 Hello, and welcome back to Beyond Networks, the evolution of living systems. So far in this lecture, we've been moving around extremely abstract spaces. So I've been telling you in very philosophical terms what an actual system is, a pattern process, and how we can model such systems using formalisms such as network theory and how we use these abstract models, idealized models as tools rather than as accurate representations. Their main use is that they give us certain perspectives that help address certain questions or problems that we may have. So what we want to do in this lecture is we want to connect to the main topics of the lecture in two ways. One is we want to reconnect this whole argument to developmental evolution. So the evolution of developmental processes. And my other aim for this lecture is to connect this very abstract and broad idea of a mechanistic explanation of a phenomenon to the idea of explaining a developmental process in causal terms. So a few years back, Gunther Wagner and a number of colleagues stated that evolutionary developmental biology is a mechanistic science. So let's think about what that actually means. So they write, developmental evolution opens up new areas of research. Fair enough. In elucidating the genetic factors that are responsible for the origin of evolutionary innovations and in particular the origin of new character, a focus on genetic factors. We can argue about that. We will argue about that. Responsible for origin of evolutionary innovations and origin of new characters, innovation, novelty. So we want to understand the causes of evolutionary novelty in a mechanistic way. What we do this day is the main methodological challenge of this research program is to link developmental mechanisms to evolutionary processes. So the mechanistic here means mechanistic explanations of development. And as we saw in the last lecture, it's not a machine-like interpretation of development, but rather a causal explanation of the phenomena of development. So to put it another way, philosopher Ingo Brigand has this nice quote here, which makes a very similar point. Evo Devo does not just lay out phylogenetic transformation sequences of morphological characters, but offers a causal explanation of how those character transformations occurred by means of changes in developmental mechanisms. So not only do we want to sort of record evolutionary transitions, but we want to explain them what caused them, how did they happen in terms of mechanisms of development. So what is a developmental mechanism? The problem is in the field, if you go to any Evo Devo meeting right now, when people talk about mechanisms, this is what they show you. Network graphs. And we saw over the last few lectures that I have to take back my earlier sort of claim that this is not an explanation. It is a topological explanation. It explains certain features of a system like antifragility or robustness in terms of the overall structure, the topology of the network. But it is not a mechanism. It does not tell us what it does. The network does. That's the challenge. If we want a mechanistic explanation, we need to know what this network does. And a graph like this does not tell us. So just to drive this, I may be flogging a dead horse, but this is a very important point. So what I'm talking about when I talk about mechanistic explanation, just to repeat that from last lecture, I'm not talking about considering organisms as machines, as clockworks. I'm not considering the computational metaphors that we use. And also, I'm not considering a mechanistic explanation that has necessarily has to be grounded at the molecular level. But I am treating mechanism as a causal explanation. Mechanisms as heuristic models that enable the explanation of causal processes, which realize some biological function or phenomenon of interest. This is very important, this distinction. And remember, those mechanistic explanations, they can be multilevel. They don't have to be molecular. They don't will come to that. They don't have to sort of interpret the development as a genetic program. So what I want to do in this lecture and then also the next module is give you sort of tools to think about how we could come up with with explanations like that, how we could move beyond these type of networks. Okay, because remember, these networks are idealized models. I'm not going to talk about myself here because these points are very, very important. So these networks are idealized models, idealized representations of an actual system, which is a patterned process. So we need to explain that process. That's the first sort of task here. And before we had this flood of networks in Evo Devo, we had a wonderful conceptual framework to do this thing. So it's high time in a lecture about developmental evolution and the evolution of systems that I introduced this classic framework of Conrad Howe, Waddington's epigenetic landscape. It's a metaphorical sort of framework. We look at several sort of attempts to make it more concrete, but it's very powerful. Okay, and of course, most of you will have seen Waddington's very famous visual metaphor. Here it is, most famous picture from the strategy of the genes published in 1957, a book which I cannot recommend highly enough. It's very short, it's very clear, and it's very, very relevant today. It lays out a research program that we have still in Evo Devo today that we should follow. And it opens up a lot of really interesting ideas that we need to investigate now. So as opposed to the network diagram, which was a mechanism schema, it depicted a sort of a complete picture of all the factors that are involved in sea urchin development. This is very metaphorical. It points us to the right sort of questions. As a metaphor, it is so powerful, because it crystallizes what is important, but also it stays sufficiently vague so everybody can interpret it in their own context. For example, look at the axis of these graphs. Have you ever thought about what they represent? The axis that comes towards us here, this is pretty clear. This is time, developmental time, right? The ball will roll down the valley as time goes on, and it will make decisions here, developmental decisions, and it'll end up in a specific valley and adopt a certain fate. But then, if you think about the x-axis here, what is that? Along it, we find different sort of fates, but it's not quite clear what it represents. And even more mysterious, what is the vertical axis here? What does the height of the landscape represent? Some sort of potential developmental value. We're not sure. What if it never tells us? If you think a little further, what is the ball? What does it represent? Does it represent a cell? Does it represent a tissue? Does it represent a system, an organism? We don't know. Also, why on earth is the ball different from the landscape? Because the landscape represents the epigenetic landscape of the system. So if the ball is the system, why is this epigenetic landscape different from the ball? Questions over questions. But let's get back at the main point, which is to crystallize a bunch of questions. And the most important here is to visualize this idea of what Warrington calls a creode. That's a developmental trajectory. It's the valley floor that the ball follows when it rolls down. And what we're trying to do if we come up with a mechanistic explanation, a developmental mechanism must explain the shape of this trajectory. Why does it end up here? Why does it take these turns that it takes? And how does the landscape look around it? Because this is a second feature that's very powerful in this metaphor. This trajectory is buffered. Warrington called the process, the mechanism that buffers it, homeoresis, an analogy to homeostasis, which maintains a specific point in the system. Homeoresis maintains a specific trajectory. It's kind of like a dynamical version of homeostasis. And he said, because these developmental trajectories are buffered, they are resilient. He called this canalization. They're canalized, okay? They're stable against perturbations. The last point I want to point out is these sort of epigenetic barriers that the hills between the valleys, okay? So they represent the energetic barriers between different states during development. I don't want to go back to all of this, but this sort of focuses very clearly on a bunch of features of the dynamics of development. Okay, the second most important picture in Warrington's work, of course, is this underbelly of the landscape, where you see the landscape is sort of connected to pegs that represent genes by this intrinsically, intricately interwoven network of guy ropes. So if you have a random mutation in a gene, it'll have very sort of complicated, non-random effects on the landscape. And the task that we have here, if we want to study the evolution of developmental systems, is we need to explain how the landscape changes in response to genetic mutation. So random genetic mutations have complex, non-random effects on the shape of the landscape. This is what we need to explain. So in terms of its conceptual depth and richness, this is much better than this world of networks in terms of getting to the point of what we really want to explain in developmental evolution. And this is based, of course, on the causal completeness principle that I've already introduced in an earlier lecture when we talked about perspectival truth in developmental evolution. This is a fundamental truth. It's that in order to achieve a modification in adult form, evolution must modify the embryological processes responsible for that. Therefore, an understanding of evolution requires an understanding of development. Okay, there is no arguing with that. So to come back to Waddington's landscape, this is a different depiction from its 1956 textbook. You can see the ball is egg-shaped here. We have different valleys that represent different tissues. We, the task is to explain the shape of the developmental trajectory, which represents what we could call a generative process. Okay, it generates a phenomenon through, it's a mechanism through generating a phenomenon. And so what we need to do is we need to understand the generative principles. This is how Brian Goodwin puts it. This is a school of thought that is called process structuralism will get back to it. So it's different from mechanistic thinking, but very similar. This quote, organization of matter is important, together with principles of directed or asymmetric time-dependent transformation. And that reminds us very much of this sort of basic definition of a mechanism in macabre at all. Mechanisms are entities and activities that are organized, so organization of matter, such that they are productive of regular changes from start or set up to finish or termination condition. So these are the principles of directed or asymmetric time-dependent transformation. So we need to understand the organization of matter that brings about the rules that shape the developmental trajectory, the generative process of a phenomenon. Okay, so this is a very sort of general way in which mechanistic explanation connects, but there is lots of pitfalls and problems here. First of all, this sort of very basic idea of a mechanism that we introduced in last lecture sort of assumes that there is a fixed stock of entities and parts that participate. So you have a parts list and that part list doesn't change over the time of development. A lot of neo-mechanisms philosophers argue that this is not a real problem, okay, but it was in the way that it was initially formulated. The second assumption, and that's a very important assumption here, is that you can find clearly delineated and separate parts in the mechanism. Who is to say that you can actually separate different components of a development mechanism? Very clearly. This is an assumption, right? And this is maybe a problem. Also, we need to be able to clearly delineate the boundaries of the mechanism. And here we're back where we were before, okay? So we're back at this problem of when we define a formal system on a real process. What we include, the nodes, the interactions, the edges of the network that we include always depend on the context on the question we're after. And there are many different choices. Okay, so this leads us to this sort of more general problem. It's a problem of correspondence. Is there really a mechanism for each phenomena in development? So this problem gets worse if we think about a mechanism must have a specific organization of its parts and activities. While in developmental biology, this is not a given. If we compare the mechanisms, for example, the mechanisms of somitogenesis, invertebrate, embryos, the process that creates the vertebrate body segments, these mechanisms are very different, but they do the same thing. And this is because of an evolutionary phenomenon called developmental system or network drift, which means it was first described by John True and Eric Hay in 2001. And they write, numerous recent studies have revealed the surprising result that developmental pathways do, in fact, diverge through time, even with no accompanying change in phenotypic outcome. This process we call developmental system drift. So you have different mechanisms doing the same thing. And all together, all these blue points on the list here, they point to one problem, which is that there is not a one-to-one correspondence between a phenomenon in developmental biology and its mechanism. There are always many different ways to define a mechanism that does the same thing. We'll revisit that problem. We call it the problem of correspondence. Okay, there is a different problem though. And that is this very simple definition of a mechanism. You know, the entities and activities, they are organized in such a way that they bring about regular changes from start or set up to finish your termination conditions. That implies a very sort of linear sequence of activities. And of course, at a certain level, we look at this trajectory. We can say one thing happens after another. But if you look at the sort of underlying process that is creating the geometry, generating the geometry of this trajectory, it is very complicated. It is not quite as simple. For example, here again, the sergin endomizoderm network. So it's not just a simple way of, you know, there's a simple task to connect this very complex network to this linear sort of causal sequence of events. We'll talk about causality and complex systems, not in the next module, but the one after that. So it is a very important topic, but it's not a trivial problem at all. I hope you see this. So we need to come up with a broader conception of a mechanism that includes this complexity. And also the fact that these complex networks are heavily driven by feedbacks. Basically, here is a model, a very abstract model of the whole organism, Ganti's chemoton. We'll come back to that when we talk about organisms. And it's basically one, all you need to know, to see here in this graph, it's one feedback cycle within another. So you have a metabolic cycle, you have an overall cycle, a closure of interactions in the whole living systems and so on, a regulatory template replicator cycle here and so on. So living regulatory systems within living beings are feedbacks within feedbacks within feedbacks. So there is no simple connection to this linear sequence of events that is suggested by the basic account of mechanistic explanation. So we need to consider all of these things. And so I'm giving you, I'm repeating here, the slightly more complicated definition of a mechanism by Bilbeck Dell and Adele Abrahamson, which says, okay, so in the beginning, it's pretty much the same as McCommer. It says a mechanism is a structure performing a function. Okay, so a structure, fine, it's a thing. In virtue of its component parts, component operations and their organizations, there's no problem, there's no discrepancy to the earlier definition of a basic mechanism. But here comes the new stuff, the orchestrated functioning of the mechanism manifested in patterns of change over time in properties of its parts and operations is responsible for one or more phenomenon. Manifested in patterns of change over time in properties of parts and operations. So the structure changes over time. Okay, the composition of the mechanism and its structure can change over time. For example, so this means that we're sort of focusing away from the components, the parts of the mechanism through the interactions and how they result in this sort of organized activities that produces the phenomenon. So there's a shift of focus in explanation from the structure, the thing that the mechanism is to what it does. Okay, and so there's a structure, a change of focus from explaining what the thing is made of to what it does. And this is exactly what we need. So Bechtel and Abrahamson call this a dynamic account of mechanistic explanation. And it solves this sort of problem that we had with the complex networks creating this linear sequence that's the problem of diachronicity again. We've been there before a static network graph, like the one I'm showing you here doesn't show you what the process is that it represents. Okay, and the other thing is there is no static network. So the network is transient. You can have the generation or elimination of all kinds of entities during development or entities change over time. So you can have a change in protein concentration, for example, or you can have a protein modification that turns one factor into a factor with a completely different activity. Okay, so you need to take this into account. We need a more dynamic idea of mechanism. Systems behavior changes as conditions change. So you get an environmental trigger and inductive signal and the behavior of this network. Remember the mantra structure does not determine function structure does not determine function. So if you get changing conditions, even a simple network can certainly switch behavior and do something different. Okay, so this is the second problem of network graphs. It's sort of the problem of diachronicity. We need to have a different approach based on something like Waddington tells us to focus on what this process does that is represented by this idealized network. The network, remember, is not the real system. Okay, so to conclude, what we need is dynamic mechanistic explanation and back down and Abramson have a really nice sort of argument on this that I'm going to present here to finish up. Basically, we need to repeat here a little bit. We need to take into account the parts and operations of the mechanism, the spatial organization and the pattern of change over time and properties of its parts and operations. So it becomes much more dynamic and those changes result in the orchestrated behavior of the mechanism. Focus is on the dynamics. The mechanism is describing a process now. Okay, so we need mathematical modeling and dynamical systems theory for this because we cannot mentally rehearse or simulate complex systems like the one I showed you for the CRG. Such modeling provides understanding beyond that which is available for my identifying the parts, operations and organization of the mechanism and mentally rehearsing or simulating its functioning. So basically mechanistic explanation not only involves the classic mechanistic decomposition. So you take the system apart into its components, you localize function activities, the specific components, but also to put those components together again to recompose on the right-hand side and recomposition works through dynamical systems model. So because if you only decompose, you get a parts list, a list of interactions that are necessary for the phenomenon that you observe. You know that these genes are involved in forming the phenotype you studied, but you do not know if they are sufficient. You only know that if you put them all together again, you simulate that and you find out that it actually does what you think and this is what dynamical systems modeling is for. There's a lot of debate about what kind of explanations are provided by dynamical models. I'm not claiming those models on their own are mechanistic explanations, but here although on their own they're not and we'll come back to that when we talk about how we classify dynamical systems. They're not mechanistic explanations on their own, but they play a role, an essential role in dynamic mechanistic explanation. So we have to take the system apart, but we also have to put it together. So counter-intuitively you need modeling for mechanistic explanation in this sense. Okay, to wrap up, I'm going to quote some of my own work with philosopher James DeFrisco. We've written about this and basically there are three problems with gene regulatory network graphs as mechanistic explanation, as they're often considered right now. If we want evo-devo to become a mechanistic science, we need to move beyond those networks. That's the central message from this lecture. Because these mechanisms themselves, these graphs themselves aren't mechanisms. There is the problem of genetic determinism. They only involve genes. Development doesn't only involve genes, they need to include non-genetic factors. We haven't talked about this yet. We'll come back to it, but it's pretty obvious. The second problem, more abstract that I talked about in this lecture is the problem of correspondence. There is no reason to assume that there is one mechanism that is responsible for one phenomenon in developmental biology. Many different phenomena can be produced by one mechanism under different circumstances. Developmental plasticity, while different mechanisms can produce the same phenomenon. This is robustness. We'll get back to that as well. Third, we have the problem of diachronicity. These graphs simply, they're static, and what we need to explain is dynamic. We need to explain the shape of the trajectory of the process, and they don't do that for us. A network graph can explain why the network is robust, resilient, modular, whatever, but not what it does. This is a basic take-home message here. So we end with a quote from our paper, and it says, Gene regulatory networks fail to provide a robust mechanistic and dynamic understanding of the developmental processes underlying the genotype-penotype map. A mechanistic understanding of more complex systems must therefore rely on computational models of network dynamics. That's it. So what we're going to do in the next module is I'm going to introduce some of the basic approaches of dynamical systems theory and the concepts, such as attractors and their bifurcations that help us explain dynamic behavior. It's going to stay at a very sort of conceptual level, no mathematics, really required, and then we'll move on and connect those modeling efforts to back to mechanistic causal explanations, and we'll ask the question, how does causality flow in complex systems, such as those involved in organism development? Okay, I hope you tune in again for that. Thanks for listening and bye now.