Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2022 Jul 8;11(8):2672–2684. doi: 10.1021/acssynbio.2c00097

Validated In Silico Population Model of Escherichia coli

Sreenath Rajagopal †, Rothangmawi Victoria Hmar ‡, Debdatto Mookherjee †, Arindam Ghatak ‡,§, Anirudh P Shanbhag †,§, Nainesh Katagihallimath †, Janani Venkatraman ‡, Ramanujan KS ‡, Santanu Datta †,*
PMCID: PMC9397402  PMID: 35801944

Abstract

graphic file with name sb2c00097_0009.jpg

Flux balance analysis (FBA) and ordinary differential equation models have been instrumental in depicting the metabolic functioning of a cell. Nevertheless, they demonstrate a population’s average behavior (summation of individuals), thereby portraying homogeneity. However, living organisms such as Escherichia coli contain more biochemical reactions than engaging metabolites, making them an underdetermined and degenerate system. This results in a heterogeneous population with varying metabolic patterns. We have formulated a population systems biology model that predicts this degeneracy by emulating a diverse metabolic makeup with unique biochemical signatures. The model mimics the universally accepted experimental view that a subpopulation of bacteria, even under normal growth conditions, renders a unique biochemical state, leading to the synthesis of metabolites and persister progenitors of antibiotic resistance and biofilms. We validate the platform’s predictions by producing commercially important heterologous (isobutanol) and homologous (shikimate) metabolites. The predicted fluxes are tested in vitro resulting in 32- and 42-fold increased product of isobutanol and shikimate, respectively. Moreover, we authenticate the platform by mimicking a bacterial population in the presence of glyphosate, a metabolic pathway inhibitor. Here, we observe a fraction of subsisting persisters despite inhibition, thus affirming the signature of a heterogeneous populace. The platform has multiple uses based on the disposition of the user.

Keywords: isobutanol, shikimate, glyphosate, Escherichia coli, knockouts, flux balance analysis (FBA), nutrient modulation, minimal media, matrices

Introduction

The microbial culture is a versatile and burgeoning community of interacting individuals that define their nature. This property helps drive the metabolite production by microbes in industries, the formation of persisters against antibiotics or biofilm formation. Bacteria have evolved to survive in the “occasionally famine and rarely feast” conditions for growth and survival. In contrast, the nutritional factors in laboratories are made optimal for procuring target biomass. Inherent in this optimizing process is often an opposite pull of flux of maximizing the biomass, which is a near precise stoichiometric summation of multiple essential metabolites or the specific maximization of any metabolite. If these nutritional conditions can be exploited, we can have a minimalistic system to carry out reactions of commercially valuable metabolites with low-cost inputs or understand the nature of the bacterial population in the presence of an antibiotic. Understandably, we must construct a virtual population model and validate its formulation through precise experiments.

One of the significant drivers for constructing in silico models for different microorganisms is to forecast the genetic changes (gene deletions and overexpressions) required to enhance the metabolite of interest. Two types of platforms are commonly used to simulate a bacterial cell. It is either based on a linear flux-balanced analysis (FBA) platform1−4 or dependent on ordinary differential equations (ODEs), where kinetic parameters (Km and Vmax) of enzymes participating in the particular reaction type (single substrate, multisubstrate, ping pong, ternary complex, etc.) are used.5−8 The popular FBA platform generally incorporates developing a stoichiometric model with genome-level annotations of pathways that map a conversion substrate to a product. The simulations thereof are used to predict knockouts of non-essential genes in specific pathways (identified during the simulation studies), which would help block the formation of metabolites and allow the cells to redirect the carbon feed into the production of the metabolite of interest.3

All types of in silico models, whether based on the FBA or ODE mathematical framework, tacitly assume that all the cells are in an identical metabolic state (Figure 1A,B) as it assumes every individual to have equal exposure to the participating nutrients.9 Such an approach is often invalidated in a natural environment, where intraspecies competition persists among bacteria, and where nutrient availability is the currency for driving a reliable binary of “haves” and “have-nots”. Hence, the bacterial population is generally asynchronous in its growth. Additionally, its reaction to a stimulus (positive or negative) at the individual cellular level is not unique and can only be illustrated by a diverse array of responses. Some proliferate in a population of bacterial cells under a standard nutrient condition, while others are stunted. Similarly, 90–99% of the population die upon exposure to an antibacterial compound, while the remaining 1–10% of bacterial cells remain unaltered due to the system’s robustness.10 In essence, a collection of cells in a group with the same genetic makeup (isogenic) has a different expression profile. For example, the heterogeneous expression of an araBAD promoter in the presence of limited arabinose quantities in Mycobacterium tuberculosis shows variations in individuals of a population.11 Additionally, as summarized in a recent review, it is a truism that in a population of bacterial cells, the gene expression levels are highly variable in lower nutrient concentrations.12 Hence, there is a pressing need to develop a population model where each cell has a unique metabolic signature. We can use a similar approach to understand the optimal flux of pathways in a population.

Figure 1.

Figure 1

Open in a new tab

Depiction of ODE, FBA, and POSYBEL models. (A) ODE model takes in the kinetic parameters and gives solutions based on those parameters. The solution is singular (green). However, gene lethality can also be found (red) for a given knockout. (B) FBA model uses stoichiometric constraints and gives a single optimal solution for the entire population. (C) POSYBEL may not predict essentiality, but it considers multiple solutions for each stoichiometric constraint and gives multiple combination of flux the best knockouts are chosen through the population. (D) Concept behind constraint-based modeling where a solution space is limited given the “dimensions” (red) by applying constraints. Off this constrained solution (orange) is the conventional FBA solution for identifying an optimal result toward a target flux. In the alternative approach, the population is chosen from the constrained region (red) using Markov model Monte Carlo simulations for further testing.

We propose a method for building a working population model, using Escherichia coli as an example, without requiring in vitro data such as gene expression profiles, enzyme kinetic parameters, biomass growth, and so forth. We use glucose as the sole carbon source, where modeling individual fluxes through different cells using the Markov chain Monte Carlo (MCMC) algorithm helps emulate a real-world scenario. We validate the model by constructing E. coli with minimal genetic modifications based on the predicted fluxes to produce homologous and heterologous metabolites such as shikimate and isobutanol at multigram levels. Furthermore, we demonstrate a natural example of mass balance (the core of any FBA system) by adapting the reduction of nitrogen in media to produce the same metabolites (as they are devoid of nitrogen). Additionally, we show our model’s robustness by demonstrating the behavior of the bacterial population upon introduction of the metabolic inhibitor glyphosate. We verify the population systems biology model’s (POSYBEL) concept of a heterogeneous bacterial population. Furthermore, the model removes the requirement of in vitro data, specifically enzyme kinetics or biomass growth, thereby allowing the user to simulate a population of any organism provided if its pathways are deciphered.

Results

Population-based simulations have been previously done in yeast and E. coli under general medium conditions, where a specific population of cells is imitated using protein expression and kinetics data to reach optimum conclusions in biomass yield and simultaneously deciphering the metabolic state of individual cells.13 Invariably, such an endeavor requires high-throughput in vitro data to reach acceptable decisions.14,15 Flux sampling, unlike other approaches, can be employed without presuming a specific biological goal, and algorithms such as the coordinate hit-and-run with rounding (CHRR) technique are efficient in terms of both run-time and multiple convergence diagnostics for complex systems.16 Our strategy follows an alternate logic to reach similar conclusions. First, unlike flux variant analysis, which finds out the range of each reaction fluxes driving the network toward a desirable optimum, we utilize the MCMC algorithm. This process stochastically samples the entire solution space to arrive at cells that mimic the real-world scenario, where none of the reactions has an absolute zero flux (each reaction should be running at the basal level at the least). Second, instead of using the kinetic parameters and protein expression data in addition to the medium conditions, we use only the latter to have a minimally constrained model.

The output of the POSYBEL model is in the form of a triangle, where the “cells”, or dots (drawn at random from the solution space), show the behavior of a possible phenotype’s reaction flow and its associated influence (increase or decrease) on the target metabolite of interest (Figure 2). An inverse triangle correlation indicates a knockout with biomass, an overexpression of the gene is predicted by a direct triangle, and a random scatter indicates that there is no correlation involved between biomass and the flux toward the metabolite of interest. In addition, POSYBEL simulations were performed with the BW25113 strain, whereas the BL21 strain was employed for isobutanol and shikimate production aside from initial testing. The models stick to the in vitro repertoire because these strains have comparable genes and pathways.

Figure 2.

Figure 2

Open in a new tab

Different states of individual bacteria in a population graph examined by the POSYBEL platform. POSYBEL analyzes the flux of a metabolite “through” a population of cells obtained using MCMC simulation. It employs FBA in numerous metabolic state combinations (matrices), resulting in a biomass or vs metabolite distribution. The platform displays diverse phenotypes of a single strain in a population. The “cells” or dots show the behavior of a possible phenotype’s reaction flow and its associated influence (increase or decrease) on the target metabolite of interest (red arrow).

Mimicking Isobutanol Production Using POSYBEL

The E. coli metabolic map reveals that a carbon flux is required to proceed toward higher isobutanol production (Figure 3). Also, the BL21 strain was used to produce isobutanol due to valine-feedback independent acetolactate synthase (ilvG).17,18 Furthermore, minimal media (M9) were used to maintain cell viability to determine a minimalistic system and prove the simulations as accurately as possible. The POSYBEL platform helped identify potential knockouts required for increasing the isobutanol yield. This selection was made by observing the flux through metabolite pathways, where their negligible flow (i.e., <10% of the maximum flux of the reaction in a population) essentially represented a knockdown. Hence, as depicted in Figure 4A, a matrix of multiple solutions using 105 iterations was executed to rule the best fluxes for isobutanol production. Each dot represents one solution among the solution space. The optimal knockouts (encircled in Figure 4A) for increased biomass against product yield for isobutanol were found to be ΔackA/ΔldhA/ΔadhE triple knockout, which has already been reported.19 This corroborated that the platform’s design was robust enough to conclude the findings published elsewhere. Consequently, shake-flask experiments were performed in minimal media to validate the results.

Figure 3.

Figure 3

Open in a new tab

Pathway for isobutanol and shikimate production. The genes colored blue are native to E. coli, whereas the genes labeled in red are heterologously expressed, and metabolites in green are products of interest. Most of the steps toward shikimate and isobutanol production are devoid of nitrogen.

Figure 4.

Figure 4

Open in a new tab

POSYBEL simulation and in vitro validation for isobutanol production. (A) Scatter plot of the population distribution for isobutanol production vs biomass comprising the triple knockouts adhE, ackA, and ldhE with the threshold considered (encircled). (B) The population of bacterial cells producing isobutanol increases as flux decreases across adhE, ackA, and ldhA genes. (C) Graphical representation of the initial shake flask experiment with the BL21 and its knockouts expressing KivD in minimal media (after growing the biomass in LB media). It represents a comparative analysis of the normalized HPLC peaks for understanding the flux of ethanol, acetate, lactate, formate, and isobutanol in wild-type BL21 and ΔackA/ΔadhE/ΔldhA triple knockout.

Moreover, to demonstrate the significance of the simulations, about 104 iterations were performed for ΔackA, ΔldhA, and ΔadhE in various combinations. This resulted in a scatter plot of the bacterial population producing isobutanol. Figure 4B shows the progression toward the “peak” of the population distribution as the flux reduces from wild type (BL21) to single, double, and eventually triple knockout of ΔackA, ΔldhA, ΔadhE. Hence, it is easier to understand the population behavior using this platform, unlike traditional simulations. Figure 4C shows the normalized high-performance liquid chromatography (HPLC) data for the wild-type (BL21) and ΔackA/ΔldhA/ΔadhE triple knockout.

Validating the Flux of the Model by Swapping away Nitrogen (N-Swap)

The root of an FBA model is the absence or presence of a metabolite of interest using mass balance. Similarly, in our population model, all individual solutions (depicting individuals in a population with varying levels of metabolites) collapse through the same mass balance. Therefore, the nitrogen swap is a practical extension of the model. Swapping away nitrogen invariably changes nitrogen-dependent metabolite flux through the population. Consequently, mass balance establishes itself through the formation of metabolites without nitrogen.

Producing Isobutanol by Swapping away Nitrogen (N-Swap)

Although the population model is not designed to simulate the nitrogen input at its current state, it is seen that the pathway toward the production of isobutanol is devoid of nitrogen. Figure 3 shows the number of nitrogen-free steps from glucose to the production of isobutanol (15 steps). Furthermore, glucose consumption remains the same under nitrogen and nitrogen-depleted conditions; however, the production of isobutanol is increased by almost 30% in media devoid of nitrogen (Supporting Information, Figure S2). Therefore, different concentrations of ammonium sulfate (the only nitrogen source in minimal media) was introduced to determine the best yield produced by isobutanol. There is an increased production of isobutanol at 3% nitrogen and a consequent decrease in acetate and lactate production in the triple knockout (Figure 4C).

POSYBEL Simulations of Shikimate Production

The POSYBEL platform was used to perform 104 iterations for finding the best fluxes, in which the shikimate yield is the highest. The optimal knockouts (encircled in Figure 5A) for increased biomass against product yield for shikimate were found to be ΔackA/ΔaroK/ΔaroL/ΔpoxB and ΔackA/ΔaroK/ΔaroL/ΔptsG. Furthermore, the E. coli BL21 gave a higher product yield than BW25113 in wild-type and knockouts (Supporting Information, Figure S1).

Figure 5.

Figure 5

Open in a new tab

POSYBEL simulation and in vitro validation for shikimate production. (A) Scatter plot for shikimate production comprising knockouts (encircled) as solutions for increasing shikimate yield. (B) Scatter plot representing fluxes through best predicted genes against the shikimate produced. It is observed that despite intuitively choosing ackA as a knockout for producing more shikimate, the POSYBEL output proved otherwise. It is detected that ptsG, poxB, pykA/F, and aroK/L can be knocked out to produce a high amount of shikimate, whereas an “intermediate” flux through ackA has higher shikimate than its knockout. (C) Normalized in vitro graphs show that acetate production is required for the flux to move toward shikimate production. Also, glucose cannot be consumed without nitrogen (LB) source.

Just like isobutanol, the individual knockouts were simulated against biomass. The individual scatter plots (Figure 5A) show an inverse triangle relationship for aroL (and subsequently aroK as it performs the same reaction), poxB and ptsG. Hence, in these cases, knockouts are essential for increasing product yield. One may assume that knocking out acetate production (ackA) in Figure 2 would divert the flux toward shikimate production. However, this is counterintuitive, as seen in Figure 5B, where a Gaussian curve is observed concerning ackA; thus, a small amount of acetate is required for producing higher shikimate. The scatter plots for knockouts showed higher shikimate production. The shake flask experiments were performed with the previously predicted ΔackA/ΔaroK/ΔaroL knockout along with ΔackA/ΔaroK/ΔaroL/ΔpoxB and ΔackA/ΔaroK/ΔaroL/ΔptsG quadruple knockouts (Figure 5B). However, the yield of shikimate was the highest in the triple knockout at 20% of LB in minimal media. This result is discussed below.

Production of Shikimate by Swapping away Nitrogen (N-Swap)

It is noteworthy that even in the case of shikimate, just as isobutanol, the pathway is devoid of nitrogen (Figure 2). Therefore, the nitrogen source was modulated in minimal media. However, unlike isobutanol production, the swapping of nitrogen is not straightforward for obtaining more shikimate. In the former, the valine biosynthesis can continue with limited nitrogen, and the carbon flux can be diverted toward the production of acetolactate (and subsequently isobutanol) using microaerophilic conditions. However, with shikimate, this approach fails as a stoichiometric deficit is observed. For increasing, shikimate production as aromatic amino acid synthesis (AAA) is downstream to it, unlike valine and other branched amino acid synthesis (BCAA).

The experiments were done with various nitrogen source (LB provides AAAs) concentrations, of which the 20% LB was found to be the optimal concentration for maximum product yield. It is also seen that there is a negligible flux of glucose without the nitrogen source. This shows that a certain amount of nitrogen is required for driving the carbon flux toward shikimate production. Figure 5B also shows the limited acetate required for higher shikimate yield in ΔackA/ΔaroK/ΔaroL. Adding poxB and ptsG knockouts reduces acetate production by 68% (see Supporting Information, Table S2). This, in turn, increases the shikimate production in the presence of 20% LB (Figure 5B).

Final Product Yield Using Bioreactor

A reasonably direct correlation was observed for producing isobutanol, where 3% nitrogen-containing M9 minimal media produced 2.23 g/L after 24 h (Figure 6A). In the case of shikimate, although the quadruple knockouts show that POSYBEL simulations help understand the population behavior and flux of carbon, we sought the best triple knockouts that gave a higher shikimate yield. The experiment was performed in a bioreactor with 20% LB and 0.8% or 1.6% glucose, respectively, where the maximum shikimate yield was obtained with the triple knockout after 24 h at 1.65 g/L (Figure 6B). Furthermore, POSYBEL simulations were performed to predict higher shikimate yield. TktA and PntAB were found as ideal candidates. Interestingly, TktA has reaction flux in both directions; therefore, overexpression of the enzyme may reduce or increase shikimate production. The PntAB expression directly correlates with shikimate yield (Figure 6C). The production of shikimate in ΔackA/ΔaroK/ΔaroL was doubled by TktA overexpression from 1.63 g/L at 12 h to 3.02 g/L at 24 h. Conversely, pntAB overexpression gave initially higher yields, that is, 1.13 g/L (at 12 h), but remained at a similar concentration (1.24 g/L) after 24 h (Figures 6C and 5D).

Figure 6.

Figure 6

Open in a new tab

Production of isobutanol and shikimate in the bioreactor under the preeminent tested conditions. (A) Isobutanol production using single, double, and triple knockouts by switching the LB grown biomass to minimal media with 0 and 3% nitrogen. (B) Production of shikimate (g/L) in BL21, ΔaroK/ΔaroL, and ΔackA/ΔaroK/ΔaroL. (C) POSYBEL simulation of genes (pntAB and TktA) possitively correlates to shikimate production. (D) Shikimate synthesis in the bioreactor using ΔackA/ΔaroK/ΔaroL with pntAB and tktA overexpression in 20% LB in the bioreactor.

Population Simulation in the Presence of a Metabolic Inhibitor

The inhibition of an essential gene by an inhibitor dictates the fate of bacteria. Interestingly, not all bacteria are killed in this niche; rather, a small population of metabolically deviant individuals survive the onslaught of the introduced compound resulting in the formation of persisters or biofilms. To observe the behavior of the populace against antibacterials or metabolic inhibitors, glyphosate was chosen as a compound of interest. It inhibits the aroA gene (3-phosphoshikimate 1-carboxyvinyltransferase), essential in synthesizing essential AAAs. Consequently, aroA flux simulations with POSYBEL were performed against biomass.

Interestingly, there was no change in the biomass until 95% (cyan) of aroA flux was decreased. Therefore, to show that the population is impacted by the 95 and 99% (dark blue) knockdown of the target aroA flux, we present it against the biomass, which shows a drastic decline at 95 and 99% inhibition (Figure 7B). We also present the input glucose flux versus biomass growth to show that the conditions remain similar, and the decline in biomass is only due to aroA knockdown (Figure 7A).

Figure 7.

Figure 7

Open in a new tab

Behavior of bacterial population toward glyphosate and glucose. (A) Scatter plot depicting the glucose flux across spectrum of aroA inhibition where, at 5% (cyan color) concentration of the metabolite (aroA), the population decreases slightly compared to higher concentrations. Comparatively, there is a significant decrease in the biomass at 1% aroA flux (dark blue), where the glucose flux still remains unchanged. (B) When exposed to a metabolic inhibitor (glyphosate inhibits aroA), at 5% aroA flux (i.e., 95% inhibition of flux), the biomass and flux through are significantly reduced (cyan). It is exacerbated further at 1% as well (dark blue), thereby mimicking minimal bactericidal concentration assays, where 1% of the population remains as persisters after exposure to any antibiotic.

Discussion

Living organisms are complex and constantly evolving biochemical entities. Even simple prokaryotes such as E. coli encompass about 90% of non-essential enzymes, which play a role in the maintenance of the organism.20 Multiple genes produce the same metabolite of interest, making the life process an underdetermined system.21 Consequently, evolution is a constant process as multiple conceivable ways are present for reaching the same metabolic state. A static biochemical map does not reveal this dynamicity, and traditional FBA or ODE models give a singular solution forecasting the average behavior of the population. In this study, in the POSYBEL, the overall presence of a metabolite “through” a population of cells is considered. Hence, it entails using FBA in multiple metabolic state combinations (matrices), resulting in a normal distribution of substrate versus metabolite. It is noteworthy to remember that FBA considers only forward reactions (k1) for the enzymes as the reversible reaction rate (k2) is comparably negligible (k1 ≫ k2).

Therefore, proceeding with exclusively forward reactions is a favorable solution for emulating an E. coli population. Although this implies intensive computational power, it helps determine combinations that are not always intuitive. When we use a scatter plot, the platform shows different phenotypes of a single strain in its population. The “cells” or points in the plot (picked from the solution space) represent the behavior of a probable phenotype’s reaction flux and its corresponding effect (increase or decrease) on the target metabolite of interest. Generally, three types of patterns are observed viz., an inverse triangle correlation signifies a knockout, a direct triangle predicts an overexpression of the gene, and a random scatter means there is no correlation involved for producing a metabolite of interest. Also, POSYBEL simulations were done with the BW25113 strain, whereas, apart from initial experimentation, the BL21 strain was used for producing isobutanol and shikimate. However, these strains have near-identical genes and pathways. Hence, the models stick to the in vitro repertoire. Additionally, the BL21 strain has a valine-feedback-independent acetolactate synthase (ilvG). This, in addition to the heterogeneously expressed ketoisovalerate decarboxylase (KivD), is better suited than the K-12 (BW25113) strain for isobutanol production.

Similarly, BL21 gave a better shikimate yield than BW25113 (Supporting Information, Figure S1), as BL21 growth in rich media (such as LB) is rapid and yields higher biomass. Moreover, rapid glucose uptake results in the secretion of a large amount of acetate in the media. However, acetate uptake is significantly more efficient in BL21 along with variations in central metabolic pathway enzymes. This is reflected in high levels of lower glycolytic pathway enzymes such as pyruvate dehydrogenase, low levels of TCA cycle enzymes, and high levels of the acetate-forming enzymes such as phosphate acetyl transferase (pta) and acetate kinase (ackA).22 Furthermore, the global transcription regulator Cra (FruR) is constitutively expressed in BL21 that is hypothesized to regulate central metabolic pathways.23 The proteome of exponentially developing cells in glucose-supplemented mineral salt media is controlled by amino acid synthesis pathway enzymes, with more balanced abundances of core metabolic pathway enzymes and a smaller proportion of ribosomal and other translational proteins.22 Hence, the strain is used for overexpressing enzymes. In this case, the metabolic “quirk” of acetate uptake is fundamental for producing both isobutanol and shikimate.

The importance of a good population model can only be determined once it mirrors a real-world scenario for validation. Isobutanol is a well-reported, important industrial solvent/alternate fuel having a higher calorific value. Unlike ethanol, no engine modifications are required, making it an ideal fuel substitute. Furthermore, it is a cleaner fuel as it generates CO2 instead of SO2 or CO upon combustion. Atsumi et al. demonstrated that integrating the Ehrlich pathway into the BCAA pathway was sufficient to generate isobutanol in E. coli under non-fermentative conditions.24 Also, the addition of the heterologous gene kivD (ketoisovalerate decarboxylase) from L. lactis is required to produce isobutanal (Figure 2). This metabolite is converted to isobutanol by multiple native isobutyraldehyde dehydrogenases such as YqhD, AdhP, FucO, EutG, YaiY, BetA, EutE, and YjbB.25 Moreover, multiple studies have predicted E. coli knockouts that increase isobutanol yield.24,26,27 Therefore, simulations for producing the metabolite using POSYBEL were done to determine if it predicted the reported knockouts. POSYBEL simulations revealed 2770 solutions of 100,000 (2.7%) iterations with negligible flux through adhE, ldhA, and ackA, giving high isobutanol yield, thereby signaling a triple knockout, ΔadhE/ΔackA/ΔldhA (Figure 4a). Other viable knockouts predicted by the platform have been listed in Supporting Information, Table S1.

The in vitro experiments were performed in minimal media (M9) after initially obtaining the biomass in LB media. The former mimics the “closest” environment concerning the in silico predictions as they are well-defined media. Therefore, it is ideal for estimating the validity of the POSYBEL platform. The BL21 ΔadhE/ΔackA/ΔldhA gave ten times more isobutanol yield than the wild type. However, the yield was increased upon limiting the nitrogen uptake by swapping away the nitrogen source of ammonium sulfate (N-swap). Of course, nitrogen is essential for cellular growth and part of proteins and DNA/RNA. Hence, under nitrogen-depleted conditions, the growth of the cell is halted. However, because carbon, hydrogen, and oxygen are present, the metabolite flux does proceed through nitrogen-independent pathways. Although we have not designed the platform to simulate this, nitrogen modulation has been used in other instances for increasing metabolite production.28 The pathway involved in producing the isobutanol does not have the element (Figure 2). Therefore, the cells were incubated in minimal media with and without nitrogen to divert the carbon flux toward isobutanol. Initial experiments showed a higher yield in media devoid of nitrogen (Supporting Information Figure S2). However, the best output was at 3% nitrogen, where biomass loss was minimal (Figure 4C). Microaerophilic conditions’ impact was found using POSYBEL simulations on oxygen uptake and isobutanol yield (Supporting Information, Figure S3). It was observed that a limited amount of oxygen could help increase isobutanol yield. The study of flux in population revealed a simple logic. N-swap under microaerophilic growth conditions results in nitrate formation that inactivates the enzyme dihydroxy-acid dehydratase (IlvD). Dinitrosyl iron complex-bound IlvD is an inactive enzyme complex making the bacteria a BCAA auxotroph, effectively halting isobutanol production. This complex is activated under the aerobic condition without forming a new enzyme.29 However, the NO formation is stunted under limiting or zero nitrogen conditions, thereby keeping the flux active through the BCAA pathway. The conditions were repeated in the bioreactor with BL21, ΔackA, ΔackA/ΔadhE, and ΔadhE/ΔackA/ΔldhA knockouts, where the maximum isobutanol yield of 2.23 g/L was obtained using the triple knockout in 3% nitrogen.

Upon validating the POSYBEL platform, it was employed to predict strains for increasing shikimate, a high-value industrial precursor for producing glyphosate and the antiviral Tamiflu.30 It is manufactured using plant sources with similar biosynthetic pathways; for example, star anise (Illicium anisatum) and sweetgum (Liquidambar styraciflua) have been used for extracting shikimate at 1.5 and 3.7% w/v, respectively.31 In E. coli, it is obtained by knocking out genes such as aroK and aroL, which participate in AAA biosynthesis.32 However, a limited yield has been reached using these knockouts; therefore, we used POSYBEL to assess if a higher product yield can be availed. About 104 iterations were done for shikimate production, of which 300 solutions signaled an increased shikimate yield (Figure 5A). POSYBEL predicted knockouts, namely, aroK, aroL, poxB, and ptsG. Although ackA knockout seems like an intuitive step for diverting the flux toward shikimate (Figure 2), the platform showed this to be otherwise (Figure 5B). Acetate was found to be required in limited quantities to increase the production of shikimate. Furthermore, just like isobutanol, shikimate production entails a lack of nitrogen as well. However, unlike isobutanol synthesis, the metabolite is a direct precursor of essential AAA production. Therefore, LB media were used to supplant AAAs and acetate in minimal media at 5, 10, and 20%, respectively. Shikimate was best produced in 20% LB media in ΔackA/ΔaroK/ΔaroL compared to 5, 10, and 100% LB media. Thereby, nitrogen and acetate play an essential role in producing the metabolite (Figure 5C). Also, ΔackA/ΔaroK/ΔaroL/ΔptsG and ΔackA/ΔaroK/ΔaroL/ΔpoxB produced less shikimate compared to the triple knockout (see Supporting Information, Table S2). Furthermore, BL21 was used to generate knockouts as it showed a higher yield than BW25113 (Supporting Information, Figure S1). The triple knockout was used in a bioreactor to produce shikimate at 1.63 g/L at 20% LB and 1.6% glucose.

However, we sought to increase the production further using POSYBEL. It predicted genes whose overexpression can help in increasing shikimate yield. Interestingly, the counterintuitive prediction of the platform again plays a significant role in this context. It forecasts overexpression of two genes, namely, pntAB and tktA, respectively. The former signifies equal forward and reverse reactions, which are generally not considered due to their dubious role in directing the flux, whereas the latter is associated with shikimate production (Figure 6C). Overexpression of both the genes yielded a significantly higher amount of shikimate (3.02 g/L) than the triple knockout, thereby signifying the importance of understanding the complexity of population for metabolite production in a bioreactor. It should be noted that after 24 h, the amount of shikimate produced by the triple knockout ΔackA/ΔaroK/ΔaroL and ΔackA/ΔaroK/ΔaroL with pntAB OE is similar. This is because the phosphorylation of NAD is catalyzed by two enzymes, namely PntAB and NadK; the pntAB overexpression is useful in maintaining the cofactor balance (thereby maintaining glycolysis and subsequently shikimate production), thus increasing the production of shikimate. However, after 24 h, this is compensated by the natively produced PntAB in ΔackA/ΔaroK/ΔaroL (wildtype). On the other hand, tktA (Figure 6B) helps direct the flux toward glycolysis. Unlike cofactors, glucose is available in surplus, thus producing an increasing amount of shikimate after 24 h in the TktA overexpressing strain.

Another commercially important aspect of bacteria is the inverse of biomass production, that is, their inhibition or death. This aspect is critical in drug discovery, where high-throughput screening of millions of compounds is undertaken against bacteria to obtain minimum inhibitory and minimum bactericidal concentrations (MBCs) of a molecule. In such a scenario, the significance of a drug target plays an essential role in shaping the structure–activity relationship of a molecule of interest. The platform has already demonstrated that varying metabolic states play a crucial role in the evolution of bacteria in a specific direction. This stands true for growth conditions, where an antibiotic or an inhibitor fails to completely decimate the bacterial population as about 1–10% of the population is at a static metabolic phase. Upon emulating the aroA inhibitor glyphosate, the platform demonstrates this phenomenon. There is little to no change in biomass until 95% of the flux is reduced (mimicking inhibition), and the biomass is significantly affected (cyan) in Figure 7A. Concurrently, at 99% inhibition of the flux (dark blue), the number and growth of the biomass are reduced to a tiny fraction of the population. As a control, when glucose flux is measured under similar conditions, despite the reduction in biomass, there is little to no change in the growth of the limited number of cells (Figure 7B). The MBC data of glyphosate against E. coli, which we reported elsewhere, clearly demonstrate this phenomenon, where a tiny fraction of E. coli is unaffected by glyphosate even after 24 h.8 This demonstrates the metabolic vigor of a bacterial population, where, often, different states play different roles in terms of the survival and evolution of the organism. Here, we have obligingly attempted to propose a policy that considers the minimal amount of input with nominal gene manipulations necessary to produce metabolites of interest or demonstrate the role of a gene in conferring persister formation.

Conclusions

Nature and art go hand in hand, and often one mimics the other to demonstrate the beauty of complexity and diversity. The POSYBEL platform is no exception, where the flux of glucose (and therefore oxygen, hydrogen, and carbon) is observed using multiple iterations, thereby giving a diverse solution space. Further modifications in the system by introducing macroelements such as phosphorous and nitrogen can help strengthen the method even more. Both ODEs that represent various enzymes that describe the cellular interactions with their intrinsic kinetic parameters or a flux-based model essentially define the functionality of a single cell and assume homogeneity in an isogenic population. Bacteria are not only asynchronous physiologically but no two cells in a population are in metabolic congruence. Living organisms evolve in a chaotic environment. Unlike conventional FBA, the optimal solutions derived from maximization or minimization of a particular reaction give an understanding of a system required to achieve a theoretical maximum/minimum in a utopian environment. Instead, it can be visualized as an ensemble of underdetermined equations that connect metabolites, producing an infinite array of possible solutions. Each cell in the population may use one set of these solutions. This tacitly necessitates that no two cells have identical expression levels in a given environment. These varying behavioral signatures enable the system to be robust enough to handle stress factors such as nutritional deficiency, osmotic imbalance, temperature shock, or the presence of antibiotics. This kind of competitive growth “advantage–disadvantage” simulation can be generated using POSYBEL. Studies show that cells diverge in their growth rate even in media with optimum nutrient availability and rapidly move away from being synchronous. This divergence is seen experimentally and is a natural outcome of our platform.

Materials and Methods

Building the Population System Biology Model (POSYBEL)

In an FBA model, a matrix (M) of size y*z is generated with stoichiometric metabolic reactions. The compounds participating in the metabolic reactions mark the rows of the matrix (y unique compounds participating in the system), and the columns represent the reactions (z overall reactions in the system). The matrix is constructed using the stoichiometric coefficients of each of the respective metabolites participating in a reaction, traversing left to right one column at a time. A positive coefficient is considered when a metabolite is produced, and a negative coefficient for consumption. The number of metabolites participating in each reaction is limited, which leads to the development of a sparse matrix. The flux through each of the reactions is an unknown and is represented by a vector k of length z. Under steady state conditions, the rate of change of concentration of each of the metabolites is 0. Thus, any solution of k that satisfies the equation M*k = 0 is part of the null space of M (Figure 1D). The system of equations obtained by steady-state mass balance is such that the total number of equations is y (equal to the no. of metabolites), and the unknown variable is the flux of each reactions participating in the system z (no. of reactions). Given any large metabolic model, the number of reactions is always greater than the metabolites. When the number of unknowns is greater than the equations, we can identify many plausible values for the unknowns that are solutions of the system of linear equations. Such a system of linear equations is known to be underdetermined. With the help of constraints, we can define the range for the solution space. The constraints are designed using the composition of the media used for the growth of the bacteria.

In a conventional FBA approach, an objective function is designed that works for maximization or minimization of a particular reaction, giving the contribution of each of the reactions in the system toward the maximum or minimum flux of the target reaction. Individually, most enzymes, intrinsically, are reversible, but the forward reaction is significantly higher than the reverse reaction. For our model, we have made it unidirectional for computational convenience. Figure 1D shows the reduction of solution space (red), where eventually an optimal solution (orange) is obtained in traditional FBA by maximizing or minimizing the target reactions. Instead, in the POSYBEL platform, the samples are fetched from the entire constrained solution space using the random walk methodology of the MCMC method. The high dimensional nature of the system of equations as found in metabolic network models makes the MCMC-based approaches to be efficient for picking samples from the null space. The inequality constraints define the boundary in the feasible region (hyperplane). All the points on one side of the hyperplane satisfy the inequality and are a feasible solution, and the points that fall on the other side of the inequality form infeasible solutions. The solutions that are considered for the analysis are those that satisfy all the inequalities and are found to be present in the constrained feasible region. A point is selected within the solution space, where all the inequalities are satisfied acting as a seed, the sampling of the new point is done from a normal distribution with mean zero and a fixed standard deviation (jump length). The new points selected randomly depend only on the previous point. The new point is accepted if it satisfies the inequality, else we go back to the previous point and try again. The xsample () function available in the R, part of the LIM package, allows us to implement this algorithm.33

We have created the shikimate model from IJO1366 with the addition of a reaction corresponding to the export of the shikimate metabolite.20 In the absence of a transport flux, the system considers the shikimate metabolite to be unused when reactions contributing to the utilization of the shikimate are knocked out, bringing down the shikimate flux to 0. This prevents us from arriving at knockouts that can maximize shikimate. In a different approach, one needs not to optimize the system of linear equation for a given objective function but try and see the solution distribution over a very large set of solutions within some boundary. This method tries to work around the optimization problem, where it is inherently assumed that the system has some definitive but unknown intelligence to work toward (turn on genes expressions suitably) to attain maximization or minimization of the unique function. The modified approach that is computationally very intensive looks to set up a bounding set of minimal constraints, so that all solutions that are possible to lie within a bound and then try and generate sample solutions from the entire solution space obtaining a matrix of multiple solutions (POSYBEL) using an algorithm based on Markov chain, and these data comprise of all the possible solution (ways), in which the organism can act (population behavior) corresponding to a given condition.34 The POSYBEL data comprising of 100,000 samples were developed (see Supporting Information, Methods 2 and 3 for scripting and output information). Once the population results are obtained (i.e., the matrix of multiple solutions), the system is filtered for the samples from the solution space with maximum production of the target metabolite. The maximum flux is identified, and all the samples which constitutes ∼90% and above of the maximum flux of target metabolite are filtered out. The maximum flux of all individual reactions is also computed. The next step is to identify the reaction fluxes that contribute least to the population level toward the production of target metabolite. We achieve this by adding a filter to sample out the reaction fluxes that run minimal ∼10% of the maximum flux for that respective reaction among the samples with maximum yield of target flux. These indicate knockouts for validating the platform in vitro.

Generation of Predicted E. coli Knockouts

Knockouts were generated using the P1 transduction method, which used donor and recipient strains for generating knockouts (see Supporting Information, Methods). The knockouts for various genes were availed from the Keio collection, which is a library of non-essential gene knockouts of E. coli and contains the kanamycin resistance gene in its place. The library was used as donor strains, and BL21 was used as the recipient strain. To further knockout the genes, the kanamycin marker is first flipped out by using the λ-red recombinase method. pCP20 plasmid is first transformed into the desired knockout and plated on the LB-amp (30 μg/mL) plate and incubated overnight. The colonies are then grown in Luria broth until they reached 0.6 OD600. Then, the culture is incubated at 37 °C for 1 hour, followed by incubation of 43 °C for 4 hours. It is then plated on plain LB media, media containing ampicillin, and media containing kanamycin, respectively. Growth of the culture on the LB plate and no growth on media having either ampicillin or kanamycin confirmed the absence of kanamycin cassette.

For example, to generate ΔackA/ΔadhE/ΔldhA knockout, the BW25113-ΔackA was used as the donor strain, and BL21 was used as the recipient strain. Subsequently, ΔadhE and ΔldhA were used to knock out the respective genes in BL21 after flipping out the kanamycin cassette before proceeding with gene knockouts. Knockouts were confirmed using colony PCR. The knockouts generated using this method were ΔackA/ΔadhEΔldhA, ΔackA/ΔaroK/ΔaroL, ΔackA/ΔaroK/ΔaroL/ΔptsG, and ΔackA/ΔaroK/ΔaroL/ΔpoxB, respectively.

Protocol for Isobutanol and Shikimate Production with Various Knockouts and Media Swap

Initially, the conformation of in silico simulations for nitrogen modulation was done in shake flasks before proceeding toward biotransformation in bioreactors. The desired knockouts and wild-type strains are then transformed with pUC57a kivD plasmid and plated on luria agar plates with ampicillin (100 μg/mL). A single transformant is then inoculated in 5 mL of LB media with ampicillin (100 μg/mL) and grown overnight. The pUC57-kivD plasmid, synthesized by Genscript, is engineered without an operator site and with a constitutive promoter, therefore, making the induction step void. For shikimate, the cells (or knockouts) are grown overnight in LB media.

Shake-Flask Bioconversion Experiments

To conduct the shake-flask experiments for both shikimate and isobutanol, the starter cultures are grown in LB (lysogeny broth) media. Later, they are transferred to LB media and grown until OD600 2.0 (secondary culture). The cells are then centrifuged at 4000g and transferred into nitrogen-deficient media (after washing with PBS to remove the carryover of spent media). In the case of isobutanol, the knockouts are transferred to M9 media with varying percentage of ammonium sulfate (nitrogen source) composition with 3.6% glucose (carbon source) and ampicillin (100 μg/mL) for sustaining the pUC57-kiVD plasmid. For shikimate production, the cultures are transferred into M9 media with varying percentage of LB with 1.6% glucose.

Protocol for Bioreactor

To get a higher product yield, the cells were grown in 500 mL bioreactor (Applikon miniBio). For producing isobutanol and shikimate, the cells were grown up to ∼6.0–6.5 OD600 in 20% dissolved oxygen (DO) at 200 rpm impeller speed. PEG400 was added as an antifoaming agent. To create microaerophilic conditions, the DO is reduced to 2.5%, and the impeller speed is reduced to 50 rpm to produce either isobutanol or shikimate.

Estimation of Cell Viability

Cell viability was estimated by plating the spent culture in LB and LB with ampicillin (100 μg/mL) plates. This is done to give a projection of the number of cells alive and the plasmid loss after various time points. Samples are sent for gas chromatography analysis/HPLC performed with the corresponding media standards. We estimate isobutanol and ethanol produced along with the remaining glucose concentration, as well as other metabolites such as formate, lactate, acetate, succinate, and pyruvate.

HPLC Analysis

For the detection of both shikimate and isobutanol, 1 mL of the culture was spun at 4000g for 5 min, followed by a further spinning of the supernatant at 14.8 g for 5 min. About 50 μL of supernatant was analyzed in HPLC. To perform HPLC, the Aminex 87H column (Biorad) was used as the stationary phase and 5 mM H2SO4 was used as the mobile phase at 0.750 mL/min flow rate with a refractive index detector, which is helpful in detecting monosaccharides and organic acids at the same time.

Data Availability Statement

Program using R programming (R development core team, 2013) and python (Python Software Foundation, https://www.python.org) is built to filter the possible knockouts from a population.35 The GitHub links for the POSYBEL platform for shikimate, isobutanol synthesis, and bacterial inhibition by glyphosate are as follows: https://github.com/sreenathrajagopal/shikimatemodel_codes.git, https://github.com/sreenathrajagopal/Isobutanolmodel_codes.git, https://github.com/sreenathrajagopal/Glyphosate_study.git.

Acknowledgments

The financial support from the Department of Biotechnology (DBT)-BIG grant (grant no. BIRAC/CCAMP0186/BIG-04/14) to Janani Venkataraman and the DBT-BBSRC for RICEFUEL (grant no. BT/IN/Indo-UK/SuBB/21SSY/2013) grant to R.K.S. and S.D. is gratefully acknowledged. We thank Swetha Suresh for adding flair to the manuscript’s graphics. In addition, we acknowledge the initial computational work done by Madhura Adavkar and Vetri Selvi and some discrete but insightful experimental work by Rajeswari Basu.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.2c00097.

  • Programming POSYBEL for isobutanol and shikimate production; influence of oxygen on the production of shikimate and isobutanol; and list of alternate knockouts for producing isobutanol (PDF)

Author Contributions

S.R., R.V.H., D.M., A.G., and A.P.S. contributed equally. R.K.S. drove the initial formulation and validation of the population model. S.R. programmed the in silico platform and conducted the cross-validation along with R.K.S., D.M., and R.V.H. validated the platform by performing the experiments for isobutanol and shikimate production, respectively. A.G. performed the scaled-up version of the experimental procedures. A.P.S. assisted S.R. and D.M. in conducting the experiments and was one of the major contributors in pooling the diverse experimental data into a manuscript format. N.K. supervised the making of the multiple constructs and directed the experimental platform for isobutanol production, while J.V. oversaw the shikimate project. S.D. conceived the idea for the population model and nitrogen swap in media.

The authors declare no competing financial interest.

Notes

Code availability: the complete source code, and the models built to simulate the knockout (KO) studies presented in the manuscript, could be downloaded from the GitHub repository. The links are https://github.com/sreenathrajagopal/shikimatemodel_codes.git, https://github.com/sreenathrajagopal/Isobutanolmodel_codes.git. The README file contains a detailed description of the usage of the scripts. Any further queries related to the codes, bugs, and the model could be addressed to the corresponding author.

Notes

The authors declare that all data generated, analyzed, and used in this study are included in this article. All accompanying information related to model outputs and raw data for the studies could be accessed via the Github repository. Any additional details are available upon reasonable request and addressed to the corresponding author.

Supplementary Material

sb2c00097_si_001.pdf (837.8KB, pdf)

References

  1. Latendresse M.; Krummenacker M.; Trupp M.; Karp P. D. Construction and Completion of Flux Balance Models from Pathway Databases. Bioinformatics 2012, 28, 388–396. 10.1093/bioinformatics/btr681. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Kujawska A.; Kujawski J.; Bryjak M.; Kujawski W. ABE Fermentation Products Recovery Methods—A Review. Renewable Sustainable Energy Rev. 2015, 48, 648–661. 10.1016/j.rser.2015.04.028. [DOI] [Google Scholar]
  3. Orth J. D.; Thiele I.; Palsson B. Ø. What Is Flux Balance Analysis?. Nat. Biotechnol. 2010, 28, 245–248. 10.1038/nbt.1614. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Raman K.; Yeturu K.; Chandra N. TargetTB: A Target Identification Pipeline for Mycobacterium Tuberculosis through an Interactome, Reactome and Genome-Scale Structural Analysis. BMC Syst. Biol. 2008, 2, 1–21. 10.1186/1752-0509-2-109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bowden A. C. Metabolic Control Analysis in Biotechnology and Medicine. Nat. Biotechnol. 1999, 17, 641–643. 10.1038/10854. [DOI] [PubMed] [Google Scholar]
  6. Kim O. D.; Rocha M.; Maia P. A Review of Dynamic Modeling Approaches and Their Application in Computational Strain Optimization for Metabolic Engineering. Front. Microbiol. 2018, 9, 1–22. 10.3389/fmicb.2018.01690. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Mannan A. A.; Toya Y.; Shimizu K.; McFadden J.; Kierzek A. M.; Rocco A. Integrating Kinetic Model of E. Coli with Genome Scale Metabolic Fluxes Overcomes Its Open System Problem and Reveals Bistability in Central Metabolism. PLoS One 2015, 10, e0139507 10.1371/journal.pone.0139507. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Barve A.; Gupta A.; Solapure S. M.; Kumar A.; Ramachandran V.; Seshadri K.; Vali S.; Datta S. A Kinetic Platform for in Silico Modeling of the Metabolic Dynamics in Escherichia Coli. Adv. Appl. Bioinf. Chem. 2010, 3, 97–110. 10.2147/AABC.S14368. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Feist A. M.; Palsson B. O. The Biomass Objective Function. Curr. Opin. Microbiol. 2010, 13, 344–349. 10.1016/J.MIB.2010.03.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Wood T. K.; Knabel S. J.; Kwan B. W. Bacterial Persister Cell Formation and Dormancy. Appl. Environ. Microbiol. 2013, 79, 7116. 10.1128/AEM.02636-13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Yang J.; Tauschek M.; Robins-Browne R. M. Control of Bacterial Virulence by AraC-like Regulators That Respond to Chemical Signals. Trends Microbiol. 2011, 19, 128–135. 10.1016/j.tim.2010.12.001. [DOI] [PubMed] [Google Scholar]
  12. Leygeber M.; Lindemann D.; Sachs C. C.; Kaganovitch E.; Wiechert W.; Nöh K.; Kohlheyer D. Analyzing Microbial Population Heterogeneity—Expanding the Toolbox of Microfluidic Single-Cell Cultivations. J. Mol. Biol. 2019, 431, 4569–4588. 10.1016/j.jmb.2019.04.025. [DOI] [PubMed] [Google Scholar]
  13. Labhsetwar P.; Melo M. C. R.; Cole J. A.; Luthey-Schulten Z. Population FBA Predicts Metabolic Phenotypes in Yeast. PLoS Comput. Biol. 2017, 13, e1005728 10.1371/journal.pcbi.1005728. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Labhsetwar P.; Cole J. A.; Roberts E.; Price N. D.; Luthey-Schulten Z. A. Heterogeneity in Protein Expression Induces Metabolic Variability in a Modeled Escherichia Coli Population. Proc. Natl. Acad. Sci. 2013, 110, 14006–14011. 10.1073/pnas.1222569110. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Zeng H.; Yang A. Modelling Overflow Metabolism in Escherichia Coli with Flux Balance Analysis Incorporating Differential Proteomic Efficiencies of Energy Pathways. BMC Syst. Biol. 2019, 13, 1–18. 10.1186/s12918-018-0677-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Herrmann H. A.; Dyson B. C.; Vass L.; Johnson G. N.; Schwartz J.-M. Flux Sampling Is a Powerful Tool to Study Metabolism under Changing Environmental Conditions. npj Syst. Biol. Appl. 2019, 5, 1–8. 10.1038/s41540-019-0109-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Marisch K.; Bayer K.; Scharl T.; Mairhofer J.; Krempl P. M.; Hummel K.; Razzazi-Fazeli E.; Striedner G. A Comparative Analysis of Industrial Escherichia Coli K–12 and B Strains in High-Glucose Batch Cultivations on Process-, Transcriptome- and Proteome Level. PLoS One 2013, 8, e70516 10.1371/JOURNAL.PONE.0070516. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Savrasova E. A.; Kivero A. D.; Shakulov R. S.; Stoynova N. V. Use of the Valine Biosynthetic Pathway to Convert Glucose into Isobutanol. J. Ind. Microbiol. Biotechnol. 2011, 38, 1287–1294. 10.1007/S10295-010-0907-2. [DOI] [PubMed] [Google Scholar]
  19. Lan E. I.; Liao J. C. Microbial Synthesis of N-Butanol, Isobutanol, and Other Higher Alcohols from Diverse Resources. Bioresour. Technol. 2013, 135, 339–349. 10.1016/j.biortech.2012.09.104. [DOI] [PubMed] [Google Scholar]
  20. Orth J. D.; Conrad T. M.; Na J.; Lerman J. A.; Nam H.; Feist A. M.; Palsson B. Ø. A Comprehensive Genome-Scale Reconstruction of Escherichia Coli Metabolism-2011. Mol. Syst. Biol. 2011, 7, 535. 10.1038/msb.2011.65. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Keseler I. M.; Mackie A.; Santos-Zavaleta A.; Billington R.; Bonavides-Martínez C.; Caspi R.; Fulcher C.; Gama-Castro S.; Kothari A.; Krummenacker M.; Latendresse M.; Muñiz-Rascado L.; Ong Q.; Paley S.; Peralta-Gil M.; Subhraveti P.; Velázquez-Ramírez D. A.; Weaver D.; Collado-Vides J.; Paulsen I.; Karp P. D. The EcoCyc Database: Reflecting New Knowledge about Escherichia Coli K-12. Nucleic Acids Res. 2017, 45, D543–D550. 10.1093/NAR/GKW1003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Li Z.; Nimtz M.; Rinas U. The Metabolic Potential of Escherichia Coli BL21 in Defined and Rich Medium. Microb. Cell Fact. 2014, 13, 1–17. 10.1186/1475-2859-13-45/FIGURES/6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Son Y.-J.; Phue J.-N.; Trinh L. B.; Lee S.; Shiloach J. The Role of Cra in Regulating Acetate Excretion and Osmotic Tolerance in E. Coli K-12 and E. Coli B at High Density Growth. Microb. Cell Fact. 2011, 10, 52. 10.1186/1475-2859-10-52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Atsumi S.; Hanai T.; Liao J. C. Non-Fermentative Pathways for Synthesis of Branched-Chain Higher Alcohols as Biofuels. Nature 2008, 451, 86–89. 10.1038/nature06450. [DOI] [PubMed] [Google Scholar]
  25. Jarboe L. R. YqhD: A Broad-Substrate Range Aldehyde Reductase with Various Applications in Production of Biorenewable Fuels and Chemicals. Appl. Microbiol. Biotechnol. 2011, 89, 249–257. 10.1007/s00253-010-2912-9. [DOI] [PubMed] [Google Scholar]
  26. Smith K. M.; Liao J. C. An Evolutionary Strategy for Isobutanol Production Strain Development in Escherichia Coli. Metab. Eng. 2011, 13, 674–681. 10.1016/J.YMBEN.2011.08.004. [DOI] [PubMed] [Google Scholar]
  27. Atsumi S.; Wu T.-Y.; Eckl E.-M.; Hawkins S. D.; Buelter T.; Liao J. C. Engineering the Isobutanol Biosynthetic Pathway in Escherichia Coli by Comparison of Three Aldehyde Reductase/Alcohol Dehydrogenase Genes. Appl. Microbiol. Biotechnol. 2010, 85, 651–657. 10.1007/S00253-009-2085-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Anfelt J.; Kaczmarzyk D.; Shabestary K.; Renberg B.; Rockberg J.; Nielsen J.; Uhlén M.; Hudson E. P. Genetic and Nutrient Modulation of Acetyl-CoA Levels in Synechocystis for n-Butanol Production. Microb. Cell Fact. 2015, 14, 1–12. 10.1186/S12934-015-0355-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Ren B.; Zhang N.; Yang J.; Ding H. Nitric Oxide-Induced Bacteriostasis and Modification of Iron-Sulphur Proteins in Escherichia Coli. Mol. Microbiol. 2008, 70, 953–964. 10.1111/j.1365-2958.2008.06464.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Lew W.; Chen X.; Kim C. U. Discovery and Development of GS 4104 (Oseltamivir): An Orally Active Influenza Neuraminidase Inhibitor. Curr. Med. Chem. 2000, 7, 663–672. 10.2174/0929867003374886. [DOI] [PubMed] [Google Scholar]
  31. Enrich L. B.; Scheuermann M. L.; Mohadjer A.; Matthias K. R.; Eller C. F.; Newman M. S.; Fujinaka M.; Poon T. Liquidambar Styraciflua: A Renewable Source of Shikimic Acid. Tetrahedron Lett. 2008, 49, 2503–2505. 10.1016/j.tetlet.2008.02.140. [DOI] [Google Scholar]
  32. Chen K.; Dou J.; Tang S.; Yang Y.; Wang H.; Fang H.; Zhou C. Deletion of the AroK Gene Is Essential for High Shikimic Acid Accumulation through the Shikimate Pathway in E. Coli. Bioresour. Technol. 2012, 119, 141–147. 10.1016/j.biortech.2012.05.100. [DOI] [PubMed] [Google Scholar]
  33. Van den Meersche K.; Soetaert K.; Oevelen D. V. xsample(): AnRFunction for Sampling Linear Inverse Problems. J. Stat. Softw. 2009, 30, 1–15. 10.18637/jss.v030.c01.21666874 [DOI] [Google Scholar]
  34. van Oevelen D.; van den Meersche K.; Meysman F. J. R.; Soetaert K.; Middelburg J. J.; Vézina A. F. Quantifying Food Web Flows Using Linear Inverse Models. Ecosystems 2010, 13, 32–45. 10.1007/s10021-009-9297-6. [DOI] [Google Scholar]
  35. R core team . The R Project for Statistical Computing, 2013. https://doi.org/Rprogram.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

sb2c00097_si_001.pdf (837.8KB, pdf)

Data Availability Statement

Program using R programming (R development core team, 2013) and python (Python Software Foundation, https://www.python.org) is built to filter the possible knockouts from a population.35 The GitHub links for the POSYBEL platform for shikimate, isobutanol synthesis, and bacterial inhibition by glyphosate are as follows: https://github.com/sreenathrajagopal/shikimatemodel_codes.git, https://github.com/sreenathrajagopal/Isobutanolmodel_codes.git, https://github.com/sreenathrajagopal/Glyphosate_study.git.

Articles from ACS Synthetic Biology are provided here courtesy of American Chemical Society

RESOURCES