 to the organizers for inviting me. I've really enjoyed all the talks and I'm really looking forward to the remaining two days. Yeah, the title is quite a mouthful but I'll explain what I mean shortly. I'll start by a brief summary of in general what I'm interested in, kind of my research group. Broadly speaking my research program is that a physicist was fascinated by the unintuitive microscopic world, except much like my colleagues in string theory, except for me microscopic world stops at bacteria. And like many people here in the audience, I'm interested in things like large an ecology, ecosystems with a large number of species or resources, and as such I'm kind of part of this community. I count myself as part of this broad community of people with shared interests, many of whom are actually here. But then also, I'm like many of us as well, I'm interested in this interplay between ecology and evolution and in fact, one way to interpret the title of the workshop, I suppose in terms of limits of community assembly, I interpreted as what are the limits to our ability to predict community assembly, what are the factors limiting our models and for sure, our understanding of evolution is one of those factors, such as ecology of players that are evolving and co-evolving, how do ecosystems evolve? Or in fact, we are not even that well, don't understand even that well, what role evolution actually plays in given ecosystems? So when I was listening to Leonardo's talk, I couldn't help but noticing at first on the slides of factors shaping microbial communities that there were multiple factors there, but not evolution actually wasn't one of them. And then of course, I waited a few slides and she blew our minds by this last latest preprint that they posted. So evolution was very much part of the story as well. But yeah, so this interplay is something I'm quite interested in. And even simpler situation than the full feedback loop between ecology and evolution, just evolution and a changing environment, which is kind of before we complete the feedback, environment can change because of the organisms that evolve, but even if you don't complete that loop and just think about evolution and changing environments, there's surprising variety of problems that we have a lot to understand about. So yeah, and all of this is kind of under the broader umbrella theme of what are the right course grain variables in these situations? And of course, there's the conversations of function versus composition, and I heard some beautiful work on that front already. And my personal kind of pet topic is to what extent, you know, the familiar language of species is the right language for some of these questions. But anyway, so that's that's kind of the broad themes. But lately, so lately, I was actually my work on large an ecology in collaboration with Raymond Monason, for example, that dates a few years back. And lately, I've been mostly interested in evolutionary questions. So what am I doing in this workshop? Well, even in evolutionary contexts, the sort of ecological motivation is never too far because I'm really interested in microbes and ecology, high diversity ecology is always kind of a persistent interest. So as an example, here's a recent story that I worked on with some colleagues at Stanford, Shaman Kachur and Daniel Fisher. And on the face of it, it was entirely an evolutionary question, which is, you know, evolution is constrained by performance tradeoffs, but performance tradeoffs themselves can evolve. And, you know, can we have some null model for this interaction? And what are some general expectations? So we have a paper on that. Also kind of describing this model for multi environment evolution, which is different from kind of officials geometric model and incorporates some ability of organisms to respond to their environment. And this kind of cute statistical physics the way I'm describing this just as an example that, okay, nominally, there's no ecology here. But secretly, this is kind of essentially MacArthur resource competition model, even though it's not in the paper. But, you know, MacArthur's resource competition has this interesting feature that the dynamics of individual species, optimize this global function, these dynamics have a lappin of function. It's something that I always found intriguing to think about. And of course, you know, maybe it's tempting to interpret that as some sort of community fitness. But of course, you can do that. And of course, in general, ecosystem dynamics don't optimize anything. But, you know, you could then take that model and basically reinterpret it as a model for an organism and its fitness kind of replacing species abundances by expression coefficients. And then you get a cool model for studying evolution, which is essentially ecologically inspired. So this was just as an example of ecological motivation kind of percolating into a different context. And what I want to tell you today is another example of something kind of similar. So this is joint work with Stefan Landmann and Caroline Holmes. We also just posted this pre-print a few weeks ago. It's very fresh. Now, in this story, again, there actually won't be any ecology and not even any evolution. So on the face of it, it'll be a question about some physiology of bacteria. But I'll ask you to bear with me through this story that I'll tell you about. There'll be a lot of ecology in this workshop. So, you know, let me tell you about something entirely different. And then I'll argue that there's actually, I'll then map this onto an ecological narrative. So the story is published is completely a physiology of bacteria kind of question. But I hope to convince you in the end that there is a, sorry, I'm trying to, my screen has so many zoom windows on top of each other that I can hardly see my own slides. But okay, whatever. Okay, yeah. So for the portion of the story where, like a purely physiological, I'll sort of change the color scheme to remind you that we're not yet in ecology and are thinking about a completely different question. And then I'll map back to ecology. So the question I will ask for the next 10 minutes or so, I tried to prepare not too many slides. So I'm counting on questions in the middle, or I'll end way too soon. The question we asked is, can bacteria learn the fluctuation structure of their environments? And I will explain what I mean by that. So if I live in a fluctuating environment, the adaptations that I might have involve not only reaction, but also prediction of the future state of the environment. So for example, you know, if I see this, I'm not, I don't wait until I get soaked. I grab an umbrella. And we are not the only ones who do this. Basically, all living systems make predictions about their environments effectively. Any predictable pattern in environment and environmental change can be exploited to improve fitness. So my, a few favorite examples of mine, and there, of course, many more, you know, quite obviously there's circadian clocks. So both plants and algae have reorganized their metabolism in preparation for the sun that's about to rise. And when it does, they're already ready to go. There's also what's called anticipatory behavior in microorganisms. My favorite example is, so E. coli, its life cycle takes it through the mammalian gut. And then they're out in the soil until yet eaten by another mammal and the cycle repeats. So there's work showing that if you expose an E. coli to a warmer temperature, it starts activating pathways that prepare it to anaerobic environment. And what's happening kind of the interpretation is that if it gets into the mouth, it's preparing for the gut environment, right? And this association can be broken in a few days evolution in the lab, where you fluctuate oxygen availability and temperature independently, you get strains that do not have that association, right? So it is a learned association. And in fact, these kind of learning these statistical features of your environment can be beneficial even if the environment doesn't change in a sort of regular predictable way, but is entirely probabilistic. So, and a good example of that is the sort of bed hedging idea. So if I, for example, if I'm in a bacterium and I live in an environment with when with some probability I might see an antibiotic, the optimal strategy, and I could adopt a persister phenotype that doesn't grow but would be resistant, the sort of optimal strategy with what probability I should be doing that depends on the sort of statistics of the environment. Okay, so all of these we know all these examples, basically there's this kind of old metaphor that evolution can learn any features of the environment. And for all we know basically any statistical feature of the environment could plausibly be learned by an evolving organism through this sort of trial and error, right? But so I'm saying can learn the statistical structure of environment, but only the kind of structures that change slower than evolutionary timescale. So, you know, the 24 hour cycle has been going on forever and ever. And so long enough for evolution to be able to learn it. But what if there are regularities in your environments that change faster than evolutionary timescale? Now, how, you know, how relevant is this question? Are there regularities in the environments that change faster than evolutionary timescales? Well, yes, of course, there are. Basically, for example, entire ecology operates in this regime, right? Who's around you changes on an ecological timescale. And the statistics of your environment thus are also changing on that timescale. So that's the theoretical question we wanted to ask in this project, which does remind you at this point is just a kind of theoretical question of, you know, what is possible for bacteria to learn about their environments? Fully within the scale of one organism, nothing ecological about this just yet, except the context, like sort of this ability would probably be most useful in an eco evolutionary context where your environment keeps changing because of eco evolutionary feedbacks or just ecological feedbacks. Okay. And so, so let me describe to you the, you know, being theorists, we made a sort of toy version of this question. So let me describe this toy version of the question to you. And then I'll pause to see if there are questions. So to model the situation where an organism lives in an environment that has some structured fluctuations, let me consider the following problem. Let's say that a cell like a bacterium has some internal quantities that I'll call P that ideally would be matched to a fluctuating external quantities, which I'll call D. So D is my environment. And P is something internal that I'm trying to match to that environment. Why P and D? Well, for example, you can think of it in some sort of metabolic analogy, where the internal you have production of amino acids, for example, right, and your environment imposes certain demand right now for all those amino acids. And ideally, you want to produce, you know, as much as needed, but not more, not less. And your environment fluctuates. So now fluctuation of the environment is encoded in this fluctuation of D. And I wanted to fluctuate in some structured way. What's the simplest way to have some structured fluctuations? Well, let's say that it's undergoing a random walk with parameters. So this is kind of a matrix of restoring force. This is a noise term. And intuitively, you know, D kind of fluctuates around filling out an ellipse, you can think of it this way. You know, in physics terms, it's random walk and a quadratic potential in intuitive terms, I have this quantity that's fluctuating around, and some directions are more likely than others, right? I rarely observe my D here. I more frequently observe my D there. And so the problem that my organism is facing, the structure of the fluctuations is encoded in this ellipse, essentially, and the strength of fluctuations. So both how the ellipse is oriented and how big it is. And the organism is trying to match P to D. And so I have a little animation describing this. So, you know, you can't react instantaneously. So you're kind of lagging behind a little bit. But your environment is, that's the sort of the blue dot here with a black trace. I meant to make it blue, but it's black. Sorry about that. And the organism is trying to do something. And maybe it could be doing what this green dot there is doing. And the question is, well, how well can I do that? And the beauty of formulating the problem in this simple way is that we actually know in the simple context what would be the optimal thing to do. And basically, kind of, you know, we know the math. So we know what the if instead of a bacterium, I had a supercomputer. Here's how it would solve this problem. From observations of the past, it would learn the statistics, M and gamma. And then it would adjust its strategy to effectively predict which changes in environment are more likely than others. And that would improve its ability to track the system. And I will postulate that basically the ability to track the average deviation is my fitness. And so that's what, you know, an organism would be trying to do. So we know how to do this with fancy mathematics. But of course, the question is, could a bacterium do something remotely like this? And the claim is that actually simple regulatory architecture can solve this task. It's effectively doing both steps here. In fact, it can do it near optimally. You can show that. But most importantly, it's a simple architecture and it's very simple to what bacteria already are, you know, the kind of networks they already know to possess. And that's why it's not a silly thing to imagine bacteria might actually be doing. So for example, you know, this is kind of intended as a proof of principle that, you know, say bacteria in our guts, could they be responding or learning not just the sort of diets that humans as a species have had for, you know, thousands of years, but also learn from recent experiences, sort of correlations and nutrients availabilities that they've experienced in the recent past while being maybe in this individual. And I don't know if real bacteria necessarily do it. This is just a proof of principle that I'll show you that in principle, kind of theoretically they could. So that's the formulation of this physiological question. Let me end. I claim that there is a simple structure that solves this. I will show you this in a second. But first, let me see if there are questions. So I don't see. Yes, there is a question in the chat by Aditya who says, does the dimensionality of P and D affect how difficult the learning processes? What is the appropriate choice for this dimension? Yes. So very good. I will consider first a one dimensional case that just simpler to sort of intuitive the picture. And the two dimensional case, which is basically contains all of the structure that the problem could have because in two dimensional case, you have both variances and correlations. And then in principle, the same structure would work in higher dimensions. But for biological sort of relevance, it seems like one D and two D cases are probably all you need. I hope I answered this question. But if sorry, so if the vector is representing, for example, production of amino acids, couldn't that be like many more than two dimensions? Yeah. Yeah. Yeah. But like so, you know, so in principle, environment is very high dimensional. But say, you know, this as a kind of learning mechanism that I'll describe to you, a given organism might, you know, in its habitat, like these two nutrients might be sometimes correlated and sometimes not. And it might, I might care to know to, you know, adjust my strategy for that. But, you know, I don't necessarily need to be such a universal learner that if anything would suddenly be correlated with something else, I would notice, right? For a given bacteria, maybe there are only two dimensions that are sometimes correlated, sometimes not. And I might care. But like I said, the same idea will work in any dimensions is just for me personally, it seems like low dimensional examples are probably the most likely places. If I were to look for this in a real life, that's probably be one dimensional to dimension. Okay. And it'll how I'll explain in a second, you know, what the solution is and how it works. And you can judge for yourself to what extent it's like biologically plausible. And I see there's another question about, oh, this is a very good question. What does it mean exactly that statistical structure changes slower than evolutionary timescale? So what in this formulation of the problem, you'll see that the fluctuations of D are very fast, and it's just bouncing around, right? But it's bouncing around with a certain preferred direction like this that's determined by the ellipse. And I'm saying that maybe, you know, I'm an organism that I'm an environment for a while, where these two resources are correlated, right? And then at some point, I'm in a different, I like my environment changes and now they're not. So that's what I call correlation structure that changed, right? And it may change. So my ellipse here, two days like this, and maybe in a month, it will be a different one. And a month is too fast for me to learn this by evolution. But I'm saying that there's a physiological mechanism for learning this. And yes, eta in my equation is a vector, very good point. Sorry about that. Okay, let me explain to you how this works, what the solution is. And then also, what's the relevance for ecology? So for now, we're still in the suspended disbelief state of I'm telling you something entirely unrelated to ecology. Okay, so the solution is actually a small modification on top of what all material already do, which is, let's forget the fancy version for a second of math of learning statistics of fluctuations and instead just say, you know, matching production to demand is something that bacteria do all the time. If I need a certain amount of some say metabolite, I'm using it up at some rate, and I want to make sure that I produce about the right amount to produce it about the same rate. How would I do that? There's a standard sort of solution that bacteria use throughout very common, which is I can place. Yeah, so by the way, I will, I won't have time to go into detail. So I'll assume that metabolite dynamics are fast. So I'll just exit just be a routine, but it doesn't really matter. The solution is let me place production under control of some regulator. And let me have the accumulation of the metabolite in my intracellar concentration of X, repress that regulator. And this is the equations. Now this, if my demand for X suddenly drops, X starts accumulating and turns off the production, right? It's a very robust mechanism and it's used throughout in kind of bacterial microbial regulation called end product inhibition. Now what I'm claiming is that if you take this architecture and you add three ingredients, you get a system that can actually learn fluctuation statistics and use that information productively to actually adjust its behavior appropriately in the scenario that I described to you. And these three ingredients are an excess of regulators. So here you would have one regulator dedicated to each metabolite, so in my conditions one or two. Let's say that we have an excess of regulators and I'll require that these regulators have self activation and some nonlinearity in how they're regulated by metabolites. So basically, and again, these are the equations, but in terms of kind of a picture, if instead of just one repressive loop here, I add these auto activation arrows and I also add, here I have two metabolites, but three regulators with some crosstalk between them. And for nonlinearity here, these equations I chose the simplest form of nonlinearity, but we can show that basically a broad class of nonlinearity will work. So I have very limited time, so I won't explain to you in detail like these equations. Let me just show you that it works and provide you an intuition for why it works. And then I'll have the time for the last couple slides to map this back to ecology and tell you what the ecological analogy would be. So, yeah, so the claim is that this system will match any static demand, but it will also learn statistics of fluctuations. So let's first discuss 1D case. In a 1D case, you have, I have just, my demand is just one quantity, D of T, some fluctuating quantity. The only thing that can change the statistics is its variance. So it was fluctuating a little bit and then it's fluctuating a lot and I claim that my system can learn that and respond to that. There's some, okay. And the claim is that faced with larger fluctuations, the system will notice this and become more responsive and be able to follow these fluctuations better. Now, how does it work? So I said that I need an excess of regulators and an excess, if I only have one dimension, I only have, I can basically have an activator and a repressor. That's the only way to have more regulators than, than one. So basically, this is kind of known as paradoxical regulation. It's kind of like I'm driving a car and I'm pressing, you know, to achieve a given speed, I can use the brake pedal and the accelerator pedal simultaneously. So I can just play, you know, press the accelerator and get the right speed, or I can really jam the brakes, but accelerate even stronger and still go at the same speed in this, what seems like a silly way to do that. But, so here the production that I need to match the demand is just the difference between my accelerator and my brake. But now I have an extra degree of freedom, which is I could press both of them more strongly. And the advantage of that is that the mean production still matches the mean demand, but I can change it quicker if I need to. And that's what will happen. So if I expose my system, and these are some simulation results of the architecture that I showed you, I showed to like increase variation of the variance of my signal. I'm kind of in an environment that fluctuates a little bit, then more than more than more. The system learns environment structure by upregulating both brake and accelerator together, which makes it more responsive, and it can track the fluctuations better, which I show here in this plot. So in blue, that's the sort of simple end product inhibition scheme, which doesn't have the other one. And as my environment fluctuates more and more, I perform worse and worse. Whereas if I can become more responsive, I can maintain the same level of performance. Okay. And in a two-dimensional case, it's kind of slightly more interesting because now I have two things that can be correlated with each other, right? And that's sort of intuitive that if I as a bacterium knew which direction of de-flatuation is more common than others, I should be able to track this environment better. And this is indeed what happens. So in this case, so what I'll do is I'll show, expose my system to like, I'll change this angle alpha. What is the direction of the preferred correlation? And as a readout, I'll ask which direction is my system most responsive to? So mathematically, there's the response matrix that has eigenvectors. And I'll ask whether the dominant response direction aligns to the dominant deviation in the environment. And in fact, like for system behaving optimally, it would have to respond quicker to the direction that's the preferred direction. And indeed, this is exactly what happens. So if you squint, there are actually two lines here. In dashed line, that's the sort of true statistics of the environment. And in black solid line, that's the responsiveness of the system, which tracks the input. Okay. And now we're getting to the important part. So if I look at the activities of regulators at this time, it seems something very strange. Like one of them goes up, the other one goes down, and actually understanding why this precise regulator is now active, it's not a trivial mapping. But it becomes clear when we look at the responsiveness of the system as a whole. Okay. So that's my little physiology story. That's how it is in this paper that's out there. And as a little bonus for something that appeals for theorists, possibly, we can show that the performance of this little architecture is actually near optimal in the sense of control theory. Okay. So it's a simple architecture, but it performs as almost as well as a, you know, supercomputer could in this situation, which is kind of surprising. But more importantly, for biological relevance, this behavior, I described to you these three ingredients that are required. And you can get that with minimal modification of circuits that bacteria are known to have. For example, things like two components signaling, which is how bacteria sense the world. This mechanism of becoming more responsive when required is very easy to implement as one additional interaction arrow in that standard regulatory architecture, which is something we are quite excited about. And we're talking to some biology colleagues to see, you know, where would it be plausible to look for something, some behavior like this would be kind of cool if something predicted by theory were actually relevant for real organisms and could explain why we have these superfluous regulators that sometimes are observed. Okay. This is the little physiology story. And I hope I'm still okay on time. I only have two slides left to tell you what, how does this map? My claim is that it maps onto, naturally maps onto an ecological narrative. And this part is not yet published, but we're basically working this out right now and planning to write this up soon. So I showed you some equations. This is a graphical way of representing these equations, a system that with these interactions between metabolites regulators and synthesis of those metabolites can learn a fluctuating structure of the environment. And the three ingredients that I said were required were self activation of regulators, which is some regulatory link in the logic of the network, some nonlinearity and an excess of regulators. Okay. So the self activation of regulators, that's the statement that this a mu dot here is proportional to a mu. So a mu dot is a mu times something. A mu is the activity of the regulator here. It takes a little bit of work to implement that in a genetic regulatory network, but there is a context where this kind of dynamics are actually extremely natural. And that is precisely, sorry, and that's precisely the ecological situation. So the ecological narrative where this maps to is the interaction between resources and species in a standard resource competition model. And the three ingredients that here I said, let's add them, we get them for free in an ecological setting. So self activation of regulators is just the natural replicator dynamics of species. The nonlinearity in activation or repression, which here we had to sort of put in by hand, that's the natural nonlinearity of, you know, this is response species growth rate as a function of food availability. Well, if food is available, I grow, but if I'm hungry, I don't necessarily die the faster the more food deficit I experience. If I don't have enough food, I sort of hunker down, shut down metabolism and, you know, die at some rate, but it's quite, you know, plausible and the dependence would be nonlinear here, possibly, and of course, you know, saturating eventually, but this nonlinearity, we don't need to work hard to have it in our equations. And finally, the excess of regulators that I required here for this learning to happen is just the same statement as I have a diverse pool of species, where the number of possible species exceeds the number of resources. And actually, why was it required here in this case, basically these extra degrees of freedom were required to implement memory in the system. And so the analog here would be that I have some slow modes in my ecosystem from something like approximate neutrality in the sense that, yeah, in the sense that there are slow modes of dynamics. So these are the ecological dynamics of resource competition model that look very similar up to a couple tweaks that were, you know, that's why the published story is yet in this case, and where the ecological narrative will get there. So what is this, what do these results translate to if I recast them in language of community assembly? And that's kind of why I'm excited to tell you about this work and see what you guys think. And so this is a reminder of what happened in this physiology language. I had a bacterium which I exposed to environment the fluctuated and structured way. And if I looked at the regulators, regulators respond to changes in, so here these are epochs where I change the dominant direction I rotate alpha. So these were regulator concentrations. This was my regulatory state of the cell. Then I changed how environment fluctuates. My cell, like all the regulators changed in their abundance. And it was hard to understand this picture. But from the standpoint of the responsiveness of the system as a whole, this was very clear what was going on. The angle of dominant responsiveness was tracking the angle of the dominant fluctuation in the ecosystem. And because it was not an evolutionary model, because I had a notion of performance and sort of fitness, I could show that this ability to learn actually improved performance of the system. The system could use the information it learned to perform better. Now, when I map it into an ecological context of community assembly, what does the same model mean in the context of resource competition? Well, in resource competition, there's no such thing as perform better, because once again, ecosystem is not unless in those remarkable conditions like associated to a host, such as Leonora's system, which is also why I really like those host-associated communities where function of the ecosystem is a well-defined thing. Here, in principle, it is not. And so there's no notion of performing better or worse. So it's not in the picture. But this part remains. And so in the ecological language, what happened is that when exposed to a fluctuating environment, if I ask, now, these are not regulator activities, but species abundances. And if I ask, why was this species successful in this epoch? What is it about, at this point, my environment was somewhere here? What is it about that environment that I observe right now that made this species successful? And the answer is nothing. It is not what the environment is right now that led to the species being successful. It is how the fluctuation structure of this environment was structured that made, that promoted this particular configuration of assembled species. And understanding the mapping in this toy model from environment structure to successive species is actually quite non-trivial. You'll have to trust me on that. It's not just that this regulator is somehow the best aligned to this angle. That's not true. It's a more complicated thing. But there's a community-level functional feature that is actually more predictable, that is easier to predict in this case, that the community as a whole, like this assembled state, is the one where the community happened to be most responsive in precise the direction where the environment was fluctuated. So, this goes back to this question of what are the right variables for predicting what is predictable about community assembly. And I kind of like these toy scenarios where some functional community-level property is easier to predict than details of composition. I also collaborated with Alvaro on some projects kind of of a similar spirit in terms of functional composition being more reproducible. I also have fewer theory paper on something similar. So, I'm broadly interested in these connections. And I'm sure that in ecological context the fact that the way environment fluctuates shapes the community is not a new thought. I'm sure there's a literature on this that I'm not particularly familiar with necessarily. So, I know I'd be grateful for any pointers of how to put this properly into context. But to conclude what I hoped to deliver today is I told you a story which is basically a story of physiology where I argued that a simple generalization over regulatory motif that bacteria are known to have can have this cool behavior. It is able to learn the structure of environment fluctuations and can act upon this information. It can sort of store and retrieve memory in this very simple form that is biochemically plausible with these three key ingredients. But I also for the purposes of this workshop wanted to map this onto the psychological narrative which is that that generates, which at this point I'll frame as a hypothesis, that if I'm thinking about a resource competition type model interaction between resources and species. I presented to you a toy model of community assembly which has this kind of interesting quirk where the species success depends not just on what the environment is but on this extra structure which is the sort of ensemble of likely well of experience fluctuations in the past and therefore likely fluctuations in the future. And it was a situation where some functional community level property was easier to predict than composition maybe in a slightly more subtle way than representation of functional pathways is more conserved than details of composition but in the sense of the dynamics of the system as a whole are sort of predictably shaped by the dynamics of the environment the system is experiencing. Which in this case comes for free from resource competition the laws of resource competition. Which yeah so I emphasize this is a toy model and of course in real life there's much more than community than resource competition that matters and I'm not claiming that you know this is how we will we can now predict things about ecosystems in general but it is a cute toy model that I'm sort of looking forward to discussion see what people think and as a result therefore in this story the statistics of environment fluctuations are sort of seen as an extra force shaping community assembly and determining success of the species. Okay and these are the people who did the work and other people in my group and acknowledgments. Thank you. Okay thanks a lot so there was a hand raised a few minutes ago so I don't know otherwise is there any any any question? I thought I thought Armoni Agat had his hand. I had my hand up I answered my question thank you. Okay but I actually you know come to think of it would you say something like depending on the trait of interest that you're tracking would tracking would look different something like more of a plastic trait would I guess allude to a different kind of memory if that makes sense. Yeah no I like this question it is it is not what I don't know how to directly link it to the sort of little narrative I described in sort of this parallel with this kind of regulatory network situation but in general I am very interested in like phenotypic plasticity as I'm not sure phenotypic plasticity is quite the right word phenotypic plasticity is usually used in the sense of different time scales like developmental time scale right if I my tree that grows at different temperature what's the value of the trait but on top of that shorter time scale responses that you know organisms respond to their environments right which I'm for example very interested in how that affects evolutionary conclusions withdraw we draw there's a lot of models that sort of assume the genotype is the same as phenotype but recently it's a kind of under rise this question of given that organisms can be flexible in their phenotype does the buffer evolutionary change does that I can see a selective sweep versus coexistence with a more plastic trait yes so once you map if I so like I said we just posted this a couple weeks ago as a narrative of just pure kind of physiology of a cell oh bacteria could in principle learn once you map it into a ecological narrative there are other forces that suddenly become you know very relevant to consider how would this thing behave in the face of yeah for example like a more plastic and that's actually okay that would have been probably a shorter response to what extent real bacteria say I'm like in a similarity context I'm like one situation where as an external variable that I might wish to track internally is something like a somatic pressure right and which is the correct strategy is that that I'm trying to match my pressure to the external pressure very well or I just develop the ability to be more tolerant of larger variations in the pressure right and I sort of phrase this question in a theorist way as if you know the goal was the first to track best right and of course ecologically the other solution is perfectly reasonable and maybe more uh more natural yeah one more question if that's okay much you said there's a one of the time periods there was a dominance of a certain abundance and if I heard you correctly you said it's not because of the the fluctuation at the moment it's because of the fluctuation structure do you mean that that's some kind of steady state such that we've already the population's already learned the structure or that I just didn't understand why the abundance doesn't correlate to some kind of fitness advantage in that time period yeah so fitness advantage is uh for in an ecological context like if I'm at an equilibrium that everybody's fitness is kind of the same uh but let me say the following so what I meant is there is a simple intuition which is kind of very ecological actually like if I have a specialist in resource a and a specialist in resource b and somebody who you know consumes a and b at the same time right I could imagine that I could cook up a model where if my availability of resources in my environment fluctuates along the dimension of every time like you know nutrients come in apples they fall in the ground and then apple has both nutrient a and nutrient b so the supply of a and b is all when they come they're always correlated with each other you could imagine that that might create you know I could write a model where then the generalist that right likes just that stoichiometry is somehow event like it's it's more advantageous uh to consume nutrients in that stoichiometry right in that sense this might well then promote the organism the species that's most aligned with how nutrients are supplied in the environment what I was what I meant by my comment there is that it's not just that so it's not just that you know the the better you your metabolic strategies align to how the environment fluctuates that promotes you because there of course ripple effects through the community right and as a whole uh like the sort of most extent way of putting what is promoted it's kind of the responsiveness of the system of the ecosystem in this case like if once a niche becomes available if it gets filled the quickest right and species abundances adjust to get you in that situation where it's filled the quickest in this toy model right it still has very much this feeling of a global optimization problem which is very special ecologically right so I'm not proposing that necessarily as a principle for understanding ecosystems I'm just saying that this is a scenario where if I look at the species and say what about this environment made the species successful that question in terms of you know the topic of the workshop limits to predicting assembly of communities in this case knowing the environment was not enough you need to know how it fluctuated in the past and that fluctuation structure could be a factor determining who survives which possibly for ecologists is like something that we all know since forever uh or maybe not so that's you know why I'm excited to just hear your guys thoughts thank you Jacopo did you freeze oh he cannot answer if he did it looks like he may have uh there are Mikael did you see there are some comments by Leonora and then by Martina there is Jacopo back yeah I just my computer crashed so okay uh well I think that there is uh we have time for one last question there is a hand raised by Shin Sun yes uh hi great talk I'm just curious about the cost of learning processes that you mentioned maybe my question was covered already but I was curious um how do you consider the cost while the microbes are learning the fluctuation because it seems very expensive to produce a lot of stuff and is it possible that the cost is so high that at some point they try to uh instead of learning but try to forget what they have learned uh when when the fluctuation see like the frequency changes or something like that mm-hmm great question and also before I answer that I saw that their question in the chat uh from Leonora and Martina I copied like chat is a small window I copied them out so I can read them properly and then I'll find a way to reply to them they seem more complicated than I can just address in a second um for this question well so I mean it's easy to put this in the model and say there's some cost of expression of all the proteins and you know do some balance analysis of costs and benefits we didn't really go there because it was sort of a proof of like the question was a question of principle is this something that we could you know if I can't if I can't think of any mechanism for a bacterium to learn you know whether its environment was fluctuating a lot or not recently right if I can't think how it could even be possible I wouldn't do that experiment right now if I know it's possible I could do the analysis of costs and benefits and say in what environments that might be useful I'm not and it's easier to put in a model as a cost of expression I'm not actually sure to what extent it's biologically that um constraining because bacteria you know spend their budget is a complicated thing um to what extent the cost of an extra protein is really metabolically a burden uh and there are actually experimental examples where there's a in fact a two component signaling system with an auto regulatory link precisely of the type we're sort of describing here theoretically um which basically serves as a phenotypic memory of a signal that recently experienced you know you respond to it in a certain way but then with a protein that's very long lived and you keep it around so that in next time you experience it you sort of already more prepared to respond which is kind of a a little bit of similar to the mechanism I just described the difference is that what I described is that you can use the same architecture that people actually known experimentally kind of exists and works to respond not just to the signal mean which basically like oh how frequently in the past that I experienced the signal but even subtler signatures such as you know which direction of fluctuation around that mean was I experienced it and is that too much of a theorist game I don't know but you know humans have diets that change in different ways if it provides a fitness advantage bacteria would make use of it right and so um yeah so that's the same and so in a way we actually know real systems that have such auto regulatory interactions and then keep this system expressed kind of you might say unnecessarily and thus paying the cost also paying the cost of the space of the membrane which is possibly actually more important than the cost of just expression of proteins but we know that real bacteria do that and so presumably it's not too much of a burden so that's my justification but excellent question and we did not do any you know cost benefit analysis here thank you great so I think we are quite out of time so unfortunately we don't have much time left for having a discussion and I think we can sort of all the speakers we are they're gonna be with us in the next days with some I mean they have also some constraints but I think there will be time to continue the discussion so with that I'd like to thank all the speakers again for their talk and all the participants for the very next discussion and we'll meet again tomorrow at the 3 p.m. Italian time for another session and a poster session that will follow so thank you very much see you tomorrow thanks bye