 Today, we start with Valentina Valdazzi that is going to talk to us about resource allocation in Ecoli. Thank you very much. Can you hear me? So thank you. So today I will talk about a work that has been recently published and that is the result of a collaboration with different colleagues, mainly from France and in particular, Petite de Jong, who is an ethereal, which is the French Institute for Applied Mathematics and Informatics. Sorry? Can you use the microphone? I have a microphone. No, I'll try to put it. I think I'll try to put it. Close up. It works. Is it better? Yeah. Yeah, okay. So as you all know, the material growth involves the conversion of nutrients into biomass, but some of the energy which is stored in the nutrients is also transferred to small energy cofactors like ETP that are needed to drive the biomass synthesis reaction. So in the cell, we have a coupled flux and then mass and energy fluxes both in the catabolic and anabolic phases. So generally, two criteria, two microscopic criteria are used to characterize microbial growth. The first is the growth rate, which is just the speed of conversion of this nutrient into biomass. And then the other one is the growth field, which is a measure of deficiency of this process. So it's generally defined as the biomass produced per nutrient consumed. So I don't have to to convince you that bacterial growth can be analyzed from the perspective of protein allocation. So the idea behind this is that proteins are the main components of biomass and proteins catalyze all the functions needed for cell growth. So the way the protein is partitioned across different functions reflects somehow the results allocation strategy of the cell. So as we already seen in the previous lecture, there is a very tight relation between the growth rate and the fraction of ribosomal protein. But there are also connections between the resource allocation strategy and the yield. So one very telling example is the so-called overflow metabolism that is a phenomenon that we can observe in many, many different microbes. So most microbes actually have different pathways for ATP energy production. So they usually have an efficient pathway, which is usually respiration in a healthy condition, which leads to the release to CO2. And then they have another inefficient catabolism, which leads to some byproducts. So for instance, in the E. coli is fermentation and the byproducts is acetate. And so what is observed is that when you increase growth rate, there is a switch from an efficient catabolism towards an inefficient one. And so this switch has been explained via protein allocation. The idea is that this switch is due to a sort of tradeoff between the ATP yield, that is of course higher in respiration than fermentation, and the corresponding proteome cost that is needed to make the pathway work. And so indeed, in a very well-known paper, Bazook workers showed that the fermentation has a higher proteome efficiency than respiration, meaning a lower proteome cost. And so at a given point, it's interesting for the cell to switch to fermentation in order to grow faster. So what I showed you are experiments that have been done by setting bacteria in different mediums of increasing nutritional quality. But actually, even in the same condition, different E. coli strains can show very large variation in both growth and growth rate in yield. So in 2017, sorry, what do you mean by that? Yes, actually during the glycolysis, yes, you make a little bit of ATP, but it's quite small compared to what you... I mean, the ATP yield of glycolysis is very small compared to what you obtain by respiration and fermentation. So yes, it's included, but I mean, the big part, yes, can be... I mean, it's really small. Yes, you... so that's one actually is constant, it's kept in any case. This switch is most upper after the glycolysis part. So, okay, so in 2016, so Moncan co-works analyzed seven E. coli strains that are commonly used in lab and industrial processes. And what they showed is that indeed these seven strains can have different growth rates. Here you see, you can range from 0.6 to almost one. And this growth rate is actually achieved using very different strategy for concerning the energy production. So some of them use a pure respiration, so all the yield, the growth yield is biomass. Where other use a mix of fermentation and respiration, so you have acetate as a byproduct of the energy producing pathway. So the question we asked is, can we reproduce the observed variability in rate yield phenotypes by changing the resource allocation strategy of the cell? And if not, of course, which part of the observed variability can be accounted by the resource allocation strategy itself? So the way we did is that we developed a model that coupled explicitly both the energy and the mass flexes based on minimal assumptions. So the model we consider is the following. So we have five microscopic reactions. There is one reaction for the conversion of external nutrients into central metabolites. So they have taken the conversion, let's say, to central metabolites. Then the central metabolites can be used to produce energy and we consider the two possible pathways, the respiration and the fermentation with the acetate as byproduct. And then also the central metabolites can also be used for biosynthesis, biosynthesis of protein on one side, but also other molecules like RNA, DNA, but also maybe lipids for the cell membrane. And so what is quite peculiar to this model is that our biomass definition includes all these compounds. So not only proteins, but also these other macromolecules and the metabolites. So in the following, I will use a few times this kind of pictogram. So here what you represent is how the total biomass of the proteins is partitioned across the different compartments. So you have in the white part, let's say, is the fraction of carbon metabolites. Then you have the other proteins in other macromolecules in the violet. And then the central part is the proteome, which can be divided into different proteomic classes. So by using five allocation parameters, so you have a fraction for ribosomes, enzymes for fermentation, respiration, central metabolism and also the protein which catalyzed, let's say, the reaction towards the other macromolecules. So in terms of equation, it's an ODE model. So here you have the carbon balance with the biosynthesis rate. It is split into the different kinds of enzymes and proteins you have. And then another point that we also consider, natural degradation for biomass, which accounts for maintenance costs. So it's our nowhere of protein and other compounds. This is just, so I will come back later. So this is just a sketch, actually. It's not meant to have a correct proportion, just for each part to be sufficiently wide. But for instance, in E. coli, I will come back later. When we calibrate the model, so for instance, the central metabolites are around 3% of the biomass, proteins, 70% and the rest. So, I'm sorry, I can't make it. So we have an equation for camber balance and then we have, yes, the same, sorry, I didn't understand. Yes, let's say we didn't have data to make them different. So most of the data actually are protein because protein are the most abundant part of the biomass. So we made a point that is the same, yes. I don't think it would make much difference actually in the model. Because anyway, C are generally small, but we will come back later. So then we have... Another issue, if we go back to the previous slide. Yes. So, when, at least when E. coli is growing on glucose, actually the synthesis of protein is coupled with the production of energy. The production of energy? A lot of energy is produced while glucose being converted to... Here actually, this micro erection includes both the amino acid synthesis and protein translation. Oh, you're not separating the residues? No, it's a little bit unusual. So the central metabolites here are more the precursor that are, I mean, small metabolites like... Before amino acids? Yes, before amino acids, yeah. So actually, for us, these two reactions, both the protein and the other macromolecules are consuming the energy which is produced here. And concerning the first question, we are neglecting the ATP that is produced during the glycolytic pathway. So it's a little bit schematically as a view. Yes, ATP, amino acids are there. Yeah. Just another question. So here you're assuming that the only thing that the cell is splitting out is actually acetate and CO2, right? Everything is converted into biomass except for... Yes, it's our, I mean, it's a theoretical model, but... No, no, no, just... Yes, it's what we assume. Just CO2 and acetate in the case of when used for medicine. Yeah, yeah. Okay, thanks. We'll come back later, but yes, the answer is this. So then we have also a question for the energy balance. We have a strange hand here. The energy balance of what we consider is the production and conversion of ATP-ADP, let's say. So ATP is produced by the fermentation and respiration flexes with different yield. These are the small N parameter and it's consumed during the biosynthesis of biomass or protein and other macromolecules. We also consider a dissipation term which is related to energy spilling reaction. It has been observed that not all the energy which is produced by the cell is actually used for biosynthesis. There is some ATP that is lost, so the mechanisms are not completely clear up to now, but so we included a dissipation term here. So we made a few assumptions. First, we assume that the total biomass concentration of the cell is constant. So what we look is how this total biomass is partitioned across the different compartments, let's say. We also consider that the total concentration of energy cofactors, let's say ATP plus ADP and equivalent, is constant. So we just look at the recycling between ATP and ADP. And so for the rates, we use Michaelis-Menten kinetics for all reaction rates, in particular for all those rates that depends on both carbon precursor and ATP. For instance, we use a product of Michaelis-Menten term, and a special case, let's say, is the flux for the uptake conversion of external nutrients. We suppose to simplify that external concentration is constant. So it is simplified to a linear relation with linear in the first kinetics with enzymes concentration. So the model has been calibrated using published data on a specific E. coli strains, BW21-15, on minimal medium. So we use data sets from which report metabolic concentration, protein concentration and metabolic fluxes. And the way we work in order to somehow ensure the consistency between these different data sets, which have been obtained, of course, in different experiments. So what we do is that we use literature information to estimate the total biomass density and the total protein in a metabolite concentration based on biomass quantification. And then we use this data from metabolic and proteomic data to define the proportion of central carbon metabolites within all metabolites and the different partitioning of the proteome into the different categories we consider. And then, so with the concentration that we have now obtained, we can derive the rate parameter using the measured fluxes. So what is important here is that actually we fix all parameters without fitting. Everything is fixed from the data with some small assumption, but we do not fit anything. No, we have different conditions. So we do what is, we have to understand, we have different, we can say different conditions to carbon sources and different culture, batch and chemostat, but we do a calibration for each situation. Yes, but it is across different conditions. So here we are not using the same way as the growth flow. Here we are, what we would, our aim is, I will come in a moment, is to see, given a single condition which is the phenotypic space of rate and yield that we can obtain by varying the resource allocation strategy. We are not dealing to the switch indeed between different environmental conditions. So it's a little bit, it's the same context, but I use differently. So okay, so I say that we have different conditions. So exactly what I was telling, so our aim is this. So what we do indeed is that we set in a condition, so for instance glucose batch, and then we make a random sampling of our keys parameter, our resource allocation parameter, and for each combination keeping all the parameters, the kinetic parameters for the reaction rate and so on constant. And so for each situation, we make a numerical simulation up to steady state, and we compute the corresponding growth rate and yield that we obtain. So there's no optimization, and there is one, let's say, random sampling per environment. Just to understand. Yeah. The proteomic and the proteomic data to get sort of effective coefficient. Yes, exactly. But it involves some degree of arbitrariness in what you call, well ribosome is ribosome, but then in terms of this other, it's a bit, so for example, catabolic enzymes. Yeah. You know, E. coli makes tons of catabolic enzymes, many has nothing to do with function, right? They're just, you know, they're just throwing all of the... Yeah, actually we started from the data by, by Schmitt for the proteomic data, and Schmitt already made a partition of the proteomic, two different functions, there was one which was... But my question is for example, for the catabolic proteins. Yeah. Do you define them as the sum of all of the catabolic proteins, or only the catabolic function that's carrying... Let's say you're growing on lactose, right? Only LACY, LACZ contributes. No, no, no. But all kinds of other ABC transport this and that are expressed. Actually, it's not me that did this work. Schmitt reported the proteome concentration for different factors. Yeah, yeah, yeah. So I use this data. Yes, it's summed up. Okay. All of... Yeah. We just used the reported fraction of proteome that, let's say, he set it to, for instance, central metabolism. So the kind of results we obtained, so it's like this. So these are the results for glucose batch. So what you obtain is a cloud. Let's say the space of rate yield phenotypes that the cell could... In principle, could reach by changing it at the resource allocation scheme. So the first comment is that... So this cloud is bounded. So there is, for instance, a maximum gross rate that is attainable, which is for us around 1.1, let's say. And there are rate... The gross yield lies between... For most gross rate between 0.3 and 0.9 almost. And what you can observe is that for small gross rate, we have a positive relation between the maximum gross rate and the gross rate, which is due to, let's say, a proportional lower burden of maintenance costs with respect to gross here, for this part. And then, instead, for the high gross rate, we have more a negative relation, which is related to the trade-off between yield and rate. I will come back on this later. So then the point was, okay, this is what we predict. How does this compare to the observed variability? Ah, yes, you're right. Yes, by David and Compt. Yes, we could try to... Yes, because, indeed, what is real, what you say, is that the mapping between the rate yield phenotype and the underlying resource allocation strategies is not too new a call at all. So there are many resource allocation strategies that lead to the same macroscopic phenotype. So indeed, we could try... We didn't have it done yet, but yes, it could be interesting to try to see which phenotype is more probable, let's say. Of course, under the assumption of uniform sampling. So in order to see the observed rate yield, we compiled data from literature from more than 50 papers. So we got data about different wild type strains, mutants, and also artificially evolved strains. Yes? One of the pathways? Ah, yes, that's a good question. It would be a lot different, but maybe can we discuss this later, because I will come back about this. So we have a picture that explains this part. So we made these data sets and we plotted so the data we could obtain over our cloud. And so indeed, we had a very good agreement between the data and the prediction, and the data fall into the expected cloud. This is true for glucose, but also for the other condition. Of course, we have less data for glycerol and chemostat in particular. We lack data from evolved strains and mutants in most cases. So however, this kind of picture can tell us something. So first, what you observe is that the evolved strains lies mostly on the right side of the wild type. So indeed, evolved strains have higher growth rates and in particular, some of them are quite close to the boundary, both for the growth rate, but even for the yield part. Instead, the mutants lies mostly on the left side. So most mutations are actually favorable to the cell. So of course, this is the picture for the growth rate, but we can go further and look for the expected space to another in the flux space, let's say. So here you have the relation between the growth rate and the uptake rate and the growth yield and uptake rate. So what is expected is a correlation between rate and uptake, glucose uptake. So this is kind of natural because the more you uptake, the more carbon you have for growth. And for the yield part, you find again this kind of tradeoff for high glucose uptake rate, which is just a mirror of the tradeoff between high rate and growth yield that you saw before. Then we can have a look this time to the fermentation rate, let's say the acetate secretion rate because that's what you measure. And so here for the yield, you see there is a very clear tradeoff, which is also expected. The most you use fermentation, the less you are efficient. However, what was more unexpected is that the model did not predict a direct relation between the growth rate and the acetate secretion rate. Indeed, according to the model, the cell can grow fast using a very wide range of ATP producing strategy ranging from pure fermentation, sorry, to a mixed of respiration and fermentation. And what is also important is that indeed we have some data that do this. We have two wild type data that grow fast without fermentation and also some of the evolving strains have very low acetate secretion rate. So the interest of course of the model is that now we have our phenotypic space with this characteristic and we can try to map them to the underlying resource allocation strategy. So as I said, the connection between the two is not easy because there are very non-linear relations between the resource allocation strategy, the biomass composition and the flaxes and so the observed phenotype. So here I pointed to some, let's say, interesting points. So BW is the point, our calibration point. Here is the point having maximum growth rate and the other is the point having maximum yield. What is apparent here is that in this model the proteome fraction is not constant at all. So in particular, for instance, for the maximum yield, you see that here we have a picture that shows how the flaxes correspond to this point and the biomass composition and what you see that indeed this point, the maximum yield is obtained by reducing all the ATP consuming pathways. So very few protein synthesis, very few other macromolecule synthesis and so of course, given our constraints of total biomass constant, we have that the biomass is composed largely by metabolites. So of course, it could be a little bit theoretical, but indeed there exist cases and mutants in which you can accumulate glycogen, which is a storage product and which is accumulated when you don't, you cannot use your caramel metabolites for synthesis, for protein synthesis essentially. And this fraction can be high as high as 30% in a few mutants. So in the following, I will focus on a specific area of our rate yield phenotype, so the phenotypes having, at the same time, a high rate and a high yield. And the way, why I do this, because indeed according to the model, this data as we showed before are obtained using very low acetate overflow, so essentially using only respiration. And this is interesting because it goes somehow against the accepted view that we saw before that in order to go fast, we need to have large ribosomal fraction. And so the switch between respiration and fermentation, according to this vision, is needed to, let's say, free some proteomic resources for the growth because fermentation will still require less protein, so we free some proteomic spaces that can be allocated to much ribosomes to go faster. So in this view, the overflow is somehow required to go fast. So of course we try to understand which resource allocation strategies can lead in our model to the situation where you are both fast and efficient. So first hint comes from the comparison between our mu max point and our calibration point because I don't know if you remember, maybe I can come back. The mu max point is, of course, by definition, the one having the maximum growth, but it's also a lot more efficient than our reference point. So when we compare the two situations, we see that indeed the mu max phenotype is obtained by increasing the nutrient uptake flux and the protein synthesis rate. But it tells when we compare the total protein concentration it's somehow lower than in the BW and as the allocation fraction for the ribosomes more or less unchanged, it means that the mu max has lower ribosomes than the BW. An ETP is also slightly reduced, but mu max has a higher metabolite fraction with respect to BW. So somehow if we come back to the expression for our protein synthesis rate, the bar is increased despite reduced air, reduced ATP because the sieve part is increased. So somehow it seems to say that there is another strategy for the cell which is not to invest more into ribosomes, but is to somehow exploit more efficiently the ribosomes, the protein that exists, increasing the saturation of the ribosomes. So of course I would just compare to what we know. So unfortunately we don't have data really here in the top of the bar or top of the cloud. So we have data for another strains which is called NCM, which grows quite fast and which has a higher efficiency than BW. So we can compare the phenotype of these two strains and what we found. So here we have the phenotype, so indeed NCM has a higher growth rate, a higher yield, a higher uptake fluxes as expected. And also when we compare the biomass composition, NCM do have a higher metabolite concentration. So it's consistent with what we expected, let's say, with the model predicted. And we also compared, we used a database of curated KM values from a paper by Dorado and coworkers for glycolytic enzymes. And so we compared the enzyme saturation level for the two strains and indeed for NCM, the saturation level is almost more than twice the one of PW. So let's say it's consistent. So of course, just the last point is what about the accepted view, so the idea of the proton. So of course this strategy exists and indeed the model we developed can be reduced exactly to the one proposed by Baz and coworkers when we make an assumption. The assumption is to assume somehow the concentration of metabolites, ATP, and other macromolecules constant. So if you do this, the total protein concentration is also constant. And of course all the kinetic rates just reduced to a linear first order rate in the concentration of the enzymes and so this kind of trade-off between metabolites and proteins is no longer possible. And so our model predict exactly as the one by Baz and that's the only way to go fast is to switch fermentation. So the message is our model has somehow less stringent hypothesis and so can account to alternative or resource allocation strategies that the cell can set up. So okay, just a conclusion to sum up. So we developed a cost grade model of coupled energy and mass fluxes. So this model has been shown to reproduce quite well the observed variability of E. coli phenotypes in different environment. So we can say that the resource allocation strategies is indeed a major determinant of this variability that we can observe across different strains. It also allows to better understand some specificity of cell function, let's say a reason at least over the cell function. So what you see is that so the common E. coli strains are not optimal for growth on a single substrate but 80 strains, all the strains can do much better. And then the central point is that fermentation is not required, let's say, to go fast. But there are some other strategies that can be set up by increasing the saturation of the enzyme. So these of course come out cold to the fact that it may be important to account for metabolites in biomass because otherwise we cannot really take the kinetics of the enzyme into account. So the perspective, so from, it would be nice to have more data to validate the model. So in particular, we would love to have more, I mean a better characterization of some of these interesting evolved strains. So namely the resource allocation, the protein allocation and their biomass composition in order to see if it fits to the prediction. Then there are also some areas in this cloud that are somehow unexplored by experiment, in particular all the rate yield parts. So it would be nice to have data there to see if it still fall in the cloud. And then the other point is that so the model we built is cost grade, is quite generic, so in principle it could be applied to other organisms provided that we have all the data we need to calibrate them. And then from the theoretical side we would like to better investigate mathematically what happens here in the Pareto front. And then the next step, ideally would be to use this model dynamically. So in particular to study the dynamics of resource allocation changes when you change environment. So because up to now we are in a fixed environment, a steady state. And so but in order to do so we have to implement somehow the regulatory function behind our key parameters so the keys are an input of the model. So we should make this key function of the variable, the inner variable of the cells. So we saw before it has been done for the ribosomal fraction with the PPGPP mechanism but it should be done also for the energy part in order to work here. And so if one disease is done in ten years I don't know and then we could imagine to study for instance the evolution that we observe so how which kind of changes in the regulatory function can lead to wild type strains to move towards fast and efficient phenotypes. So I conclude here, I conclude on a small announcement. We are looking for a PhD student in collaboration with colleagues from Area Bordeaux. So multiomics data integration for the analysis of microbial communities lives. So if someone is interesting or you know someone who is interesting please tell me. Thank you very much. Thank you. We have time for a couple of questions. So I think it's very interesting. I was wondering whether you looked at this protein concentration behavior in the evolved strains like how do they... We don't have data on evolved strains. Actually we have data just on their phenotype growth and yield but no proteomic studies on the strains. So it would be very interesting to have this data. Interesting study. So several comments. First of all the difference between the growth of NCM and BWMG is a question that's been bothering us for like 20 years. So I'm glad you're taking a look at it because in many aspects we looked at we couldn't see any difference. But then I have a comment on what you pointed out about the metabolite concentration difference. I think you're basing this on the two sets of experiments. NCM metabolites were done by Josh Rabinovich's group and BW was done by Uri Sauer's group. They basically have the same measurement but divided by different water volume. Sorry, Norma. Big work was on this. We go into this in detail. I think that in NCM we used a dataset by Park. That's Josh's app. So these measurements are done by measuring the amount of metabolites per cell mass. But then you have to divide by cell volume somehow. They used two different... They divided by two different volume. So if you talk to Uri's lab there was just number somehow high just because they divided by different volume. It doesn't mean I think it needs to be done consistently by the same lab. It was a very big problem when we tried the dataset because everyone is... We can metabolite concentrations really. Maybe the biomass density at the front and I guess we're here from the next talk. This kind of issue. We assume a constant. But the other thing is if you are going to vary metabolite concentration then there is another important aspect that I think we could add to the model because of course the cell should know that increasing the concentration they could get a boost of activity for free. But on the other hand there is a penalty for increasing the concentration. That's why E. coli and many cells they fill their metabolite mostly with glutamate that keep the other concentration at low level. It's for a reason because if you use something else other than glutamate then protein activity change. It's not a free parameter you can just increase. I agree. There are many other mechanisms that are missing regulatory mechanisms. That's why I'm not sure that the extreme phenotype are probably not completely smaller kind of theoretical view of the... But can be interesting to reason how it works. But yes, I agree. I just wanted to ask in your talk the variability between strains comes mainly from random sampling of the different allocation parameter, right? And you keep everything else fixed between strains. Can the other parameters change between strains basically? Yes, of course. Somehow we use the more strict hypothesis. The hypothesis are very strict. As you say we assume that everything is constant and probably there are metabolic regulation that can change the effective kinetic parameter. But despite of this the consistency with data is quite good. So yes, I agree. It's a theoretical view but as long as it's quite consistent we think that resource allocation changes is however a big part of the expected phenotype. You could check if you get the same cloud if you change some set of parameters. So what we need is the reason we made another calibration for another strain, not the BW, the MG I don't remember the number, the MG strains and what we got is actually I have a picture I guess. We got more or less the same pattern. It's here. There are small changes actually on the shape so one is a little larger than the other but more or less it's almost consistent and we keep the same pattern increasing and decreasing. So I agree with you. It's not perfect. It's a course-grade model so if you want to really have predictive fit on a specific strain it's not the way you want. Thank you. Thank you. Excellent talk. I was trying to put it in more ecology context in my head. So when you're quantifying the yield you are assuming that everything is converted to biomass both the acetate and the glucose if you are in a community or maybe I misunderstood it so you had this barplot for different strains and you had a part of the barplot from acetate. The monk studies you mean this not just before I guess yeah this one so this is not my plot it's the plot from this study so here indeed in the yield part they put biomass for the cell and plus the overflow by product so how would it change again how would your diagram change if you will calculate the total yield including the acetate if you are growing alone you are consuming your own acetate if you live in a community other strains will eat your acetate likely. The shape would probably change so I can make a prediction but I'm not sure because I didn't test but so what I expect because here I have maybe I can go to this picture it's better here you see how the different points use acetate so if you the point which use secret acetate are in the bottom part so if you include acetate into the biomass you would increase the lower yield I guess so it will get slimmer shrink shrink the cloud will shrink I guess but I didn't try this is a good question so I have a follow up question to what we saw before with the overlap of BW and MG 1655 so I guess these strains are rather similar do you have any option to look at strains that are not that closely related because I'm not so surprised that the clouds look similar for MG and BW the problem is that we need in order to set the model that we did here we need for the same straight metabolic concentration, protein concentration and flaxes and I I promise I really look in everything and I couldn't find data except for these very well known strains that I agree with you are very probably quite related so yes I couldn't find data it would be interesting and also not for NCM for example for NCM what? for NCM we got the 3722 so how would the cloud for this look like? it's more or less the same do you have picture or not the shape we tried with all of them the shape keeps the same what changes a little bit the upper bound value can be a little bit higher a little bit lower but the main feature I guess for NCM it's a bit larger than for BW so the data are even more within the cloud okay thank you if there are normal questions oh no there is one more it's the last one thank you it's very exciting I see in this model an opportunity to explore mutations and basically trying to understand the effect of mutations and that's usually very hard with experiments so I think having this model is really great usually so how do you know if you change the parameters for the regulations size and that's a small change do you get a small change in the G and the yield so is the mapping continuous or differentiable? we didn't try to change the parameters both yes I guess it might be quite continuous but keep in mind that our parameters are effective parameters so it's difficult to relate this parameter to a very specific mutation in a given gene it's like that's why we have a calibration for each environment because in our for instance the update pathway is not at all the same for glycerol or whatever so you use different enzymes so it's somehow hidden within this parameter I think if you use it so I was just curious about this parameter that you had in the beginning which is like a wasted ATP rate can you learn something by then fixing this parameter probably you have only a couple of values because you fix it with the reference actually we don't have really values I mean the way we fix these parameters that we have the actually okay back to the balance we have the ATP so we are so we are a steady state so the left side is zero and then so we have the flaxis for MR we have everything except this and so this is somehow how we match the balance so we don't really have data it's just the values I get so we get the value I can tell you later what is interesting maybe is that at a given point I try to see which were I compare the value obtained for different strains on a given environment and indeed this is one of the value that we change much so for instance for NCM is a lot lower than for BW so maybe there is something there I cannot say more right there are no more questions for sure and let's thank Valentin again