Glycolysis/gluconeogenesis specialization in microbes is driven by biochemical constraints of flux sensing

Abstract Central carbon metabolism is highly conserved across microbial species, but can catalyze very different pathways depending on the organism and their ecological niche. Here, we study the dynamic reorganization of central metabolism after switches between the two major opposing pathway configurations of central carbon metabolism, glycolysis, and gluconeogenesis in Escherichia coli, Pseudomonas aeruginosa, and Pseudomonas putida. We combined growth dynamics and dynamic changes in intracellular metabolite levels with a coarse‐grained model that integrates fluxes, regulation, protein synthesis, and growth and uncovered fundamental limitations of the regulatory network: After nutrient shifts, metabolite concentrations collapse to their equilibrium, rendering the cell unable to sense which direction the flux is supposed to flow through the metabolic network. The cell can partially alleviate this by picking a preferred direction of regulation at the expense of increasing lag times in the opposite direction. Moreover, decreasing both lag times simultaneously comes at the cost of reduced growth rate or higher futile cycling between metabolic enzymes. These three trade‐offs can explain why microorganisms specialize for either glycolytic or gluconeogenic substrates and can help elucidate the complex growth patterns exhibited by different microbial species.


Response to reviewers
We thank the reviewers for their carefully assessment our work. The reviewers have raised several helpful points that we address below. We believe that the changes improved the quality of the manuscript and we think it is now ready for publication.
Reviewer #2: In this work, the authors developed a relatively simple coarse-grained kinetic model of central carbon metabolism that is combined with equations for allosteric and transcriptional regulation. The main purpose of this model is to describe the lag time that is observed when bacteria switch substrates from glycolytic substrates (e.g. glucose) to gluconeogenic substrates (e.g. acetate), or vice versa. First, the model was trained using experimental data from previous publications. Then, after performing a wide range of simulations using the model, the authors conclude that there are inherent trade-offs between the length of the lag phase, futile cycling, and growth rates of microbes.

Concerns:
1) While it is interesting that a relatively simple model that combines coarse-grained kinetics and regulation (only six equations in total) can capture key experimental observations such as changes in concentrations of intracellular metabolites FBP and PEP and the length of the lag phase after a carbon source switch, it is important to consider that the model was specifically constructed for this purpose and that the parameters in the model were fitted using experimental data from substrate-switch experiments. So, it should not be surprising that simulations are in fairly good agreement with experimental observations. I would argue that the majority of simulation results that are presented in this work are not independent model predictions, but rather reflect the fact that the authors did a good job fitting the model parameters. Yes, the model was designed for the specific purpose to describe the dynamics after nutrient switch. Not every model would be able to do this, and we believe that we added just enough components for the model to be able to describe all phenotypes while at the same time being flexible and general enough to scan through parameters to understand what types of kinetics would in principle be possible. A full model of metabolism (one that contains our equations and more) would certainly be able to do the same, but boiling down the model to its key components allowed us to identify fundamental regulatory principles that might have been extremely difficult to identify in a full model.
To be absolutely transparent and reflect that our model is fit to experimental data, we added the following clarifications. Highlighted text: underscored We also removed the absolute scale of the metabolite measurements, because absolute quantification based on published data is uncertain (see also comment from reviewer #3). This should further focus the reader's attention to the fact that Fig. 1 is a calibration of the model and to highlight that we want to reproduce the differential changes of abundance and concentrations of enzymes and metabolites in glycolysis and gluconeogenesis.
In my view, the most interesting (and truly independent) predictions of the model are regarding the activity of metabolic futile cycles. Unfortunately, none of the predictions regarding futile cycling were experimentally tested. It should be feasible to test at least some of the futile cycling predictions using the experimental methods described in Nature chemical biology, 12 (7), 482-489, 2016.
Futile cycles are key to this work, and we would love to measure them. Unfortunately, they are notoriously hard to measure, in particular the ones at the irreversible reactions Pfk and Pyk that are so important in this paper (even the paper cited by the reviewer did not measure these!). We have thought several years about mass spec labeling strategies how to measure these futile cycles at these nodes, but have concluded that it is not possible. To our knowledge previous studies have only quantified futile cycles at the Ppc node and these strategies do not work at either Pyk of Pfk.
To establish a causal link between futile cycles and lag times (without directly measuring them) we have shown in previous work that overexpression of either of the enzymes Pyk or Pfk that are contributing to futile cycling in E. coli during shifts to gluconeogenesis dramatically increased lag times [see Extended Data 2) The authors mention that they ignored the Entner-Doudoroff pathway in their simplified E. coli model. This is probably fine, since it is known (e.g. from 13C-flux studies) that the ED pathway is not very active in E. coli. However, the authors do not mention that they also ignored the pentose phosphate pathway, which is highly active in E. coli, e.g. 13C-flux studies typically show that around 30-40% of glucose is metabolized via the pentose phosphate pathway. The authors should discuss what is the expected impact of ignoring the PPP pathway on their model simulation. For example, by formally considering the PPP pathway, a significant fraction of glucose carbon would be by-passing the FBP node, i.e. since about half of the carbon flowing through PPP will end up producing GAP, and this will likely change the form of the equilibrium equation 7 (which will no longer be quadratic).
The reviewer is right to point out that the precise mathematical form of the growth rate -lag time relation depends on the equilibrium equation. We note that this is not the primary focus of this work. We consider the obligatory existence of the three tradeoffs as the main finding of this work and this result will be unaffected by an altered equilibrium equation. This is also why the model naturally extends to other organisms.
We also want to point out that the equilibrium relation between FBP and PEP is most important during the middle of lag phases. However, in mid lag phase net fluxes are very small and the equilibrium is mainly determined by thermodynamics. Even if the PPP pathway or other pathways were to distort the equilibrium relation during lag phase, as long as the reversible enzymes connecting FBP and PEP via glycolysis are fast or high abundance, FBP and PEP would stay very close to their equilibrium relation.
During growth, the PPP and ED pathways are of course important for metabolism. During exponential growth of E. coli about 29% of flux from G6P goes through PPP pathway and 11% through ED (Gerosa et al., 2015, Cell Systems 1, 270-282, http://dx.doi.org/10.1016/j.cels.2015.09.008.). As the reviewer pointed out, some fraction of the metabolic flux from PPP will go directly to GAP (~6%), while the rest (~15%) is cycled back to F6P. Similarly, for the ED pathway flux will go to GAP and PYR. Overall, about 89% of the flux to GAP stems from F6P (Gerosa et al). Because in our model we do not discriminate how flux arrived at F6P, but only describe how it gets converted from F6P to GAP.
We added a note in the manuscript to reflect these comments. Neglecting these parallel pathways is one of many simplifications of the model, but we do not think that our main findings are affected by any of these simplifications. 3) The authors also suggest that their model predicts that there is a trade-off between the length of the lag-phase and growth rate of cells. Unfortunately, this prediction was also not tested experimentally. One way to evaluate this prediction would be to study E. coli strains that have been adaptively evolved to grow on different carbon sources, e.g. determine the length of lag-phases for strains that have been adaptively evolved to grow on glucose vs. strains that have been adaptively evolved to grow on acetate. What does the model predict that should be expected for such evolved strains? How well do the model predictions match with experimental observations?
In our paper we propose a total of 4 trade-offs 1. Between lag time to glycolysis and lag time to gluconeogenesis ( The reviewer is right that evolution experiments would be one way to test these trade-offs. Evolution experiments tend to be difficult, expensive and very time consuming. To be insightful such experiments must be repeated many times, spanning many different evolution conditions, e.g. different intervals between growth and switching to really map out a Pareto front and establish these tradeoffs. There are not many works in the literature that pass this bar and have actually managed to firmly establish tradeoffs. The experiments in the literature that is closest to what this would require is Lenski's long term evolution experiment. We actually think that the co-existence of glucose and acetate specialists in the Lenski experiment can be explained by the tradeoffs outlined in this work (fast growing glucose specialist with long lag time to acetate and a slower growing acetate specialist with short lag to acetate naturally), but a detailed analysis of these evolution experiments would be necessary and is beyond the scope of this work.
But even in an evolution experiment as expansive as Lenski's, the resulting strain is unlikely to be optimized for growth and switching across different growth rates. The required evolution experiment would therefore need not only flipping substrate preference, but also adaptation on diverse growth conditions, in order to show that the growth rate vs lag time relation reappears. 'Simply' evolving one strain for growth on acetate and measuring a lag phase to glucose would not be very conclusive.
So rather than doing an evolution experiment, we asked how evolution has solved this problem in the past. We found that Pseudomonas aeruginosa preferentially uses gluconeogenic substrates (and even grows faster than E. coli on glucose!), which gives us an ideal strain that is evolved for gluconeogenic substrates. In addition, we use the fact that bacteria grow at different rates on different substrates, presumably because they have evolved differently on these substrates. Both E. coli (Fig. 6D) and Pseudomonas aeruginosa (Fig. 6F) show increasing lag times with growth rate. We also found that another species, Pseudomonas putida, shows moderate growth and lag times in both direction (Fig. S8). Even for P. putida lag time still increases with growth rate. The fact that this relation exists for all species we tested, despite their different evolutionary adaptation is remarkable, and we think it is a strong confirmation that the trade-off between growth rate and lag time exists.
Finally, we want to address the last question of the reviewer, the hypothetical 'what would we expect for evolved strains and how well would model predictions match experimental observations'? Any evolution experiment will lead to a number of mutations. If we evolve for fast growth on acetate, we will not only see changes in metabolic enzymes, but also changes in the rest of the proteome. For example, expression of non-metabolic and non-biosynthetic proteins will likely decrease, because they decrease both growth rate and lag time (e.g. Balakrishnan et al (2021) https://doi.org/10.1101/2021.04.28.441780). Because of these global changes, that are unrelated to the trade-offs we describe, we would need to fit a new parameter set to the evolved strains. In addition, we expect that unless we evolve for a really long time and in multiple conditions, that growth rate and adaptation are not optimal in such an evolved strain. In fact, from our parameter screen (Fig. 5) we know that most parameter combinations are actually pretty bad in most conditions. Reviewer #3: The manuscript by Schink and Christodoulou et al. describes a kinetic model of growth shift between glycolysis and gluconeogenesis in bacteria. The aim is to understand the mechanisms determining the duration of lag phases in either direction (glycolysis->gluconeogenesis, gluconeogenesis->glycolysis), and the difference between these two as observed for E. coli. The authors use a simplified central carbon metabolic model by lumping together irreversible reactions and thereby focussing on the key network (pseudo-)nodes -upper and lower glycolysis -and associated (selected) regulatory links. The model is then parameterised using experimental data and used to explore the relationships between metabolite concentrations, protein abundances and flux shifts. The key finding is that the lag time for the shifts have trade-offs with growth rate, energy efficiency and increased lag time for shift in the opposite direction. Metabolomics data from two Pseudomonas sp. is used to demonstrate the trade-off.
Overall, the study provides a new insight into an interesting biochemical problem. While the Results section of the manuscript is generally clear, there are several instances of ambiguity, typos, and impreciseness in Title, Abstract, Introduction and Discussion. Some important biochemical considerations also seem to have been overlooked (details below) and the manuscript would therefore benefit from a thorough revision.
We thank the reviewer for these helpful suggestions for clarifying our writing and statements.
1. Title: the title is misleading as, a) substrate specialization has a very different meaning in ecology / biochemistry (preferential/exclusive utilization of a few substrates from a range of those simultaneously available in the environment) than what the authors study (i.e., switching to by-product utilization after using the primary substrate); b) "flux sensing" is neither evident in the presented data nor adding to the presented concepts, and thus I strongly recommend to avoid using this term. This is what we mean with the 'biochemical constraint in flux sensing'. Therefore, we think that flux-sensing is the appropriate term.
We changed the title to: Glycolysis/gluconeogenesis specialization in microbes is driven by biochemical constraints of dynamic flux sensing. (underscored reflect changes) 2. Abstract, Discussion: "....turning the cell 'blind' to...". In my opinion, this is a shallow metaphor that unnecessarily adds to the imprecise anthropomorphising of bacterial physiology. It is shallow because the model does not account for extracellular sensing and several other cellular mechanisms that respond to nutrient and metabolic changes.
Good point, we changed it to: "[…] rendering the cell unable to sense […]" Abstract: 3. "very different ways" -> fuzzy We reworded this sentence to : Central carbon metabolism is highly conserved across microbial species, but can catalyze very different pathways depending on 4. "different bacteria" -? How many, which?
Changed to: "Here, we study the dynamic re-organization of central metabolism after switches between the two major opposing pathway configurations of central carbon metabolism, glycolysis and gluconeogenesis in Escherichia coli, Pseudomonas aeruginosa and Pseudomonas putida." These two species is where the majority of the focus is on in this work. We also used P. putida see Fig. S8. We have removed this sentence.
11. Line 54: why "naturally"? In nature, it is often that some other species would utilize the by-products of primary fermentation.
Not every microbe lives in a microbial community, and in not all community fermentation products are all used up before the primary product runs out.
We added a qualifier: "This naturally leads to bi-phasic growth, if no other microbe is around to utilize this biproduct, where initial utilization of glucose is followed by a switch to acetate." 12. Line 71: "Why are..." -> the question is ill phrased as the model cannot show "why" but rather "how".
This is a matter of taste. For us, we want to address why-questions in biology even if it is very difficult to obtain definitive answers. The question we are trying to address is: "Why are microorganisms not able to overcome long lag phases using the existing allosteric and transcriptional regulation?" The answer is that trade-offs between lag times, growth rates and futile cycling prevent microbes to optimize all traits simultaneously 13. Line 103: "sense fluxes" -> what is meant by this? Is there really a physical way to directly sense fluxes that is applicable here? Metabolite level as a proxy for flux is a consequence of biochemical configuration and not sensing by the cell.
The term flux sensing is established in the field (e.g., Kochanowski et al https://doi.org/10.1073/pnas.1202582110) and we are using the term is in line with the previous literature. The network described by our model detects both the direction and the magnitude of flux and modulates enzyme activity and enzyme expression levels accordingly (see Fig. S1). For all practical purposes and in linne with previous works we call this kind of regulation "flux sensing". We think it would actually lead to confusion if we changed the term.
There are potentially many different mechanisms how cells can measure flux. The mechanism in our model is that metabolites can be used as a proxy for cells to sense flux. The metabolite FBP has been experimentally shown to reflect glycolytic flux and directly affect expression of dozens of genes.
14. Lines 107-8: "...in the absence of information" is inaccurate as no external sensing or other cellular mechanisms (e.g., translation response to low substrates/co-factors etc.) are accounted for.
We changed the text to: "This choice of direction at low metabolite concentrations becomes the 'default state' of central metabolism and determines the substrate preference." 15. Last Introduction paragraph -the argument if "could have evolved" is spot on but conflicts with the narrative above on "why" etc.
The point we are trying to get across is that glycolysis-specialists or gluconeogenesis-specialists could evolve. But why not both? Why are there no bacteria that can grow fast and switch fast on/to all substrates?
We believe that the trade-offs that we describe in our work give some good reasons for that. In the discussion section, we pick up this point again and discuss what kind of problems different regulatory architectures would have, and why we think that the trade-offs that we describe are really uncircumventable.
16. There are several typos/grammatical errors etc. throughout and for the sake of time I will not report them all here -most critical being Figure 5 is referred to as Figure 6 on page 17, and on page 13 end I think 'PEP' is meant in place of 'FBP'?
Thank you for pointing this out. We have proofread the text to try to root out remaining errors. To be transparent, we added a statement to every figure explaining how we generated data, where it is from and why on some figures error bars are missing. In addition, we removed the absolute scale on FBP and PEP (which we are not using in the model) in Fig. 1D and plotted the data on a relative scale. We this this reflects better that there is uncertainty in the quantitative data, and puts the spotlight on the qualitative changes.
18. I did not find any data/code availability statement. Is the data available in Metabolights? Is the Matlab code available to reproduce the results?
We now included a Data availability statement. We uploaded code to github and metabolomics data to Metabolights. 19. The model and the interpretation of the results overlooks the role of transporters and transport efficiency. Thus, statements like on line 263 ("must be") etc. are incorrect and Discussion needs to be carefully revised (or a new model built including transporters).
We thank the reviewer for raising this potential point of confusion. Transport processes are coarse-grained into the model as direct flux to the substrates of the irreversible reactions, either glycolytic carbons or TCA carbons. If there are insufficient transporters present, they will certainly slow down lag times. But note that even most transport systems of glycolytic systems are repressed in the absence of the carbon, but lag times are still very short (Fig. 1H). Thus, the long lag times we observe must be caused by something other than transport. For that reason, we coarse-grained transport into uptake of glycolytic/TCA carbons and focused on central metabolism as the potential culprit for long lag times instead.
To clarify the text, we added a sentence about transport in lines 151-153 and also added a qualifier statement: Line 263: "The absence of transient futile cycling, despite the symmetry of regulation and metabolic reactions, means that, according to the model, it must be the allosteric and transcriptional regulations that 'prime' central metabolism of E. coli for the glycolytic direction." 20. 488: "ATP production at relatively low proteome cost" -not entirely true as it assumes that higher proteome will lead to higher flux, the lack of this is exactly why the modelling is being done.
This statement was meant to say that proteome cost to supply the additional energy required for futile cycling is relatively modest, according to the available estimates for ATP production from glycolysis and the TCA cycle [Basan, M., Hui, S., Okano, H., Zhang, Z., Shen, Y., Williamson, J.R., and Hwa, T. (2015a). Overflow metabolism in Escherichia coli results from efficient proteome allocation. Nature 528.]. These estimates already included the proteome cost of uptake transporters.
Reviewer is right that only more proteins doesn't mean more ATP. In addition to proteome costs directly associated with ATP production, there is also a contribution from transporters and metabolic enzymes. We adapted the discussion to address this point: The energetic cost of such a wasteful strategy would be relatively low. Because energy production pathways only constitute a relatively small fraction (around 20% (Basan et al.,  2015)) of the total cellular proteome and nutrient uptake even smaller (around 1% for glucose uptake (Schmidt et al., 2016)), the cell could compensate ATP dissipated in futile cycling by increasing nutrient uptake and ATP production at a relatively low proteome cost. 21. 499: "what about enzyme parameters?
The trade-offs are not properties of specific enzyme parameters. We showed that they persist even if we screen a large number of parameters (Fig. 5).

500: could IN THEORY use
We changed the sentence to: In theory, the cell could use a direct input signal from the carbon substrate to allosterically inhibit or even degrade undesired metabolic enzymes.
We scanned parameter space for a given regulatory architecture and demonstrated the tradeoffs (see Fig. 5). Because the trade-offs are only based on the regulatory architecture, we would like to leave this unchanged.
24: "conservation of the phenomenology of shifts.." -> why?? This is not striking at all -what else one can expect?
There are plenty of alternative outcomes that could in principle occur. There is a priori no reason to expect the same phenomenology in totally different metabolic shifts.
For example, it might be that all organisms lag long in the same shifts and show similar growth rates on the same substrates, simply because these substrates are 'good' or 'bad'. But this isn't the case. Different organisms show completely opposite patterns of substrate preference and lag times (see Fig. 6 where we compare E. coli to P. aeruginosa). This means that inherent quality of nutrients is not responsible for these differences.
We believe that the fact that E. coli, B. subtilis, as well as S. cerevisiae all show long lag times only in shifts to gluconeogenesis, but short lag in the other direction is already striking. Why has none of these diverse organisms been able to overcome the problem if such regulation is easy to evolve? Is it a fundamental biochemical property of gluconeogenesis substrates like acetate that makes it unavoidable for all organisms? The answer is no, because other bacteria have overcome this problem, e.g., P. aeruginosa switches almost immediately to gluconeogenesis. But as we show with the model, this comes at the cost of lag phases in the opposite direction and the same trade-off only in the opposite direction. We think that it is remarkable that the P. aeruginosa shows precisely the reverse phenomenology of E. coli.
The fact that lag times increase with growth rate for all organisms, independently of their specialization, is another remarkable phenomenon in our opinion.
25. The last concluding sentence "environments, ecology and evolutionary origin" is not supported by the results here.
We have changed the formulation from "we argue" to "we believe". This should clarify that this is our opinion. We think articulating such an opinion is acceptable in a discussion paragraph.
Conversely, we believe that the quantitative phenotypes exhibited by microbes in such idealized growth shift experiments in the lab can reveal much about their natural environments, ecology and evolutionary origin.

9th Dec 2021 1st Revision -Editorial Decision
Thank you for sending us your revised manuscript. We think that the performed revisions have satisfactorily addressed the issues raised by the reviewers. As such, I am glad to inform you that we can soon formally accept the study for publication, pending some minor editorial issues listed below. We would ask you to address these issues in a minor revision. Do the data meet the assumptions of the tests (e.g., normal distribution)? Describe any methods used to assess it.
Is there an estimate of variation within each group of data?

Reporting Checklist For Life Sciences Articles (Rev. June 2017)
This checklist is used to ensure good reporting standards and to improve the reproducibility of published results. These guidelines are consistent with the Principles and Guidelines for Reporting Preclinical Research issued by the NIH in 2014. Please follow the journal's authorship guidelines in preparing your manuscript.

B-Statistics and general methods
the assay(s) and method(s) used to carry out the reported observations and measurements an explicit mention of the biological and chemical entity(ies) that are being measured. an explicit mention of the biological and chemical entity(ies) that are altered/varied/perturbed in a controlled manner.
a statement of how many times the experiment shown was independently replicated in the laboratory.
Any descriptions too long for the figure legend should be included in the methods section and/or with the source data.
In the pink boxes below, please ensure that the answers to the following questions are reported in the manuscript itself. Every question should be answered. If the question is not relevant to your research, please write NA (non applicable). We encourage you to include a specific subsection in the methods section for statistics, reagents, animal models and human subjects.
definitions of statistical methods and measures: a description of the sample collection allowing the reader to understand whether the samples represent technical or biological replicates (including how many animals, litters, cultures, etc.).
The data shown in figures should satisfy the following conditions: Source Data should be included to report the data underlying graphs. Please follow the guidelines set out in the author ship guidelines on Data Presentation.
Please fill out these boxes ê (Do not worry if you cannot see all your text once you press return) a specification of the experimental system investigated (eg cell line, species name).
Sample size of all experiments was chosen to be 3.

Data
the data were obtained and processed according to the field's best practice and are presented to reflect the results of the experiments in an accurate and unbiased manner. figure panels include only data points, measurements or observations that can be compared to each other in a scientifically meaningful way.