 and a few years ago, he's gonna talk to us about ecology, geometry, or economy in a synthetic microbial ecosystem. All right, can folks hear me in the back? All right, excellent. Yeah, it is just an amazing pleasure to be here. This is an incredible place, and there have been some really fun talks. I'm real honored to be able to contribute to this as well. So, I obviously don't need to convince this community that microbial communities are important, but I always like starting here just to give us all on the same page. So, we know that microbial communities are really important for everything from individual nutrient uptake to global nutrient cycling. And additionally, microbes are these amazing chemists, and so there's a lot of excitement about trying to harness the power of diverse microbes for a whole array of industrial processes. Now, because of the importance of microbial communities, we'd really like to understand what determines the composition and function of communities. We'd like to predict how those emergent properties will change their time, and ultimately we'd like to be able to manipulate both the composition and function of microbial communities. And sequencing tools are obviously giving us some really powerful approaches for confronting this challenge of understanding, predicting, and manipulating. So, we're now getting parts lists at the level of both the single genome and the microbial community. So, the list of all the genes that make up a single organism and the list of all the species that exist within the community. And my lab is attempting to connect dynamics across these levels of biological organization. So, specifically, we would like to be able to start with a genome, use that to understand the physiology of a single organism, understand how this intracellular metabolic physiology generates species interactions, and then ultimately understand how these species interactions generate emergent community properties. My background is in evolutionary biology, so ultimately we'd also like to understand how selection at each of these levels of biological organization feeds back to change the genomes that encode life. And this is a fun integrative challenge that allows us to use tools and approaches from lots of different areas of science. So, today I'm gonna tell you just sort of two brief stories about some of the ways we've started to try and confront this integrative challenge. First, talking about the interaction between genomic and ecological robustness. And then the second half, talking about the effect of location on microbial interactions. Taking a big step back, from study of microbial organisms, we know that genetic variation within a population can be really important for structuring communities. So for example, Tom Whittem has shown that genetic differences between poplar trees, so mutations that arise in different individuals of poplar, play a really important role in structuring the communities that are associated with those trees. So genetic variation is driving or structuring community ecology. Conversely, there are lots of great examples of ways that community ecology shapes the genetic structure within a population. One of my favorite examples is the study or the work by David Resnick in which he showed that the life history strategies of guppies are strongly shaped by the presence of predators in their environment, within their community. So we know that genetic variation interacts with community structure. But I think that the exciting next question is to ask, can we predict the prevalence of these interactions between ecology and evolution? And when thinking about dynamics, I think there have been two sort of archetypal bodies of literature which have addressed this. So there's been some great work looking at the effects of genetic robustness. So for example, looking at the distribution of fitness effects, much like we just saw from some of Roy's work. And these types of approaches allow us to answer questions such as what is the likelihood that a mutation will affect a phenotype? And what is the average effect size of a mutation? In parallel, there's been some fantastic work looking at ecological robustness. Asking questions such as how do species interactions influence the stability of emergent community properties? But I think that there's been less integration of these two approaches. So we were really interested in asking, how do species interactions influence the genetic robustness of a system? So specifically, how does ecological context influence the robustness of a single population? And conversely, how robust to genetic perturbations are different types of ecological interactions? So we carried out a study in collaboration with Juan Chubes, Brian Granger, and Daniel Segre. And we used one of my favorite systems, a system that we developed, to ask, what is the relative robustness of interactions? And the system involves a strain of E. coli which can't produce its own methionine and a pairing it with a strain of salmonella that excretes methionine. So if you grow these pairs of strains and lactose mineral media, they form an obligate mutualism. The salmonella relies on carbon from E. coli, E. coli relies on an amino acid from the salmonella. And one nice thing about this system is that because we know exactly the metabolites that are being exchanged, we can simply add those metabolites in the media and convert this system from one in which the species are cooperating to one in which they're competing. So it's the exact same genotypes in both cases. All that's changing is whether they're getting the metabolites from each other or from the media. We then made use of the computational platform that Daniel stepped through yesterday, which is called COMETS for computation of microbial ecosystems and time and space. And just as a brief review, the way this works is you start with a genome-scale metabolic model of one of your species. And some define metabolic environment. You then run flux balance analysis to predict which of these metabolites will be utilized by the species, how much that species will grow, and what byproducts that species will spit into its environment. You can then throw multiple models within a single simulation. And then species interactions emerge as a result of the utilization and exchange of metabolites. So this ecology emerges from the intercellular metabolic mechanisms. And as was previously mentioned, we've previously shown that this computational approach can quantitatively predict species ratios. Further, we've shown that this computational platform allows us to track the concentration of all metabolites within the system through both space and time. Okay, so stepping back to our initial scenario, we have two different bugs and two different environments. We created genome-scale metabolic models of each of these, threw them into COMETS, and then we're able to predict growth through time under each environmental condition. We then went in and created all possible single reaction knockouts. So in essence, we created a library of 2,583 mutants and then re-ran simulations for each of these and said, how do each of these properties or how does growth change as a result of each of these mutations? And so what we're interested in is we can think about, so in the paper we actually did, so we only do one mutation at a time. In the paper we did both E. coli and salmonella, and I'd be happy to show you the salmonella day at the end. At the moment I'm just gonna tell you about the E. coli story, but it'd be fun to come back to that. So what we're looking at here is the variants that arises. So when we're thinking about genomic robustness, we can say as we introduce each mutation, how does that influence the variants in final E. coli population size? If E. coli is very robust, you're going to see a very small variance, and if E. coli is less robust, you're gonna see that this variance increases. And we can pay attention to that both for just the population size and thinking about E. coli, we predicted that we would see that the variance, the robustness of the E. coli genome was approximately the same in both of these conditions because E. coli is using the exact same metabolites in both cases. So the metabolism should be equivalent, so we thought, well, mutations will have roughly the same effect in both cases. In terms of the variances, so what we're tracking is actually the population size at the end of a single growth phase. So it's not equilibrium actually, it's just at the end of a growth. But we're not considering evolutionary stable strategies at this point. Going forward we will, but here we're not. Okay, so we had that prediction about how species interactions might influence the robustness of a single genome. We were also interested in how mutations might influence community properties. Again, what we're looking at is a variance, but here we're plotting variance in community properties as opposed to population size. And here we had some varying predictions. As I just showed you, when organisms are engaged in mutualism, they tend to stabilize species ratios. And so if you could think about this as if we plot the final E. coli population size against the final Salmonella population size in the absence of mutations we might find a point here. And then as we introduce mutations which reduce the final E. coli population size, that should also reduce the final, reduce the growth of its obligate partner. And so as we throw in a bunch of these mutations we expect to see some distribution that looks like this where species ratios are being stabilized. So we expect cooperation should stabilize species ratios in the face of mutations. That's right. So here what we'd be looking at is species ratios. We'd plot species ratios on the x-axis and say how much change do we see in species ratios? So there we're looking at the final size of just the E. coli population. So the contrast is we're looking at the variance within a single population versus the variance in community properties. Conversely, there's been some great work by LaRoe and others suggesting that if you have competing organisms that should lead to compensatory dynamics such that you end up stabilizing biomass. So if E. coli takes a hit and uses less of the resources those resources then just get taken up by its competitor. And so in the competitive case we thought that what would be stabilized is the total biomass. So we thought that cooperation would make species ratios robust to mutation while competition would make total biomass robust to mutation. So we then ran the simulations and tested these hypotheses. I'm showing you similar graphs again where final E. coli density is on the x-axis while final Salmonella abundance is on the y-axis. The green dot is the case of no mutations or sort of initial condition. And then we'll step through them. So for the competition case again what we're expecting is biomass to be stabilized so mutants to lie along this axis. And we found roughly that pattern. There's definitely a spread but we do tend to see that as E. coli as mutations decrease the abundance of E. coli they increase the abundance of Salmonella. And we found that over 400 mutations had a greater than 1% effect on the community. Switching to the cooperation case now we expect the opposite pattern or we expect mutations to lie and a diagonal along this line. But here we saw a divergence from our expectations. So now instead of when E. coli gets decreased instead of decreasing we see that there's this handful of mutations which actually increase the abundance of its obligate partner. We also see a change in the number of mutations which have a significant effect. We're really interested in these mutations which seem to break this constraint on species ratios. So we went in to take a look at what they are and we're really interested to see that all of these mutations change the excretion profile of E. coli. So they're changing the currency through which or the molecular economy through which these species are interacting. Going back to our original questions and plotting the exact variance we predicted that genomic variance would actually be equivalent in the two scenarios but we found that that wasn't true. There is a significant difference in the robustness within the E. coli population such that when E. coli was cooperating with Salmonella there was smaller variance so the genome was more robust. Stepping to the ecological scenarios so how do mutations influence emergent community properties? We found that in terms of species ratios along with what we predicted actually cooperation was more, did stabilize species ratios despite those outliers and looking at biomass we predicted that competition would stabilize biomass and we found that that wasn't actually true. But if anything competition leads decreases the stability of biomass within the system. So in summary of this first part genomic robustness was altered by the community this is perhaps something we should have known ecology is really important and the robustness of a genome is going to be dependent upon the ecological context in which you study it. Second we found that lots of mutations influenced emergent community properties. And specifically we found that the cooperation tended to stabilize species ratios. It made species ratios more robust to the genetic perturbations. But we found that mutations can break that constraint by changing the molecular economy or changing the metabolites through which species interact. Conversely we found that competition need not stabilize biomass as was expected. And so I think this points to the fact that metabolic mechanisms can be really useful for connecting genotypes to community properties. That there are times in which we can't simply generalize interactions to lockable terra but that some of these metabolic mechanisms can be really useful for understanding the dynamics of systems. We're now just got some funding to continue this and experimentally test these computational predictions. And so for this we're using TNC to create the library of all possible, so you throw in a transposon which inserts randomly into the genome and you create libraries of 50,000 mutants all with unique insertions. You then take this pool of mutants and can grow it in different ecological contexts and then look to see how the frequency of these mutants changes and whether those changes are different in the different ecological contexts and use that to understand the selection on every gene within a community. And so one important thing that I think this suggests is that there's been some fantastic work in evolutionary ecology demonstrating that it's possible for the effects of genes, so mutations, to cascade up all the way to influencing the properties of ecosystems. And then conversely ecosystems can influence the frequencies of genes. But I think we're now at the point where we can say not only is it possible for a mutation to influence ecosystems, but we can say what is the distribution of those effects? And just as the distribution of mutational effects are really critical for understanding the evolution of a single population, I think that understanding these distributions within a community context are going to be critical for understanding the evolutionary ecology of microbial communities. So for example, when we design a community to create biofuel or for some other industrial process, we need to start understanding when our mutations likely to perturb that community in various ways. Okay, so now stepping forward to the next brief and yet we are interested, so in this first pass, all of this was in mass action. So we didn't incorporate any spatial structure, but we are very interested in how spatial structure might mediate some of these interactions. And again, as we just saw, space can be really important. Often microbial communities or microbial ecology assumes well-mixed communities. So historically, what we've done is we've gone out and grabbed samples from our community of interest, ground it up and sequenced it and said, oh, all these bugs are there, they must all be interactive. This sort of harkens back to the soup of enzymes that Daniel mentioned at the beginning. I would argue that many microbial communities exist in scenarios something like this false picture in which there are lots of species that are at different locations in space. And I think that in most cases, microbial communities will not actually be soups of enzymes, but that the location of each of those enzymes will be critical. And so to start understanding the effect of location in complex scenarios like this, we wanted to simplify and take a more bottom-up approach. And so we went with a simple experiment we could think of of let's just take a single species, throw it on an auger plate, allow colonies to grow up and say, to what extent can we understand the variation in colony size as a function of location? So this is an example of E. coli colonies growing on LB. And as any microbiologist knows, you've played out some colonies, you're gonna get colonies of different sizes. So like this colony is much smaller than that colony. So it was our first just totally simple experiment. We wanna say, can we a priori predict the difference in these colony sizes by understanding their location? And so a fantastic postdoc in my lab, Jeremy Chacon, developed a really nice system in which we use basic office scanners. We house them in warm rooms. He wrote then a program, which causes those scanners to fire every 30 minutes. And then also wrote custom software, which analyzes each of these images, identifies the location of each colony and tracks it through time. And so as a result, we can look at the variance in colony size and relate that back to location. We did this for both E. coli and salmonella on multiple media. And one interesting thing that we found was that the variance in colony size was really different on different types of media. If we look at the coefficient of variation in colony size at the end of the experiment, we can see that that changes dramatically both as a function of species and as a function of a metabolic environment which they're growing. To start to understand some of this, we again returned to comets and simply simulated each of these conditions. So for each plate, we seeded biomass in our simulations at the same location. And then we said, how much of this variation, experimental variation, could we have predicted from interest of their metabolism? And we found that comets did a remarkably good job. So what I'm presenting here is the simulated, the relative simulated colony size mapped against the relative experimental size. And so things that fall on a one-to-one line like this mean that we're able to predict with a high degree of certainty, how big those colonies will ultimately get. Yes, that's exactly right. And that's where I'm going in the next slide. So simulations did a relatively good job of predicting it although interestingly there were some cases like salmonella growing on glucose, where we saw more divergence than we would have expected from our simulations. That is true. So in simulations what we're tracking is total biomass of a colony and in experiments we are looking at area. So we standardize, so it's relative changes in size, but we don't transform the biomass into the area. Yeah, so we standardize so that the maximum is one, the biggest colony is one in each of our metrics and then map that against each other and multiplying again. Yes, that's true, that we could modify this in various ways, the reason we didn't do that is because the transformation you'd want to use would depend a little bit on the shape of your colony and we're unable to at this point give very good metrics on the shape of the colony. So we did this as a first pass and it seems to have a strong correlation, but it would, yeah, you're right, we could certainly do some additional transformations and that might improve some of our predictions and it'd be fun to talk to you some more about ways that we could do that if you have thoughts. Okay, so we have all this data on colony sizes and locations and as was just predicted, I thought this is gonna be super simple. All we're gonna have to do is look at the distance to the closest competitor and map that against, or graph that against colony size and we're gonna be able to explain all the variants. And we found that that explained some of the variants but we're still leaving over half the variants on the table, so that wasn't terribly satisfying. So we then dug into the ecology literature. These sorts of things have been measured in trees and in all sorts of vegetation for years and in those cases, often what's suggested is that you need to not take the distance to the closest competitor, but consider the distance to all possible competitors and that can either be a linear distance or the distance can be weighted as a square as a result of diffusion. So we thought, ah, now we got it. It's gonna describe all of it and in fact it describes less of the variants. This left us a little puzzled but then I was talking to a collaborator of mine, Wilfromobius, who's both a great guy and has the coolest name in science and he's a biophysicist and as many of you probably have already intuited, he said, ah, physics figured this out a long time ago. What you need is veronoi areas. So you need to map out the area on the plate that's closer to a focal colony than any other colony on the plate and if you graph that area against the biomass then you'll be able to explain more and of course he was right. So now we can explain sort of a satisfying amount of the variation. And I think this result is really cool because one of the things it suggests is that colonies, so the variation in colony size is influenced only by the location of colonies that determine your territory size. In essence, the variation of this dark blue colony is only influenced by the location of these red circles. So it's geometry, not distance, that is driving the interactions in these spatially structured environments. And just to convince you of that, we ran simulations in which we simply removed one colony at a time and said how much of an effect does removal of an individual colony have on all the other colonies on the plate and basically we were able to show that the final size of a colony is only changes if one of its neighbors is removed. So the size of this colony is not changed by removing any of these dark colonies. Further using simulations, we're able to look at how the strength of spatial, yep. So as long as you've used all of the metabolites within the system, so as long as you completely draw down resources, it actually doesn't matter what the distance is. It's independent of distance. On opposite sides of the plate and as long as you let them grow long enough that they draw down all the resources, they are the only thing that's gonna, yep. So we're then able to also look at how certain parameters within the system influence the strength of spatial interactions. And so for example, demonstrated that as you increase the maximum uptake rate of resources, that increases the strength of spatial interactions. Conversely, as you influence the diffusion rate of metabolites, that decreases the strength of spatial interactions. And an interesting one that I hadn't really appreciated in which relates back to this is that the strength of spatial interactions increased through time because you're decreasing the amount of resources that are available. Interestingly, it's these two effects that cause metabolism or they cause these different patterns of spatial effects in different metabolic environments. So because bacteria both take up different metabolites at different rates and grow at different rates with those metabolites, that's what generates this divergence in spatial effects in different metabolic environments. Finally, we were able to, we were interested to see that there was one pretty strong outlier of our expectations, which was salmonella growing on glucose. That was also one of the big outliers when we mapped our simulations against our observed. And specifically, you can see that these, there are some colonies that we predicted would get much bigger than they actually did. We then were able to go in and, so we hypothesized that this might be because of toxicity building up. So acetate is generated as a byproduct of growth on glucose. Acetate can change the pH. And we found that indeed, when we grew colonies of salmonella on glucose plates and threw in a pH indicator, they generated a strong change in the pH indicator. They made the environment much more acidic. And if we incorporate this toxic effect into our simulations, we can suddenly explain a great deal more of the variation and we can manipulate that through changing the pH of our environments. So in line with what we saw yesterday about pH being really important for structuring the composition of the gut and maybe some of the things that we're going to see from Jeff Gore coming up, we found that pH was a really important determinant of the strength of spatial patterns. So in conclusion, we found that interactions were determined by geometry, not distance. That metabolism alters the signature of the effect of location. And that these quantitative predictions can allow us to identify interactions that we didn't realize were occurring within the system. And so going forward, we're excited to extend this through collaborations with Rachel Dutton to try and understand the spatial patterns that arise as we introduce more competitors to our system as well as introducing different types of interaction. So mutualism or facilitation in addition to competition. We're also doing a lot of work to look at, to understand how these effects of locations mediate the selective pressures within microbial communities. And then we're also having a lot of fun talking to some of the great ecologists at the University of Minnesota who to compare how these spatial patterns at the microbial level relate to the spatial patterns that are observed at microbial levels. And so in conclusion, I hope that I've convinced you that synthetic systems can allow us to understand, can allow us to confront some of our initial goals for managing microbial communities. So it allows us to understand some of the mechanisms that generate emergent community properties, metabolic models can allow us to predict how genotype connects to ecosystems and therefore how communities will change through time. And integrative approaches I think are whole promise for allowing us to develop methods for manipulating microbial communities. And so with that, I'd like to thank collaborators and the fantastic folks in my lab and I'd be happy to take any questions.