 v tem seminaru. Zelo se prof. Cikara Furosawa z Universtju Tokio, in Riken v Nosaka. Zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, zelo se pravič, groj načianje in je No. Čutfpsu do�� fiskar. Zelo soki,�� колeko? Zelo soki. Koraj se prostoso zelo skleperer bej a prihčetima tristima. Zeloo se Sikara Furosawa z Universtvo Tokio in Riken. Zelo se profiti. Zdaj sem pričočila, da sem zelo v KUNI, in tukaj razdaj sem tukaj zelo v komputationalnih biologi, zelo vsega modela, zelo vsega replikacija modela, in da vsega vsega in biologičnih sistem in biologičnih sistem zelo vsega zelo vsega in biologičnih sistem in biologičnih sistem, or log-normal distribution or something like this. So this is the universal statistical laws in the replicating cells. And I also studied about some self-organized criticality in the replicating cell model. And also I studied that noise-driven adaptation and that differential dynamics triggered by cell-cell interaction and so on. And maybe that these topics are already explained by the queen's lecture. Yes, to be three. And the last one I will talk about. I hope so. So, but basically at that time that simultaneously I want to see that the real biological system see and analyze the real biological systems. And so after getting my PhD around 20 years ago, I started my own wet experiment, experimental studies. And now still continue that enjoying the both theoretical studies and experimental studies in my lab. And today, so because of the almost all that the theoretical study are already explained by Kuni, so today I would like to present my experimental works, especially that experimental evolution of the microbial cells to investigate that some nature of that adaptive evolution. And I afraid that some topic, because that today's topic is about the experiment, so that I afraid that some topics or terminology or concept are not so familiar with you. So if you have any questions, so please stop my presentation and catch the question. OK, so that the question to be addressed today is that the dimensionality of the cellular state changes in adaptation and evolution. So that of course that the empirical systems has the ability to adapt and to evolve to various environmental changes. And so for example, so this is the illustration of the equalized cells. And today, so when we cultured that these equalized cells in the some, for example, stress environment, some evolutionary dynamics occur that to overcome this stress environment. And the question to be addressed here is how we can describe that phenotypic changes during the evolution. So in other words, for example, how many freedoms are necessary to describe, and the phenotypic changes, how so we can extract some variables, so essential to describe that evolutionary dynamics. So this is the today's topic. And of course, even though that simple bacteria cells, like E. coli, consists of huge number of components, like protein and the metabolite so on, so this system has a intrinsically so very high-dimensional system. But when we consider that the complicated cellular functions, like cell replications, that the amount of such huge number of components need to be changed, so somehow very co-ordinate manner. And such correlated dynamics can reduce the number of effective degree of freedoms. So for example, so let us consider the very simple case in which the cellular state is represented by the expression level of the three proteins. So we can describe the state of some cells in the three-dimensional space. And its time development can be described as a trajectory in the three-dimensional state. And if the time development of the state in the evolution can be described like this kind of line, so this dynamics should be described in the three-dimensional dynamics. So degree of freedoms is necessary to be three. But if the expression level of the three proteins are somehow correlated like this, so we can set new axis on this line, so the state change can be described as a single degree of freedoms. So the question is that of course, that such correlated dynamics can reduce the number of effective degree of freedoms. And the question is how many degree of freedoms are necessary, for example, to describe that evolution dynamics of the equalized single bacterial cells. And of course, there is some theoretical background. So I guess that Kunier already explained by these theoretical studies, but very briefly, that by assuming that the steady growing state and possible phenotypic changes somehow consolent on the three-dimensional dynamics under this theoretical prediction such as protein expression changes for protein expression changing ratio are common for all proteins. This is a corresponding, which corresponds to the gross changes, that this is a theoretical prediction, and it agreed well with that experimental data. And we also studied how such that the low-dimensionality emerges and during evolution by using that very simple cell model, we infer that very simple catalytic networks serve as simple catalytic networks evoked and after the evolution, the possible phenotypic changes consolent on low-dimensional dynamics. This is a theoretical work. And with these backgrounds, so we tried to analyze that such low-dimensionality of the evolutionary dynamics in that by using that experimental evolution of that simple bacterial cells. And experimentally, so, okay, let me move to the wet part, wet experimental part, so that I use that the experimental evolution of the bacterial cells, and here I would like to present that methodology. And the experimental evolution is a very simple method. Basically that we culture that cells for a long time, for example, typically that with that constant time interval, so we take that some part of the cell to transfer to the fresh medium, so interacting this. And they proc, for example, that the propagation, so we could observe that evolution dynamics. I mean that when, so with that iteration of that expansion and the selection for the long time, so once some cells has a higher fitness, like, for example, that higher growth rate, that progeny of these cells increase the ratio in the whole population, and eventually take over the whole population. And so this kind, by using that this long-term cultivation of the cells, so we can observe the evolution dynamics in the wet lab. So this is a very simple experiment, so of course it's a bit tough work, but methodology itself is very simple. And this experimental evolution has a very several advantage in comparison with that ordinary method for to analyze that the evolution dynamics. So basically that the study of the evolution dynamics is based on the reconstruction of the past event, based on, for example, fossil data or genomic sequencing. And of course that such reconstruction of the past event of the evolution is suffered from that, for example, missing data or some missing links or something like this. But experimental evolution has many advantages that we can directly access that the initial state or final state or transient state of that evolution. So we can store that all samples during the evolution, so we can access these all samples and also so we can control the evolution environmental conditions. And also that by using the replica experimental evolution, so here that the replica experimental evolution means that starting from that identical initial condition and identical environmental condition, so we can maintain that the replica experimental evolution series. And by comparing the result of this replica experimental evolution so we can discriminate the fat data necessity changes, fat phenotypic changes or genetic changes that occurred unnecessarily or fat changes occurred by chance. So this is a big advantage to understand the fat data universal nature of that evolution dynamics. So answer, okay, so that here I would like to present some two examples for that experimental evolution of the bacterial studies. And one is that the very pioneering works by Richard Rensky for that long term cultivation, long term so experimental evolution of the E. coli for that 30 years. So he started in 1988, so that he started that long term cultivation of the E. coli cells. So basically that they transfer the E. coli cells for the 24 hours time interval. So here the working volume is around 100 milliliters that medium and that transfer and transfer the long term. And by you and for every one week so they store the frozen stock that each samples. And by analyzing these huge amount of samples so we can analyze how the fitness changes, how many mutations are fixed during this long term cultivation. So here in this figure that the X axis is a generation here in the R and R, 2,000 generation, this is a very long term, but maybe this is around 10 years or 15 years or something like this. But anyway that this green dot is at the increase of the fitness. So it early stages that the fitness increased rapidly, but the gradual increase that continue for a very long term and the blue dot is a number of the mutations and it increased almost linearly or something like this. And so this is the analysis of the changes of the genetic changes during this long term cultivation. In this case is that the X axis generation for the 60,000 generation. This is around the 30 years cultivation. And here that each figure show that the result of the replica series of this long term experimental evolution and why axis is a little frequency. This means that the sum, the ratio of the equalized cells having one mutation. And so you can see that the sum rapid increase of the allele frequency means that here that one mutation emerges and this cells with this mutations occupy that increase the ratio and that eventually that all cells has this mutation or something like this. So you can see that many times that some mutation arises and fixed or the whole population like this. And this is a very informative data to understand how the equalized cells adopted one environment. And there is many findings, unexpected findings. For example, in this region, so you can see that the sum mutation occupies around 80% and some others does not have this mutation by around 20% or something like this. So in this region that the two different, so state, state with two different states coexist. So very long time. So here I did around more than several years so they coexist. Even though that this experimental evolution started single genotype, but after that long time cultivation that there is two cells with two state emerges and coexist. And this coexistence is very interesting because if that there is only one carbon sources like glucose, so if some cells has a higher gross rate for this environment, they eventually occupy the whole population. But in this case is that coexistence can be maintained a very long time. And in this case are two phenotype emerges. One is that good at utilizing the glucose as a carbon source. And other is that some other phenotype that utilizing that byproduct of this major population. Here the byproduct is acetate. So that this major population that takes the glucose and some part of the carbon transferred acetate and this minor population to utilize this acetate as a carbon source or something like this. So actually the more complicated interaction might be around these two populations, but this is a very non-trivial. I expected to do that in this length-skid long-term cultivation. So the next example is that around the visualization of the antibiotic resistance evolution of the E. coli done by the lojkation is grouping and in this study, so they prepared a huge gel plate. It is more than 100 centimeter, so very big, so aga plate in which they prepared a step-wide increase of antibiotic concentration. Here that so 3,000 or 300 is a concentration of the antibiotic to limit of prim. And both end of this huge aga plate, so they inoculate E. coli cells and observe that how this E. coli cells spread on this huge aga plate. Here I would like to show that the movie of this study. And here this is a huge aga plate with the step-wide increase of the antibiotic concentration. And at the both end, so they inoculated E. coli cells and without antibiotic, they spread. And here is some step-wide increase of antibiotic concentration, so they cannot grow at first. But as you can see that, at some point, E. coli cells start to grow in the presence of the antibiotics. Here at several point, they acquire the resistance somehow and the spread on this aga plate, like this. And the top point reaches that the next step-wide increase of antibiotics, but some cells overcome this increase of the antibiotic concentration, antibiotics. And eventually that so they can overcome that the step-wide increase of the antibiotic concentration, like this. And as you can see that some patterns here, that this represents the evolution dynamics of the antibiotic resistance evolution here, around here. And so from this aga plate, so we can take the samples and analyze that genetic changes like sequencing analysis by which so we can draw this kind of the family tree. So that some one single cell takes the antibiotic resistance around here and the spread, that progenia spread around here and so on. So by this methodology that they can visualize that the resistance to evolution, like this. Okay, so that is clear. So if you have any questions about this kind of experiments, so please stop me. Okay, so, and so, okay, let's move to my own studies. So by using this kind of the experimental evolution of the bacterial cells. So, and to understand that the dimensionality of the evolutionary possible phenotypic changes in evolution, so we started that the systematic experimental evolution and that the various different environmental conditions. I mean that, okay, let us consider again that some state space, phase space in which that the axis correspond to that, for example, gene expression level. And that here is that the parent strain that this state space. And by performing the experimental evolution and that the different environmental conditions. And like for example, that the environmental condition is so that the state changes under this direction and the environmental condition be for this direction and so on. And by performing that this kind of the experimental evolution was that the enough large number of the different environmental condition, different selection pressure and quantifying that the phenotypic changes or genetic changes. So we expect that we can draw that how we can analyze the how that the state spreads in the high dimensional state space. And by this information, so we can understand that the structure or dimensionality of possible phenotype space of this equalizes. So this is the fact we want to do. And the question is how we can estimate the effectivity group of freedoms or how we can extract some macroscopic variables to describe that the possible phenotypic change of the equalize. So this is the fact we are trying to do. And so now, okay, so today so I would like to present at this kind of the ongoing works. And as I said that the method of the experimental evolution is a very simple that the culture in the equalizer for example, that a very long time in the given environment. But in this project, so we want to analyze that the evolution dynamics with a many different environmental condition. So this is a bit tough work for wet biologists. So to maintain that such a large number of the experimental diseases. So this is a very, very, very hard work. And such hard works can limit that the number of the possible environmental number of the possible environmental condition were number of replica experiment. And to overcome, so this limitation. So we developed some device as an automated system designed for the experimental evolution. Fitch consists of the liquid handling system connected to the shaker incubator. Here is the shaker incubator. And here is the micro-plate reader by which we can quantify that the self-concentration. And based on the result of the quantification of the self-concentration that some algorithms so we transferred the fresh medium and adding the stress chemicals and go back to that incubator. And here in this incubator so we can maintain around 40 micro-plates so byfetch that more than 10,000 in depended character series can be maintained in a fully automated manner. So what wet biologists need to do is just supplying the fresh medium or new chips or something like this and all other things were done by this automated system. And by using this system we started the experimental revolution under the various different environmental conditions. And as a first run so we performed the experimental revolution under the 95 different stress environment which included that many inhibitor of the cellular functions like cell role synthesis inhibitor or protein synthesis inhibitor or replication inhibitor so many compounds that inhibited these cellular functions in the different action mechanisms. And also these stress involved that the surfaxant or creator metal or some organic gaseats or others included the many compounds that surprised that the cellular growth of the E. coli. So by using these many stresses with different mechanisms so we tried to investigate that how E. coli cells can overcome these stresses. And this slide showed that the method of the experimental revolution. And the cells were cultured in the synthetic medium so here synthetic medium is a very pure medium. It's all component that are well defined like carbon sources or powerful somaticities. But anyway that we cultured that E. coli cells in the synthetic medium with the concentration gradient of the one stressor. And in low concentration range of the stressor E. coli cells can grow. And in high concentration ranges, they cannot grow. And every 24 hours from where with the highest concentration of the stressor at which cells were able to grow, cells were transferred to the fresh medium with the concentration gradient. And by iterating this daily propagation, so we expected that this concentration that with the highest concentration for the cell growth is increased. This means that E. coli cells acquire the resistance of the corresponding stresses. And to quantify that the resistance we use the minimally inhibitory concentration and the concentration of the stressor around here, okay. And so this is a result of that one example of the result using that the cell for stressor. This is a kind of the antibiotics that killed the E. coli cells. But anyway, that here that the X axis is a time so that we perform that 27 days daily propagation, which roughly correspond to the 300 generations of E. coli. And Y axis is that the log-transformed minimally inhibitory concentration of this surface stressor. And to check that the reproducibility of the evolution dynamics, so we maintain the six independent character series starting from the same E. coli cells. And we draw that the six line aren't here. And as you can see that the MYC gradually increase like this. This means that the E. coli cells acquire the resistance to the corresponding to the surface of the stressor in this case. And so we performed this kind of that the serial transfer experiment and that addition of many different types of the stressor like this. This is only the part of the result of this systematic experiment of the E. coli cells. And as you can see that for some stress, let's see that the MYC gradually increase like this. And for some stress that they increase stepwise manner like this. Maybe that in this case that the E. coli cells has a single mutation stochastically, but in this case that by in this case is a many complex philosophical changes of the surface like this. But anyway, so among the 95 stresses we inspected, we obtain that we can observe that the senior count increase of the MYC for that 87 stressors. And among them, so we selected that the 48 stresses and the four independent evolved line and 192 evolved lines in total for further detailed analysis. So to analyze that the phenotypic change or genetic changes. Okay, so first to analyze the phenotypic changes. So first we quantify how the acquisition of the resistance to the one stresses changes the resistance or sensitivity to other stresses. For example, in the case of the chloramphenic resistance strain, so in this case is after the 27 days, days, daily propagation that we obtained that the four chloramphenic resistance strains and this chloramphenic resistance strains show that the significant increase of the MYC to the acrylic flavin to different type of stresses. Such increase of the different type of stresses is so called the closed resistance. Ansah, we quantify that the MYC level for that the many stresses indicated and investigated 47 stresses. So among that the 47 stresses, we found that for 15 stresses that this chloramphenic resistance strain showed a significant increase of the MYC. This means that the closed resistance can be observed for that 15 stresses. But for some other stresses, this chloramphenic resistance strains became more sensitive to the some stresses like this. Here that the MYC, so significant decrease for the some stresses, which is called the collateral sensitivity. In this case, so chloramphenic resistance strains showed that the collateral sensitivity for nine stresses out of the 47 stresses. Ansah, we quantify that the changeable resistance that all possible pairwise combination of these 48 stresses fit is more than that 2,000 combinations. And so this is a bit tough work, but by using that automated system, we can quantify this kind of that systematic quantification of the closed resistance, the collateral sensitivity and found that around 25% of the combination, pairwise combinations, they showed significant changes of that resistance level, like a closed resistance or collateral sensitivity. So this results suggests that equalized cells cannot change the resistance levels for that one stresses independently of the other stresses. Instead, the mechanism of the resistance equation or whether that became sensitivity is somehow very tightly interconnected, which constrained that possible phenotypic changes were equal in this case. So this results, so suggest that somehow possible phenotypic changes are constrained on the some, some low dimensional patterns. Okay. And so this figure showed that the position, that the mutation identified that this is 192 resistance strains. And here that the x-axis is at the genome position. Equalized cells has around 4 million base pair genome. And the y-axis is just index of the evolved strains. The first four is at the chrono-phenicol, evolved strains, second four is at the refunction, evolved strains and so on. And red dot represents some position of the mutation. And for some stresses, the number of mutations is much larger than others. So this is that this stress cell, it can be acted as some compound causing that some mutations that are so called the mutagen. But anyway, we found that the number of fixed mutations of these evolved strains are relatively small that more than 80% of resistance strains have less than five mutations. So that this is the distribution of the number of mutations for these around 20 evolved strains. But anyway, that the number of the mutations is almost always at the less than five. So a very small number of the mutations identified. And so among them, so among these mutations, we selected that 64 mutations feature commonly fixed to different evolved strains, that to investigate the effect of these mutation. And the data is very much that, but we found that many mutations that affects the resistance level changes. Like, for example, in this case is that, we introduced that mutation at the RS genes which is involved some stress response machinery. But anyway, that by introducing the mutation that this RSSB genes into the parent strains, that the MRC level to the various different stress. Here is that the x axis is the name of the stressor. And the y axis is the changes of the resistance level. So relative MRC to the parent strain. But just introducing the single mutation that is for some stresses that MRC increase, this mean that the resistance acquisition were observed and for some stresses that this mutant, mutant became more sensitive to some stresses. So we obtained this kind of data for the many mutations, but data is too much to present here, but anyway, that this kind of the many mutation observed. So for the days around 200 resistant strains, 200 resistant strains, 40 environment and for individual evolved strains. So we quantify that the change of the resistance are sensitivity to many stresses and that we analyze that the genetic changes for... And also we quantify that how that the gene expression changes in comparison with the parent strains. And by using these high-dimensional data... Excuse me. So you have four evolved strains for the same environment. So do they have some common mutation or they are often different? Yes, so in some cases, so it is a bit difficult, but in some cases that the common mutation were found in the same, so evolved strains at the same environment. But for others that many different mutations cause that similar phenotypic changes. So here that the genetic phenotypic mapping is very complicated. OK. So basically that several mutations are common, but many other mutations exist. OK. Variety of mutations are observed. Any other questions? At this moment? No. OK. OK, so... So in this... OK, so in this case is that... I have a question. Yes, so the way to analyze to extract your data is that you determine for each stress what is the minimum concentration at which there is resist... I mean, the minimum concentration at which there is no replication anymore, there is no evolution anymore, right? You have for each stress there is a different... Exactly, that's my question. The minimum inhibitor concentration you have to determine for each type of stress. Is that correct? Yes. And this is what you analyze. So every day, so we quantify that cell concentration with different drug concentration, right? And so by this data, so we can determine that the minimum inhibitor concentration and below the minimum inhibitor concentration we take the cells to transfer the fleshy material. And from this sample you do the genomic sequencing, right? Yeah, so after that 300 generations, so we take the single clones at the end point of the experimental evolution and apply these samples to the genomic analysis. OK, thank you. So in this case there is no heterogeneity. So we analyze the heterogeneity of the genotype or phenotype at the end point, but in this case the selection pressure is very strong so that not so high heterogeneity were observed. OK, so thank you for your question. So we analyze that from the clones that isolated at the end point of the experimental evolution around 200 clones. So we quantify that the expression changes and that the quantify that the change of the resistance to the various different stressor and that the change of the genomic sequence and analyze that the relationship between that these phenotype, phenotypic and genetic data. And so we have tried many methods to analyze these phenotype-genotype mappings, but basically as a question about the genomic correspondence between the genetic changes that the phenotypic changes is not so clear, I mean that the many different mutations cause similar phenotypic changes. But these two data that the resistance profile and the transcript data is very well correlated. So by using this data so then we analyze that some constraint on the possible phenotypic changes. And for this purpose, so we tried to predict that the change of the resistance level. So we analyze the feather that this change of the resistance level to the 48 environment by this environment can be predicted just based on the changes of the gene expression level. So here that the objective variables is that the resistance MYC for the 48 environment for the 292 strains and that the explanatory variables is that the expression level of the 1000 genes. And we check the feather that this change of the resistance can be predicted based on that the expression level of the 1000 genes. And for this purpose, so we use a very simple method, so-called partial list square regression, PLSR. So in this case, in this method, so both the high dimensional expression matrix and expression data and the resistance data are projected onto the small number of the component with some loading. So this is a linear, basically linear algebra. So these high dimensional data are projected on the small number of component with some loadings. And in PCE, a principal component analysis, the PCE are determined by that to maximize the variance on this component. But in PLS, the loadings are determined by the maximizing the covariance between that these products component. And by feature that we can select is that the good axis to represent is a relationship between the two these to a high dimensional data. And then that after that this predictions, so we use that linear regression to predict that changes of the resistance level based on that the expression level. And here the question is that how many components are necessary to predict that change of the resistance level and to check that this we use that so-called cross validation that the data are split onto that the test data, training data and the test data and the parameter that fit by the training data and the prediction accuracy evaluated by test data. And in this figure that the X axis and number of component, number of component, number of axis is used for the prediction and Y axis is a prediction error. And as you can see that the prediction error became minimum around nine or 10. So this mean that this is the best for this degree freedom, the best for the prediction. And this figure showed that the prediction accuracy here that the X axis is observed resistance, here that the observed resistance is quantified but log transform and MIC ratio to the parent strain. Here that the positive value is that the resistance acquisition and the negative values became more sensible due to the collateral sensitivity. And the Y axis that predicts the resistance. And as you can see that the prediction accuracy is aware. So even if we use just nine components, so very high dimensional, so expression data somehow projected only that the nine or 10 axis is component and by feature that we can predict the changes of the resistance level to the various different stress. So this mean that the possible expression changes somehow constant on the low dimensional space. So by feature that we can predict the behavior, change the behavior for the stress resistance. So this results also suggested that possible changes are somehow constant on low dimensional data like a nine or 10 degree freedom. And the next question is that how we can interpret such axis components for such low dimensional dynamics. And this figure showed that expression profiles projected onto that first three component of the PLS analysis and each dot represented the state of the evolved strength. And the question is how we can interpret these axis. And for this purpose, so we performed so called enrichment analysis, I mean that the fat gene function significantly enlist on the high loading genes for each component. And as a result of this enrichment analysis, so we found that for the component ones, the first component that the genes are related to the stress response has a strong significant contribution to that component one. This mean that this axis represents the how much stress response machinery of E. coli are somehow activated. So this is a stress response axis. And in the same way that the component, in the component two that the aerobic genes are related to the aerobic respiration and the membrane transport are significantly enlist. So this suggests that this component two are somehow related to the growth activity because that this aerobic respiration is very important for the cellular growth and the membrane transport is also very important. And in fact that the score on this component two correlated well with the growth rate of these evolved strains. So this result supports that this component two is that represents the growth activity. So this is a growth active, growth axis. And in the same way that the component three are that the related amino acid biosynthesis, especially that the branch to change amino acid, there of course amino acid is somehow very important for the cellular growth, but at the same time that amino acid is a production amino acid is also very important for the stress protectant. So that third axis is amino acid biosynthetic activity. And other axis also can be interpreted by, for example, so this axis corresponds to an aerobic respiration where this axis corresponds to glutamic acid, the synthesis is something like this. So we found that the possible expression changed somehow constant on the low dimensional space and I infer that each axis can be interpreted related to the cellular function. So we named the major somehow, so major axises, but anyway that the, and evolution dynamics somehow constant on this, these major axises related to the cellular functions like stress response or growth or something like this. So we can describe that the evolution dynamics to the various different environmental condition by using these small number of the axises. So this is what our systematic experimental evolution studies suggest it, okay. And so in another, in other words, so this, our result indicated stats that the two high dimensional phenotype space, so one is that the expression space in feature that each axis represent is that the expression level of the different genes are having several thousand dimensions and in the resistance space in feature that each axis corresponds to the resistance that the different stresses in our analysis around 100 or something like this. But anyway, so these two high dimensional state space can be connected through the low dimensional feature space rather than having that nine or 10 degrees of freedoms, something like this. So this is that our experimental evolution that indicates. And for the next step, for the next step, so we started, we try to analyze that the trajectory of the evolutionary dynamics. So in the previous analysis, so we isolated the clones at the end point, at the end point of the 300th generation of experimental evolution, but actually that evolution, within the evolutionary dynamics, where phenotype is gradually changes, this can be described by the trajectory in the phenotype space. But to analyze such trajectory in the expression space, this is possible, so for example, by taking the messenger RNA samples for the everyday and by which we can draw the trajectory on this high dimensional expression space. This is possible, but it's a bit too hard work, too tough work for the experiment. So, but instead of the trajectory in the expression space, so we can analyze the trajectory in the resistance space. This is relatively easy by using the automated system we developed. For example, let us consider the cases and then the in-feature stress one is used for the selection stresses. So equalized, selected by the resistance to the stress one, but at the same time, that resistance of the stress two to stress N can be so quantified. So MIC level is quantified. And by after the selection of the stress one that the resistance of the stress one gradually increase, but at the same time, due to the close resistance or collateral sensitivity, that the resistance of the other stresses will changes. So by fit, we can reconstruct it, we can reconstruct it trajectory of the evolution dynamics in the resistance space, high dimensional resistance space. Okay, and so to test this analysis of the evolution trajectory, so first we use the very simple cases in fit that we use a two dimensional resistance space in fit that the X and Y axis is the represent the minimum inventory concentration to the chloramphenicol and amigasin. And our previous studies showed that these two antibiotics, where both of the features are protein syncyl inhibitor, and these two antibiotics show that the collateral sensitivity to each other, I mean that when the cell that creates the chloramphenicol resistance, it became sensitive to the amigasin and vice versa. These are the tradeoff between the resistance equation. And here is that some examples of trajectory on the two dimensional resistance space. And this result is showed that the selection by amigasin and this blue dot is initial point and we overwrite the four trajectory on the selection by amigasin for 30 days. And because we selected cell by amigasin resistance, so that the MYC to amigasin gradually increase, but at the same time due to the collateral sensitivity that the resistance to the chloramphenicol gradually increase like this. And in contrast, when we selected the cell by the chloramphenicol, that we put that cell to the upper side and the chloramphenicol resistance increase like this, but at the same time, that the amigasin resistance decrease like this. And by merging that these two trajectories, so we can see that some clear tradeoff line between the amigasin and the chloramphenicol resistance like this. And the questions here is that whether that this phenotype, the possible phenotype is a constraint on this tradeoff line or as a direction diversion is possible. So to check that this possibility, so now we are trying to develop the method to control the evolutionary trajectory towards the desired target like this. So let us consider that the high dimension, yes? So in this case, so collateral sensitivity to each other, that means if it is strong to A, it's also strong to B. And if it's evolved to strong to B, A, then it's also strong to B. And if it evolved to strong to B, it's also strong to A. So it's not to each other, it's not like that, it's more, if you try to evolve collateral to strong to CP, then it's weaker to AMK, okay? Yeah, yeah. So when we put the selection pressure to the strong to CP resistance, that equalized cells decrease the resistance to AMK. Okay, so each other means that if it's strong to CP, it's weaker to AMK. And if it's stronger to AMK, it's weaker to CP, I think. Okay. So here is a tradeoff like this line. So if that equalized cells increase the AMK resistance, they decrease the Kranvenko resistance like this. And when we put the upper side, that they decrease the Amikashin resistance like this, okay? Okay, so and, okay. And the question is that whether we can overcome or this tradeoff, we can control that, we can realize that evolution trajectory towards a different direction. And to check this possibility, so now we are developing the methodology to control that evolution trajectory towards the target state. Here that let me consider that some resistance space and here we put the target state and the current state here is a problem that how we can design that the selection pressure towards the target. So by changing the drug concentration independently. And we realize that this kind of the regulation by using the very simple feedback regulation. But anyway, so we can design that the selection pressure towards the target by combination of the concentration gradient of the drug sees multiple drug sees. And here is that the example of the such controlling the evolution target state. Here in this two dimensional resistance space of the Amikashin and Kranvenko that we put that the target state here and we can see that the trajectory towards target like this, okay? And in this figure so we put the target state right here and in this case is that evolution trajectory towards the target seems bit difficult but anyway that some slide changes towards the target were observed. And this direction is different from that the single drug condition like this. This is just a little fly. So we can design that the evolutionary trajectory towards the target now. And this is an example of the two dimensional resistance space but now we are trying to expand so this methodology to more high dimensional resistance space. For example, this is an example of the three dimensional resistance space and the controlling the evolution trajectory towards the target. Here that the initial state is around here and the target state around here. So that this little dot is a target and that we put that eight trajectory of the eight replica lines. As you can see that some trajectories are overshoot it but anyway that we can see that some trajectory towards the target. So now so we can put that many different targets and many different trajectory can be observed. Now, so in currently that we are try expanding that this method for the eight dimensional resistance space and we achieved that the sum controlling the phenotype by using this kind of methodology. And in near future so we will try to investigate that effect of the changing the target structure to quantify that the availability. Here I availability mean that okay, let us consider that the two dimensional resistance space and we can prepare that the many different initial state and for each initial state we can design that evolutionary trajectory to different direction and for some direction evolution is easy or for some direction the evolution is easy and for some direction it's evolution difficult. So some that the difficulty of the evolution can be quantified by this methodology that control the evolutionary target. So that one good point for that this method is that we can set that the direction of the selection pressure by ourselves. So by feature we can design that this kind of that experiment and by based on this data so we might be able to estimate the random scape so feature direction, feature direction the evolution is easy or the feature direction is difficult or something like this. And this random scape so at the glance it looks like a fitness random scape but it is not fitness random scape instead. So this is a random scape representing that availability. And so actually that for some directions the evolution is possible some direction is difficult and so as maybe as Kuni explained that the possible evolutionary trajectory might be constant on low dimension dynamics this means that for some direction that evolution is easy and so for some direction for example also another direction that the evolution might be difficult and by using that this kind of methodology we expect is that such consulent can be accessed directly, quantified directly. So this is our ongoing project to understand that the low dimensionality of the evolutionary dynamics. Okay, so maybe time is over I guess. So, okay let me summarize my presentation so we set and we develop the term automated system for the high-slip experimental evolution and by feature we can perform that we can analyze that phenotypic and genotypic changes for the values different environmental conditions and by analyzing the data so we can see that some major axis of the phenotypic changes in feature the possible phenotypic changes are constrained and by using that this kind of methodology is that we can make the sum more so quantitative theory so adaptation of the evolution by macroscopic variable and so it might it will so make that the possible to the prediction and control of the evolution. And so this is acknowledgement so many level members contribute to this project and thank you for your attention. Thank you very much. Beautiful talk. Thank you, you have a question regarding this plot of the observed resistance or the predicted resistance versus the observed resistance. You had this plot. I don't know if you can show it again. You had these red dots experiment after the analysis of the minimum of the maximum correlation. There was this predict. Exactly this thing. So, yeah, this in the right plot, in the right plot. So when the observed resistance so I understand that here there are two kinds there are two sets of stresses. Those that improve resistance and those that worsen resistance. Is that correct? I mean, when the observed resistance is the logarithm of the ratio of the minimum inhibitory concentration between the mutant, the both mutant and the pattern. So in principle, if this concentration decreases of the mutant relative to the pattern, the logarithms is negative means that the fitness has decreased. I mean, how this relates to the fitness, the fact that this observed resistance is positive or negative? Yes, so, of course, after the experimental evolution and that the resistance will increase. Okay, and under that drug sees used for the selection that the resistance is generally increase, always increase in my samples. But at the same time, so for some stresses that the resistance will decrease for that, for one, okay, so let me take that some resistance of strength for the stress A. So, of course that the MYC for the stress A will increase, this means that the positive value around here, but for same strength that for other stresses it can decrease for other stresses by that acquisition of the resistance A. So, here it's a many negative values. So, we put that all data coming from that single evolved strength. So, for each strength, we put that 48 dot days, which can have that both positive and negative values. Is it the answer for your question? Yeah, yeah, so my question also was, is there a direct relation between these measurements at this stage that you are showing here in this slide? Is there a direct inference we can make or about the fitness, can we say something, can we quantify fitness in a way from these measurements of resistance, is there a direct relation between resistance and fitness or fitness is a different quantity? I see, okay. So, this depends on the fitness, this depends on the definition of the fitness in this case. And I use, of course that in my experiments that the fitness is that the concentration of the MYC, the minimally heated concentration corresponds to the fitness because we take that only as cells close to the MYC. So, that the fitness corresponds to the MYC in this case. But at the same time, the different kinds of fitness can be, for example, just for example, the growth rate without adding that drug sis. And such growth rate is generally decreased by the acquisition of the resistance drug sis. So, I mean that the fitness depends on the definition. So, fat fitness, so now you are asking. I don't know. I'm asking. The question is precisely that, what is the fitness here? Yeah, in this case fitness is the concentration of the drug sis, that minimally heated concentration drug sis. This is a fitness. But other fitness can be possible for the other environment. Okay, I see, I see. Okay, so you are not measuring the fitness, let's put this way. So, now that you mean that the fitness, if that the fitness is a minimally heated concentration, so we can measure that that is how that the cell concentration depend on that the drug concentration. So, higher concentration ranges that cells cannot grow, that the low concentration ranges, low drug concentration ranges cells can grow and high drug concentration ranges cells cannot grow. And there is some boundary. So, we can define that the minimally heated concentration around here. So, this is a way to determine that the minimally heated concentration and here we use this value as a fitness. Okay, thank you, thank you. Can I ask a question? Maybe not a question, but just to make sure that I understand the key message of your talk. It's very interesting talk, by the way. So, the message of the talk is that control and prediction are possible not because of the existence of some fitness, well-defined fitness, but it's rather because of the low dimensionality in the genotype mapping, right? Yeah, yeah. This slide, I mean that slide. Because in the feature space, so the feature space has small number of components, so now in the case that you show, because of that, the feature space is the way you map between the genotype and phenotype and since that feature space is low dimensional, then it is possible to make prediction and control. If not in the case, then no way to make the prediction and control, right? Yeah, yeah, yeah, in that way. So, what I want to present in this study is that predict and because there is a low dimensionality between the phenotype, so it makes us to the, it makes us possible to predict the behavior. If that this relationship is really high dimensional, it is almost impossible to predict the behavior, right? But at the same time the correspondence between the genotype and phenotype is much so complicated in my data, so that the relationship between the genotype and phenotype might be more complicated and more high dimensional, but phenotype is somehow constrained on low dimensional space, so by feature that we can, so phenotype can be predicted, genetic change is difficult, rather difficult to be predicted. This is what our data saying. Thank you, thank you. Maybe. Ah, actually I have many questions. Okay, so referring to Felix' questions, so maybe one confusing point is that that plot of this prediction, error plot, that is obtained for, that plot is given for a single strain, for given strain. And across many different environmental conditions. Yes. Ah, so this is that this plot is obtained that all 192 evolved strains for each stresses, so that the number of data points is very huge. So actually you have many, many strains and many, many environmental conditions and you can plot all of this and then, yeah, then this is well predicted. Yeah. And that means so later you show that if this is strong to chromophonical and this is strong to MK or something like that. So that is also predicted from this nine dimensional model. Yeah. Yeah, so this change, so at least in my, at least end point, so this changes is, this relationship also can be predicted that, for example, this point, this point can be predicted by the, just expression changes. So if we have the data on this, that expression changes on this trajectory, so we can predict how the MRC changes. So basically this kind of trade off structure in this case is given by this nine dimensional as a result of this nine dimensional dynamical system. Yeah. And then I often ask about this on sagare sprosity works or not. And so if this is strong to A, evolve to strong to A, it's also strong to B. So in this case, somehow it works. If evolve to A, then it's negative to B and B evolve to B, it's negative to A. And this is true to many other cases. So, or if it is strong to A, it's also strong to B, B evolve to strong to B, also strong to A. So either plus plus or either plus minus minus plus. That's always. Yeah, plus, of course, plus plus exist and plus zero exist. Yeah. So, yeah, relationship, of course, we can see that this kind of minus relationship in some cases, but not always. So how common is this the case that on sagare sprosity works? So that means this dynamics is somehow, how kind of represented by nine dimensional potential structure or something like that, isn't it? Yeah, yeah. Yes, so actually it, how to say? So it's difficult to say that some, how many fraction that the on sagare sprosity is, but it's, but it's difficult to answer. Of course, I can say that this completely the correlation matrix as well, but now, so what I can say is that for some cases that this kind of relationship exist and for other cases not, it's difficult. But that should be predicted from this nine dimensional dynamical system. So if you have some kind of cyclic flow around there, maybe some, yeah, kind of that violates on sagare sprosity. Yeah, cyclic flow, yeah, yeah, okay, so, yeah, I will try again. And another, so prediction in this case, so you, so try to evolve this, but before evolution, so in this direction of this trade-off direction, so trade-off direction, so before evolution, if you take this kind of some strain and then just by mutation, I don't know, just by noise, in that direction it's more easily to change. So it's, in that sense, so some kind of relationship to above-ability direction and the change-ability direction in noise, that is correlated? Yeah, actually that, for that expression noise, so that the data is a bit difficult to take, but so we have many data for that environmental response of the equalizers. I mean that the transcriptome changes or transcriptome change of the equalizers obtained under many different environment is available, and we can see that the clear correlation between the direction of the easy, the direction between the phenotypi, how easy to the phenotypic environmental response and how easy to the evolutionary response. There is a clear correlation for that variance, for that variance between the environmental response and the evolutionary response. And also it can correspond well with that, for example, nine major axes in the semester case. So in the next part you try to make this to evolve to the cells that are both strong, both to A and B, so called C. And of course that is more difficult. But can you quantify such kind of difficulty in some way of this nature of these dynamics? Yes, this is what we will try to next. For example, this direction might be difficult and this direction is easy or something like this. So this is what we will try to next. By taking the data from many different points in this resistance space and smudging these data so we can see that this kind of the landscape. So for this direction that evolution is difficult and this direction is easy or something like this. So but if you know that you can make a very dangerous bacteria that is strong to any biotics or something. Yeah, but that might be interesting but of course it is bit dangerous. Yeah, actually this trade is very... I feel that this trade is very strong. Actually we know that molecular mechanism how this trade emerges. This is just based on some metabolic shift that depends on the proton transport through the membrane. But I bit surprised that some cells can overcome this trade off. This is very interesting and we will try to understand how such difficult evolution is realized by the sun. Genetic change or genetic changes. So that might be make some dangerous stories. I'm not sure. Ok, so any other questions? Ok, so thank you very much. Ok, thank you for your attention. Kešpo.