architecture situated in these portions of the genome, which have previously just been impenetrable. So to really get at this, what our lab did is we ended up actually applying FibroSeq to the exact same cell line that was actually used for the T2T genome. To essentially ask the question, which is, what is the utility of having truly a paired perfect genome with a sort of paired sort of chromatin long read sequencing approach for looking at these really difficult and challenging portions of the genome? So for this, you know, we did our standard FibroSeq. We mapped it to the T2T genome. We got 1.7 million single molecule architectures and average size of around 16 KB consistent with our HiFi sequencing. And essentially, what I'm showing you here is this is how well we were able to actually map our data to the genome. And essentially, what you can see here is that we were able to perfectly map all of our reads to the eukaryotic genome. We had no issues with that, which is what we'd expect. But also, we were able to map quite easily to telomeres and centromeres, sort of really traditionally impossible to map areas in terms of our chromatin understanding. And we were able to actually undercover sort of the architecture of all telomeres and centromeres within these cells using this approach. So one initial question is, do we actually need to go through the trouble of actually having a paired epigenome and genome from the same sample? Can you say mix and match, say take an epigenome from GM12878 or another cell line and mix it with the CHM13 genome and start to look at these more challenging areas? So to accomplish that question, what we ended up doing is actually just applying our FibroSeq methodology to GM1287 cells and mapping that to the CHM13 genome to actually see how good the CHM13 reference was in terms of actually looking at some of these very, very challenging portions of the genome. What you can see here is that for the euchromatic portion of the genome, the CHM13 reference was great. Even using GM12878 data, we're able to essentially map to the entirety of it. However, once you start to look at telomeres and centromeric areas, we saw that really that took a big hit. And really what this sort of highlights is the fact that there's likely a large amount of genomic diversity that exists within these telomeres and centromeres that even though the CHM13 genome is really a complete genome from T to T, it is a complete genome from T to T for CHM13 cells and for the individual that those derive from. It doesn't necessarily represent sort of the population that's out there. But nonetheless, it is a much, much, much better genome build than HG38, and I can tell you that. But still, it's just sort of a highlight that really having these paired epigenomes and genomes from the same individual can be quite beneficial when you're trying to look at some of these very challenging portions of the genome. So, you know, really I think we can say that, you know, we can now look at chromatin, you know, within all of sort of the, you know, genome in this, or almost nearly all of the genome. And so for this, I'm just going to go over sort of two vignettes in terms of what we've been able to look at. You know, one is sort of for telomeres, and the other is for short repeats. And these are some of the people from my lab that's really been essential for getting these stories up and running. So in terms of telomeres, you know, from a human genetics perspective, you know, telomeres are quite important, not just because they're essential for genome biology, but they actually, you know, multiple disorders are associated with disruptions within telomere function, as is listed here. Telomeres are interesting when you think about it from a chromatin perspective, because actually telomeres have a very unique sort of structure within the nucleus, which is essentially that they're actually covered by a telomeric cap, which is this sort of not fully appreciated sort of structure, which involves a sheltering complex, and potentially nucleosomes or histones that are sort of intermingled in there. But really, it's not been that well sort of situated exactly how the structure exists on individual molecules of DNA. And obviously, as you can imagine, that the challenges associated with sequencing TTAGGG repeats that are over, you know, several kilobases in length has sort of interfered with a lot of the technologies to really uncover this from both a genetic as well as chromatin perspective. So really, the two questions that we really wanted to go after with this dataset were sort of what is the impact of telomere shortening on cap formation? And the other is how did genetic alterations in the telomere actually impact the telomere cap itself, really trying to get at sort of the core biology of how exactly the genetics of the telomere are actually impacting its function, which really is being read out here as the telomere cap itself. So what I'm showing you here are actually what the telomere looks like from a chromatin perspective. So these are actually a little over a dozen reads mapped to the P arm, the P telomere of chromosome 12. And I think what you can sort of readily appreciate here is that actually the chromatin architecture of telomeres is actually divided into three separate domains. And so what you can appreciate here is that at this end of it you have these long extended regions which are entirely protected from methylation corresponding to actual telomere cap that we're observing on a single molecule level. And then what you can see is that this actually transitions into this relatively highly accessible regulatory element, which I'll go into a little bit more in the next couple of slides, which is then sort of has a very clear and striking boundary, at which point that transitions actually into sort of a traditional heterochromatin environment, which is what often people think about when they hear telomeres as sort of being this heterochromatic structure. Strikingly enough, actually, this sort of tripartite chromatin architecture is actually the norm. So actually over 52% of the fibers that we actually see arising from telomeres have this sort of very distinct chromatin domain structure that is observed here. So the interesting thing is when you're looking at these actually sort of chromatin sleeves or telomere caps is that you can appreciate they're actually quite heterogeneous. So individual fibers actually show striking differences in terms of the size, as well as sort of where they terminate and sort of start to transition into sort of standard nucleosomes, or even having sort of spurious what appears to be nucleosomes present within what appears to be sort of a telomere cap on either side of this. And if you actually quantify this, you can see that on a per-molecule basis, you see over 1,000 base pair difference in terms of where the telomere cap ends and where sort of the nucleosome pattern actually begins. And sort of to answer sort of the first question that we had is what we actually were able to show is that, yes, if you actually do have shorter telomeres, because we're not only measuring the chromatin, we're actually in the DNA, we're also measuring the telomere size, as is appreciated here. But these shorter telomeres are truly associated with actually a shorter chromatin cap. So you actually are diminishing the chromatin cap size on a per-molecule basis when you're actually having shortened telomeres. So now we want to ask, okay, great, we've sort of seen this on a sort of length basis, but now can we actually start to look at the genetics of this to actually look at the genetic architecture of these individual telomere molecules? So this is sort of showing you just the data, you know, where you're looking at the telomere on a length basis. But we can now actually leverage the fact that, you know, this is packed bio-sequencing, we have the DNA content, and we can simply ask is, are these telomeres actually made up of TTAGGG repeats? Yes or no, and where are they, and where are they not? And so what I'm actually showing you here is on a per-molecule basis, in blue, are actually areas of each fiber that actually contain TTAGGG repeats. I'm actually more so limiting this to TTAGGG TTA, because that's actually the minimum sequence required for shelter and occupancy. And in red are actually sites in where you have non-TTAGGG repeats. So what you can sort of readily appreciate is that telomeres are not just TTAGGG repeats, but rather you have these sort of non-TTAGGG repeat sequences that actually punctuate and that are actually quite sort of uniform to some extent across these individual fibers for on a per-chromosome basis. Now the question that we had is, can we actually simply ask this question, which is do these TTA, sorry, non-TTAGGG repeats actually disrupt the function of the telomere, i.e. when you actually see these non-repeat structures here, is that actually causing these telomere caps to actually break at those sites, which is what you would expect if you sort of had this sort of naive model, which is just shelterin, which binds only to TTAGGG repeats, is actually mediating the telomere cap. And what you can appreciate is that actually, yes, when we see disruptions, they tend to actually occur at these non-TTAGGG repeats. So yes, these actually are disruptive in terms of the ability of the telomere to form caps. However, for most fibers, they don't really care, which is kind of striking, that you actually can see the telomere cap formation actually bridge these sort of invoking that actually it's not just shelterin that actually is forming the cap type structure, but actually multiple other proteins are actually doing this together. This is sort of just highlighting this right here that it's probably just multiple proteins beyond just shelterin, which is really enabling these caps to form and to bridge these non-TTAGGG structures. Okay. So now I'm going to zoom in to actually not the telomere itself, but actually this area right next to this telomere, this highly accessible region that we identified. And one of the striking things is when we zoomed in on this sort of boundary right here between this element and actually this heterochromatic domain, is we actually saw that this is perfectly bound by CTCF. So essentially this transcription factor, which is known to be sort of this master regulator of the genome, appears to be actually almost universally bound right at this telomere-sub-telomere junction. And not only is it bound there, but actually what we see is that, excuse me, let me back up a second. So these are actually contained within repeats, which are called TAR1 elements, or TAR1 repeats. And these TAR1 repeats, which we actually see abutting the telomere, appear largely unique to actually telomeres. So when we see them elsewhere in the genome, they appear to be actually broken or fragmented, and also do not appear to have CTCF occupancy at them. But the striking thing is when we actually see CTCF bound at this telomere-sub-telomere boundary, it's not just CTCF bound as a loner. It's actually often CTCF bound with multiple CTCF occupancy events happening immediately adjacent together on the same molecule, which is what you're appreciating here, where you can sort of see footprints for three CTCF proteins on the same molecule. And not only that, but they're actually all oriented in the exact same orientation whenever we see this, such that the CTCF is actually oriented with its N-terminal domain sort of abutting the sub-telomeric area. And so the question that we have is, given the fact that CTCF is known to be sort of this master regular, important sort of in terms of chromatin organization, is this actually having a functional impact in terms of the telomere itself by having this sort of sub-telomeric CTCF bound there? And so what we actually were able to identify is that truly that when we see these CTCF bound fibers, that actually is associated with telomere sequences, which actually have a much higher burden of these non-TTAGG repeats. So essentially, there's this crosstalk between the chromatin state and then the actual genetic architecture of the telomere itself, which we still are trying to sort out the mechanism of this exactly, but there's some sort of ability for these elements to essentially protect the telomeres from actually undergoing degradation and sort of de novo reconstruction of the telomere in and of itself. Okay. So that's sort of an overview of sort of some of the stuff I've told you about for the telomere, and now I'm going to switch over in the last couple of minutes in terms of actually now what you can do when looking at actually repeat expansions in and of itself. And, you know, as I think I really do not need to go over this, you know, because it's really been sort of well talked about in the last couple of slides, there's multiple different repeat expansions that are involved in numerous different diseases. We're probably just, you know, there's a lot which we're still learning about these and many more that are left to be discovered. So in terms of the mechanism of repeat expansions and how they cause disease, you know, that's definitely worked out, but one of the sort of things that, you know, obviously happens is that they need to expand. You know, repeat expansions, you know, half of the words or half of that statement is expansions, and how do they expand? There's multiple different mechanisms that can happen, but especially for CAG repeats, one of the major mechanisms that enable them to expand is by forming these slipped DNA structures. This is sort of a diagram which is shown here, and essentially it's been sort of well worked out that these slipped DNA structures can cause repeat expansion instability, and the longer repeats are more likely to form these slipped DNA structures. However, our view of sort of thinking about these slipped DNA structures is sort of, you know, molded by looking at the genome in this sort of context. However, as I mentioned earlier on, our genome is not in this context. It is actually bound by proteins. So now you have to ask a simple question. It's great if these repeat expansions, or sorry, if these slipped DNA structures are essential, how are these slipped DNA structures actually interacting with the chromatin environment to even enable them to have enough exposed DNA to actually undergo this slipped structure? So we specifically were interested in HTT for a number of reasons, but one is actually we had this sort of earlier observation that the CAG repeat in normal cells happened to perfectly overlap what appeared to be this intranucleosomal linker region. So essentially the promoter is located right here. You have this plus one nucleosome, which just refers to the first nucleosome after the promoter, and this appears to sort of create this intranucleosomal linker region that overlaps the CAG repeat on a number of different reads. And really we wanted to know is the actual chromatin architecture at this site actually potentially impacting the ability of the site to form these slipped DNA structures, and also how does this actually occur on disease alleles for individuals who have Huntington's disease? And so to do this, we ended up actually applying FibroSeq to this quad samples, which are individuals sort of with and without Huntington's, as well as with juvenile onset Huntington's, and also at risk. And to really make this a little bit more efficient, we ended up actually not just doing our standard FibroSeq, but we ended up developing a targeted FibroSeq approach. This way we can actually pool and multiplex these samples into the same smart cell and give us sort of improved coverage and also reduce sequencing costs altogether. So in doing that, what I'm showing you here is these are actually the individual molecules, the repeat size of the actual CAG repeat. And you can sort of clearly see, sort of we're able to see two distributions for this individual with adult-onset Huntington's disease versus unaffected individual. You can sort of pull up both of the, you know, sort of unexpanded alleles right here. You know, if you look at this at-risk daughter, you can sort of similarly see both copies. And then this individual juvenile onset, you can see that there's been this expansion selectively with that individual. And you can also start to appreciate some of the heterogeneity that's actually existing, you know, even within this size range on a per-molecule basis. So now I'm actually showing you some of the data from this where you can actually now look at the per-molecule chromatin architectures. But not only that, but now we can actually look at the chromatin architecture in the repeat itself. And so in this situation for this individual who had the expansion here, you can sort of appreciate that you have your promoter architecture. You have a nucleosome. You have another nucleosome. You have a large expanded inter-nucleosomal linker region. You have another nucleosome versus you have these large expanded regions right here. You can actually look at chromatin within these highly repetitive portions, which you've never been able to sort of really look at before. And one of the striking things that came from this is that when we look at repeats that say are only less than 26 CAGs in length versus those that are expanded, we see that the actual size of the accessible piece of DNA within that repeat actually significantly expands when you actually have this expansion. And really this expands to a range which is known to be suitable or compatible with actually forming these slip DNA structures. So really what we're showing here is that the chromatin context at the HTT for repeat expansion is actually creating an environment that is conducive for actually the somatic formation of these slip DNA structures to form. Okay. So that's sort of it for this presentation. So just in summary, really, you know, you can leverage these single molecule chromatin fiber methodologies, you know, to look at these allele-specific chromatin architectures across entire regulatory domains. You can map chromatin architectures within these previously uncharted genomic regions, you know, in terms of the telomeric stuff, you know, showing that, you know, the sequence in chromatin architecture of this telomere really sort of has a big impact in terms of its function when you're measuring that in terms of the cap formation and then sort of looking at these functional interplay where I showed you in terms of the chromatin state as well as the genetics of these repeat expansions. And, you know, this is work that could not have been done without the massive support of my lab as well as these funders here, and I really sort of have everyone to thank. I just started my lab less than two years ago, so it's been just a huge effort on everyone's part to sort of get this, and I'm really very, very thankful for everyone and as well as our collaborators. So thanks. I'll say congratulations on starting your lab two years ago, which was exactly when COVID hit, so that is remarkable progress. July 2020. Yeah. Any questions from the audience? There was one question online. It's fairly technical about what high molecular weight genomic DNA isolation strategy you used for your epigenome sequencing. We've used a bunch, and we've essentially found that most all of them will work quite well. It all just depends on terms of like what your individual tech is most familiar with working with as well as how much yield you want to get at. Right now we're using a combination of sometimes we'll end up using chloroform for some of our stuff where we really want to get really high molecular weight stuff versus we also use the Monarch, as was mentioned earlier, kit as well as a couple of other stuff. But really, I found that that's actually not that critical, especially once you're going after maybe the 20 kb reads. I think once you start getting much higher than that, it gets a lot more critical in terms of the genomic kit you're going after. All right. Well, thank you for your questions for the panel discussion at the end. We'll do our last talk, and then bring everybody back up. So our last talk is from Dr. Frank Bennett, who is from Ionis Pharmaceuticals, and is going to talk to us today about allele-selective antisense therapies for rare neurological disease. Thanks. I'd just like to thank the audience for sticking to the end of the talk. I know it's tempting to escape now, but what I'd like to do is talk about what's downstream. So if you have a genetic diagnosis and you've phenotyped what impact that genetic diagnosis in, the next step is obviously to develop a therapy for the patient. And so that's what I'd like to focus on. And so just, I think it's obvious, but I'm an employee and shareholder of Ionis Pharmaceuticals, so I'll just leave it there. So I'm old enough to remember these two papers that came out over 20 years ago, and they were really landmark papers, and they gave us the first draft blueprint to what the human genome was. And for those of you who were working at the time, you may remember how controversial it was to spend $3 billion to get this information, when it was questionable what value was going to come out of it. And so there was a tremendous amount of controversy there. I think that controversy is behind us now. We do really recognize the value that's come out of this effort, and it really was a yeoman's effort with the sequencing methodologies that were available at that time. And so what I'd like to do is kind of focus where I see sort of the future of medicine is, is really, and I think this audience well recognizes this, that I don't think we're far away from having whole genome sequencing on almost all people, at least in the United States. There's obviously some ethical and, you know, genetic counseling that needs to be done, and I won't touch on that because it's not something I think about, but it's really what I think is the next step, is that it will really impact how we're treating disease. And, you know, I believe that if you know what you're at risk for, you can start impacting lifestyles fairly early in the process, in childhood even. And then if you know, if you have a disease, and you know a genetic basis for the disease, or modifier genes that may be modifying your potential for the developing the disease, there's obviously a therapeutic potential. And really where I think medicine is going to start going is more towards prophylactic treatment, and I think, you know, the work that we've done for spinal muscular atrophy, where we're starting to treat pre-symptomatic patients that have a very severe neurodegenerative disease, really highlight the potential of prophylactic treatment, where you can, in many cases, you can prevent the children from ever developing the disease. And so I think that's really been phenomenal. And then as far as the therapeutic potentials, there's, you know, small molecule type drugs, there's protein-based drugs, and where I think, again, the field is moving is more towards genetic medicines that can really take advantage of the sequence information that we have. And so I think genetic medicines are going to be an increasing part of our future in developing therapies. And so what I'd like to do is discuss one type of genetic medicine that I've been working on for 32 years. It's a technology called antisense oligonucleotides that we're designing synthetic oligonucleotides that will bind to RNA through Watson-Crick base pairing. And once it binds to RNA, it can invoke a number of different effects on that target RNA. Perhaps the best characterized effect is by causing degradation of the RNA through endogenous enzymes that exist in cells. So we use an enzyme called RNase H. It's RNase H1 is the isoform that is responsible for degrading the RNA upon binding an oligonucleotide to it. And it's found both in the cell nucleus as well as in the cytoplasm. So RNase H can target a pre-messenger RNA, so before splicing is occurring, while it's still being transcribed off the chromatin. Or it can target mature messenger RNA in the cytoplasm. The other major mechanism that people are using today is siRNA that uses an enzyme called ALGO2 to cause degradation of the RNA. So they're both catalytic mechanisms. They both have nuances in whether you'd prefer to use RNase H or ALGO2 mechanisms, but they both work well for causing degradation of the RNA. The other thing that we do, and it's actually what I spend a lot of my time thinking about, is how to use oligonucleotides to modulate RNA in cells that doesn't cause degradation. And so one way that we could do this is to modulate splicing, where you can cause either exon inclusion, as in the case of spinal muscular atrophy drug, Nusin-Hirsen, or you can cause exon skipping, as in the Duchenne's muscular dystrophy drugs are doing. And so depending on where you bind on the target RNA, you can direct it to either include or exclude an exon. You can also modulate polyadenylation site selection. So most transcripts have alternate poly-A sites that in many cases will eliminate regulatory elements in the 3' UTR of transcripts. And so you can use oligonucleotides to direct which poly-A site is being selected. We've shown biochemically we can do that. It hasn't been exploited broadly, but I think as we better understand RNA metabolism, it will become a more important process. And then finally, you can use antisense oligonucleotides to inhibit translation in the cytoplasm. And there are a few mechanisms that you could do to increase translation, either by blocking microRNA binding sites or alternatively these upstream open reading frames that many transcripts have that tend to limit translation. So there are a variety of ways you can use antisense oligonucleotides to modulate RNA function in cells and modulate the output. So I mentioned earlier that antisense drugs are chemically modified oligonucleotides. And this is the rest of my lecture is going to focus on medicinal chemistry of oligonucleotides. Just kidding. Sorry. But there are a number of chemical modifications that you could use. And it's important that you do use chemically modified oligonucleotides. WAL-type DNA or WAL-type RNA gets degraded within a matter of a few minutes in biological systems. So you have to invoke some chemistry to stabilize the drugs. And they include, as shown in the bottom left-hand side of the slides, the phosphate modifications, in particular replacing one of the oxygens with a sulfur atom. That's called a phosphorothioate modification. That's commonly used in most antisense drugs that are in the clinic today. There are base modifications. And you've already heard about 5-methylcytosine. We use 5-methylcytosine on RNA largely because it tends to mask the oligonucleotide from toll-like receptor systems or pattern recognition receptors. So you tend to decrease the pro-inflammatory effects that you see with nucleic acids. And then there's a whole series of sugar modifications that are being used today that all have the potential to further stabilize the oligonucleotide against being degraded. Also, importantly, many of them will increase binding affinity for target RNA, so therefore you get an improved potency. And they further reduce some of the pro-inflammatory effects you see with modified nucleic acids. And then finally, an area that's really up and coming within the antisense field is sticking biomolecules onto the oligonucleotide to help with targeting to different cell populations in the body. And I'm showing here a conjugate called GalNac that binds the acyclic glycoprotein receptor that's expressed on hepatocytes. So you get very good improvement potency for antisense drugs targeting liver hepatocytes. And so we and others are looking very broadly at other ligands that we can use to further enhance the potencies for a variety of different cell types throughout the body. Today, there are 12 approved antisense drugs. Most of them are for rare diseases that range from familial hypercholesterolemia, Spinraza, Teplersen for Duchenne's muscular dystrophy, as well as other Duchenne targeting drugs. There is one drug that's approved that in principle would be applicable to a very large market called Inclisirin. That's a PCSK9 inhibitor that's used for hypercholesterolemia as well. And there are more than 50 antisense drugs, and I'm including siRNA in that category, that are currently in clinical trials. So it's a very robust area of research right now within our industry. So some of the attributes of antisense drugs is, as I mentioned, a direct translation of genomic information to a drug. So if you know the sequence of a gene that's causing pathology, you in principle can design an antisense drug to modulate that gene. And importantly, it can target any RNA. So both protein-coding RNA as well as non-coding RNAs can be easily targetable with antisense drugs. It's a very efficient drug discovery process, and I'll touch on that in a second because it's really enabled thinking about targeting small patient populations because of the efficiency. And in contrast to gene therapy, antisense drugs, the effects are titratable. So it's like any drug you administer orally, the more drug you get, the more effect you see. And if you remove or stop administering the drug, the effects are reversible. So again, contrast to gene therapy where the effects are permanent. So if you're dealing with a drug where you really need to titrate how much inhibition or activation you're seeing, antisense technology is a good technology to use for that. And then finally, we're making a lot of advancements. I already talked about tissue targeting that's going on, but there's a number of other advancements that we're seeing with the technology. So the profile of the antisense drugs that exist today will be different 10 years from now, and we'll have improved drugs that I think better treat our patients. So what I'd like to focus on now for the rest of my talk is CNS delivery of antisense drugs. And I mentioned that the antisense drugs have a phosphate backbone or a phosphorothiolate backbone that has a negative charge. And so that negative charge prevents these drugs from crossing an intact blood-brain barrier. And so to deliver the drug into the CNS, we're using lumbar puncture as a way to deliver directly into the cerebrospinal fluid. And then once it's in the fluid, we've been able to document that it very rapidly distributes from the base of the spine to the brain and then penetrates into brain tissue where we can target almost every cell type in the brain. So today we have experience in well over 11,000 patients. Most of those are from Spenraza or Nusinirsen that's used for treating spinal muscular atrophy. But we also have experience, a growing body of experience in treating patients with ALS, Huntington's disease, Alzheimer's disease, and Parkinson's. So it's being broadly used for neurodegenerative diseases. And we're beginning to do work now in neurodevelopmental disorders as a technology. And so we were really the first company that was pushing forward using antisense drugs in the CNS. And I think with Nusinirsen's success as a commercial success, it's brought a lot of competition into the field. And you can get a sense for the breadth of which antisense drugs are being used today for potential therapies for a variety of different neurological diseases, including neurodegenerative diseases, as I mentioned, neurodevelopmental disorders like Angelman syndrome. And then a number of programs are in drug discovery that will hopefully enter clinical development in the next couple of years, ranging from, again, a number of triplet repeat diseases for spinal cerebellar ataxias, prion disease, leuphora disease, different forms of autism, et cetera. So as I mentioned, one of the attributes to antisense drugs is that, compared to traditional pharmaceutical drug development, it is a very efficient drug discovery process where a small team of folks can identify an antisense drug that's safe and well tolerated. And there are two early examples of actually going after a single patient with an antisense drug. And that was really unthinkable with more traditional drug discovery platforms, that you could develop a drug just for a single patient. And so the one on the left is an example that Tim Yu published a few years ago, where he developed an antisense drug for a patient with Batten's disease that had a SVA insertion into an intron that caused alternate splicing to occur and prevented the expression of the CLN7 protein. And then the second one is a program that we've been involved with, where Dr. Neal Schneider was developing, we developed with Dr. Neal Schneider a antisense drug for a very rare form of ALS due to patients that have a mutation in the FUS gene. And so it really laid the framework for thinking about how to use our technology for treating patients that commercially you really wouldn't want to think about them. You couldn't approach commercially. And so one way that we've done this is our CEO set up a foundation where the goal is to provide access to antisense technology, free of charge to the patient, working with Ionis to identify the antisense drug. And the mission is really to apply the versatility and efficiency of antisense oligonucleotide technology to treat patients with these ultra-rare conditions that, again, from a commercial perspective really aren't feasible. And we've arbitrarily set a size of 30 patients in the U.S. It's not something that's written in stone, but it's really 30 or fewer patients is really what we're trying to target within Lorem. We founded the foundation in 2020. And in the first two years, we've reviewed over 100 cases that have been submitted for consideration. We're roughly running about 40% acceptance rate. So looking at the genetics, there are some diseases where the target tissue really isn't amenable to antisense technology today, so we decline them. Or there's homozygous loss of function mutations that antisense technology really isn't the best technology to treat. So there are a variety of reasons why we're not accepting them. But by and large, as we've been working with investigators and getting the word out, they're getting more sophisticated in the cases that are being presented. 90% of the cases are neurodegenerative or neurodevelopmental disorders. So neuro is still a big focus of where the cases are coming from. And they're really a mix of splicing modulation. So there are several cases that have mutations or insertions in intronic sequence that cause aberrant splicing. And so we're trying to block that with antisense drugs. The downregulation with RNSH. And what I'd like to do is focus on allele-selective downregulation with RNSH. And I'll give you an example of that next. So one of the advantages of antisense technology is it can discriminate a single nucleotide change between two transcripts. And one of the early cases that we used was Huntington's disease, where there's a CAG expansion that occurs in the first exon of Huntington. That CAG expansion, because there are a lot of other proteins that have CAG repeats or a lot of other genes, really wasn't amenable to targeting. And we found that CAG targeting, all of those tend to have pretty profound toxicity because we're probably modulating some of these other genes in the body. And so what we reasoned is that there are common haplotypes that are present in the population that we could target single nucleotide polymorphisms that distinguish between the mutant Huntington gene and the wild-type gene. And so the RNSH would cleave the RNA that has a perfect match for the mutant gene and avoid cleaving the RNA for the wild-type gene. Therefore, we're selectively lowering the mutant Huntington allele. And this is some preclinical work that we did in mice that expressed both—it's a humanized mice, so they're expressing two copies of the human gene. No mouse gene is present. And one copy is a wild-type as 18 CAG repeats. The other copy has 118 CAG repeats. And you can see for an oligonucleotide that was designed to be non-allele selective, we get a very potent down-regulation in both the wild-type and the mutant Huntington gene shown by the western blots here. And if you look at an allele-selective oligo on the right, you see that we have a very equally potent oligonucleotide that down-regulates the mutant gene but preserves the wild-type gene. And we've seen about a 50-fold selectivity and cell culture between these two alleles using this specific antisense oligonucleotide. And so with—in lorem, we've had, as I said, about 75% of the cases are cases where you want to inhibit production of the allele that has the mutation or the pathogenic allele and preserve the wild-type allele. And these are examples of various cases where there include a number of different ion channels that are gain-of-function mutations, so you want to preserve the wild-type ion channel. You don't want to completely knock it down but have—remove the gain-of-function channel. And then there's a number of other proteins like TDP-43 that causes—associated with ALS that, again, you don't want to completely remove TDP-43 from your brain, but you want to preserve, you know, some function associated with that. And so one of the challenges has been, because we are—we need to first identify SNPs that we could target, and I should mention that because our drug, the RNase H, works in the nucleus, we could target intronic sequence that dramatically opens up the sequence space that we can explore. The challenge has been phasing these SNPs to the mutant allele. And so working with Illumina, they've kindly donated their services. We've been able to phase about 30% of the alleles using their Illumina sequencing, but the other 70% of the cases, it hasn't been possible. And so we've—are beginning to look at alternatives, and we're working with Children Mercy as doing long-read sequencing as a strategy for this. And then our colleagues at Illumina have also developed a new technology that they're working on that they call Infinity that uses their Novaseq technology to be able to phase SNPs appropriately. So we're getting—I think we have one case that we haven't been able to phase, and that's because the mutation has about 25,000 nucleotides away from the nearest SNP, and so we just haven't been able to connect the dots on that one. But pretty much every other case we've been able to phase accurately and give us a number of, you know, different target sites that we could design antisense drugs to as a way to provide a therapy for these children. It's still a little cumbersome, and right now it's the rate-limiting step for us to begin working on developing antisense therapies for these patients. And in a few cases, these are patients with really severe disease that's rapidly progressing, so time is of the essence. And so I still, you know, think that there's a need to further optimize both types of technology so that we can get the patient DNA to the sequencing lab and then get returned on the sequencing information to get the drug there. So with that, I'll stop and just acknowledge my many colleagues at Ionis that have worked with me for many years on developing antisense technology. The NLORM Foundation, Stan Crook, who is the CEO of Ionis, is now the CEO of the foundation and is really driving the NLORM Foundation and colleagues there. And then our colleagues at Illumina, Ryan Taft and Ali Crawford, and then at Children's Mercy, Emily and Tommy have been very gracious in donating time and effort to help phase some of the patients that we're taking on. And with that, I'll stop. Thank you. All right, thank you to all three of you and to our previous three speakers. We have plenty of time for questions, and I think, you know, please ask questions and discussion. We have a couple online, but are there any from the audience? I'm back again. Great. You asked good questions, that's great. I'm happy to be shot down on this. My question is, the chromatin architecture is kind of an average, so can you talk about your pre-analytical considerations for that, and is this better suited to a single-cell analysis? That's a great question. So definitely how we traditionally think of chromatin architecture is an average, because usually when we're thinking about ATAC-seq, DNAseq, things like that, we're putting in at least 10,000, if not million, cells into that, where we're getting sort of an aggregate picture of what's going on. When we're looking at this on a single-molecule level, it's no longer an average. On a per-molecule, we're actually looking at where the individual nucleosomes and transcription factors, polymerases, you name it, are bound along that single molecule, which you can then sort of create an average at the end of the day. But the other thing you have to think about is, like, from a gene regulatory perspective, you know, that promoter doesn't necessarily care what's going on three cells over at that same promoter. It cares what's going on on that single molecule of DNA, where those nucleosomes are actually bound, where the transcription factors, and also whether or not that enhancer happens to be accessible on that same molecule of DNA. So getting it to the single-molecule approach, you really gain a lot more insight in terms of what's actually going on at the promoter, which is the sort of truly sort of functional unit when you're thinking about the output of gene regulation. And methods like what Andrew's talking about are going to be really important. You know, we're playing with cell lines to develop the methods. But then when we get to clinical samples, to heterogeneous samples, where the epigenome is heterogeneous, as I was sort of alluding to, these kinds of methods become incredibly powerful because, as you're saying, like single-cell, but each molecule will look a little different. And they don't care what the other cells are doing. And from a single-cell perspective, so these methods are all theoretically possible to combine with single cells, but I think that from a practicality perspective right now, they're just not that practical to do it, just because of sequencing costs. So essentially, like, you could do a single-cell, you know, nanonome or FibroSeq or anything like that. It's definitely, there's no technical reason why you can't do it. You're just going to be having to put a lot amount of sequencing to get in there to actually get enough coverage from each cell, because essentially what you'll end up having is sort of like, you know, maybe five reads or a hundred reads from each cell that are sort of sporadically across all the genome, unless you get really deep coverage sequencing. I would actually suggest that the best methods often are orthogonal approaches. So I would suggest something like 10X's ATAC-Seq single-cell method combined with something like FibroSeq or nanonome, and then you're looking at the problem from two different angles. That's going to reveal way more than looking from either by itself. Okay. Thank you. Yeah. I have a question. I'm building on that. I think the application of long-range sequencing for things like structural variance and single-nucleotide variance for diagnosis is obvious to me and I think to most of us here, but can you speculate on where you think some of the approaches that you're using to look at the epigenome are going to be most impactful in the diagnostic setting, if at all? In the diagnostic setting. I mean, I think the one thing I would just highlight is that you're already spending the cost on sequencing. In doing these sorts of spray painting, the stenciling with nanonome or FibroSeq, it's essentially a prep issue where you, rather than just going directly from blood to DNA, you have one sort of step in the middle, which is you just take your, you isolate nuclei, you permeabilize them, which takes around three minutes to do, you add in an enzyme, and then you go to DNA. So it's like, you know, it's definitely a processing step that has to occur, but it's something that can easily be done by other people. And I think that the critical thing which I'm trying to say with this is that if you are going to spend the money on sequencing and if you can add this sort of, you know, step to the processing and you get all this additional information, this essentially now enables you to have an orthogonal, potentially, methodology at the exact same time that gives you an ability to interpret a variant. So you're still going to be going the same bioinformatic approaches where you're probably going to look for, is there a variant in a known regulatory element, is there a structural variant, what's going on, you name it, and then you layer on top of that, like we've seen earlier with the CpG methylation, where now you can say, great, I see this variant, I see altered CpG methylation, this is an orthogonal dataset from the exact same sample which gives you some sort of validity or potential validity in terms of whether or not that is truly something that could be impactful or not. And I think the same thing is with the chromatin, is this is just an orthogonal dataset which you're collecting at the exact same time, which gives you that additional potential validity in terms of understanding or interpreting a given genetic variant which you're looking at. I think going after, you know, sort of epigenetic variants, whether or not it's not necessarily something that is truly genetic but something where you're just sort of noticing the chromatin architecture is altered, that's going to be a little bit more challenging. I think that that really, the challenge with that has been alluded to with CpG methylation is you need to know you're normal. You have to have good understanding of what's normal and also the epigenome changes based on which cells you're in. And so you not only need to know you're normal, you need to know you're normal for that sample you're going after to really say definitively that that is the case. But I think that from a sort of medical utility, it's really going to be in terms of the interpretability of variants. Yeah. That's great. So I would add to that two things. One, I agree with Andrew completely. Like the epigenome can be free if you're just looking at methylation on either soon to come PacBio or Oxford Nanopore, albeit some computational work. And chromatin is a little bit of add. You know, you do a little bit of enzyme treatment, you get that too. And then I would point back to something Danny talked about in an earlier talk where, you know, when you have that data, you can interrogate the methylation. You can say things about like structural variants and methylation that are happening at the same time. All that information can feed into sort of your diagnosis. And then the final point I'll make is sometimes the methylation or the epigenetic state that we see can inform you about which genes are actually being expressed without you having to do a separate run of RNA-seq. I'm not saying that's definitive, but it can hint to you about things to look at, especially things where the genes are very homologous, like a segmentedly duplicated genes. That's great. Question from the audience. Hi. My name is John Vanderbilt. Mine's about antisense oligos. And in the UDN, we have, of course, a cohort of patients who have rare and sometimes new diseases. So I'm interested, are the ones that might be considered as candidates by N. lorium gain of function, loss of function, or families where there are previously affected sibs that might be used as controls in natural history studies? The majority are either mis-splicing that results in a loss of function, and they tend to be de novo, or gain-of-function mutations. Again, the majority, I think, are de novo mutations. I would say half of the cases that we have submitted to us are through UDN, that patients were identified through the UDN, and the investigator worked with us to submit the cases. Great. Okay. Thanks. Sure. A couple online questions as well. One interesting question about ASO is related to how does the specificity change, or does it, as the ASO degrades? And what effect does that have on the expression of the target, of non-target DNAs? Or is this prevented with the modifications that you talked about? The modifications largely stabilize the oligonucleotide, but for the ones that I've mentioned that work through RNSH, there's a central DNA component into it, in the middle of it. So we put the sugar modifications on the end, that DNA is susceptible to nucleases that degrade the DNA strand, so you end up with short fragments that really are too short to be pharmacologically active. Like for the SMA drug, we had to look at a variety of different metabolites, because that's a uniformly modified oligonucleotide. And we had to look at theoretical metabolites that came from that for a potential for off-target effects. So it's something that we look at, but so far it hasn't been an issue. Thankfully, right? And then there's also a question about how is the FDA looking at these N-of-1 type therapies as far as the approval process? And I know a lot of people are working on this, but... Yeah, so I think you said something that's really important. So we're not trying to get drug approved. We're not doing, quote, placebo-controlled clinical trials. What we're trying to do is provide a treatment that would help the patient. And so the FDA's been very good at partnering with us on that exercise. They've put through a number of position pieces to give guidance on how to approach these N-of-1 cases. Really focusing on the talks, the manufacturing, some of the clinical endpoints that you might want to consider trying to capture to show that the patient's benefiting from the drug or not. But it's really not trying to get it approved, it's really just providing a potential therapy to the patient. Got it. And then one last quick question, related question about do you have information about the half-life of these in vivo, the ASOs? Yeah, in the CNS, the drugs have a relatively long half-life where we can dose quarterly to, in some cases, every six months. Depending on the severity of the disease, we maybe go through a loading paradigm to get to a steady state quicker. But we're really looking at quarterly dosing for most of our drugs going forward. Great. Question from the audience? I have a question on ASO Honey Ready from the Medical College of Wisconsin. So you talked about them being very specific to the DNA sequences. It works for single gene disorders, I'm very naive to this field, so trying to understand how it would work if there are multiple genes involved, because one of your slides talked about autism and there are quite a few genes involved in that. We would basically subset autism into what the causative gene is that's been identified. And so for autism, there's a couple dozen different candidate genes. Some are more common than others, and so as a company, that's what we would focus on for enlorum. We would focus on the rare mutations. So we have a couple of enlorum cases that are autism where it's a single gene for that patient. Are there any scenarios where you've had more than one gene, and how would you look at it in that instance? Yeah. So in theory, you could. There have been a couple of clinical trials that I'm aware of where people have taken two different or three different antisense drugs, mixed it together, and delivered it to patients. So unfortunately, they weren't successful for treating the disease, but there's a precedent for being able to do that. Thank you. Another question from the audience, and we've got several online as well. Sure. Ben Wiltshire, Nationwide Children's Hospital in Columbus. Again, an ASO question. How long does it take to go from a target to an agent that you could deliver for an NF1? I ask mostly in the context of fetal exome and genome are rapidly approaching, and particularly for neurodevelopmental disorders, you might imagine that delivery of therapy in utero would have better effect? Yeah. So it's in part because of the tox requirements. So the FDA still requires us to do tox studies, and a three-month tox study takes six months to conduct by the time you get final reports, but I would say the shortest you would expect is a year, and it can go just depending on whether the science works for you on the particular target. It could be longer than that, but I would say a year is probably a very optimistic best case scenario. So longer than nine months. Yeah, unfortunately. Right now, yeah. As we get more comfortable with the technology, my hope is that we can skip some of these steps, but we're not there yet. I suppose, would there be potential for some, you know, you talked about designing them to SNPs in the, if you knew the phasing of your pathogenic variant with a common SNP you've already developed an ASO for, would that be a potential? Yeah, if we've already developed an ASO to that specific SNP, that would clearly be available immediately for the patient. I think that's a really interesting approach. Our partner, Biogen, is beginning to explore in utero therapy for Neuson-Earson. It's still really early exploratory work, but our colleague Charlotte Sumner's been doing some work in mice, and we're trying to translate that now. Great. All right, more ASO questions, but we'll come back to that. We're going to go back to, I know you didn't talk a lot about the long-read RNA-seq, but it's an option to expand here. Someone's asking about, yeah, comment on the coverage of RNA sequencing and if you can get differential gene expression using long-read RNA-seq. Yeah, so that's a great question. The coverage depends on how you approach it. I would say that from a good run on, like, Promethean, I can get like 100 million cDNA reads, and if I want to do direct RNA sequencing, I can get more like 20 million direct RNA sequencing reads. Now, those are full-length reads, but then if you think about this compared to, say, Illumina Novaseq, where you're getting, you know, billions or trillions or billions of reads, then it depends on what you're targeting. If you're doing, like, three-prime and sequence tagging on Illumina, then you're going to get way more gene-level representation, but if you want to get a transcript isoforms, you can get it, like, 100 million reads from a good Promethean cDNA run. But throughput is somewhat of an issue. I would remark that another long-read technology, PacBio, there's a recent paper, the Broad Institute, where they ligated together cDNAs, because cDNAs are short, from my perspective. They're, like, 5 kb, you know, 3 kb. Fully-spliced transcripts aren't that long, usually. And so if you stick them together, the length of the read doesn't matter, and then you can get a lot of transcripts together and then just later split it apart computationally. So that's also an approach that's being tried, is sort of ligating molecules together to improve your counting, to improve your numbers of molecules yielded. I think that answers the question of sort of coverage? Yeah, I think so. There's also a question for either of you on using the telomere-to-telomere CHM13 reference to align long reads to the SM1 and SM2 genes, but I suppose that could be for anything. A very specific question. Yeah, this is a very specific question, but I think, I guess, maybe some comments on just using the T-to-T genome compared to the current reference, what the differences might be. I mean, the T-to-T genome is better, it seems to be more representative of the VACR copy numbers than, say, than HG38, certainly. It allows us to align, you know, outside of, like, the specific, as Andrew was pointing out, outside of the specific centromeric and telomeric regions, alignment works really well, even the segmentally duplicated genes and things like that. It doesn't preclude the need for a more diverse set of samples to generate sort of human genomes, and the Human Pan-Genome Reference Consortium is working towards this goal, because CHM13 is from a very specific genetic origin, and making sure that we're getting representatives of different human superpopulations involved is going to be important so we have an accurate representation so we can align, so we can look for sequences that may be missing from specific, you know, people, and we can get them from a more diverse platform. That said, this is still a far better tool to be used than the older genomes in terms of mapping efficiency, I would say. Male Speaker Yeah, and I completely agree with that, and I'd also just add that, you know, segmental duplications, which SMN1 is part of, you know, that is, those are one of those areas where we're seeing the most diversity, when you actually look at the Human Pan-Genome Research Consortium and the studies that are coming out of that. So these are areas where, you know, yes, you can use long-read sequencing data to map to that with the TTT or CHM13 genome, but I really do think that these sort of additional resources, which will be coming online hopefully within the next year or two, will be even sort of improved in terms of improving that. The one thing I will also mention is not all segmental duplications are made equally in terms of your ability to map, even using long reads. So like, even using long reads, you know, with PacBioSequencing, you may not be able to approach some of the really, you know, highly homogenous segmental duplications, you know, that, you know, require your ability to have reads that essentially span the entirety of that segmental duplication and actually have unique sequence on each side of it to give you your genomic position in the genome. So that's where ultra-long nanopore comes in, but on the flip side, there are some locations where it's only different by, like, a couple of snips, and nanopore is not going to save you because of its error rate, and then you want PacBioHiFi. So I'm sorry that this isn't a one-size-fits-all for everybody, but there's right tools for the right job. I will say, though, again, CHM13, very nice genome. It's going to – it's much better, especially if you're looking – trying to look at some of these questions, because HE38 sometimes was misassembled or had gaps at regions which were interesting and relevant clinically. Back to Dr. Bennett. There's an interesting question here about the potential of antisense technology to treat viral diseases. Yeah. One of the approved antisense drugs is for treatment of CMV retinitis. It's commercially – it's no longer relevant because the disease doesn't really exist, but that was the first. There's a couple of antisense drugs being developed for hepatitis B, and so I think in principle you can. It's just until COVID, everybody was sort of ignoring viruses largely. Yeah, so the second part of that question was how about for COVID? In principle, it could. Yeah. Great. Any other questions from the audience? There's a couple more here. Well, this is a question you probably get a lot. Is there an estimated cost or target cost for antisense treatment in the future for N-of-1 or for potentially for more common diseases that have a target? Yeah, for the N-of-1, we're providing the antisense drug free of charge to the patient, so it really is a charitable contribution. I don't set costs, so it's not something that I really get involved with. I'll let other people figure out what to charge for the drugs. Got it. Can I ask you a question? Yeah. I was really like the work that you guys were doing on DNA methylation and chromatin structure. The question I had is that you're using peripheral cells, I assume PBLs are for most of that, and that a lot of the biology could be tissue specific biology. How do you reconcile those two that you're not going to do a brain biopsy in somebody to look at chromatin structure, or at least easily? That's the fun part about being at Hopkins, is that I get access to lots of tissue samples. Many of these methodologies are applicable either first you do a nuclei isolation, then you can treat to probe chromatin or look at methylation, but you're absolutely correct. Epigenome, transcriptome too, you really often want to do tissue specific. You want to answer tissue specific questions, so then we're going after the same tissue. I did show in my talk, we did have a breast normal and tumor sample. We do explore using tissue samples. Brain's a little harder to get. Blood's the easiest. It's also the easiest to work with, so it's a good first test for methodologies, I mean for the methods we found, but we do try and work with solid tissues a lot as well. We've shown that we can use tissues. We've actually done this on brain samples, and it works actually quite nicely. You're absolutely right. Tissue specific epigenetics and transcriptomics for that matter is critical. I don't know that I have a specific question, but one thought that came to mind about using when I asked the question about clinical utility of some of these approaches is the potential for using some of this information as biomarkers for disease. So for example, we talked about the epi signature earlier and methylation patterns in the disease that may or may not be causative, who knows, but we're thinking about could you use those as a way to monitor response to treatment, for example, especially if you had a gene targeted therapy that's really getting at the disease, but I don't know if you have any thoughts about whether or not any of this would be useful for thinking about biomarkers. I always think about circulating tumor DNA when I think about biomarkers because, I mean, cancer screening tests from relatively non-invasive gang blood is always a thing, but oftentimes the DNA, circulating tumor DNA is short. There have been some papers suggesting that you can use methylation to predict cancer incidence from blood, but then the molecule's short, so why are we necessarily doing, trying to use long read on it? I mean, there's an argument to be made that, oh, but I could also have this little USB stick, but then you need to validate a clinical assay, as everybody in this room knows, and so if you want to develop something like that, it might be better either on a microarray or even on Illumina whole genome bisulfite sequencing to demonstrate that it's applicable. Maybe in the long run, there's reasons to do this, but most of the circulating tumor DNA I see is pretty short. That's what I think of. And I think we probably got at this with my question, but someone asked specifically about any known applications for M6 RNA sequencing by long reads and inference back to potential DNA pathogenic changes in the germline. Like methyl 6-adenine in RNA and then going back to the germline, that's the question? I think using the M6A approach for long read RNA-seq and whether or not that'd be useful for... Oh. Well, I think one of the challenges with M6A is actually an endogenous methylation in RNA, so you would have to use a different enzyme to actually modify the RNA that was actually not endogenous. Since there are so many different RNA modifications out there, I think that's going to be one of the big challenges you're going to run into. And also the fact that there's so many RNA modifications, you're now going to have to distinguish what is a... This base is already modified. You can't modify it again, or if you modify it again, that may cause an unclear signal change either with Nanopore or with PacBio. It's definitely probably possible, but I think it's just a little bit more work needs to be done along those lines. You might have to do two runs. You might have to do a run of unmodified versus modified and then compare the two to really get at that. And then another very technical question, but is there a big difference in the time of the pause of the polymerase between different methylation moieties on a single nucleotide? And are there methylation changes that cannot be differentiated? It's a very good question. There's a huge difference. So this is actually one of the big things that's been known as... So M6A causes a very distinguishable pause when actually the polymerase hits the adenine and doesn't really impact the neighboring bases around it in terms of its pause state. In contrast, 5MC, which is really sort of quite readily detectable with ONT technology, has traditionally been very challenging to detect with PacBio because it actually doesn't really impact the polymerase or cause a pause at the polymerase as the polymerase is sequencing that specific base. Rather, it causes a subtle change in the pause of the polymerase at the neighboring bases where if you aggregate those information together, you can then subsequently compute out whether or not that specific base has been modified. But yeah, it definitely makes it quite different. So traditionally, I always sort of said, 5MC, go ONT, 6MA, go with PacBio, and sort of there's no overlap. But I think that that's starting to break down, and I'm glad to see that that's starting to break down in both directions. Maybe we'll get to one-stop shopping someday. There's other modifications out there as well. These are not the end-all, be-all. My hope is that over the next couple of years that we'll be able to be talking about not just these modifications in terms of probing the epigenome. Rich Roberts at NEB will tell you a lot about modifications of bacteria, because restriction modification systems is their jam at NEB. Yeah. Cool. All right. Well, the last question is getting back to some ethical considerations about the amount of data being generated from long-read sequencing, for example, and how to think about returning results. This is, again, in the clinical setting. So a question about long-reads with respect to telomeres, for example, and telomere length and aging, and those types of things that we can find with long-read sequencing. Have you thought about that and clinical implications? And then the last one is about, for antisense drugs, any thoughts about implications of harmful drug design? I'm not sure exactly what's meant by that, but we'll let you take that. I think Emily did a good job sort of describing this earlier. She did. I think it's important to really ask, what are you doing this test for? What did you consent the patient into in terms of the protocol? And what is the actual utility of this variant, and is it even a variant that, in a normal clinical lab, they would report out? In terms of telomere length studies, we do order those clinically for certain individuals for very specific clinical conditions where we have good control sets that have been developed using sort of standard telomere fish-type assays. If there's not a good control set for doing telomere length assays for PacBio or Nanoports, probably those are computable. Like, you know, you can sort of move one to the other, but we don't know that yet. So I would not necessarily return anything I was seeing right now unless I had that good. And also, I had a clear indication that this is actually pertinent from that patient, and we consented them into the right protocol that enabled that. There's a lot of basic science to be done on telomere length biology, and before we know that with confidence, you've got to be careful. And the other thing I'd also mention is most telomere methodologies that exist clinically for measuring telomere length are all going after this TTAGG sequence. So all these non-TTAGG sequences we're seeing are actually not even considered to be part of the telomere length traditionally, because they're not actually—if you're using a flow fish-type signal, those actually wouldn't hybridize to the probe and would not actually give you that signal to come out there. So there's a lot that I think needs to be done now as we're starting to think about telomeres as not just a sort of fluorescent signal that binds to an oligo. I think there was a question about oligotox. Is that basically— Harmful drug design was the phrase used, but I think, you know, in the context of drug design— It's just—it's something that we've, over the 30 years, have taken into consideration and do a lot of preclinical testing in both mice, rats, and non-human primates before we get to patients. So it's something we've tried to eliminate as a potential risk to a patient. But these are novel drugs, and so there's always, you know, a potential that something comes up, and you just have to, you know, approach it clinically very carefully and in a very controlled manner. Okay, technical issue. I'm trying to get the CEU code up, if there's anybody who can help me do that. So after that, the ERA says, currently, there is no presenters, so we need to do more on the presentation. So is that— Which one? Sorry? The TPMF. Oh, no presenters with the file. Thank you. Okay. There you go, folks. All right. Any other last questions from the audience? I think this has been a really great session and generated some good discussion, good food for thought. And I like the way we came full circle at the end there to think about applications of what we are finding with long-read sequencing for potentially helping patients. So thanks to all of you for your attention and sticking with us. Get your credits here. It's great to see everybody, and I hope you enjoy the meeting. Thank you.

genomics

molecular biology

long-read sequencing

rare diseases

diagnosing

challenges

genetic variations

targeted long-range sequencing

nanopore sequencing

structural variants

DNA methylation

chromatin accessibility

FibroSeq

telomeres

antisense oligonucleotides