13:05:10 Originally I thought well I'll give a completely new chalk. And then that know that will be a complete disaster. So instead I'm going to give three new talks and see how that works out.
13:05:14 Thanks to the organizers for putting this together, especially thanks for inviting me and then lastly for allowing me to talk with all of you.
13:05:21 I think there will be many questions of course what these are all motivated and stimulated by the discussion yesterday about how perhaps we need new models for studying microbial communities.
13:05:35 I'm going to show you a lot of data related to things that we have been doing.
13:05:39 Of course, ask questions, if you so desire. But I would encourage you to see some of the data before asking the questions because at least for me, the data was very surprising, and I think it will make movies to be different questions for all of you.
13:05:52 But I just spent like two minutes, telling you about myself. I know some of you but not all of you.
13:06:11 Some of you will appreciate that that is not me now because I don't have any gray hair, but I started as a theorist, this was me right before I started at Stanford, and I was a postdoc with Ned where I fell in love with bacteria in particular questions about cell biology ourselves organized how they
13:06:15 How do they grow with some spatial structure and I sort of conceptualize my whole lab around this idea of combining bottom up, and top down, approaches, and somehow linking the middle in the middle where theory and computation were of course, really critical.
13:06:29 And for us, you know things like modeling the bacterial cell wall was kind of our, our specialty. And I will just flashed up there for 30 seconds while I talk.
13:06:41 So, we're a champion to combat our lab and sort of captures this whole aspect of like multi scale modeling and experimentation, we're really interested in particular and how to incorporate both genetics and mechanics into this picture, to try to build
13:06:55 some fundamental ideas about how cells grow, and I bring all this up because I think it's very relevant for the discussion about microbial communities.
13:07:05 We have a close collaborator Justin Sonnenberg a good friend of mine who's a glycol biologist studies the role of diet in the gut microbiota, and we taught a class together, specifically to try to bridge this boundary between physics and modeling and
13:07:16 biology, and we start to conceptualize the same picture that, you know, microbial communities are made up of units Let's call themselves but maybe you have other ideas, and we want to understand phenomena at the tissue and organism level, and perhaps
13:07:29 we need new tools to do that as well. So, I'm not a department bioengineering I don't think of myself as an engineer, but actually a lot of this talk to me about like how do we actually design something that is useful in some particular way that I'll
13:07:40 lose a date, but I love this idea of model systems, probably many of you can see your favorite model here. And so the question that we've been trying to ask for several years now or in one of the properties of a model community and this sounds somewhat
13:07:55 simplistic but I think it's kind of weird that we don't have really a model community maybe you, you do and I'd love to talk to you about that.
13:08:03 But, in part, I think it's because no one has really elucidated what exactly the end game is.
13:08:10 So throughout the course of this talk, I want to discuss three things.
13:08:15 One, actually I'm going to mostly skip over this but it sort of is going to be in the background of everything I'm going to tell you about which is to try to understand how communities like the gut microbiota are robust perturbations like antibiotics
13:08:27 and pathogen invasion.
13:08:29 But most I'll spend most of my time talking about the development of synthetic communities for interrogating gut, physiology and then at the end, and I definitely want to get here because I think it's a good lead into packages talk so I may do some rejiggering
13:08:44 rejiggering in the middle, to make sure we get there, I want to tell you about some recent work that we've done on the physical properties of longitudinal dynamics in the gut microbiota.
13:08:53 So that's just where we're going. and the first part of this talk is all work in close collaboration with Michael Fischbacher who's a chemist at Stanford, with his postdoc, Alice and Andreas in my lab and Poli in my lab, and it all revolves around this
13:09:09 idea of trying to develop a highly complex synthetic community of gut commensals. By the way, I probably should have had a slide in here about why the gut microbiota is important.
13:09:21 I will just tell you there's a lot of, there's a lot of different species they do a lot of stuff. There's a lot of genes, they affect your health a lot in a lot of ways, you've probably read something about it.
13:09:29 You probably think some of it is not scientifically well founded I share your concerns about some things, but I think it's a fascinating field. So, that's my intro to the gut microbiota.
13:09:40 I'm sure many of you could have done it better.
13:09:42 So
13:09:45 I for a long time have been thinking to myself, we've been doing mouse experiment and the first time I talk which I cut was all about mouse experiments, and we spent five years getting to a point where we thought we understood how antibiotics affect the
13:09:57 gut microbiota, and I thought, oh my god this is painful. And there's got to be a better way of trying to understand what's going on. And I started asking people like.
13:10:10 Could you just do everything in a test tube and overwhelmingly the response was well you just cannot capture anything about a guy, you know, that's what you don't even have a host, and I couldn't argue they're wrong but they couldn't argue that I was
13:10:20 wrong and I so we started to conceptualize well what is it we're actually looking for, well. So we certainly would like to be able to study a complex community, get to a complex means in a second, it really needs to be stable and reproducible for the
13:10:35 results to mean anything.
13:10:37 It needs to be store level and scalable so that you can actually do experiments.
13:10:41 But perhaps from a biological standpoint, you'd at least like the composition of the community to in some way mimic the composition in the gut. And even more importantly likely to have functional characteristics that might mimic say the dynamics of perturbation
13:10:55 in the gut, and of course I'm alluding to the fact that I was going to tell you about my experience with antibiotics, and I'll just show you the highlights in a little bit.
13:11:02 But first I'm gonna talk about a bottom up approach to building community so the idea is, let's say you take humans or mice and what you might want to do is you might want to say, I have a whole lot of isolates, and perhaps I can put them together into
13:11:20 communities and put them into mice, or even humans one day and study them. And this is not something new, people have been doing this for ages in the gut microbiota field, there are certain well defined communities, for instance, is a collection of 12
13:11:33 members have been studied for a long time, one called author Chandler flora, and a lot of discoveries have been made, but there are certain aspects which 12 species, simply cannot recapitulate at all.
13:11:46 And so there's a question of scale and completeness.
13:11:50 So, can we generate a community that somehow can recapitulate all or nearly all of the functions of the gut microbiota, and also then if we had such a community, you can imagine all sorts of games you can play.
13:12:03 So, for instance, you can take a species out and ask, how is that important for the structure of the community. You can modify a species Yellen secretary and make a mutation in one organism if it's genetically tractable, you can add a species in a pathogen
13:12:18 or something otherwise and study colonization resistance. And so, I think all of these are kind of like the dream and I'll show you where we've gotten in all this, it's very first.
13:12:32 So, causation resistance is one.
13:12:37 So, you know, colonization resistance to certain bugs is first for sure.
13:12:44 causation resistance to every organism is not achieved by these 12 members species I don't think has been talking about that, you might say, well, that's kind of too lofty have a goal, I'll actually show you experiments in mice where we think, in some
13:12:55 sense, we've achieved a complete community.
13:12:59 There is this conversation.
13:13:00 Okay. Yeah, I mean, I'd love to hear other ideas, because I think what we're trying to do is get to an enabling point where, you know, we tried now the next big causation was in such an obvious thing because so well studied, there's clear hypotheses,
13:13:14 some of the hypotheses turned out to be wrong, as I'll show you but there's certainly like a well defined course forward. Yeah.
13:13:27 I'll tell you how we chose our species that's probably more important for the rest of the talk, but the 12 is easy, could you summarize the question, if the oh yeah sorry the question was, how are the 12 species chosen, there are aspects of that which
13:13:39 are interesting and uninteresting but we're not going to talk about it all for the rest of the talk. So let's talk about the hundred and four species that we started working with.
13:13:46 So basically what we did was look at the Human Microbiome Project, and we took all the most abundant organisms and all the most prevalent organisms, at least as many as we could get our hands on and that ended up being 104, you could argue that we missed.
13:13:58 I don't know 15 or 20 that might have been nice but we just couldn't get them, so that was that. Um, so, the main point that I want to highlight is that you cannot read this, don't worry.
13:14:10 probably everything you think of is in here.
13:14:11 The color is the file is colored by phylum, and every major file is there, and there's a lot of diversity in particular in the Firmicutes in red, and the Bacteroidetes shown here in green, and we'll come back to this.
13:14:25 So, I want to just say that issues that arises when you deal with a complex community is that you might expect that the abundances are now going to range or many orders of magnitude, right, this is probably going to be logged normally distributed and
13:14:42 most analysis platforms crap out at like point 1% ish, which is highly problematic for us. So we spent ages developing this platform called ninja map which is enables highly accurate remapping.
13:14:57 I'm happy to discuss the details for anyone that's anyone's interested. The main point is that it takes into account everything that you logically think of the main downside is it requires sequence genomes for everything is there, so for an undefined
13:15:10 sample it's completely useless, but it's great for what we're doing.
13:15:13 and to show you what we can do. So this is now a community of 104 species. It's colored by the abundance. After 48 hours in rank order. And you can see here that it starts out relatively clustered as it should, because we, as close as we could just normalize
13:15:33 to od. And then we grew it and we sampled at 1224 and 48 hours, you can see that most of the dynamics happen, relatively early on, as you might expect as well because the nutrients are getting consumed and the main point here is that indeed the abundances
13:15:48 are very over like six orders of magnitude.
13:15:52 And here is a comparison of two technical replicates, so this is two combinations of 104 species on the same day. And this is a biological replicates where we grew them on completely different days and combine them together, the technical replicates are
13:16:06 highly reproducible, a biological replicates are okay.
13:16:10 But the main point here is that, certainly for the technical applicants that we can measure, very far down. And so, this is at least giving us some confidence that we can make measurements.
13:16:25 So what do you grow in them in like media was yeah so this is this.
13:16:27 Hi me first.
13:16:31 I think this is in sa sec. It's a minimum media, we did, mostly all experiments in sec and mega media. As it turns out, as I'll show you in the second part of the talk, I wish we hadn't done that but it was sort of beside the point for what we were trying
13:16:48 to do but these are defined media, not.
13:16:51 Some are defined and some are not a mega media is not defined. Okay. sec Yes. Okay. Yeah, that's a specific to define because it enables to do dropout experiments really easily.
13:17:01 So this is where you can see one second oh yeah sure sorry question on from Julia. Julia Can you go ahead. Oh hi Casey, I was just wondering if you have you quantified whether there's active fish infection in these communities.
13:17:19 Yeah, that's a good question. um, we actually haven't even though we have massive amounts of metronomic data so it would be quite straightforward to do so, there hasn't been an obvious reason for it, thus far, there's no obvious signatures of dynamics,
13:17:34 and we actually sampled all of our.
13:17:36 All of our communities at 1224 48 hours so we would at least have the chance of seeing some dynamics, so we just haven't been motivated through it but I think it's an interesting question.
13:17:55 Yeah.
13:17:55 No, this is literally just one batch culture of growth. And so there's lots of questions about packaging that I'll answer shortly, but this is put it this way.
13:18:07 This is, in retrospect, not how I would have done the experiments, but it ended up still giving us a lot of interesting results.
13:18:14 It's just growing for 48 hours from stationary face. Exactly. Yep.
13:18:19 And so this is an experiment just shows you what you can do, and I just, the main thing I want to highlight is that, you know, nothing crazy is happening.
13:18:27 So, you know, with the complexity, you might worry that you know interactions would cause a lot of results. So what you're seeing here is the effect of dropping out each individual amino acid one by one, and then grow in 104 member of community.
13:18:43 They are ordered in order of perturbation or response, as I'll show you on the next slide, and.
13:18:54 But again, you know, the overall structure is relatively concerned with the same distribution of abundances, but nevertheless there are straight nutrient interactions.
13:19:01 So this is just a heat map of where most of the interesting interactions come from.
13:19:06 You can see here the distribution of how many significant changes there were basically just calculate the z score because we have all these communities that are all logged normally distributed in the, in the distribution of each individual species, we
13:19:19 just ask how many have a significant change. And you can see here that some amino acids, cause no significant changes whatsoever. Basically that means you drop out an amino acid, nothing changes by more than like 30%.
13:19:32 Yeah.
13:19:35 Yeah, so one passage is effectively like seven generations. Yeah, so it's not a ton of growth. And that's a very valid point that maybe would see something more interesting if you, if you grew from more generations.
13:19:48 The nice thing is that some of the interactions that showed up, were both robust and interesting. So, this is a case study where when you drop out leucine an Argentine, you see effects on broad knees and costumes for our journeys.
13:20:06 And what I'm showing you, it was actually not known that Spurgeon is required Argentine as a carbon source.
13:20:14 So we were able to show that indeed growth without a carbon source without art Sorry, I said current projects will not grow without Argentine. You can see here in batch culture as an isolate it won't grow it all without Argentine.
13:20:28 And we were able to find the gene, the two genes here that are involved in Argentina Argentine synthesis. So I think this highlights two things that one.
13:20:39 You can actually see a dependency in the with within the ocean of all of the other metabolic processes that are going on with the other species, and to then that it enabled us to zero in on something, you know, new biological aspects of metabolism.
13:20:57 Yeah. Just a question. What should I be the mindset of, like, evolution.
13:21:03 Evolution not happening here or are you taking measures for relational Yeah, I doubt evolution is happening, we have actually looked at the metronomic data.
13:21:11 So, I mean, all of these quantification has to be done with metagenomics because 16 us both would run into problems with overlapping 16 sequences, and it just wouldn't have the resolution to see a community within a community of this diversity.
13:21:25 So, we have actually looked for mutations, but even in.
13:21:30 We put it this way more likely to happen or mouse experiments I'm going to tell you about in a couple of slides. And it's really hard to tell. So I think the adding the answer is in on these timescales you're not going to see anything.
13:21:45 Yeah, and I'd be curious to get people's impressions of whether you, we should be trying harder in the context of the mouse experiments. And can I just quickly, nobody in the community was able to compliment out that Argentine.
13:21:59 I'm able to, I mean you're in a complex community here and no one was able to separate like I'm surprised that it wasn't able to get Argentine from out of its friends, I know it's happening.
13:22:06 Yeah, it's pretty wild, actually, we were pretty surprised about that as well. Yeah. And, you know, there's any, I think, one, one complaint which is very valid about all these dropout experiments within a very complex community is that of course redundancy
13:22:21 kills your ability to see anything. And so you know you might be doing a lot of experiments for nothing.
13:22:27 You know, I think, the argument here someone surprisingly is that well this is like a really efficient way to at least find one guy that was dependent RG and maybe this is like, you know, highlighting many aspects that way.
13:22:38 Yeah.
13:22:41 So, I mean, I take your point about nobody complimenting but at the beginning, you don't have a lot of other cells right you're, you're nervous at the high dilution yeah so by the time they would be complimented yeah there are no other nutrients for them
13:22:55 to grow on. Yeah. Good point.
13:23:00 You have a similar story for leucine or
13:23:04 dependency on Lucene was known.
13:23:06 Yeah, I guess maybe following up with what Nick just asked, instead of starting from equal amounts of the initial inoculate if you just pass it to the actual endpoint.
13:23:16 Yeah, they just maintain the same ratio growing up or DC entirely different things stock yeah yes and no. There's a lot of media dependency and save that question for the second part of my thought.
13:23:29 So, we also looked at strange strain interactions so we grew 104 cultures where we removed. Each strain one by one. And here, I have to say going into it, we kind of expected like a lion king effect, or like you drop out one guy and there's all sorts
13:23:43 of reconfiguration. I think now that was kind of silly but still the answer was the exact opposite. You can see here the distribution of each species across every one of those hundred dropout experiments that are really tightly distributed all the red
13:23:57 dots are just the abundance of that species in the suppose a drop out. And that actually tells you something about like the degree of Miss mapping, so we actually incorporate that degree of Miss mapping into our subsequent analyses because in some cases
13:24:12 you know we can't tell two species apart and so you know we acknowledge that.
13:24:16 I'll just highlight a couple aspects that, you know, that I think are interesting about this. So one is the this is a case study, where you see that a drop out of this Hasidim in Akaka strain causes a large causes a large change in in a start.
13:24:36 This is the drop out here, and then it causes an increase in the other very closely related acetaminophen aka strain. So, that's something in the redundancy which you'd expect to see what you see here the positive and negative interactions.
13:24:50 For dropouts of each species, and this a Kak a dropout is perhaps a positive control, because it was the most dominant member, and since we're measuring relative abundance of course you see a lot of statistically significant changes when you knock it
13:25:03 out. But it's not just that we're getting rid of high abundance members and then the whole thing has to reorganize because you can see here that the fraction of strains that have some effect here.
13:25:15 So these are like all the guys that are preservative as a function of their abundance, here's the really high abundance guy but a lot of these guys are have abundance like 10 to the minus three.
13:25:23 And so they're not causing statistically significant changes just by being gone. Yeah. Can you tell me about resolution of measurement like we got error bars like what's the technical noise can I take.
13:25:34 Yeah. So really resolve that much like 10 to the minus two to tend to them. Right, so I'm just like, yeah, would, you know, right. So, um, Here's what I can say.
13:25:48 One of the aspects which is interesting and frustrating about the in vitro side of all this is that there can be huge variation from day to day.
13:25:56 And I showed you that I showed you the data that are starting communities on the same day there, they're like, you know, 99% correlated. And so, we think that what we're doing here, essentially, all of the other communities are replicates except when
13:26:13 there is some statistically significant change. So it's similar to what you would do in like a functional genetics like chemical genomics study where you basically just assume that the underlying distribution has no changes, and you use everything else
13:26:25 for statistical significance.
13:26:26 But the day to day variability is just like it isn't chemical genomics as well. The really remarkable thing is that all that goes away mice, everything in mice is absurdly reproducible, and I have no idea why is he can ask a question before you go on
13:26:40 to the mouse.
13:26:41 What are the two points statistics of the dropout experiment look like is there a covariance structure there that's simpler to understand or is that just point we haven't looked at that but for the, for that for the same reason we have not.
13:26:54 I've been aware of reading too much into it because because of sorts of noise that would get amplified. Yeah.
13:27:01 So, yeah.
13:27:06 So they're a little strange to interactions with about competition so not actually not competing for shared resource yeah so there are strange, strange interaction.
13:27:15 So, for instance, This is now categorized.
13:27:19 This is basically, I want to. Yes, explain the y axis yeah so so this is the this is this traditional metric of binary co culture interaction, where alpha greater than zero represents a represents, and a negative interaction in binary code culture where
13:27:41 the sum is less than you would expect based on the sum of the parts of course all sorts of reasons for that, like, like resource competition.
13:27:49 But we, what we did was we attempted to validate the predicted interactions from the strange dropouts, and many of those do have a positive action not all, not all of them.
13:28:01 Many of those do have the predicted interaction, not all of them you can see here that many of them are centered around zero. and that's not surprising because you could imagine that in the these either require a community context, or there, they are
13:28:13 just noise in our measurements. But, you know, for instance, we were able to take one of the things that Michaels I was really interested in is clustering storage nice.
13:28:21 We have a lot of genetic tools for this organism. And so there was an unknown interaction with lactic caucus like this, which we then attempted to validate both groups through CPUs and through liquid co culture in both cases you can see here, for instance,
13:28:35 that the laxest has very little ability to grow in the absence of his progeny, and that's exactly what you see here with the, with the CO culture, there's more growth, and if you measure CF use the lack this will only grow more in the presence of spotty.
13:28:52 Yes, sir.
13:28:54 I was just wondering, so you didn't do serial dilution yet right so you don't know how many of those threads will survive if you were to do a serial delusion, know but what I will show you in a second is that Sarah dilution in certain contexts can easily
13:29:07 preserve 50 to 70 organisms. Okay, good. And the strain and directions would be amplified I guess and serial dilution because yeah now you see very modest effects but once that's it that's entirely possible.
13:29:19 Yeah, so I don't, I don't want to. I don't want to overemphasize the aspect that the stranger interactions are rare, what I would argue with that this is the least a starting point, or things haven't blown up in your face.
13:29:30 Is there a question from the online.
13:29:38 How you're normalizing things so if there's a large number of each of which are interacting with many of the others than the individual interactions are going to all appear pretty small.
13:29:55 I don't know how to answer that because I don't have a prior for what the magnitude should be so we are we are literally basing this the interactions, simply on this assumption that I said before, statistical significance, based on an underlying null
13:29:58 Are you normalizing in some way that takes into account. Yeah. No, I mean,
13:30:10 hypothesis that mostly interactions are zero, simply because that is what we appear to observe.
13:30:19 I don't I don't necessarily argue the best way to do it is.
13:30:25 It's sort of our first stab
13:30:30 Casey I have a question did a bag of your online.
13:30:35 So the community is always collapse to the same input and point if I understand that correctly so that would suggest that there's really no statistically so that the way that the community.
13:30:45 Yeah, develops from the inoculation to the endpoint is completely deterministic in this case. Yeah, and then that would go with what you said that from day to day.
13:30:55 So, the same day. It's deterministic but from between different days are some differences. In fact, that's a great, that's a great point of the elephant.
13:31:05 And if I if you walk away from this talk with any message. What I would say is that, at least it came as surprised as a surprise to me that deterministic aspect is pervasive.
13:31:17 In fact, the only the very few like three out of 10,000 community examples where they're stochastic city are all interesting, primarily because there was a very rare extinction event, but it basically never happens.
13:31:35 Yeah, good following up on Daniels question I suppose it. Have you guys looked at all at sort of collective modes in these fluctuations that you know any change would be small, but if you put them together, we've done that in the context of our in vitro
13:31:47 communities, and because there I think the packaging becomes really important. And there are really interesting collected modes, where for instance like there's an Enterococcus which clearly inhibits the growth of many species that you never would have
13:32:00 been able to see. And so, to Daniel's point there probably are low.
13:32:06 There probably are low magnitude interactions that are being masked and the same thing, of course happens in like, yeah, like a chemical genomics, email, so many types personally could get pretty good effect size estimates they are based on.
13:32:21 Yeah.
13:32:23 Can you just talk about what the protocol is for growing the in vitro system like you, you straight them out and then you make three cultures and you dilute them to equal optical density or something like this like Yeah exactly.
13:32:34 Okay. And so, do you know if the day to day variability is due to your inability to get sequel od or something else. I'm almost sure that the inability that the day to day variability.
13:32:45 that was a worst case scenario where
13:32:50 humans got better and better at doing it, and then we just decided that humans weren't still weren't good enough and we bought a robot. And so the all the data from the robot is a lot better.
13:33:02 I'm sorry, one more question. Yeah. Are y'all when your computing this is this just relative abundance Are you taking in some absolute biomass madman into it, we haven't been taking into account absolute biomass for the other in vitro communities I'll
13:33:16 show you we do, and it's really important. And I wish we'd done that here but hindsight is 2020. Yeah, thanks. Yeah, sure.
13:33:36 Given what you've observed that there's no dramatic variation with these dropouts and things like that. I'm wondering how much.
13:33:43 If you just take the initial measurements you have not at the end point of the seven divisions but at the early points, and you just take the assumption that you know the doubling times of all your strains.
13:33:53 Yeah, and you just propagate them assuming there's no competition at all like hold that hold that thought Hold that thought that's exactly what Yeah, we're gonna get into that.
13:33:59 Okay, so, but first let me talk about what happens in the mile in the mouth.
13:34:03 So, this is where things were pretty wild so here's the protocol where we took the frozen stocks, we normalize the cultures we colonized mice, we sampled mostly every week for eight weeks.
13:34:18 And this is the distribution of abundances of each of the organisms across across life, there's like a few organisms where there is outliers.
13:34:30 We don't understand those yet, it does happen, but, for instance, here you can see that the initial community the inoculate them by week four it's again spread out of like six seven orders of magnitude.
13:34:43 Here in a week. You can see what happens over the weeks. It's basically settled down there are some things that come in and out in abundance, but they're low abundance members, and basically by week two it's settled down.
13:34:57 And you can see that it settles down into a structure that roughly mimics these are now colored, not by rank but by filing a structure that looks like what you'd see with humanized patients with patients samples, and I'll just highlight here that we're
13:35:10 going to use three humans fecal samples, they're all quite different.
13:35:15 And we're going to use them for what's called a backfill strategy so this comes back to Terry's question of, like, what does it mean to be useful or, or have a functional significance.
13:35:25 So what we did was we said okay, we're pretty sure this isn't everything that you need. So what's missing. And so we took these mice, and then we challenge them after four weeks, either with a PBS negative control, or we challenge them with the vehicle
13:35:41 sample.
13:35:42 Right. And then we asked what we could get in. And you can see here that by week five, all the green dots are new stuff.
13:35:49 Sorry PBS is phosphate buffered sailing, it's just, yeah. Yeah.
13:35:54 And we don't see anything coming in in the PBS negative controls which is great, and not surprising, but you can see here that, you know, even after eight weeks, there's still other stuff that's entrenched, and it's a different set of stuff with each
13:36:08 of the three samples so you can see here, these basically are colored by whether they weren't there before the number invaders.
13:36:16 And, but there are a couple things I want you to know. So first of all, a lot of stuff from the original community sticks around it, roughly speaking, retains the same structure.
13:36:26 So the color here is colored by week four before the challenge, and you can still see that the Reds on the top and the yellow the bottom and the blue sit in the middle of the blues at the bottom.
13:36:35 So, that says if you haven't like massively disrupted the whole computer community, but new stuff is coming.
13:36:40 So what we did was we said, Now, let's take those mice.
13:36:44 Let's figure out what we're missing and let's try to get as many of those strains as possible. Right, so let's backfill the community with this stuff that we know wants to come in.
13:36:55 Yeah.
13:37:00 Well, while you're in the microphone, let me let me show you. Let me show you the second part of the data about this. So, this is just to say, that is a quantitative statement what I said about the community not being so affected.
13:37:14 Those are basically a correlation coefficient of abundances where you can see here, that when we challenge. The with, with the human sample 123, if you include the invaders.
13:37:28 The correlation coefficient is around point seven compared to something in the high 90s before the challenge so that says, something's changed.
13:37:37 But if you consider only the inputs basically the correlation coefficient of their abundance is still around 90%.
13:37:43 They're not bad. Yeah.
13:37:44 In those previous parts Can you remind me what the colors represented where they are the colors are basically always going to be rank ordered based on some route and in this case it's week four before the challenge.
13:38:01 So basically I just said like how similar are things to week four well to be civil they have to read at the top and blue at the bottom, I see.
13:38:06 It's almost like kind of visualizing a rank some sort of missing the reason why you want to backfill, is there some aspect of the community that you think you're missing and then this is motivating.
13:38:19 So I think even like at a higher level.
13:38:23 The question is, you know what, what should happen with the backfill right, is this community somehow robust to invasion. And I would argue the answer is, so So, because it doesn't largely get disrupted.
13:38:35 but stuff comes in.
13:38:39 And then the question is, if we can make it can we make it better, or we just started taking a random walk in the space of community center would mean I combine it with another human community and there's less that I would need to back so that's what
13:38:50 better means. Yeah, well I'll show you what we think better means, and then, you know, but.
13:38:56 So now, we better mean a mouse community instead of human community.
13:39:03 Well, so, I think no because there is definitely going to be aspects of a mouse community that would change this picture, I do you think this process is a very viable way of creating a complete mouse community as well.
13:39:21 So I think if it for instance if your goal is to study something like composition resistance. In either case, that that's sort of where we're going with it.
13:39:33 Yeah.
13:39:39 Oh yeah, it's all the way at the top, like, yeah, yeah, yeah. So the highest gray is like below. It's like a few percent of all right. And in this case, yes.
13:39:53 Yeah, this is one mouth.
13:39:55 Okay, you have a you have 1% that have invaded. I actually think that, actually.
13:40:10 Actually this is probably a, this is, I think this might be a just a.
13:40:11 We should probably fix this to make it better. I think there is a great dot even higher. But the reason I'm bringing this up is because the most abundant organism then well okay, sorry.
13:40:21 No, no, because this is a little bit confusing, because this is the abundance. After rebuilding the community, but so I'm not sure it's going to be complete answer your question, but let me tell you what we did.
13:40:30 So we got an augmented community where seven species dropped out anyway. So we just left those out, and then we added in 22 new species, and so that ended up giving us 119 species that we did the same process to, and then we challenge with the same people
13:40:46 samples. And you can see here after four weeks, basically this is the new distribution, and all the red ones are the invaders. So now, one of the invaders is the most dominant organism.
13:40:58 I don't know that's interesting or not. But, but again, the distributions of each organism are pretty stable, and you can see it here, where it's basically the same context as before, but here's the part which I really want to emphasize.
13:41:14 So for instance, here is, after the challenge, comparing week for week eight. Okay. And all of these guys in the black circles are the invaders.
13:41:26 But every single one of these invaders except these two guys right at the bottom.
13:41:32 We're invaders that we detected before. There's ones that we didn't happen to get a strict we just didn't get an isolate for them.
13:41:40 And so, in some sense, if you're willing to buy that had we put them in, they would have ended up just like this which I think is a very reasonable assumption.
13:41:49 It basically says that if you go to the effort of identifying invaders, and then getting them, then you're done.
13:41:55 At least we would have been done in this case, I guess these are individually raised my so if I let them share food. Yeah. Would you just get this for free.
13:42:05 Like, if I took your synthetic community yeah and I just didn't have very clean cages.
13:42:11 Well, would you get it just from from exchange you just fill in changed from the colorful.
13:42:20 Yes, but then you wouldn't have a defined community, you probably would. I mean, that's how Yeah, like, I think that's how that's how we got to the point where we are with the gut microbiota, the whole point is to have a defined community.
13:42:34 So if you re ran this.
13:42:49 instead of instead of instead of doing a garage with this patient sample Yeah, we just put it with somebody else. Sure, yeah, I mean, the reason we didn't do that is because this is actually easier, because, Yeah. Yeah. But I think your point of know
13:42:52 But I think you're pointed no I think your point is well taken that like, it's all about it's all about just having a reservoir, and then seeing what gets in, and then the backfill process kind of gets you to completeness, which we found pretty surprising,
13:43:05 like we thought, okay, you know, you change the community to a reasonable degree, you know maybe now someone else gets it.
13:43:10 I think it comes back to the point where like interactions are not so crazy that like everything changes it's, it seems, you know, relatively, for lack of better word deterministic.
13:43:28 Yeah.
13:43:29 It is neither is just three human samples. We don't even know what's in there.
13:43:51 hundred and 404 we picked two were from the Human Microbiome Project, based on abundance and prevalence. I see and but those are, but I just like I'm trying to understand with a pic from similar patients or like the co worker patients are in the Microbiome
13:44:01 Project I guess I don't know the answer to this. They're kind of all over the place that I mean here everything we chose to a large degree was prevalent, so they were there in most people, which was kind of the idea.
13:44:16 Sorry.
13:44:17 Oh, yeah, sorry, the straight Yeah, strange specificity was actually from all over place and.
13:44:25 To be clear, actually what's interesting is that what we did not do was isolate from the mice, right, because I felt like cheating.
13:44:32 So we actually went and got strains from people of the same species. In order to build the augmented community.
13:44:43 So, this is just the deterministic aspect, so this is the correlation coefficient amongst abundances. And this is actually technical replicates and this is across five biological replicates where we literally took different sets of mice and completely
13:44:57 different months, and basically you just get the same thing every time. It's really studying in fact, there's no higher correlation with education there is across experiments, which I find totally wild.
13:45:09 I mean, it's like, that's not what you see in a normalized, you know in a normal humanized analysis experiment. Yeah.
13:45:18 What do you like give an idea of what is driving this like super high coral Is it like the copper ag that they just sort of recycle everything or.
13:45:28 Yeah, I don't know. I mean, like, I almost don't want it to be the case that the undefined aspect of a normal experiment, somehow, is what is what gives rise to it.
13:45:35 Although it is it is.
13:45:39 I don't know, we were quite shocked.
13:45:42 Tonight, but of course it makes them for a great experimental system.
13:45:46 Yeah, yeah.
13:46:08 Well, no, it's using a different Eloqua each time because we had a bit of the same have the same value movement.
13:46:19 If you have some weight sorry if you have someone else ratio if you if you challenge with a fourth person.
13:46:24 Yeah, good, good, good question. Yeah, yeah, we. So, this question of causation resistance we actually tried to address with a pathogen. And so, the main point here is that, you know, they're this 12 member community coming back to where we started, is,
13:46:39 is that that's in red, and basically it has little to no colonization resistance to the heck.
13:46:48 Whereas, our community does pretty well to keep out, he heck all together, and it, notably it has no protein bacteria in our, in our community so it's not that there's just some closely related organism which is competing with the, with the call I sorry
13:47:04 he heck is no remember the call I. And so I think this is at least you know one step towards the idea that we can start to now piece apart. What gives rise this colonization resistance.
13:47:19 And one of the things we've done is to drop out each file and separately to see what contribution each file of makes and quite interestingly, several of the five not all of them but several the Fila, give a signal, they have like intermediate causation
13:47:32 resistance, which means that it isn't just like one bug or one metabolic processes having effect. and that's something you just couldn't do without a defined community that's, that's like why we want the defined aspect.
13:47:46 Okay. So all of that was to get us to the stage where I at least want to spend like 15 minutes dissecting.
13:48:11 that I've heard thus far, for instance, looking at steady states to do multiple passages, how can we understand like the diversity that persists and all that.
13:48:15 So this is Andreas, he's, he was a fantastic gratitude in my lab, he calls his work the ecology of poop tease actually poop T is something you can buy in what looks sort of like a Trader Joe's box here with this label, but poop T is actually been around
13:48:32 since like the, at least the fourth century BC in China, where the Emperor would, you know, drink poop tea for various ailments. So now this is a top down approach to community generation.
13:48:45 And the idea is, let's say you take a sample from wherever you want. And you ask what kind of communities can be derived. And then of course, like you can generate isolates and do the bottom up approach as well.
13:48:58 But the question is what does this intermediate look like you know what's kind of the space of communities and to what extent can we forward predict what's going to happen.
13:49:07 So what we did was we may use of a mouse experiment we had going on, which I thought was kind of the perfect marriage. So, we were looking at so called humanized mice so these are germ free mice that have a human fecal sample just like the sorts of things
13:49:20 we've been talking about, they were on two different diets, and then they were treated with super Parkinson for five days. And what we did was collect fecal samples from the night before the treatment.
13:49:31 At the time of what turns out to be peak disturbance, what we call the residual treatment, right after treatment ends and then post treatment after two weeks, and the main point here is that the composition changes a lot on it.
13:49:43 as you would expect due to separate treatment. You can see here that species like the verruca microbiota expand the factories drop out and then come back again.
13:49:53 But even after two weeks things are different. So, this is basically a set of fecal samples that this is two different mice the pairs of bars. That's reproducible across mice but lots of changes across time points, but they're all coming from the same
13:50:05 human donor right so all mice, had the same theses going in the same principle all the same reservoir of organisms. So hopefully that gives us an opportunity to get rid of aspects of like strain level changes or anything like that.
13:50:19 So, there's also differences across diet. And so what we did was we took all these people samples. And then we asked how they would behave under batch packaging so not keeping stats.
13:50:34 Basically every 48 hours we dilute them one to 200 to come back to win games question that we undergo roughly 50 generations of, as a community, we pass them in many media but these are the four I'll tell you something about because they're the most interesting
13:50:48 for different reasons. And so this ends up being like 296 small plates of samples that are being passage every 48 hours. And it's interesting actually for all the grad students the audience that this is a really neat scale, these sorts of experiments
13:51:02 now we do very regularly across all sorts of different initial conditions. And, you know, you can.
13:51:09 It's interesting to think that what you end up generating are depending on how you count hundreds to thousands of communities with, you know, like efforts on the level of, you know, a couple hours a day for you know for two weeks.
13:51:22 What we did was try to look, do you typically in phenotypically through microscopy flow cytometry growth itself and 16 A sequencing.
13:51:33 All the cultures are kept anaerobic the entire time. They are never frozen and 37 degrees the whole time in fact they're like rushed from the mouth to the anaerobic chamber for the experiments.
13:51:46 And the neat part is that after two weeks, there's a ton of diversity, so this is just a phase contrast image of one of the first samples we did this with you can see all kinds of crazy stuff here.
13:51:57 Right. So I think it's really encouraging you know this is this is maybe more or less surprising to some of you, the part that's also really nice is that they're really stable, and they vary quite rapidly transition, which you can see here is that each
13:52:12 What you can see here is that each color is a family, and within two to three passages. The competition is stabilized and basically the only thing that's really changed about the initial structure of the community is that the introductory ACA which is
13:52:26 like the Nikolai has substituted the verruca micro via. That's not so surprising because the Broncos needed museum to grow and in fact we know now that we can stabilize the verb goes with with music.
13:52:36 They're also incredibly reproducible from a technical point of view, you can see some differences but they're basically they're almost indistinguishable.
13:52:45 And in fact, these are the correlation coefficient of the three technical replicates so these are being passed it separately for two weeks, but you end up with the same endpoint.
13:53:00 Yeah.
13:53:00 Yeah, this is a VASV level. So actually, basically the only.
13:53:07 Yeah, so, so the level, the resolution reproducibility is far higher than if Yeah, exactly. I should have mentioned that thanks for doing so basically, this is the this is a correlation across as vs, which are, you know, thanks to Ben Callahan and the
13:53:22 audience. You know, we think of them as having species level resolution. You can see here that things that appear in one replica and not the other all below 1%, and basically everything that's there is at the same abundance in, in both in both parents.
13:53:36 Yeah, I mean.
13:54:02 I mean,
13:54:02 have you been able to identify what exactly those entire Rebecca ratios are. Yeah, okay, like because there is at least for the 16 s there's usually a lot of degeneracy yeah we have actually, in in a second.
13:54:15 I'll show you that if you look across human donors, there's really interesting variability in the interbank Tracy effect of Sony i is one that appears in a vast number of donors.
13:54:26 But, for instance, this particular donor, he forgot Sony is, is the dominant and our vector ACA, except that after antibiotic treatment, it clearly goes extinct is replaced by another director ACA, a club clo, and that change actually is incredibly important
13:54:45 for things like colonization resistance. Yeah. Yeah, good question.
13:54:53 What's in the media, and what's trending sectors and yeah so in this case all the media are undefined. But bhi is
13:55:04 brain, heart and fusion. So that tells you a lot about what's in it. TYG, Scott. Jones and yeast extract glucose.
13:55:13 Why CFA interestingly is a media that was designed to enable culture ability of a large range of organisms. I'll show you in a second. It's absolutely the worst are culturing communities.
13:55:26 And what did I what I say and gamma is.
13:55:30 It's another, it's another, basically we chose a whole range of media that had been shown to
13:55:38 in isolation promote growth of certain classes, and we wanted to see like, you know, is, is that aspect important for any particular reason, I'll just show you hear that one of the exciting parts of the up bhi is I told you, we've got different time points
13:55:51 right from the antibiotic treatment, and that changes the composition right so each one of these points in the background, are the shapes are the inoculation right and so the innocuous actually moved this way and then back this way.
13:56:08 And then the shapes the lines a big thing what happened to the communities. After passing. And basically, for all of these different starting points, the communities do the same things as their inoculate.
13:56:22 So, for reasons that we're beginning to at least have hypotheses for this is at least a media that for these different in ocular drove the communities to retain the same starting point so it's not as if you just selected for the same set of organisms,
13:56:37 regardless of the fact that you had different abundances in in the initial an ACO. I guess the reason I'm asking is, it seems like something about the mouse gives you reproducibility.
13:56:49 Yeah. And you also get that reproducibility from an anaerobic chamber. Yeah. And so I'm wondering if you know like in a mouse it's obviously it's anaerobic but there are other electronic sectors in the mouse, right.
13:57:01 If there's something about the respiratory pathways that has to do with the reproducibility like the traffic cascade that describe test and so on. That's a great question.
13:57:11 I can't say anything too intelligent I'd love to talk about it more
13:57:18 can say.
13:57:18 I hate to break, break the flow, but just getting a little lost in the.
13:57:27 He could help me get my bearings. Yeah.
13:57:29 So you started by saying, what are like property, good properties a model community. Yeah. And then there was a part of the Talk where you were building them bought like the big pictures there was bought them up and other section of top down.
13:57:41 Yeah. And like the big point, you're driving home was the reproducibility across yeah so like manipulate ability or the phone yeah so here what I would argue is the great part about bottom up and said you kind of get what you put in, and crazy ship doesn't
13:58:00 come out come out that you'd never would have expected, which is like, good and bad like as mentioned to me that, that, you know, you can see organisms pop in and you're like, holy shit I didn't, we didn't even know that was there before hand.
13:58:13 The.
13:58:15 The downside of the bottom up approach is that you have to know what you want to put together.
13:58:21 Right.
13:58:30 So the answer to that answer that question.
13:58:31 My question for the first part is Yeah. Did we win, like, was it was it good was it like so.
13:58:34 the answer to that answer that question. Rest soul in the hands of the three reviewers right now.
13:58:39 But no in all seriousness, I think we like.
13:58:43 I mean we gave ourselves a pat on the back.
13:58:46 And then, we're moving on to the next, you know, the next step in this process. I mean I think of it as like, I've been surprised at how well things are going so far.
13:58:58 And so I'm more convinced that this is like a profitable way to try to understand.
13:59:05 Okay, let me tell you the things that I think is useful for. So, first of all, I think it tells us something about an argue it tells us something about the niches, of the various organisms that are there and how they can, how they potentially can compete
13:59:21 with each other.
13:59:23 I think it tells us something about you know Kenyatta obvious point.
13:59:29 I don't, I have now more convinced that there's like a hierarchy of variables that are controlling community assembly, and you just have to get like the ones at the top, you know, for instance like digital communities are totally different, but I think
13:59:46 I'm more convinced now that that's not, because there's like a whole new world we have to explore we just have to like, identify a few key features.
13:59:55 I realize these are like, you know, slippery sorts of goals, but I think it sort of as I said at the beginning, the bottom up, step, we want to have something that we can manipulate and we know that we can.
14:00:10 And for the top down. What I'm going to show you is that I think it gives us a sense of some of the rules across many many communities, which is very difficult to do in the context of bottom up because, like, just don't, you don't know what start with
14:00:28 that when you
14:00:34 Mike duty
14:00:50 I really want to see it's a follow up so you, you change the ratios of these species. Yeah. At the beginning, and you see what they go to stable fixed point right so that that is another measure of like robustness.
14:00:52 Yeah. Um, I'll show you some data about a second but yes that absolutely is the case.
14:00:56 I'm going to skip over this.
14:00:58 And I'm going to skip over the here, Terry has a question. Yeah, it's, it's clear what game you're playing, was a bottom up at the top guy is not clear what what you're what what's the degree of freedom, you have to play with.
14:01:16 I mean, you've shown us for out of the many medium you've, you've used and okay so for those, we should the most interesting in a sense, reproducible, right, I know that right.
14:01:28 But then, but what, but then you don't really care about what's in the media.
14:01:38 So I'll show you those two data sets in a second. Those are the functional aspects that that this allows us to address, rather than know there's something, there's something you need to do to see weather resistant.
14:01:46 You know, the way you're describing what I'm playing with, I wouldn't say, I wouldn't say we don't care but I'll show you in a second. So, I mean, for instance we're interested in similar questions about both antibiotic sensitivity and colonization resistance.
14:01:57 Yeah.
14:01:57 Yeah.
14:01:59 So, just really quick. one of the aspects that was also encouraging, is that basically, you can put these communities back into mice, they will call actually the, not only colonized but they'll basically bring the inner vector ACA back down which is for
14:02:16 all sorts of known mechanisms, and then we were able to compare the community colonized to humanize. And basically, in particular, especially related to like immune system function that the proteomics the host proteome from the community colonized mice
14:02:34 is much better than injury mice as you can see here. And so, you know, we think that things are going roughly back to normal. Once you put the community.
14:02:43 Okay, so.
14:02:46 So one question along Terry's point is, what can you use this for, and in some sense that's what novel you're trying to to and so in our case, we're going to ask, Well, you know, how can we take a community, and then perturb it in some way and asked whether
14:03:00 it's predicted. And so, what we can do is to just apply the same Cipro treatment that we did in the mouse to communities that did or did not experience, Cipro in the mouse and ask, How much would you have a look.
14:03:16 Can we have learned through from these five years and mouse experiments. And the answer is a reasonable amount.
14:03:22 years of mouse experiments and answers, a reasonable amount. But we can also do things that you couldn't really address in the mouse, for instance, ask about those dependence and ask about questions of resistance versus resilience because if you see something.
14:03:33 Stick around, is that because it was resistant, or is it because it was able to bounce back after a perturbation, and of course here, we can do things like apply treatment continuously, or we can apply to treatment leave it out, then we can even apply
14:03:45 the treatment again.
14:03:47 And so, first, as you probably are not surprised to see if we compare the pretreatment community, to the residual communities, it's a community generated from a fecal sample that experienced Cipro, then the pretreatment community is more sensitive to
14:04:04 the Cipro.
14:04:06 And you can see here the diversity loss that occurs as a function of concentration so this illustrates you know the sort of tight tradable aspects of the experiments, you can do in vivo, and then you can see that Cipro basically ends up, selecting for
14:04:22 certain species.
14:04:23 So, here it depends on the dose. You can see here that the blackness Bray ca are the ones that you know really are able to stick around at high concentration.
14:04:36 And that's not so surprising because blackness race here like Gloria are very are very resistant. So this is the MIT of each individual species color with his fractional change, and overall In fact, it's pretty remarkable that you'd be able to predict
14:04:52 a lot just by knowing what each individual organism,
14:04:57 what it would do during growth with with Cipro.
14:05:01 So, so what we did then was a transient treatment, where we just apply one set of Cipro. I'm going to argue that that one does have Cipro is similar. One sec is similar to five days in a mouse just based on the number of doubling, and then we released
14:05:18 the Cipro and Ascot recovers Yeah, could you go back a slide I didn't went by too fast I didn't understand what I was supposed to learn. Yeah, explain.
14:05:24 It's basically just saying that like the guys that went up during during separate treatments in vitro are the guys that you would have predicted because they had higher Nic.
14:05:34 So, this one means you went up in vitro to in vivo no sorry this is an isolation, it's sensitivity to how much it changed in the community. Okay.
14:05:45 Yeah. So, the ones over here are the ones that are resistant. In, in isolation and went up in abundance within the context of a community I see so the x axis is.
14:05:56 Yeah, I mean this of course, there's lots of resource competition questions that would perturb this relationship. And I think that's a really interesting set of questions but, yeah, but in large part, sensitivity does tell you a reasonable amount.
14:06:12 And so, in the transit treatment I'm going to skip over this part just to say that there's other features than that, that occur for instance. So Patrick ACA, and the Erica Casey are actually able to bounce back after, after the treatment so they're basically
14:06:29 undetectable they're below 10 to the minus three, and then they come back to you know 1% afterwards and so these are organisms that they still could be resistant right and you could have slept with their growing back up again.
14:06:42 I can tell you they're not because then you treat them again and they go back, they go away. And so I think we were excited because this sort of allows us to separate the different organisms, into the ones that are going to go up or down.
14:06:55 Based on features that we wouldn't be able to see in vivo, yeah.
14:06:59 In this.
14:07:01 In the previous night we were talking about MSC presumably you also had growth rates from just doing odd. Yeah, right. So, do you Did you check if the growth rates were, like, the growth rates are around was sufficient because I would assume, either at
14:07:16 sensitivities of the strains are a little bit, you know, low if they're just much faster growing, so it was super Fox isn't growth rate and in this case was calculating Am I see from.
14:07:30 I believe we're calculating from finally the measurements.
14:07:33 But they're basically the same. It's not true for all drugs. So, some drugs because of tolerance and things like that the two aren't correlated.
14:07:42 But yeah, it was separate system, you're right it growth rate would have told you just as much. Yeah.
14:07:47 So, okay, I am.
14:07:53 I would just say as so okay let me, let me just pause for a second and collect my thoughts.
14:08:00 So I think one aspect that is
14:08:06 encouraging, is that, basically, this panel summarizes what we thought we learned from the in vivo experiments and where, you know, we were able to dissect across lots of different perturbations what was going to happen to each organism, as a function
14:08:26 of separate treatment. And the thing that's exciting is basically, not only can you recapitulate all of that in vitro, but also there are nuances that as I sort of hammering on now that, that wouldn't be accessible in vivo.
14:08:42 And one of the things that is really apparent is that if you focus in on like one family like the factory DCA. This is all this is even true so so this is what happens in vivo, but then in vitro, what we can see so yeah so in vivo you see that if you
14:08:55 break down the back Troy is shown here in gold into its constituent species that are all labeled by this is Bo goddess in uniform us and khaki and so on, that basically we can see all of these features emerging in different ways.
14:09:10 For instance, be theta is able to stick around at low doses khaki and uniforms are able to stick around at intermediate doses. And same thing with a bogus.
14:09:18 And if you look across all of the, the factories, basically we break them down into resistant recovered and extinct strains, and that's exactly what you see in vivo.
14:09:35 And so, so I think, you know, to a certain extent, we think that we're getting to the same point that we would have gotten to in vivo. Yeah.
14:09:48 Because the right 12 species.
14:09:50 Yeah, so, but bottom up and top down are not inherently different totally right. You don't need to do this well but I would argue that you need to top down, in some sense to tell you what to do bottom up.
14:10:09 Well, Okay. Well, yeah, I know you're taking your mouth result, yeah and then just just take individual ones and then look at their individual redundancy okay there's a good correlation and that's it.
14:10:20 I don't argue with that. So for instance, one argument against that is this is way easier, and way and enhance enables a much larger scale for those experiments.
14:10:30 Right, so the act of even deciding to do what you're what you're saying, would add months to the process.
14:10:37 So I have not argument that isn't a good way to put it, and we do do that. But there that's the reason why why we progress in this direction. Yeah.
14:10:45 Okay. Okay, sorry. Yeah. Okay, so I'm just gonna say one more thing about in vitro communities.
14:10:52 Because I certainly thought,
14:10:58 oh yeah sorry.
14:11:01 I think we've gone through all of this I mean I'm sort of hammering on a lot of I've been, like, Yeah, sure. So, um, there's a section of my talk about not going to talk about, which is now about trying to understand some of the rules, where we now very
14:11:15 the media across 48 Media conditions. And it turns out, I will show you the very last slide, as a teaser, that it's like a Marvel movie after the credits, where you want to know Kevin foggy.
14:11:30 But I will just say that.
14:11:35 Where do we go, because I want to get into something that I think at least Pankaj will find provocative or interesting, I don't know, maybe one or the other, where we're basically.
14:11:50 The goal was to ask if we were to say grow communities across carbon sources. To what extent can we predict exactly who's going to win. And the answer was, basically, that.
14:12:05 Well this is how what we could do this was our prediction here compared to what happened. I'll you know, let you ruminate on that for 10 seconds, but the punch line is that all of this prediction is based on the growth of single Iceland's, but it does
14:12:19 so in a way that sort of, it's, it's sort of like, it sort of like cone Sham theory for for community assembly, where there's like a, an objective function that lumps in everything about about the interaction.
14:12:34 So, I'm happy to talk with anyone. And if you're annoyed that I'm not showing you more. I apologize, can you just define what when means, sorry, when.
14:12:42 But what, what does it really mean.
14:12:49 Did I, what did he say when he went Oh god, what does it mean to win the prediction here is the ratio of what turned out to be the two dominant species, which are the inner vector AC and purple and the Inner Circle K CAA in.
14:13:02 In, breath. And so this is the prediction, versus the data for the ratio of the two.
14:13:14 I will be happy to talk with you about that lunch but you will not be able to understand from that side. That's why I said, if you're if you're annoyed, I apologize.
14:13:22 Casey before you before you delve into the next year. I'd like to get you to a bit more on the determinism in the mouse can.
14:13:31 So just to recapitulate early in the synthetic community experience of the hundred four strands. It's not really reproducible day to day.
14:13:38 Okay, well okay so here's, here's.
14:13:42 I wouldn't I wouldn't say it's not reproducible day to day.
14:13:45 You know, I think, I think that's about as bad as you can do.
14:13:49 I will say that the one of the big differences is that both in the experiments I just showed you where, you know, from a top down point of view, the correlation coefficients are as high you know 99% 95 99%.
14:14:03 Okay, I think it's because it does take time to reach it does take time to read steady state, and we haven't really given it that in the context of the hundred and four member community at the beginning.
14:14:15 And then in all these media conditions that you propagated these other communities in it, going back to Terry's question, is it, do you get reproducible structures and all of them all just different.
14:14:26 Yes, I see all of them. There's, there's basically one set of three communities that turned out differently, and it was all based on whether one Enterococcus train went extinct or not.
14:14:39 Okay.
14:14:40 So it's a recap, you know this is related to the work of gosh and Albert Oh of course.
14:14:44 Yeah, in some way, you can write or stand right So Josh and all bros case is like totally different right like they do technical replicates, and every single one has a different set of species that come up.
14:14:55 Now, I still haven't, I mean, I mean, I've talked about this a lot, and I still haven't figured out whether they think it's because of that there's there is like this massive extinction event that they have at the beginning because they treat with a ton
14:15:07 of drugs to get with things like the fungi, probably has an effect. There's also a delusion I can remember. I mean, in the in their paper, they did a pretty big delusion.
14:15:18 And so, I don't really remember if they knew for sure like how close to extinction points they were so it's possible that there's nothing surprising about the fact that they saw a lot of variability at the single species level, and we don't see any towards
14:15:33 is, is there something to understand about those final structures from the media compositions, Allah, you know obvious question. Yeah, there are some electronics afters, and then the software requires show you, you know, is there some yeah there's simple
14:15:46 or not so simple story that one can tell or you haven't tried it I thought level. Okay, I mean, I haven't tried for not some not great.
14:15:57 at all. And so it's I mean it is like, it's like one of my great failings and sad parts of my life but the, but you know that is not an excuse so much as a fact.
14:16:20 I will say, you know, you could say, well, I'm pissed off that used all these undefined media. And, you know, because like it sort of puts this black box that's hard to deal with.
14:16:31 That's true.
14:16:32 We did actually all those carbon source experiments in em nine as well. And what happens is somewhat akin to what overseas and not surprisingly, diversity is a lot lower.
14:16:46 Now it's not that know there's some carbon sources and communities that will support.
14:16:51 Dozens of species and 15 families. So that's really not bad at all. But still the best and nine community is worse than the worst community where we just added a little bit of complex media to get things going.
14:17:09 Where Best and Worst diverse is just diversity. Yeah, so I don't know there's like a conceptual divide in terms of media composition where, if you restrict yourself to define media.
14:17:21 You are going to be studying fundamentally different communities, then what you would study otherwise. That's not bad, it's just not what we've been doing.
14:17:31 You can make synthetic complete media.
14:17:35 But that does not mean that that it will support community for reasons I don't understand it was, yeah.
14:17:42 I totally agree, but but if it's not what I can give you an answer to. Yeah, yeah, yeah.
14:17:47 Okay, so in the last 15 minutes.
14:17:50 This is sort of switching gears, but it's something that I think is very much in the spirit of what you're going to hear from package and in the class.
14:17:58 So this is work with Ben and postdoc Povey where, you know, this is really kind of about why you get plots like this. So, what this is illustrating our abundances of various organisms in one individual across hundred and 20 days.
14:18:21 Okay. And so these longitudinal time series now are there not that frequently published but there's enough of them in different communities that we found this very interesting.
14:18:33 And there was a paper from Dennis vid cups lab that was published last year, where they noticed some interesting statistics.
14:18:42 And so for instance if you plot the daily abundance change. Okay, on a log axis, relative to this business right, this is the histogram of daily abundance changed.
14:18:54 For all these reasons, you get this across distribution.
14:18:58 There's also all sorts of other features that are interesting so this is the residence and return time, because you'll see species seemingly go extinct and then come back again.
14:19:07 And then this is the distribution of a standard deviation of abundance changes, relative to the abundance. There's scaling walls here This seems attractive.
14:19:19 Yeah.
14:19:27 Loot abundance relative abundance, they're all relative abundance.
14:19:27 Yeah, that's the that's the data we have for all for all the cases. Yeah.
14:19:30 Yeah.
14:19:34 Exactly.
14:19:35 So, perfect. I was just about to say that. So then there's a paper on my archive maybe it's published now, but then said, Well, this is all bullshit because you're telling me about daily changes, but I scramble the time axis, and I get the same thing.
14:19:50 That's interesting. So, I will hopefully explain to you how those two are not actually contradictory.
14:19:56 Okay, so, and in part, the way we came to that was, because that's looking at this data set, but other data sets have similar statistical qualitative properties that are not quantitatively the same and that was very revealing.
14:20:09 Okay, so here's what you did. And so he took a consumer resource model, and which you're going to hear a lot about so I'm not going to explain it but basically, the idea is that you have equations that dictate how each species expands as a function of
14:20:25 the concentrations of certain metabolites or resources. And in particular, one thing that he did, which I think is very biologically motivated is not I think done.
14:20:37 Traditionally, is to just apply a sparsity criteria, basically to say that instead of just having, say a random matrix. We're going to say that this matrix also has the property that for each species, some fraction of the nutrients are simply inaccessible
14:20:50 to it. And that becomes a parameter in the model.
14:20:54 And so, so what do you get if you do that well okay so and then the second thing he did was to say, Well, what could give her I mean, you know, this is deterministic problem.
14:21:06 And so what would give rise to any fluctuations. Well let's assume that all the fluctuations, come from the environment.
14:21:13 Okay, So let's assume that all the metabolites.
14:21:17 Why are fluctuating.
14:21:19 And, but there's also a driving force to come back to some mean level.
14:21:24 Okay, I think I have that written out somewhere here.
14:21:30 I do not I apologize but anyway yeah the basic idea is that there's some noise strength in each metabolite abundance. And then there's some spring constant k, that tells you how quickly noise is going to, or what's the correlation timescale the noise.
14:21:45 And so then what he does is he now assumes that there are these repeated cycles where the system will propagate to something near steady state, and then the metabolite abundance, will change.
14:21:58 It will reach a new equilibrium or new steady state, and that will be like the daily abundant change. It makes sense.
14:22:07 Okay.
14:22:10 So, the output is an abundant time series, just like the one I was showing you experimentally, and I'm sure you can you were not surprised to see, to hear that basically for two of the most sort of eyecatching results in the, the experimental data in
14:22:29 particular syllabus distribution, representing the histogram of the abundance change is is preserved and, for instance the abundance variance sigma squared, has a linear relationship with the mean abundance.
14:22:43 So, question is, how can we actually compare that to experimental time series from this one from the human gut microbiota. And so, what he does is he first asked what what are we going to try to compute.
14:23:02 And so one of the things that he computers, the thing that I alluded to just last slide, the variance scaled with the abundance as I'm showing you here schematically here, and you know the model can fit the data, extremely well.
14:23:18 You can also.
14:23:20 You can also reproduce the disruption of abundance changes, as I said before, you can also reproduce, quite a lot of other features like for instance you can compute the residents and return times.
14:23:30 How long does it take things to become extinct and how long does it take to come back, and as I showed you the data. So the solid, the circles and squares represent those two distributions.
14:23:41 And again, it managed to capture those features fairly well.
14:23:46 There's some other things that can capture as well that weren't addressed in the paper, like the distribution of restoring slopes, so you can calculate for each species.
14:23:58 What is the timescale over which it returns back to its mean, and that matches the data fairly well in fact, this is basically like all the things we could think of calculating, nothing is perfect, but we were pretty happy with how things, how things
14:24:16 appeared. And in particular, related to Michael's point, all this data is very sensitive to the model parameters, which means that there might be something interesting about the model parameters.
14:24:27 So, here what he did was to vary. What he realized were the main features the model so anagram is the ratio of the number of species to the number of metabolites a sparsity s, the noise strength that tells you how much fluctuations you get in the nutrients
14:24:43 and then the correlation coefficient k for for nutrients and basically what's neat is that there are various features of the statistics that all seem to strongly depend on one of them, and not the others.
14:24:57 And so, in some sense, it's just a relatively simple fitting to try to identify where one should sit along each one of these parameters in order to generate all of a fitting to all of these to all the statistics.
14:25:11 And so, where this all where this all comes to is, you know, Michael your point is why, why can we not why can we get any of this by scrambling the data, and of course, that all comes back to you know how big is K.
14:25:27 Right. And so what I'm going to show you in a second is that for all the gut microbiota it says K is very close to one, meaning that, actually, it looks like the day to day correlations in the nutrients are are basically uncorrelated.
14:25:44 Whereas for other microbiota data sets. There, that's not true. And some of these statistics have nothing to do with the scrambling anyway right so I think what we'd like to argue is that it's actually a feature not a bug that you that you can scramble
14:25:58 it, because the models telling you about the daily fluctuations.
14:26:07 So, in that case, there's a. So, I thought that was a beautiful paper. The one thing I was confused about when it was reading it was kind of the setup, which you might think, just a choice but what he just said, Now, kind of makes it, I guess a relevant
14:26:18 question. So naively, I might have thought that to set up a model like that you have some model for how your environmental parameters walk around and restoring for some blah blah.
14:26:29 And then you have these passages that okay this is the environment. Now, you initialize you get acid. And now your environment changed you initialize passage, and instead you have this world that lives like that.
14:26:40 Yeah, but then for every point of the world you go and do passages until it's stable as then that's your measurement and then somehow you forget it all and repeat.
14:26:52 Yeah. Could you comment on that. Oh, I can comment on it. Um, so I mean, that's actually, that's actually a point that Ben and Paul you and I talked about quite a lot, because the details of that process are also important.
14:27:04 And, for instance, the degree to which you dilute become quite important. In this process, you know, because it basically tells you something about the degree to which you can reach a new a new state.
14:27:21 And so, I can. What I can tell you is that again it is, it is sensitive to that process. And so if you think of this as like a phenomenal logical model, then that's, I would argue good because it tells you you've fallen in something which you couldn't
14:27:35 get any other way. But from a biological standpoint, I agree that like it's it's it either is reflecting something we just totally don't understand, or it's telling us that, Like, what happened.
14:27:49 I mean, fundamentally, I don't know what happens in the gut, so if you wanted me to like model like for instance, if you, if you said Well, like I could say to someone disingenuously well no that was motivated by the fact that you eat three meals a day.
14:28:04 And so there is this like sort of, you know, but the food comes in, in a way that kind of is, you know, Well, yeah. Yeah, that's good we're measuring daily.
14:28:15 And so there is kind of these multiple ways of delusion and, and, but I don't necessarily think that's true when you said sensitive to the process. Do you mean that if you change that to something simpler, it will not, or if you do with a single dollar
14:28:38 so well I tell probably it's like the reviewers are gonna like this paper a lot better, and I'll feel more comfortable to, if you can get all of this, by just doing one delusion, and then you're done. And he basically can't make that work at all.
14:28:45 Yeah. So, I realized I am close to time so I will just say that, like, I'm going to go through these slides but they're all meant to make one very simple point, which is, you can reproduce a whole lot of stuff this is another gut microbiota.
14:28:56 Everything looks the same. In fact, I'll just highlight that like things like rank statistics, that's like one that doesn't have anything to do with scrambling are captured.
14:29:06 Very well.
14:29:08 The same thing can be captured in human saliva statistics, which, again, like, a lot of stuff is different here right the rank or statistics are different.
14:29:15 The width of this distribution is very different. The restoring slopes are different. And same thing in the vaginal microbiota said this is, in fact, what's interesting is there are many different diversity types of national microbiota.
14:29:30 One of the interesting aspects though is if you look at them, statistically, they're indistinguishable.
14:29:35 And so, I think, If you sort of collapse out all these different microbiota onto the space of all the parameters that we're measuring. Basically, the different microbiota do explore many parts of this space for instance sparsity is one that's particularly
14:29:50 interesting where some of the microbes have very low sparsity, and some of them have very high sparsity. And so for instance if you look at the human gut microbiome is very low sparsity me lots of things can eat many things, the saliva is the exact opposite.
14:30:05 And in particular, this noise correlation K, you'll notice it is very high. For many of the data sets which exactly addresses your point Michael, but that's not true, or some of the other data sets like the saliva, and saliva absolutely depends on whether
14:30:20 you scramble or not. For this reason, I mean, not for this food but it's consistent with this with this picture. So, I will.
14:30:30 I think I've settled these points. In conclusion, so I will just end by saying, so this is not just my group is will Luddington slab.
14:30:38 And Fred Shanks I have at our retreat, a couple years ago and sadly we won't be able to do a retreat last year but I really have a really fun group of people and if any of the students are interested in, in particular talk more.
14:30:51 Definitely also looking to get more people involved. Thanks for the attention great questions.
14:31:04 Five more minutes. If people want to keep going. can I ask about the elephant in the room.
14:31:10 Yeah, please Daniel what which elephant, he walked out evolution to remember is about Gordo has a ton of various experiments on human bacteria in mice which showed rather fast evolution.
14:31:26 now with such different conditions that that's irrelevant or, or, Absolutely not. So,
14:31:33 we do similar experiments as Isabel's which are with her experiments are beautiful and if you look with high resolution at barcoded strains. It's pretty clear in every organs we've looked at that, there is selection on the week long timescale for sure.
14:31:52 In Mondo colonized mice.
14:31:58 And Daniel I like in like three weeks will be done with the mouse experiment to really nail how this depends on the presence the rest of the community, I think, a question that we're trying to answer which I'd love to hear your input and anyone here is
14:32:14 you know there's like two pictures or okay so. So, to address evolution, the way we think about it now in a monetize case is that there are niches that each one of these bugs can learn to fill.
14:32:31 And they do, if you pass them in in vivo, and even a bug like be feta. It seems that within a week you know you you will see a new, new functions that that have evolved.
14:32:48 So, what does that mean in the context of a community, well you could argue that that doesn't happen in a community because there's someone already occupying that niche and so get shut down and you don't see any selection, that's reasonable, or you could
14:33:01 argue well, and then is the one who is is you know, they may see the light about about all these various aspects that, you know, the rest of community creates metabolic potential.
14:33:16 And so, I, so I don't know but I think we will know fairly soon because we've used some of these communities, to try and control the degree to which are barcoded strains will or will not have to compete in America.
14:33:33 How's that for a non answer Daniel was only a good start.
14:33:39 But I will look forward to chatting with with real data soon.
14:33:46 So I guess in the spirit of going back to, can we actually have a model community. If you take the community that is backfield. I think version two which was relatively complete pretend that is now the community that was in the second part of your talk,
14:34:01 which was the one that you treat with Cipro, and then put into all these media, do you see basically exactly what you predict should happen, which is the exact same strains are lost the exact same responses are curved growth rate predicts everything that
14:34:14 are not really any interactions, even in that context, I think it's a great set of experiments and one that we are we are doing.
14:34:27 The reason of course it happened this way, was because we were drawn to this from the mouse experiments and not the other way around, but your motivation is well taken.
14:34:30 I don't know.
14:34:32 But, but I think it's I think it's a great test, or, I would like to argue it's a great test. Of course I argue that it was a great test for paper, and some reviews agree with me and some didn't and presumably for good reason.
14:34:43 And like Terry's point is, I mean, I keep not answering his question, but it doesn't mean I don't appreciate it that, like, like exactly what we want to do it with these communities, is not totally clear to me but I mean that that's that's where we ended
14:34:59 up and that's, and we're doing that, too, but like what the next step is I'm not totally sure yet. And I guess maybe from my perspective, given the, from what I've heard and read from other people who study microbiome the expectation is that there are
14:35:12 all these strong interactions between players in the microbiome yeah was what use, at least I think you told partly into us is that they're not really that many strong interactions in pairs that a lot of sort of behave, I can tell you they're definitely
14:35:26 not lots of strong and. So, we have, we have now taken like large sets of these and done like every binary co culture, and if you if you quantitatively account for resource competition.
14:35:37 That's all there is for the not the faster but like, like 80% of the pairs are more, I guess what they're not suggest then that most people's microbiota is like buffered by their environmental load rather than some intrinsic dynamics that are occurring,
14:35:52 once they get colonized because presumably if there's not that many strong interactions, by chance if I lose a particular bug in my gut. There's no reason, it should be brought back up by its neighbors to reach a steady point versus just somewhere in
14:36:05 the environment I get receded, so it's more a question of like sort of like continuous receding that stabilizes you know rather than just intrinsic maintenance.
14:36:15 Okay, I like that interpretation, but let me, let me play devil's advocate to that, because maybe you can tell me something, intelligent about it. So, so what's, what does extinction look like in vivo and in vitro.
14:36:27 Well,
14:36:27 obviously, we sometimes think that things have gone away, but a perfect example of getting food is anchorman two yet, which is the Muse integrating microbe that I mentioned before, anchorman sia is undetectable from sequencing, in some light in a normal
14:36:48 diet. But if you switch your Paul soccer deficient diet booms to about 500%.
14:36:55 In Vitro those the this communities I showed you.
14:37:01 Within one passage, they seem to have lost documented.
14:37:04 So we actually took one of those communities after seven passages, so So, 50 generations of growth. and it didn't have Ackerman see after seven generations.
14:37:14 And we Adam Joosten acrobats he was president of 20%.
14:37:19 And so, uh, I mean I think there's probably a simple resolution to all of this, which is that, well, our level of detection isn't as low as we'd want, and in the case of in vivo any special nice will provide a reservoir for the future.
14:37:35 But I think it does highlight the fact that your question may like.
14:37:43 except it seems really, it seems like really really prevalent, sort of scenario, but it may be less prevalent than we think.
14:37:51 And you are worried about really big numbers in us. Right. I mean, yeah.
14:38:05 Alright. Okay, so I appreciate that you have your hundred and four strains that you bought from various sources and found all these amazing, amazing patterns with them, but do you think that if you had isolated strains that had a more recent evolutionary
14:38:25 history we've got another that you might see more of these strong interactions.
14:38:30 So, that was also another motivation for all the top down studies that, you know, if, if there were cautionary aspects that we wouldn't see them in the bottom up communities, I will just say that this isn't totally addressing your question, but this.
14:38:50 Okay, so, so we did ask if we make a bottom up community.
14:38:55 This is a simpler one that's motivated by a top down communities, either reconstituted from strains that we got from that host or from reference strains that we, you know, basically just ordered from from wherever.
14:39:09 Did they grow any differently. And the answer is yes, but they seem to grow differently. Only because the inner backer ACA actually has a different capacity for growth so that's the purple dots there.
14:39:21 That was the dominant member in the normal community, and then you get a reference stream, and it grows like crap. And that might just be one isolated incident, but actually we noticed that when we took a different human donors and we grew them all the
14:39:35 carbon sources, one of the carbon source that had a huge amount of variability was Mel BIOS.
14:39:41 And we found so what we did was we we isolated enter vector ACA from all from seven out of the eight, at least five of those are unique speech there's at least five unique and our vector ACA, out of all of those.
14:39:57 And those introductory ACA all grow differently with little bias, as you can see here on the left and
14:40:08 on the right, and they basically that difference in growth explains the difference in the community growth. And so I do think there's like, in certain scenarios there's massive impact or strain level changes is can you I can I attribute that to conclusion.
14:40:27 Know, but I thought I'd like to think so, I mean, I guess. Like, I kind of if you allow me to make the kind of crazy statement that, you know, all street level difference happened for a reason, and that reason is most likely recovered and then I would
14:40:42 say that that that this is like my best example.
14:40:47 I am totally okay with that. Yeah, so it is a great question. I would love to hear your input on what it would mean to have a experiment to try to answer that.
14:41:00 Yes.
14:41:01 Okay. Um, so I, I have some questions about the last part of your talk.
14:41:07 So you said you could was a very resource consuming resource model, and there was some regularization yeah and you will be able to capture statistical properties of time series.
14:41:18 And so the question is that if you see that different communities have different statistical properties of time series. Yeah, what do you can you infer back about biology.
14:41:27 Yeah, so yeah sorry I know I went by, like everything too fast but, um, so like here is basically, I would, I would call it our, our hypotheses about what this all means.
14:41:43 So, one of them is that like human gut microbiota are under more intense resource competition than than other than other microbiota is mouse gut microbe mouse gut microbes are under less resource competition.
14:42:00 So they're sparsity is higher.
14:42:04 I think there's I think there's a way one could imagine testing that the saliva seems to be like, like, totally metabolically distinct microbes, almost, and then the this range of noise strength is basically the part that was motivating us in at the beginning,
14:42:22 which is to say that like the environmental aspect. Tape putting aside the microbiota, the environmental aspect, which is what's being captured by this noise correlation and this noise strength is quite variable across the different, the different communities,
14:42:36 and there's also I mean, there's also some, some nuance things that we allude to in the paper that that I like, but I don't think we've like nailed, which is like for one thing we do this, we actually the one community where you can't get perfect fitting
14:42:53 of all the statistics is in the rise of sphere.
14:42:57 But there's also reason to believe in the rise of sphere that you're not at any sort of. There's other papers that have suggested that the rise of spirit is generally not a steady state.
14:43:07 And so unless of course like a strong assumption that law so I don't know.