 Thank you, Nico, for the kind introduction. It's after lunch for me, too. But I'm here to actually, I'm just going to say it, I'm going to make a pitch for introducing genetic diversity into the kinds of gene function screens that the COMP program has pioneered. Before I get into my pitch, I want to pause and let's all look at this iconic diagram. I'm sure everybody in the room knows what it is. This is a gene-gene interaction map of essentially all the genes in the yeast genome. I was very lucky at the point in time when Charlie Boone, Brenda Andrews, and Company had done all the single knockouts in yeast to be invited to be on their advisory board. And I got to watch them wrestle with, OK, we've knocked them all out, so what? And then they came up with the idea, well, let's put them all together in all possible pairs. And clearly, that was going to be impossible. There was no way we could make all 36 million. And let me just say, creativity, corner cutting, and brute force, they did it. Now, they didn't study ribosome function or RNA processing or any of the functions that you see listed on this diagram. All they looked at is growth in permissive media. And they tracked pairwise epistasis between the genes. And the graph itself is constructed in a particular way. It's called a topological overlap map. And it's actually the same algorithm that Facebook uses to suggest friends to you. If gene A is epistatic with a bunch of genes and gene B is epistatic with a similar bunch of genes, then you draw a connection between A and B. And then doing that, miraculously, genes that are sharing the same function start to cluster together. And you can start to do this guilt by association and assign function to genes of unknown function. So just hold that thought. And we're not going to propose to knock out all the genes in the mouse two at a time. That's clearly impossible. But here's what I do want to talk about. Is the mouse a good model for human biology? And I'm just going to say mostly yes. I'd also like to just make the point that the genetic background matters. And then I'm going to propose some study designs that I think are eminently practical and show you how those study designs can lead us to better understanding of function. So as a mouse geneticist, I often encounter people who kind of want to break into the field and study their favorite trait in the mouse. And they always ask me this question, which mouse is most like the human? And the answer, obviously, is C57, Black 6, J, or N. Take your choice. And we always have the concern. I always have the concern that even though it's very important to study C57 Black mice, you're really studying one mouse and one genome. And it's a room full of Waldos. But if we could look at genetic diversity in mice, we might more accurately model the kind of diversity that we see in human populations. And if I had a laser pointer, I could show you where Waldo is, but he's in there. So we're not missing out on anything. But by bringing in lots of genetic diversity, we actually get to see many, many new and unexpected phenotypes. There are at least 20 papers in the past two years that address the question, and they go back further, too, of whether the mouse is a good model for humans. I chose this one because I like the picture, the little cartoon that I stole. And there are just two points that I want to emphasize that were made in this paper. And one is that the allele and the strain in which the target mutation will address the leels in a minute, in which the target mutation is maintained can make a big difference to the phenotype. And as a minor point, but one that's going to play into my hand here, literate controls are ideal. So a Palmer's group, many people have done this, too. But a couple of years ago, did a really heroic experiment where he took some of his favorite mutations and crossed them onto a whole panel of inbred lab strains. The design here is pretty important for us. I think this works. No, I'm going to go back. Anyway, the donor of the mutation is a heterozygote. Yep. And by crossing this heterozygote to a number of strains, he generates a whole panel in which, among other things, one has literate controls. These animals are heterozygous, and these animals are basically wild type for the mutation. Every one of them is an F1 hybrid, half C57 black 6, half something else. And you can go look at the paper yourself. These backgrounds all affect the trait, but the backgrounds also affect, in some cases, the direction in which the mutation changes the phenotype. So it's just one example, and there are many, many more. This is also true. I think I'm not a medical geneticist, but I suspect that it's pretty universal that when diseases are segregating in families, there's always variability in the expression of the disease. One explanation for that is, well, there could be many explanations. There's environmental things that could be going on, but one very likely explanation is that there are other genes segregating in these families that are modifiers. And this is a very tip of the iceberg list of human diseases that are not only known to have modifier genes, but the modifiers have been identified and the mechanism by which the modifier acts has been identified. And in my opinion, modifier genes are just an open door to therapeutics. If you have a gene dysfunction and you have another gene that modifies it, you have at least one angle to take to how do I address the primary mutation with the therapy. So human diseases have modifier genes. So I'm going to segue a little bit and talk about genetically diverse reference populations. Anybody who's heard me talk in the past 15 years has heard this ad nauseam. I apologize. But there's something that I would like to say here, and that is we attempted to make many, many collaborative cross strains. Our collaborative cross paper came out right after the comp paper. And if you look at them, they're very similar. They both were vision papers. And at the time, we had no mice. And in fact, we had no mice 10 years ago. We had no mice five years ago. We had no mice two years ago. But pretty much now, except for people at UNC who have first dibs, if you want a collaborative cross mice, you couldn't get them. So there's no way this thing could be a failure because it didn't exist. And it does exist now. There are many, many fewer strains. And we thought there would be. But along the way, we did something sneaky and we siphoned off mice. And we made an outbred population. And the outbred population is easily available. And it might seem a little scary, but I'd like to talk you through it. So in a cartoon here, we have basically three genetic resources. We have eight strains that are the founders. We have approximately 70 plus or minus 20 inbred strains that have the same genetic complements as the outbred strains. And there are infinitely many outbred strains, as many as you care to make, and they're all unique. Like I said, if this makes you nervous, just think about when you order a strain from the Jackson Lab, there's a strain number. You call up services, and they send you a box of mice, and they're all black. Or if they're Balb C's, they're all white or something. Well, if you order strain number 009376, and that's all you have to do, they're readily available, they're going to come in all flavors. And these are actually single litters of Dio mice. There's nothing special about them. They're just a lot of mice, and you can phenotype them. In fact, we did that. We've done that in a few studies now. And one of the big concerns that people raise about outbred mice is, oh, the studies can't be reproduced. They are not reproducible. This is a chart from a plot from a paper that we published on benzene exposure using Dio mice. At low exposure, there's very little DNA damage. But when we get up to 100 ppm, some of the mice show high levels of DNA damage. We did the experiment once with about 300 mice. And then six months later, we did the experiment again with 300 more mice. And in my mind, that is highly reproducible. We have reproduced exactly the same mean phenotype and the variants and their mice, and they're reproducible. So how can we use the outbred mice to do phenotype screens? There are many, many ways that you could come up with in your imagination. And I just want to show you a couple, one very simple one. There's a dominant mutation phenotype screen. Let's imagine that we have a mutation. This yellow star here, and it's on a C57 black 6n background. And I have a heterozygous breeder. And I cross it to a Dio mouse. And I get four offspring. And I cross it to another Dio mouse and another Dio mouse, say 20 or 30 or 50 of them. I'm going to get a whole population of animals that are F1 hybrids, one half of their genome being B6, and the other half being diverse. So we're putting this mutation on a diverse genetic background. And we're also getting built in littermate controls. And if I were to do this experiment, I would do it blind. I would just take the litters. I would phenotype them. When everything is said and done, I would go back and genotype them. And I would know then which ones carried the mutation and which ones didn't. It's a very clean, very simple experiment, not unlike the APOMR experiment, but much more practical logistically because the Dio mice are just a strain, right? And you can just order them. And these matings, they're going to be highly productive. One has to be concerned about sample size because outbred mice are more variable. This is a very generic sample size curve, but I want to point out that the effect size of a mutation should always be measured in standard deviation units. Standard deviations are going to be bigger in an outbred population. And an effect size of one standard deviation is a pretty good target. If you get much smaller, you have no hope. And if you get much larger, you're going to get it every time. So really, when you're evaluating sample size, you want to be looking right around here on the curves. And what you're going to see is that at a reasonable alpha level, you want to do some multiple test corrections for your multiple phenotypes. But between 25 and 50 mice is going to be per group. So you'd have to generate 50 or 100 per mutation. It's going to be an excellent sample size with sufficient power to do a good mutant screen. Now, I want to come back to this diagram from the benzene experiment. And you could think not of benzene as an environmental exposure, but if benzene were the mutation that you introduced onto the mouse, what you're going to notice is these mice that are barely or not at all exposed to benzene, they're just nothing's happening. But in the exposure group, the variance goes up. The mean goes up for sure, but the variance explodes too. And that is the key to understanding when there might be modifier genes in the background. And so you want to have an experiment that's sufficiently power to detect changes in variance. And those are a little trickier to detect. But if you look for a two-fold change in variance and you want 80% power, you're going to need 100 mice. So it's not ridiculous. It's not out of reach. You could do a screen to look for a shift in the mean trait and a slightly bigger screen using the same mice to look for a change in variance. If you see this change in variance, you're going to be pretty sure there's some modifier genes in the background. Oh, you can do this with recessives too. It takes two generations instead of one. And there are multiple ways to do it. And I'm just going to move on and say if you suspect there are modifier genes in the background, you can go after them. We did this recently with a modifier of PYMT and cancer metastases. This is a really nice effort by Nigel Crawford, who validated the genes in human cancer patients. But what I really want to talk about and to wrap up is a cross that was done in Ron Costania's lab to look at Alport syndrome. Alport is X-linked. And X-linked genes give us just another variation on how we can do these modifier crosses. So if we take an X-linked Alport mutation, which is a collagen mutation, cross it with some Dio mice, and we get a bunch of offspring. We get females that are het. And we get males that are hemizygous for the mutation. The mutation causes a disruption in the basement membrane in the kidney, which causes the kidneys to be leaky. So protein gets into the urine. It also causes inflammation and the glomeruli drop off. And so you get reduced filtration. And you see females are mildly affected and males are severely affected. And that's no surprise, because the males are hemizygous. But what we wanted to do next is take those. I see the cross here. I left it off. I'll show you this one. OK, that's where the cross went. So what we wanted to do next was validate that this gene, RFX3, which we mapped using 100 Dio mice, is actually a modifier. We kind of had a good idea, because we mapped it, but mappings never proof. So what we did is we took a B6 knockout mouse, probably a comp. More likely we made it with CRISPR, Rhondibus. And we cross it to a col-4A5 mutant. And we get four groups of mice. We get the female. All the mice have the col-4A5 mutant. We get the females with and without the modifier, the males with and without the modifier. This is an ongoing experiment. But this group here, females with the modifier, you can see them. They're living a long time. This group here, males without the modifier, these are the ones that are dropping off fast. So there's a way to validate your modifier gene candidates. We also looked at the proteinuria phenotype. In this case, we found two modifiers. I'd like to point out that these LOD scores are not earth-shattering. They're pretty weak candidates. But we have functional evidence. And we're going to go forward and validate them. These validations are ongoing, but just like the other picture. In this case, the only background we had the modifier on was an FVB background. So this is a slightly mixed background validation experiment for the modifier. So phenotype screens using single-end bread strains may miss important effects or may fail to generalize. I guess I didn't talk about that much because I think it's kind of obvious. Phenotype screens with genetic diversity are reproducible. They're simple, and they require manageable sample sizes. And if you do such a screen, you can then go on to identify modifier genes. You can detect them. You can identify the genes. And you can validate them. And they add significant new knowledge about gene function. In fact, what they're doing for you is they're prioritizing which pairwise knockouts you should make in your all-pairwise knockouts of the mouse model. Gets us a little bit closer to, we all have a little yeast envy here, right? So it gets us a little closer to being yeast geneticists. And there are many people to thank, but I just really want to acknowledge primarily Ron Costania for the Alport studies, which he was really bold to undertake. And just to make a final pitch for genetic uniformity versus genetic diversity. And I talked a long time, so we're going to save questions for later. Thank you.