13:06:12 Alright, so I wanted to tell you today about some of the work we, so give you a brief introduction, my lab fundamental is trying to understand how micro communities form, how they they work together, and how they function, and how they evolve.
13:06:33 And what I want to tell you today is about one of the directions in my lab that I'm most excited about.
13:06:39 We're fundamentally trying to understand how one may use evolutionary engineering partners like black revolution, or breathing. To find microbial communities that have properties and functions that are desirable or that are optimal given environment.
13:07:00 And one of the main things I want to see if I can convince you off is that evolutionary engineering of microbial communities is can be conceptualized as a Guided Navigation of a structure function landscape.
13:07:17 So, I think To that end, I think is good that I start this district by introducing you to this idea of what our structure function landscape is, and in ecological ecological significance landscape is what it is is essentially a map between the structure
13:07:34 of our community and the properties of the collective purpose of that community, which can be referred to us, its functions.
13:07:45 And
13:07:45 what the structure is, is, is a different question but you can define it in various different ways and various levels of biological organization.
13:07:56 You can think of the structure of a community as the community compensation, what is there, the least of four cell types that are present, and their abundance.
13:08:05 But of course you can define it in different ways you can define it as the least of all genes and their relative abundance and the community or the list of for pathways are the list of four streams or species or, you know, higher and higher up throughout
13:08:19 this talk about we talking about it as the basically the, the, the abundances of all the, the cell types which I'm going to use a species in in our communities.
13:08:40 And if we have a good map between community competition and function that will allow us to predict the function of our community, knowing its composition, which of course is incredibly useful, and it will also help us engineer microbial communities.
13:08:50 And I just wanted to point out that there's various different levels of organization that could be understood. Also, and the reason why this idea of ecological structure function landscape is so appealing is because if we had a good model of it right.
13:09:02 Now, before I continue.
13:09:06 I think it's helpful and important to take a bit of a break from the talk and discuss whether sets a landscape is actually well defined, whether one should expect that community functions should depend in an ambiguous manner.
13:09:23 on the species composition of a community.
13:09:29 And I think that if you think about it for a little bit on. In general, the answer to that question has to be now. Right.
13:09:37 The value of the function of our community at any one time should depend on on yes have could have a component of depends on their composition of the community that time.
13:09:51 But more broadly is going to depend on their own internal dynamics of that function, up to the point where we're measuring it right.
13:09:58 And that's going to be different for different functions, right, and it's going to be affected by the history of community assembly. I just to give you an example imagine that the by the function, we are examining the concentration of a particular metabolite
13:10:11 that is produced by the community.
13:10:17 And, and the function is the concentration of of that metabolite.
13:10:21 Well the metabolite is going to be produced by some community members and might be updated or degraded by others, but then the tablet may also have its own chemical.
13:13:16 a function so in function space that depends on things like substrate concentration, so even for a given Community Edition in history. Yes, your function still in this absolutely and you're absolutely right right so that that you can only define your
13:13:21 function is going to be a function of the composition right, but it will be, you know, affected by any number of other mental parameters to. Right.
13:13:34 So let me give you an example right and the.
13:13:40 Just think, for instance of a metallic dollar on the one I'm going to give you an enzyme that is released by their community members right and we are trying to understand that we tend to measure the catalytic rate of those that the collective, the collection
13:13:57 of all incentive that I've been released by the community, right.
13:13:59 So, in that case, we can think of is. of them that their, their concentration that is another one time is going to be, of course, a function of the rate at which that enzyme is produced, and there were there was that enzyme is in activated right through
13:14:17 any means right and those means could include you know that diffusion out of the volume of interest could could include degradation by other enzymes of protease that might be present in activation by absorption to the surface of God knows perhaps right.
13:14:32 So there's going to be any number of processes that can give rise to the decay of that enzyme that are not directly mediated by the, the species abundance, right.
13:14:43 So, and that is what massively one good reason why the structure and function, not necessarily one to one. Of course, when we should say more cautiously is that it's it's, I think would be better defined as the relation between competition and function
13:15:00 of composition and function in a given environment, right, given that all alone in the parameters are are constant that is the hard part of it right because then the species themselves are changing the environment as they grow on it.
13:15:13 Right. And it's very difficult to define what the environment is so even know. Exactly. And that is one of the complications why structure function landscapes are not necessarily as trivial as they might seem right and why they the function, and any property
13:15:28 of the community right, any environment the variable, if you will, right, will not necessarily be a unique not be uniquely determined by competition at one time.
13:15:40 Right. But by the history by other environmental variables, many of which may not be known.
13:15:47 And, in a long list of other other issues.
13:15:52 And I was, I was in fact, you know, trying to discuss this. I think this was there, this late I was, I had I don't know if you got to see it. But yeah, I guess what I was trying to say is that the answer is no, right, that that community functions are
13:16:05 are not necessarily uniquely the tournament, by composition composition of the community, right, could become a term by at least have a number of other factors.
13:16:17 However, I going to argue about this talk, right that, and I'm trying to put together a perspective from the Web Player to discuss this point in more and more depth.
13:16:27 At the moment, and, and I'm reaching some conclusions that are at least at this point I'm thinking, which is that there are a number of situations where set a relationship between structure and function can be a good approximation can be approximately
13:16:42 well defined.
13:16:45 One needs of course the to the cases that I think we're these ratios of is well defined would include communities that dynamic is table where.
13:17:01 Whenever the external selective pressures are kept constant, of course, the communities would read state of equilibrium, and other medical sense. And, and that would allow all the environmental variables to equally rate as well as the completion of variables,
13:17:17 and it is possible, and there are circumstances you could have a one to one relationship between both.
13:17:23 I think it's also possible that when the functional dynamics are very fast compared to the original nomics.
13:17:29 You could have the grid query. Right.
13:17:33 In the time skill of when opportunities don't change very much and that could lead to some sense a, an approximate mapping between both, and the other case I'm going to discuss today are single bad communities.
13:17:47 When whenever you have reproducible enough population dynamics that so that whenever you start the community with a given inoculate them.
13:18:01 You can guarantee they're going to have a specific dynamical trajectories of all the species and environmental variables. As a result, so whenever that population dynamics within a single box is very reproducible, it is possible to link the initial composition
13:18:13 which you might know, and the final, you know, say for is the accumulation of a molecule, or environmental the pH or any other environmental variable that is the result of the metabolic activity of the community.
13:18:26 I'm not saying it's trivial to do that.
13:18:28 And in fact, one time to argue in this before I start my talk is that even though I'll be talking about a structure function landscape, as if it were a well defined magnitude that it is not necessarily so right and we need to keep.
13:18:43 I want to make sure that I, I am skeptical of this concept. As much as I can be right because as, as you're hinting. It is definitely more complicated than simply, or you have a particular composition therefore you must have this property associated to
13:19:01 it right, you could have other determinants of function that are not directly linked or uniquely linked with the composition. Yes.
13:19:14 Hi, I'm just wondering, and you already touched on this a little bit. When I think of this very defined structure function landscape should I be thinking of it as being ready to find in a particular environment or sort of independent of environment, right
13:19:28 I am thinking of, in a particular environment but it's it's a bit tricky right because what I mean is in a particular set of
13:19:37 of external constraints, right, like a particular temperature particular set of, you know, supply nutrients or particular, you know, geometry that you supply of course the community itself will change the environment where it's growing right.
13:19:52 And it's often not possible to know what that, you know, environmental production will be, and often we don't know. Right. So, there will even if I'm thinking of it as a as a community the marketing comes to function in a given environment, what the environment
13:20:08 is is often very difficult to decouple from, or to know right right and and and that's the environment itself has been constructed by the community as it grows on it.
13:20:20 Right, and environmental variables interact with one another right so so it's it's it's a bit of a subtle concept, and, and, and what I'm going to try to argue is that despite all the subtleties there are conditions what I think is still helpful.
13:20:36 Or maybe helpful. Right.
13:20:45 But, but that's that's a good point right because it should be as you know when we think about fitness landscape right from this just to give another another example.
13:20:51 And, and I'm going to try to tell you where I am in terms of the way I'm thinking about it.
13:20:53 Fitness landscape is a ratio between genotype phenotype in a given environment, and often we take that out but the environment. This is a given right and we know what it is.
13:21:03 And often to what approximation we do, right, but the truth is that organisms change the environment as they grow on them right so so that that makes that relation between you know different phenotype in a given environment bit less.
13:21:16 You know, more complicated than it seems. And I think the same thing happens for ecological structural functional landscapes right there there's a nice construction or environmental modifications can alter that, you know, given environment close, right,
13:21:29 we put at the end.
13:21:31 at the end. So I mean, all these are, I was starting with this.
13:21:36 This kind of discussion of caveats before I jumped in right just to make sure that we all keep that in the back of our heads, as I continue telling you what we found, right and practical examples.
13:21:49 Because it's, and also because it's a it's a topic I'm thinking a lot about lately, and I'm trying to put my thoughts in order and I in a perspective speech so if you guys have any anyone has any thoughts about this I would love to get feedback after
13:22:02 after the talk.
13:22:04 So I wanted to just be face to top by by this right by me trying to build a bit skeptical about this concept I'm going to be arming my book around, but I think that I hope I will convince you throughout this talk that despite all these caveats.
13:22:19 There are still circumstances where startups landscape can be well defined in fact has a very powerful predictive powers, right.
13:22:29 So, so I don't want to get ahead of myself, and so let me show you what I what I wanted to tell you and, and then definitely we can have a more in depth discussion about how these planets might affect our findings.
13:22:42 Okay, so in this talk, I want to, I'm going to focus on situations where I think there's two different landscape is well defined, well enough to find at least me as unique might be doing compositional function.
13:22:55 And he's dividing two parts. First I'm going to be telling you about how these interactions govern the shape of the structure functional landscape. I want to give you a specific example of an empirical structures landscape we have characterized.
13:23:08 And what we have learned about it.
13:23:10 And then in the second half of this talk I'm going to give you up our current state of our understanding of how want to navigate these landscapes, through essentially evolutionary engineering takedown approaches.
13:23:26 And for the first part, this is in stuff has been published in a couple of papers, primarily, this is the work done by Alicia Santa Clara and George if he were to postdocs in my lab Alicia left.
13:23:48 And the paper was published a couple of years ago right before the pandemic complex biology and given his reference in case you're curious and we have followed up on another study in a very similar lines, and that was published later.
13:23:51 And he's now a staff, scientists that center in Madrid and Georgia and is still a Yale but will be finishing his persona is amazing.
13:24:03 Earlier this year.
13:24:05 Alright, so the first question is, what are the shape. What did the structure Fox landscape, look like and how many wanted to understand how species interactions affect the shape of these landscapes right.
13:24:22 And the, the system that we had studied is a model the composer Consortium, that is made up by six different species of bacteria.
13:24:30 Five of them are Seelye, and different species of the seals and a sixth, which is our panelists he was Polly mixer.
13:24:40 And all of these bacteria that you see here is a corner bits of the bacteria. Are you see this halo around them.
13:24:47 This Halo is caused by the action of extracellular emulators which are enzymes that the bacteria secrete to the environment.
13:24:55 The answer these enzymes to up to dump starts and break it up into smaller molecular weight components mostly sugars the sacrifice and other smaller molecular weight components and all six of these materials are capable of expressing and secreting these
13:25:14 enzymes that are called emulators.
13:25:21 And so that's in terms of their composition that's different subsets of these bacteria going to be the composition of our of our consortium is we inoculate the beginning.
13:25:31 And as for the function but we're going to do is the following. We're going to take known upon the amount of these cells and when I put them into an environment that is a liquid media that contains starts as the primary carbon source, we're going to allow
13:25:46 the consortium that we are going to assemble.
13:25:49 We're going to allow them to grow in that environment for 24 hours after which we're going to take all the liquid media and want to filter it to remove all of the bacteria.
13:25:58 We're going to add different, the illusions of it to media that contains starts.
13:26:04 And then, as a function of time. We're going to use scoring a tree. So, sounds very fun so it's super simple technique to monitor the, the rate at which starts to break down maybe we can add a reagent that binds to the start.
13:26:19 And it only binds to Bali sacrifice that about certain length, right, and and doesn't for a shorter ones. So as the larger molecular weight compounds are being chewed up the amount of this region that gets absorbed diminishes, and we can tell it through
13:26:38 changing color of their of their ultimate, so it's a quantitative very inexpensive as a way to monitor stats the relation, both in semantically and, and through all the processes.
13:26:53 Yes, maybe I missed this first.
13:26:56 What was the proportion of the species.
13:26:59 Is it the same for the species or did you try different proportions in the in the experiments we did with your friend seminar, the same proportions, but we could have tried others right.
13:27:12 That's, we had to choose one particular ratio for is approximately one to one to one, but it's not quite almost there.
13:27:20 And. And what we did is basically assembled coming authorial sets of this, this consortia right so it took every every monoculture pairwise three member consortia for memory consortia and so on.
13:27:33 Right.
13:27:34 But we only tried one initially knocking but we know, so we know how many cells of its bacteria we're adding at the beginning. Right. And then we'll let them grow and after 24 hours we stop.
13:27:45 And, and then that's when we measure the young.
13:27:49 This is my duty.
13:27:50 Right, so I wanted to first of all. Before I continue, and show you how that what this starts the graduation reactions look like.
13:28:00 If you take this isn't even for them, the emulator. Just one I mean a suite that we is commercially available, which is pretty fired from the satellites.
13:28:13 What you. We did a number of experiments and found out that the mechanism of starts the relation by this amylase is a two step by chemical change that.
13:28:26 configure and calculate what is the fraction of starch, that has been degraded as a function of time, or you get an expression that look like this, where, which has only has a variable which is a time that is has time units but has been, you know, realizing
13:28:50 this weight really doesn't matter right, unless you're very curious secondary what that is, but this is basically I'm every scale time, and then a parameter, which is called v here which you can think of it as the velocity of the, of the reaction.
13:29:05 And, and this visa is a fitting parameter right so T is the variable and visa fitting parameters.
13:29:11 So, if you if you fit that data to this model it does a very good job.
13:29:17 In fact that that's a much better job than if you try to feed a single, single step model. And actually, in fact, it, it does make a lot of sense and it's consistent with people had seen the literature before.
13:29:28 And the nice thing about this model is that it does a very good job when you look at the activities of all the enzymes are released by each one of the bacteria in monoculture.
13:29:40 And if you fit the starts of relation curves to the model is the foods are very reasonable and allows you to determine for each one of these bacteria, what is the analytic rate of the monocultures.
13:30:00 seeing each other in any way. Right.
13:30:01 So, this practises what is our empirical system or structure from the landscape that we're going to do you say in response to your question is that we are going to do single bats cultures are we going to mix together.
13:30:15 Bacteria at fixed initial abundances, right.
13:30:19 And we're going to integrate together subsets of this bacteria right we're going to either have every culture, every possible pairwise culture last number of years for member communities and so on and so forth.
13:30:33 And for each one of those.
13:30:35 We're going to determine what is what is the, we're going to do to kind of fit the, the resulting started relation kinetics to this model that I showed before I were going to obtain the community function which is going to be there, the analytic rate
13:30:52 of the enzymes, be right. So, the structures the bonuses pieces. Yes. Can you go back to the previous slide.
13:31:00 It was looking at this be Polly Mexico. Yeah, it seems to suddenly shoot up at the end.
13:31:07 Right. and the answer right it is somewhere in between.
13:31:11 What do you think is going on there is this, is that some weird kind of cooperatives at this maybe model fit. No, not at all it's it's almost certainly an experiment that artifacts or politics or forms biofilms when it grows in isolation.
13:31:25 And it's kind of hard to filter the medium through. So there's a little bit of error when you, When you you measure it right.
13:31:36 This is, we have other data and this results actually are very much in line with other experiments we've had right. So, it's almost certainly just error for this particular data point here which is the one that shoots up.
13:31:49 And we did sensitivity analysis to see how much of this particular not value was affected by by this data point, and it's not really much right so the, the rate they're.
13:32:01 These feeling parameter is consistent with other independent experiments we've done for this, but they were bacteria coded Thank you.
13:32:10 Alright, so again going to structures the abundance of species, and the beginning so as we know what that is right because we mix them together. And then after 24 hours we simply measured a function right, that's the structure function landscape here.
13:32:23 Now, I'm going to put this out here. I'm happy to, to, to discuss if you want.
13:32:32 But we, there are essentially three different types of interactions between the members of the consortium that can affect the, the function. Right. And I think that really is our only three.
13:32:48 At least main ones that I can tell.
13:32:51 The first type of interaction that one can potentially see in these experiments are a biotic interactions which are biochemical interactions between the incense themselves as well as between the enzymes and the other molecules that are released by the
13:33:03 other bacteria right, it is perfectly possible for instance the one bacterium might secrete a, an inhibitor for for an in some of our competing microorganism.
13:33:16 I mean, as I said what I showed you that doesn't happen but in principle is perfectly possible.
13:33:21 One bacteria might be secreting, you know maybe acidifying the environment and that made the nature of the enzyme or they may be submitting a protease that might be also messing up with them for the enzyme, simply by another bacterium.
13:33:34 There are also mechanisms by which enzymes can act synergistically.
13:33:39 For instance, endo annex Swami laces are types of enzymes that act in different ways. One twos from the, from the tips, and the other one breaks the starts at random points in the middle of a chain.
13:33:54 So one generates substitute for the other, right. So it is it is in principle possible right that enzymes as well as I'm thinking of isms here in a very loose term right so you could have interactions between the enzymes, and the other molecules that
13:34:10 the bacteria might be releasing right. And those interactions are independent, on any biological interactions between the cells themselves right it's just a purely by chemical interactions between the molecules that they have released.
13:34:34 But there are also ecological interactions that can affect the contributions of different, different members of the consortium to the activity right. For instance, one species might actually through the metallic it secretes to environment that some of
13:34:41 them might induce the expression of the enzyme by another member of the community. Right.
13:34:48 It is also finally possible, that even if none of this other two interactions are present another type of a positive interaction would be that one bacterium could interfere or promote the growth of another restaurant species might inhibit the growth of
13:35:03 another species, or it might promote it right and simply because there's more cells you might imagine that there might be more enzymes secretive and that could also lead to higher or lower activity, right.
13:35:13 So, I, you know, given this topic of thought and this is pretty much all that I can think and happen right so you could you could have either evading interactions between the molecules that are released.
13:35:26 Or you could have that more enzymes are released or have higher activity right as a result of either simply.
13:35:35 There are more cells in culture or fewer than there are a monoculture or that they're being the same cells but itself produces more enzyme. Right.
13:35:45 And the reason why I say that these are pretty much all the project and how does it because these characterized by the total amount of time, right, and by the catalytic rates of answers right so.
13:35:58 These can be manipulated by simply the other molecules that are could be inhibitors which could be stabilizers or the nature of the enzyme or whatever.
13:36:08 And then the other possibilities that you could have is a change the total number of offense and that is released, which is what would result from an ecological interaction.
13:36:17 Right.
13:36:18 So that's, yes.
13:36:20 Three Can you can you show them the equation.
13:36:24 This sorry visa Right, right.
13:36:40 The, is that right yes. So this is the way you get an Nicholas mental kinetics I think when you are looking close to low and some concentrations. But would it will be reasonable to consider the effect of this saturation.
13:36:45 Right, right, right, it is and we we did for this particular case, and it doesn't fit the data better than than the model without it, right. So, and it's because of the specific concentration of substrate we have, and the values of KKMKM.
13:37:01 Make put the system in the, in the case where the substrate is is below situation. Right.
13:37:19 I mean, you're right, it could happen. Absolutely. It doesn't in that particular case of our experiments for the presence of us Yes. All right, thank you.
13:37:22 Right, so, um, anyway so these are the three types of interactions that we have.
13:37:29 And the first thing I would want to argue is that if no interactions exist so if if there were no interactions at all right, none of the three WordPress and then what you would expect is, first of all, and another, it will be the same as the sales everyone
13:37:45 together but it would be the same as if they were growing and different test tubes right so that they do not affect each other ecologically in any way so there's no ecologic interactions, right.
13:37:53 So if we did an experiment where you say take two monocultures right so there's an ecological tracks and then you took that you filter the, the volume right with enzymes are, and then you mix them together so you add them together to the two.
13:38:05 If there were no interactions that are biochemical in origin between them, then you would expect that the monolithic rate should be that the some of the rates from it.
13:38:15 Right.
13:38:17 And these principles should apply to any number of enzymes that you're adding together right to the enzymes are acting on the substrate independent from one another and not affecting each other right, and there are no other molecules on the enzymes right
13:38:33 that are interfering with the metallic catalytic activity of the other enzymes, severe trauma species.
13:38:40 Then the expectation should be that, then you can, you know, the math is quite straightforward, but that that there can only be great should be the sum of all the rates from all the different pathways right it's catalyzed by different type of it.
13:38:55 So, so that's what gives us this unknown model right of what kind of behavior we should expect to see if a there's no ecological interactions and me there's also no biochemical tractors right because the enzymes are acting independently from one another.
13:39:12 Right.
13:39:14 And the test is why what we did is we, we, for instance here in this one example, we combined incense from beam Avensis, which we live in the house, em.
13:39:27 And so this should be a people in the exam artist be we combine them both, right here in grey, we show the 95
13:39:37 confidence interval of the from an old model, right in red dots you see the experimental data, right.
13:39:49 It's very well described by the by the empty model in this case. And if you repeat the same exercise where we separately culture, it's Bs and that makes every possible pairwise asthmatic cocktail from the filter superintendents, and you compare what you
13:40:07 would predict from the note model right, which is that their rates are additive from a you obtain when you measure the actual cocktail of enzymes together.
13:40:18 And there it's it's spot on. Right, so the model really explains very well, the activity of cocktails which indicates that there should not be stung by chemical interactions between the enzymes secretive out or between the incentive we have a different
13:40:36 species, and the bio, the other molecules that the committee members might also secrete to the environment.
13:40:44 Right, so that that in tells us that there's no.
13:40:47 Yes, sorry this.
13:40:50 I'm still a little bit confused so I'm following up on autos question.
13:40:55 The, the, the, not a model that you presented, where you said that everything can be additive is the assumption there that you are still in sort of the linear regime of this, this Nicholas mental kinetics because in principle if you get the amount of
13:41:14 enzyme you would assess it Yes, yes, you are here with me, because, again, this doesn't have to be the case right but it is for our for our under conditions.
13:41:23 Right.
13:41:23 And if your question if you if I can, or if you're still answering our kids and please finish ages right I was going to say that that the in the supplement of the paper, I can derive all this and explain why we are in a, in this room in case you're curious
13:41:41 to see exactly why we're making the statement.
13:41:47 Maybe I don't understand something, but if you put two species in the badge, they will compete for a resource so neither of them will grow to the same abundance, as when it was a monoculture how I'm going to talk about that but at this point I haven't
13:42:03 tried. So what I'm doing is this right and growing each of the two separately. Right.
13:42:08 And then I'm taking. I'm filtering this filtering that and mixing them together right so that's I mean there's no ecological interactions at this point all there is, is if there's anything is going to be biochemical right so that's how isolating.
13:42:20 So what I'm trying to do now is I want to see separately quantify the different types of interactions that there aren't right.
13:42:27 Okay.
13:42:29 One more quick one before you move on.
13:42:31 be positive or negative interactions between the enzymes themselves right for us to simulate says do interact with one another right indirectly right so there, that this is well known right so there are endo and exercise replaces the endo cell is is basically
13:42:59 create ends right for the excess for exercises to start chewing up and possibly move cleaning off sugars of the tips. Right, so that that's a mechanism of synergy between enzymes right where one insane produces the substance for the other, right, so it
13:43:14 is possible you could have that. But there's also interactions that are possible that are not between enzymes right that are between molecules and concentrates for instance, this is not the case for these bacteria, but there's many bacteria that acidify
13:43:29 the environment right and and lowering the pH. My very well destabilize the, you know, that means for or unfold the proteins that are simply by the bacteria, right.
13:43:42 So, there are potential interactions between the molecules that are secreted by not not just the insert themselves which of course could aggregate to right and this, as far as I know this one America, but you know there's no reason why they couldn't potentially
13:43:56 aggregate and not even with a I mean listen for the greater good with all the material with others or they could be brought a stressor this really large number of possible mechanisms through it.
13:44:09 The, the enzymes, they act as a medic activity of a species may be affected by the, the extra metabolism, and a separate on have another, right.
13:44:16 So that's where we're trying to pull up. In this case,
13:44:22 right.
13:44:23 So, there's, there's no stomach chemical reactions. And in this case, but how about ecological interactions, right.
13:44:32 So, to evaluate the strength of this, the place of self is because contraction, what we did is two things which is now we combine utterly assembled consortia.
13:44:42 Right. And then we compared the they observed activity to the prediction of the interaction female model right now because we know that there is no biochemical interactions, any deviation from the, from additive model can be attributed to ecological interactions
13:45:00 of the two types we discussed before right that 111 species may actually promote or inhibit the growth of another and therefore that might lead to more or less and being secreted or that one of the species might be influencing the gene expression, or
13:45:22 this question, the submission rate of an enzyme by another species, right. So, any deviation from the norm model could be accurate of biochemical reactions as well as the clinical practice but because we know there's nobody, nobody can make interaction.
13:45:31 Everything that we see that is deviation from the norm model must be caused by positive interactions, right.
13:45:37 So that's, that's it.
13:45:41 So, now in here what we did is that we contract together.
13:45:45 What do we find, right.
13:45:45 Be more haven't system between Genesis and approximately one to one ratio, grow for 24 hours and then looked at the biochemical activity of the pairwise culture.
13:45:58 And what we do here is in gray, we're plotting the expectation from the additive model that assumes that there's nobody can be contraction and also know it was contractions of any kind.
13:46:16 In read by you see his first the activity of different agencies alone. And my husband says alone, of course that's the same as the model right because, you know, we're not adding anything yet.
13:46:21 And here, this is activity of the periscope culture of heaven system through in Genesis me.
13:46:27 And it is extremely close to the predicted value from an old model that assumes there's no interactions whatsoever, neither biochemical nor, which we know they aren't, but also there's no
13:46:42 economic contraction.
13:46:57 No a surprising perhaps at least was to us is that there, the predicted this not additive model was extremely successful for all cases, when put people in this particular bacteria was absent from the consortium.
13:47:01 Right. So whenever you took this is the last version of bears three years four member five members, and communities that do not include politics. And in all cases, the solid line, that's the, the equality between measured competitive.
13:47:16 Right.
13:47:16 And what you can see is that, what if we're all of these communities there observed.
13:47:23 I mean elliptic rate is extremely close to you predict from the non model that assumes there's no interactions.
13:47:32 Now these might be wanting to partition is that, that there are also very weak interactions, across the interactions between the 5% strains. That's one possibility.
13:47:43 However, when we actually went in and measured it to see okay how much are these bacteria are interacting with one another and giving you one example here is for this series and building a theorem is this point that you see over here, you see that there
13:47:59 is a substantial decline of the scenarios right almost tenfold. And a five fold increase of material income culture would be serious relative to what it has a monoculture.
13:48:09 So what we're finding is that in this consortia that lack people it makes sense, we are finding that there are in fact stronger cause contractions where species are growing more or less single culture than they do in monoculture.
13:48:24 Right. But despite that, we find that the that the non model that assumes that there's no interactions. That's a really good job of predicting the function.
13:48:34 So, what gives.
13:48:38 Well what we had to scratch our head a little bit and eventually we descend to this and we grew.
13:48:45 All of our backs, all the all of the speakers in our communities to different abundances final abundances right so we studying academies the same, but we were able to modulate the final density, by, by adding the super Nathan's of other species.
13:49:02 Using the trick.
13:49:03 And what we find is that for all of this.
13:49:08 This is true for all six species.
13:49:13 There, the amount families produce so that if you will the rate of amylase activity saturates at high population sizes final police and sizes so you know collate and, and you might imagine that if after 24 hours, and like an amoral without the, the more
13:49:32 bacteria have at the end, the more enzyme or the higher analytic activity, you're going to have as well, right now, what we find is that any kind of a positive interaction that would be that would move the fine population size in this region over here,
13:49:54 you will have a very modest effect on the amount of immediacy created by the bacteria. Right.
13:49:59 And and this is what we end up seeing right is that if we have that a species, a monoculture is reaches up hyper business size, but it is somewhat repressed its growth is inhibited or through competition for resources and for any other mechanism.
13:50:18 It's in mixed culture, the same bacteria grows less, as long as that effect is occurring in this flat region over here, the effect it has on the on the on the enzymatic rate is going to be a small, so it's perfectly possible to have ecological interactions
13:50:38 that could be strong and modulate the amount of bacteria or growth of a single bat, but yet we have a very small effect on on the amount of families expressed.
13:50:52 There are very actually many different types of models of growth that would give such a saturated response.
13:51:00 For instance, whenever the expression of an enzyme is down regulated by the product right and it is known for instance that emulates expression is done regulated by the amount of glucose in the environment of glucose a byproduct of families, right.
13:51:15 So, whenever the expression of an enzyme is under negative feedback with itself.
13:51:21 It's, it's very possible and there is no guarantee depends on the on the details, but one of them finds this kind of saturated dependence between problems and size and the semantic great with you also observed experimentally for all six species in our
13:51:38 communities, right.
13:51:40 So, can I ask a quick question. Yes, It's a clarifying one with this figure here that we're looking at in the preceding figure. Yeah so empirically that window.
13:51:49 I want to make sure I understand so there's this relatively small window where no interactions should be taking place between population sizes of one to five times 10 to the four cells per milliliter.
13:52:03 And so it looks like the most of the most of the time, we're looking at velocities in population size ranges where there would not be no effects. There he said that our ranges.
13:52:14 So, there are cases a good question. So, this block here where the one I'm showing, And I'm showing the section is for people in Excel because this is the one for which were able, and I'll explain why in a minute.
13:52:24 We were able to tie trade, the population size that the most right by the streak of adding the byproducts of other bacteria.
13:52:34 But for all the others, what we, you can determine a, you can estimate where the breaking point is right so where this, this, this car would start coming down right and and that's actually their range in density that we observe in our experiments where
13:52:59 we see no effect coincide with the values for different bacteria where you should they, they should still be to the right of that, that inflection point when you see basically half point of this, of this, of this curve, right.
13:53:18 So analogy, you could fit this to like a McCandless met and type function. It doesn't quite fit, it's interesting because it doesn't quite fit the line here, right.
13:53:26 It seems to be more steep and and we have, we have actually the function we tried for this, it's in the it's in the paper, when it actually matters well what you would expect from a model that like I described before, when you actually self inhibit, but
13:53:43 I want to thank you again assault right because that there are some assumptions in this a very very simple model right and there's all kinds of other things that could be going on.
13:53:52 But now it doesn't it doesn't fit well, a lamb or no.
13:53:57 It's deeper than that. Okay, thank you.
13:54:03 All right, so when people in excess is absent what we have is that I broke down all possible interactions, and before. And we find that the, the, an additive model that assumes no interaction work extremely well.
13:54:20 And the reason is that there's no chemical interactions. There are psychological absurd employees have flipped right there, there are some ecological effects Bobby's and sizes but they are inconsequential for them out events and producers because they
13:54:34 saturation effect. And there doesn't seem to be strong effects of gene expression for these five particular procedures that we try right now what happens when people in Excel is present in the consortium.
13:54:48 The situation is completely different.
13:54:49 So, this is our culture of people who make sense in Genesis, and here now, the engrave that is the additive structure functional landscape model.
13:55:01 and you read is, this is actually the observe people in exam, featuring Jensen's culture.
13:55:08 The function of that of that, Periscope culture, right, that helps us define but is there the size that the the strength of ecological interaction, they have.
13:55:18 And if we repeat that not only for bears but also for three years for member 500, or six member communities. Whenever there is politics and we find that whenever politics is part of the consortium and not always, but often, we find that there are pretty
13:55:32 substantial deviations from the data model which you know that shouldn't be surprising right sometimes that's what I had expected to see when we started experiment for all right now.
13:55:44 The main reason why we see this effect. We did a bunch of experiments.
13:55:49 We know that actually through in our papers that in fact people in mixer. We noted that people who makes that was very poorly in monoculture right that's one, I was one one hint.
13:56:00 And we found that the main reason is that people in excess of IoT knocked off so it grows very poorly in the absence of biotech are biting precursors that are added to the, to the medium growth is rescued if you had my attempt to the media.
13:56:15 And one of the things we noticed was that people in executive grow very well in in the company of every other bacillus in the community. And it also grows well in, when you filter the superintendent and added to polemics indicating that there is a cross
13:56:35 feeding mechanism of biting from every other community members to people in accelerate and that explains why we see that the functional, when you have Linux exceeds what you would expect from from from a new model.
13:56:52 In this case, where we seize the polemics a monoculture grows very poorly right, it would be in on this kind of lower side of fact, if you want to see that a proper brother it's, it's here right people to make someone a cultural was very poorly.
13:57:06 But when you supplemented with when you grow it in pairs with others, or Superman with super Nathan's right you could see that it actually moves is moved over here right and that's part of reason why the more superintendent of the other monocultures you
13:57:20 add to the people, it makes the larger growth you observe and also you see higher levels of
13:57:29 offensive any activity.
13:57:31 So while we find that in the case of of a poor grower that is being whose growth is being rescued by facilitation from other committee members, is that there are two different types of interactions here, on the one hand, the population size is increased
13:57:48 when, in periscope culture relative to monoculture so you're moving here right now, if there were no behavioral interaction of any kind. So if the only interaction was ecological interaction was through population dynamics, then this would be where you
13:58:03 would observe right. In practice, we find for putting the Excel is the test a little bit lower.
13:58:09 And that that decline can be attributed to behavioral interaction, which is caused by by people in mix expressing less amylase Inca culture than it would for the same amount of growth in monoculture.
13:58:24 Right, so we can compare how much amylase a similar amount of politics would express in monoculture to the amount of expression. You see, when it is a single country with another bacteria, and the difference can be attributed to behavioral interaction
13:58:40 of an ecological type right this.
13:58:43 This type of ecological effects on gene expression.
13:58:47 So, we can, through this process, and again this is more detail in the paper if you want to look at more closely. We were able to separately quantify the pairwise and higher the interactions that are due to both on population level interactions as well
13:59:05 as behavioral again expression level interactions.
13:59:09 As we had argued before there's very weak and barriers are higher than interactions, when politics is absolute when he says present you see that population over here interaction actually have different signs right and how they affect how they affect growth,
13:59:26 So, what I was getting to hear is that on the first question how space and directions govern the shape of a structure function landscape in our conclusions is that, that, that this structure landscape can be both quite ragged or surprisingly additive,
13:59:33 right.
13:59:43 simply by, by having one committee member that are not right, which means that for any start from there you can have regions that are very smooth right and additive, and other regions that are very rugged right so there.
13:59:58 One needs to take this into account right because it is it is very possible that to have that a, we think of ruggedness as a, as a collective, you know, as a general average ruggedness of the landscape that there could be different regions in the landscape
14:00:12 that have very different levels of randomness, or smoothness.
14:00:16 We also have seen that the effects of this three different types of function of interactions that can affect function.
14:00:23 Are they responsible to separate them and briefly and quantitatively. And finally, that even we have strong across contractions. And what it is not necessarily true that that leads to either lower perfectibility more record finish structure from show
14:00:38 landscapes, right.
14:00:39 So these are all some of the lessons we took from this exercise that I just described.
14:00:46 So, after after these what I wanted to jump on there on the second part of the totality so to once we establish that, you know, what functional escapes are, and I hope I have convinced you that at least in some cases, right.
14:01:00 They may have meaning, despite the fact that one should not simply assume right that they always do right not always they want you to expect to see a landscape as a well defined mathematical object right, but sometimes it can happen right and and they
14:01:17 can be useful.
14:01:18 Right now. Yes. Can I ask a question before you move to your next. Yes, yes, of course, have you, or how do you see the possibility for dimensionality reduction on the X side of things.
14:01:33 So, the possibility that g of x is actually can be written as g of f of x and f maps to a much lower dimensional.
14:01:45 Right. Oh, and so you're you're you're seeing all this complexity and recognizing the x base but if we were and it would be great in fact if f maps to space that we can measure and I make about you know functional composition and stuff like that for as
14:01:59 as an example, what do you think of that that's that's a fantastic question right. You might you might define x has been for instance teams right there on the sub specific teams specific pathway, or you know, or cells of the same species of cells of the
14:02:11 same dinos or, you know, higher up.
14:02:15 And I, I don't know. Yeah, I don't know what is the right way to describe the structure of our community. Right. And I think that's an open question that is super interesting.
14:02:25 And I suppose that it depends on the systems, I know this sounds like a cop out.
14:02:30 But I am not convinced that there is one right way to describe an ecosystem I think it depends right of what question you're asking about it's right, but that that these questions are very important one on one that needs to be thought of, more, more,
14:02:45 more closely and in many ways I think that I, I know that my health economic give a talk and I don't know if he brought some of these questions up here and I have talked a little bit about this.
14:02:56 And, and it's a question where I think physicists could really contribute to my opinion.
14:03:02 To help us think through that, that the issue. But yeah, I wish I knew whether tell you I think there's all kinds of really interesting possibilities of how ruggedness depends on the scale at which you're looking at right.
14:03:19 And whether you're looking at genes for species, do ecosystems become less rugged, the more you coarse grain or more I really don't know. Right.
14:03:27 But there's all kinds of really good questions there, by all means.
14:03:44 All right. So on the second part of stuck I wanted to discuss how one might navigate these landscapes in search of communities that have higher function, right, and through, you know, top down evolution engineering approaches, on a person is, you know,
14:03:50 know, different, I think complimentary to more synthetic approaches for the engineering of consortia.
14:03:56 This work has been done, primarily by two amazing residents in the lab, young Villa and chang chang, and is now being continue the next round of front by a super talented poster quantity of Columbia who joined us in this very difficult pandemic year and
14:04:10 he has made already, all kinds of really excellent progress, and the many of these ideas are elaborated on in more detail industry papers that came up over the past year and a half.
14:04:25 I'm happy to talk about any one about this offline, if there's any interest.
14:04:33 Anyway, this is a via.
14:04:35 If you have a question that is too technical, I might refer you to the paper and then have a conversation offline.
14:04:41 So, the idea of using evolution engineering. Other organism a level, right, is, is a very old one, right, as we are, it's perhaps one of the oldest human technologies that our ancestors in the Neolithic used even, way, way before they had any understanding
14:04:59 a lot of where the balance go basis of a biological function, let alone the fact that DNA even existed and was the basis of genetic inheritance for that.
14:05:09 And I think that that that's neat right because many ways it's almost where we are with micro communities at the moment right we don't we, I mean, and maybe it's a bit of a joke right but but but there's still a lot we don't know about making this degree
14:05:22 how they work.
14:05:24 But one thing we could try to do is think of whether we might even though we, there's a lot of leaders want to know about how function emerges from structure and if and when it does, is that we might still be able to engineer communities from the top
14:05:39 down without that knowledge right through engineering.
14:05:43 So that's at least.
14:05:46 I think it's a provocative and an exciting possibility.
14:05:50 And so how how does it work right how do we have the, the, the nutritional marvel that is modern wheat and corn right from
14:06:04 grasses and that our ancestors were able to read.
14:06:09 It was through a process known to all selective breeding six exceedingly simple other organism high level you employ you make use of a natural variation that occurs in the, in the population that you're stuck with, and then you interfere within the natural
14:06:36 endogenous reproductive cycle of the organism, you select seeds for reproduction, that account came from individuals that had trades that were closer to what you desire. And then that process is repeated over many, many generations and.
14:06:40 And that has led to basically all the farm animals crops that we have that we have now.
14:06:48 That sounds great at the at the organizational level right because organisms have a reproductive cycles that we can interfere with but how are micro communities they clearly do not right they do not micro greens don't breathe are made to one another.
14:07:01 So, so many times in my scientific career because I wanted to the rescue selection can actually be applied to eligible system at any level of organization as long as it fulfills three simple rules right.
14:07:18 The first of all you need to be able to make copies of it. Right.
14:07:23 And this is for instance using the evolution of biomolecules right, you must be able to make copies of, of, of biomolecules that don't have the reproductive cycle so that works for that and they should work for any biological system at any level of organization.
14:07:40 If you're able to make copies you also need to have some phenotypic variation, along the trade of interest in the parent population, you must have some population, made up by multiple different units of whatever system you want to do that evolution.
14:07:59 and or was engineering, and those must differ from one another in the trade that you want to. They want to select.
14:08:08 And finally, that variation must be hurdle, right, which means that when we make copies, as we describing the first step.
14:08:15 The, the copies must resemble their, their parent, more than they would resemble any randomly drawn individual in the population right so that means that that variation must be heritable.
14:08:29 So as long as communities are fulfilled is three rules theory tells us that our faith selection, could be possible. Right.
14:08:39 So how would it work in practice so the one, this idea, had been floating at lower levels of a classical organization so people like
14:09:05 But in the early 2000s.
14:09:07 They have Islam Wilson and Robert Elias and William Swenson extended those ideas to Hollywood systems, right, and apply our feel section of the ecosystem level, that there is very simple.
14:09:09 to good night for instance microwave has done these experiments in the 70s and 80s and 90s with populations of of single species of animal or plant.
14:09:22 You start by for instance calculate about two different microbiome micro communities in a number of different habitats.
14:09:28 You allow them all too well for some time.
14:09:32 And then you score them based on the truth and their selection to you start from multiple different microbiome so let them grow and you culture them in some container.
14:09:43 And then after that you can choose a particular property of our microbiome, which could be a direct property of the microbiome for instance the pH of the medium or the amount of a bio by the relation of a pollutant or where it could be an indirect
14:09:57 property of the microbiome like you know when for host associated micrograms it would be appropriate to the host.
14:10:03 That is induced by the microbiome right now, you once you have allow the mechanisms to develop in their environments. What you do is a you rank them based in the function of the selection and they select those that are closer to the side trade, and then
14:10:17 use those selected communities to inoculate a new generation of habitats, right.
14:10:23 Now, this last step is, is key right because that's one of the three elements that we've had before, is how do you actually make copies of the selected communities, in a way that also serves hurdle variation.
14:10:39 There's been two methods that have been tried out in past studies, and the first one to receive the wisdom of the propriety of strategy, and the strategy, what you do is that you rank, all of them, all of the communities by the function of interest, and
14:10:54 then you select those that have a higher function or one closer to your target. And then, each one of those, you use without mixing, as the inoculate them.
14:11:05 The source of bacteria and microorganisms, from which you are going to see a bunch of new, new habit. That's right. So we just grew. This community has the highest function when you just sample sales from it and add them to another flash, that is otherwise
14:11:22 identical. The first one you had in the previous generation is the same thing with, you know, any other selected community just passes them in many copies.
14:11:32 The second method is called the mixed pull strategy or the magnetic pull strategy and, and it works in a very similar way you first rank all the communities.
14:11:41 But then, before by you next after you run them you take the ones communities the community you select them and you, you pull them together you make some homogenized them, right, and then you use that mixed pool, as the inoculation from what you're going
14:11:53 going to be seeding.
14:11:58 The new generation of communities, right. So that's called the mix prosperity and these are the two methods there's variations of these two methods have been used in every study.
14:12:08 So, how well does this work, right.
14:12:09 To date, including those who are not.
14:12:10 Let me show you an example from my lab, and that I think is fairly representative and I'm going to give you some other examples from other labs. Afterwards, and it goes back to this, our, our community solve I mean only thing bacteria that I described
14:12:24 in the first part of the stock in this case we took four of those bacteria.
14:12:30 And we, all of which again are sacred enzymes that amylase is that break them starts.
14:12:38 And what we do is we we mix them 1212121, right in the center we make a for a member of the consortium. And then what we do is we blow for 48 hours just as before and after this case was 40 hours, 48 hours not 24.
14:12:56 Number 48 hours we filter the spent medium and and monitor stats in relation kinetics, and that way we determine the function is basically very similar to what I described before.
14:13:04 Well it is explained what we did is we started by growing this for monocultures, then we mix them 121212 a single tube and from that to be inoculated 24 different replicant so we started from 24, very genetically homogenous initial populations.
14:13:18 Right.
14:13:20 And then what we do is we would culture them for 48 hours and after 48 hours, we would measure the analytic activity of all, ultimately for communities.
14:13:29 And then we would rank them, and then we would select the top for communities for the highest activity, and each of them, we use this proposal seeding method that I described before where you take the top community and then you use it to randomly sampling
14:13:44 cells, see the six national offspring communities. And same thing for the second best, and the third best on the fourth best and all the others but discarded.
14:13:56 Now, what we would do then is that we would then grow all those 24 communities for 48 hours, and then repeat right we again phenotype them all rank them on the basis of the front of their activity selected the four and use them to seed, without mixing
14:14:13 using the methods for before 24 new community so we're repeating this process for 17 generations.
14:14:22 This.
14:14:24 This procedure was compared to the outcome of an experiment.
14:14:27 That was identical with the only exception that every generation will randomly tues for communities for production I went different each generation, right.
14:14:38 Or it could be the same if by chance that happens to be the same number but I mean like it's every generation just think for random communities regardless of what their rank is.
14:14:48 And you, you use them as they propel you seed for to generate a new generation of communities, right. So we have a selection line, and a single controller right there's a random selection.
14:15:02 So, what will we expect to find if our feel section experiment was working right.
14:15:07 If we plot here on the y axis then the mean I'm going to activity in the 24 metal community as a function of the number of selection rounds that we would apply.
14:15:19 Like a successful experiment something like this right, we should expect to see a relative increase in function in the deal section treatment relative to the random selection control.
14:15:30 Right.
14:15:30 But also we want to see that the, the main function is increasing over time, right with a we see a positive response to selection, right, as well as a relative difference between both lines.
14:15:41 What do we actually see what we find this when I was a biology is called predefined selection, right.
14:15:48 So, there we find that the random selection control does decline right and that leads to a statistically significant difference between both lines that grows over time.
14:16:01 But the reason for that is not that there's any improvement in their field selection line. Right. They provide your method is not improving the function of that line.
14:16:10 What it's doing is preventing its collapse, right, which is happening in the random selection.
14:16:16 Right, right.
14:16:18 Yes, of course, it kind of relates to the three
14:16:24 prerequisites that you listed so beautifully.
14:16:29 Namely, the source of standing variation. After a few selection cycles. So, where, why would you expect in this particular experiment.
14:16:43 What would you expect be the source for ongoing variation to push you towards an ever increasing function for example in selective breeding of sheep or corn or whatever, there's there's three combination and there's mutations.
14:17:00 Absolutely. Yeah. But here, you start at most with what you already like you move to the next step with at most what you already had before. I suspect you don't expect much mutation to happen between selection rounds, or maybe you can correct me on that
14:17:16 ever expect the blue curve to go all right so you you you hit this, you're like absolutely 100% right and I'll, I'll touch upon that in a throughout the rest of the stock.
14:17:26 But yeah, absolutely correct. Right. And there is an infected mutation right and that's that's seeing the fact that random selection control is going down, right.
14:17:34 So, so that means that there is variation that has been introduced by mutation among the communities. Right.
14:17:41 I mean, and I think my hypothesis, but I haven't tested this right is that these emulators are extra several molecules and probably expensive to produce.
14:17:50 So, there might be some of the lines in some of the communities that are appearing here if you want to call them tears but bacteria that do not express as much or that do not express amylase at all.
14:18:01 Right. And that those are not being selected in in the in that field selection treatment. And there'll be randomly selected in the other one right so that you're the accumulated deleterious mutations in in your community.
14:18:15 Given the understand the community as a whole. Okay.
14:18:19 It's like a you have accumulation of the materials mutations there right.
14:18:23 So yes i mean i think what this experience shows is that there is an effect so that mutation, just random mutations are enough to create variation between communities.
14:18:33 Otherwise you would not see a difference between both, but it's not helping us right so those division of fitness effects of us mutations, I mean I'm making this air quotes here right because it's not really fitness but you know what I mean by this series
14:18:48 fusion of functional effects of those mutations is clearly skewed towards deleterious effects so we're not seeing any much in the way of positive selection here right meaning that there's no variation has been to this mutation, that is, has a positive
14:19:04 sign. Right.
14:19:06 But, but that's, but you're you're hitting the nail on the head, right, the generation of new variation is critical. Right. And it's perhaps the most critical thing and I'm going to show you hopefully in the next few minutes.
14:19:20 Why, I think right that our experiment is expensive work, and why other experiments didn't didn't work either. Right, precisely for that point. Yes, thanks.
14:19:27 That brings me to another question and I realized this might actually open huge philosophical Pandora Boxx but could you be creepy that you're running into a conflict between individual level selection and group selection here.
14:19:45 So, yeah, you're imposing a selection at the group level for certain trade.
14:19:52 But actually, because the components of your system. Each selfishly replicate.
14:19:57 There's somehow a conference here that that as an individual level of the selection actually drive them down.
14:20:04 I wonder how much you've thought of that. Yeah, that's, that's my hypothesis for this particular result. But I haven't tested it right and in fact I'm writing now some proposals to test those, those things to use the system for for to study precisely
14:20:17 that that question right.
14:20:20 It's interesting to write because politics is the is the one comedian will express the most emulates, right, but it doesn't grow on its own. Right, so it's it's a tricky system as well right because it's not just the singles strain isolation is a community
14:20:36 right and i don't think i mean i.
14:20:42 At least, my understanding that we have a very good understanding of how it project interactions might for instance modulate the costs and benefits of cooperation, right.
14:20:50 So, so there's all kinds of questions What can address with the system.
14:20:54 But to get to your to your question i agree i think that that's very likely the reason, right, that there is a conflict in this case, between individual level selection.
14:21:04 Right.
14:21:05 and the fact that, you know, mutations that would that would for instance lead to polemics expressing more might not be might not be selected. But we know I had to hope that perhaps you could have mutations where that influences the the amount of biotech
14:21:19 produce or that, or that could actually wear one bacterium might be stuck to produce a cost less by product that somehow induces the expression of amylase for instance right so you could you could having direct effects where, where I mean, politics, I
14:21:35 might actually secrete more, right, not because my mutation, it has it because I'm a mutation in another bacteria, right, which might not actually carry any cost.
14:21:45 Right. So I it's at least theoretically possible that those might occur right although clearly they didn't happen in this case, but it's it's it's a question that I you know I don't think there's any reason why they wouldn't happen I think it could be
14:22:00 all those are fantastic questions. Thank you.
14:22:05 Alright so, so we fail, right, biggest pain and failed.
14:22:10 But I still think I learned something, I don't know, I don't know what you guys think. I felt good about it.
14:22:17 I also felt in good company because or back I don't know, yes, that he was, it was within.
14:22:23 par for course. When you compare our lack of lackluster effect with frankly, most other experiments have been done right, I'm going to show you a few examples right.
14:22:34 This by the way this is a very cool idea, right, I think, and the fact that it has been, there's only like a handful of papers that are published that have done it.
14:22:44 I, I, I don't know, I don't have any events for these I suspect that people have tried and didn't it didn't work right. In fact, I have heard stories like that people have tried it didn't work they didn't publish it.
14:22:56 But, but in a way, all the studies that have, you know, modest success and definitely mixed results. All of them were using.
14:23:04 Yes.
14:23:06 With respect to the point that you were making before.
14:23:10 What was the time between like consecutive selection rounds and thinking what if you had more time in between consecutive selection rounds to generate great variation who would, would that possibly help.
14:23:27 Right, absolutely.
14:23:31 Yeah, yeah i think i think it could.
14:23:33 I want to show a few examples of how one could generate a variation. Right. In fact, that's kind of.
14:23:40 Probably the final point when I'm making it up but you're right but I think there are there are things you could try to go beyond, beyond this actually this is this is how this slide kind of speaks to what you're saying so.
14:23:50 So, all of the studies that they've described, except for one right, that it's time to play for instance increase their information length right have been performed the almost exactly the same way as I described before, right, and followed followed up
14:24:04 what spend Sunday, they use the proposed strategy or the mix bull strategy. And in both cases, with a fixed incubation time.
14:24:12 And, and a bunch of other caveats that I will, I will discuss in a minute. Right. And in all cases, all of them, compare their, their outcome with a random selection control and normal about it but whenever there was a control he was have this kind of
14:24:26 run such.
14:24:28 such a problem. Now, let me show you what people found. First of all, this is a result from the very, very first study from Spencer all.
14:24:42 Here they were selected for a microbiome that would increase blood biomass, right, they will do selection on the microbiome not on the plants the plants came from the same seed stock, right, that wasn't changing between sex around so only the microbiome
14:24:48 was being selected and used to manipulate another pot.
14:24:53 And here, essentially to replicate experiments for intents and purposes,
14:24:59 where they selected for a line that increased blood biomass and another line that have a microwave or some for microbes that would lower plant biomass, and you see there there's fluctuations.
14:25:12 Over time this experiment took a year.
14:25:15 Every, every, every plant growth period was 30 something days. So yes, it was a long time so you know what I'm saying half, I think, when it's all said and done so there's, I don't know how much of the footage that might come from seasonality or what
14:25:29 have you, but where you can see is that, that there there there may be in one of these two experiments, some difference between the highway amass selected community and the law biomass like the microbiome.
14:25:41 But in the other one wasn't seen.
14:25:44 And, but in any case, what you see is that the, even though there might be a relative improvement.
14:25:50 The weight of the plant after 16 rounds of obstruction was about the same as it was at the beginning, right. And it's also seen here, right.
14:26:03 There are fluctuations but, for instance, that the low biomass is also increasing together will have our most correlated fashion. It's hard to say that this has anything to do with selection in my opinion, as opposed to any other environmental effects
14:26:18 that that are affecting both lines in a coordinated manner, there's been other other papers that have been published some have had some other successes and other paper where microphones were selected for their effect on flowering time in in various.
14:26:35 And I think it was anatomy lapses and brassica rapper two different type of grasses. And this is an example I can think was the opposite in this case, this is an days to flowering and this was the line that was selected for late flowering from this is
14:26:52 a line that was selected for early flowering and you do again see a difference between both the late flowering line increases flowering delay, right, as a function of generations, but the early flowering also increases although follower extent, right.
14:27:09 So, it again works in a relative sense also an absolute sense for one of the two lines, but not for the other line, right, and there is no no selection control to compete with me so it's kind of hard to to really make a very conclusive.
14:27:25 And this also one line right because these experiments are very time consuming, so it's hard to draw conclusions.
14:27:31 But these almost I would say I'll be the one exception and probably one of the one of the most remarkable success stories I know there's a couple others, but most of them are in some cases you see like some flooded negative results experiment data work
14:27:45 and even the title of this paper is the microbiome wants wants microevolution overtakes experiment the host minute indirect selection. This is a, you know, the next experiment didn't work right this case they were trying to read a microbiome that would
14:27:57 modulate fight development. And when they compare our control line here in red and orange with the selection line there you'll see any difference between both, right.
14:28:09 So now the question is, there is why this experience when more successful. Right. And I think one very clear candidate of something that could be done differently, is the method of our production the propaganda strategy and the next posts priority right,
14:28:36 Right now, that's our hypothesis and perhaps the Commedia production methods are not ideal. And, but really tested this hypothesis of these experiments are a ton of work.
14:28:41 They're very very time consuming.
14:28:53 And it's hard to, it would be very difficult to do experiments with enough replication enough parallel lines to draw to explore the space of possible modifications to the protocols, you could do, right.
14:28:57 So we started to do these computationally right.
14:29:01 To that end, we used in silicon customer resource models, which I believe I don't need to explain to you guys you have been hearing about from pan catch and George both for has been there too and Michael different have probably also discuss them to some
14:29:15 extent, So I'm, I'm going to gloss over this.
14:29:17 So we were using customer resource models with cross feeding and where consumers interact bye bye competition and secretion of resources. And in our simulations we could take for instance synergy through his way, and our simulations, we find that we,
14:29:33 we take that one of the nutrients, we consider it an environmental pollutants, right, that we want to remove from the environment. And the goal is to engineer a community for by a degradation of that one polluted.
14:29:48 All this is done in Silicon Valley so we can establish multiple communities. Bye. Bye.
14:30:00 And we can, there's no evolution in our simulations right.
14:30:06 But we still have a biological, we can introduce new genotypes into each community through ecological processes like migration and, and various other processes.
14:30:14 Anyway, at the beginning of this experiment, what we did is we just replicated all the experiments, or the experimental protocols that have been published today in our in Sligo communities where they only caveat that there is no evolution in our in our
14:30:29 experiments right so we're trying to see to what degree, simply through ecological sorting.
14:30:35 This could potentially work.
14:30:37 If there is no innovation at the species level revolution this business but.
14:30:44 So under these conditions, what we did is, for instance, we were comparing a pagan strategy that has been proposed by one of the authors, potentially our own group, with no selection control which is a controller has been absent from all the microbiome
14:30:59 selection studies that we're aware of, not sure exactly why because that that control was present in some of the early work in group level experimental group level selection.
14:31:10 So, in this novel section but all we're doing is we're passing every community without selection right so it's like every community for instance in this business provide your method but so we are simply passing in the proposed strategy.
14:31:22 Only we select a few communities and those communities we get past it in many copies right for no selection control every community is selected and passes to exactly once.
14:31:30 Right. So, there, there is no group level selection in this experiments. But there is individual level selection on the starting variation that we added the beginning right so this is going to be psychological dynamics and each other communities as they
14:31:44 will totally agree. And in both of these what we do is we do 20 rounds of committee on the selection or 20 rounds capacity with our selection. And we simply allow their, their communities to equity rate, and after that point where we're going to do is
14:32:13 we're going to compare the community after it's all every protocol has been applied to our competition or communities. We're going to compare the best performing community in the provider strategy protocol with with testing with the best performing community
14:32:15 in the, in the control where we did nothing we just simply pass in the community, as many times as we did before, right.
14:32:20 Again this comparison between the best community in one line and the best community that control is often missing or not reported in previous experiments and we think that this actually is important.
14:32:30 So, what kinds of articles we expect right so if we now try all of the various protocols that use their patio and then, Mexico strategy.
14:32:40 Yes, come back to the question about how much evolution is going on within the communities. Right, sense of that from the early experiments.
14:32:52 No, no, no one has no one has quantified and all the to my to my knowledge, when you're picking the best, and then replicating that one. Whether you're each of those will just be basically the same going forwards, or whether they're going to change.
14:33:10 Jimmy and experiments run the simulations.
14:33:17 The experiment I just described, where we use the defined community. Right.
14:33:25 And and attempted to do quintiles select an experiment is the only experiment has been done to date with a defined community where you could compare us ancestors with the derived straight after the experiment is done.
14:33:35 So in all the experiments that have been done with with basically and which may not EULA me was, was very difficult to do it right and and most expensive.
14:34:00 But no one has really quantify how much evolution there was or how much of the scene effect has come from ecological change right in a competitive way.
14:34:03 sequencing and metagenomics So, so no one has really done it right. And for our experiments we didn't do it right because the experiment didn't work.
14:34:03 Though in hindsight, we should and we are doing that now, right in the expense we're doing at the moment.
14:34:14 Yeah.
14:34:14 But yeah, but in this.
14:34:18 Sorry, go ahead. No, I was I was going to say that in this simulation there's, there's no evolution at all. Right. And we, we only have ecological processes here Right, so the species are in the provider method all that's happening chat population then
14:34:33 I make something. We are sampling stochastic so you could have species lost during during something so it's something is done this quickly. Right.
14:34:40 Yes.
14:34:43 Okay, so.
14:34:46 Okay so, so when we do, we compare our selection protocols, what we would expect if if a selection protocols are working better than a simple screen we would expect to see that if you compare the best community in the selection line with the best community
14:35:01 in the screen that we should have that that difference is larger than zero, right, that, that the selection produced of one community that is best.
14:35:09 Better than the best community we have seen we have done nothing. right. And, by contrast, if protocols didn't work, right, there will be no better than a simple screenwriter you would expect to see them the best community in the selection line should
14:35:21 be the same as a community we have gotten if you have done nothing and we've been lazy and just passing your set your, your communities, right.
14:35:28 So what do we find, you know, consistent with the lack of success of previous work we find that very rarely this protocols, again, in the with a caveat that we didn't have evolution going on, but simply for ecological mechanisms are just not enough right
14:35:43 to generate variation that could give rise to higher vibration capability.
14:35:50 After all the trouble of they're applying the selection protocol, then you would have seen.
14:35:57 Otherwise, and here that's true regardless of whether you use a mixed pool strategy or the particular method, are in purple, right.
14:36:06 So, and now we're running out of time.
14:36:09 So I'm just gonna tell you why we think this is. And one of the big problem we find that in the simulations, is that all of our communities were selection started before the communities had had tend to find an equilibrium.
14:36:24 Right.
14:36:25 So these what happens if you take the top performing community in a no selection control. Right. And, and you track it after services after the first incubation you take their community with how has the highest vibration function.
14:36:38 And if you track it over time, that community answer being once after multiple passages community reach the state of of equilibrium generation like glue as we call it, that becomes a completely mediocre community.
14:36:52 Later on race so by selecting before communities are finding equilibrium state, you're jumping the gun and selecting for communities are not going to be very good.
14:37:02 When they stabilize.
14:37:14 The second point is that selection requires variation of communities but it requires that variation is heritable right only little variation counts and hours means that only variation in function that comes from the fact different communities have different
14:37:21 species that write our own is useful right for for evolution to work right.
14:37:28 So then, in other words, you need high heritability right at the community level.
14:37:34 High Community High heritability is necessary right because it.
14:37:38 You can interpret the heritability up a particular way of thinking of community and equilibrium, by, by thinking on their structure functional landscape is when you have communities of high function, and are stable, when they are surrounded by other communities
14:37:49 that are also high function, right, whereas communities that have lower function are surrounded by other stable communities that also, on average, lower function right so when that is true, right, then selecting on a high function community is better
14:38:06 than setting a random community from the population. When that is not true. Then I heard that with these very low, that it doesn't matter whether you're selecting a high functional function community because there's no correlation between the distribution
14:38:20 of, you know, fitness effects, if you will, right the functions that you see in neighboring stable communities that with the function that you that you have so we have a very rigorous fitness landscape, or you should have very low heritability and adequate
14:38:36 level and that should lead to the failure of our divisional community level selection.
14:38:40 One of the, we have examine the degree of predictability if that is and how these two different reports are starting to factor into reality. And they both performed very poorly, and I think I we haven't tested this experimentally and I don't have time
14:38:55 to go through it running out of time but but we do see a very stark effect where when you start communities with a lot of variation, at the beginning of the experiment where you have a lot of where you see communities from different pools of species so
14:39:12 so you have a lot of variation at the beginning, how by applying the prepackaged strategy iterative Lee.
14:39:19 What at the beginning you have a very high heritability and you see that artificial selection performs a lot better that compared to our control. Right, random selection of job but after a few selection rounds, all the communities are descendants of the
14:39:33 same initial innocuous, right. And after that point when the frequency race to get all of your communities in the pool, come from the same apparent.
14:39:42 Then heritability collapses and when that collapses are if your section stops working, right.
14:39:49 So, so there is a heritability at the community level, that, that, you know it's it's a bit of a nuance term right it's not the same as readability in England through genetics, but it's, it's I think there's some pilots that are that are interesting.
14:40:03 And that we are now trying to explore, understand a bit better and how that correlates with the, with the program, but basically the structure of history, what's the landscape of how ragged it is versus how small it is.
14:40:15 Small the landscape should give rise to two more heritable tweets at the community level.
14:40:22 And the opposite would be true.
14:40:24 Alright so, just to sum up, we took a prisoner doing selection before committees we took weariness about idea, and that they provide you and the mixed methods are pretty bad reintroducing heritable variation after its selection well, right.
14:40:43 And that leads to them not being very good and those are the methods that have been used so far.
14:40:50 We have been introducing a bunch of basically our our our ideas is actually quite simple for how to this better. Well you're proposing is that you should start communities from as many diverse parts of speech as possible.
14:41:01 Let the molecule liberate right. Let the find find a state of stable equilibrium, and then select those that have the highest function, and then perturb them right i into this perturbations by by either introducing new genotypes from for ecological means
14:41:19 or even stimulating mutation. And then after that let those community second fine state of equilibrium, and then select those stable communities that have higher function and iterate that process always select them hi function, and let, let, let, the
14:41:35 community equally rate between selections.
14:41:37 There's very use ecological mechanisms that one could use to reintroduce variation that are not the provider or the mixed bowl strategy, you could for instance do coalescence where you can take the top, the highest performing community and then mix it
14:41:52 with every other community in the pool, or even with those select a few.
14:41:57 You could just rent to these migrants from the regional pool of various different regional pools. You can apply severe bottlenecks that can also lead to that principle One could think of knocking and knocking out single species.
14:42:12 But I'm sure you can you can think of, of any various other processes for instance increasing the incubation time or decreasing it and having variation that way.
14:42:23 All of these methods we've tried so far actually work better under the same conditions than the propeller on the next bill strategy. And when, when they fail, these that we just described before a workout.
14:42:37 I think that that's kind of the final thing I want to say right that despite the fact that we, people have not have generalized success when they have attempted to do this, this procedure.
14:42:52 I think that there's a lot of potential here.
14:42:55 And it will really help the efforts of were able to translate what we know for population genetics or to draw parallels perhaps from publicity genetics to, to the structure function landscape and community dynamics right there it's not as straightforward
14:43:09 thing to do but I think it's worth the effort. And I think there's a lot of open questions there and a lot of open field for for people to, to help us crack this problem.
14:43:23 We are currently doing experiments. So hopefully, if I have the chance to tell you in the future. Our, I will tell you what we're finding.
14:43:32 That's it, I'm running out of time so I just wanted to thank every all the folks in my lab.
14:43:37 This has been a very collaborative and fun place to work, and that's it I think we have time for any more questions but I'm happy to take them offline, if, if, if necessary.
14:43:51 So I think everybody.
14:43:58 Thank you.
14:44:00 It was really, really interesting. So yeah, we're out of time for questions but I'm sure people can follow up later.
14:44:06 Tomorrow for you because really later. Yes.
14:44:13 Thank you. That was me thank.
14:44:13 All right. Thank you. Bye everybody.