 Tya ne zelo, že smo v certains nekaj bilovali, na zelo donovanju, to nekaj nekaj ne, samo ponreadnjo, ki so bomo, dve raztej, tamto pro Professionalismi, tako, da je nreset, čkori bilo, so so bojni, ki so bomo, kaj smo, pri diamondsj, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, nekaj ne, Tukaj, vznikam, da smo vsega vsega vsega. In z 5 rovnih različenih, smo začali, da smo se zelo zelo. Zelo je to, da je to, da je teori. Zelo smo počkali počki. Tukaj je to, da je vsega vsega vsega. Zelo smo počkali v Ludo Vico, ki je tukaj, inkl adding with Luca, who is also here. Ludovico will give a talk about this. Here the story... I will not tell you about the story. So you can see the talk. The story is that we have a proposition on how both messenger RNA total levels and ribosomes could affect growth rate. I don't know if limit growth is the good sentence there. We have this expression basically. Tukaj, da je zelo pravda, da Ludo je tudi početil, da je to druga, ki je vsega vsega vsega vsega vsega in koncentručnja. To ne zelo z vsobnega vsega in vsega, ne zelo z vsega in vsega modelu, da se početim vsega vsega vsega vsega in vsega vsega vsega. in to naredal držav, zo nige glasbo za oddaj, ne glasbo, rožav, ne glasbo za zboje. Res nevom mali postavnem, z šteh, dozno ili Ludo in Luka. To je ležno, če zač Sogroth. Ne, to je zač, imaš jedno, da je dačo data, da je več bilo vsozben, že Ludo je pozdravil za najbolj slavnem všem všem, da je tudi izbrat, neselo se, da je in Nikolaj, pa tako in ist, in nekaj, da se vse zepravimo, in pravda bakterija, je zelo zdaj, da je bolje načinil na življenje. In nekaj nekaj, da je in izgleda, da je in izgleda za to, ali da je, nekaj, da je, Count? I don't know. I think he's working on this. It's a very good question up there. And then we propose that when you have slow growing ecology, you have sequester ribosomes, as was explained this morning, but you also have a fraction of the ribosomes that are doing turnover. So they're actually balancing your degradation rate. In da sem bil, da so prič naredil, da naredil se včučne in viške zelo. In jasno, da se naredilo, da je odrožen, zelo nekaj karjut, nekaj v jezice in vseh, vsaj način, za nekaj karjut in malo bo nekaj karjut. Zato da, da je, da je vedno, da je za nekaj karjut, nekaj karjut, je nekaj karjut. vsak se tomi sezologiči predstavljeni z antivbitičnim, kternje poda fenojtipicom strategijami vzpeš Otpoznega, kaj bi se mogli evaditi izgačen so v nekaj naša svojjega. Krije, kot sem se ztenil, vsak z vsega, tako sem ztenil, sečeno prijevanje izlišnji, kljer, objevali vsega, prikaj, o vašci pizdler, kar je tako vsej začeljene vsej, pa jaz najbolj je, da je začeljene vsej, da je začeljene vsej, in z njegalima vstavnosti in med njegalim sem jaz. Vsej smo se vzpešni, ne zemljamo drž, da smo vzpešni, da smo glasbni, in se veči, neč nekaj renejo in neč nekaj rov, neč nekaj renejo in nekaj renejo in nekaj renejo, nekaj renejo in nekaj renejo in nekaj renejo, Vseč, da je nožno, da je to nekaj način, kako je prišljene, da se počutno vsoči, bo je nekaj, da je zelo, da je zelo, da je nekaj način, da se počutno vsoči, da je zelo vseč, da je izglednje biogenizie, je počutno način, da je nekaj način, da je nekaj način. To je bilo vseč, da je zelo. the first part and second part will be related to more details stories, where we try to look at something that is similar to what Terry discussed this morning. You have a sensing machinery and then your location machine that sets your growth rate, In in skupn charge in danes in kaj je vse boče pa izgledati, kako je karipв, karipn kaj je kako, kaj je kaj jaz ne zašličen. Morajo jaz ponukujeli tukaj lajer, tega ne bo zsledat kako se zelo roli v konzirajno, kako se voljimo iz večjave. tudi dvise z vrste. Ovo je, kaj sem bilo, kaj sem bilo, kaj je vzivno, vzivno, in kaj je dinamiklj, nekaj sem vzivno način, nekaj sem vzivno, nekaj je vzivno, nekaj je vzivno, tudi, da je narednega vsega tudi, in z tudi tudi vsega tudi, skupajte proti. To je tudi, da počutno vsega, je tudi, da je nekaj zelo, da je to zelo, da je tudi, da je tudi, tudi, da je tudi, tudi, da je tudi, tudi, da je tudi, proč v praktaji Xingi. Your have this Sen thing, and in this 2017 paper they, they deal with it. I would say in a, in ha, phenomenological way. so I have this allocation function which is the fraction of ribosomes that are making ribosomes, and then I have to link it to translation efficiency, and I do it by looking at the steady state data, and with this I have a useful model that doesn't need any extra parameters and I can use to describe nutrient shifts. So the question is, but then there's a mechanistic layer of this, and Terry described this mechanistic layer which has to do with PPGPP. in vzve, če mi Jorgeš Češčen naši všeč se umožim, tudi in tudi, da smo pa, niekaj, nečešči vse da se kratimo. The second, the second, the third, the third and the fourth is the third, the fourth isth. The eighth is the q1 chip, we are going to make this jazza in now is what we will be doing. If we are going to make this jazza, we are going to make this jazza and we are going to do this jazza. Ur the lab, where I contestall this over, so we're just having to focus on taxi on. So there is a sensor system, there is a controller, p 6genommener and there is an actuator system that we wanted to put in a model. And there is like, I think ours is the first model. Let's say, this was done in the early days, for example, there is a model by Mark in 1991 that actually does this, dodači drugi subsob! The PPP-PP dynamics is the same as the VU paper Teri described today, the transcription regulation of ribosomes, and the balance of the flux equation for amino acid, which is the focal length of the voters from the od vsezajstva, iz inštenju. Zato, imaš vsezajstva podi katabolic, in uvršč eny Nr singogren, vsak bomo po vsezajstvu. Tukaj imaš vsezajstva, s njem ni ne hvala, bo Coň složena vo sreč, bo ju osredini, evoč. Ceseno je teoretica, s časim imam, na zelo pesku je to pogledajte, da je to spodelo. Titul Ti vse morate vsepranje vsepranje, in odnošli moji izvedajte, ogledajte vo načejce osilatori uranjeli korene podlije. Tegem pogledajte. Tako človek, Western is that, because you have mixed feedback, if you put the modules together. It works like this sensor controller, a Twitter system with a delay. And they will find that if we reduce the delay, we get the oscillations. Then we look at data. There are data around from Terry's paper, where we think maybe there are oscillations, maybe not. There is an early paper from the 80s, where you see oscillations, but if you compare the time scales, they are much faster. And then we have this data that comes from our collaboration, which was actually, I didn't tell you, but we didn't start by cooking up this model just by theoretical considerations. We cooked up this model because we had this data. And these data are actually madder machine data, where you do see these oscillations. And we don't know if it's madder machine specific behavior. I will tell you more about this data, but we see strong oscillations. And again, these oscillations are on a time scale that is different from the other two conditions. So experimentally, this is, let's say, I think it's still an open question. And here I should tell you how do we measure phi r. And I will tell you in a minute, but we use a reporter, a GFP reporter of our bosomal RNA gene. So we have fluorescence here, which we believe should be roughly proportional to ribosomal allocation, at least at the transcriptional regulatory level. So this is the first story that I wanted to tell you, but again you will get a lot of more details from Rosana, so I didn't want to get in too much detail on this story. And please interact me if you have any questions, I don't know. But now I'm going to get to the newer data that we have, which is related to this experiment, actually this experiment is already published in this paper, so it's public data that you can also analyze if you want. But in this paper we just looked at the cell cycle properties of the shifts, and then we started to look at, let's say, resource allocation properties later. So this data set is done in these microfluidic devices, where you can grow bacteria in these channels, flowing nutrients around the channels, so theoretically it's very steady conditions. And then you can do segmentation, tracking, and get data of fluorescence and volume along lineages, and your data looks like this. So we looked at different strains, where we constructed different reporters. Essentially we had two promoters. One is ribosomal promoter, this RNB P1, which we call P1, and it's ribosomal RNA, and we have two versions of this promoter. One is kind of the full version, which also includes regulation by fees, some transcription factors that are in HNS, physiologically probably relevant. And we also have a stripped-off version, but there will be, I think, no story about this today. And we put it close to where ribosomal operants are, so close to the origin, but also as a control close to determinants. And then we had a promoter that should be just constitutive, so not regulated by PPGPP through the KSA. This P1 should be strongly regulated by PPGPP, so it should behave like a ribosomal protein. This one, P5, should not, and again we can put it close to the origin or close to determinants to look at dosage effects. And then we can look at different observables. Of course, we don't have mass, we just have volume to measure cell size, so we can derive a single cell volume growth rate as this volume-specific derivative of our volume, tracked over five minutes frames. Then we have the concentration of our reporters, which are built into separate strains. We cannot measure the two promoters in the same strain, which we use as proxies of, let's say, a p-sector fraction and an r-sector fraction. And we can calibrate, so the experiment was run, as I told you, through a shift between two steady conditions, and we can look at the average values of the fluorescence across these two conditions, and we can use them to calibrate. Or we calibrate with one condition, and we can predict the average within the other condition, and we can ask whether they fall on the FIAR versus lambda growth law for bulk data, and they do. So it seems, at least from the data we have, it seems quite consistent for the averages. And then we can also look at the volume-specific production for the fluorescence-specific production rates of the two promoters. We get lineages like this, usually 8 to 10 generations in this data set, so not very long, but also not very short. Up to 15 generations, let's say, where we can measure different quantities, like the ones that you see here. This is, for example, just cell volume. This is our proxy for FIAR, and this is growth rate along the same lineage, and we have a lot of data like this. So these students, Simone, spent some time analyzing this data, and I think we have interesting results that, let's say, let's look at it together. So the first result we have is that we can analyze these quantities along different lineages, and we can ask about how different is the behavior of these variables along different lineages. And to do this, so what we want to ask from the physics viewpoint is whether we have self-averaging or not. So whether if I average along a lineage, I get the average of the whole sample of all the lineages pulled together or not. And ideally also look at the time scale where I reach self-average, which, as I told you that in our data, the lineages are not very long, but we've also looked at published data where the lineages are longer, although the things that you can measure are not the same. So to cut a long story short, you can do time averages along the lineages, and then you can compare the time averages. If you have an ergodic system, the time average should correspond to the ensemble average, so all the gray lines, lineage-specific averages, at some point they should converge on the purple line, which is the global average of our data. But they don't. Neither for growth rate, nor for the proxy of ribosomal mass fraction. So on this time scale, you can see that there is a lineage-specific behavior that is quite strong. And then you want to quantify it. So one way to quantify it is to look at the CV of the time averages. So now we take the time average and until the time t we have a distribution of time averages across lineages, and we can look at the CV of that, and it's depending on the data. Growth rate is about 10%, whereas proxy of ribosomal fraction is about 20%, depending on the replicate. And then we can compare the CV of the time averages of the lineages to the global CV of all your data. And this is a measure of how much you break self-averaging, because if I have self-averaging, basically there is no CV, so this is zero, and at the most you can get to one. And we are about 0.7 for the fluorescence proxies and for the sort of volume-specific derivative proxies, we are around 0.4 in our data. OK. And then we said, OK, so this is just 10 generation lineages. We should look at longer lineages, but we don't have longer lineages. So this is what other people do. There is this paper out from Hannah Salman as the main author, where they were able in this similar device to get lineages that are as long as 250 generations, and most of their data are longer than 100 generations. They don't measure their fluorescence proxies that we measure, but they only measure growth rate, but we can do it with growth rate. Here, for example, you see, well, the same time averages, and here you see the CV of the time averages, and you see that we are here, so where we are, let's say they are about 10% like us, but then it goes much lower, so we believe that this is basically self-average. There might be some error related to measurement as well. Basically, what you can say is that in 10-15 generations, you have this individualistic behavior, but then in 100 generations, you sort of have average behavior. You can ask something else, because again, in 10-15 generations, these cells are diverse. Let's say these lineages are diverse in the growth rate, so maybe this has an impact on fitness, for example. So you can try to compare the variability of growth rate here we are comparing along single lineages, with the global variability of all lineages. This is different from what we looked at before. We know that a lineage has a lineage-specific average, but we want to ask, there are also fluctuations within lineage around this average. We want to ask how these fluctuations compare to the global fluctuations of growth rate, and they are similar. So here it says that 80%, they could recapitulate 80%, or even more than 100%, because the standard deviation of a subsample can be larger than the standard deviation of the whole sample of the global fluctuations. So basically it means that they fluctuate around the lineage-specific growth rate, but also they explore a wider range of growth rates while they do that. So this is basically what we see in terms of loss of self-averaging. And then we also see oscillations, which is something that people have already reported. These are sustained oscillations, as you see here from the Fourier transform. And if you go to the long lineages, you see oscillations also from the autocorrelation function. Whereas if you go back to the short lineages, both in our data, and in the data published from our collaboration, and in the data that we took from Hannah Salman's lab, you see a diversity of non-self-average behavior also in the autocorrelation function. So these oscillations are sustained oscillations. So they are not the kind of damp oscillations that we see in our model, but people have already reported the presence of these sustained oscillations with cell cycle period. And there are different hypotheses for those. We don't have the data actually to test these hypotheses. One hypothesis is that they are due to non-equal ribosome partitioning at cell division. And the other hypotheses are related to the fact that you have gene dosage changes at specific times in the cell cycle, which drive these oscillations. So we also report that sort of the feedback response system of results allocation might have its own damp oscillatory behavior, but it's hard to test it with this data. Then we looked at sort of Fajar-Landa correlations across and within lineages. So within lineage cross correlations is in line qualitatively, but not quantitatively, because you see that the positive correlation is very weak. So with, let's say, a positive change of growth rate lambda with the positive delay in response to a positive change of Fajar, which is the peak that you see here. Yes. And then we see this negative peak. This negative peak, which we think is due to dilution, and we compared with this paper where they looked at, this is a paper from Sander Tan's lab and Dan Kiviet was the first author, where they looked at similar cross correlations where the reporter was a reporter for a catabolic enzyme, so different from our own. Can I ask a question? So just if I understand this plot, so what you're showing here is like whether the fluctuations of Fajar and lambda, if you want to see whether these fluctuations respect some sort of growth laws, and what you see is that actually not really, but the best you can do is if you have a leg between... Yes. Well, I do expect a leg between, let's say, ribosomal location and growth. So if I look at this leg, I find the positive. It's similar to what they found in this study. Now I'm going to show you a little bit more of their plots. But let's say if you have, for example, a catabolic enzyme, you could expect that you would have this sort of causality where you increase the concentration of this enzyme and after a while you increase growth rate. And this is the positive delay. But you could also have the opposite causality where you increase growth rate, you dilute the cell and you dilute your catabolic enzyme and then you find this kind of peak, the negative peak in the past. And yeah, sorry, I don't know if I answered your question. OK. OK. And this positive peak, just to compare, is something that we do find the dilution with the other promoter, but we don't really find this positive peak. And we looked at their plots and this kind of positive peak, so maybe it's remarkable. We don't know. This is really new data for us as well. I mean, the data are very old, but we started looking at this data seriously only recently. OK. Then we asked, we also have this lineage-specific behavior, so we can ask if we do the time over this 10-generation period. If we do these time averages for every lineage, we can get, for every lineage we can get a value of phi r, which is the time average, and then a value of lambda. And we can do the scatterplot, are they correlated, and they are not correlated. If anything, they are anti-correlated. So phi r time average, proxid by r, concentration of GFP from this P1 promoter, slightly decreases with increasing growth rate time averaged. OK, that's it. That's basically the whole story. I think I'm early. I have one last thing that I wanted to tell you, which is we tested something else with this data, which is related to the nature of the growth rate fluctuations. So basically you compare your model with the stochastic growth model of this kind, where gamma is a parameter that you want to try to learn from the data. So if gamma is zero, basically you have an additive noise. If gamma is one, basically you have a purely multiplicative noise in your growth process, and then you can analyze your model the specific intermediates, and it turns out that if you look at the phano factor of your logarithmic size increase conditional to size, then you get an expression dependent on this parameter that you can test with your data, and then you can guess a value for this parameter, and we find that our data are close to one-half which is a stochastic autocatalytic model that people have already proposed in the literature. So maybe we can show that the data are in line with this proposition. So that's it. That's all I wanted to tell you. I told you that we are doing several things and it would be nice to talk about it if you're interested in more detail. I told you that we have this story where we try to put the modules together in the oscillations, and then I told you about this sort of exploration of single cell data where we have these observations of allocation and growth rate that we can correlate, but we find a reality that is quite complex. It's intriguing, it's quite complex. It teaches us a lot I think about how single cells are behaving, but it's very hard to relate to these macroscopic models that we are working on. So, yeah. Thanks for your attention. I want to thank the collaborators that worked on this project. Experimental data come from a collaboration with Pietro Cicuta and Bianca Sklavi, but they are published. So basically you could analyze this data just as we did. We do plan to have open positions in the coming year if you want to join us and these are all the people that worked including Jacopo and Luca. Thanks Marco. We have plenty of time for questions. Hi Marka, nice work. I was trying to understand the connection of it to the nutrient upshifts and downshifts and I guess I got lost somewhere. So in your mother machine it's always the same nutrient, right? The single cell data that we do shift nutrients but in the single cell data that I showed you are only steady state data before the shift. This is the averages across the shift that you see. And what do you shift? Is it an upshift or an outshift? It's an upshift and we go to M9 plus glucose to M9 plus glucose plus amino acids. Why you didn't study the downshifts because that's where the lags which are kind of can be quite heterogeneous if you we know that cell history or even small variations between strains can affect the time lags between shifting from one nutrient to another. So is there any plans to move to downshifts? This is all the data with this PhD student. She tried some downshifts but for some reason it didn't work out. At the moment actually we are asking our collaborators not to even try to do it's a mess to do shifts in the mother machine because the cell size changes a lot with growth rate so you need to have channels that are of the good size to have cells that are not tilted basically or stuck in both conditions. So now we are asking if anything we want more data like this let's say like this and so we are asking them to do steady conditions and we might be able in some time to get more data like this with the same strains. Is there any difference in the ribosome growth rate correlation in the cell? The lack of correlation. This one? Or this one along lineages? Did you look at for your strain you could of course also grow in batch culture? Yes. And just measure how does GFP change? We haven't done that. We are doing that but these experiments were not done in the published data there are no data like this. Yes. We asked the same question and in particular what we are doing we are also trying to calibrate expression with direct RNA over proteins, measurements in batch cultures with these strains and Luca is already maybe Philipp who is a student in Luca's group will show this data or maybe not. But anyway we have some preliminary data but something we can do is that we have the same hypothesis that is just protein useless protein burden because we have this data in R e and T e so with different dosage and different GFP expression and we haven't done it but we can ask whether they fall let's say in the same kind of protein burden master curve. So exactly in the group of in the group of the young there have been experiments that show exactly the same thing that the growth flow does not hold for the distribution in the population. And they were single cell. And did they have an explanation for that? Other questions? Ok, so let's thank Marco. Thanks.