 Okay, so let's get started. So as you remember in my first couple of lectures, we talked about how the early stages of embryonic development are characterized by these very rapid proliferations that are all driven by maternal products in the presence of very little gene expression. And this program is sort of executed to make sure that the embryo is ready to take on morphogenesis. And what I want to do today is to discuss the challenges that are put by doing actually cell proliferation and cell division as you are actually undergoing the morphogenetic process of gastrulation. The reason why this is interesting is that if you think from a structural perspective, mitosis puts a lot of challenges on keeping the integrity, for example, of an epithelium or essentially any tissue because as a cell undergoes mitosis, any possible thing, any possible structure both in the cytoskeleton and nuclei get completely rearranged. This is just a cartoon of a cell and dividing in an epithelium and what you see is that instead of having now a cuboidal cell, these cells round up a lot of the structure. For example, microtubules move from being around the lateral and basal membrane and now attach to the chromosome to pull that apart. So you really want to make sure that you make this kind of morphogenetic changes at a time that it doesn't interfere. If the cells are also trying to move or to change shape, you want to be sure that this is timed correctly so that you don't run into problem. And gastrulation is a time where there's a lot of this cytoskeleton rearrangement happening. So let me show this movie again. And this is actually a movie that was taken by Archbishop. I was sitting right there. And it's one of the most beautiful movies that we actually have. That's why we all love to keep showing it. And because it shows this, it starts from very early on. It shows that actually cortical contraction. And then it shows this really beautiful synchronized mitosis that happened on the surface. And I hope to have convinced you yesterday that these are driven by this excitable system that can generate waves. And then somehow, magically, after they've done 13 divisions, they stopped. And we talked a little bit about how they count to 13. Or how they know that they've done 13 divisions. And then after they've ingressed every nucleus in a membrane, they start this beautiful process of gastrulation. And I'm sure that Eric is going to tell you a lot more about this. And but if you see now, you don't have all the cells dividing at the same time, but you get different group of cells divide at different times. So maybe I can just play this again. One never gets tired to look at this movie. At least I still haven't after six, seven years now. So this is really like a beautiful example of the synchronous mitosis, which are mainly driven again by maternal product, very little gene expression. It's maybe a little bit of gene expression, but it's actually not driving any significant cellular process. And then as soon as this then just stop dividing, you get membrane. So now this is not a whole continuum cytoplasm. Now there is membrane separating nuclei in separate cells. And then really this beautiful process of morphogenesis in which you can see the cells have been programmed to be differently. So now some cells need to move in one direction, other in other, and they're programmed to divide at different time. So no, no, that's so called the cephalic furrol. So there is some cell, there is multiple imagination, or tissue bending. And one of them is called the cephalic furrol. And it's the region that would eventually divide the ad region from the trunk. And this is one of the earliest morphogenetic events. And yeah, Eric may have a lot more to say about that. Yeah. It is true, the cephalic furrol is a transit structure. What's its function, Eric? They serve as an anchor. So you can actually see in this other area the time the furrol is extended, furrols go like this. Something like this. But mostly it moves the furrol in the other direction. So when you move the furrol, the energy expansion goes like this. And it's able to move time. That's the decalic copy number you don't see there. I don't see that. Yeah. Maybe. Yeah, so you have to wait. You have to wait for Eric to learn about that. But now what I instead want to talk about is how do you make sure that you don't do cell division as you are also doing these morphogenetic changes? Because if you are trying, let's say, to constrict your apices to do this type of imagination and you divide, that is just going to destroy the tension that you need for having apical constriction, for example. And naively, you can think that it's two ways to do this. Either the two processes are talking to each other so that the cell knows I'm now doing a morphogenetic event that is compatible with cell division. That event sends a signal to the cell division machinery at tell stop. And so that the other way around that does the cell division machinery is operating can stop and delay a morphogenetic event. Another possibility is that everything is timed so accurately that cell division and morphogenesis never run into problem. And the fly embryo offer an example of both. My lab mainly work on the second aspect about things are timed so accurately that there is never a problem. But before going into that, let me just I will want to give a very nice example actually from Eric's lab about this. But before doing that, I will remind you again of the cell cycle and the movie you have just seen. So nuclei first proliferate inside the embryo. They come to the surface. They undergo this synchronous mitosis we talked about yesterday. They stop and they make membrane. And then they undergo gas relation when you can see now this group of cells is dividing. This is dividing, but it's a lot of cells that are not dividing. So now cells have been programmed to divide at the same time. From the cell cycle perspective, we talked about the fact that this early cycle, there is only DNA replication mitosis, DNA replication mitosis. They're trying to go as fast as they can. The way they introduce this extra layer of control in which now they can program cells to divide differently is by inducing what is called a gap phase. So as soon as they've completed DNA replication, they don't enter the cell cycle right away. And they don't enter mitosis right away. They sit there and they wait for some developmental cues that is going to tell them now it's the right time to divide. And I'll show you how that developmental cue is actually linked to the kind of patterning that, for example, James and Thomas were talking about. And this reprogramming all happened at this transition. The mid-blastula transition, I talked about it in the first lecture. But just as a reminder, the mid-blastula transition is not only a time when you remodel the cell cycle in the way I was just telling you about. This is also the time where zygotic gene expression really picks up. And a lot of maternal mRNA, not all of them, but a large fraction of this mRNA that were allotted in the egg to drive early development, get degraded. So this is really a great example of the reorganization of the genome at a developmental transition. And probably Eric again will talk about this and the regulation of the activation of zygotic gene expression in more detail. I will not. But this reprogramming happened at this time. And a lot of people work on it. And then after the embryos done this reprogramming, made cells and undergo gastrulation, now actually what you see is that cells divide a different time. But they actually divide a different time in a spatially coherent manner. What I mean by that is that there are groups of cells that are programmed to divide about the same time. There is 25 of these groups. And they are called the mitotic domains. And they are extremely reproducible. They really show the level of precision and accuracy and reproducibility which the embryo can be regulated, which the cell cycle can be regulated. And I'll get back to that in a second. And now the last reminder for this lecture from the previous one is about the cell cycle machinery. And you remember, hopefully, from all lectures that really the master regulator of the cell cycle is a kinase, an enzyme that is called CDK1, which needs a partner. So when cycling and CDK1 are together, they can have a biochemical activity that can phosphorylate, substrate, and drive cell into mitosis. There's a lot of regulation of CDK1. The one that is will be mostly relevant for this talk is a post-translational modification in which this complex that is present at eye level can be inhibited by an enzyme called V1. It's a kinase that will put a phosphate inhibited. But then there is two phosphatases. But actually, for this talk, all is the one called string. It will be relevant. So there's a phosphate string that can remove this phosphate. And there is also some feedbacks here to make this transition sharper. So what Victoria Fo and Bruce Edgar, when he was a post-doc in part of our lab showed was that this is an extremely reproducible process. This is what I was telling you before, before I showed you in the movie when selling mitotic domain 1 and 2 divide. But actually, there is 25 of such domain. So this is just a view of an embryo, which you have anterior, posterior. It's a lateral view, so you can see all the domains. But you can already appreciate from this picture how intricate that is. So there would be this group of cells that divides as domain 1, these as domain 2, 3, and so on, and so forth. And what is impressive about this program is that it's always the same. So it's always cells in domain 1 that divide first and then cells in domain 2. And remarkably, this all very complex, special temporal program seems to be mainly encoded by a single gene in the sense that when cells in the embryo need to be programmed to divide, they express this gene string. And that's what drives them into the cell cycle. So somehow, all this timing information is all being funneled and talking to a single mRNA. So in a way, that makes the problem of understanding timing of the cell cycle simpler in a way that we just have to study the expression of a single gene. And this is what happens for most of the domain. But there is one notable exception, which is mitotic domain 10, which actually, and I'll show in a second, is the first mitotic domain to express string, but it's only the 10th mitotic domain to divide. And the way that that happens is that there will be some extra delay. But an interesting possibility is that maybe, and this is not fully sort out, so it's an open question, it's an interesting one. And I'm not sure if anybody's really addressing it at the moment. But there is a possibility that actually, the cell cycle and the cytoskeleton are coupled. And there is one notable example of this that is from actually the work on my colleague and good friend at Duke-Denny Lab. And what he showed was that the binding is seems to have a morphogenesis checkpoint. So he has an ability to know if the acting cytoskeleton is properly patterned before cell division. What this is does is that you have what is called the mother cell. And when he wants to give rise to another cell and enter the cell cycle and start replicating DNA, he creates what is called the bud. And then only the bud grows while the mother cell stays the same size. And the way that it does so is by polarizing all the actin and shuddering all the vesicle polarly so that all the essentially is just inflating the daughter cell. And somehow before deciding to enter mitosis, the yeast cell are sensing that the cytoskeleton and the bud have the proper morphology. And only if they do, they will undergo mitosis. And they do this by talking to CDK1. So what is happening is that there is some sensing mechanism that acts at the level of the acting cytoskeleton that can talk to V1. And if you now remember, V1 was the inhibitor of CDK1. So what's happened is that there is a checkpoint that makes sure that the shape of the cell is correct. If it is not, it activates this inhibitor of CDK1 and you cannot enter the cell cycle. Cooled something like that be going on during morphogenesis. We don't know, but there is an interesting suggestive process for this. And he has to do, and this is, again, Eric's expertise. And he will probably, again, talk about this. And this is a process. And we were already talking a little bit about this, that morphogenesis is often characterized by tissue bending and by tissue making folds. The most well-studied during early gastrulation is the ventral furrow. So this would be the mesoderm and the way that it ingresses is by essentially making this furrow that will go in and this cell will internalize. And this was the other furrow, the cephalic furrow that you saw before. And Alessandro was asking about. But this is where the ventral furrow was formed. And if you take a cross-section of the embryo, this is how it will look like. And you can appreciate in this, you are a little better with the shape and the morphology of the cell. And this is now another picture from the developmental biology book of Gilbert that shows in a slightly more detail what this happens. These cells are programmed because certain transcription factor or pattern in gene are restricted to the cells that have to become the mesoderm. They are called twist as nail. And the expression of this transcription factor reprogram these cells. And so that the actomyosine contractility or myosine moves from being at the. So this is oriented in a way that this is the basal side of the cell. That's the ethical side. Myosine moves from the basal side to the apical side and constricted. You have a lot of actomyosine contractility that constrict the apices. And that drives. And I'm not going to tell you much about this because you're going to hear much more thorough and better understanding on this from other speakers. But this caused the tissue to bend and internalize. And only after the tissue has folded, the cells will start to divide. So what will happen is that the tissue first has to fold, cell have to enter, and then they undergo the cell cycle at about the same time. And I'm not sure if it's known what comes first. They undergo an ampitilia to mesenchymal transition, at which point they spread around. And one end up with the mesoderm being internalized. What is remarkable about this, and was shown by George Grossens when he was a postdoc with Eric about 60 years ago, is that actually, if one looks at a string expression, which I told you is the transcription factor that times the cell cycle, actually, you can see that this will be. So this is a lateral view. This is the dorsal side. This is the ventral side. So these will be the cells that will ingress. And this cell seems to be the first one to express string. So there is a very, very high level of expression of string already, as the other domain are barely starting to come up. Yet these cells will only be the tenth one to divide. So what you can see is that essentially these cells have a longer delay. But then what was shown by George was that this seems to be under genetic control. There is certain genes called first start and triples. So that if you delete them, now these cells start to divide much earlier. So these are comparable stage. These cells are not dividing yet. These are dividing. And the way we know that these are dividing is that there is a methodic specific marker called phospho-eastern. And if you stain, you can see that these cells are entering mitosis, while at the similar stage they are not. And as a result of entering mitosis early, while these cells, you can see a form of furra and have gone in so that you have really only few left on the surface, all the cells of the miso del near are still on the surface. And essentially what is happening is that they were trying to bend the tissue, they started dividing, and that destroys the integrity and the tissue doesn't go inside. Now if this, the delay that is mediated from these genes is just an hardly encoded, it's just an extratimer or a delay. Or if this is more similar to yeast, a way in which cells are sensing that there is some mechanical signal that should slow down the cell cycle, it's unclear, and I'm not going to talk about this because I don't have anything original to say. We are not actually focusing on that, although I still think there's an important open question that the field needs to resolve eventually. What I want to tell you about instead is how what I think is the major way that embryo solves the problem of making sure that mitosis and morphogenesis don't bump into each other, which is by timing everything very reproducibly and accurately. And so the first experiment I actually did, as I was a postdoc in Eric's lab, was I focused on two domains that are nicely on the surface before gastrulation has progressed further so they are easier to visualize. And I really wanted to put a number on how accurate is one cell division from one embryo to the next. So what you do there, you just develop some computational method to identify where mitotic domain one and five are. This is a little bit of a self-consistent definition in the sense that you look for a group of cells that divide about the same time and is a coherent object so that the neighbors are programmed to divide sufficiently in different time. And then you track these cells and you score when they divide and you get something like this. So you get the fraction of cells that are divided as a function of time. And you can see that all the cells in one mitotic domain divide in about 10 minutes. Mitotic domain five, it's programmed to divide after mitotic domain one. And here you can already appreciate the high level of reproducibility in the sense that mitotic domain five always divide about five minutes after mitotic domain one in all five of these embryos. So it never happens that I mistake my cells in domain five divide before domain one. But actually what we really wanted to do was to get a sense of how reproducible every single cell is. For that you have to be very careful. You can just look at this in the side that is 10 minutes because what is possible is that what I'm calling a mitotic domain as a group of cells that divide synchronously may be an oversimplification. Maybe there is a structure within the domain. This is in fact what we found. So what happens if you look at these domains is that they don't all divide at the same time. So what we did here was taking those five movies, overlapping them at best that we could and look at the relative timing. And what you will see is that essentially there is wave-like pattern on mitosis. So these are the cells that divide first within a domain and then the cells that are at this end or more toward this edge are reproducibly programmed to divide later, similarly for mitotic domain five. So what you really need to do is to take care of these deterministic pattern if you want and only see how much does a cell deviate. So what you really wanna do is align one embryo to the next, go at the best you can to the level of a single cell and say how precise was this cell from one embryo to the next. And when you do that exercise, the best that we could, we come up with something like this. Essentially this is the deviation of how much noise there is in timing for individual cells once we have perfectly aligned them. And what you get is that this is a distribution with a standard deviation of about two minutes. This is over a cell cycle, it's about 90 minutes or so. So this seemed to be a very reproducible program. So where is this precision and reproducibility coming from? And I've argued before based on what was known in the literature when we started, I see these all seem to be linked to the expression of a single gene. But again, these were all based on like in C2's expression and very, very crude, like fixed materials in which it's very difficult to determine really if something is precise down to the level of two minutes. So we wanted to relook into this more quantitatively using live imaging. And the first question we try to see is that is really string transcription what encodes the timing of cell division? Or could there be another process? So to do this, what we did was the following experiment. We generated flies that had un-stringing answers. So now you remember from both the lecture of a lot of people at this point by Ken Post, Thomas Greger and also James that there is pieces of DNA that encode for the spatial temporal expression of genes. And often what you can do to, and this was, I mean I did this experiment several years ago, sometimes it's very difficult to genetically type genes in organism, it's getting easier and easier but it was very difficult at the time. So what you can do to get an idea when a gene is expressed is to get this an answer and have it drive GFP. And when you see that GFP is expressed, you can get a sense of when that gene was activated in that particular cell. It's not perfect, but I think it was at least at the time one of the best thing we could think of and now we have better tools actually for this, but this is the experiment that I could actually have enough data to show you. And then on top of having flies pressing this construct so that we have a proxy of when string is expressed in individual cells, we also have a express system tagged with direct P so that we can follow the nuclei and see when they divide essentially. And if I play this movie, so what you will see, let me just, I don't know, I can't see, right? Okay, so what you can see here is that the outline will be blue and it's gonna turn from blue to white when I detect expression and the gene turned on, once I don't mind, sorry, I cannot see my arrow anymore, so I don't know where to click, let's see if this works. Yeah, sorry. So, okay, I'm not playing movie, yeah, okay. So I guess I don't have the movie here. So, okay, I could try to download it. I can maybe show it to you later, I'm sorry, I should have tested it at least. I don't know why it's not working, but what you would have seen is that as time goes by, this cell will express some GFP and then about five or 10 minutes before it divides, I can show you the quantification of this movie. So this is, well now you can imagine seeing, or imagine all the data will look like, but if you look at cell in my target domain one or five, what you see is that there is very little GFP expression as development goes on, and then at some point the gene is turned on and GFP accumulates. That is specific data and answer that we used is specific for actually my target domain one, two and five, we did not image domain two, but if you look at cells that do not belong to that domain or that they're not dividing, they don't express GFP, but you can see GFP accumulating in this my target domain one and five, and then at best we called you like computational algorithm to detect when this gene is turned on and then you correlate is the time when the genies turn non-predictable when cell divide, and the best that we could do is the correlation where rather strong, so a cell that turns string on early seems to divide early. So this is sort of suggesting that maybe most of the timing is encoded to string transcription. The other thing that we noticed, and it was true for both my target domain one and five, is that the relationship seems to be the same between right, so there is a given delay between when string divide and when mitosis happens, but that delay is about the same in both domain one and five. So for us, what then the way that we interpret that is that we could treat these as two separate problem now. There's two problem, the first problem is how do you decide when to turn on string and then there is a second problem is like once string is turned on, how does it determine when mitosis happen? And my work as a postdoc was mainly on the second step and now in collaboration with a graduate student who just graduated with Eric and now in my lab, one of my students that tried to tackle the first problem which I actually think is the hardest one is like, how can you time transcription so accurately and precisely in embryo? Of course, I'm less brave than my students, so I tackle the easy problem first and let them do the hard ones. So what Eric and I decided to do was to analyze this question of is string sufficient to encode the timing on mitosis or are there other inputs? The best way we thought to do this was they just make a constraint in which string is expressed uniformly everywhere and let's make sure that you get that all the cells divide at the same time. That would be the prediction. If a cell was seeing, if cells seeing the same amount of string will divide at different time, or will have different sensitivity to string, that will imply that we are missing something. There are other inputs. So the way we did this experiment, how I hope that this move is not here. So this is the wild type and what you will see is again that they undergo one last synchronous division and then you will get this different group of cells that divide at different time. And now let me show you, hopefully, what will happen if I'm not blaming you. Let me just find this one second. I'm really sorry about this. Okay. Okay, so it's not projecting the best but you should be able to see. So actually the first division you see is the last division, synchronous division cycle 13 and then about at this time, the old nuclei or all the cells in the embryo start expressing string at the same time. And I hope you can now realize that the old embryo divide about the same time. So now you don't have group of cells that divide at different time. So if you give all the cells in the embryo the same amount of string, they will all divide at the same time, which suggests that really there is no other inputs. And we have done more quantitative work in which we have looked at the sensitivity in different region. You can look at the anterior cell more or less sensitive than the posterior cell or the middle cells to string. And what you found is that it's not, they are really statistically, in a statistically reproducible manner all responding and showing the same sensitivity to string expression. You still see a wave and the reason why I think you still see a wave is that the delay that was set by the previous wave. So for reason that we not fully understand or we partly understand, but I can say fully. So what happens is that a cell divides and then it undergoes these maternal to zygotic transition when it can start transcription. But that transcription starts about 15 minutes. And we don't know if that's the transcription in part is also the product is being degraded actively. But essentially there is a process that will only let string accumulates about 15 minutes after the previous division. That process of course, reflect the previous delay. So if you see that there's a wave at cycle 13 and you still have a wave at cycle 14, but that wave is not determined by string. What that tells you is just that the cells in the anterior express string a little bit earlier than the cell in the middle. And that is because they've divided a little earlier that clock was a little advanced. So this wave-like mechanism that you see at cycle 14 is not the result of cell communication. It's actually just of the result of the previous wave. But the cells, they will all have the same sensitivity to string. Yeah, there's a question there in a second. Yeah, I'll show you in a second. I should have put this next slide first. So as you have seen gas relation is a pretty complex process. Oops, let me just play this movie again. This is a good question. Why 1 and 5? The truth is because they were easier to image. And the reason why that is is that if you look now, the embryo is a nicely organized array of cells or nuclei on the surface. But as they start undergoing gas relation, you get a lot of folding and tissue moving. So the mitotic domain 1 is this 1 and 5 is there. They are still very early. It would be very difficult to follow mitosis at this stage when a lot of folds are forming. Or cells in between? Oh, yeah, yeah, the distance. How much is the distance? Right, so in terms of distance, it's about, so let them undergo this division and then I'll show you where they are. So mitotic domain is around here, and mitotic domain 1 is here, and mitotic domain 5 is around here. So they are probably three or four cells apart, which at this stage will be 15 to 20 microns. So they are sufficient, and some of the cells in between don't divide. So it's very easy to determine the boundary. It's not that difficult ones have made few of this movie to identify which domains they are in, also because they are so reproducible and so stereotypical. But the reason why we chose them is not anything special about their geometries. It's the fact that among the earliest ones that divide, they are the most at the surface and they are a little easier to image. Sorry. Oh, yeah, I should have put it here. So for that, you use a trick which it's called the GALFOR UAS system. So the way this works is that you use this idea of an answer. So GALFOR is a transcription factor that flies non-express, fry.nf. This comes from East. And you remember, I've told you that when the mother lays a egg, it lays a lot of mRNA in the egg. But usually it is a mRNA with some notable exception of the one that makes gradient. Most of the other mRNA are uniform everywhere. So what will happen if you have a mother that expresses maternal, what is called maternal GALFOR, as he laid the eggs, he puts a lot of mRNA and protein already. So now you have GALFOR protein everywhere. Then what you do is that you cross these flies to a male that is able as essentially binding site for GALFOR. So now the transcription factor, GALFOR, can recognize this piece of DNA and bind to it. By binding to it, it can activate gene expression. But because this is uniform and because this piece of DNA is encoded genetically, so every nucleus or every cell is expressing it, this is essentially a way to give the same amount of strength to all the cells at the same level. And the trick of introducing this by the father is just because if you put any gene in through the males because the early stage of development that driven maternally, you get no expression, right? The zygotic genome or any paternal contribution will only be expressed at the maternal to zygotic transition. By doing this, we essentially let the early stage of development be driven all by the maternal product and only see zygotic product later on. And Eric is gonna explain you how that idea was actually seminal to find all the gene that control early embryonic development which he actually discovered himself. Right, so these cells divide about 25 minutes, 20 to 25 minutes before any endogenous cells will divide. And it takes about 10, 15 minutes realistically for from string expression to cell division. So at the time of which these cells are dividing, nobody's expressing string yet. If you wait 10, 15 minutes, you might start getting interference, but if you do it at the stage of which we did it, essentially there's no endogenous string because it's not being expressed yet. So string is degraded at the maternal to zygotic transition, there's nothing left and then you re-express it in the cells that you want to divide. That is very accurately programmed and only happens few minutes before mitosis. So we are driving these mitosis sufficiently early that you don't run. So if I were to, if I gave you a timeline, you have the maternal to zygotic transition or MBT, you have cellularization about 50 minutes and that's where our UAS string gets expressed and then the cells will be, in this experiment the cell division will lap in here. So this will be probably about 60 to 65 minutes. If I show you what will happen endogenously, maybe I don't sure. Can I raise this? Did everybody write down information about the trips? Well, I'll try to squeeze it here. I just, maybe somebody wants to go back and get. So if I were to look again, you'll have the MBT and then you'll have cellularization about 50 minutes and then the earliest mitotic domain would be maybe about 75 to 80 minutes but will only start express string about 60 to 65. So what you can see is that as this cell divide is the earliest that you'll ever see string expression. So this divide early enough that you don't have to worry about the endogenous contribution. Yes, no, so this is a maternal specific promoter, maternal to oblique promoter, it's already expressed in the mother and in general. The problem will stick around for a while and this fly will not survive. That the program of gene expression may some morphogenously significantly, I believe that then they will not survive. But you don't need to get adults to do the experiment, because you just pick up the right female, you close them to the right male, you do your experiment and you're done. But yeah, you just use a maternal promoter, you don't need sophisticated, inducible way to do this type of experiment. So this is the way we drive expression and what you will see, it's also not plain, but what you will see if you look at this or when you quantify these movies is that if you look in individual cells, then you have, you can get traces like this and so this is about the time from unaface of cycle 13. So you can sort of see the timing about 45 to 50 minutes, which is about the end of cellularization, the string problem start accumulating and you can see go up and then cell enter mitosis and you can do this sort of exercise for a lot of cells and what we do, what we did was just holding this, the time between my string is expressed in cell division and you do this for a lot of cells and you measure this rate of activation of string and you measure the time it takes and then you can really see that the cells are seems to be sensitive to string activation rate and here now we wanted to go from these to try to maybe say something a little more of how you make this transformation accurate. So the way that we thought about these or at least to start was that maybe it was convenient to kind of borrow some pages from books of quantitative biology or control theory starting maybe about input output relationship, right? So you have a signal that is going up linearly. What kind of computation is the cell doing to make this decision and that is the computation in any way optimal to assure some precision? And so the first thing we asked is like what kind of transformation is the cell doing? Is it maybe just a linear transformation or just linear with saturation or you have some abrupt or what is called a switch? Is the cell just sitting there and then there's a magic threshold at which you go? It commits to divide. Maybe not surprising we have talked about the need of making transition varies which like and rapid and what that will give to the advantage that those will give to the cell. Also James has talked about by stability as a fisher that will also give this kind of sharpness and insensitivity to noise at least at the transition after the transition has been made and in fact we found that this decision is where is which like? So the way that this is usually quantified is using a ill coefficient. Essentially this number is the number such that the IST is and actually I've done a question here for it. So the IST is what you can convince yourself very easily is that the IST design is the more rapid this is. This is actually with n equal one and you see the growing up is very slow but if you have an n much larger than one then you get something that's a lot more abrupt. And in this case we have a n of seven which is rather sharp. A different way to think about this is that you could see how much string do I need so that I have a probability of dividing of about 10% and how much of that I need to have a probability of dividing of about 90% and that's only about two fold. So a change of two fold in the concentration of this enzyme makes you to have a very little probability of being in mitosis to high probability. But such a if you have such a rapid response you might think that, and again a lot of people I touch on this you might have sensitivity to noise because if there is a little fluctuation that brings you right above that level then you will just go because you're trying to commit very quickly. So is there a way in which embryonic development optimize this need to be fast and commit to cell transition rapidly and also do it with some precision. And again we were thinking about this from a point of view of signal processing and the first thing that we thought is that if you read any engineering book about control and signal processing what usually they tell you is that there is a trade off between sensitivity and accuracy. So if you have a signal that is very noisy and you're trying to make a decision based on it is very difficult. One way that you can improve that signal and this also is that you could not respond to a instantaneous value but you could have some built-in system that is sort of responding to the integral of this. And actually if you take this curve and you do this integral you end up with something like this. This is a lot smoother than this so you won't have the problem that if the treasured threshold was by mistake here you land there exit enter again maybe you may even exit again and so you don't really get confused around here but this temporal integration will make your decision very reliable. The problem is that if you have a very rapid change as you have it here it's very difficult to be sure that there was a rapid change there. So by integrating you can improve precision but you sort of become slow. So you can be fast and inaccurate or you can be slow and accurate and of course there is a whole range of solution in between and there is a trade off and what I'll show you is that we think that the cell do something in between that we will speculate could be a way to improve precision. So that will enter my tosses and we really have not followed that after but what will happen is right so we are only looking at the first cell side we are only looking at the response of this ramping up because this is what happens endogenous physiological. So we only look at 14 cell cycle we don't look at what happens later. The level supposedly may get degraded once they exit my tosses but then they accumulate very fast again if you do it through the dark for UAS but this is not what we are following. Okay so what we did was to mine our data and try to ask it okay so what is the cell really computing to make this decision? Can we extract the kind of signal processing the cells do from data? So what we did the first was ask that would an instantaneous model work a model in which there is really an art threshold the cell accumulates string to their level and they divide. If that was the case you will think that independently about quickly accumulates string you will always divide at the same level. So a way to plot these are independently of what this lag in time is actually you'll always divide at the same level so if I plot the amount of string versus delta T I should get a flat line because all the cells are dividing at the same time another possibility is that they are integrating if they were integrating then it should be the integral that should be constant and if you plot the data you find that neither one works so they are neither integrating not taking an instantaneous measurement this is probably not surprising but again cells that activate take a little longer to divide they seem to do it with a little bit less string cells that take long and by the integral is still much higher. So then what we said was like okay if that's not explaining it let's try a simple function that might explain it and what we thought is like an instantaneous model means that you are integrating over a time scale of zero an integral model you are essentially you have infinite memory maybe what the cell does is integrating but over just a short period of time so you can easily write that as a say you may you have your string dynamics as a function of time you deconvolve that with an exponential kernel with a given time scale and you ask what kind of memory does the cell need to have to explain the decision and we got a time scale that is about two minutes which is may seem fast but it's not that much faster than the time scale we are interested in so we think that the cell is trying to compromise between still there we want to respond on a time scale of about 10, 15 minutes but so it's in doing so maybe you don't want a memory of 10, 15 minutes because that's equivalent to just doing an integral and you run back into the same problem of being slow but they still don't integrate over seconds on which maybe noise will be too relevant really if you want to be skeptical and the experiment that we don't have but I think we should do here that will be crucial is to measure what is the time scale of the fluctuation of noise because if the fluctuation of noise was on time scale of 30 seconds the time correlation time on noise fluctuation was of order of one minute this system will be perfect it will filter out the noise and still give you speed if the fluctuation correlation where of order of 5, 10 minutes this will really not help and we haven't done that experiment yet but I think we really should too was there a question? okay so now how do you build a biochemical signal a biochemical circuit integrate signal over time and turn out that the solution is rather simple it's actually kind of obvious in a way but so what is happening is that you have an enzyme Cdk1 that could either be in an active state or an inactive state and the transition between these two states is driven by these two enzymes and what is happening is that we want is constant as string is going up now if you think about how much of Cdk1 will be active at any given time that is not necessarily going to follow instantaneously how much string and we want you have because these are enzymatic reactions that take time so there is a built-in time scale into these which is the inverse of the enzymatic activity or the rate at which these enzymes are operating and so that time scale could give you the integration and because how much string you need for division is kind of set up by how much B1 you have this sort of make a prediction that if I can do an experiment in which I can change B1 level I should be able to change for how long they integrate and so these we were successful in doing and we did a very simple experiment of looking in a V1 heterozygous and what we found was that if you do an experiment in V1 heterozygous and you ask for how long they integrate well you will suspect because the integration is the inverse of V1 concentration and you will suspect this integration to get longer and this is sort of what we see at two minutes integration doesn't fit the data anymore but if we go up to about five to six minutes now we have a longer integration we could over express we want a little bit and these data are a lot less convincing because at some point it's almost saturated so it's very hard to know but at least there seems to be a trend this model actually by having a kinase phosphatase cycle that is operating out of equilibrium also gives you some switch like response not a very high one but this is sufficient maybe to explain some data we had about the relationship between CD25 level and CDK1 activity so what we speculated in this paper and I agree that it will be great to show that the integration time is the relevant one but what we speculate is that if you control a cell decision by having a kinase phosphatase cycle what you get automatically you can get a switch or at least a moderate switch but you also get an ability to reach a compromise between speed and accuracy and this is what I tried to depict here so what I did here this is the same signal I showed you before but in red now instead of showing the integration as I did before I'm showing you the integral of this activity with a time scale of about two minutes and you know at some short term integration and what you see here is that this is a compromise in the sense that this curve is much smoother but it's still able to sort of follow or just follow with little delays and decrease in sloppiness it's very, it's pretty much able to follow rapid changes in activity so this could be a compromise you filter out noise but you still have some rapidity in response and will be very interesting to see if these kind of strategies are actually used over and over again as a mechanism to achieve precision and filter out noise okay okay yeah uh... yeah okay go ahead right, so this is well, I mean this is actually and we are not the first one to propose this idea we are probably a mo- I'm not sure even if you are the first one there is experiment probably in development we are among the first one we try to quantify and show some evidence that this is actually the way that they do it in a contest that has to do with time but there's a lot of theoretical work in fact Boris Shryman who is speaking next week he is, he did a lot he had the theoretical paper in which all these ideas that are sketched were very nicely explained and the compromise between speed and accuracy was and this was actually in part what inspired us in doing this work so yes, this is a lot of work in engineering and control theory and even work done by people who were thinking exactly about this type of circuit before so I'm sorry if I gave the impression that we invented this probably not referencing all the relevant work done before but this was our attempt at figuring it out the possible role of this kind of mechanism for timing the cell cycle in development you also have a question? right, so we think you integrate you filter out the noise but you don't integrate over the whole history you integrate with a two minutes exponential kernel what that is essentially doing if you want is that you are only integrated over about few minutes so you are, right so it's not, it's an averaging but you're only averaging over what happened in few minutes so that you don't need to keep all this memory of past events that are gonna preclude you from being able to respond very quickly to what if things change very quickly and so we think that this is a compromise and you can make some of these argument a lot more rigorous mathematically and some theories are doing it but we just want it to be as simple as possible in our thinking because in any case we were we are very limited maybe now we can improve in how well we can measure these things so we couldn't be too sophisticated in our theory because we couldn't be able to distinguish sophisticated methods with the data that we could generate okay so this took a little longer than I wanted but so this is our answer to the second problem how do you go from turning on string to making a decision that we think is accurate and fast but now we still don't know why cells divide so accurately because we don't know it's how do they know so accurately when to turn on string transcription and this is a much harder problem and we are still working on it I think it's an exciting one the first cue from this came from again Bruce Edgar just if you do fly self cycle you can escape him and what Bruce showed was that if you mutate the patterning genes these are the same genes that Thomas and also James were talking about these are genes that give special information to the embryo these genes somehow inform the cells of where and when to divide I think what this really shows is that informed cell where to divide and then I'll now I'll go over the work that Amir was a former graduate student of Eric and collaborated also with me who did to try to figure out if they also encode time but what the Bruce showed here was this is mitotic domain two and would be one of the major projects of my talk now if you look at this mitotic domain two and you look in a mutant for one of these genes that is expressed only in a certain region of the embryo called gap gene and you mutate the gene then mitotic domain two doesn't is not there anymore these cells are not programmed to divide anymore similarly, now the transcription factor to specify the message that Eric will talk about this these cells do divide but if you don't have twist they don't divide and what you are looking at here it's a technique called in situ for string and marinade so string is not expressed in mitotic domain two if you have a button and all it's not expressed in what is mitotic domain 10 if you don't have twist so this seems great the pattern in gene control string but there is a problem with this type of experiment is that you really don't know if you just do this mutation if this factor are really what encodes time because you know that they are required but once you get them off these cells are not dividing anymore so you don't know if what is timing them you just know that this factor is needed for them to be expressed but it could be possible that this factor just needs to be there and then there is another queue that is giving you time so what we decided to do and we thought was better is that anything that is a real limiting regulator of any process if you change the rate at which it accumulates or gets degraded if you change its rate you should be able to see a shift in time so the way that it's easy to do this genetically is that you look in an aetherozygot so now instead of having a read of a gene we have just made it off so you don't have as much of a gene product as you want, you have alpha these two leaders slow down or accelerate processes if that's a real limiting regulator if it is not nothing should change and so the first experiment Amir did was taking the botanette gene that I just showed you and he compared so I told you that it was required to divide the metotic domain too these are the blue cells here so now what you would predict and it actually also had a way to give him more botanette and I'm not gonna tell you the genetics it's a bit complicated but believe me that he could do this and what he noticed is that if you have only one copy a lot of cells in metotic domain once start dividing and metotic domain two follows but if you have four copy of botanette then metotic domain two divides first and then metotic domain one follow you can quantify this but he could show that by changing the dosage of this gene he could shift the time on which cell divide so that's when a wild type will divide if you give it a little less is later if you give a little more is earlier so these genes seem to be a real limiting activator so then what Amir decided to do and that's a picture of him is now graduated and move on a postdoc and Eric was sitting in the audience what Amir decided to do was to do a genome widescreen he was gonna look for all the genes in the genome that will do this that they will act as really limiting activator but he wanted to do it in a way in which he only changed dosage a lot of this experiment that be done obviously which should take out completely the activity of a gene as I told you and I hope I've convinced you that is not gonna tell us about time we needed this more sophisticated and precise quantitative measure which is exactly what Amir developed and so here he use a great tricks flies are great genetic organism and you'll hear more about that but the people have engineered flies in which let's say this is a fly chromosome so it's a large structure you can essentially delete a big piece of it sometimes even up to 1% and the fly is able to leave with this genetic deletion because he has a functional copy of that gene on another chromosome so the fly essentially can survive as by only having one copy of even like a lot of genes like cumbreads which is remarkable and is a remarkable genetic trick because now we don't need to scan every single gene we could use this deficiency to scan regions of the genome that are larger and then once we found interesting regions we can go in and look for the genes and so what Amir decided to do to get a lot of embryos he will actually not use light imaging now he will fix embryo and use the phospho-eastern reagent I introduced before something that marks specifically metotic cells and what he will do, he will fix embryo and then he will count how many cells are in mitosis in metotic domain 1 and 2 and will do that over time and this is not projecting well but yet the proxies have morphological proxies that allowed him to estimate time very accurately so he could do this over time and essentially he got a map these are many cells in metotic domain or one or two divide as a function of time and now he goes he builds a very nice map for wild type and then he goes to the deficiency and what you expect to find are three classes either deficiencies that don't change time these should be most of them hopefully there will only be few real limiting regulators so that the problem is simple and we can actually treat it and solve it and I'll show you the evidence that this seem to be the case thankfully another possibility is that a deficiency actually does advance or delay timing so that either the time they divide a little later or a little early if you want this could be a real limiting activator or a real limiting repressor there's another thing though that is interesting here is that the slope of this line is the same what that means is that the time it takes for selling one domain all the selling one domain to divide is the same you are just making them divide few minutes earlier or few minutes later so you have shift time but you have not changed the overall rate at which they enter mitosis there is a third possibility of course in which they start dividing at about the right time but then the mitotic domain progress lower or faster so what Amir did was he covered about 85% of the genome so it's not full coverage but it was still quite impressive because it required him to screen more than 200 deficiency and you have to count many embryo for each deficiency so this was a lot of work but I think he paid off because he didn't get it especially for mitotic domain 2 it didn't work quite as well for mitotic domain 1 or mitotic domain 1 only as one regulator it's hard to believe but for mitotic domain 2 we got a lot of it and so now what we can do of course is essentially what we have done now we know that there's a piece of DNA somewhere on a given chromosome that seems to be acting as real limiting but that doesn't tell us the gene so what Amir did for the gene for finding the gene he went on into... so no, this is different deficiency so this is a different... what this is is a different genetic manipulation so this line... okay so what he does he fixes a lot of embryo and then he counts and this is telling you this is a map and it's extrapolated so these are not the raw data I'll show you some raw data in a second this is just something in which you because they will look too clouded but let's say stains 30 embryo of a certain genotype and then he counts and many cells divide as a function of time and build this map and then every one of these line is lacking remember I told you that we could delete a piece of chromosome so this one will maybe miss a piece of chromosome somewhere on chromosome 2 at the beginning this will miss a piece on the end of chromosome 2 so this is sort of like the way that you could think of this is walking through the genome and delete piece by piece try to find which piece will give you a shift in this map the grey stuff are all the deficiencies... sorry the grey stuff are all the deficiencies that they did not do anything so the grey majority like about 90 something percent of the deficiency don't do anything which is great which means there is only few specific genes not that every time you delete any gene you mess up with the mitotic program that will mean that we will never figure it out what it is so this is actually a great thing that Eric will probably also talk about and he's going to talk about genetics which is the importance of every process in which only few genes are really important for something because then you can really get specificity and you can understand what is happening so then to find the gene what Amir decided to do was to zoom in a region and look at which genes were expressed in that region and this is what you get you get this big list this really looks overwhelming but because we work with flies and we can stand up the shoulder of so many good people that have done genetics and a lot of other work you can just guess and the way you guess you look is there some transcription factor that we know that is important in patterning the embryo and is expressed specifically in this region and for each one of these deficiencies you couldn't find such a gene so for this particular deficiency it looked and he found that there is a gene called empty spherical and this gene is expressed as subtly mitotic domain too and then what you do you get a mutant from the mutant collection or from you know you go into Eric stocks and you find the mutant for the gene and you see if repeating the experiment rather than with the deficiency but with the mutant gives you the such same phenotype would you still see a delay in the time when cell divide and it could show that all these deficiencies were recapitulated by single mutants so now he has found the genes this is actually how the raw data looks like and this is the name of the genes for deficient others I already showed you bottom ed so we already had one activator for domain two and now I show you two other empty spherical and knirps what will happen in here is that the black curve is the wild type curve so this is typically our cells will divide but now if you and this is what is plotting here it's doing something a little different it's plotting metodic domain two versus metodic domain one and what happens is that if you have a little bit less empty spherical what will happen is that selling metodic domain one now start dividing a little earlier with respect to selling metodic domain two so that the curve is shift up and you see again as I mentioned you get the same curve for the deficiency that you get for the mutant so this is really the gene in that region you get the same thing for knirps there is activator there is also repressors and so what we saw is that there were three genes that shift this curve down what was interesting was although that all this gene they all belong to the class that I told you before in which you shift the curve but you don't change the slope so somehow just changing the dosage was not sufficient to make the cell the overall rate be slower you just program the cells to divide five minutes later or five minutes earlier however if we did a double perturbation if we mutated let's say if we got rid of two repressor or two activators then we start seeing some condition in which the slope became statistically in a statistically significant manner either smaller or larger so it does look that maybe it's an abuse term but a way you could state this is that the system is robust if you just go and break a single activator a little bit you will still you still get a very precise or a group of cell that divides with its own temporal precision but if you start messing with two then maybe then also it takes someone longer anyway this is the data what it really means we're we don't know yet but so now what we know we know that there is certain genes that are limiting but we still don't understand how time is encoded so precisely and I don't have an answer yet but the way then my student Patrick who is here although he's not here right now and and I have been thinking about this building on of course already on a lot of discussion that I've had over the years with both of me and Eric about this and it's not I mean it's not that of original ideas just let's start this description factor and see what they do and I show you movies for two of them I show you some quantification and I stopped there because we haven't done enough analysis to have anything actually intelligent to say but hopefully I will next time this should play because I know that I just put them in so this is a botanical one of the major activator and I'll play this again so what you will see is that when these movies start you hardly see any fluorescence in yellow and these are the Eastern markers so that you can follow the nuclear and then as time progresses you'll see GFP appear in a stripe and then most of the cell in this area will undergo mitosis and as they undergo mitosis the nuclear envelope breaks and so the signal disperse and that's actually a great way to see when they enter mitosis this is instead what happens for a repressor it's called law prepare one now the mitotic domain is going to be around here and I hope you what you'll be able to see is that when we start the movies law prepare this is pressing most of the selling domain to and will kind of specifically disappear only from the cells that are programmed to be metotic so this appears here and these are the cells that will divide so this is about what my topic domain to ease and these are the cells that have lost this pressure and again we don't know how accurate it is a one-to-one mapping or how accurate this is this is already preliminary but what we do have and it's working is that we have our algorithms to very simple I mean we can use all our algorithm to quite simply segment and track those cells and so this is the kind of dynamics we see bottom head starts very low and this is what happens if you look in a region when there is no expression so this could serve our zeros for expression of the noise level background level bottom head goes up with time as well prepared goes down and now what the question we want to ask is how does a cell sense these changes in concentration to make an accurate decision to time and uh... you know are there accurate switches are they actually integrated doing similar kind of integration or is there enough dynamical range into this that this decision could be accurate without needing specialized mechanism so this is most of what I have of our time to understand my doses are timed and now one could generate a precise temporal program for gene expression and the conclusion I guess for this part that it looks like that the same prescription factor that control the special pattern or encode the special information in the embryo somehow also encode the temporal pattern of my doses so you could think of maybe the mitotic program has been part of the differentiation program as they're having these clocks that are built in to assure that the cells never run into problem of dividing and doing morphogenesis at the same time and uh... there is both activator and repressors and we can measure the dynamics and hopefully next time I give a talk you will be able to hear something more intelligent about it and uh... so Eric and Amir so all the work that I have presented except for the last part that is now the work of my graduate student Patrick who most of you have met was done in collaboration with Eric the first part on our cells go from activated string to mitosis was actually my first postdoc project when I was with Eric and then the big screen and probably I will say the biggest breakthrough we have so far about this problem of timing was really done by a great graduate student Amir was now postdoc in Colombia that's it