 at the beginning of our session, Roy Kishoni from the Technion Israel Institute of Technology, who is going to talk to us about coexistence and stability in antibiotic-mediated microbial communities. Okay. It's on? Okay. Thanks, Daniel, and thanks for organizing the organizing committee. I actually noticed that you were wise enough to invite Kalina Sijian from previously from my lab, and in discussion with him I thought he would be a much more perfect person to tell you about this story, so I'm actually going to change the title on you guys. Okay. And instead of that title, I'm going to switch to this title, and I'm going to tell you about some studies that we do on trying to understand evolution of antibiotic resistance and evolution towards other stress that bacteria encounter both in the clinic and both in the lab and in the clinic. So we know that antibiotic resistance is going to the major public health concern. This is a quote from the World Health Organization saying a post-antibiotic era in which common infections and minor injuries can kill far from being an apocalyptic fantasy is instead a very real possibility for the 21st century. So with that motivation and other, we're really trying to look in detail how bacteria evolve resistance to antibiotics in laboratory setting, in the clinical setting, how they evolve resistance to other stresses that they encounter within our body. And then to think about possible ways by which we might be able to intervene, slow down, perhaps even reverse that process and drive evolution backwards in time. Most of my talk is going to be about this first part, and I'm just going to touch with a few slides at the end about possible ways to counteract resistance. Okay, so how do we study evolution of resistance in the lab? And I'm mainly going to focus on evolution through de novo mutations. There's, of course, another mode of evolution which is acquiring horizontally transfer-acquired resistant genes. But I'm going to focus this talk on acquiring resistance through de novo spontaneously appearing mutations in selection. So, typically, if I want to study that, I'm going to take bacteria, put them on a plate, resistant mutants are going to appear, and I'm going to pick up these resistant mutants and see what they are. So, this is great. It's great if you want to see the first adaptive step towards antibiotic resistance. But if you want to be able to see how bacteria keep accumulating more and more mutation to become more and more resistant, we need to keep ramping up the selection pressure, ramping up the antibiotic stress as the bacteria evolve. So, I want to show you two different devices, kind of, I like to call them benchtop evolution devices that allow us to do that. One in which we're going to increase antibiotic concentration in time. And the other one kind of in a well-mixed environment. And then the other device we're going to increase antibiotic concentration in space. And in both of these, we're going to let bacteria evolve high-level resistance by acquiring more and more mutations this time. Okay, so here is the first device. We call it the morbidostat. It's kind of a continuous-culture device, kind of like the chemostat or the tubidostat, if you think about it. But instead of the bacteria being limited by a limiting nutrient, instead, they are limited, the ghost rate is limited by drug inhibition. And the way that works, there's a computer that constantly monitors how fast the bacteria are going. If they're going too fast, it adds more drug. If they're going to slowly dilute the drug out. So the bacteria are always kind of partially inhibited. They go like that. At one point, resistant mutations arise. They take over the population. The protein starts to go faster. The computer sees that. Increase the drug concentration and inhibit again the population. And the process repeats itself. Numerations arise and so on. Okay, here's kind of a cartoon that kind of, I think, explain a much more vivid way what's actually going on. It's from a review by Natalie Balaban. It's essentially kind of a walk-a-mall. But every time they come, we kind of hit them harder. Ghost rate is constant, right? But not nutrient-limited, low concentration constant. Here's the implementation of the device. We have multiple tubes, like that. Each one fed with multiple sources with drug-free and drug-free media and media with antibiotics. And there's optical density meter on each one that's fed to a computer which decide what to do at any given moment. So when we take bacteria and put them into these devices, we see really a fairly dramatic increase in antibiotic resistance. We see some, in this particular case that I'm showing you, which is evolution to trimesoprine, diadropholic reductase inhibitor. We see some 1,000-fold increase in resistance in the MIC in the minimally inhibitory concentration. And we see that that increase does not happen through one big step, but rather through accumulation of multiple mutations. Each one provides a mild benefit, but together accumulating to a large effect. And you see those individual mutations accumulating here. So about four or five mutations accumulating. You can also see that process happens fairly fast through some two or three weeks of selection. We get this large increase of 1,000-fold. And you can see that every time we repeat the experiment, we get fairly similar adaptive paths. We never get a path that's kind of stuck in the middle. We reach about the same high level of antibiotic concentration. At that point, we are almost at the solubility limit of the drugs. We really just cannot add more drugs. Trimesoprine. Yeah. So diageopharynibitor. But we see similar patterns for many different drugs that we've tried. Yes, please. The population size is about 10 to the 9, I would say. About 10 to the 9. It's a large population size. You can see that we are walking here right at the limit of mutation limited. So for other drugs, here we are kind of almost like there are periods of stagnation waiting for mutations to appear. So we are right there at the population size. For other drugs, we don't quite see those steps. We are not mutation-limited. And I'm going to show you in a second. This is a very different target size for mutations that appear for different drugs. So that's the first device that I was talking about. And of course, what we want to do is sequence these along the way and see what are the trajectories. Genotypic trajectories that we see. But before that, I want to show you this second device that I've mentioned, which we call the mega plate. The microbial evolution goes arena. What it is, it's essentially a petri dish. But sized up to about the size of like a desk. And we pattern it with slabs of increasing amount of an antibiotic and simply put bacteria on the sides where there is no antibiotic and let them swim towards every increasing antibiotic gradient. Is there a way to dim off the light here, if anyone knows? The guard of light? No. Okay. Not on my side. Oh, it is. Okay, it's good. It's good. Thank you. And then I'll do all on afterwards. Okay, so here is a plate. And in this particular experiment, it's set up in a symmetric way. There's no drag on the side. And that consultation, again, this is a time mess-up in the same drag I've shown you earlier, is increasing towards the middle, again, increasing by a lot. This is just the amount of drag needed to kill the bacteria, 10 times more, 100 times more, 1,000 times more. And we're going to put bacteria here and there and let them swim. This is just condensation on the lid, so just ignore that. Okay, so I'm going to put the bacteria, let them swim. And I'm going to stop the movie here. So what happened? They grow in the area that is drag-free. They're very happy. The easy thing is they actually chemotact. So they get stuck on that line. They depleted the nutrient where there was no drag. And the cannophonic go to the area with that, but they get stuck on that line because every time a single bacterium go in, it gets killed or inhibited by the drag. Okay, so the only way to go in is through evolution. And if I play it, you'll see mutant appearing and giving rise to whole lineages like in here and on the other side, giving rise to whole lineages that penetrate. They get to the second step. Second mutations appear, these mutants are not capable of going. A 10-time fold, the concentration, the process repeat itself, a 100-fold and 1,000-fold increase in drag resistance. So the whole video is sped up. This experiment took about 10 or 11 days to run. There are actually two reasons why we take a very large plate. One is related to the question that was just asked earlier, is to increase the population size and therefore the probability of mutations. But the second is to avoid or at least minimize a lot the effect of diffusion, because the bacteria swim linearly with time and diffusion goes like square root of time. In large plates, diffusion doesn't do much. But yes, there is some little diffusion going on in the plate. Let me run it one more time. I personally enjoy seeing this movie, so I'm just going to enjoy it with you. And I think what happened with that video, besides being a very strong scientific tool for us to follow multiple trajectories to antibiotic resistance, it also, what it does, it takes concepts in evolution that are fairly vague in our mind, it's surely in the mind of the general public, concepts of mutations, selection, diversification, competition, clonal interference, and mainly, probably most importantly for the general public, antibiotic resistance and just how quickly it can happen, it takes all of these concepts and puts them in kind of a thing, is believing demonstration. The whole thing took 10 or 11 days. Everything I've shown you is 11 days. Time is linear in this movie. This video, we published it some half a year ago or something, it was viewed over 24 million times. Just to tell you that it's something I found it kind of very educating for me how visualization is, in fact, very important to communicating science and concepts that tend to be fairly vague and obscure. So when the experiment is done, we open up the lead and we pick up mutants, every time there's a funnel like that, and in one experiment like that, we can trace many different trajectories leading to high-level resistance. There are beautiful phenomena that happen here, such as, if you look here, there are mutants that penetrate and compensative mutations that appear. There are mutator alleles that increase a lot the frequency of mutation that also appear in some of these lineages and so on. And if we look across the trajectories, we can map in a fairly systematic way what is a collection of pathways that can lead to antibiotic resistance for many different drugs. This actually from the Mubidostad, but a similar picture appeared also in the mega plate. We see that for some drugs there is a fairly large target of alternative genes, an alternative pathway that can be mutated to allow resistance. And for other drugs, like thymosoprim, almost all the mutations appear in one particular gene, the target of the antibiotic, and they even tend to appear in exactly the same residue, exactly the same position in the gene. It is in these drugs that we are somewhat on the limit of mutational limit, and in these drugs you don't see those steps anymore. There's much more potential to go. For that particular drug, for thymosoprim, it's kind of interesting. You run evolution many different times and essentially you get almost exactly to the same solution. It's a genotypic level. It's a amino acid changes level. If you look at what actually happened, these tend to be all of these mutations in the active site of diatopholic redactase, the target of the antibiotic, and you see that each time we run the experiment, we get kind of partially overlapping final states of the evolutionary trajectory. Not only evolution gets to very similar solutions, it gets in a fairly similar order in which mutations are required. So here, if we follow the order of mutations by actually sampling from a population to multiple clones over time, this is in the mobility stat, you see that, for example, if you look at these two, the color stands for the different residues changes. They require the same mutation in exactly the same order. And overall, if you look, there's a strong signal for conservation of order. It's not a strict, but there is a tendency for if two mutations appear, let's say imitation A and B appear, in one replicate of the experiment, it's quite likely that I'm going to see these two mutations appearing in that same order in another replicate of the experiment. Okay, I'm actually going to remind myself to turn on the light here. Good. Okay, so what I've shown you so far is, first, those devices, benstop evolution devices in mobility stat and the mega plate that allow us to trace evolution multi-step adaptive path leading to high level of resistance. We see fairly dramatic increase in drug resistance in these machines. We see some 1,000-fold increase across two or three weeks of selection. And we see that for some drugs and not for others, there's some signal of reproducibility or visibility of conservation of the final state towards which evolution is going and even of the order of accumulation of mutations. So what I want to do next is to focus on this idea of parallel evolution if you wish. Same thing happen over and over. Again, not for all drugs, but for some drugs. And try to generalize from that. I would say that when you see a pattern like that, two things happen. One, it gives some sense of predictability, right? We know what will happen before it actually happens or we have some good guess, good idea of what might happen before it actually happens. So maybe we can prepare for it. Maybe we can kind of anticipate the evolutionary process. Again, not for all drugs, but for some drugs. And second, if you see a pattern like that, same mutations appearing over and over and over, you say, huh, these mutations must be very important. They must be the drivers of the evolutionary process. They are not appearing by chance. They are not hitchhiking mutations. They are not appearing by drift. These are the important mutations that are required for adaptation in these particular conditions. So what I want to do next is to take this very simple concept, an evolution equal or signifies adaptive mutations. It is a signal for adaptive mutations. I want to take that concept from the very simple environment of bacteria in the tube to the much more complex environment of bacteria going within us during infection. But the concept is very similar. We're going to ask what mutations appear when bacteria go and infect multiple people and which of these mutations appear repeatedly or which genes get mutated repeatedly in multiple people. And these would be candidates for the genes that are important for the bug within that particular environment. So to do that, we need to follow bacteria as they go and infect multiple people and we need to follow the same strain or the same clone as it goes and infect multiple people. And the way to go about it is to look for cases like that and that essentially happens all the time when a pathogen goes and spreads in the population. Every time we have an outbreak or an epidemic. We have the same or almost the same clone going and infecting multiple people and if it starts colonizing and forming a long-term infection we can follow the evolution of the bug of the same bug in different people. For some drugs mutations are very specific and for other drugs the target size is much larger. And I think with respect to if I try to pinpoint more towards what you're asking, for these drugs it's more likely that mutations providing resistance to one drug would also cause cross-resistance effect to another drug. Sequencing time. I would love to know the answer to this question. This is something we actually think about right now and right because I think the question of we actually know a lot about just taking bacteria, putting them on an antibiotic plate picking mutants and asking are they resistant or sensitive to other drugs? What I think we don't know is if you keep going whether those more specialized or more generic mutations tend to appear first or last. And I think we have an opportunity to do that with either one of those devices. I don't know the answer to this question but we are actually thinking about it quite deeply. Okay, so we looked for a case like that of a small outbreak and focused on an outbreak of a bug called Bocoldea dolosa. Fairly rare pathogen that infect primarily people with cystic fibrosis whose lung is much more... the mucus is much more thick and they are prone to infections by many bugs and sometimes actually not often they get infection with that bug Bocoldea dolosa that is really quite dangerous. It goes in effect from patient to patient. It's very hard to treat because long-term infection resistant to almost all non-antibiotics and often after several years unfortunately makes it from the lung to the bloodstream and cause sepsis and death. And because of the severity of this disease the hospital that we worked with the children hospital in Boston kept clones, isolates for multiple people that were infected across time, actually across several years from the lung in blue and from blood in red and all of that was sitting in the freezer and what we simply did was to take those clones from the minus 80 and sequence them. And first thing that you see is that mutations actually do appear to accumulate over time while with inpatient doing infection with inpatient this kind of a sense of a molecular clock some two clicks per year that get accumulated and that of course lead to the diversification of the strains within between different patients and we can actually use these mutations to then reveal the underlying phylogenetic tree that connects those multiple isolates that were collected from different patients. The patients are labeled here with light color on the back and here is the phylogeny. Okay, so now what we were asking we were asking can we find mutations or maybe specific genes that get mutated repeatedly in different patients and here is one such gene, GIRA. It's the target of one of the antibiotic used in treatment of these patients, Ciprofloxacin, and you can see that for that gene besides the wild type allele there are actually four other different alleles that are labeled here of amino acid representing different amino acid changes. And because we have a phylogeny we can actually count how many independent denoval mutational event give rise to that diversity. So all of these appeared from one but these orange one appears to one, two, three, four, five different denoval mutations actually appear in different patients repeatedly. And all together we can say, hey, that gene got mutated one, two, three, four, five, six, seven, eight, nine different time independently in different patients. So when I see something like that I'm going to say, wow, that's very strange, right? So there must be something important going on. There must be selection acting on that gene. Selection for what? Well, in that case I can quite easily guess it's probably selection for resistant to the antibiotic to Ciprofloxacin. And we can test that in the lab and that's exactly what happened for measure resistance. All of the one that got in any one of these mutations increased resistance actually by quite a bit by 10 or even 100 fold compared to the wild strains. If I want to generalize from here I basically just want to ask my analysis to ask my algorithm, give me all the genes that have such a pattern that are mutated repeatedly in multiple patients. And turns out that there are not too many some 17 genes like that and I want to show you the list but before that you might want to ask is it possible that these genes actually have higher signal, higher chance of mutation not because of selection but rather because of hotspot for mutations? And to tackle that question we looked at the rate of non-synonymous to synonymous substitutions in genes that got mutated multiple times the candidates or candidates for selection compared to all other genes and you see very strong in reach for non-synonymous changes in genes that get mutated repeatedly in different patients. So what are these genes? Here they are and when you look at it essentially you see the hallmark for pathogenicity you see antibiotic resistance like including the genes the gene I've shown you the GILA but other genes are presenting resistance to other antibiotics you see outer membrane changes in the outer membrane structure which is important for evading immune response and for resistance to adrophobic antibiotics you see genes related to secretion and one thing that was actually surprising to us but in fact appeal to be the most mutated across almost all patients is a two component system for oxygen sensing and when you think about actually in retrospect it makes a lot of sense. You have the pathogen coming suddenly to the lung low oxygen, very low oxygen and also fluctuating oxygen environment and it seems it's the main thing the bug is busy with is sensing oxygen and regulating gene expression based on oxygen concentration. If I try to summarize from here what that approach allows us to do is to really look from the eye point of view of the pathogen what are the main selection pressures that it sees that it encounters what are the main challenges that it needs to overcome during infection with inpatients and we can read that information just by looking at the genomes or in fact by looking at the diversity of genomes across multiple patients. But if you follow everything and the premise of how we've laid down the foundation for doing that it is based on comparing evolution in multiple infections in different patients. But what if I have a different scenario I have a single patient coming coming to the clinic and I want to identify I want some diagnostic tool that would tell me on that particular single patient what are the main challenges that the bacteria are facing can we still do something about it? And the answer is we might we might if evolution within patient doesn't look like that but rather look more like that if evolution within a single patient is actually based on mutations not taking over, not sweeping to fixation but rather leading to diversification and multiple lineages co-exist and co-evolving within any one single patient. That was the case then we can ask what mutations appear in that lineage what mutations appear in that lineage and are the mutations that appear repeatedly in two lineages within the same patient these would be the signals for positive adaptive evolution. So in some way we want to ask is evolution within the patient more similar to evolution in the morbidostat well-mixed device that I've shown you or is it more similar to evolution on the big plate, the mega plate? Excuse me? We cannot say for sure we kind of make that assumption in some way and it's consistent with what I'm going to show you but can I rule out that it actually got infected with multiple strains? I think if it got infected with multiple strains that are different enough typically I can recognize it if it got infected with multiple I wouldn't even call it strains but different clones of the same strains maybe it's going to be hard to reject. I'm going to try to touch on that in a second. Yes, but it's a very good point I think it's hard to reject the possibility of multiple clones slightly diverse infecting a person but I think the data is consistent perfectly consistent with that phenomenon not happening and let me just also add if that part did not happen within the patient but rather these two clones came in I can still ask the same question do these two lineages evolve in parallel? In some way, I care a lot about this question but in some way it doesn't it's not going to affect what I'm going to show you. What I'm citing there is actually a review that presented that question that's what I'm citing here but most of the studies on CFR with pseudomonas the studies I've shown you earlier was like I said with Bocoder Dolosa with Dolosa fairly rare and I'm going to show you one other pathogen in a second but still within Dolosa we went back to these patients and said what is the diversity within a single patient? How do we do that? Instead of the standard clinical practice of streaking and picking one colony we're going to streak and pick many colonies but essentially we take a sputum sample from the patient, spread it on a plate and we actually do one of two things we either pick individual colonies and hold genome sequences or we scrape the whole plate and deep sequence it to find the diverse locations and what do we see? So here's what we see this is one particular patient but similar things in different other patients you see that the population is actually very diverse these are different clones taking the same single time, same clinical sample very diverse in fact if I try to reconstruct the phylogeny of these clones from a single time point and because I have the molecular clock of the disease from what I've shown you earlier I can date it to about in this case nine years which is what we know about the time the patient got infected with that pathogen so again that's where I said it's consistent with infection with a single clone only very few mutations actually fully fixing in the population and since then the diversification and multiple lineages coexisting for ages coexisting that's all the mutations the phylogeny is based on all the mutations and the molecular clock is based on all the mutations okay so now given that that's a picture we can ask other mutations that appear repeatedly or genes that get mutated independently in different lineages within the same patient and the answer is yes and not only that the answer again is that these genes and not other are enriched for high ratio of non-synonymous to synonymous substitutions so the picture we kind of want to have in mind is it's kind of similar to the picture of the mega plate sometime in the past a stress appeal new antibiotic is being used because the population size is so large multiple different clones discover a way to overcome that stress both of them go in frequency but neither of them take over and diversity maintains itself for a long time what maintains the diversity well could be clonal interference could be that these are actually very similar could be some cross feeding could be diversification in space I'm going to touch on that a bit the real answer is we don't know but independently of the mechanism that keeps those lineages coexisting for such a long time independent of what is the mechanism just the case that is a phenomenon we can come today in a single clinical sample that we take actually contains information we would say the diversity in that single clinical sample contains information on past selection pressures acting on that population within that individual with that idea we can now go and start assigning genes under selection not at the population level like before but at the single patient level we see some of these genes like the GIA that I've shown you earlier are shared between many patients but some are patient specific and may represent stress that are unique to that patient maybe have to do with specific metabolic requirement with specific treatment with specific challenges of innate immunity ok so what did we do? we took the concept of a parallel evolution and we said what are the genes parallel at the whole population level between different patients and identified drivers of selection of the epidemic or the outbreak as a whole we then zoomed in and say given that as we found there are actually multiple lineages coexisting within a single patient can we use parallel evolution in these lineages to identify patient specific selection pressures now we can actually keep going deeper and ask within given organ or even given site in the lung are there specific adaptation are there kind of site specific selection pressure or site specific adaptation that allows the bacteria to overcome challenges that are unique to the very specific location in which they are proliferating how do we do that? so we were actually fortunate enough to have one of the patients agreeing to give us his expanded lung after a lung transplant surgery and we took that lung and did a many small biopsies cross-section biopsies, cross-section many biopsies like that in the lung now this is actually not with the bugs that we've shown you earlier with the LOSA but rather with stenotrophomonas maltophilia but you'll see actually very similar phenomena of the specification happening in that we collected multiple clones and looked for phenotype, antibiotic resistance and for genetic diversity what we see again very similar again this is a single patient now multiple location in the lung but again in a way single patient single time point and you see again huge diversity multiple very distinct lineages with distinct phenotypes of antibiotic resistance coexisting in that patient you also see mutations that again repeat between those different lineages the same mutation again representing stress that you would say are probably important for resistant to antibiotic and giving rise to the differential profile of antibiotics but now unlike in the previous case we actually have the distribution of those clones in the lung so first thing you want to ask is what happened there are two lineages here how are they distributed in the lung left lung, right lung or different location or the different lobes so here's the answer to that mostly not always but mostly every site that we looked at even the tiniest one millimeter biopsy actually contain diversity that almost like represents the entire diversity of the lung it's not separated in lobes it's not separated into the two lung there are some sites I can hear that are clonal for one of the lineages but mostly sites tend to be diverse and contain clones or presenting those different core existing lineages because that happened just like we did in the single patient level because we have diversification within a single patient we can do selection pressure acting within it's a patient level now because we have diversification within a single site we can do the same we can ask what are are there any genes that evolve in parallel in those different lineages yet are in parallel and independently in those different lineages yet are co-localized to the same sites can't hear you right, right I think it's a very good hypothesis that we have evidence for against so here's one such gene very interesting in matter resistance and you can see this is the entire collection of sites that we have 20 sites and you can see that almost in all sites that gene got mutated in the pink lineage it also got mutated in the green lineage two different lineages acquiring mutation in the same gene and in the same site so when we see something like that we say that's a signal for two things for adaptation selection but it is site specific representing overcoming challenges that are unique to that patient but also to specific locations within the body I think I'm almost like converging here we are done with a part of selection I see one more story that I'm actually going to skip but I've just put that slide here to say that diversity of the pathogen within a single sample not only allows us to say a lot about the selection pressures acting at population level patient level, single site level but also allows us to identify transmission network of who infected, who at the population level and at the single patient level and you can imagine looking at ancestry that's possible and it's really enabled by looking at not at single clones but rather the diversity of the population I think if I try to summarize what I've shown you is that the genome of pathogens taken from the clinic contains beautiful information if you know how to read it if you give me a genome from that from the Dolosa epidemic outbreak I can tell you what date it was isolated it kind of has a molecular clock stamp in it I can tell you which patient it came from I can tell you who infected that patient and I think more importantly I can tell you what are the selection pressures, the main challenges that the pathogen experience at the population level, at the single patient level and with inpatient at the single site level some of these have to do with resisting treatment, antibiotic, antimicrobial therapy and other challenges that the bugs see within our body and and we think moving forward that we can even use that information to say not only what the pathogen is capable of doing today but also to kind of foresee the future say something about anticipating what will happen next, what are the next genetic changes that we might anticipate if we were to use that antibiotic or another okay time is done so actually I'm going to end here I'm going to end here, not talk about that just think that many people involved I'm really very very fortunate to work with amazing people in my lab starting at Harvard and now we're moving to the Technion the part that I've shown you with the selection, the evolution devices if you wish is a work started with in the top part Adam Palmer on the morbidostat and the mega plate was really spearheaded by Michael Bame with help from Tommy Lieberman and now continuing in my lab in Technion with Autumn Gross and Edan Yelin and with inpatient evolution I'd say this is where you really need amazing collaborators at the clinic and I was really very very fortunate and very fortunate to work with Alexander McAdam, Greg Bebe the children hospital with Ted Coyne and Doug Wilson on a work that I've went through very quickly on tuberculosis and that work was spearheaded in the lab with Tommy Lieberman and and and now continuing with Olga and Edan Yelin ok thanks much for your attention thank you