 Okay, so I'm really happy to be able to talk about our experience and as Jason and Colin mentioned that, you know, that we've been working to some degree in parallel with this, thinking about how can we use CRISPR to model phenotypes directly in the mouse embryos. And we've been doing this for several years now and it's nice to see this convergence here with this, with a set of genes that we think that this would be particularly applicable for. So I thought it would be useful to talk about our experience with this and to maybe impress upon the group that it's actually quite feasible to do this and that we do indeed identify interesting and relevant phenotypes. And so this is an approach that I think is definitely translatable to what we're trying to accomplish with the hapleninsufficiency pilot. So I wanted to step back and talk about sort of our inspiration to begin this exercise. And I think this plot from Jen Posey's review just this year about the CMG, which the Centers for Mendelian Genomics Progress, illustrates the interactions that I had over the last several years with the human genetics community and really high throughput sequencing in the big programs around patient disease gene discovery has not only ramped up the rates of discovery. You can see that here with the red bars. But what you're not seeing here is the numbers of variants that are suspected to be causative for different clinical conditions but are still as yet unsolved. So there's a similar increase in the rate of variants of unknown significance or at least presumed causative alleles. And the challenge there is how do we help them move the dial towards diagnosis? How do we move past presumptive disease-causing genes to actual disease, to full diagnosis or at least solving the cases? And so the challenge there is that we have this big gap in the genotype-phenotype causality between these mutations and the phenotypes of interest, and that's because often the numbers of patients are very small, pedigree information is limited, or very little is known about these genes. So in a lot of conversations that we had with discovery partners, there was often a lot of frustration that there's a lot of limitations to the existing approaches that we've been using to solve this. We have in vitro screens, and these can provide a lot of useful information, but they often lack that biological context for a specific morphological phenotype, for instance. Non-memelian vertebrate models are great, and actually there are some good examples of using a similar F-zero screen in Xenopus from Mustafa Takaoka and others have actually adopted this approach, but they often lack certain anatomical features and being a non-memel, they're not as closely related to human. Often people consider mouse models the ideal choice, and I certainly think I'm preaching to the choir here, but they are relatively slow and expensive, and we know this, and even in the context of a high-throughput program like COMP, the comparative cost is still quite high, and more important, I think, to clinicians who are trying to close the loop on the diagnostic odyssey for their patients, the time is really what's constraining here. And so the question is, can we really rethink how we use the mouse in order to help move the needle, as it were, towards solving some of these cases for our clinical partners? And so our inspiration for maybe how to approach it really just comes from observations from our own animal model production, and that is that we frequently see in any production attempt what we presume to be homozygous, or at least highly edited embryos. And so this is just an example from Richard Berenger's paper several years ago targeting tyrosinase locus, and you can see frequently in your founder population, you see what are peer-to-be, bilally-edited mice that are fully white, and we've seen this, and we've all seen this, we've seen affected mutants, it becomes an eye-to-cook challenge sometimes when you have mice that are displaying the phenotype immediately after production of this. So it really triggered us to think that would we be able to, in these founder populations, particularly in the phenotypes that we're interested, directly model those phenotypes and gain some specific information that is sufficient to corroborate the causality of a given variant or a given gene for a phenotype. And Jason showed this plot as well, we're similarly seeing it just using attrition, just using death or lack of production as our phenotype of interest. So we approached this by basically saying, why don't we just try to model these particular human variants directly in mouse embryos and just phenotype the embryos themselves, and not go any further than that, and not have to put mice on the ground? And is this feasible? Is this something that'll work, and can we create sufficient evidence, corroborating evidence? And we knew there were some questions with this, and we weren't certain about, and I'm hoping that I can convince you that we were fortunately right in this. We weren't sure if editing efficiencies would be high enough to practically test these hypotheses. We don't really know, we weren't sure how many animals we would have to use, how many embryos would you have to screen through in order to find that evidence. And finally, we weren't certain about mosaicism. This was the biggest concern, I think, amongst people we talked to was, would this confound experimental interpretation? And I hope I can convince you that while the answer might not be universally no, or might not be universally yes, it's certainly not universally no. And we have many cases where this has worked quite well. So we wanted to start with a proof of principle experiment. We wanted to start with something we knew quite well, and see if we could recapitulate that phenotype. So we started with a gene that's been knocked out and characterized many, many times of forestonic hedgehog, and it has a very distinctive phenotype at E12.5, and you can see the whole of personcephaly that characterizes this phenotype. We took two approaches initially, and this is several years ago, so early days before R&Ps were readily available. We did either double-guide injection with Cas9 mRNA, or we actually took advantage of a mouse that expresses Cas9 maternally and used, just injected the two guides, and much to our delight, I guess, is that we were able to identify individual mutants within this group of embryos harvested that displayed the characteristic sonic hedgehog phenotype with varying degrees of efficiency. But you can see, using the Cas9 mouse, it was actually remarkably efficient, about 13 out of 21 of these embryos displayed the expected phenotype, which was very buoying from our perspective. And so, fortuitously, right at that time, we were collaborating with Cecilia Lowe at the University of Pittsburgh, who had a pedigree that, or a collaborator with Chris, who was collaborating also with Chris Gordon in France, and they had a large pedigree, or set of pedigrees of individuals who were affected by heterotaxi, and they had a candidate gene, MMP21. And at the time, despite their large pedigree, they really wanted additional corroborative animal model evidence to show that this gene was causative. And so we worked with them closely, and we selected two individual mutations from the 11 different mutations that they had identified in this gene, and engineered them directly into mouse embryos. So this was a knock-in experiment using HDR to generate the specific mutation. And as you can see, I've taken out the animation, so the punch line is right here for you to see. This was an extremely efficient process. We saw editing in a high percentage of our embryos. It depended a little bit on which experiment we did. But of those that were edited, a vast majority showed the phenotype, and the phenotype, this is just to show you what we saw. What we could see is in our edited embryos, you can see that at the top, we noted both cytosine versus in heterotaxi. Heterotaxi being defined as randomization of the C-tues, and cytosine versus being complete reversal for at least for the organ's score. And so you can see in the cytosine versus the stomach is on the wrong side, and the heart is pointed the wrong way. And we would see all mixes and matches of these. Most importantly, though, when we looked at the heart, we could see that it actually displayed the individual congenital heart defects that were comorbid with the heterotaxi, including transposition of the great artery and double outlet-right ventricle. And so we were really happy about this because we were able to, within a matter of approximately three to four months, and this is our first attempt at it, support this publication and getting it out really quickly in a fairly competitive landscape. So I'm just going to give you a couple more examples and show you some of the things I think we've learned along the way. We're also, my lab is particularly interested in craniofacial dysmorphology and craniofacial syndromes. And so we were interested in this Anderson-Towell syndrome phenotype in the gene KCNJ2 that's disrupted or in the syndrome. And it has a number of postnatal phenotypes and particularly concerning our cardiac phenotypes, but these patients are often are typically fairly affected in terms of their craniofacial development. And so we were interested in this one causative or one mutation that has been reported as causative for this gene that is proposed to be a dominant gain of function, but we were also interested in testing a few other features of our process along the way. And so we took sort of three approaches. And so as similar to the last example I showed you, we did a knock-in experiment using a single-stranded oligonucleotide donor to create that specific mutation. We generated knockouts the way as we did with Sonic Hedgehog. But we also incorporated sea-base editors as a different approach to create stop codons that we thought could be more efficient. And actually it was a proof of principle for using these sea-base editors not just to create stop codons, but of course to make specific mutations if the guide and the window of editing turned out to line up perfectly. In this case, it lined up perfectly, but the codon was not orthologous between mouse and human, so the editing actually creates a stop codon rather than the R2W transition. But we still thought this would be a good use of this or test case for testing this out. And what you can see in the table below is that we had very good editing efficiencies overall. And I want to note that this is irrespective of the type of editing, and I'll get more into that in a moment. And of those that are edited, we had really a high correspondence of phenotypes. So these edits are highly efficient, and they were edited to the degree in which a phenotype could be observed. And I'll show some of that here. So these are the embryos, E18.5-day embryos, from the different classes of editing. And you can see in both the cases of the knock-in, which was a rare event overall, the knockout in the sea-base editor, what I'm showing you here is the cleft secondary palette indicated by the yellow arrows. It's hard not to point, but I'm describing it to you and virtually pointing. So you can see the cleft secondary palette. Some interesting observations that came out of this is that the numbers of edits, as I mentioned in the previous slide, with the knock-in experiment were quite high, but the number of knock-ins was not nearly as high. So this efficiency is much lower than the knockout efficiency overall, but the knockout efficiency is incredibly high. I will also make an addendum here that these appear to be biloloc edits. But with a tip of the cap to Lydia, these, of course, could be a knock-in mutation over a larger deletion, which we know does occur at some frequency. The knockout efficiency was also quite high, as I mentioned, similar to what we saw in other examples. And the really interesting thing was the base editors showed very high biloloc base editing or base editing over large deletion. But we saw quite a few of these. What was interesting was we saw a lot of indels in this base editing as well. And in conversations with David Luzlab, it appears that the speed at which the cell divides is quite highly related to the number of indels you get from the Nikkei's function that's in the base editor. So his suggestion was to try in a slower dividing cell. And unfortunately, I can't really control that with zygotes. But I mean, I guess we could put them at a lower temperature or something like that. But biology dictates this here. But at least in theory, the approach looks to be quite efficient. And we do have some RMPs for base editors now, which is really useful. The other nice thing is within a spectrum of individual samples, you can occasionally find these low penetrance events that are actually quite interesting and important for human relevance. The patients in this syndrome actually have quite a variety of phenotypes there. The severity of their phenotype varies significantly across the patient population. And we're seeing something similar in our embryos as well. So we only had two cases where we actually had a severe phenotype of micronathia, which is what is seen in some of the individual patients in a glossia, which is loss of your tongue. And we saw this in a small number of cases. There was no, the degree of editing wasn't any higher in these particular embryos. But it's just showing you they were actually able to capture that variable expressivity that we see in some of these phenotypes. So one of the things we've started doing, and it may be worth mentioning Jason's sort of primeness, is how do we approach our genotyping? It's really important for us to understand and quantitate the genotypes that are coming out of these embryos. And we've been using an approach lately from Synthego, which is they call ICE. It's basically deconvolutes the Sanger traces. And it was a really sophisticated system to really try to interpret those Sanger traces and actually estimate the relative contributions of different alleles that are buried underneath that mixed trace. And what we found is that it's actually quite accurate. It's really good. It works most of the time for both, for knockouts. I think larger deletions and knock-ins tend to be a little more difficult for it to handle. But I think it's actually something thinking about what approaches we're going to use for our haplen and sufficient pilot. This really may suit us quite well, because the evidence thus far suggests it's quite accurate for knockouts. And I will say I also really like their knockout score approach, which is basically summarizing all the frame-shifting indels with larger indels of at least, I believe it's either 21 or, I think it's about 21 base pairs, is their limit. So anything with a bigger deletion than that or a frame-shifting indel, they give it a knockout score. So I think it's a good way to summarize and this is something we really need to think about in this pilot. It's how we communicate the specific genotype that we're seeing for each specimen. OK, I have one more example, and I really like this example, because it actually, in this case, it doesn't start with a human mutation. It starts with a mouse genetics problem. And so this is also a collaboration with Cecilia Lowe. And she had found a mutant, an ENU mutant, in her screen for cardiac dysmorphology that shows a phenotype that looks very much like, and I think they've gone very far to prove that this is a clinical model of hypoplastic left heart syndrome, which you can appreciate by the left ventricle way over on your right. The top is the control, and the bottom, I guess I can try to do this, and the bottom, you can see the left ventricle is completely occluded by the myocardium. So this is when co-morbid with other great vessel architecture defects, this would be clinically diagnostic for hypoplastic left heart syndrome. So she had an ENU mutation, but the challenge for this ENU mutation is that it was two mutations, and these two mutations were very tightly linked. And so from their perspective, they had a very hard time proving which mutation was causative or both. And so their hypothesis is that both mutations were causative. And so basically, we engineered mutations and edited the embryos directly with both guides and donors, and we were indeed able to recapitulate these mutants only when we saw both of the mutations in these embryos, and we were able to help them with this publication as well. So that was really exciting. And it illustrates one of the potential utilities of this F0 approach is that it makes these genetic interaction experiments actually feasible. If you can think of trying to create a biolink mutation, especially a linked one, using normal breeding, it's something that I think would be quite daunting. We also were able to use this particular experiment to start piloting some of alternative approaches to genotyping that we think would be perhaps more quantitative at this time. ICE wasn't an option for us. So Kevin and Lab went and was really interested in seeing if we could use a high throughput sequencing approach that would allow us to quantitate using the extreme read depth of MySeq or any of one of the Illumina platforms. And so the idea here is that you is basically based off of the metagenomics pipeline is you PCR, and then you add the barcodes to the PCR, and you're able to sequence. Now, each barcode is an embryo. So as you know, these library preps are not cheap, but this could be multiplex. And so the idea is with multiplexing, you can really expand up a number and make this a cost-effective feasible approach. And what's really nice is because of the read depth, the extreme read depth that you get, you can quantitate with a high degree of accuracy, the individual species of mutations that you have within this. And similar to the knockout score, but without the interpretation of a Sanger trace. And I think this is more accurate. And even though this seems like it's more expensive and more difficult, actually the various high throughput sequencing platforms have a lot of automation in the entire pipelines. And so really if you think about a scaled up program, you'd want something that can take advantage of these previously developed high throughput platforms. So it does require multiplexing for it to get the sample costs down to where we'd want it, but if we're doing a large project, that shouldn't be a problem. And one of the things that came out of this experiment as well that was particularly interesting and Jason alluded to this is the regional mosaicism. And so the genotype that we see, or at least the specific collection of alleles that we're able to observe, does seem to in some cases vary based on where you harvest the tissue to genotype. So it's not a uniform throughout the embryo. And this shouldn't be surprising, giving that we know that CRISPR is editing throughout the early stages of embryo division. But so this was a little concerning to us when we first did this, but really we come back to the fact that in a high number of cases, we're seeing the phenotype. And so, my personal opinion here is this is not particularly confounding in that when you're doing a screen, you have a lot of samples to work with and it's basically developing sufficient evidence to make a causal assertion. So based on all of this that we've worked on, and I think Jason showed some of this, we've kind of adopted our workflow for any of these experiments, but this is very similar to what we would do for the Appalachian pilot, for our individual human disease allele experiments. We like to start with sort of a guide testing and protocol because we do know that not all guides are equal and high efficiency is really important for the knock-in part of this process. So we typically will culture to blast and then genotype the blast. And so I can at least say from our experience that this is quite feasible. You do get dropouts from the blast genotyping, but it's almost usually manipulation based. It's not actually sequencing based. So the PCRs are pretty robust straight from blasts and none of this requires any amplification step apart from a standard PCR, which is really important because of for costs. And then we take those, once we learn which guide donor combinations are most efficient, we then can transfer. And then of course, as I showed you, we have lots of options for both phenotyping and genotyping the embryos. So I think I hope I've convinced you to some degree in this short little presentation that we have this ability to maybe think differently about working with some of our human disease colleagues. I think they're different ways you can use mouse models. You can use mouse models for discovery and potential hypothesis validation. You can also use it for mechanistic study. So I would say on the discovery side, something like the F-zero pipeline is quite feasible and actually in many cases advisable given the increase in speed and the reduction in cost available. I think there's some other advantages here. As I mentioned, the multigenic modeling would be very difficult and challenging to do through standard breeding. I think we haven't tested past two alleles, but theoretically we could add a lot more. And that would be an interesting thing to test. This is easily adaptable to alternative genetic backgrounds. We don't need an ESL for this, although we do have all those ES cells now. We also know from our experience at Jax that you can edit pretty much any genetic background with differing degrees of difficulty, which is an interesting discussion for our pilot about which strains make our core shiver, I guess at the prospect of working on. I think what is really exciting about this process is that validation of a novel variant is theoretically achievable in three to four weeks. Realistically, it's more like two to three months, but the KCNJ2 experiment I showed you was just under two months from concept to phenotypes to genotype ascertainment. So it can really be compressed and if you're in a real hurry, it can probably be compressed more. And there still are some challenges that we'd like to think about. Knock-in efficiency, not relevant for the haplosecentral project, but certainly relevant to human disease modeling. The efficiency is still not there. We haven't found the magic bullet. Base editors, though, at the first glance appear to be quite efficient and so we may be able to take advantage of that to make point mutations more readily. The challenge with those, prime editing perhaps will save this, is that they don't cover enough of the genome at this point, so you can't make any mutation of base editors at this point, but we'll see what prime editing can give us. And then the question about generalizability across different gene functions and one that's very interesting to me. A lot of the genes that we've worked on are secreted factors. They're not cell-autonomous and it makes sense that knocking down in a field, mosaically, you just have to reach a threshold where you'll begin to see a phenotype. We don't know really how this will behave with some core essential cell-autonomous functioning factors, we're not sure. So this is what the pilot's for and I think it'll be really interesting to see how that plays out. Okay, and with that I'm just going to thank the people who did the work, particularly Kevin and also Caleb Hefner who's a research assistant in the lab along with other members of my lab and Cecilia Lowe and of course funding as well. Thank you. Steve, this is great. But I just wanna make sure I'm clear. So when I was, when you were getting towards the second half of your talk coming to the end I was thinking, well this is gonna be good for confirming or valid, confirming a phenotype that you may be after in a patient, but I thought, but it's not gonna work for a discovery because there's gonna be a lot of phenotypes or phenotypes that you're gonna see in a patient that aren't evident, they're not anatomical or they're not even evident till long after birth or when it's a teenager or a young adult. So, but then at the end you said this is gonna be great for discovery. So now I'm confused. Well, I think there's no theoretical reason why this couldn't be done on postnatal adult animals. You could make your founders, phenotype your founders the way we phenotype them now. My interest, our interest is in these developmental phenotypes, it doesn't have to be restricted to that. We've not tested that, but it does start with some of our original observations that we can make albino mice quite readily and that's one phenotype. So I don't see any reason why we can't do this in adult animals and it would be a really interesting pilot to see if we can recapitulate some of our existing knockouts for which we have, we know what the phenotypes are. Can we edit them, run them through our comp pipeline to see if what we get? The question is really how you design the experiment. How many individual mutants do you need in order to feel confident in your phenotyping call? And the second most important thing, Steve Brown brought this up when Caleb gave this talk at IMGC, this is great when you know what your phenotype you're looking for and that's really helpful. We don't know what phenotypes we're looking for. So this will be one of the challenges and that might change the structure of the experiment. We might need a larger end. I can say one thing is that at least from morphological phenotypes in our unedited embryos, they typically look quite wild type. So I think seeing any phenotype will be likely, I don't see this as being a major problem, but it's something to consider as we're going through the pilot. Maybe I'll just add to that. One of the things that we do is establish the window of lethality and I think what's, while some phenotypes can have variable expressivity or we can see a lot of variation because of the mosaicism, death is pretty finite. That's true, yes. And will allow us to determine exactly where the effort and time should be spent in trying to gather more information, which is I think also a very good thing to be able to report to the community. So I'm coming at this from the perspective of this is great. So my questions are not to say. But again, I'm just trying to make sure I completely understand because when you're talking about a knockout which is not clinically relevant, generally, versus a variant which is very clinically relevant, I can see this where you are confirming a phenotype you're looking for. Like you said, when you know the phenotype it's a lot easier. And therefore from that perspective I can see where this is very, very useful for recapitulating a human variant in which you know the phenotype you're after, you do target, you got it very early. That's clear. I guess I'm still struggling with the how this can reveal and discover unknown phenotypes from a knockout when you don't know what you're looking for. Well, let's say for example, Other than, other than, I can use a, so the two things I would like to, I want to comment on the point mutations because I think that is a perception but I think our evidence so far is that a lot of these missense variants are loss of function because we're getting the same phenotype in the nulls as we are in the missense variants with the proviso that it could be missense over large deletion but we're seeing this happen. Now I know that's not always the case, there are certainly missenses that aren't loss of function but they, we are seeing frequently that the loss of function phenotypes are actually quite relevant and quite informative to whatever that human disease variant is. So that's useful, it may be a selected bias case, set of cases, but I can't be certain. The other comment, oh now I've, now I've, oh what you, when you don't know, just think of a hypothetical novel gene that you've never seen that gives rise to cleft palate when mutated. I feel like we can really reliably phenotype cleft palate. So we know we will look for these phenotypes. It's only what you'll look for. It's really limited to the same degree as the depth of phenotyping we do today. So. I also add my two cents because I'm jumping out of the chair. First of all, we see variable expressivity in the recessive lethals all the time. We report it, we have adequate numbers. We, you know, for at least describing it, the question will be what we think are adequate numbers with this degree of mosaicism and the phenotypes are clearly discernible. So it's not like we don't know what they are. We can clearly, you know, we can provide, we'll provide the imaging data sets and people will see the variation. The second point is that people, so, you know, human people who work in the clinical genetics community are looking at de novo mutations often. They see clearly definable clinical phenotypes for which there is a large degree of variability and can still, you know, assess probable gene interactions from those data sets. So it's not like we're the only ones to do this. The third point I'll say is that in the human data, what's, some of the things that are being borne out now is that there has been a bias to looking at specific phenotypes at specific times because of the way that geneticists are, okay. But, maybe I shut myself off. But I think as you look more and more at the data from early neonates, and some of those studies are coming out, there's one such study at Baylor, where when you start looking at those phenotypes in early neonates, you discover phenotypes associated with well-known disorders that were not identified related to that disorder because it looks so different from how those kids present at a year of age or two years of age. So I think it's really important that we open up our data and open up the idea that some of these things are really, as the human geneticist community gets more and more data, all of the things that we think about in the mouse are really there and we can add to this data set and we are in particular maybe modeling genovo mutations in a way that has not yet been done in the mouse. So, one sort of hot topic that was going on at ASHD was this idea of somatic mosaicism and identified rare disease cases where they were saying this is a genovo mutation because we don't see it in the parent but they sequence different tissues of the parent to be there, they sequence, so those ended up being germline but by the same thing, they're seeing mosaicism in the affected patients. Just as a result of somatic mosaicism. So this idea that in these F-zeros we might see mosaicism and stuff may actually in some instances be a better model for the disease than a whole body knockout. Yeah, certainly, especially when we're making missense and so I can step out of the compact for a moment. Some of these are dominant gain of function types of mutations so they just literally would not be modelable by any other mechanism and so it's just one of the tools in our toolbox and I would say if we're looking trying to understand mechanism obviously, this is not an efficient or particularly useful way to go about it, it's to help with the disease discovery and genotype-phenotype correlation part of the puzzle that helps our clinical colleagues move forward with diagnosis or at least moving it into the solved column. I think the other thing is about mosaicism is and I said we see a lot of mosaicism. We don't see, we see many cases where the mosaicism is extensive but they're all mutagenized alleles. So we see several different frame-shifting alleles and one thing I didn't mention here is the tools are getting better and better for designing these things. So I didn't mention that we now use this tool called in Delphi which is a sort of machine learning derived program that basically allows you to predict the outcome of a cut for a given guide will be a frame shift or it'll be an in-frame deletion and so you can optimize which guide you select based on the likelihood of getting a frame-shifting mutation. So we can really dial in this for knockouts and so that's why I'm really enthusiastic about this for the Hapel Essential Pilot because I think with all the tools we have, the efficiencies of the system, the new ways to design, I think we're really in a good position to make sure we have within a group of specimens a number of them that are sufficiently edited for us to begin to make some degree of causal assertions. So which is really where it's, I just think the way we represent the data in the database is going to be the thing that we need to think about. Are we gonna put the full spectrum? Do we include the partially edited embryos in there? How is that all going to play out? Because we do see partial edits with partial phenotypes but those are fascinating, right? I think that it's almost like a dose curve, right? Dose curves are very convincing to me, right? So you see partial phenotypes leading to more robust phenotypes and it correlates with the degree of editing that you achieve, that's better than bimodal, in my opinion. So I think how do we represent that? That's an ongoing discussion, for sure. Yeah, I mean, I think mosaicism was scary in tumors five or 10 years ago but it's not scary anymore. It's actually really interesting. In the genome, people are really interested in the role of semantic mosaicism in all biology. Right, right. It's really appealing. All right. Okay. On the knock-in, the knock-in zero, the SNP is a little more... Are you guys genotyping for those embryos that are just the HDR knock-out with the knock-out? Right. The mosaicism is actually... Yeah. While type mosaicism, where a lot of the mosaics and the embryos would be potentially knock-out with knock-in. Yeah, the knock-in definitely makes it more complicated. So we do genotype, we try to assess to what degree are the contributions of the knock-in allele versus wild type versus some sort of indel. I think we've been fortunate in that the interpretation has been relatively simple because they are SNPs, but they're loss of function. And so the knock-out phenotype is basically the same as the knock-in phenotype and we don't have... So we don't have a lot of those cases that we've tried. Yeah, I think it's a really interesting point. Both dominant genome... This case that we're talking about, both dominant genome... This case ENJ2 allele that we were trying to model is presumed to be a dominant gain of function, but everything looked the same. All the phenotypes are identical to each other except for the rare, variably expressive phenotype. So we haven't stumbled upon those cases. We are interested in doing that in proof of principle. We wanna... There are some proven and known examples out there and we can try to model those and see how well we do in a mosaic embryo. Something that's either a known hypomorphic SNP where the knock-out phenotype perhaps is early lethal and you only see the phenotype later in development or if it's some sort of dominant gain of function, which I think some of those have been modeled as well. So we can take, you know, do some more proof of principle and that's stuff that's ongoing work right now. Yep. Yep. Okay, good. So lunch break. We actually have a reasonable amount of time an hour and 15 minutes. Carolyn's scheduled to be here at two o'clock so you wanna be back for two o'clock. I was asked to draw your attention to the AMRA meeting in Taipei, which will be happening right after our spring meeting in Korea. So a couple of trips to Asia, that's not bad. And I was gonna introduce TC who's actually in the room for the moment. Sheethol's replacement. We have a new program analyst, TC, who's gonna be splitting his time between comp and SCGE, actually. But he's here. Any other? No, I think we're good. Yep. So enjoy your lunch and we'll see you at two o'clock for our forward-looking discussions.