 Good morning, everyone. So today, I tell you about the genotype phenotype map. And especially what I think is an important part of my talk today is the mutational properties of a developmental system. So first, I will give you two parts which are introductory on the genotype phenotype map and on developmental bias and mutational variance. And then we'll come back to the Volvo and then explore different sensitivity of the system of the six cells I told you about, the evolution and sensitivity to different perturbations. So first, some definition for those of you who come from far from biology, maybe. So the genotype is basically the sequence of DNA. With the phenotype, you can define a phenotype at any level you want. So it can be a level of gene expression. It can be a final morphology of the organism, so behavior. It can be the activity in the biochemical reaction or the concentration of the product. It can be anything, or the shape of a cell, or so on. And so throughout, I could, I think, genotype in blue and phenotype in orange and later, environment in green. So this is a slide I showed you yesterday. Basically, in evolution, most of evolution occurs through genetic variation. You also have evolution of the environment. But genetic variation is the predominant source of evolution in phenotype. And the relationship between variation in genotype and relation in here, I'm talking about a final phenotype, like morphology, after development, has to go through development. And the problem of the genotype to phenotype map is how do you translate this genetic variation into phenotypic variation in phenotype space, whatever the space you chose to study. And so this is not a rigorous science, the science of genotype to phenotype mapping. And some of you may take on this. I think this is a call for this. So what I'm going to show you in the next few slides are representation of the problem. It's not, in any way, a mathematical transformation. So this is a very well-known drawing by Richard Leventine in his book of 1974 on the genetic basis of evolutionary change. So here is representing a genotype space here, phenotype space here, in a very abstract manner. What's important is that when you do this representation, you can either represent one individual or here what he represents is the mean of a population during evolution. And so these arrows represent evolution. So you start with a population which has a genotype here, G1, that translates. And this is the genotype to phenotype mapping that we're concerned with into an average phenotype. Then you have a transformation T2. So this is transformation T1. Transformation T2 is the result of natural selection that may favor some phenotypes of another. So at the end, you get an average P2. And then this is the inverse transformation to T1, going back to genotype space. T4 is reproduction. So reproduction, you may have, this may cause a change in average genotype because of, for example, what's called assortative mating, the fact that some genotype like to mate with each other or disassortative mating. So there is no reason that it will stay at the same place here, and so on with cycles like this. What's missing in this representation? And I think this is really not that Levantine was not aware of it. But Levantine is an evolutionary biologist. And evolutionary biology so far has been carrying a lot about evolution in genotype space. So that's population genetics. A lot about evolution in phenotype space. That's often quantitative genetics. And as you see, Levantine is carrying about the mapping. But usually they don't go in between in the intricacies of cell or developmental biology. So representation that's now becoming more common is the following. Where you have a genotype space here, it's inverted compared to Levantine's direction. The genotype space in blue here, the output phenotype here. But you can consider in the middle some phenotypes, which may be developmental phenotypes such as position of cells or signaling pathways activities. And so the structure here is that you have an organism that's in genotype space. It's also in the part of an environmental space. And this is translated in both genotype and the environment acts in the process of phenotype construction, yielding an output phenotype. An important point when we come to genetics is the notion that all this only makes sense if we consider differential genotypes. Because basically the whole genotype to phenotype relationship is variational. So what counts is that when you have differences in genotype, so this can be natural variation of individuals in the wild. Or it can be mutants in the lab. The reasoning is the same. Or if you have change in the environment, this acts in changing the phenotype. And so this is the same for laboratory mutants. What we are looking at when we look at a knockout of a gene is the variation in phenotype that results from a variation in genotype. And I think this is the only thing important to remember about genetics is that it's really a science of variation. Even though we tend to shortcut this and say that if a gene knockout has a phenotype, the gene has a function in this. Or it's the role of this gene. This is all based. It's fine to say this to be short. But the important thing to remember is that it's a variation that we looked at. Yeah, sure. Good point. So I think we could write a book on the representation of genotype to phenotype. So here the difference is that here I'm thinking of it as an individual, not a population. And so no. I'm thinking of it as a population of individuals with the same genotype. And so they may have different phenotypes for a given genotype. But that's a good point is when we look at this representation, we did a World Journal Club in my lab on this. You have a very different ways of, I mean, what is represented differs a lot. So it's not because we're interested in genetics that we should forget another important part of parameters that act during development, which is the environment. Again, there are many parts of processes of development that we study, like volva formation, for example, which is not very sensitive to the environment. So we tend to forget it because people to do developmental genetics in the lab usually have chosen systems where the environment has not a strong effect. But in real life, environment may have a strong effect on development. And so the environment acts in many places in evolution. It acts in the construction of the phenotype. And when I say construction of the phenotype, you can equate this to development. It's just if you're with a unicellular organism, construction of the phenotype is cell biology or it can be behavior if you're interested in behavior and so on. So it's a more general term. Then the phenotype of the organism can feed backs on its environment, either because it modifies the environment or because it migrates to a new environment. And then environment, of course, is involved in the selection process by providing the context of selection. So for example, it can be a pathogen or it can be a high temperature. And finally, there might be a lateral transfer of genetic material from the environment to the genotype. But this is a relatively minor contributions at least in eukaryotes, okay? So this is a representation of the evolutionary process which really emphasizes the involvement of the environment. Now coming back to genetics, here is a representation of genotype phenotype map that emphasizes two different phenomena that I touched on yesterday, at least this one. So here what is represented are different genes and the arrows indicates that variation in these genes affects this trait one, two, three, four, five. And so in this mapping here from genes to traits, you can observe two different phenomena, one that a given genetic variation can affect two different traits. So this is called pleiotropy. And you can also observe that variation in two different genes, for example, here one and five, both affect trait one. And so this may often interact in a way where the variation in one and the variation in five are not additive. And so this is what's called epistasis in evolutionary genetics or genetic interaction. Now if you go to actual models of genotype to phenotype map, I think, I mean, if you're interested in this, I advise that you read this review by Walter Fontana on taking RNA folding as a toy model of genotype to phenotype mapping based on work he did with Schuster before. And so here if you think of RNA folding and the shape of the RNA fold, and these are not experimental shapes, it's just computer based. You can derive a lot of properties that are true for more complex genotype to phenotype mapping from using melting profiles, for example, that's plasticity. So the sensitivity to the environment of the shape of a given RNA sequence. You can get epistetic relationships. So for example, here you have a mutation that doesn't change the shape. So it's a neutral mutation. But if you add another mutation, which is the same here and here, it produces two different shapes here. So here you have an interaction between this mutation and this mutation. You can also represent what's called neutral networks. So neutral network, genotype space, here color coded, is the network of mutation. So each of these dots is a genotype that's related to the next one by one mutation. And they may not change the shape, like this one. But you can move along this, for example, if you have selection for this shape, for stabilizing this shape. And it matters whether you are here or here to know which is your nearest neighbor. So that's an important property, is that you can have completely neutral evolution here, yet this matters for future evolution of the system. Then you have more complicated representation of this map where you can, in between the genes, so this comes from the same review here, you have the genes and the traits, you can try to also introduce in the middle developmental processes here, which are all affected by the environmental context. If you're thinking in terms of modeling and parameter space, which may be your way of seeing things, you can introduce, in between the genotype space and the phenotype space, a parameter space in some, that's another representation. And the final one I'll show you as cartoons, is that one that comes from the paper I mentioned yesterday on evolution of this morphology, where you don't, you have genotype and phenotype, and then you can add one map, which is the fitness of the different phenotypes. So this is called a genotype-phenotype fitness map. Okay, so enough with these cartoons. Do you have questions on this? And we'll go to defining what is called developmental bias or mutational variance. And here I'd like to step back and ask, basically, if you look at phenotypic evolution in the living world, you can ask what are the roles in this evolution of natural selection, that's probably the first thing that comes to your mind, of historical contingency, and what I will define better, but developmental biases or constraints. So natural selection, I guess you all know what this is about. That's okay. Yeah. And so historical contingency, I think, is very important. Biology is a historical science. And because of also chance processes due to mutation and due to many other processes in biology, you do have a strong historical component in biology. And yeah, I think this cannot be overemphasized. Biology is one history on our planet. And there are a lot of contingency here. So what is this? So the idea of developmental constraints is the first expression which came, which was popularized by a paper by Stephen J. Gould and Dick Levantine in 1985, where they made a very strange, well, we're not far from San Marco, so I think it's good to talk about this. But so they talked about developmental constraints by taking the analogy with the dome of the Cathedral of San Marco in Venice. So if you have a dome that's on four pillars by architectural constraint, you end up with these four spandrels here, okay? And so the question is, so they were making fun. So the goal of these two persons was to make fun of people who think that natural selection is the main driver of phenotypic evolution. And so what's called the adaptationists, or what they call people adaptationists who always find that things are there because a natural selection put them there. So here, the reason the adaptationists would say, oh yes, this was adapted to fit the four evangelists, which actually are on the four pillars. What they say is no, the spandrels are the necessary consequence of the architecture, and then they put the four evangelists on it, which looks more reasonable also for architecture. And so they were left, and this paper, to be frank, was written by Stephen J. Gould, not by Levantine, and this actually matters a lot because Stephen J. Gould is not a geneticist at all. So this doesn't help you a lot to translate this in genetic terms. And so on the next drawing, I'm going to try to translate this in genetic terms, which is if you start from a genotype, which is the black dot, and you, in genotype space, you look at random around this genotype, you can ask what's the distribution, the probability distribution of phenotype you get at the end, okay? And here you may have direction of phenotypic space, which are extremely constrained, where you can hardly change by random mutation, and you may have others which are very easy to evolve in. So here, instead of talking only about constraints, which is a very negative word, we can also talk about bias, which I think is a more generic, yes. No, one genotype can give rise to many phenotypes. I'll get more into that later, yes, sorry. Yep, maybe it's not a logical order, but yes. So there is one dot here in genotype in black. You have already a distribution in black, but then up on mutation you have yet another distribution. Again, this is cartoonish, and this is not a mathematical formulation. And then, if you look at the evolutionary process, is once you have these accessible phenotypes, you can have selection on them, okay? So this is undrift. So stochastic drift is the fact that when you have finite population sizes, you have stochastic processes in determining what comes to the next generation, and therefore you may have evolution that's purely stochastic without inference of natural selection, and this is particularly strong in small populations. So at the end, you have what's geneticists called standing variants, standing in population. It means the genetic, the phenotypic variants that's standing in a population, okay? And so one important definition is that when you hear about these genetic variants, environmental variants, mutational variants, standing variances, these are all in phenotypic, or square of the phenotype matrix. So this is really, so here we're defining something which is called a rotational variance. So it's in square of the phenotype in terms of unit, okay? And for example, a genetic variance is also square of the phenotype. It's the proportion, it's the part of the phenotypic variance that's due to genetics. So how do you access this experimentally now? And basically there are two ways, and I will show you an example of each. One is, and they are not equivalent exactly. One is to do artificial selections in the lab, experimental evolution in the lab where natural selection is replaced basically by the experimenter selecting. And here you start from genetic variation that's in your starting population. Or you can have another method which is to start from one genotype and really explore around this genotype by mutation. So in this case here you're starting with a population that's already full of genetic polymorphisms and presumably phenotypic polymorphism. Whereas here you create mutation in OVO spontaneously, okay? And then there is an extension of this mutational variance when you are in several dimensions which is called the M matrix which has all the genetic, the mutational variance for each of the phenotype and then the covariance between phenotypes outside the diagonal. Okay, the first examples, it's beautiful, it's a butterfly that's found in Africa which is called Biciclus and Inana. And here look at the spots on the wings. So the question these authors raised was if you take the four wings, you see that in their population they had one small anterior spot and a large posterior spot. And the question they asked is there a constraint on the respective sizes of the two eye spots? Okay, so what they did was to try, this I think are Photoshopped wings but what they tried is to get, to try to select in the direction of having only the anterior one and no posterior, only the posterior and no anterior or reduce both or increase the size of both, okay? So here you're trying to see and here again, so here they started from a population of butterflies that came from the wild which had genetic variation which is in some way very representative of what occurs in natural population and they ask if there is, I mean we are going to impose artificial selection for in four directions of space. So here you have the size of the posterior eye spot and the anterior eye spot, you start from here and so each experiment runs with several parallel lines which run for example for selection for no eye spots or for two large eye spots or for dissociation of the size of the two eye spots. So you take a population which crosses naturally with each other, yeah, sorry. And so at each generation you look, so you have your population to start with, you measure the eye spot size and then you take, so you can decide, you take the 10% which are closest to this angle and you see the next generation here and then here you're going the next generation to take the 10% which are closer to here and so on, okay? And you do this in parallel lines as well, so that's why you have the deviations. So in this way basically you're testing from the original composition of the population, can you get anywhere in phenotypic space, okay? So here I think it's a question of a half full and a half empty glass. Their conclusion was that there is no constraint because you can go in the four directions. If you go look at the rate, they're different, okay? So I think there is still, so it's the problem that people have been talking about absolute constraints that it's impossible to do something. Here it's not impossible to go in this direction, it's just more difficult. Yes, excuse me, it's each generation, sorry. The size of the eye spots, no, no, the size in the next generation. So you have a distribution of, I mean here there are two but if you are just selecting for one dimension and you measure, this is your parameter that you want to increase for example, at each generation you make what's called truncation selection. So you take this one, then you seed one population and at the next generation you may have as distribution exactly this. Well maybe it's impossible developmentally to get whatever, an entire eye spot but not a posterior one. Especially given the genetic composition of the population, exactly, yeah. Yeah, so this is exactly, this is the basis, I think I have some slides on this later. It's the basis of quantitative genetics is that you may have this due to stochastic variation between individuals in which case you get exactly the same curve at the next generation. So you cannot select on it. But if part of the, so that's why the partition phenotypic variants in genetic variants, that's a problem of breeders in agriculture for plants or animals. That's exactly to know what they should take here to be able to switch their distribution to the fastest as possible. How big, what is the genetic variation? Yes, although what you can get and these are the wedding and experiments you probably know, but you perfectly can go at the end further than when you were at the beginning because you combine, even without mutation, obviously mutation occurs, but even without mutation you may get combinations of alleles which were not present in the original population through recombination, okay. So you can go further than your original population by doing this, okay. That's, sorry, because of the blowing, I don't hear you very well. So I think what you're asking is that maybe you're not getting them because they're dead or you're asking whether this population at the end if you release them in the wild they would survive. They're two different, so there are constraints that you can only get those which are alive and can reproduce. Sure, but here you're not in the wild because you're doing the artificial selection. But that's exactly, you'll see that's exactly the point later. Yes, yes. Yes, exactly. No, no, it's because of their constraining. I mean, is there experiments that design to stay on this line, right? Yes, even though probably, I mean, first they are not perfectly on the line. But yeah, so they need to compromise when they choose for the, so I think it's the ratio probably that they picked. So in which case it stays on the line, I think. So now this contrast with another experiment that the same authors did later where, so this is the experiment we just saw that's repeated actually, it's a replicate of this experiment where they could select along this diagonal and it was more difficult here. But here what they tried to do was to change the color of black versus yellow of the eye spots, this yellow ring here. And basically they could not dissociate the colors of the two eye spots they were looking at here. So it was always changing if one was gold, the other one was gold, and if one was black, the other one was black, it was very difficult to move in this direction. So here that's what you could call a developmental course train. And it's biased in this direction to change colors together. So when you think in terms of developmental biology, there is no mystery here, it's pretty logic that it's easier to change because the colors of the different eye spots are probably in part regulated in the same manner by the same gene network. This is logical, but it's just the way to prove it experimentally. And then what you can see is to compare what you got here with what actually happens in natural populations. And so what they plotted here is against for the size of the eye spots, the posterior and the anterior. The species they started with is this one, the red star, and they asked, can you find all the possible feeling of space in all directions using the different populations of a species and also the related species? And here, their conclusion was that the space was not completely filled, so especially in this direction. And therefore, the fact that you can attain it, but you don't find it in the wild, is the result of natural selection. So the reasoning here is that it's genetically by crosses you can obtain here this direction, but in the wild, it's much more difficult to get it, especially this corner. And now against, I think it's a story of, if you look quantitatively, it's not completely true, but here the point is the reasoning here of trying to compare what is in natural population to what can be obtained by artificial selection in the lab. Yes, that we don't know. So in these cases, these ice spots are probably recognition by predators. This is kind of irrelevant here, but I mean, it's highly relevant for, but yes. Yes, but here you focus on the one phenotypic trait, which is these ice spots. Yes, but here you focus on one phenotypic trait, which is these ice spots. Well taken, so completely well taken and the next section will not really answer your question but take a different approach. The question is if you are a population biologist and you go to see a population of these butterflies in the short term, this is very realistic. Now again it's the problem of evolutionary scales and here we are at a small evolutionary scale. So this is the second way of doing it, I mean both of the advantages and disadvantages and both more represent different processes in the wild. So the second approach is to do what's called mutation accumulation in the lab and you can do it using just spontaneous mutation or you can also take artificial mutagens if you want. So here and this is particularly powerful if you take organisms which are isogenic to start with. So this is the problem of fly or butterflies or mice or zebrafish or whatever is that they are never totally isogenic because they always mate male and female and very often you have what's called inbreeding depression. So the fact that you cross brothers and sisters is not very healthy and you usually have part of the genome that cannot become a mosaic because of recessive deleterious mutation. So this can work for example in unicellular organisms or especially C. elegans is an excellent organism for mutation accumulation lines or Abidopsis as well as the news which is also a sulphur. So what you do here is that you transfer at each generation you transfer a single individual that you pick at random and you see the next generation. So here you have the minimum natural selection because you don't have the population size is one you don't have competition here. So you accumulate the mutations at each generation and then you can continue up to 400 generations for example and accumulate mutation each generation. So here you're not relying on the genetic variation in the population you're relying on de novo mutation. You can also do the same experiment by just dumping mutagen on them if you want. Spontaneous mutation is very strong in terms of accumulating deleterious mutation because very often these lines actually die out after a few hundred generations. And so here I plotted you experiments such experiments that were done with a virus or deployed of yeast and what you're seeing here is the distribution of fitness of different lines after mutation one being the ancestor here and so you see it's a multimodal distribution especially you have some which are zero meaning they are lethal and then you have a given distribution and that's very important to understand in terms of evolutionary dynamics. So here fitness is defined I guess as a competition fitness in this case. I don't know our growth rate maybe. So here in C elegance what they measured was the brute size over time again fitness decreases not very surprisingly. So here what I want to make these are very old data which look a little outdated but it's for the reasoning I want to point this out. So these are data from the Denver on mitochondrial DNA sequencing. So first of all you can get at the mutation rate by sequencing the lines of the X generations and again you can use a comparison of the pattern with. So here it's a mutation pattern but it could be a phenotypic distribution pattern if you compare it with wild isolates so you sequence your mutation accumulation lines and then you also sequence different wild isolates of the species and you compare the for example the rate of transition versus transversions and what you see here is that both upon spontaneous mutation without selection you have a bias towards transitions in the ML lines and in the wild strains. So here it's a mutational bias. Now if you look at silence mutation so silent mutation is a mutation that due to the redundancy in the genetic code doesn't have an effect on the protein level and you see here that in the ML lines you tend to have the numbers are not very high in this old paper but that you tend to have at least as many silent mutation as amino acid alteration whereas in the wild you have counter selection against amino acid alteration and so here the difference in print here of mutation and mutation plus selection in the wild reveals the action of selection in the wild. So that's again the reasoning of comparing what you get just by transforming you know genotypic variation, infinotypic variation without selection and letting a natural selection act. So before I turn to the vulva just one example that's easy this is body size, body volume. What you're seeing here in the X axis is body volume and this is a histogram of distribution of parallel lines that have accumulated mutations of their volume. So in black you have the ancestor line being so duplicated for measuring size I mean replicated artificially here and then you have the distribution in the mutation accumulation lines. So from this you can derive two things is the change is the variance increase among the lines and then you can also derive a rate of going here towards the left the mean rate the mean among the lines that's moved from this position to probably this position and you have the number of generations so you can derive a rate. And at the final time looking at 200 lines this one I forgot I think it's 400 generations and maybe 100 lines of the individuals. So you accumulate let's say 400 generations and you do this in parallel in 100 lines and then for each of them you assign you measure body volume giving random number in 15 individuals. So here the data points are for given line you have a mean here which can be you know 8 here, 12 here and so on and you make the histogram. A bottleneck in which way sorry I don't answer the question you rephrase it just yes it's not a population each one is a single individual so you have a bottleneck at each generation here you transfer at each generation a single worm does this answer your question what do you call a bottleneck maybe yes so I think what you're talking about is a reduction in fitness right that's what you bottleneck right so yes you do have a strong reduction in fitness so sorry I passed on this because I'm running late but so for example here at generation 214 this is the distribution of one fitness proxy which is the rate of increase the R in a demographic model and you see that some of the lines you have a distribution among the lines obviously each line has a history of mutation so they're all different and some of them are doing pretty well but some of them are very miserable in terms of fitness sorry these are jam line mutations yeah folk of course yeah other questions on this okay so here I'm my second part is related so I'll come back to this here but now I just like to come back to the Volvo and look at the six cells now we're not going to forget the the entire one p3p and look at properties of variation of sensitivity to different perturbations so first we look at the actual evolution I told you for the five posterior ones they don't change but we'll see for the entire one and then measure that sensitivity to different perturbations and so everyone has a definition as its own is our definition of robustness so for me historically because I'm very much in the quantitative genetics community as well that this community doesn't use the word robustness it's just measures variance phenotypic variance and says it's low and the word that was used was more canalization in the sense that it's a reduction of variance I think robustness comes from physicists as far as I know maybe in chemistry would talk about buffering the problem I have with the use of the word robustness is that I find it I understand that it can be used in a very generic manner but I think it makes it extremely confusing so I tend to either avoid it or now use it exactly as saying insensitivity to a given parameter and so it's basically that each time so it's so it can be robustness to stochastic variation to a given and raw mental variation or a given genetic variation but again within this category it's useful to tell which ones because there is there are many cases where I think people are shifting in the same sentence about the concept of robustness to for example environmental variation and genetic variation or a given genetic variation and genetic variation in general and this is not helping the literature at the moment I think okay so I'll try to to sage time the robustness of which phenotype to which perturbation and if possible measure it how much it's not infinite either so if you look at the developmental system you have yes sorry now he's here you have genetic variation in the population that's a source of variation and raw mental variation you have parental effects which in turn can be genetic or environmental and then you have stochastic noise that you can only measure if you have an isogenic population in a given environment you can also call it micro environmental variation it's basically what you cannot control practically if you look at development it's also important to make the distinction between variants and invariants at different levels in this genotype to phenotype mapping so you may have variation in an intermediate developmental phenotype but no variation in final phenotype okay or you may have this case where you have variation in development that results in a variation in the output so coming back to to see elegance so this is this tree I showed you yesterday of birth of all the cells during development and so the vulva is this one and here I'm going to concentrate on this cell on the entire side called p3p and in reality John Solston always said that see elegance has 959 somatic cells and he didn't miss what I'm going to tell you but what he showed is that actually in about these cells only divided in about 45% of the animals in the strain he was using and in the other half of the animals it doesn't divide so actually half of the animals have 958 cells so this is a typically a case of stochastic variation here because you have an isogenic population in a given environment so considering now these six cells we are going to look and so I represent this no division with a gray shade here we can look at their properties of variation in evolution so once natural selection has acted as well and sensitivity to different perturbations so I told you yesterday this is totally in variance this self-made pattern for these five cells now if you look at the division frequency of this cell it's extremely labile evolutionary so it's true inside the species species so intraspecific manner inside see elegance if you take different wild isolates of see elegance the the the so here it's a frequency so histogram of the frequency see elegance is in blue you see that some wild isolates of see elegance have very low rate of division and some have a higher one okay and if you look at different species close to see elegance they have different distributions and this one they tend to be at 100% okay so this is really there is action going on here in evolution in the synoditis genus what's interesting to me and and this is something we'd like to look at in the future is that here this was the synoditis genus here p3p varies and p4p and p8p don't vary if you look at another genus here that we study in the lab O'Shaiaz here p3p is invariant and p4p and p8p vary so here it's a case of what you could call an evolution of the evolutionary tenancy so within one group of organisms it's common also in paleontology that you find that there is a tendency of one phenotype to evolve and sometimes the tenancy is directional or sometimes not and so here we have this but in addition we have an evolution of this tenancy because different genera have different tenancies so our goal is to try and explain these properties and so far we concentrated on the synoditis genus in the future we'd like to compare the two genera so sensitivity to noise and now environmental variation so what we did and this is the work of a former postdoc in my lab Christian Brenda what he did was to devise specifically six environments in the lab three temperatures liquid culture starving the animals or letting them go through another developmental stage the dower and measure for the six cells on thousands of individuals in each conditions what kind of variants he would get in the pattern and for sure the most sensitive is p3p so it's not only sensitive to stochastic noise it's also the frequency depends a lot on the environment so here you're looking at for example in the end to background the variation depending on whether you give them food or no food and for example this is another species he breaks a you see the actual inverse thing so if you starve them you have a higher division here if you starve them you have a lower division so the environment effect also varies with the wild genotype you have an interaction between the two so again p3p is sensitive to noise and to the environment now if you look at the five other cells by looking at thousands of animals Christian could see variations which you can call error but it's a little anthropomorphic but where you didn't get this exact pattern for p4p to p8p so you got cases very very few cases where the pattern was incomplete this we actually never saw it's a thought drawing or the this cell was only partially developing to the vulva and this is the proportion so here you have thousands individuals per environment the colors are replicates on different days and you see here the percentage in the different environment not much is happening now you have other types of variants especially centering of the pattern on p5p instead of p6p or on 7 instead of 6 missing cells hyper induction here with with an additional primary which is not harm as this is very harmful this is not and so on and the remarkable and the fact here that p4p and p8p behave like p3p and you get here in some environment and some genetic background a very specific for example here you have five percent in this genetic background upon starvation where you have centering on p5p instead of p6p this does not occur in this other way to isolate now for the fact that these cells don't divide this occurs pretty often 10% of the cases in this background and much less in that one so here it actually shows that there is some cryptic evolution I mean that's that you can reveal here in these environments between the while isolates and that the system is not infinitely robust right so these variants here where you have a miss centering of the pattern they actually occur because the anchor cell so this cell I showed you yesterday sends a gf signal to p6p is not always located just above p6p so what you're seeing here is in two in these two species the position each dot is a position in one individual of the anchor cell compared to p5p and p7p and you see that in and to it tends to be displaced towards the anterior side and it brings in here towards the posterior side those explaining these variants yeah so to sum up what I showed you so far you really have a distinction among the six cells in p3p being sensitive to noise and environmental variation and varying in evolution and these ones which are relatively robust to some degree to noise environmental variation and that don't that have any stasis means in evolution that it stays the same they don't change in the genus so now what happens if you submit the system to random mutation so here it's genetic variation in the form of spontaneous mutation so we do the same experiment as before using mutation accumulation lines we start with a given genotype and actually in this case we're going to start from four different genotypes and as say the distribution of phenotypes after mutation accumulation so these were lines that were made by Charlie bear in a Michael Lynch slam it took two wild isolate of C brick day two of C elegans and then replicated the hundred lines for 250 generations of mutation accumulation the great advantage of C elegans actually is that you can freeze the lines so you can also even have a fossil record of the experiment but we are said after 250 generations a Christian Brandler has said all the different the six cells to see what had happened in average and in variance among the line so what is represented in here is the mutational variance so the increase in variance in line means for different cells so here you have four p3p p4p p8p and basically this are the the variance that's really affect the self-heat pattern like centering and so on and what you see is that p3p is that evolving fastest much more than p4p and p8p and irrespective of the starting isolate you can also express it in the in the in the displace displacement of the mean from the control to the mean of the lines and you see here that p3p really evolves much faster than the other ones so here you have a system with six cells which come from equivalent points in embryogenesis and that have a very different sensitivity to run the mutation in terms of the fate they acquire in development you have six five cells which are not extremely sensitive I mean on a log scale you would actually see something here they are variance but one cell is by far much more sensitive and again here and here because we have some variance here you can compare the distribution of these variance with the wild and derive the fact that despite the fact that we didn't see a lot of variance here which have a defective pattern or a centering variation there are even more rare if you look at wild isolates and so you can infer that there is selection against these variance in the wild while the distribution in the wild among the wild isolates is very similar as the distribution after mutation accumulation so it's consistent it's not showing it but it's consistent with the idea that this is neutral that this evolution here is neutral so yes no this is not it's the same slide basically showing that this is under stabilizing selection and that's this may be neutral so in conclusion to this mutational impact on volva development what it shows that the system of the five cells is under stabilizing selection because it degrades in mutation accumulation lines although slowly most importantly I think is that different volva traits vary at very different rates and the most viable a cell fit is that of p3p and this I will show later and I didn't show you the data but if we compare the four different starting genotypes there is actually a signal that the mutational variance is different between them so that the mutational variance evolves and so our goal now is to look at the mutational variance in the other genus Oshaeus to see whether this may explain the evolutionary trends. The next question we asked on this line is what causes the difference in mutational variance among PNP cells and for this we have two approaches one is to identify the genetic mutation that caused the change in phenotype and the second is a more developmental biology approach. So this genomic approach was the work of Fabrice Benard who recently left the lab so he asked the fact that we have more higher mutational variance of p3p fate is that due to one or few highly mutable loci for example so micro satellite repeats a repeat in DNA like AC AC AC AC and this is easier to mutate and there are cases that have been shown I think shall be here an example in dogs where you have in a transcription factor that's important for morphology of the head of the dog that is actually your fast evolution perhaps in part due to variations in this micro satellite and The other hypothesis is that you just have many genes that when mutated cause P3P to change fate. So I will not enter I think in the details of the genomics because this is not the point. So basically we sequence the lines. So these lines have a lot of mutations in them. So sequencing them is not enough to find the causative mutations. We also had to map genetically. So to use recommendation mapping in the lines to identify where the mutations that were causal in changing of these P3P phenotypes were. And first of all we found them on basically all different chromosomes or chromosome arms here showing that they could not be a single locus that was mutated like crazy, okay. And we also identified the mutations. So so far we have this five. And one of them is something we expected. I explain later why. The other ones are very general regulators of transcription translation. And here they are completely unannotated. So we have no idea. It's not something we expected from developmental biology would affect P3P. But it goes with the idea that you indeed have a very large mutational target. So a large number of loci that when mutated affect the development of this cell. So of course it's a jamline mutation. So the DNA is the same in all cells, right. So it affects the fate of that cell. Or not, yeah, that's okay. It affects the DNA inside these other cells. But the change in the DNA doesn't affect the cell fate. A lot of changes in DNA. Yes, yes, for sure, yeah. It's mostly the position in space. But I'll show you just in now why there are difference due to position in space. I'm saying they have no phenotype, for sure. Yes, not in the six cells, in the five other cells. I'm not saying they have no effect elsewhere. There might be other, I mean, like, sorry, body size is something which is affected by a lot of things, for example. So I know that at least one of them has an effect on body size. You can, no, when the environment acts on P3P fate, it's directly not through mutation. Can you rephrase your question, sorry? Yes. Okay, let's come to this, yeah. So it's the developmental biology which is behind this. So here we're, I mean, so we basically, the developmental biologists have worked a lot on the signaling pathways that affect these cells. So I showed you yesterday the EGF notch for these cells. And I will show you that for all of the cells, actually, but P3P is the most sensitive, you have another signaling pathway called WEND. So here what we're changing is perturbation, not by random mutation, but by manipulation by the experimenter of gene dosage or gene activity in these signaling pathways. And so in animals, very often, you have a gradient of WENDs that come from the tail of the animal towards the anterior. And so this is representation. So in silicon, you have five different WEND genes. So WENDs are signaling molecules that are secreted by cells and signal to other cells. And here you have a staining of one of them that comes from the posterior and going to the anterior. On this image, it looks like it's rich, hardly P3P. I'm not going to make a lot of this, but so all the cells, the six cells are in this region, yeah, the six cells we're talking about. In this experiment here, you have the percentage of division of P3P. Sorry, I forgot to write it. In the normal situation, you have about 50%. Now if you manipulate the gene dosage of either this WEND or another one also from the posterior CWN1, what you see is that if you have one of them, not the two, just one of them, you have a strong effect on P3P division frequency. We didn't do the curve very well in the top. We just know that if we continue in multi-copy, you'll reach 100%. So here, variation, you are in a very sensitive part of this curve of answering, of response to this signaling pathway activity. So anything that vaguely changes, for example, the WEND gradient itself or response to the WEND gradient is going to affect P3P. P4P can be also affected at very, very low levels of WEND, actually here, but it's so much less. Really WEND is in a very sensitive part here of the gradient. So here we have an example of being in a sensitive part of the curve. There are many ways where you may not be. So for example, this is all drawings by a flight geneticist called Randall, where he was imagining cases where you have thresholds. And so here in the case of one threshold, and this may be the case for P4P, for example, where usually it's always dividing because you have enough wind to be here on the plateau. Okay. For, I find these two drawings interesting because there are other cases in developmental biology that the plateau also, there is a second threshold and you can get too much. And this is really reminiscent of this curve of EGF. So when you plot induction, I mean the number of induced cells in the vulva as a function of EGF, you get this. And so Randall derived this from counting these large bristles here on the part of Josephina, it's called the scutellum here. So the scutellum bristles. Here you have four bristles. Here he was varying using different genetic lines. He was varying the level of activity of the pathway that's relevant here and seeing that there was a plateau here and it was much sharper before and after. So what we aim to do with here was to measure basically the widths of this plateau. So asking if you could count the number of messenger RNA of the blue signal here EGF in three, how wide this plateau here is. Okay. So you, here we changed the dose of EGF and measure the number of induced cells. So three is the wild type, it's the red plus blue. This index is counting the number of red plus blue. Okay. That's yellow. And we wanted to look at the widths of this plateau. And so this was done by Michaelis Bar-Coulas who used a lot of genetic tricks to increase or decrease expression of this linseed EGF from the anchor cell and then measured by single molecule fish, the number of messenger RNA molecules in the anchor cell. And he came with this curve in black here which is plotting the induction index here with three being here as a function of transcription of this gene here in the anchor cell with this plateau. So first of all, this plateau is about four fold. So this shows that in the white type we are actually in a zone unlike the wind curve which was in a very sensitive part of the curve. Here for linseed we are on a plateau. The actual, the wild isolates are actually just in the here. It's a two fold distance from either side of the plateau. The second thing you can see from this which is also interesting I think is the red curve which is the variance among isogenic individuals. In different parts of the curve. So here you have no variance because they are all identical with this, was here and which is very common with mutants in the lab. You do have a variance among individuals. So some of them are not going to be induced at all. Some of them will have one cell induced. Some of them will be perfectly induced normally. So you have a stochastic noise among individuals that results in this very sensitive part here of the curve in a high variance. Same here. And then when you reach zero or six you have no variance anymore. It's the stochastic deviation associated with each point. I mean the variance, yeah. It's experimental. I think the interesting thing here is that as you notice as soon as you have a mutation that makes you go from here to here you increase the variance like crazy. Same on the other side. But it's very difficult to increase the variance when you're on the plateau. So for the purpose of my talk today what I want to compare here is again this cell where we are in a very sensitive zone of response to the wind path, these two different winds. And with this system which has a relatively invariant self-heat pattern due in part, I mean the industry is not the only important factor obviously but it's kind of the one which starts the whole self-heat decisions here that you are on the plateau here instead of being here on the very steep part of the curve. So to sum up basically what I showed you is that you have, I mean if you look into the developmental details here you can find mechanistic properties which explain the level of variation of different volva traits. So what we call the class D, so P3P variation is due to sensitivity of P3P in the wind gradient. The class C which is variation in P4P and P8P is still relatively sensitive but much less than P3P in the wind gradient. Then the class B variants that we saw were due to this variation in relative positioning of the anchor cell and the PNP cells. And finally the most rare class of variants that we found, the class A, are variants in the actual self-heat pattern and this is in a domain which is quite robust to those variations in the EGF pathway and therefore also in the response to the EGF molecule itself. Okay, so the two main message here is that's... Yeah, sorry, right. No, no, it was not a variation in a repetitive segment. So the mutations were in genes that we don't know anything about basically. So one of them is a wind inhibitor and makes total sense but what we think is that the other ones are... To say that you have a developmental pathway acting in making a self-heat decision is a simplification. Of course you need a lot of regulators in the cells to around that and so I think here we are mutating things which have slight effects on the interpretation of these gradients which are not known from C. elegans genetics so far and have weak effects. I mean, you would never keep this in the mutant screens in the lab because it goes from, for example, 45% or 44% here to 20%. There are very significant effects statistically but they are mild in terms of phenotype. It's variation whether partially it has some, what we used to call hypo-morphism, there's mutations that contain some wild-type level. Excuse me, if there's a zebra or a venom in a zebra more invariant than 30% wild-type that's a mutation. No, no, so... No, so... I think coming back to the GF curve rather than this, no. Yes, so that's a good point and I would say no. It doesn't matter and I'm very strong on this. So basically what Eric is asking, so of course if I have a mutation in GF receptor and it's a null, it will go to 0.0, no variance. We agree. If it's a hypo-morph, so reduced activity but not null, it will be here, actually one of these dots is a tissue-specific hypo-morph with a very high variance. Now you have regulators of the GF pathways where which nulls have mild effects of moving this and they will also have a strong effect on variance. It's a variable way of thinking of a critical variance or a critical cosmological, but are there certain components that give you some questions to ask that help to sort out the ones that with a null variation are they weren't directly involved in particular developmentals? Yeah, so this is... You're asking the reverse question of what people usually ask but it's no, no, so if you look at here for example the vulva index of three, so the wild type is here and the variance is tiny. Most mutations here have a high variance except if they are null in in three. So some people first ask whether, I mean they're looking for robustness regulators which I don't think exist personally but so they're looking and I think if they take this that have a high variance they say this gene is a robustness regulator because they have a higher variance but for me it's trivial, it's just because then a hypomorph in RAS is a regulator so this is what we're looking for is either that it stays at three and has a high variance or what you're asking is that it's low like this and have a low variance. So it's these two situations and basically no I don't know of any in this system. I think it's kind of an inbuilt... I mean it's not completely true I can know somewhere like this really to be frank as you know variance is something that has not been looked at a lot in developmental genetics so I don't think they exist but to be sure I don't think people have systematically looked for them it's just difficult to believe that you are in a situation where you're close to inducing as I told you yesterday it's very easy when you've induced the blue cell to induce the red cells but you can also induce half of a cell actually at the next division you can have a lot of variation so when you are in this very sensitive part of the curve I don't see how you could fix it but they've never been experiments devised at looking for this for example it would be a mutant screen for example for that piece where one can think about this this has been done more recently now in yeast if you take a GFP reporter and you have a variance and you want to increase or decrease variance yes so this was my last slide to conclude that knowledge of the developmental system so this is more a slide in that evolutionary biologist just to convince them that it's important to look in developmental systems and also that the mutational variance matches the evolutionary trend in PNP self-evolution so this was work in the lab of many people over many years and we are continuing on this thank you very much