 Hello everyone and welcome to this mini-series on modeling developmental systems small and short and brief and simple explanation of why we need to model to understand developmental processes. I have about two hours to introduce a topic that takes several semesters of lectures usually so what we're doing here is clearly vision impossible. It's going to be very condensed, it's going to be very superficial, it's going to be very broad. So fasten your seat belt for a little introduction what I want you to sort of get out of this is not expertise in how to model a developmental process, a developmental system. But I want to sort of instill in you this sort of understanding why it's a good thing to model. So as an experimentalist as a maybe bioinformatician so that you can go and find sort of the colleagues in your department that can do the modeling for you can understand what the purpose of modeling is what it can and what it cannot do. So it's going to stay at this very sort of conceptual level high level, and I'm trying not to be very technical. The strategy is to sort of introduce the general purpose of modeling in this first lecture, and then provide two examples of what I think a really successful and illustrative cases of where modeling was used to do something that you couldn't achieve with sort of molecular and genetic experimental work alone. So this lecture is going to be busy with the question. Why should anyone model of course, nowadays it's good to have a model in your paper because that increases the likelihood of it getting accepted by editors but that's not the point and often these models that are just added on top of some experimental work. Anything to the understanding of the developmental process under study. So basically what I want you to understand here is when do we need to model and what does that get you that you cannot get with experiments alone so we do not just want to have a model because it's fancy. We want to model only when it's necessary and when it's beneficial and whether it contributes to some sort of explanation or understanding of a developmental process. The way I would like to frame it because I am coming from the field of evolutionary developmental biology, I would like to see development as the sort of process that mediates between the genotype and the phenotype. So in my head I contextualize development in terms of the genotype phenotype map, which is what maps variation at the genetic level to phenotypic variation, which is ultimately what population processes such as selection work on this is where the variation occurred. Variant phenotypes that are being selected. And of course, this is just a sort of a thinking help. It's not very realistic. In reality, animals are actually phenotype to phenotype maps right so you go from one generation to another generation that resemble each other, but it's helpful nonetheless. If you sort of consider that it's a cartoon and also that there's some additional factors here. You have an organismic and external environment that contributes to tissue level factors environmental triggers to contribute to this mapping. It's not a closed mapping this sense. And also you will notice that in the drawing I am showing you here several different genotypes sometimes map for the same phenotype here. And that is of course robustness, canonization or resilience of development, while sometimes the same genotype can map to different phenotype under different environmental conditions. And this is called developmental plasticity. So there's a whole lot that goes into this very simple picture, but it's, it's nevertheless good to structure your thought this is the main task we want to understand how genetic variation maps onto phenotypic variation. The structure of this highly complex and degenerate map determines the variational properties and hence the evolvability of the development system. This is my own personal interest, but I would like to make an argument not only that modeling is important to understand development, but also the comparative analysis of developmental processes will come to that later. What's really important here is that you can interpret the mathematical structure of a map in two very different ways. So this sort of mapping here, it's also called a morphism in mathematics, could be a correlation. So we could just be measuring genotypic variation and correlated to phenotypic variation. This is what, for example, quantitative genetics does. But development, of course, looks at this map in a clearly different way, a more mechanistic way. We're going to try and understand what it means to explain this map in a mechanistic sense. And so what is really hidden here is not a correlation, but sort of a causal process. Okay, so we want, we don't want just correlational explanations. But we want a causal mechanistic explanation of this map. And then we need to realize that what's hidden behind this very abstract formulation is a process, of course, very beautifully illustrated by Conrad Howe Waddington's epigenetic landscape. And I'm showing you a rather less well known example, an illustration of his landscape here, where you can see two little X shaped sort of shapes here and then different valleys that lead to different tissue types, or even regionalizations in the embryo. So you can build a thorax, a wing, different kinds of leg parts, antennae, mouth. And what is represented in this landscape is a trajectory. Okay, so this, this sort of visual metaphor that Waddington is presenting here is not an exact model. It remains at a metaphorical level. It's very powerful exactly because of that. The visualization is very compelling. And also, the formulation is sufficiently vague, so you can actually adopt this picture to many, many different contexts. Just think about it. What are the different axes here? So this sort of depth axis here is time, of course, time flows downward. But then it gets more difficult if you think about the height, or sort of this X axis here, what is the difference between those different fates. And also, if you think about it a little more, why are these sort of balls that are rolling down or egg shapes that are rolling down the landscape? Why are they separate from the landscape itself? What do they represent? Do they represent a cell, a tissue, an organism? It's not clear. So this is both the power and the limitation. The genius of Waddington was to not explain exactly what he meant by this picture. So everybody can adopt it to its, their own purposes. But on the other hand, this is not a precise model of development. What it tells us and what the power of a metaphor like this is that is what we need to understand is not a thing. Okay, it's a process and Waddington called his developmental processes the process of rolling down the valley. He called it a creode. This is a word that's no longer in use. So we will call it a developmental trajectory. We need to explain the shape and the dynamics of this sort of trajectory here. That is our task. And this is where modeling becomes crucially important. So let's, let's have a sort of a historical look at how do we explain these trajectories historically. And we could arbitrarily start maybe in the 1970s when people were starting to describe developmental genes and in particular they were discovering Hawks genes and found out that they're active in all animals. They're active in the specification of whole body regions. And if you actually mutate one of these Hawks genes, you switch the identity of one body part into the identity of another. So the first view, the first type of explanation for these trajectories is the genes act as sort of switches that put you in one of these valleys or another. So development is just sort of a sequence of switches, one switch after the other, and what we're called selector genes like the Hawks genes were executing those switches between the valleys. Okay, that's, that's fair enough for the sort of knowledge at the time. So what I'm showing you here is a beautiful study by Michaelis Averhof and Michael Aiken on the identity of different arthropoc segments and how they correspond to the rearrangement of different Hawks genes. But it leaves many, many things, many answers open. For example, what, so you now know how you switch from one valley to the other, but the shape, the exact shape of each of the valleys is not clear. So let's move forward and develop this, this sort of type of explanation a little for further. And throughout this lecture, I'm going to talk a lot about one of my favorite animals here. This is a vinegar fly, not a fruit fly, fruit flies, parasites, fruit, tephritis, flies like the MedFly in California. This is Drosophila melanogaster, of course, one of the most commonly used systems. So this lecture is aimed at sort of early stage PhD students. So you're probably familiar with the animal we're going to use the segmentation gene system that creates the beautiful segmented body plan. Visible here in California. And by the fact that different appendages are attached to different thoracic segments here. We're going to look at that as our example system. So here is the segmentation gene system of Drosophila melanogaster, very brief sort of just reminder. It's active during the blastoderm stage of development during the first three hours of development, when the embryo is a syncytium. There are nuclear divisions, but there are no membranes yet between the different nuclei, and the nuclei migrate from the center of the embryo to the periphery where they form the blastoderm, which is visible beautifully here. The advantage of this embryo is obviously that it's developing very fast. We're looking at here about one and a half hours of developmental time, and also there is no growth or sort of cellular rearrangement involved. So what's happening is that you can have transcription factors expressed that diffuse through the embryo and directly regulate each other's expression in this spatial temporal context. It's very simple. And this is great to study pattern formation because you can focus when you have a mutant of one of these factors, segmentation genes. The only effect it has before gastrolation is that it affects the expression of other genes, no other morphological or any other. So a very simple developmental system, that's its basic power. And you have this cascade of maternal coordinate genes, such as bigwig, which is shown here, RNA localized at the interior of the embryo, then the protein is translated and diffuses through the embryo. You can see the single nuclei here and the different concentrations of this transcription factor in the nuclei. And these maternal coordinate genes activate transcription factors that activate at a first step. The gap genes, hunchback, crudable, shown here in the middle of the embryo, canerbs, giant tailors, and huckabine, and these genes are expressed in these sort of broad overlapping domains. Together, the maternal coordinate genes and the gap genes regulate the peril genes, the first sort of periodic expression pattern here, seven stripes, even skip dots, skip hairy, run fuchsia terrazu, and paired are examples of peril genes. And all of those together finally regulate the expression of segment polarity genes, in particular, engrailed and wingless, which mutually inhibit each other and form these sort of parasygmental boundaries, which are a pre-pattern, a molecular pre-pattern that reflects the formation of the segments later on. And you can see this is already after gastrolation, the embryo has wrapped around the posterior. Now here, the interior is here, and you get 14 stripes in this molecular pre-pattern. This is a classic, of course, system, and we know a lot about the genes, their interactions, and also sort of the principle by which it works. And for years and years, it was thought to be the sort of French black patterning system. So big weight is a classical morphogen. So it was sort of the discovery of big weight and its expression pattern was a triumph because it confirmed a model that was formulated by Lewis Wolpert much earlier in 1968 and 1969, in which he said he wanted. So Wolpert's idea was to refocus the sort of attention of developmental biologists from temporal gene regulation, which came from Jacoby's and Mano's paradigmatic example of the lack opera. Wolpert wanted to refocus our attention back to spatial, problems of spatial pattern generation. And what he did is he assumed a very simple sort of situation in which you have a tissue, there's a source of a substance called morphogen, a morphogen on one end of the tissue and a sink at the other end. And if you have no degradation in between what you get and no production of the morphogen in between what you get is a linear grade. So this sort of linearly decreasing slope of morphogen concentration across the tissue. You can then imagine. So this is sort of the first step development involved with model is fundamentally seen as a two step process in this first step. The cells in the tissue can sense different concentration thresholds. Here is one here is one in that morphogen gradient, and they will then in a second step, the differentiation step switch on different target genes which are represented by the different colors blue, white, and red here. And that forms, of course, the French flag what is very important in this very simple. This is a model of development. It's a static model, but it's a cartoon model, but it's a very strong tool, thinking tool that influenced generations of developmental biologists who are looking to try and identify different morphogens and different developmental systems based on the prediction that you should be finding those. What's important here in this very simple model morphogen thresholds correspond exactly to where the target domain boundaries will lie in the tissue later on. And in this way the morphogen gradient can be said to encode positional information, especially the thresholds here. By reading the concentration of the morphogen, the cell somehow knows where it is in the tissue. So the gradient contains information for the cell to read and interpret by switching on differentiation genes, target genes. The French flag is a classic sort of model of spatial pattern formation and the segmentation gene network with its maternal gradients and the gap genes that were switched on later on was sort of a paradigmatic experimental confirmation of this classical model. Of course, there's much more to the segmentation gene network and luckily we have this absolutely beautiful Nobel Prize winning work by Ms. Lime Bollhardt and Eric Grishaus, who performed the saturation mutagenesis screen, and then classified the resulting mutants into these different classes that we saw before, but also examined every single sort of mutual interactions between different genes. So we get a very complicated network that not only involves these sort of hierarchical interactions, but also cross regulatory interactions among gap genes, for example, and even auto regulatory interactions of the segmentation genes themselves. So we get a rather complicated network that creates a pattern. And this of course is already a big step forward from this sort of simple explanation of genes as switches. Okay, so we could call this genetic mechanism version 2.0, where you are saying we have now a genetic regulatory network or GRN very popular now. These genes they interact together and what's important are the interactions that determine the sort of pattern and the dynamics of the process. This was famously first suggested that development works like this by Britain and Davidson in 1969 and this absolutely fantastic pioneering science paper I recommend you read it. And this is one of the figures where they tell us that development is basically to be explained as a sort of an electric or electronic circuit. So this is a completely different type of explanation where the focus is no longer on the activity of single genes, but an activity that comes out of the organized sort of interactions between different fact. So, but still, all of this is based on genetics only. So what you do is you perturb healthy system, you check what's going wrong, and then you try to sort of read out what the healthy system is supposed to be doing. So this is a reductionist approach. You're trying to decompose the system into its component factors. So you try to identify the different genes, come up with a parts list of different genes, and then interactions between them to reconstruct the network. This is a massively sort of successful research approach. The history of genetics is 100 years old now. And, well, if you go back to Mendel even more and has been absolutely, fantastically successful, but it has its limitations. So let's think about those limitations. One of them is that many genetic perturbations are lethal. So they're lethal in a way that's not very interesting. Drosophila is a special case. The embryo doesn't grow, while the segmentation process occurs in the flower beetle, triboleum, the embryo elongates while it's segmenting, and most of the mutations, or a lot of the mutations of segmentation genes simply have a growth effect. So if the embryo doesn't elongate, there is no segmentation, but that's not very interesting for reconstructing the network that creates the pattern, because you simply have no pattern, not very informative. So we need informative mutants that have a specific effect on the process that you're studying. But many mutations, on the other hand, making things worse, do not have a phenotype, especially bad invertebrates with their genome duplications. So about a 30% at least of the genes in mice do not show any detectable phenotype at all. But to assume that these genes don't do anything, that's premature. We don't know that. There's a lot of redundancy built into the system, and genetics really hits a limit when trying to decompose such redundant systems. Furthermore, many interesting developmental phenomena are not associated with any specific mutation. In the second example that I'll show you, I will have a specific case point to make the point. So it has been a little bit lost on us, not the classical embryologists, that not all phenomena we want to explain in development are genetically determined. Okay, so we need to come back to that. Another really big problem is that if you've already gathered some experience in a lab, then you know that it is really difficult to interpret mutant expression patterns. Again, an example here from the gap genes in Drosophila. Let's take a wild type embryo, here's my simple depiction of it and the expression pattern of the gap gene Kruple in the middle. Remember that picture that I showed you earlier. Kruple interacts with another gap gene called hunchback, which is both maternally and zygotically expressed, which makes the genetics really complicated. And so if you mutate both the maternal and the zygotic contribution of hunchback you will see that Kruple is moving into the interior of the embryo here. So it's expanding into the area where hunchback used to be expressed. I should say it's expressed in the interior of the embryo. So how do you interpret this? Okay, so you can do the reverse experiment. You can say, okay, I overexpressed hunchback using a heat shock promoter. So now hunchback is everywhere. And in this case you can see compared to the wild type, the interior boundary of Kruple has been pushed back towards the posterior of the embryo. How would you interpret this? You would say, okay, hunchback is gone, more Kruple in the interior, hunchback is overexpressed, less Kruple in the interior, hunchback must be a repressor of Kruple. However, at the same time, in the hunchback mutant Kruple gets a lot weaker. You can see that even with qualitative assays. And also if you overexpress hunchback, Kruple expands very much to the posterior where hunchback didn't used to be. Okay, so this indicates that hunchback at the same time as being a repressor is also an activator. Okay, what's going on? There are two possibilities we could imagine. One is that hunchback acts as a repressor at certain concentrations, but maybe as an activator at different concentrations. So interpreting this, people have said hunchback acts as a repressor at high concentrations, as an activator at low concentrations. Could be. But it could also be that we've forgotten the third factor. There is another transcription factor encoded by the gap chain Knurps, which is expressed just posterior Kruple. And it is a repressor of Kruple. And at the same time, strongly repressed by hunchback. You can see if hunchback is gone, Knurps expands all the way through the area where Kruple is expressed. Since Knurps represses Kruple, hunchback is gone, you derepress a repressor, you get an activation. Double negative is a positive interaction. And consistent with this, if you overexpress hunchback, there's no Knurps, which explains the posterior expansion of Kruple in December. So basically the activation could be indirect through the repression of a repressor. There is no way you can do genetics and molecular biology and get around this problem of not knowing exactly whether an interactionist, director or not. Molecular asses will show you whether Kruple, Knurps, and hunchback bind to each other's binding regions, but they won't tell you whether they're activating or repressing. Okay, so to piece a whole system together again is very complicated, especially if you consider that there's more factors here. Another gap gene, giant and terminal gap genes that are also involved in a regulation of these genes. So the situation is much more complicated, even in this very simple system. The last problem with genetics is once you've sort of worked out doing a lot of genetics and molecular biology, you've worked out an entire network. So what I'm showing you here is a bit of a dated sort of picture in 2010, but it nevertheless gets the point across. It's a mapping all the interactions between maternal coordinate genes. It's just a great coddle. And then gap genes, giant Kruple Knurps, hunchback here, and other factors and you can see it's very complicated. Basically, you've done all the work, you've drawn the summary diagram of a network, and now you're supposed to tell me how this works. Okay, it's very, very difficult to understand what this network actually does. It's much easier to reconstruct it to get to this point than it is to understand what this does. This is the grand challenge. So what we're having here is not an explanation in this sense, but the starting point, a challenge to understand what such a network is actually doing. These are all sort of limitations that are intrinsic to any reductionist approach. You're decomposing the system, you're localizing function, but the task is now to sort of put all this together again and find out what it actually does. And one of my favorite philosophers of science, Gandalf from the Lord of the Rings, has put this beautifully to the point. He that breaks a thing to find out what it is has left the path of wisdom. This is the Saruman and where reductionism leads can be seen in this beautiful illustration of Isengard flooded by the ends. It does not end well. So we cannot stop there by just taking everything apart. We do not understand anything. That's the take home message here. So to put it slightly differently, this is how we understood sea urgent development in 1967. You have a beautiful sort of series of stages. We have an understanding of certain cytoplasmic determinants that are localized and that lead to the segregation of different lineages, such as the micro mirrors here in red that will go and ingress into the embryo during gastrulation and then form the skeleton of the embryo. Beautiful. And it hatches here. So this is a sort of a descriptive understanding of development with a bit of a mechanistic understanding that there is something like these determinants that differentiates different areas of the embryo. This, of course, is how we understand sea urgent development. This is a picture that I downloaded in 2016, but it hasn't changed much since then. This is how we understand sea urgent development today. And you could argue that, you know, I mean, you have a gene regulatory network, the different boxes here represent different tissues, different time points, and you have a lot of interactions. You can argue that this isn't, in a sense, progress, right? Well, we've abstracted away a lot of things like the geometry, the cellular interactions, the sort of, you know, complicated behavior of those micro mirrors ingressing, crawling into the inner, the seal of the embryo. And it's all gone the tissue geometry, this sort of very proactive behavior of the cells, all of that is gone out abstracted idealized away out of this model. So not only have we gained a lot of genetic knowledge, but we have lost some context. And how can we put this together again? So I would say if you do this, this is not system spellage. And so this, this sort of advanced from the action of single genes to networks hasn't really overcome reductionism at all. What it is is doing reductionism at a genome-wide scale. So you look at all the factors involved, you map all of their detailed molecular interactions, you decompose the system, you've localized functions, but you still don't know what the whole system does. And I would call this, to quote a dear colleague of mine, Janne Hofmeyer, I would call this system-wide biology, not systems biology. The difference between that and true systems biology is simple. Systems biology worries about what this network does. But if you go to any sort of systems biology meeting, often what you get is you get a lot of these sort of network graphs thrown at you, they're called hairball graphs, just like something that your cat would puke up after eating a bird. And people are impressed because they show, you know, the system is complex, you've done a lot of work to draw this graph, fantastic. But it doesn't really tell you what the system does. And to do that, that's a really grand challenge. So let's take a very, very, very, very simple network. Three genes indicated by different color dots here and interactions between them. All of these genes are constitutively activated, and they are transcription factors that repress each other in the way that's depicted here. So you have these sort of mutual repressive interactions. And it has a funny name, this little circuit, it's called an AC-DC circuit. And it was named like that by James Briscoe and his group when they discovered and described the circuit in 2012, 2012. And it's called AC-DC because it does both positive and negative feedback. So just like the current AC-DC, it does sort of steady switch between the green and the red gene. And if these interactions here are very strong, you have a double negative interaction. Remember, double negative is a positive. So this is a positive feedback group. And because these two factors hate each other, you get an expression of either the green one or the red one, but never both at the same time. So it's like a switch, a toggle switch. While this sort of second feedback group here between the three genes, three negative interactions that's a negative feedback group, they can create oscillations just like you have AC or DC current in an electric circuit. So a very simple circuit, but it can show many, many different behaviors. So it can switch, it can oscillate. So what you see here is the situation of the switch. Interestingly, it can have a switch that's not stable, it goes back or back and forth. This is called a relaxation oscillation. It can just create, if the interactions in the negative feedback group are stronger, it can create oscillations that are damped, that are stopping after a while, or that are sustained or stable oscillations. So you have at least four different qualitatively different behaviors that come out of the simple system. And I haven't even mentioned the most common and most typically absurd behavior, and that is that all genes are just dying off and nothing happens at all. Okay, so five, in reality, this simple circuit can exhibit five qualitatively different, what are called behavioral regimes, completely qualitatively different types of behavior. And so imagine that if you want to understand this huge Eric Davidson network about the sea urgent, you need to understand how this sort of behavior comes about, which is difficult for even such a small circuit. So this figuring out what the circuit, what the system does is is true systems biology in my opinion. So, to wrap up this lecture, we've come to a place where we have to introduce mechanism 3.0 if you want to have a causal a mechanistic explanation of a developmental process, you need to do modeling. Surprising. Okay. And here is why this is work by two philosophers Bill Bechtel and Adele Abrams Abrahamson and very, very interesting work. And it's about how do we explain things in biology. Okay, and so they say we need to take into account the parts and operations of the mechanisms of the parts or the genes, the operations are their interactions. They're spatial organization. Of course, if you have an embryo like the Drosophila embryo, you have to take into account where the interactions are happening. And the pattern of change over time in properties of its parts and operations. That's very important. So the interactions, the concentration of the different components changes over time. And so the interactions will change over time and you need to track that as well. And that results in the orchestrated behavior of the mechanism. So how do you track those spatial temporal interactions? You need mathematical modeling and dynamical systems theory for this. So not just any mathematical model, but you need a model that tracks all these different interactions happening at the same time in different locations at different times. Okay, you cannot do this in your head. As soon as the system is above a certain complexity. So you don't need to model if the system is very simple. But in the case of the gap genes here, we've seen that it's very hard to track all the interactions at the same time. And they say such modeling provides understanding beyond that which is available from identifying the parts, operations and organization of the mechanism. So deep composition and mentally rehearsing its function. So your ability to mentally rehearse the functioning of a mechanism is extremely limited in other words. And you need computational or mathematical modeling to help you. So here's a basic recipe for what they call dynamic mechanistic explanation. You need to do all the decomposition. You need first the parts and their interactions. You need to draw this hairball. You need to reconstruct that. But, and that's very important. That's just the first step after decomposing the system you need to re compose it again. So you need dynamical systems modeling, which is not in itself a sort of a causal mechanistic explanation a model without evidence is not an explanation, but a model combined in this way with mechanistic decomposition amounts to an explanation and mechanistic explanation of the system. This is the last point I'm going to make so that the sort of, you know, reductionist perturbation approach like genetics will never give you sufficient explanation so you will never be sure whether the parts that you've assembled are sufficient to explain the behavior of the system. It's not necessary. You've knocked them out in a mutagenesis screen. You've targeted them by reversing genetics, whatever you do, you know they're necessary, but they're not sufficient. You need a mathematical model to get a sufficient explanation. So, if you want to make an explanation of a developmental process counterintuitively, you not only need to let Humpty Dumpty fall off the wall, but you need and that's the much more difficult task to put Humpty Dumpty together again and see if the interactions actually do what you think they're doing. To summarize this part of the lecture, I told you that development provides a mapping between genotype and phenotype. Developmental processes can be represented by interacting dynamic regulatory networks. That's how we like to represent them. The components and interactions represented by the network implement a developmental mechanism. So different components, their activities and their organized behavior. And these mechanisms are too complex for us to decipher just by mental simulation or rehearsal and we need computer models to understand. Very simple. I'll provide two examples in the next two lectures that illustrate this point. So if you have any questions, comments, please email and you can follow me on Twitter at this Twitter tag as well. Okay, I'll see you for the next lecture. Bye bye.