Engineering microbial chemical factories using metabolic models
- 251 Downloads
Living organisms in analogy with chemical factories use simple molecules such as sugars to produce a variety of compounds which are necessary for sustaining life and some of which are also commercially valuable. The metabolisms of simple (such as bacteria) and higher organisms (such as plants) alike can be exploited to convert low value inputs into high value outputs. Unlike conventional chemical factories, microbial production chassis are not necessarily tuned for a single product overproduction. Despite the same end goal, metabolic and industrial engineers rely on different techniques for achieving productivity goals. Metabolic engineers cannot affect reaction rates by manipulating pressure and temperature, instead they have at their disposal a range of enzymes and transcriptional and translational processes to optimize accordingly. In this review, we first highlight how various analytical approaches used in metabolic engineering and synthetic biology are related to concepts developed in systems and control engineering. Specifically, how algorithmic concepts derived in operations research can help explain the structure and organization of metabolic networks. Finally, we consider the future directions and challenges faced by the field of metabolic network modeling and the possible contributions of concepts drawn from the classical fields of chemical and control engineering. The aim of the review is to offer a current perspective of metabolic engineering and all that it entails without requiring specialized knowledge of bioinformatics or systems biology.
KeywordsSystems biology Metabolic engineering Flux balance analysis Control theory
Dynamic Flux Balance Analysis
Flux Balance Analysis
Metabolic Control Analysis
Max-min Driving Force
Metabolic Flux Analysis
Chemical engineering draws from a wide range of disciplines such as physics, chemistry, computer science, mathematics, operations research, and the life sciences. Almost seven decades ago during World War II the rising demands for penicillin was met by co-opting chemical engineering principles to carry out large-scale fermentation. A quadruple rise in production was achieved by refining the original mold species, developing an effective growth medium, and changing the fermentation process from rudimentary surface culture to submersion in tanks .
These early efforts were the vanguards of current mechanistic descriptions of biological processes. Cellular metabolism is a temporally-varying process that gives rise to a wide variety of dynamical phenomena such as multiple steady states and temporal oscillations. The elucidation, and subsequent prediction of the behavior of metabolic systems is one of the major challenges of the postgenomic era [2, 3, 4]. To this end, significant strides have been made in recent years to construct and investigate detailed models of cellular processes [5, 6, 7, 8]. Such a model can be treated as a “virtual laboratory” that allows one to build a characteristic description of the system and elucidate an understanding of the design principles of cellular functions, robustness, adaptability, and optimality. The insights gleaned can then be translated into the rational engineering of microbes to serve as miniature chemical factories to produce products of interest. Microbial fermentation is a popular production mode for many biofuels and biochemicals as it generally (1) relies on a sustainable feedstock (i.e., usually sugars), (2) involves a reduced environmental footprint, (3) is easily scalable, and (4) bypasses the need for complex separations.
The goal of this article is to review how chemical engineering is playing a germane role in the study of metabolic networks. We first describe the use of principles such as reaction kinetics and linear programming in modeling metabolic networks. This is followed by a description of the tools used to identify the processes that control and limit flux in biological systems. Finally, we discuss the challenges and opportunities associated with the successful establishment of microbial chemical factories.
Steady-state analysis of metabolic networks
A genome-scale metabolic model (GSM) includes internal metabolic conversions as well as the reactions transporting metabolites in and out of cells. Thus, reactions can be limited by the resources available in the environment, resulting in a flexible network structure that can adapt to different ecological niches and perturbations. For instance, an E. coli cell can survive with or without oxygen, investigations of which using FBA captured the varied routes of energy production and protein biosynthesis employed under the two regimes . The same metabolic model can be used in both cases by simply adjusting the bound of the reaction that transports oxygen into the cell.
FBA has been used to make significant contributions to understanding biochemical networks and metabolic engineering. Its primary goal is to design engineered organisms that can achieve higher efficiencies in metabolite overproduction through alterations in the flux distribution. This has been realized in numerous cases such as lycopene  and butanol  production in E. coli, and lysine production in Corynebacterium glutamicum . FBA can also be used to improve productivity such as by optimizing process parameters and rationally designing the cell culture medium. Optimal uptake rates were first computed using FBA and then used to alter the nutrient feed composition in a hybridoma culture which reduced waste accumulation by several folds . A metabolic model of E. coli was used to determine culture conditions that increased strain stability by optimizing the glucose to oxygen supply ratio .
FBA predictions have been found to achieve biological fidelity -- its ability to identify essential genes (i.e., genes whose deletion negates biomass synthesis) exceeds 90% in many metabolic models . Thus, such analyses can be used to glean insights into an organism’s physiology by examining its metabolism quantitatively. For instance, input vs output trade-offs can be computed that describe the impact of nutrient supply rates on by-product secretion and/or growth. Such a study in E. coli predicted growth rates to increase with the nutrient supply, as is expected, but at higher growth rates the secretion of metabolites such as acetate was also predicted . This is because the oxygen utilization capacity is reached at higher growth rates and the cell thus possesses surplus reductive potential. This leads to a redistribution of metabolic fluxes and by-products (such as acetate) are secreted so as to eliminate the surplus redox potential [20, 21, 22].
Dynamic models of metabolism
FBA considers metabolism using only reaction stoichiometry and reversibility subject to a steady-state condition. Thus, such a framework performs well when predicting the long-term response of the system to environment or genetic perturbations  but is unable to explain transient behavior or even how a specific metabolite state is achieved [24, 25]. In the absence of any regulatory or mechanistic information, the in silico solution space available to constraint-based models is much larger than the biologically feasible sample space [14, 15]. Thus, attempts to engineer microbial systems as in cell factories to overproduce metabolites must consider the kinetics associated with their production pathways along with the interaction of the designated pathways with the entire metabolic network.
Where X is the vector of metabolite concentrations (Xbiomass represents the concentration of the biomass metabolite), μ is the growth rate, and wj is the vector of weights associated with the objective function in the current time interval ∆T. t0 and tf represent the initial and final time points. The non-linearity in this formulation arises from the kinetic expressions for flux rates contained in the vector c(v(t), X(t)), usually used to constrain input fluxes. The flux values so obtained are considered to be constant over the current time interval and are then used to solve a dynamic model describing metabolite time profiles. Mahadevan et al.  used DFBA to describe the biphasic growth of E.coli on glucose and acetate. They could successfully predict the onset of acetate production and sequential substrate utilization where E. coli preferred glucose followed by acetate instead of simultaneous utilization. Furthermore, they could also qualitatively match the predicted rates of metabolite consumption to those measured experimentally. Since then DFBA has been used to predict the accumulation of high-value storage compounds in microalgae under varying light and nutrient conditions , determine optimal aerobic and anaerobic culture times and thus scale-up a batch culture of ethanol production in S. cerevisiae by 5-folds , and optimize glucose and ethanol production in E. coli by calculating the optimal flux profile for reactions controlled by genes under genetic perturbation .
Thus, DFBA as an extension of classical FBA can indeed be used to analyze the dynamic reprogramming of a network , especially in response to external perturbations. However, it still hinges on the inherent assumption that the time constants related to intracellular dynamics are much smaller than those describing changes in external concentrations. This is not always true for biological systems as they exhibit control on various levels and thus a direct kinetic description of metabolism that incorporates regulatory mechanisms would likely lead to a higher biological fidelity. For instance, analysis of a hybrid kinetic-FBA model of S. cerevisiae demonstrated that the inclusion of a relatively small number of enzyme kinetic expressions substantially improves the predictive accuracy of FBA, especially if these are used to describe initial reactions in the metabolism of exogenous substrates and reactions at crucial metabolic branch points .
Where the reaction flux v is a function of metabolite concentrations X, kinetic parameters k, and enzyme concentrations E, and X0 represents the initial metabolite concentrations. Because many of the enzyme kinetic parameters are unknown, approximating kinetic mechanisms is a way to improve the tractability of these models when applied to large networks . These include substitutes such as power-law, lin-log and log-lin kinetics, and modular rate laws (reviews of the various rate laws and their differences can be found in [38, 39]). Due to a lack of experimentally-measured kinetic parameters, many times in vitro kinetic data is used to fit and approximate in vivo enzyme kinetics [40, 41]. The thermodynamic feasibility of kinetic models can be enforced using the generalized Wegsheider condition: BT log Keq = 0, where B is the right null space of the stoichiometric matrix S [42, 43]. This enforces thermodynamic feasibility by constraining the values of Keq for every reaction such that a reaction flux can be non-zero only if the corresponding change in the reaction’s Gibbs free energy is negative.
To counter the paucity of data, a top-down modeling approach is used, wherein model parameters are adjusted iteratively using an optimization routine until the model predicted flux distribution matches the one observed. This method uses Monte Carlo modeling to reduce parameter uncertainty by replacing the use of a single set of parameters by multiple parameter sets obtained through random sampling [44, 45, 46]. The feasibility of these methods is improved by defining the parameter space on the basis of known in vivo information before sampling commences. Model validation in this approach involves comparing against data from a different physiological state, usually nutrient stress or a mutant strain . There are three primary frameworks in existence for Monte Carlo modeling in genome-scale reconstructions – ORACLE [48, 49, 50], Jacobian-based structural kinetic models (SKMs)  and Ensemble Modeling [52, 53, 54, 55]. The ensemble modeling approach was recently used to construct a genome-scale kinetic model of E. coli called k-ecoli457, consisting of 457 reactions, 337 metabolites, and 295 substrate-level regulatory interactions . Model parameterization was done via a genetic algorithm where all available fluxomic data was simultaneously imposed. The k-ecoli457 model  was able to capture a wide array of perturbations, with a Pearson correlation coefficient of 0.84 between the experimental data and predicted product yields for 320 engineered strains spanning 24 product metabolites. More recently, the K-FIT decomposition based parameterization approach was introduced  which offers orders of magnitude improvement in parameterization times allowing for detailed a posteriori local sensitivity analyses. Despite their obvious virtues, constructing detailed kinetic models remains challenging. For instance, the latest E. coli constraint-based model contains 2719 reactions involving 1192 metabolites and spans 1515 genes (accounting for ~ 34% of the genome) .
The overarching goal is to be able to capture the hierarchical organization seen in biological systems, where the overall phenotype is a function of the inherent cooperativity between layers such as the transcriptome, metabolome, and proteome. Whole-cell computational models are a step in that direction and are able to predict a wide range of cellular behavior by incorporating the function of each gene, gene product, and metabolite [58, 59]. Cellular functions are split into independent modules describing processes such as DNA replication, segregation, and repair, RNA transcription, protein folding, ribosome assembly, and biochemical conversions modelled via metabolic networks. These modules are then integrated and the overall model performance validated against known properties such as organism doubling time, cellular chemical composition, and gene expression. Thus, whole cell models herald a new age of biological discovery driven by in silico modeling, but the data- and computationally-intensive inputs are a major deterrent to their construction and accessibility [60, 61].
Metabolic control analysis
Reaction properties such as rate and efficiency can be tweaked by altering metabolite concentrations and the catalytic and binding properties of enzymes. Enzymes enable high metabolic flux rates by acting as primary ‘levers’ governing a cell’s metabolism constrained by the thermodynamics and stoichiometry of the system. Drug design was one of the first paradigms under which modification of metabolism was attempted: by identifying and therapeutically targeting ‘essential metabolic pathways’ in a pathogen. This rational drug design was based on Hans Krebs’ proposition that once the exact description of a metabolic pathway has been given, the ‘pacemaker enzyme’ (i.e., the rate-limiting step) can be identified and this controls flux through the entire pathway.
However, the flux through most biological pathways has been seen to be controlled by more than one enzyme. Glycolysis, which is responsible for breaking down glucose to release energy for cellular processes and provide precursors for downstream metabolic pathways, has been found to be controlled by three enzymes – hexokinase (HK), phosphofructokinase (PFK), and pyruvate kinase (PK) . These enzymes are the slowest in the pathway (Vmax values lower than the rest by at least an order of magnitude) and catalyze reactions that operate far away from equilibrium (ratio of mass-action ratio to equilibrium constant ~ 10− 3 to 10− 4). Hence, it becomes necessary to quantitatively determine the degree of control that a given enzyme exerts on the flux through the entire pathway and/or the final rate of product formation so as to be able to rationally reengineer metabolism.
If a small change in Ek causes a significant variation in flux vj, then the enzyme exerts high flux control. Similarly, if a minor change in vj is observed when Ek is varied a lot, then the enzyme does not exert significant flux control.
MCA has indeed been used to analyze control in a metabolic pathway and rationally identify potential enzymatic intervention targets, such as in the parasite Entamoeba histolytica which causes amebiasis in humans. E. histolytica does not have a functional mitochondrion, TCA cycle enzymes, or the oxidative phosphorylation pathway. Thus, glycolysis is the only route of generating ATP, the primary energy ‘currency’ of the cell, and thus inhibiting this pathway will potentially cripple this parasite. Saavedra et al.  constructed a kinetic model of this organism’s glycolytic pathway in order to determine its distribution of control. They found that the majority of control is distributed between HK and phosphoglycerate mutase (PGAM). The amoeba model was also used for evaluating the effect of inhibiting individual enzymes and its effect on flux through the entire pathway. Model predictions indicate that in order to reduce glycolytic flux and ATP levels by 50%, HK and PGAM should be inhibited by 24 and 55%, respectively, or both enzymes by 18%. In contrast, if other enzymes such as phosphofructokinase and pyruvate phosphate dikinase are targeted, their activities have to be inhibited by over 70% to achieve similar glycolytic flux reductions. These predictions were later experimentally verified  using reconstitution experiments which expressed glycolytic enzymes in vitro at near physiological conditions of pH, temperature, and enzymatic activity. PGAM was found to be the main flux controlling enzyme for lower glycolysis while upper glycolysis was controlled by HK, and to a much smaller extent, PFK, and ALD.
However, this homogeneity is hard to realize as a change, say, in the concentration of enzyme Ek will also change components such as the concentration of free substrate which will, in turn, affect the reaction rate vj. Only for systems where enzymes are substrate saturated, i.e., where the concentrations of metabolites exceed those of the enzymes by two or three orders of magnitude does the above serve as a fair approximation (such as 70% of the enzymes measured in E. coli, yeast, and a mammalian cell line ). This is exemplified in the case of Rubisco, the most abundant enzyme in the world, as efforts to engineer it by modifying its catalytic efficiency or by overexpressing it, have been largely unsuccessful. Its concentration was measured to be 3–4 mM in the stroma of spinach chloroplasts , whereas the concentration of its substrate ribulose 1,5 bisphosphate is often lower than 0.5 mM. As expected, the kinetic rate law of the corresponding carbon-fixing reaction was found to be non-homogeneous w.r.t. Rubisco .
Thermodynamic analysis to predict pathway bottlenecks
MCA uses a given steady state to describe how changes in enzyme abundance (or activity) affect the pathway flux. This requires the careful calculation of enzyme kinetic properties which are experimentally laborious to measure and also differ between organisms and isozymes, which further confounds the ability to compare across pathways and/or multiple organisms. Using thermodynamics and stoichiometry to calculate a single metric called Max-min Driving Force (MDF)  alleviates these problems. MDF identifies reactions within a pathway whose rates are constrained by a low thermodynamic driving force and thus would need the catalyzing enzymes to be present in higher concentrations and/or be engineered to achieve higher turnovers. Furthermore, MDF can also predict if a pathway is likely to support high flux under cellular conditions, thus allowing one to select among alternative pathways as intervention targets [69, 70].
Minimal protein utilization drives cellular metabolism
MDF exploits the fact that the thermodynamic driving force behind a reaction dictates its rate, where higher forces correspond to high forward and low backward fluxes. This translates into efficient enzyme usage by decreasing the amount of enzyme needed per unit of metabolic flux. However, computing enzyme demand from metabolic fluxes is not trivial as enzymes tend to not function at maximal capacity. This is mainly due to metabolites causing incomplete substrate saturation and acting as allosteric regulators (affecting enzyme turnover by binding to sites other than the active site). This becomes a cyclic inference problem as steady-state metabolite levels depend on enzyme profiles. Thus, in order to arrive at a single solution, one can look for the enzyme profile with the least cost which is needed to realize a certain flux distribution. This is well justified in biological systems, where metabolic enzymes are a limited resource and thus cells economize by synthesizing the right enzymes in the right amounts, and adapting their levels when conditions change.
This function q(x, v) represents the trade-off between the fluxes which can be realized and the enzyme levels required to sustain that. Noor et al.  used enzyme-cost minimization (ECM) to predict enzyme levels and metabolite concentrations in E. coli using fluxes found by 13-C MFA . They found that the prediction fidelity increases monotonically as more complex cost functions are used. The root-mean square error varied from 1.35 (when enzyme levels are considered to be proportional to reaction fluxes) to 0.42 (when the modular rate laws  are used and the form of ƞkin(c) is determined using the reaction mechanism and order of enzyme-substrate binding). However, the caveat of ECM is the a priori knowledge of reaction fluxes, which is difficult to realize on a genome-scale. Although it is true that metabolic states with a maximal specific rate constitute an elementary flux mode , but their enumeration is computationally intensive . Furthermore, ECM relies on the assumption that the metabolic states of a cell are optimized for enzyme levels, which isn’t always true. Cells often function at sub-optimal levels for robustness or maintaining the metabolic flexibility needed for tidying over future perturbations .
Summary and perspectives
Metabolic engineering has been used for the analysis, design, and optimization of metabolic pathways with significant successes [13, 14, 80, 81, 82]. In this review we discussed metabolic engineering tools (using flux balance analysis) that enable the formulation of a cell’s metabolism as a resource allocation problem driven by biological objectives such as maximizing growth rate or energy production. The construction of genome-scale models of metabolism require, as input, the set of all known metabolic conversions (or reactions) occurring inside the organism, and the thermodynamic favorability of each. Although such constraint-based models of metabolism have found wide use and adaptation, their primary drawback is an inability to capture the dynamic behavior evinced by biological systems. To this end, conventional FBA has been augmented such as by incorporating pseudo-kinetic reaction descriptions for a subset of reaction fluxes (dynamic FBA). Kinetic models take the next step in this direction by modeling reaction fluxes as a function of metabolic concentrations, enzyme kinetic parameters, and enzyme levels themselves. Such models are able to predict the dynamic behavior of metabolic networks but at the expense of intensive data-driven or computationally expensive parameterization. Nevertheless, a kinetic description of reaction mechanisms can be used to identify major flux-controlling steps  and identify pathway bottlenecks (MCA and MDF). Modeling regulations at different levels of metabolism, such as enzymatic or gene expression regulation, draws heavily from control theoretic approaches and can be further extended using classical concepts such as proportional and integral control. This will enable the study of cellular processes such as robust adaptation to environmental perturbations within the well-established fields of control systems for both steady and transient states.
In nature, organisms rarely exist in isolation but interact with others in a variety of biological and ecological niches. Microbial modeling allows us to explore the co-product potential of such communities by modeling the dynamics of inter-species interactions. Microbes can interact with each other and their hosts via processes such as metabolite cross-feeding, which can link disparate pathways from individual species to give rise to novel emergent metabolic functions . By intelligent design of the growth media , standalone growth can be negated and thus co-culture growth and product secretion can be made to be an obligatory outcome of microbial biomass synthesis. The composition of a synthetic consortia can be further tuned by using genome-scale metabolic models to scan potential members and subsequently determine the ability of the culture to synthesize desired compounds.
Thus, a thorough and mechanistic understanding of an organism’s cellular processes would revolutionize our abilities to repair or even guide metabolism. Synthetic biology offers the promise of replacing traditional, high carbon footprint processes fed by unsustainable feedstocks with tunable microbial reactors. Using rational approaches derived from metabolic engineering, designing clean processes that use renewable feedstocks as the raw material can also help provide feasible solutions to current problems of global warming and fossil fuel exhaustion. Indeed, a number of instances where metabolic engineering has helped sustainably improve the economy and efficiency of production processes are already available. Engineered bacteria being used to produce energy from sunlight, water, and organic wastes; synthetic molecules produced by biocatalysts being used as novel drugs and vaccines; and increasing the productivity of existing crop systems by implementing an optimal set of genetic interventions – these are but a few of the possible applications of metabolic engineering [30, 86, 87].
C.D.M. and D.S. generated the ideas for the manuscript, and wrote the manuscript, and edited it. Both authors reviewed and approved the final manuscript.
This work is funded by the Center for Bioenergy Innovation (CBI) (DE-AC05-000R22725) and NSF (DE-AC05-000R22725). This funding source had no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
- 5.Saha R, Liu D, Hoynes-O’Connor A, Liberton M, Yu J, Bhattacharyya-Pakrasi M, et al. Diurnal Regulation of Cellular Processes in the Cyanobacterium Synechocystis sp. Strain PCC 6803: Insights from Transcriptomic, Fluxomic, and Physiological Analyses. MBio. 2016;7(3):e00464–16. https://doi.org/10.1128/mBio.00464-16
- 9.Fell DA, Small JR. Fat synthesis in adipose tissue. An examination of stoichiometric constraints. Biochem J. 1986.Google Scholar
- 15.Vallino JJ, Stephanopoulos G. Metabolic flux distributions in Corynebacterium glutamicum during growth and lysine overproduction. Biotechnol Bioeng. 2000;41:633–46. https://doi.org/10.1002/(SICI)1097-0290(20000320)67:6<872::AID-BIT21>3.0.CO;2-X
- 19.Varma A, Boesch BW, Palsson BO. Stoichiometric interpretation of Escherichia coli glucose catabolism under various oxygenation rates. Appl Environ Microbiol. 1993.Google Scholar
- 20.Bajpai R. Control of bacterial fermentations. Ann N Y Acad Sci. 1987;506:446–58. https://doi.org/10.1111/j.1749-6632.1987.tb23840.x PubMedCrossRefGoogle Scholar
- 22.Varma A, Palsson BO. Stoichiometric flux balance models quantitatively predict growth and metabolic by-product secretion in wild-type Escherichia coli W3110. Appl Environ Microbiol. 1994.Google Scholar
- 30.Flassig RJ, Fachet M, Höffner K, Barton PI, Sundmacher K. Dynamic flux balance modeling to increase the production of high-value compounds in green microalgae. Biotechnol Biofuels. 2016;9:165. https://doi.org/10.1186/s13068-016-0556-4
- 33.Serrano-Bermúdez L, Barrios AFG, Montoya D. Clostridium butyricum population balance model: predicting dynamic metabolic flux distributions using an objective function related to extracellular glycerol content. PLoS One. 2018;13. https://doi.org/10.1371/journal.pone.0209447 PubMedPubMedCentralCrossRefGoogle Scholar
- 41.Teusink B, Passarge J, Reijenga CA, Esgalhado E, Van Der Weijden CC, Schepper M, et al. Can yeast glycolysis be understood terms of vitro kinetics of the constituent enzymes? Testing biochemistry. Eur J Biochem. 2000;267:5313–29. https://doi.org/10.1046/j.1432-1327.2000.01527.x PubMedCrossRefGoogle Scholar
- 42.Liebermeister W, Klipp E. Bringing metabolic networks to life: convenience rate law and thermodynamic constraints. Theor Biol Med Model. 2006;3:41. https://doi.org/10.1186/1742-4682-3-41
- 46.Janasch M, Asplund-Samuelsson J, Steuer R, Hudson EP. Kinetic modeling of the Calvin cycle identifies flux control and stable metabolomes in Synechocystis carbon fixation. J Exp Bot. 2019;70(3):973–83. https://doi.org/10.1093/jxb/ery382
- 52.Khodayari A, Maranas CD. A genome-scale Escherichia coli kinetic metabolic model k-ecoli457 satisfying flux data for multiple mutant strains. Nat Commun. 2016.Google Scholar
- 54.Zomorrodi AR, Lafontaine Rivera JG, Liao JC, Maranas CD. Optimization-driven identification of genetic perturbations accelerates the convergence of model parameters in ensemble modeling of metabolic networks. Biotechnol J. 2013;8(9):1090–104. https://doi.org/10.1002/biot.201200270 PubMedCrossRefGoogle Scholar
- 56.Gopalakrishnan S, Dash S, Maranas C. K-FIT: An accelerated kinetic parameterization algorithm using steady-state fluxomic data. bioRxiv. 2019. https://doi.org/10.1101/612994
- 61.Goldberg AP, Chew YH, Karr JR. Toward scalable whole-cell modeling of human cells. In: SIGSIM-PADS 2016 - Proceedings of the 2016 Annual ACM Conference on Principles of Advanced Discrete Simulation; 2016. https://doi.org/10.1145/2901378.2901402
- 73.Zheng T, Olson DG, Tian L, Bomble YJ, Himmel ME, Lo J, et al. Cofactor specificity of the Bifunctional alcohol and aldehyde dehydrogenase (AdhE) in wild-type and mutant Clostridium thermocellum and Thermoanaerobacterium saccharolyticum. J Bacteriol. 2015;2610–2619:197. https://doi.org/10.1128/jb.00232-15 PubMedPubMedCentralCrossRefGoogle Scholar
- 75.Haverkorn Van Rijsewijk BRB, Nanchen A, Nallet S, Kleijn RJ, Sauer U. Large-scale 13C-flux analysis reveals distinct transcriptional control of respiratory and fermentative metabolism in Escherichia coli. Mol Syst Biol. 2011. https://doi.org/10.1038/msb.2011.9 PubMedPubMedCentralCrossRefGoogle Scholar
- 82.Suástegui M, Yu Ng C, Chowdhury A, Sun W, Cao M, House E, et al. Multilevel engineering of the upstream module of aromatic amino acid biosynthesis in Saccharomyces cerevisiae for high production of polymer and drug precursors. Metab Eng. 2017;42:134–44. https://doi.org/10.1016/j.ymben.2017.06.008 PubMedCrossRefGoogle Scholar
- 86.Cases I, De Lorenzo V. Genetically modified organisms for the environment: stories of success and failure and what we have learned from them. In: International Microbiology; 2005.Google Scholar
- 87.Twyman RM, Schillberg S, Fischer R. Transgenic plants in the biopharmaceutical market. Expert Opin Emerg Drugs. 2005.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.