1. Computational and Systems Biology
Download icon

Control and regulation of acetate overflow in Escherichia coli

  1. Pierre Millard  Is a corresponding author
  2. Brice Enjalbert
  3. Sandrine Uttenweiler-Joseph
  4. Jean-Charles Portais
  5. Fabien Létisse
  1. TBI, Université de Toulouse, CNRS, INRAE, INSA, France
  2. MetaToul-MetaboHUB, National Infrastructure of Metabolomics and Fluxomics, France
  3. RESTORE, Université de Toulouse, INSERM U1031, CNRS 5070, Université Toulouse III - Paul Sabatier, EFS, France
  4. Université Toulouse III - Paul Sabatier, France
Research Article
  • Cited 0
  • Views 496
  • Annotations
Cite this article as: eLife 2021;10:e63661 doi: 10.7554/eLife.63661

Abstract

Overflow metabolism refers to the production of seemingly wasteful by-products by cells during growth on glucose even when oxygen is abundant. Two theories have been proposed to explain acetate overflow in Escherichia coli – global control of the central metabolism and local control of the acetate pathway – but neither accounts for all observations. Here, we develop a kinetic model of E. coli metabolism that quantitatively accounts for observed behaviours and successfully predicts the response of E. coli to new perturbations. We reconcile these theories and clarify the origin, control, and regulation of the acetate flux. We also find that, in turns, acetate regulates glucose metabolism by coordinating the expression of glycolytic and TCA genes. Acetate should not be considered a wasteful end-product since it is also a co-substrate and a global regulator of glucose metabolism in E. coli. This has broad implications for our understanding of overflow metabolism.

Introduction

Overflow metabolism refers to the production of seemingly wasteful by-products by cells during growth on glycolytic substrates such as glucose, even when oxygen is abundant. Overflow metabolism has been reported for most (micro)organisms: yeasts produce ethanol (a phenomenon known as the Crabtree effect; Crabtree, 1929), mammalian cells produce lactate (the Warburg effect in cancer cells; Warburg, 1956), and bacteria produce small organic acids such as acetate (Harden, 1901). This ubiquitous phenomenon has been extensively investigated because of its fundamental and applied importance, but its origin and regulation remain to be clarified.

Acetate production by Escherichia coli has been studied for decades as a model of overflow metabolism. The recent rise of systems biology, which combines experimental techniques and mathematical modelling to achieve a mechanistic understanding of processes underlying the function of biological systems (Bruggeman et al., 2007Shou et al., 2015), has provided a quantitative understanding of the cause of acetate overflow in E. coli. Acetate production is considered to result from an imbalance between E. coli’s capacities to produce and assimilate acetyl-CoA, the main precursor of acetate. Stoichiometric models suggest that this imbalance could be caused by various cell-level constraints, such as energy conservation (de Groot et al., 2020; El-Mansi and Holms, 1989), recycling of cofactors (Wolfe, 2005; Vemuri et al., 2006), membrane occupancy (Zhuang et al., 2011; Szenk et al., 2017), and resource allocation (Basan et al., 2015; Zeng and Yang, 2019). These models – and the underlying theories – capture the essence of the overflow process and successfully predict the growth rate dependence of acetate production in E. coli (Basan et al., 2015; Zeng and Yang, 2019; Renilla et al., 2012).

However, none of these theories explain why increases in extracellular acetate concentration can interrupt or even reverse the acetate flux in E. coli (Enjalbert et al., 2017), which is able to co-consume glucose and acetate even when glucose is abundant. Cell growth is not affected when acetate production is abolished (Enjalbert et al., 2017; Lee and Liao, 2008), suggesting that this overflow is neither necessary for E. coli to sustain fast growth nor strictly imposed by the proposed constraints. This phenomenon cannot be reproduced by stoichiometric models since they do not account for metabolite concentrations. Kinetic models are therefore a necessary step towards a comprehensive understanding of the dynamics, control, and regulation of metabolic systems. For example, a kinetic model of the Pta-AckA pathway has successfully been used to predict the reversal of the acetate flux at high acetate concentrations (Enjalbert et al., 2017), a significant advance compared to stoichiometric models of acetate overflow. This effect is caused by thermodynamic control of the acetate pathway by the concentration of acetate itself, a mechanism that operates independently of enzyme expression. Still, thermodynamic control does not imply that enzymes exert no control whatsoever over acetate flux, though the degree of this control remains to be clarified. Moreover, this local mechanism does not explain the toxic effect of acetate on microbial growth (Wilbanks and Trinh, 2017; Luli and Strohl, 1990).

In this work, we used a systems biology approach to clarify the origin, control, and regulation of acetate overflow in E. coli over the broad range of acetate concentrations this bacterium experiences under laboratory and industrial conditions as well as in its environmental niche (Wolfe, 2005; Shen et al., 2012; de Graaf et al., 2010; Kleman and Strohl, 1994; Jones et al., 2007; Cummings et al., 2001; Macfarlane et al., 1992; Cummings and Englyst, 1987; Fabich et al., 2008).

Results

Construction of a kinetic model of E. coli metabolism

Following a top-down systems biology approach (Bruggeman et al., 2007), we constructed a coarse-grained kinetic model of E. coli metabolism that links acetate metabolism with glucose uptake and growth (Figure 1A). This model includes three processes: (i) glucose transport and its conversion into acetyl-CoA by the glycolytic pathway, (ii) utilisation of acetyl-CoA in the TCA cycle (energy conservation) and anabolic pathways (production of building blocks) for growth, and (iii) acetate metabolism, i.e., the reversible conversion of acetyl-CoA into acetate via phosphotransacetylase (Pta), acetate kinase (AckA), and an acetate exchange reaction between the cell and its environment.

Figure 1 with 3 supplements see all
Representation of glucose and acetate metabolism in Escherichia coli (A), in Systems Biology Graphical Notation format (http://sbgn.org) (Le Novère et al., 2009).

We performed 13C-labelling experiments to calibrate the model and evaluated the goodness of fit for different topologies (B). The initial model (model 1), which does not include inhibition of the glycolytic pathway and TCA cycle by acetate, did not fit the data satisfactorily. Adding inhibition by acetate of glycolysis (model 2) or of the TCA cycle (model 3) improved the fit, but both pathways had to be inhibited (model 4) for the goodness-of-fit criterion to be satisfied. In (B), the horizontal line represents the 95% confidence threshold for the variance-weighted sum of squared residuals (SSR). The best fits of the experimental data obtained with model 4 are shown in (C), where the shaded areas represent the 95% confidence interval on the fits. The best fits obtained with the alternative models (models 1–3) are shown in Figure 1—figure supplements 13.

Figure 1—source data 1

Experimental data used to calibrate the model.

https://cdn.elifesciences.org/articles/63661/elife-63661-fig1-data1-v2.xlsx

This initial model included two compartments, 6 species, 6 reactions, and 24 parameters. The value of 14 parameters were taken directly from the literature (Enjalbert et al., 2017; Millard et al., 2017; Kadir et al., 2010). To estimate the remaining parameters, we performed three growth experiments on 13C-glucose (15 mM) plus different concentrations of 12C-acetate (1, 10, and 30 mM), as detailed in Enjalbert et al., 2017. These experiments were designed to demonstrate that E. coli simultaneously produces acetate from glucose and consumes it from the environment (Enjalbert et al., 2017). They provide information on the forward and reverse fluxes of acetate between the cell and its environment (Enjalbert et al., 2017), and thereby improve the calibration of the model. The model was extended with isotopic equations to account for the propagation of 13C label (Enjalbert et al., 2017; Millard et al., 2015), and parameters were estimated by fitting concentration time courses of glucose, biomass, and acetate and 13C-enrichment of acetate (see Materials and methods). This initial model (model 1) did not fit the data satisfactorily (Figure 1B, Figure 1—figure supplement 1, Materials and methods). Adding inhibition of the glycolytic pathway by acetate (model 2) to account for the reduction in glucose uptake at high acetate concentrations improved the fits of the biomass, glucose, and acetate concentration profiles but not of the acetate labelling profiles (Figure 1—figure supplement 2). Similarly, adding inhibition of the TCA cycle by acetate (model 3) slightly improved the fit of the acetate labelling profile but not of the glucose and biomass concentration profiles (Figure 1—figure supplement 3). A sufficiently accurate fit was only achieved when both pathways were inhibited (model 4; Figure 1B,C). These modifications suggest the existence of an unknown regulatory program that makes E. coli sensitive and responsive to acetate concentration, and provide a mechanistic rationale for the reported ‘toxicity’ of acetate to E. coli (Enjalbert et al., 2017; Wilbanks and Trinh, 2017; Luli and Strohl, 1990; Pinhal et al., 2019).

Acetate regulates glucose uptake, glycolysis, and the TCA cycle at the transcriptional level

A key hypothesis required for the model to fit the experimental data is that acetate inhibits the flux capacity of both the glycolytic pathway and the TCA cycle in E. coli. In this model, the aim of the simplified description of inhibition by acetate was to represent the integrated response of E. coli metabolism to acetate, with no a priori knowledge of the underlying molecular mechanism(s). To determine whether this inhibition actually occurs in vivo at the transcriptional level, we monitored gene expression in E. coli grown on glucose (15 mM) plus acetate at different concentrations (0, 10, 50, or 100 mM).

Transcriptomic results revealed a global and progressive remodelling of gene expression at increasing acetate concentrations (Figure 2A–B). We noted that the presence of acetate modulates the expression of genes involved in various biological functions: motility, biofilm formation, translation, stress response, metabolism, transport of carbohydrates, amino acids, and ions (Figure 2C). Importantly, the observed changes in gene expression cannot result solely from the inhibition of growth by acetate, as shown by comparing gene expression levels on glucose plus 100 mM acetate with those on glucose alone at the same growth rate (0.35 hr−1) (Figure 2B).

Response of the E. coli transcriptome to changes in acetate concentration (0, 10, 50, or 100 mM) during growth on glucose (15 mM).

The changes in gene expression are shown in (A). Each line represents the expression of a single gene relative to its expression level measured in the absence of acetate. Up- and downregulated genes are shown in red and green, respectively. The Venn diagrams (B) represent the total number of genes upregulated (left) and downregulated (right) by at least a factor of 2 under each condition and during growth on glucose in the absence of acetate but at the same growth rate as in the presence of 100 mM acetate (0.35 hr−1, extrapolated from the data from Esquerré et al., 2014). Biological functions modulated by the presence of acetate (based on Gene Ontology analysis) are shown in (C), with the corresponding p-values. The expression levels of central metabolic genes are shown in (D). The data were obtained from four independent biological replicates for each condition.

Figure 2—source data 1

Response of the E. coli transcriptome to changes in acetate concentration (0, 10, 50, or 100 mM) during growth on glucose (15 mM).

https://cdn.elifesciences.org/articles/63661/elife-63661-fig2-data1-v2.xlsx

At the level of the central metabolism (Figure 2D), acetate reduces the expression of all genes that code for the glucose phosphotransferase system (PTS) (ptsGHI, crr), without inducing the expression of alternative systems of glucose internalisation and phosphorylation (galP, mglABC, glf, glk, manXYZ) that could have compensated for reduced PTS activity and the corresponding inhibition of glucose uptake (Lim and Jung, 2017; Ruyter et al., 1991). Expression of upper glycolytic genes remained stable, with the exception of two isoenzymes that were overexpressed in the presence of acetate: fbaB, a gluconeogenic enzyme, and pfkB, which contributes little to phosphofructokinase activity on glucose (<10%) (Fraenkel, 1986; Scamuffa and Caprioli, 1980; Long and Antoniewicz, 2019; Daldal et al., 1982). In contrast, the expression of most of the lower glycolysis genes (pgk, gapA, gpmA, eno, pykF, aceE) was reduced by 15–40% at 100 mM acetate. At this concentration, acetate also inhibits the expression of virtually all TCA cycle genes (gltA, acnAB, icd, lpd, sucABCD, sdhABCD, fumABC, mdh) by 30–67%. In terms of acetate metabolism, the expression of pta and ackA, which code for enzymes in the Pta-AckA pathway, remained remarkably stable at all acetate concentrations. Expression of pyruvate oxidase (poxB) was increased by a factor of 8 in the presence of acetate, though this promiscuous enzyme contributes to acetate metabolism mainly in the stationary phase and apparently not under our conditions (Enjalbert et al., 2017; Pinhal et al., 2019; Martínez-Gómez et al., 2012; Dittrich et al., 2005). Expression of acetyl-CoA synthetase (acs) – which converts acetate to acetyl-CoA with a high affinity – decreased with increasing acetate concentration, indicating that under glucose excess, the presence of acetate does not activate acetate recycling through the Pta-AckA-Acs cycle (Enjalbert et al., 2017; Valgepea et al., 2010).

These data confirm our hypothesis that acetate gradually modulates both the flux capacity of E. coli to produce acetyl-CoA from glucose and its capacity to utilise acetyl-CoA as a source of energy and of anabolic precursors for growth. Acetate does not appear to influence the flux capacity of the acetate pathway itself, which is also consistent with the proposed model.

Testing the model

We tested the model by predicting the response of E. coli to new perturbations and comparing the results with experimental data (Figure 3).

Figure 3 with 4 supplements see all
Comparison of model predictions with experimental data.

We used the model to simulate (i) steady-state glucose and acetate fluxes in glucose-limited chemostat cultures at dilution rates of 0.1–0.5 hr−1 (A), (ii) the growth rates and glucose and acetate fluxes during growth on glucose at various acetate concentrations (B), and (iii) the time courses of the changes in glucose and acetate concentrations during exponential growth on glucose after a pulse of either acetate or water (C). Model predictions are represented by lines and experimental data are shown as dots (the error bars represent one standard deviation), the shaded areas represent the 95% confidence intervals on the predictions. Predictions obtained with the alternative models (models 1–3) are shown in Figure 3—figure supplements 13. The predictive accuracy was compared between models based on the variance-weighted sum of squared residuals between simulated and experimental data (Figure 3—figure supplements 14).

First, we checked that the model could reproduce the established growth-rate dependence of acetate production in E. coli. We modified the model to simulate glucose-limited chemostat experiments (by adding reactions for glucose feed and medium outflow), and we predicted the steady-state glucose and acetate fluxes at different dilution rates (from 0.1 to 0.5 hr−1). Since the present model cannot account for the activation of acetyl-CoA synthetase – involved in acetate assimilation – under glucose limitation (Valgepea et al., 2010), model simulations were compared to experimental data collected on a Δacs strain (Renilla et al., 2012). The simulated profiles of acetate production as a function of growth and glucose uptake rates were in excellent agreement with experimental data (Figure 3A), indicating that the model accurately captures the effects of glucose limitation on growth and acetate fluxes.

Second, we evaluated the response of E. coli to changes in acetate concentration. We simulated steady-state acetate, glucose, and growth fluxes under glucose excess over a broad range of acetate concentrations (between 100 µM and 100 mM) (Enjalbert et al., 2017). The growth and glucose uptake rates decreased monotonously with increasing acetate concentrations, in agreement with the experimental data (Figure 3B). The acetate flux profile was also well described by the model, with a progressive decrease of acetate production with increasing acetate concentration, and a reversal of the acetate flux above a threshold concentration of about 10 mM. As observed experimentally, the growth rate was not affected when acetate production was abolished, confirming that acetate production is not required to maintain fast growth nor is it imposed by any intracellular constraint.

Third, we tested the dynamic properties of the model by predicting the response of E. coli to sudden variations in the acetate concentration, which is not possible using existing models. In the exponential growth phase on glucose, we simulated the time courses of the changes in the concentrations of glucose and acetate after a pulse of acetate (30 mM) or of water (control experiment), as described in Enjalbert et al., 2017. Here again, the model accurately reproduces the experimental profiles, in particular the rapid reversal of the acetate flux and reduction of glucose uptake after the acetate pulse (Figure 3C).

Overall, the predictions of the model are in agreement with experimental findings. The model accurately predicts the steady-state and dynamic relationships between glucose uptake, growth, and acetate fluxes in E. coli over a broad range of glucose and acetate concentrations. Importantly, the model was not trained on these data (i.e., they were not used to calibrate the model), so this test is a strong validation of its predictive power.

To assess the functional importance of regulation by acetate, we predicted the response of E. coli to the same perturbations with alternative models (i.e., without inhibition of the glycolytic and/or TCA cycle pathways, models 1–3). The simulations in the absence of inhibition are qualitatively different (Figure 3—figure supplements 13). For instance, model 1 predicts a constant glucose uptake flux at all acetate concentrations, and even, contrary to observations, an increase in the growth rate with the acetate concentration. We compared the predictive accuracy of all the models by calculating the variance-weighted sum of squared residuals between the simulated and experimental data (Figure 3—figure supplement 4). The most accurate predictions for all perturbations are those from the model with dual inhibition (model 4), followed by those from the single-inhibition models (models 2 and 3) and finally those produced by the no-inhibition model (model 1). The predicted data are thus only qualitatively and quantitatively consistent with observations when inhibition by acetate is included, indicating that the regulatory role of acetate in E. coli is functionally important.

Intracellular control of acetate flux is distributed around the acetyl-CoA node

We used a metabolic control analysis of this kinetic model to determine the degree of control exerted by each reaction on acetate flux (Figure 4). Flux control coefficients (CEJ) quantify the impact of small changes in the rate of each reaction (typically from a change in the enzyme concentration E) on each flux (J) (Kacser and Burns, 1973; Heinrich and Rapoport, 1974).

Heatmap of flux control coefficients during growth on glucose (15 mM) and acetate (0.1 mM).

Each column represents a controlling reaction (E) and each row, a flux (J). Red and blue cells represent negative and positive flux control coefficients (CEJ), respectively, with darker (lighter) tones indicating stronger (weaker) control.

Metabolic control analysis revealed that rather than being controlled entirely from within the acetate pathway, acetate flux is controlled to a similar degree by the acetate pathway (with a flux control coefficient of 0.76), glycolysis (0.88), and the TCA cycle (−0.64). The balance between acetyl-CoA production and demand (i.e., between glycolytic and TCA fluxes), therefore has a strong effect on acetate production. As expected, this control is positive for glycolysis (since it produces acetyl-CoA) and negative for the TCA cycle (since it consumes acetyl-CoA). Still, acetate flux is controlled to a large extent from within the acetate pathway, mainly by AckA (0.69), with a small contribution from acetate transport (0.07). Overall, acetate flux control is distributed around the acetyl-CoA node.

Intracellular flux control patterns shift with the acetate concentration

Since control of acetate flux is distributed around the acetyl-CoA node, we tested how flux changes around this node might affect its control properties. Given that the concentration of acetate is a major determinant of acetate flux (Figure 3), we performed metabolic control analyses for a broad range of acetate concentrations (from 0.1 to 100 mM).

The control exerted by the acetate pathway on acetate flux decreases non-linearly as the acetate concentration increases (Figure 5A), highlighting a progressive shift of the control from inside to outside the acetate pathway. This is contrary to the classical behaviour in which the control exerted by a metabolic pathway increases as it becomes saturated, i.e., where the flux reaches the capacity of the uptake pathway at high substrate concentrations (Wegner et al., 2015; Moreno-Sánchez et al., 2008). The control exerted by the acetate pathway gradually shifts towards the glycolytic and TCA pathways (Figure 5B,C). Discontinuous control patterns are observed for the latter pathways at a concentration of about 10 mM, i.e., around the concentration threshold at which the acetate flux switches from production to consumption and is zero (Enjalbert et al., 2017). The sum of the control coefficients of the glycolytic and TCA blocks, which represent the overall control exerted by E. coli metabolism on the acetate flux (Figure 5D), increases with the acetate concentration and compensates for the decrease in control from the acetate pathway. The ratio of the control coefficients of glycolysis and the TCA cycle show that control shifts from the glycolytic pathway to the TCA cycle as the acetate concentration increases (Figure 5E).

Figure 5 with 3 supplements see all
Control of acetate flux over a broad range of acetate concentrations.

The shaded areas represent the 95% confidence intervals. Flux control coefficients calculated with the alternative models (models 1–3) are shown in Figure 5—figure supplements 13.

Overall, these results reveal how intracellular flux control patterns are strongly modulated by the extracellular concentration of acetate, with the control exerted by the acetate pathway being transferred progressively to the glycolytic and TCA pathways as the acetate concentration increases.

Here again, we tested the impact of regulation by acetate on the control properties of E. coli by comparing the results obtained with different models (Figure 5—figure supplements 13). The individual control profiles of the glycolytic and TCA pathways are qualitatively similar in all the models. However, the control profiles of the acetate pathway differ drastically in the absence of inhibition (model 1) and with partial inhibition (models 2 and 3), from the results with dual inhibition (model 4). In models 1–3, the control exerted by the acetate pathway remains constant at low to moderate acetate concentrations and tends to increase at high acetate concentrations, indicating that the progressive transfer of control from the acetate pathway to the rest of the metabolism as a function of the acetate concentration emerges specifically from the inhibition by acetate of both the glycolytic and the TCA cycle pathways.

Regulation of E. coli metabolism by acetate

Metabolic control analysis identified the controlling steps, i.e., the reactions that alter the acetate flux if their rates are modified (e.g., through changes in enzyme concentrations). However, more information is needed to understand the mechanisms involved in flux responses to perturbations of an external parameter such as the acetate concentration. Indeed, while a metabolic reaction may exert some control on a given flux, this does not necessarily mean that it is involved in the observed flux response. For instance, AckA exerts some control on the acetate flux, but its expression appears to be constant at all acetate concentrations (Figure 2), hence it does not regulate the acetate flux at the transcriptional level.

To identify the regulatory routes that are actually involved in the response of E. coli to changes in acetate concentration, we used the concept of response coefficients (RAceJace), which express the dependence of a system variable (here the acetate flux, Jace) on an external effector (the concentration of acetate). The partitioned response relationship (Kacser and Burns, 1973; Cornish-Bowden, 1995) allows the flux response to a perturbation in acetate concentration channelled through a given reaction i (viRAceJace) to be quantified, as detailed in the Materials and methods section. Since acetate regulates more than one reaction, the partitioned response coefficients provide a quantitative understanding of the different routes through which the acetate flux is regulated by the acetate concentration. We calculated the partitioned response coefficients of the acetate flux to the acetate concentration via (i) the acetate pathway, which represents the contribution of direct metabolic regulation, and via (ii) the glycolytic and (iii) the TCA pathways, where acetate acts indirectly by modulating the flux capacity (Figure 6A). As expected, regulation is minimal when acetate does not significantly modulate the acetate flux, i.e., at low and high acetate concentrations, and is maximal at concentrations that strongly modulate its flux (Figure 6B–D). Setting the response threshold of 0.1 – i.e., that a relative change of x % in the acetate concentration should lead to a relative flux response of at least (0.1×x) % via the regulatory route considered – we determined that acetate acts as a regulator over a range of concentrations spanning three orders of magnitude (between 0.2 and 100 mM).

Regulation of acetate flux in Escherichia coli.

The different routes through which acetate flux can be regulated by the acetate concentration are shown in (A). Dotted lines represent indirect (hierarchical) regulation, and straight lines represent direct (metabolic) regulation. The strengths of the three regulatory routes are respectively shown in (B–D), and their relative contributions are shown in (E). The shaded areas represent the 95% confidence intervals.

Based on the regulatory strength of each interaction, we determined the relative contribution of each interaction to the observed flux response (Figure 6E), revealing two distinct modes of regulation that depend on the acetate concentration and remain remarkably stable over a broad range of concentrations. At low acetate concentrations (<10 mM), the main regulatory route (50–60%) is direct metabolic regulation via the acetate pathway, with the remaining flux response being accounted for by indirect regulation through the TCA cycle (35–40%) and glycolysis (5–10%). This regulatory pattern remains stable before changing abruptly at ~10 mM when the acetate flux reverses. Above this threshold, direct regulation by acetate accounts for about 40% of the observed flux response, glycolysis, about 20%, and the TCA cycle, roughly 40%. This pattern then remains stable up to 100 mM.

These results provide a comprehensive and quantitative understanding of acetate flux regulation by acetate through the coordinated effects of direct and indirect mechanisms, with distinct regulatory programs at low and high acetate concentrations.

Discussion

Two independent theories have been proposed to explain acetate overflow in E. coli, but neither global regulation of central metabolic pathways nor local control of the acetate pathway can account for all observations. Using a systems biology strategy, we reconcile these two theories, clarify how acetate flux is controlled and regulated, and establish the pivotal role acetate plays as a global regulator of glucose metabolism in E. coli.

The kinetic model of E. coli metabolism developed in this work links acetate metabolism with glucose uptake and growth. This model suggested the existence of a global regulatory program whereby the extracellular concentration of acetate determines the flux capacity of the glycolytic and TCA pathways, whose inhibition was required to explain the dynamic profiles obtained in 13C-labelling experiments. This inhibition was also necessary (and sufficient) to predict the steady-state and dynamic responses of E. coli to a broad range of perturbations. In this coarse-grained model, the simplified formalism used to consider acetate inhibition represents the integrated regulatory response of E. coli metabolism to acetate, regardless of the underlying molecular mechanisms. Regulation may indeed take place through a variety of mechanisms acting at the (post-)transcriptional, (post-)translational, and metabolic levels. These regulatory networks are tightly intertwined and act concertedly to ensure a coordinated response of E. coli to perturbations on the seconds to minutes time scale (Enjalbert et al., 2013Enjalbert et al., 2017; Chubukov et al., 2014; Lempp et al., 2019; von Wulffen et al., 2017; Gerosa and Sauer, 2011). Transcriptomics experiments confirmed that acetate regulates the flux capacity of these two pathways at the transcriptional level by modulating the expression of most of their genes. This is a highly efficient way for cells to adjust their fluxes while maintaining metabolite homeostasis (Kacser and Burns, 1973; Heinrich and Rapoport, 1974). The metabolic response of E. coli to changes in acetate concentration thus involves a global reorganisation of its metabolism from the transcriptional to the flux levels, contrary to what has been suggested (Pinhal et al., 2019) and providing a mechanistic rationale for the reported ‘toxicity’ of acetate. The coordinated regulation of the glycolytic and TCA pathways cannot be explained by the actions of known transcriptional regulators, suggesting that these regulators act in concert or that there are additional regulators present.

Transcriptional regulation occurs in a few minutes (Enjalbert et al., 2013; Lempp et al., 2019; von Wulffen et al., 2017; Morin et al., 2020), which is sufficiently fast to explain most observations (Figures 1 and 3A,B). However, the immediate decrease in glucose uptake flux in response to a pulse of acetate (Figure 3C) suggests the existence of additional regulatory mechanisms that act on a faster time scale. These mechanisms could operate at the post-transcriptional level (possibly via the carbon storage regulator system [Revelles et al., 2013] or the BarA/UvrY two-component system [Chavez et al., 2010]), at the post-translational level (possibly via protein acetylation by acetyl-phosphate [Weinert et al., 2013; Ren et al., 2019]) and/or directly at the metabolic level itself (possibly via α-ketoglutarate [Doucette et al., 2011] or ATP [Koebmann et al., 2002], which control the glycolytic flux). Further work will be required to identify the complete regulatory network of sense and response to acetate in E. coli. This work may be facilitated by the coarse grained model developed here as well as by more detailed kinetic models of E. coli metabolism (Millard et al., 2017; Kadir et al., 2010; Jahan et al., 2016). These models should help to computationally evaluate the biological relevance of detailed regulatory mechanisms, as illustrated here.

Our results confirm that the direction of the acetate flux is controlled thermodynamically and is fundamentally determined by the extracellular concentration of acetate. Moreover, we found that intracellular control of the acetate flux is distributed around the acetyl-CoA node, i.e., between the glycolytic, TCA, and acetate pathways, such that none of the three processes can be regarded as the sole kinetic bottleneck. The inhibition of the glycolytic and TCA pathways by acetate gives rise to unusual control properties. The intracellular flux control pattern is indeed determined by the concentration of acetate itself, with a progressive shift of control from the acetate pathway to the rest of the metabolism when the acetate concentration increases. This means that metabolic engineering interventions aiming at minimising acetate production in biotechnology should rebalance the flux capacity of both the glycolytic and the TCA pathways, possibly via global metabolic regulators, while accounting for the concentration-dependent effect of acetate on intracellular control patterns. The proposed model may thus be used to guide rational design of optimised strains for synthetic biology applications.

This study also identifies the regulatory scheme involved in E. coli’s response to changes in acetate concentration, which largely determines the acetate flux itself as well as the cell’s physiology. We demonstrate that this scheme involves direct metabolic regulation of the acetate pathway in combination with indirect – at least partly transcriptional – regulation of the glycolytic and TCA pathways, and we show how this equal contribution of direct and indirect regulatory mechanisms explains the diverse patterns of acetate flux observed and forms part of E. coli’s global physiological response to changing acetate concentrations. The distinct regulatory patterns at high and low acetate concentrations also determine the metabolic status of acetate as a co-substrate or by-product of glucose metabolism. This work shows that the regulation of acetate flux is far more subtle than previously considered.

These results call for a reconsideration of the role of acetate, and more generally of overflow metabolism, in E. coli. Acetate should not be considered a terminal, wasteful product of glucose metabolism since it may equally be a co-substrate of glucose. Moreover, it is also a key regulator that induces a global remodelling of E. coli physiology and modulates several cellular functions (motility, biofilm formation, transport, metabolism, translation). Acetate regulates E. coli metabolism over a range of concentrations spanning three orders of magnitude, and its regulatory action involves a combination of direct and indirect mechanisms. In the laboratory, E. coli is typically grown on glucose at low acetate concentrations, which enhances its production. In the intestine however, E. coli’s environmental niche, the concentration of acetate is high (between 30 and 100 mM) (Macfarlane et al., 1992; Cummings and Englyst, 1987) and well above the ~10 mM threshold at which the acetate flux reverses. These levels of acetate should thus favour its co-utilisation with available sugars, suggesting that acetogenic species may be an important source of nutrients for E. coli. Besides its importance in terms of understanding cross-feeding relationships, this work also stresses the need to further investigate the regulatory role of acetate (and of acetogenic species) in the intestine. This will require a quantitative characterisation of the available nutrients and of their dynamics in this particularly complex environment.

Microorganisms other than E. coli are similarly capable of co-consuming glucose with its metabolic by-products, even when glucose is abundant. Saccharomyces cerevisiae can co-consume ethanol and glucose (Raamsdonk et al., 2001), and mammalian cells co-assimilate lactate with glucose (Hui et al., 2020; Hui et al., 2017; Rabinowitz and Enerbäck, 2020). These similarities, which have important implications for cell energy and redox balance, indicate that overflow metabolism should be considered a reversible process and point to the existence of a universal phenomenon similar to the one described here for acetate in E. coli. The quantitative systems biology approach developed in this work can be used as a guide for future investigations of overflow fluxes, their regulation, and their biological implications in E. coli and other (micro)organisms.

Materials and methods

Strain and conditions

Request a detailed protocol

E. coli K-12 MG1655 was grown in M9 medium (Nicolas et al., 2007) complemented with 15 mM glucose. Sodium acetate (prepared in a 1.6 M solution at pH 7.0) was added up to the required concentration. The cells were grown in shake flasks at 37°C and 200 rpm, in 50 mL of medium. For the isotope labelling experiments, unlabelled glucose was replaced by U-13C6-glucose (Eurisotop, France). Growth was monitored by measuring the optical density (OD) at 600 nm using a Genesys 6 spectrophotometer (Thermo, USA), and a conversion factor of 0.37 gDW/L/OD unit (Revelles et al., 2013) was used to determine the biomass concentration.

Transcriptomics experiments

Request a detailed protocol

Cells were grown in flasks in M9 minimal media with 15 mM glucose and 0, 10, 50, or 100 mM acetate. In mid-exponential growth phase (OD600nm = 1), 4 mL of each culture was centrifuged for 90 s at 14,000 rpm, the supernatant was discarded, and the pellets were immediately frozen in liquid nitrogen. Total RNA was extracted using a Qiagen RNAeasy MiniKit and quantified using a Nanodrop spectrophotometer. Double-stranded complementary DNA (cDNA) synthesis and array processing were performed by One-Color Microarray-Based gene Expression Analysis (Agilent Technologies). The images were analysed with the software DEVA (v1.2.1). All array procedures were performed using the GeT-Biopuces platform (http://get.genotoul.fr). Four independent biological replicates were analysed for each condition. For each data set, the log2 intensities obtained in the presence of acetate were divided by the log2 intensities obtained in the absence of acetate. These ratios were then normalised by the log2 median intensity. Genes whose expression level differed by a factor of 2 or more between the two conditions were used for further analysis. Gene ontology analyses were performed using Ecocyc (https://ecocyc.org), p-values were estimated using Fisher exact test with Bonferroni correction. The transcriptomics data can be downloaded from the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-9086. Theoretical expression data at a given growth rate were obtained by extrapolating the data from Esquerré et al., 2014.

Metabolomics experiments

Request a detailed protocol

Extracellular concentrations of labelled and unlabelled glucose and acetate were quantified during growth by 1D 1H-NMR on a Bruker Ascend 800 MHz spectrometer equipped with a 5 mm QCI cryoprobe (Bruker, Germany), as detailed previously (Millard et al., 2014). Briefly, 150 µL of filtered broth (0.2 μm, Sartorius, Germany) were mixed with 50 µL of D2O containing TSPd4 (3-(trimethylsilyl)-1-propanesulfonic acid-tetra deuterated, used as internal standard) at a concentration of 10 mM. A sequence using presaturation (ZGPR) was used for water signal suppression, with a 30° pulse and a relaxation delay between scans of 10 s to ensure full signal recovery. A total of 32 scans were accumulated (64 k data points with a spectral width of 16 ppm) after 4 dummy scans. From each spectrum, we quantified glucose and unlabelled and labelled acetate. Spectra were processed using TopSpin v3.2 (Bruker).

Model construction and analysis

All models are provided in SBML and COPASI formats in the Supplementary Information (Supplementary file 1) and at https://github.com/MetaSys-LISBP/acetate_regulation; Millard, 2021; copy archived at swh:1:rev:c8bcf8fad3459269df30dba7e52c81e62ca181d0. The calibrated kinetic model has also been deposited in the Biomodels database (https://www.ebi.ac.uk/biomodels) (Chelliah et al., 2015) with the identifier MODEL2005050001. Model analysis was performed using COPASI (v4.27, Hoops et al., 2006) with the CoRC package (COPASI R Connector v0.7.1, https://github.com/jpahle/CoRC, Förster and Bergmann, 2021) in R (v3.6.1, https://www.r-project.org). The scripts used to perform the simulations, to analyse the models and to generate the figures are provided in the Supplementary Information (Supplementary file 1) and at https://github.com/MetaSys-LISBP/acetate_regulation to ensure reproducibility and reusability.

Model construction

Request a detailed protocol

The model contains six reactions, six species, and two compartments (the environment and the cell, with a cell volume of 1.77 × 10−3 (L/gDWChassagnole et al., 2002; Figure 1A). Model units are litre (L) for volumes, hour (hr) for time, and millimole (mmol), and gram dry weight (gDW) for amounts of metabolites and biomass, respectively. All reactions are listed in Table 1. Glycolysis (which produces acetyl-CoA from glucose) and the TCA cycle (which utilises acetyl-CoA to produce biomass) were modelled using irreversible Michaelis-Menten kinetics. Growth rates were calculated from the flux of the TCA cycle assuming a constant biomass production yield from acetyl-CoA, in keeping with observations (Enjalbert et al., 2017; Pinhal et al., 2019). Acetate exchange between the cell and its environment was modelled as a saturable process using reversible Michaelis-Menten kinetics (Millard et al., 2017), and the Pta-AckA pathway was modelled using the detailed kinetics of the Pta and AckA enzymes used in previous models (Enjalbert et al., 2017; Millard et al., 2017; Kadir et al., 2010).

Table 1
Reactions included in the kinetic model of glucose and acetate metabolism of Escherichia coli.
NameReactionRate lawComment
Glucose_feedø → GLCCGlucose inflow and medium outflow to simulate chemostat experiments
Acetate_outflowACEenv → øMA
Biomass_outflowX → øMA
Glucose_outflowGLC → øMA
GlycolysisGLC → 1.4 × ACCOAIMMStoichiometric coefficient taken from Millard et al., 2014
TCA_cycleACCOA → øIMMUtilisation of AcCoA by the TCA cycle
PtaACCOA ↔ ACPRMMRate law from Enjalbert et al., 2017; Millard et al., 2017; Kadir et al., 2010
AckAACP ↔ ACEcellRMMRate law from Enjalbert et al., 2017; Millard et al., 2017; Kadir et al., 2010
Acetate_exchangeACEcell ↔ ACEenvRMMRate law from Millard et al., 2017
GrowthX → 2 × XMARate calculated from the TCA cycle flux, assuming a constant biomass yield (Enjalbert et al., 2017; Pinhal et al., 2019)
  1. Abbreviations: C: constant flux; MA: mass action; RMM: reversible Michaelis-Menten; IMM: irreversible Michaelis-Menten.

The differential equations, which describe the progression of the variables over time as a function of the system's rates, balance the concentrations of extracellular (biomass, glucose, and acetate) and intracellular (acetate, acetyl-CoA, and acetyl-phosphate) species:

dGLCdt=vglycolysisXVcellVenv [+vfeedDGLC]
dACEenvdt=vacetate_exchangeXVcellVenv [DACEenv]
dXdt=Xvgrowth [DX]
dACCOAdt=1.4vglycolysisvTCAcyclevPta
dACPdt=vPtavAckA
dACEcelldt=vAckAvacetate_exchange

Terms within square brackets are required only to simulate chemostat experiments (at dilution rate D and with a glucose feed defined by vfeed).

Reaction rates were modelled using the following rate laws:

vglycolysis=VmaxglycolysisGLCGLC+Km_GLC [11+ACEenvKi_ACE]
vTCA_cycle=VmaxTCA_cycleACCOAACCOA+Km_ACCOA [11+ACEenvKi_ACE]
vAckA=VmaxAckA(ACPADPACEcellATPKeq)Km_ACPKm_ADP(1+ACPKm_ACP+ACEcellKm_ACE)(1+ADPKm_ADP+ATPKm_ATP)
vPta=VmaxPta(ACCOAPACPCOAKeq)Km_ACCOAKm_P1+ACCOAKm_ACCOA+PKi_P+ACPKi_ACP+COAKm_COA+ACCOAPKm_ACCOAKm_P+ACPCOAKm_ACPKm_COA
vAcetate_exchange=VmaxAcetate_exchange(ACEcellACEenvKeq)Km_ACE1+ACEcellKm_ACE+ACEenvKm_ACE
vGrowth=vTCA_cycleY

As detailed in the Results section, we constructed four different versions of this model, i.e., with or without (non-competitive) inhibition of the glycolytic and/or TCA cycle pathways by acetate (terms within square brackets). This regulatory term represents the immediate, integrated response of E. coli metabolism to changes in acetate concentration, with no a priori knowledge on the underlying molecular mechanisms. The controlling species is defined as being extracellular acetate, i.e., the initial environmental signal sensed by the cells. Similar results and conclusions are obtained if the controlling species is the intracellular acetate pool.

Finally, these models were extended with isotopic equations for parameter estimation, as detailed in Enjalbert et al., 2017 and Millard et al., 2015. Briefly, all reactions (except biomass synthesis) were considered separately for unlabelled and labelled metabolites. Rate laws of reversible reactions were decomposed into their forward and reverse components to account for the bidirectional isotope exchange that arise from reversibility and significantly affects the distribution of isotopes through the network. For instance, the isotopically extended balance of acetyl-CoA corresponds to:

dACCOA0dt=GLC0GLC0+GLC11.4vglycolysis+ACP0ACP0+ACP1vPtareverseACCOA0ACCOA0+ACCOA1(vTCAcycle+vPtaforward)
dACCOA1dt=GLC1GLC0+GLC11.4vglycolysis+ACP1ACP0+ACP1vPtareverseACCOA1ACCOA0+ACCOA1(vTCAcycle+vPtaforward)

where subscripts 0 and 1 refer to the unlabelled and labelled metabolites, respectively.

Concentration of cofactors

Request a detailed protocol

Concentrations of cofactors were taken from a published kinetic model of the Pta-AckA pathway (Enjalbert et al., 2017; ADP = 0.61 mM, ATP = 2.40 mM, CoA = 1.22 mM, P = 10 mM).

Parameter estimation

Request a detailed protocol

The values of 14 of the 24 parameters were taken directly from the literature (Table 2). Parameters whose values are not available from elsewhere, which do not have a real biochemical value, or for which biochemical measurements are generally not representative of intracellular conditions (e.g., Vmax) were estimated to optimally reproduce 152 experimental data points obtained from E. coli K-12 MG1655 grown on 13C-glucose (15 mM) plus 12C-acetate (1, 10, or 30 mM). These data included time-course concentrations of biomass, glucose, and acetate and 13C-labelling of acetate. The parameters p were estimated by minimising the objective function f defined as the weighted sum of squared errors:

f(p)=i(xiyi(p)σi)2

where xi is the experimental value of data point i, with an experimental standard deviation σi, and yi(p) is the corresponding simulated value. The objective function f was minimised with the particle swarm optimisation algorithm (2,000 iterations with a swarm size of 50). The experimental and fitted data are shown in Figure 1 and provided in the Supplementary Information (Supplementary file 1). Values and 95% confidence interval of the estimated parameters are given in Table 3.

Table 3
Values and 95% confidence intervals of the estimated parameters.
ReactionParameterValue95 % CI
 AckAVmax3.4 × 1052.8 × 105 – 5.5 × 105
 PtaVmax9.8 × 1054.9 × 104 – 9.9 × 106
 GlycolysisVmax5.6 × 1035.3 × 103 – 5.9 × 103
Ki_ACE36.730.9 – 46.9
 TCA cycleKm_ACCOA24.88.4 – 615.4
Vmax7.4 × 1052.4 × 105 – 1.7 × 106
Ki_ACE2.31.8 – 3.4
 GrowthY1.0 × 10−49 × 10−5 – 1.1 × 10−4
 Acetate exchangeVmax4.8 × 1058 × 104 – 1.5 × 106
Km_ACE33.21.5 – 99.8

Goodness-of-fit analysis

Request a detailed protocol

We used χ2 statistical tests to assess the goodness of fit of each model and determine whether they described the data with sufficient accuracy. The minimised variance-weighted sum of squared residuals (SSR) is a stochastic variable with a χ2 distribution. The acceptable threshold for SSR values is χ²(α, d), where d represents the number of degrees of freedom and is equal to the number of fitted measurements n minus the number of estimated independent parameters p. The parameter α was set to 0.95 to define a 95% confidence threshold and models with SSRs above this threshold were rejected since they cannot accurately reproduce the experimental data.

Model validation

Request a detailed protocol

A total of 170 experimental data (extracellular fluxes and concentrations) were collected from the literature to evaluate the predictive power of each model. These data, which were not used to train or fit the models, were obtained for E. coli K-12 MG1655 and its close derivative strain BW25113 grown on different concentrations of glucose and/or acetate, as detailed in the Results section and in Renilla et al., 2012; Enjalbert et al., 2017; Pinhal et al., 2019. The steady-state and dynamic responses of E. coli to the corresponding perturbations were predicted using each model (Figure 3). The residual error (SSR) was calculated for each model from the simulated and experimental validation data (Figure 3—figure supplement 4) to identify the model that yielded the most accurate predictions.

Metabolic control and regulation analyses

Request a detailed protocol

Scaled flux control coefficients (CEJ), which represent the fractional change in the steady-state flux J in response to a fractional change in the rate of the reaction E (Kacser and Burns, 1973; Heinrich and Rapoport, 1974; Cornish-Bowden, 1995), were calculated as follows:

CEJ=lnJlnE

Similarly, we defined the response coefficient (RAceJace) which represents the dependence of the acetate flux (Jace) on the extracellular concentration of acetate (Kacser and Burns, 1973; Cornish-Bowden, 1995):

RAceJace=lnJacelnAce

The partitioned response relationship (Kacser and Burns, 1973; Cornish-Bowden, 1995) was used to quantify the flux response (viRAceJace) to a change in acetate concentration channelled through reaction i. The effect of the acetate concentration on the rate of reaction i (vi) is described by the elasticity coefficient εAcevi, and the resulting change in vi then propagates through the system depending on the control exerted by reaction i on the acetate flux (CviJace):

viRAceJace=CviJaceεAcevi

with

εAcevi=lnvilnAce

Global sensitivity analyses

Request a detailed protocol

We used a Monte-Carlo approach (Saa and Nielsen, 2017) to determine the 95% confidence intervals on (i) the fits of the experimental data (Figure 1), (ii) the estimated parameters (Table 3), (iii) the predicted responses to perturbations (Figure 3), and (iv) the flux control and regulation coefficients (Figures 5 and 6). For this purpose, we generated 500 simulated sets of calibration data with noise added according to experimental standard deviations. For each of these artificially noisy data sets, we carried out complete computational analyses (i.e., including parameter estimation – starting from random initial parameter values, simulation of the responses to different perturbations, and metabolic control and regulation analyses). We calculated the mean value and 95% confidence intervals around each parameter and each variable (i.e., concentrations, fluxes, flux control coefficients, and response coefficients) from the distribution of values obtained for the 500 data sets.

Data availability

Transcriptomics data have been deposited in ArrayExpress under accession code E-MTAB-9086. The calibrated kinetic model has been deposited in BioModels database under accession code MODEL2005050001. All the scripts used to perform the simulations, to analyse the models and to generate the figures are provided in the supporting files and at https://github.com/MetaSys-LISBP/acetate_regulation (copy archived https://archive.softwareheritage.org/swh:1:rev:c8bcf8fad3459269df30dba7e52c81e62ca181d0/). All data generated or analysed during this study are included in the manuscript and supporting files.

The following data sets were generated
    1. Millard P
    2. Enjalbert B
    3. Uttenweiler-Joseph S
    4. Portais JC
    5. Letisse F
    (2020) ArrayExpress
    ID E-MTAB-9086. Response of Escherichia coli to acetate concentration during growth on glucose.
    1. Millard P
    (2020) BioModels
    ID MODEL2005050001. Calibrated kinetic model of glucose and acetate metabolisms of Escherichia coli.

References

    1. Cornish-Bowden A
    (1995)
    Metabolic control analysis in theory and practice
    Advances in Molecular and Cell Biology 11:21–64.
  1. Conference
    1. Kacser H
    2. Burns JA
    (1973)
    The control of flux
    Symposia of the Society for Experimental Biology.
    1. Lee SG
    2. Liao JC
    (2008)
    Control of acetate production rate in Escherichia coli by regulating expression of single-copy pta using lacI(Q) in multicopy plasmid
    Journal of Microbiology and Biotechnology 18:334–337.
    1. Warburg O
    (1956)
    On respiratory impairment in cancer cells
    Science 124:269–270.
    1. Wolfe AJ
    (2005) The acetate switch
    Microbiology and Molecular Biology Reviews 69:12–50.
    https://doi.org/10.1128/MMBR.69.1.12-50.2005

Decision letter

  1. Michael Doebeli
    Reviewing Editor; University of British Columbia, Canada
  2. Naama Barkai
    Senior Editor; Weizmann Institute of Science, Israel

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This paper provides an analysis of the metabolic feedback mechanisms regulating overflow metabolism during glycolysis in E. coli. The authors present a kinetic metabolic model incorporating glycolysis, the TCA cycle, and acetate metabolism, and argue that acetate needs to suppress glycolysis and TCA cycle activity for the model to be able to reproduce experimental findings. The assumption of acetate activity is further validated through experiments showing that acetate reduces expression of the glucose phosphotransferase system and of most TCA cycle genes. Another finding is that the control of acetate production shifts from inside (low acetate concentrations) to outside (high concentrations) the acetate pathways.

Decision letter after peer review:

Thank you for submitting your article "Control and regulation of acetate overflow in Escherichia coli" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Naama Barkai as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional work is required before it can be published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Essential revisions:

1) Modeling choices, model robustness and the question of alternative models need to be more thoroughly addressed.

2) Interpretation of experimental results need to be strengthened, in particular with regards to the time scales of transcription and regulation.

3) All comments made by reviewers need to be thoroughly and constructively addressed.

Reviewer #2 (Recommendations for the authors (required)):

The manuscript delves into a major classical question in the study of metabolism, namely how organisms balance the decision of respiring or fermenting carbon sources, and what role fermentation by-products play in this process. Specifically, the authors focus on E. coli and the dependence of its physiology on acetate present in the environment. This is an interesting topic relevant to a broad spectrum of questions, including cancer metabolism. The approach is original, and the specific hypothesis that acetate influences transcription and the metabolic flows in previously unknown ways is well-posed and intriguing. I must admit though that I am not totally convinced that the results provide conclusive evidence about their hypothesis, ruling out other possibilities. Clear comparisons with alternative models, parameter sensitivity analyses, and a clearer expression of possible caveats would alleviate this concern, and help readers clearly understand the possible limitations of the approach employed. The work should be of interest to a fairly broad audience of biochemists, microbiologists and systems biologists.

Subsection “Testing the model”: the authors mention an excellent agreement between model and experiment in Figure 3. I think the agreement is not bad, but I would not call it excellent. I am not sure a p-value is absolutely necessary, but it would be good to at least have an idea of how much variability the experimental data may have beyond the shown error bars, and whether the agreement is really unexpected, also in light of the parameter fitting presented in Figure 1.

It would be useful to see Figure 3 in absence of the acetate inhibition effects. Would the results look only quantitatively different or also qualitatively different? Can the authors comment on the intuitive meaning of these curves. Is any of these curves dramatically affected by the new model, relative to alternative standard models?

As opposed to what is mentioned in the Introduction, stoichiometric models could in principle help address the questions posed by the authors. This would not be the case for regular FBA models, but would be true for dynamics FBA models, in which the extracellular concentration of substrates is explicitly accounted for.

The “Metabolomics experiments” subsection in the Materials and methods is very brief, and does not get into the detail of flux measurements reported, which are crucial for the model and its interpretation.

I am curious about whether acetate inflow and outflow are separately measurable (and measured) in the system, or whether the reports are for the net flow. I am not sure if multiple acetate transporters exist, and whether it is in principle possible for E. coli to simultaneously secrete acetate due to fermentation and take it up from the environment, but it would be good for the authors to comment on this.

No parameter sensitivity analysis was performed. I would recommend that the authors either add it, or comment on why it would not be useful.

The authors' conclusions are presented in an absolute way, that leaves no room for alternative explanations. Partly due to the issues mentioned above, after reading the paper I felt highly intrigued by the results, but not 100% convinced that their explanation is the only possible. I would suggest that the authors (a) take steps to present in a more critical way the results in light of other models that were not explicitly tested, and (b) acknowledge explicitly possible limitation of this study, and avenues to further corroborate the proposed mechanisms.

Introduction, second paragraph: it becomes clear from the context, but the authors should clearly mention that they refer to extracellular concentration of acetate.

Reviewer #3 (Recommendations for the authors (required)):

The manuscript presents an extension of previous work of the group (Enjalbert et al., 2017; Millard, Smallbone and Mendes, 2017). The authors present an interesting dynamical model of overflow metabolism, producing new insights into a long-standing question in bacterial physiology. The model is confronted with new data that appear to confirm specific model predictions. The major novelty of the model is the addition of a direct control by acetate of the activities of glycolysis (or glucose uptake) and the TCA cycle. Very interesting transcriptomics data appear to confirm these regulatory relationships. However, transcriptional regulation and the direct regulations suggested by the model act at different time-scales. In conclusion, the work is very well carried out and interesting. The connection between the transcriptomics data and the model need to be clarified.

1) The first major argument in favor of control of glycolysis and the TCA cycle by acetate is the quality of fit to bioreactor data. It does not seem surprising to me that the quality of fit (Figure 1B) is improved by adding additional regulatory interactions. Even though the figure shows a significant improvement of the fits, it would be nice to see the alternative fits (the equivalent of Figure 1C) in supplementary information.

2) A strong claim of the manuscript is that a kinetic model is needed to account for experimental observations. The fits in Figure 1 use the dynamical model at steady state. How do the predictions compare to the alternative models (global control of metabolism, local control of the acetate pathway)?

3) The transcriptomics data are very interesting. However, these data were acquired at steady state, reflecting a slow adaptation (order of magnitude of the generation time) of physiology. The regulatory interactions (square brackets in the equations in the subsection “Model construction”) are instantaneous effects of the concentration of acetate. I fail to see how the transcriptomics data can argue in favor (or disfavor) of the kinetic model.

4) The perturbation experiments (Figure 3) are probably the best test of a dynamical model. The model correctly predicts the observed dynamics, which are clearly on time-scale of minutes, precluding transcriptional regulation as a cause. The authors could suggest possible molecular mechanisms.

5) The remaining results of the manuscript address model properties and are very well carried out. Under the hypothesis that the model is correct, these analyses yield interesting insights into the physiology of overflow metabolism.

In conclusion: the manuscript describes a very interesting extension to current models of overflow metabolism in E. coli. The experiments are very well carried out and pertinent new data are presented.

My major concern is that the transcriptomics data seem disconnected from the kinetic model. Long-term adaptation can be achieved by modifications in gene expression. However, this does not correspond to the mechanism described by the model equations. The manuscript would be much stronger if the authors could show at least one of the molecular mechanisms by which acetate controls metabolic pathways. I realize that this would be "another" manuscript and cannot be reasonably covered by major revisions. The authors could at least discuss possible molecular mechanisms of this regulation.

https://doi.org/10.7554/eLife.63661.sa1

Author response

Essential revisions:

1) Modeling choices, model robustness and the question of alternative models need to be more thoroughly addressed.

In the revised manuscript, we provide additional details to clarify our modelling choices, in particular regarding (i) the formalism used to model the regulation of metabolism by acetate and (ii) the definition of extracellular acetate as the controlling factor. We have also extensively tested the robustness of the model by performing global sensitivity analyses for each step, from model construction to the analysis of regulation, which confirm that the model and its predictions are robust to parameter uncertainty. Finally, simulations performed with alternative models highlight the biological importance of the identified regulatory interaction on the properties of the system. We appreciate these constructive suggestions, which significantly improve the manuscript.

2) Interpretation of experimental results need to be strengthened, in particular with regards to the time scales of transcription and regulation.

We have strengthened the interpretation of the experimental results, in particular with regards to the time scales of the different regulatory mechanisms.

3) All comments made by reviewers need to be thoroughly and constructively addressed.

All of the reviewers’ comments are thoroughly addressed below.

Reviewer #2 (Recommendations for the authors (required)):

The manuscript delves into a major classical question in the study of metabolism, namely how organisms balance the decision of respiring or fermenting carbon sources, and what role fermentation by-products play in this process. Specifically, the authors focus on E. coli and the dependence of its physiology on acetate present in the environment. This is an interesting topic relevant to a broad spectrum of questions, including cancer metabolism. The approach is original, and the specific hypothesis that acetate influences transcription and the metabolic flows in previously unknown ways is well-posed and intriguing. I must admit though that I am not totally convinced that the results provide conclusive evidence about their hypothesis, ruling out other possibilities. Clear comparisons with alternative models, parameter sensitivity analyses, and a clearer expression of possible caveats would alleviate this concern, and help readers clearly understand the possible limitations of the approach employed. The work should be of interest to a fairly broad audience of biochemists, microbiologists and systems biologists.

We thank reviewer 2 for their insightful comments and constructive suggestions.

The revised manuscript contains detailed comparisons with alternative models (i.e. models without inhibition of the glycolytic and/or TCA cycle pathways by acetate), which highlight the importance of the identified regulatory interaction for the functioning of the system and its control properties. We have also performed a global sensitivity analysis, which confirms our conclusions are robust to parameter uncertainty, and we discuss the limitations of our approach.

We would like to stress that the simplified formalism used to consider acetate inhibition represents the integrated regulatory response of E. coli metabolism to acetate, without any a priori knowledge of the exact mechanism. Experimental results confirm that acetate does indeed regulate gene expression in E. coli, but this does not rule out other possibilities. In fact, as stated in the original manuscript, we believe that other mechanisms work in concert, and that acetate may “also regulate E. coli metabolism at the post-translational level (e.g. via the carbon storage regulator system [Revelles et al., 2013] or acetyl-phosphate [Weinert et al., 2013; Ren et al., 2019]) and at the metabolic level (e.g. via the control of glucose uptake by α-ketoglutarate [Doucette et al., 2011] or of glycolysis by ATP [Koebmann et al., 2002])”. These regulatory mechanisms are likely also involved in the response of E. coli to acetate, and this should be explored in future work. We have made this point clearer in the revised manuscript.

Subsection “Testing the model”: the authors mention an excellent agreement between model and experiment in Figure 3. I think the agreement is not bad, but I would not call it excellent. I am not sure a p-value is absolutely necessary, but it would be good to at least have an idea of how much variability the experimental data may have beyond the shown error bars, and whether the agreement is really unexpected, also in light of the parameter fitting presented in Figure 1.

We agree with this recommendation. In the revised manuscript, we have removed the term “excellent”, and have carried out global sensitivity analyses to determine 95% confidence intervals around the model predictions. We now evaluate these predictions based on a quantitative metric, namely the variance-weighted sum of squared residuals between predicted and experimental data. This metric is useful to compare the proposed model with alternatives. Results indicate that the model with inhibitions, which is already the only model that satisfactorily fits the data shown in Figure 1, is also the one that most accurately predicts the data used for validation (Figure 3). We present and discuss these new results in the revised manuscript (in the Results and Discussion sections).

It would be useful to see Figure 3 in absence of the acetate inhibition effects. Would the results look only quantitatively different or also qualitatively different? Can the authors comment on the intuitive meaning of these curves. Is any of these curves dramatically affected by the new model, relative to alternative standard models?

We have followed this suggestion and have predicted the responses to the perturbations shown in Figure 3 with the alternative models, i.e. in the absence of acetate inhibition of the glycolytic and/or TCA pathways (see Figure 3—figure supplements 1–4). The new simulation results are quantitatively and qualitatively different, which confirms that the curves are drastically affected by the absence of inhibition, further supporting the regulatory role of acetate. We have also carried out metabolic control analyses for the alternative models (see Figure 5—figure supplements 1–3), which also indicate that the control properties of the system are qualitatively affected by the absence of inhibition. These new results are discussed in the revised manuscript.

As opposed to what is mentioned in the Introduction, stoichiometric models could in principle help address the questions posed by the authors. This would not be the case for regular FBA models, but would be true for dynamics FBA models, in which the extracellular concentration of substrates is explicitly accounted for.

We agree with reviewer 2 that dynamic FBA models can be used to estimate and predict dynamic flux reorganizations. However, dynamic FBA models are not purely stoichiometric since they include a kinetic description of the uptake pathways, and thus can equally be considered (partly) kinetic models.

One of the main objectives of this work was to better understand the control and regulation of acetate metabolism. We could have used dynamic FBA models to predict the response of E. coli to perturbations, but not to obtain a quantitative, mechanistic understanding of overflow metabolism. Indeed, to the best of our knowledge, dynamic FBA models lack regulation of intracellular pathways (such as the inhibition of the TCA cycle and of glycolysis identified in this work and implemented in our model) and do not account for phenomena such as enzyme saturation. Since these models cannot be used to compute elasticity coefficients for the intracellular pathways, they cannot be used either to quantify the flux control exerted by each pathway, or to determine the regulatory routes activated in response to changes of acetate concentration.

The “Metabolomics experiments” subsection in the Materials and methods is very brief, and does not get into the detail of flux measurements reported, which are crucial for the model and its interpretation.

The set of 152 experimental data used for model calibration consist only of time-course exometabolome concentrations and labellings, no flux data (though this model can be used to simulate such data, as shown in Figure 3). All the flux measurements reported in this study were obtained from the literature. As detailed in the Materials and methods section, to validate the model we used “A total of 170 experimental data (extracellular fluxes and concentrations) obtained from the literature for E. coli K-12 MG1655 and its close derivative strain BW25113 grown on glucose with or without acetate [Renilla et al., 2012; Enjalbert et al., 2017; Pinhal et al., 2019]”. The methods used to obtain these flux measurements are described in detail in the corresponding publications. We have made this point clearer in the revised manuscript:

“A total of 170 experimental data (extracellular fluxes and concentrations) were collected from the literature to evaluate the predictive power of each model. […] The steady-state and dynamic responses of E. coli to the corresponding perturbations were predicted using each model (Figure 3).”

We have also added the following details in the “Metabolomics experiments” subsection of the Materials and methods section:

“Briefly, 150 µL of filtered broth (0.2 μm, Sartorius, Germany) were mixed with 50 µL of D2O containing TSPd4 (3-(trimethylsilyl)-1-propanesulfonic acid-tetra deuterated, used as internal standard) at a concentration of 10 mM. […] Spectra were processed using TopSpin 3.2 (Bruker).”

I am curious about whether acetate inflow and outflow are separately measurable (and measured) in the system, or whether the reports are for the net flow. I am not sure if multiple acetate transporters exist, and whether it is in principle possible for E. coli to simultaneously secrete acetate due to fermentation and take it up from the environment, but it would be good for the authors to comment on this.

The dynamic 13C-labeling experiments performed in this work can indeed be exploited to determine acetate production and consumption fluxes separately. In fact, in a previous study (Enjalbert et al., 2017), we integrated these experiments into an isotopic (stoichiometric) model to demonstrate, as suggested by reviewer 2, that E. coli simultaneously produces acetate from glucose and takes it up from the environment. We have added a couple of sentences about these results in the revised manuscript:

“These experiments were designed to demonstrate that E. coli simultaneously produces acetate from glucose and consumes it from the environment [Enjalbert et al., 2017]. They provide information on the forward and reverse fluxes of acetate between the cell and its environment [Enjalbert et al., 2017], and thereby improve the calibration of the model.”

No parameter sensitivity analysis was performed. I would recommend that the authors either add it, or comment on why it would not be useful.

We agree, and are grateful for this suggestion. In the revised manuscript, we present the results of a global sensitivity analysis (Monte Carlo approach) to determine 95 % confidence intervals on (i) the fit of the experimental data (Figure 1), (ii) the estimated parameters (Table 3), (iii) the predicted responses to perturbations (Figure 3), and (iv) the flux control and regulation coefficients (Figures 5 and 6). These procedures are detailed in the revised manuscript. The results confirm that our conclusions are robust to parameter uncertainty.

The authors' conclusions are presented in an absolute way, that leaves no room for alternative explanations. Partly due to the issues mentioned above, after reading the paper I felt highly intrigued by the results, but not 100% convinced that their explanation is the only possible. I would suggest that the authors (a) take steps to present in a more critical way the results in light of other models that were not explicitly tested, and (b) acknowledge explicitly possible limitation of this study, and avenues to further corroborate the proposed mechanisms.

In the revised manuscript, we provide a more critical presentation of the results in light of other models that are now explicitly tested, we discuss the limitations of the approach, and we discuss alternative hypotheses and avenues to further investigate the regulatory role of acetate.

Introduction, second paragraph: it becomes clear from the context, but the authors should clearly mention that they refer to extracellular concentration of acetate.

We now refer explicitly to the extracellular concentration of acetate.

Reviewer #3 (Recommendations for the authors (required)):

The manuscript presents an extension of previous work of the group (Enjalbert et al., 2017; Millard, Smallbone and Mendes, 2017). The authors present an interesting dynamical model of overflow metabolism, producing new insights into a long-standing question in bacterial physiology. The model is confronted with new data that appear to confirm specific model predictions. The major novelty of the model is the addition of a direct control by acetate of the activities of glycolysis (or glucose uptake) and the TCA cycle. Very interesting transcriptomics data appear to confirm these regulatory relationships. However, transcriptional regulation and the direct regulations suggested by the model act at different time-scales. In conclusion, the work is very well carried out and interesting. The connection between the transcriptomics data and the model need to be clarified.

We thank reviewer 3 for their positive feedback and for raising important points that we address and discuss in the revised manuscript. We have clarified the fact that regulation as encoded in the model does not aim to specifically represent transcriptional regulation (which is in fact fast, see our detailed response below) but rather the integrated response of E. coli to acetate. Transcriptomics data confirm that acetate regulates E. coli metabolism at least partly at the transcriptional level, and this does not exclude the possibility that other mechanisms may be involved in this integrated response. New analyses with the alternative models confirm that the proposed regulatory interactions are required not only to fit the data shown in Figure 1, but also to accurately predict the response of E. coli to perturbations (Figure 3 and Figure 3—figure supplements 1-4). Finally, we also demonstrate that unexpected control properties emerge from regulation by acetate (Figure 5 and Figure 5—figure supplements 1-3). The results of a global sensitivity analysis (suggested by reviewer 2) also strengthen these conclusions.

1) The first major argument in favor of control of glycolysis and the TCA cycle by acetate is the quality of fit to bioreactor data. It does not seem surprising to me that the quality of fit (Figure 1B) is improved by adding additional regulatory interactions. Even though the figure shows a significant improvement of the fits, it would be nice to see the alternative fits (the equivalent of Figure 1C) in supplementary information.

We have followed this suggestion and now provide alternative fits (equivalent to Figure 1C) in Figure 1—figure supplements 1–3. These new figures indeed help to visualize the extent to which the alternative models do not reproduce the data, as confirmed by the goodness-of-fit statistical analysis (Figure 1B).

2) A strong claim of the manuscript is that a kinetic model is needed to account for experimental observations. The fits in Figure 1 use the dynamical model at steady state. How do the predictions compare to the alternative models (global control of metabolism, local control of the acetate pathway)?

The fits in Figure 1 are not based on the dynamic model at steady-state. The simulated and experimental data are the time-course dynamics of biomass, glucose and acetate concentrations and of acetate labelling. We now provide the predictions of alternative models (responses to perturbations shown in Figure 3, as well as the control properties shown in Figure 5) in supplementary figures. The predictions are drastically affected by the absence of inhibition, confirming that regulation by acetate is necessary to explain the observed behaviours.

3) The transcriptomics data are very interesting. However, these data were acquired at steady state, reflecting a slow adaptation (order of magnitude of the generation time) of physiology. The regulatory interactions (square brackets in the equations in the subsection “Model construction”) are instantaneous effects of the concentration of acetate. I fail to see how the transcriptomics data can argue in favor (or disfavor) of the kinetic model.

As noted by reviewer 3, the inhibition of glycolysis and of the TCA cycle by acetate is considered immediate.

However, transcriptional regulation acts much faster than assumed by reviewer 3 and is not of the “order of magnitude of the generation time”. For instance, Kresnowati et al. (doi: 10.1038/msb4100083) found that glucose-limited Saccharomyces cerevisiae cells exhibit a major transcriptional reprogramming within the first 5 minutes after a glucose pulse. The authors showed that most TCA genes “were immediately downregulated” after the pulse. A similar time-scale has been reported for mRNA dynamics in E. coli, with a median mRNA half-live of just 3.1 min during exponential growth on glucose (Morin et al., 2020). This fast mRNA dynamics explains how the transcriptome of E. coli cells respond in less than 5 minutes to external perturbations, such as glucose exhaustion (Enjalbert, Letisse and Portais, 2013; Lempp et al., 2019) and changing oxygen levels (von Wulffen et al., 2017). As shown in the latter study, “significant differential expression of several genes within 30 seconds indicates immediate cell adaptation to the changed environment”. These findings indicate that the metabolic and transcriptional responses of E. coli and other microorganisms are tightly intertwined rather than occurring sequentially on different, independent time scales.

It seems important here to clarify that when constructing the model, the aim was not to specifically represent transcriptional regulation but rather the integrated response of E. coli to acetate. With this objective in mind, the model indeed provides new insights on global rules by which acetate governs E. coli metabolism. Transcriptomics experiments confirm that acetate control regulates the flux capacity of the glycolytic and TCA pathways at the transcriptional level, but we do not rule out other hypotheses. As discussed in the manuscript, we also believe that transcriptional regulation is not the only mechanism by which acetate regulates E. coli metabolism and physiology. We stated in the initial manuscript that acetate may “also regulate E. coli metabolism at the post-translational level (e.g. via the carbon storage regulator system [Revelles et al., 2013] or acetyl-phosphate [Weinert et al., 2013; Ren et al., 2019]) and at the metabolic level (e.g. via the control of glucose uptake by α-ketoglutarate [Doucette et al., 2011] or of glycolysis by ATP [Koebmann et al., 2002])”. These mechanisms act on a similar time scale – seconds to minutes – and are also represented by the feedback interaction included in the model. We have made this point clearer in the revised manuscript. Further work will be necessary to determine whether additional mechanisms are involved in the response of E. coli metabolism to acetate. We thank reviewer 3 for raising this important point, which we have clarified in the revised manuscript by adding the following sentences in the Discussion:

“In this coarse-grained model, the simplified formalism used to consider acetate inhibition represents the integrated regulatory response of E. coli metabolism to acetate, regardless of the underlying molecular mechanisms. […] These models should help to computationally evaluate the biological relevance of detailed regulatory mechanisms, as illustrated here.”

4) The perturbation experiments (Figure 3) are probably the best test of a dynamical model. The model correctly predicts the observed dynamics, which are clearly on time-scale of minutes, precluding transcriptional regulation as a cause. The authors could suggest possible molecular mechanisms.

As noted by reviewer 3, the perturbation experiments are indeed the best test of a dynamic model. The dynamics observed in response to an acetate pulse (Figure 3C) occur on a time-scale of 8 minutes, which cannot be explained by transcriptional regulation alone. In fact, as discussed in the manuscript, we also think it likely that acetate regulates E. coli metabolism via other mechanisms operating at the post-translational level or at the metabolic level. We have added a paragraph discussing in detail the time-scales of regulation and other possible regulatory mechanisms in the revised manuscript (see our answer to comment #3).

5) The remaining results of the manuscript address model properties and are very well carried out. Under the hypothesis that the model is correct, these analyses yield interesting insights into the physiology of overflow metabolism.

We thank reviewer 3 for this positive feedback on the analyses performed with the model.

In conclusion: the manuscript describes a very interesting extension to current models of overflow metabolism in E. coli. The experiments are very well carried out and pertinent new data are presented.

My major concern is that the transcriptomics data seem disconnected from the kinetic model. Long-term adaptation can be achieved by modifications in gene expression. However, this does not correspond to the mechanism described by the model equations. The manuscript would be much stronger if the authors could show at least one of the molecular mechanisms by which acetate controls metabolic pathways. I realize that this would be "another" manuscript and cannot be reasonably covered by major revisions. The authors could at least discuss possible molecular mechanisms of this regulation.

We thank reviewer 3 for their constructive suggestions. Using the model, we determined that acetate inhibits the glycolytic and TCA fluxes, and identified its global impact on E. coli cells. Importantly, we show that this inhibition involves at least the transcriptomics response, which is in fact fast and tightly intertwined with the metabolic response (see above). We do not rule out other possible explanations and are working on identifying additional mechanisms involved in this response, which are indeed beyond the scope of this manuscript. In the revised manuscript, we discuss other molecular mechanisms possibly involved in the previously unknown regulatory role of acetate, and how the present model can help to address this question.

https://doi.org/10.7554/eLife.63661.sa2

Article and author information

Author details

  1. Pierre Millard

    1. TBI, Université de Toulouse, CNRS, INRAE, INSA, Toulouse, France
    2. MetaToul-MetaboHUB, National Infrastructure of Metabolomics and Fluxomics, Toulouse, France
    Contribution
    Conceptualization, Software, Formal analysis, Supervision, Funding acquisition, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    millard@insa-toulouse.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8136-9963
  2. Brice Enjalbert

    TBI, Université de Toulouse, CNRS, INRAE, INSA, Toulouse, France
    Contribution
    Formal analysis, Investigation, Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1291-1373
  3. Sandrine Uttenweiler-Joseph

    TBI, Université de Toulouse, CNRS, INRAE, INSA, Toulouse, France
    Contribution
    Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9019-4766
  4. Jean-Charles Portais

    1. TBI, Université de Toulouse, CNRS, INRAE, INSA, Toulouse, France
    2. MetaToul-MetaboHUB, National Infrastructure of Metabolomics and Fluxomics, Toulouse, France
    3. RESTORE, Université de Toulouse, INSERM U1031, CNRS 5070, Université Toulouse III - Paul Sabatier, EFS, Toulouse, France
    Contribution
    Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3480-0933
  5. Fabien Létisse

    1. TBI, Université de Toulouse, CNRS, INRAE, INSA, Toulouse, France
    2. Université Toulouse III - Paul Sabatier, Toulouse, France
    Contribution
    Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1490-0152

Funding

Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement (MICA-JC)

  • Pierre Millard

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors thank MetaboHub-MetaToul (Metabolomics and Fluxomics facilities, Toulouse, France, http://www.metatoul.fr), which is part of the French National Infrastructure for Metabolomics and Fluxomics (http://www.metabohub.fr/), funded by the ANR (MetaboHUB-ANR-11-INBS-0010), for access to NMR facilities. JCP is grateful for funding from INSERM for his temporary full-time researcher position. The authors are grateful for Vincent Pierunek’s help with the isotope labelling experiments, and wish to thank the following INSA Toulouse students for help with the transcriptomics experiments: Leidy Caraballo, Xavier Caron, Pauline Chanut, Sarah Guiziou, Ngoc Thu Hang Pham, Diane Barbay, Céline Ben Hassen, Thomas Cerutti, Cécile Roland, Sarah Srour, Audrey Baylet, Mathilde Beraud, Claudie Bosc, Lilas Courtot, Violaine Dolfo, Anna Kaci, Manon Chevallot-Beroux, Sarah Colom, Fanny Leclerc, Zihan Liao, Grégoire Quinet and Mélina Vaurs.

Senior Editor

  1. Naama Barkai, Weizmann Institute of Science, Israel

Reviewing Editor

  1. Michael Doebeli, University of British Columbia, Canada

Publication history

  1. Received: October 1, 2020
  2. Accepted: March 12, 2021
  3. Accepted Manuscript published: March 15, 2021 (version 1)
  4. Version of Record published: April 4, 2021 (version 2)

Copyright

© 2021, Millard et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 496
    Page views
  • 88
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Computational and Systems Biology
    2. Neuroscience
    Brian Q Geuther et al.
    Research Article Updated

    Automated detection of complex animal behaviors remains a challenging problem in neuroscience, particularly for behaviors that consist of disparate sequential motions. Grooming is a prototypical stereotyped behavior that is often used as an endophenotype in psychiatric genetics. Here, we used mouse grooming behavior as an example and developed a general purpose neural network architecture capable of dynamic action detection at human observer-level performance and operating across dozens of mouse strains with high visual diversity. We provide insights into the amount of human annotated training data that are needed to achieve such performance. We surveyed grooming behavior in the open field in 2457 mice across 62 strains, determined its heritable components, conducted GWAS to outline its genetic architecture, and performed PheWAS to link human psychiatric traits through shared underlying genetics. Our general machine learning solution that automatically classifies complex behaviors in large datasets will facilitate systematic studies of behavioral mechanisms.

    1. Computational and Systems Biology
    Michael S Lauer et al.
    Research Article

    A previous report found an association of topic choice with race-based funding disparities among R01 applications submitted to the National Institutes of Health ('NIH') between 2011-2015. Applications submitted by African American or Black ('AAB') Principal Investigators ('PIs') skewed toward a small number of topics that were less likely to be funded (or 'awarded'). It was suggested that lower award rates may be related to topic-related biases of peer reviewers. However, the report did not account for differential funding ecologies among NIH Institutes and Centers ('ICs'). In a re-analysis, we find that 10% of 148 topics account for 50% of applications submitted by AAB PIs. These applications on 'AAB Preferred' topics were funded at lower rates, but peer review outcomes were similar. The lower rate of funding for these topics was primarily due to their assignment to ICs with lower award rates, not to peer-reviewer preferences.