Logo of aemPermissionsJournals.ASM.orgJournalAEM ArticleJournal InfoAuthorsReviewers
Appl Environ Microbiol. 2009 Sep; 75(18): 5831–5839.
Published online 2009 Jul 24. doi:  10.1128/AEM.00270-09
PMCID: PMC2747866

Increased Malonyl Coenzyme A Biosynthesis by Tuning the Escherichia coli Metabolic Network and Its Application to Flavanone Production


Identification of genetic targets able to bring about changes to the metabolite profiles of microorganisms continues to be a challenging task. We have independently developed a cipher of evolutionary design (CiED) to identify genetic perturbations, such as gene deletions and other network modifications, that result in optimal phenotypes for the production of end products, such as recombinant natural products. Coupled to an evolutionary search, our method demonstrates the utility of a purely stoichiometric network to predict improved Escherichia coli genotypes that more effectively channel carbon flux toward malonyl coenzyme A (CoA) and other cofactors in an effort to generate recombinant strains with enhanced flavonoid production capacity. The engineered E. coli strains were constructed first by the targeted deletion of native genes predicted by CiED and then second by incorporating selected overexpressions, including those of genes required for the coexpression of the plant-derived flavanones, acetate assimilation, acetyl-CoA carboxylase, and the biosynthesis of coenzyme A. As a result, the specific flavanone production from our optimally engineered strains was increased by over 660% for naringenin (15 to 100 mg/liter/optical density unit [OD]) and by over 420% for eriodictyol (13 to 55 mg/liter/OD).

Development of efficient recombinant production platforms for natural product biosynthesis is often limited by the availability of precursors and cofactors derived from the host's native metabolism. This limitation is generally addressed through modifications based on ad hoc predictions or random genetic perturbations, often overlooking the myriad of interactions within the global metabolic network. Advances in computational systems biology have yielded more systems-based approaches (8, 17, 37, 38), providing a means to analyze genome-wide reaction networks using limited parameters and assumptions (19, 25, 35). Using constraints developed from the network architecture to define a metabolic flux space and optimizing these by means of a prescribed metabolic objective, such as biomass, known as flux balance analysis (FBA) (39), one can explore unique aspects within the solution space (30). Gene deletion mutants are typically investigated using a quadratic objective, termed minimization of metabolic adjustment (31), although numerous other optimization routines have also been developed (5, 26, 27, 32). All routines face the same challenges in application, particularly those related to computational costs associated with the combinatorial explosion of phenotypic possibilities in large networks. FBA has largely remained at the theoretical level with few examples presenting its successful experimental verification, especially toward heterologous product biosynthesis (1, 2, 10).

Previously, a novel approach termed OptGene, which integrates an evolutionary algorithm within constraint-based modeling, was utilized to identify in silico gene deletion strategies for the microbial fermentation of vanillin, glycerol, and succinate (26). Presented here is an alternative, independently derived model to similarly investigate the impact of multiple gene deletions in microorganisms by coupling an evolutionary search to constraint-based modeling, termed the cipher of evolutionary design (CiED) model. This method differs from OptGene in two significant factors. First, CiED includes a unique mutation function to retain beneficial mutations, thus expediting evolutionary convergence. Second, CiED includes an optimality assessment from an evolutionary search by means of a frequency analysis.

In the past decade flavonoids have emerged as potential candidates for the treatment of various human maladies, and as such their biosynthesis in microorganisms has been extensively studied (7, 11, 16, 23, 24, 40). Formation of flavanones, the common precursors of all flavonoids, occurs through the action of 4-coumaroyl:coenzyme A (CoA) ligase (4CL), flavonoid chalcone synthase (CHS), and flavonoid chalcone isomerase (CHI), requiring 3 mol of malonyl-CoA and 1 mol of each CoA and ATP for every flavanone molecule generated. Thus, by grafting this biosynthetic pathway into the native metabolic network of Escherichia coli, a significant metabolic burden is imposed (33). During exponential growth in glucose-supplemented media, acetyl-CoA is the dominant component of the total CoA pool, with malonyl-CoA appearing only as a minor species (13, 14). Therefore, engineering the E. coli metabolic network to channel more carbon into malonyl-CoA, CoA, and ATP requires identifying and manipulating key reactions and pathways within the global network architecture.

In this study, we experimentally validated genotypes predicted by CiED for the optimized biosynthesis of flavanones in E. coli. Such genotypes were identified by both flavanone production potential and a minimal growth requirement by evolving random populations of in silico strains under an artificial selection process. CiED-predicted genotypes were then constructed and coupled to overexpression of acetyl-CoA carboxylase (ACC), biotin ligase (BPL), and the flavanone biosynthetic pathway as previously performed (22). Finally, since the CiED algorithm predicted carbon flux increases through the CoA biosynthetic pathway in concert with gene deletions, sequential overexpression of native E. coli enzymes responsible for CoA biosynthesis were performed, resulting in an optimal strain with improved flavanone production levels.


Cipher for evolutionary design.

The CiED stoichiometric model was prepared by expanding an existing E. coli model (EciJR904 [28]) to include additional biochemical processes identified since that model was published. In addition, reactions of the heterologous flavanone pathway (4CL, CHS, and CHI), assumed diffusion fluxes for metabolites within this pathway, and exchange fluxes for those metabolites were also introduced. All reactions added are described in Table S01 in the supplemental material. The additional genes were identified by manual comparison of the EciJR904 model to the available online databases (15, 18). Our expanded model (EciZF922) was compiled and solved using CiED for the steady-state FBA solution and, in the case of gene deletions, the minimization of metabolic adjustment solution. The built-in genetic algorithm function within MatLab 7.1 (Mathworks, Inc.) was modified to include custom crossover and the mutation functions described below. Formulated linear programming and quadratic programming problems were solved in CPLEX 9.1 (ILOG, Inc.) via a MEX file (12). All simulations were performed on an HP Compaq dc5000 system running a 2.80-GHz Pentium 4 processor with 512 Mb of RAM.

The schematic of CiED is provided in Fig. Fig.1.1. Each call to CiED begins with a randomly generated initial population represented as strings of integers that correspond to the deleted genes. Flux profiles for each individual are then calculated and used to assess each genotype's fitness. To eliminate the selection of nonviable phenotypes, gene deletions leading to zero cell growth are automatically assigned poor fitness scores (set to zero). Those with elevated growth and production fluxes are assigned good fitness values by using either the production rate alone or biomass-product-coupled yield (BPCY), defined here as the product of the flux through CHI and growth rate (26). The current population (“parents”) is then ranked using this fitness score before the crossover and mutation process forms the next generation (“children”). In CiED, the five best strains, also known as an elite count, are selected to continue directly into the new population unchanged while the rest of the new population consists of a fraction made via crossover from two parental strains or via mutation from one of the parental strains. To determine the fraction of the new population made by crossover (fcross), CiED uses a function (equation 1) based on the current generation (Gcur) as opposed to a fixed value like OptGene.

equation M1

This function is designed to limit the amount of crossover children generated in early stages of the evolution process, while building slowly as the generation limit (Glimit) is approached. The constant τ is the total number of genes in the metabolic network under investigation, with constants k1 and k2 set to 0.3 and 0.4, respectively, to give a mutation fraction of over 0.7 at the start while nearing 0.3 in later evolutionary stages. Once this fraction is determined, parents are selected for crossover using the common roulette wheel method, where the likelihood a parent is selected is proportional to the magnitude of that parent's fitness score. Mutations are conducted on a least-sensitive approach in an effort to retain beneficial mutations while eliminating others. Each multiple gene deletion parent is decomposed into single gene deletion individuals (“subpopulation”), each containing one of the deletions that existed in the parent strain without repeating. The fitness of each individual in the subpopulation is found, and then the mutation with the lowest fitness score is removed from the parent and a new random point mutation is introduced, ensuring the same number of gene deletions in the child strain as in the parent yet retaining advantageous mutations.

FIG. 1.
CiED simulations. Schematic representation of the CiED with the traditional genetic algorithm elements (gray box).

The selection, crossover, and mutation process continues through generations until one of two termination criteria is met for each genetic algorithm run within a single CiED run. The first is when the generation limit for the natural selection process has been reached, with a default limit of 1,000 generations set. The second focuses on the rate of change for the best individual in the population. If the best individual ceases to improve over a specified stall limit (defaults to 50 generations), the population is considered optimal and the genetic algorithm returns the current population. This parameter has been shown to be difficult to assess, since sharp changes in the optimal value have been identified in previous studies (26). To avoid such concerns, CiED executes a series of sequential evolutionary trials using different starting populations to identify individuals with improved genotypes. The results of successive runs are then used to give the frequency of optimality for each individual found. In other words, this is the fraction of times a particular genotype appears as the best individual over a number of different evolutionary trials. All MatLab m-files for CiED are available upon request.

Strains and media.

All strains and plasmids are listed in Table S02 in the supplemental material. E. coli TOP10F′ or JM109 was used for plasmid cloning and propagation, while BL21Star (Invitrogen) was used for recombinant molecule production. The recombinase system plasmids pKD46 and pCP20 or the template plasmid pKD4 were obtained from the E. coli Genetic Resource Center, Yale University, and then isolated and transferred to the JM109 or BL21Star strain for use. Plasmids pCoLADuet-1 and pCDFDuet-1 (Novagen) were used for cloning and subcloning. DNA manipulations were performed according to standard recombinant DNA procedures (29). Restriction enzymes and T4 DNA ligase were purchased from New England Biolabs. Genomic DNA (gDNA) was isolated using a PureLink genomic DNA purification kit (Invitrogen), and colony PCR was performed using HotMaster DNA polymerase (5 Prime), while all other PCRs were performed using Accuzyme DNA polymerase (Bioline). Plasmid DNA preparation was conducted using a Zyppy plasmid miniprep kit, while fragment DNA was isolated by gel extraction using a Zymoclean gel DNA recovery kit (Zymo Research). Luria-Bertani (LB) broth (Sigma) and M9 minimal salts (Difco) were used throughout with the additional supplementations noted. Cultures of recombinant strains were grown in media containing the required antibiotics: ampicillin (70 μg/ml), kanamycin (40 μg/ml), chloramphenicol (20 μg/ml), and/or streptomycin (40 μg/ml), with uniform concentrations used throughout. p-Coumaric acid and caffeic acid were purchased from MP Biomedicals, and the flavanone standards naringenin and eriodictyol were purchased from Indofine. Media glucose levels were measured using a OneTouch Ultra glucose meter (Johnson & Johnson). Detection of acetate in culture medium was performed using an Enzymatische BioAnalytic kit (Roche).

Plasmid and strain constructions.

All PCR primers used in this study are listed in section S03 in the supplemental material. To construct a plasmid allowing for the expression of E. coli pantothenate kinase (PNK), its encoding gene, coaA, was amplified from isolated gDNA with restriction enzyme primers to insert pnk under the control of the T7 promoter in pD-4CL2. Amplified DNA and vector were double digested with EcoRI and SalI restriction enzymes, gel purified, and then ligated with T4 DNA ligase. Clones were screened for positive insertion-forming vector pD-PNK-4CL2 by restriction digest. A similar process, inserting at the same restriction sites, was used to clone both dfp and coaD into pD-4CL2, forming vectors pD-DFP-4CL2 and pD-PAT-4CL2, respectively. Each plasmid was then cotransformed separately with pA-ACC, pC-BPL, and pE-CHI-CHS to generate strains E2NC (coaA), E2ND (dfp), and E2NP (coaD). To incorporate an acetate assimilation pathway, acetyl-CoA synthase (ACS) from E. coli gDNA was amplified by PCR, digested with NdeI and KpnI, gel purified, and ligated with the digested vector, pC-BPL, using T4 DNA ligase. The resulting vector, pC-ACS-BPL, was then isolated and cotransformed with pA-ACC, pD-PNK-4CL2, and pE-CHI-CHS to generate strain E2NAC. All clones were screened by restriction digest mapping.

Deletion of identified gene targets was completed using PCR fragment recombination with λ-Red recombinase as previously described (9) with a few exceptions. Briefly, E. coli harboring pKD46 was grown at 30°C with 50 mM arabinose and then made electrocompetent by washing three times with sterile cold deionized water. Primers with homologous flanking regions were used to PCR amplify the FRT-flanked kanamycin cassette from pKD4, which was then used as the insert for recombination by electroporation using a GenePulser X-Cell electroporator (Bio-Rad), incubated at 37°C for 1 hour, and then plated on kanamycin selection plates. The antibiotic marker was removed after colony purification and screening by transforming positive strains with pCP20 and growing at 30°C. All deletions were verified by both colony PCR and PCR analysis of gDNA. For colony PCR, single colonies were resuspended in 15 μl sterile deionized water and subjected to three rounds of boiling for 1 minute and freezing at −70°C for 5 minutes before use as the DNA template in a standard PCR. Verified mutants were transformed with up to four plasmids harboring the flavanone pathway, overexpression of ACC and BPL, and/or overexpression for PNK and ACS.

Flavanone fermentations.

Strains were maintained by growth in LB broth at 37°C unless otherwise noted. All flavonoid production fermentations of engineered strains were induced in LB for 3 h and then transferred to M9 modified medium (1× M9 salts, 1% glucose, 1 mM MgSO4, 6 μM biotin, 10 nM thiamine, 1 mM isopropyl-β-d-thiogalactopyranoside [IPTG]), normalizing to an optical density (OD) of 0.6. Strains overexpressing CoA biosynthetic genes were also supplemented with 50 μM pantothenic acid, while strains harboring ACS were supplemented with 2 g/liter sodium acetate. For all flavanone-producing strains, the final concentration of the phenyl-propanoic acid substrate was 1.5 mM after two additions. Flavanones were isolated after 38 h of fermentation, at which time E. coli cells were removed from the culture by centrifugation and the medium was then analyzed using high-performance liquid chromatography (HPLC) with 0.1% formic acid in acetonitrile and 0.1% formic acid in water as the mobile solvents, as previously described (22). Batch fermentations were performed in triplicate. Results of each run were grouped and analyzed using a standard Student t test with a P value cutoff of 0.05.

CoA-containing compound quantifications.

The analyses of CoA and its thioesters were done using HPLC with an Agilent 1100 series instrument, running samples through a reverse-phase ZORBAX SB-C18 column (4.6 by 150 mm) maintained at 25°C. Isolation of CoA-containing compounds was performed using a total OD of 30 (the volume [in ml] × the OD at 600 nm) from supplemented M9 minimal medium cultures. Shake flasks with 110 ml of M9 modified medium were grown at 37°C until the OD reached 0.6, at which time a 50-ml sample was taken, pelleted in a precooled 4°C centrifuge, and maintained at 4°C during the extraction process. Cells were then resuspended in 1 ml of 6.0% perchloric acid and neutralized by dripping 300 μl of 3 M potassium carbonate into the cell suspension while vortexing. The suspension was centrifuged for 1 minute at 14,000 rpm on a precooled benchtop microcentrifuge, after which the supernatant was passed through a 0.22-μm syringe filter. From the filtered cell extract, 20-μl samples were injected into the HPLC column, using 0.2 M sodium phosphate (buffer A) and 0.25 M sodium phosphate in 20% acetonitrile (buffer B) as the mobile phases with a flow rate of 1.0 ml per minute. The mobile phase composition profile after 1.5 min of 97% buffer A (3% buffer B) was divided into a linear gradient having end points of 4.5 min (82% buffer A), 6 min (72% buffer A), 9 min (70% buffer A), 15 min (60% buffer A), 15.6 min (58% buffer A), and 21 min (10% buffer A). At 21.6 min buffer A was returned to the original 97% and maintained for 6 min to fully purge the column. Standards were given the same extraction treatment and used to find the following retention times for the CoA thioesters: malonyl-CoA, 8.9 min; free CoA, 10.8 min; acetyl-CoA, 14.6 min.


Discovering improved genotypes.

The artificial selection process carried out with CiED evolves theoretical populations to identify optimal mutant genotypes. The CiED methodology was run and found single, double, triple, and quadruple gene deletion strategies to identify genotypes with increased flavonoid production potential (Table (Table1).1). Figure Figure22 shows the location of the predicted gene deletions in relation to the central metabolic pathways. Deletions are made at the gene level and then mapped to the enzymatic reaction network in an effort to replicate removals in actual network modifications. By increasing the number of mutation children at the start through the crossover function, a greater range of the genetic landscape is searched by CiED. Mutants predicted via production alone were considered viable only if the growth flux was greater than 15% of the theoretical wild type, a limit similar to that used in previous studies of cell viability (1). Using FBA with the stoichiometric model developed here, the theoretical yield of flavanones from glucose is over 1,100 mg per mg of glucose; however, the cell growth rate was zero under this condition. The lack of growth in the optimal theoretical case indicates that flavanone production and cell growth are competing objectives, as with many secondary metabolites and recombinant natural products. Therefore, creating a balance between cell growth and flavanone production is essential, thus implicating the use of BPCY predictions when selecting genotypes to experimentally construct. Additionally, the biomass flux represents the metabolite drain for a host of other biological functions difficult to model by stoichiometry, such as DNA synthesis and repair.

FIG. 2.
Metabolic reaction network. A reduced central carbon network and relevant pathways are shown, including genes identified by CiED for deletion (bold) and those overexpressed (underlined).
Escherichia coli CiED simulationsa

Importantly, since the genotypic landscape that results from making gene deletions to metabolic networks is highly complex and irregular, early terminations of the genetic algorithm may be premature, resulting in only local optimal solutions. Due to the uncertainty with which the global optimum can be found through such stochastic searches, CiED uses an optimality frequency (Table (Table1)1) for predicting deletion targets, providing a measurement of the network's tendency to these solutions. Including this algorithm minimizes the impact of combinatorial explosion that arises when investigating deletion schemes with a multitude of knockouts. As an example, consider just the primary deletion identified, sdhCDAB, encoding succinate dehydrogenase, whose deletion would be expected to create a surplus of acetyl-CoA, which then might be directed into flavonoid biosynthesis. The sdhCDAB deletion was found to improve flavonoid levels with a frequency of 98%, indicating the network's internal preference for this local optimum. To corroborate this, searching the whole primary solution space showed sdhCDAB is in fact the global optimum. The optimality frequency becomes more important with increased complexity in the deletion landscape due to the increase in modifications introduced. Frequencies of optimality for all genotypes identified by CiED are reported in Table S04 in the supplemental material.

Traditionally, metabolic engineering strategies have relied on inspecting metabolic networks or limited kinetic trials of localized pathways to infer the selection of gene deletions and/or needed overexpressions (3, 4, 6, 20). Therefore, a number of the predicted knockout strategies directly correspond to alterations in the flux through the acetyl-CoA node, the immediate precursor of malonyl-CoA in glycolysis. For instance the primary deletion target glyA, a subunit for serine hydroxyl methyltransferase, is expected to increase the availability of glycolysis intermediates pyruvate and 3-phosphogluconate, thereby indirectly affecting the carbon flows toward the acetyl-CoA and CoA pools. On the other hand, a full perspective of these deletions is only found by looking at the total cellular effects of these deletions. Consider the sdhCDAB deletion, which appears to cause a reduction of the citrate cycle activity, thereby reducing the drain of the acetyl-CoA pool. One might expect that any other deletions within the tricarboxylic acid (TCA) cycle would have a similar effect on reducing the acetyl-CoA drain. Further simulations, however, showed no other deletions in the TCA cycle, resulting in high flavonoid production or viable phenotypes (see Table S05 in the supplemental material). A closer look showed that the sdhCDAB deletion plays a role in the yield per unit glucose of ATP, another critical cofactor in the synthesis of flavanones and CoA. Stopping the TCA cycle at succinate dehydrogenase causes only two fewer moles of ATP to be produced through reoxidation of electron carriers, as opposed to the complete cycle, while other TCA cycle deletions have a more significant impact on cellular growth or a minimal impact on production in comparison.

While such deletion mutants intuitively make sense and their global impact can be recognized, other promising deletion targets or optimal combinations are not as apparent. For example, the most efficient of all the primary deletion mutants predicted from a production standpoint corresponds to the removal of an ammonia AMT transporter, a mutant seemingly unrelated to malonyl-CoA or acetyl-CoA levels (data not shown). Unfortunately, such a mutant did not appear viable from a growth perspective and was not investigated. Mapping these mutants by comparing their theoretical production and growth rates gives rise to three classes of genotypes (Fig. (Fig.3).3). First, and relevant to this study, are viable deletions with elevated production levels. The second class is composed of mutants with no or minimal improved production potential, and the third class contains mutants which most interestingly appear to be nonviable but show improved production potential. These classes are easily identifiable when using BPCY as the objective function, as it is a function that includes specific growth rate; however, using production alone restricts, perhaps unrealistically, the optimization to predict low growth rate strains as optimal genotypes. Investigation of high-production, nonviable classes of mutants, such as the ammonia AMT transporter (amtB), could be performed using strong specific inhibitors to limit the activity of such enzymes. In fact, recently the inhibition of fatty acid biosynthesis by cerulenin was shown via fabB and fabF gene products, deletions that fall in the same category as amtB, to result in increased flavanone yields in recombinant E. coli (23). Interestingly, glucose uptake rates were similar for all mutants identified by CiED (Fig. (Fig.3,3, inset).

FIG. 3.
Genotype analysis. The genotype distribution for single gene deletions as calculated using CiED ordered into viable and productive (x), nonviable productive (□), and (non)viable nonproductive (○) is shown. The lower limits of 15% ...

Higher-order deletions identified by CiED highlight the important aspects of searching the whole solution space to find improved genotypes, not just sequentially building off of previously identified genotypes. CiED also highlights the importance of constraint-based modeling in general to recognize optimal combinations of gene deletions, something that is difficult (if not impossible) by intuition or metabolic network inspection. For instance, not all higher-order deletion schemes included the sdhCDAB gltP genotype most frequently found as a double deletion strategy (Table (Table1).1). A glutamate/aspartate DAACS transporter is encoded by gltP. In the optimal higher-order deletion schemes, suboptimal double deletion schemes were combined by CiED to find both the two triple and the one quadruple deletion genotypes with the optimal levels of growth and production via BPCY. In contrast, maximizing for production alone resulted in just combining the two optimal primary deletions of sdhCDAB and glyA, a subunit of serine hydroxymethyltransferase, for the optimal double deletion scheme. Triple and quadruple schemes were similarly found by stacking mechanisms (see Table S04 in the supplemental material for production fitness function results). To validate the metabolic potential of the predicted deletions, a selection of the best gene deletion strategies was made based on CiED predictions.

Experimental validation of deletion targets.

It was expected that, in order to increase flavanone production, gene deletion schemes would have to elevate intracellular malonyl-CoA and CoA levels during aerobic growth. To verify that the CiED predictions did in fact improve malonyl-CoA and CoA levels, intracellular concentrations were measured from both the E. coli mutant and wild-type strains while overexpressing ACC and BPL (Fig. (Fig.4).4). These overexpressions were included because a previous study suggested flavanone production increased as a result of increased malonyl-CoA levels due to ACC and BPL overexpression (22). Here we indeed found a significant change for the intracellular concentration of malonyl-CoA during growth post-IPTG induction when comparing the parent strain BL21Star to the same strain harboring genes for overexpression of ACC and BPL (strain E2M).

FIG. 4.
Intracellular concentrations of CoA thioesters. Shown are concentrations of malonyl-CoA (A), acetyl-CoA (B), and free CoA (C) from the following experimental strains: BL21Star, the commercial strain; E2M; Z24M (ΔsdhA ΔcitE); Z40M (Δ ...

More importantly, a greater increase in intracellular malonyl-CoA levels was seen in the mutant strains Z24 and Z40, with overexpression of ACC and BPL (Z24M and Z40M, respectively) postinduction, where the intracellular malonyl-CoA concentration increased from 0.95 nmol/mg of dry weight of cells (DCW) to nearly 2.6 nmol/mg DCW (Fig. (Fig.4).4). The mutant strains also showed a decrease in acetyl-CoA levels, from 1.25 nmol/mg DCW to around 0.4 nmol/mg DCW in the first 3 hours postinduction. The gene deletion strains represented an over 200% increase in the metabolite pool and also exhibited a twofold-higher level of free CoA than the wild-type strain. The low malonyl-CoA levels found for all strains at 19 h postinduction can be attributed to malonyl-CoA consumption for the generation of fatty acids and lipids, a trend that has been previously reported in E. coli K-12 (34). All levels for the CoA thioesters reported here are within the reported ranges for E. coli K-12 (34), with the slight differences observed attributed to use of the E. coli BL21Star strain. While both mutant strains Z24 and Z40 were deficient in sdhA and citE (citrate lyase), as those mutations were assumed to be responsible for directly affecting levels of acetyl-CoA, the additional deletions in strain Z40 were suspected of channeling added carbon to the acetyl-CoA node by, in the case of adhE (acetaldehyde dehydrogenase), limiting degradation of glycolysis intermediates and in the case of brnQ the return of underutilized amino acids back to the carbon pool. Based on the significant differences in malonyl-CoA levels seen for the three strains overexpressing ACC, one might suggest that a reduced flux through the TCA cycle for Z24 and an additional increase in carbon channeling for strain Z40 are responsible for the increased levels of malonyl-CoA.

To exploit the increased malonyl-CoA levels within the gene deletion strains, we extended the application of these mutant genotypes to the production of flavanones by incorporating the recombinant plant biosynthetic pathway. Table Table22 lists the flavanone production levels with overexpression of ACC, BPL, and the flavanone pathway for all deletion mutants and the wild-type strain. Note that the increased number of genetic modifications resulted in increased flavanone production levels. Of the single gene deletions performed, the sdhA mutant (Z11N), as predicted by CiED, performed the best with an increased production of 102 mg/liter of naringenin. Continued improvement was seen in higher-order deletion strains, with the best producer, the quadruple deletion strain (Z40N), achieving production levels of 215 mg/liter of naringenin and 114 mg/liter of eriodictyol. Perhaps more importantly than total production, the specific flavanone production from these recombinant strains was increased by over 530% for naringenin (15 to 80 mg/liter/OD) and by over 320% for eriodictyol (13 to 42 mg/liter/OD).

Production characteristics for Escherichia coli mutants in batch fermentationsa

For comparison to simulation data, the biomass-flavanone coupled yield (BFCY) (the OD × the mg of naringenin per liter per hour2) are qualitatively similar to the BPCY from CiED results. For strain E2N the initial growth rate of 0.044 OD/h and production rate of 4.44 mg/liter/h give a BFCY of 0.195 OD·mg/liter/h2. Strain Z11N has a BFCY of 0.355 OD·mg/liter/h2 (growth, 0.036 OD/h; production, 9.86 mg/liter/h), and strain Z40N has a BFCY of 1.08 OD·mg/liter/h2 (growth, 0.069 OD/h; production, 15.7 mg/liter/h). The BFCY is fourfold higher in Z40N than in E2N, and while neither directly related nor quantitative, they show a qualitatively similar trend to the BPCY.

Extending simulation results.

Constraint-based modeling methods may also provide insights to pathways in need of carbon flux enhancement in order to fully exploit the metabolic capabilities of the organism. Genetic perturbations, such as gene deletions, do not simply result in mere local flux readjustment, but due to the interconnectivity of larger cellular networks they result in a more global rearrangement of the flux profile and more often than not invoke significant increases in several metabolic pathways. For this reason, we introduce a parameter, the necessity ratio (ηi), to compare a pathway flux (vi) under two different objective functions (Z) given the same reaction network (equation 2):

equation M2

As the measure of a reaction's response to a change in metabolic objective and dependency on that objective, a pathway's response for flavanone production can be determined by computing the necessity ratio for fluxes under flavanone maximization (product) versus the flux under growth rate maximization (biomass). Similarly, direct comparison of the pathway flux profiles between the theoretical wild-type strain and CiED-predicted mutants under cell growth maximization conditions could be used to investigate optimal carbon utilization during flavanone biosynthesis. Using the native in silico network, Table Table33 shows the necessity ratio of four critical reactions leading to the biosynthesis of ATP, acetyl-CoA, and malonyl-CoA. Not surprisingly, ACC had a necessity ratio of 10.5, indicating that a large increase in carbon flow through the pathway may be needed to facilitate high flavonoid production. Additionally, Table Table33 shows that the two mutants, Z24 and Z40, have elevated levels of ACC flux, signifying a need for increased metabolic flux through the pathway in which the reaction is contained. The gene deletion schemes predicted show up to a twofold increase in ACC activity when compared to the wild-type during maximization for cell growth.

Reaction fluxes and necessity ratiosa

Similarly, ATP and CoA biosyntheses also show significantly large necessity ratios under flavanone maximization, indicating these pathways as potential overexpression targets. Previous studies reported E. coli strains overexpressing PNK, when supplemented with pantothenic acid, gain 10-fold in the CoA pool and 5-fold in the acetyl-CoA pool with minimal acetate accumulation (36). Strategies predicted by CiED also indicate increases in CoA biosynthesis are needed to accommodate acetyl-CoA pool growth. Therefore, overexpressing PNK is important for achieving optimal acetyl-CoA levels and in turn high-yield flavonoid biosynthesis. Interestingly, ACS, while having a necessity ratio of zero, has been previously shown to improve flavonoid yields upon overexpression by promoting acetate assimilation (23). The zero necessity ratio found is a result of the interplay between ACS and the alternative acetate synthesis pathway via acetate kinase, a reaction with no necessity value (see Table S07 in the supplemental material). In the CiED mutant strains, increases in ACS flux indicate an enhanced drain of carbon to acetate. However, previous experimental work with acetate assimilation has shown that ACS can act in an anaplerotic-reversible manner under high acetate concentrations to exploit excess acetate by refilling the acetyl-CoA pool (22). This highlights the inability of constraint-based modeling techniques to deal with regulation or in situ conditions that arise during actual fermentations. Thus, while ACS was included here as an overexpression target based on previous results, ACS was in fact a deletion target predicted by CiED.

Experimental validation of overexpression targets.

Validation of the necessity ratio was done by sequentially overexpressing reactions within the CoA biosynthetic pathway to elevate levels of acetyl-CoA. As suggested by the CiED genotype analysis, overexpression of the CoA biosynthetic pathway to increase the carbon flux can lead to further enhancement in flavanone production. Doing so was expected to improve intracellular levels of free CoA, thus augmenting levels of not only acetyl-CoA but also coumaroyl-CoA, the enzymatic product of the ATP-driven 4CL. Figure Figure5A5A shows the production levels of natural flavanones during overexpression of the CoA biosynthetic pathway when combined with the overexpression of ACC and BPL from Photorhabdus luminescens. PNK has been previously reported to be the rate-limiting step in the biosynthetic pathway; therefore, the resulting increase in flavanone production with PNK overexpression included (strain E2NC) was not surprising. Notably, the supplementation of pantothenic acid alone increased naringenin production from 69 mg/liter to 89 mg/liter (strain E2N), yet E2NC had elevated production titers up to 114 mg/liter of naringenin. Similarly, eriodictyol titers increased from 61 mg/liter up to 86 mg/liter in E2NC with medium supplementation. The two other reactions in the CoA biosynthetic pathway, namely, pantetheine-phosphate adenylyltransferase (coaD) and phosphopantothenoylcysteine synthetase (dfp), not surprisingly resulted in a minimal increase or reduced level of flavanone production (strains E2NP and E2ND, respectively). These results indicate that control of CoA biosynthesis lies solely with PNK. Together with the results from the intracellular CoA pools, these results suggest the cofactor pools of CoA esters are undersized for high-level biosynthesis of flavonoids and potentially other natural compounds that require CoA esters as precursor metabolites, such as fatty acids.

FIG. 5.
Overexpression of cofactor and assimilation pathways. (A) Production levels with strains overexpressing CoA biosynthetic genes coaA (E2NC), dfp (E2ND), and coaD (E2NP) and the control (E2N). Production of naringenin (dark gray) or eriodictyol (white) ...

In an effort to expand the flavonoid production potential of E. coli, the mutant strains were combined with PNK overexpression and then further modified by including the overexpression of the native ACS previously shown to improve flavanone production (22). Figure Figure5B5B shows the control strain E2 and selected mutant strains Z24 and Z40 harboring overexpression for ACC and BPL (strains E2N, Z24N, and Z40N) as well as PNK and ACS (strains E2NAC, Z24NAC, and Z40NAC, respectively). With inclusion of ACS and PNK, the mutant strain Z40NAC produced up to 270 mg/liter of the flavanone naringenin and reached just over 150 mg/liter of eriodictyol. These represent increases of almost 400% for naringenin and over 250% for eriodictyol from the original E2N strain. Note that while the overexpressions incorporated play a critical role in flavanone improvement, the CiED-identified deletions have a quantitatively similar role in achieving the high production levels required for industrial application. Importantly, acetate consumption levels were the same in all strains; however, glucose uptake was 5 g/liter lower in the deletion strains over the fermentation period (see the supplemental material). This indicates that the deletion strains reduced the flow of carbon away from pathways needed to maintain high cellular biomass and toward pathways leading to malonyl-CoA. As such, the yields per gram of glucose in the mutant strains were significantly better, though still well below the theoretical optimal, reaching 29.1 mg naringenin per gram of glucose in Z40NAC, while a yield of only 12.1 mg naringenin per gram glucose was seen in the control. Flavanone production levels reported here are the highest to date not requiring the addition of expensive specific enzyme inhibitors, such as cerulenin. Further refinement and expansion of the stoichiometric model, optimization of media and fermentation conditions, and application of a fermentor-based production process could potentially increase flavanone yields.


The results of the CiED suggested that the deletion of citrate cycle genes sdhCDAB and citE, the amino acid transporter brnQ, and the pyruvate consumer adhE, together with the suggested enzyme overexpressions in CoA biosynthesis and acetyl-CoA carboxylase, generate an efficient genotype for the production of flavanones. CiED predictions included altering distant metabolic routes rather than neighboring pathways that directly impact the formation of malonyl-CoA, the major indigenous precursor in E. coli used for flavanone biosynthesis. Batch fermentations confirmed that simulation predictions led to improved phenotypes with the most promising strain, Z40NAC, increasing specific flavanone yields by over 600% when combined with the overexpression targets of acetyl-CoA carboxylase and acetyl-CoA synthase as well as the peripheral metabolic enzyme pantothenate kinase. Importantly, as seen in Table Table2,2, combining the two best single deletions, sdhA and glyA, resulted in lower levels of flavanone biosynthesis and minimal growth in glucose-supplemented medium. This highlights two important points. First, since the genotype was also predicted as one of the best double deletions when using only production as a fitness function, coupling growth to production when scoring the genotype's fitness may be critical to identifying beneficial deletions for pathways tied to growth rate. Second, the mere stacking of predicted primary deletions leads to only limited increases in recombinant production, as global rearrangements in the carbon flows are not considered for the added genetic modifications. Therefore, algorithms such as CiED that look at global changes offer important advantages in designing optimal genotypes.

Continued development of E. coli as a small-molecule production platform requires identification of metabolic network bottlenecks within the native metabolism to reroute carbon flows with inventive applications of established methods before higher-yield biosyntheses can be realized. Prior studies have alluded to the limited ability of network inspection to identify these bottlenecks, even when comparing them to efficient metabolic producers of the target metabolite (10, 21), while computational methods looking at global changes in network topology have a higher potential to identify more influential genetic perturbations. In a broader context, we have shown that in silico evolutionary design of gene deletion mutants can lead to elevated levels of intracellular metabolites and the rewiring of carbon flux from distant metabolic channels to improve recombinant natural product biosynthesis.

Supplementary Material

[Supplemental material]


We acknowledge financial support from the Independent Research and Development Fund awarded to M. A. G. Koffas. Z. L. Fowler acknowledges financial support from a Mark Diamond Research Grant and the help of undergraduates Ellen Cardone, Ryan Tomko, and Kyle McHugh.


Published ahead of print on 24 July 2009.

Supplemental material for this article may be found at http://aem.asm.org/.


1. Alper, H., Y. S. Jin, J. F. Moxley, and G. Stephanopoulos. 2005. Identifying gene targets for the metabolic engineering of lycopene biosynthesis in Escherichia coli. Metab. Eng. 7:155-164. [PubMed]
2. Alper, H., K. Miyaoku, and G. Stephanopoulos. 2005. Construction of lycopene-overproducing E. coli strains by combining systematic and combinatorial gene knockout targets. Nat. Biotechnol. 23:612-616. [PubMed]
3. Atsumi, S., A. F. Cann, M. Connor, C. R. Shen, K. M. Smith, M. P. Brynildsen, K. J. Y. Chou, T. Hanai, and J. C. Liao. 2008. Metabolic engineering of Escherichia coli for 1-butanol production. Metab. Eng. 10:305-311. [PubMed]
4. Barkovich, R., and J. C. Liao. 2001. Metabolic engineering of isoprenoids. Metab. Eng. 3:27-39. [PubMed]
5. Burgard, A. P., P. Pharkya, and C. D. Maranas. 2003. Optknock: a bilevel programming framework for identifying gene knockout strategies for microbial strain optimization. Biotechnol. Bioeng. 84:647-657. [PubMed]
6. Chemler, J. A., and M. A. Koffas. 2008. Metabolic engineering for plant natural product biosynthesis in microbes. Curr. Opin. Biotechnol. 19:597-605. [PubMed]
7. Chemler, J. A., Y. Yan, E. Leonard, and M. A. Koffas. 2007. Combinatorial mutasynthesis of flavonoid analogues from acrylic acids in microorganisms. Org. Lett. 9:1855-1858. [PubMed]
8. Covert, M. W., C. H. Schilling, I. Famili, J. S. Edwards, I. I. Goryanin, E. Selkov, and B. O. Palsson. 2001. Metabolic modeling of microbial strains in silico. Trends Biochem. Sci. 26:179-186. [PubMed]
9. Datsenko, K. A., and B. L. Wanner. 2000. One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. USA 97:6640-6645. [PMC free article] [PubMed]
10. Fong, S. S., A. P. Burgard, C. D. Herring, E. M. Knight, F. R. Blattner, C. D. Maranas, and B. O. Palsson. 2005. In silico design and adaptive evolution of Escherichia coli for production of lactic acid. Biotechnol. Bioeng. 91:643-648. [PubMed]
11. Fowler, Z. L., and M. Koffas. 2009. Biosynthesis and biotechnological production of flavanones: current state and perspectives. Appl. Microbiol. Biotechnol. 83:799-808. [PubMed]
12. Giorgetti, N. 2005. CPLEXMEX, MatLab MEX interface for the CPLEX Callable Library, Boston, MA. http://www.dii.unisi.it/∼hybrid/tools.html.
13. Jackowski, S., and C. O. Rock. 1986. Consequences of reduced intracellular coenzyme A content in Escherichia coli. J. Bacteriol. 166:866-871. [PMC free article] [PubMed]
14. Jackowski, S., and C. O. Rock. 1984. Metabolism of 4′-phosphopantetheine in Escherichia coli. J. Bacteriol. 158:115-120. [PMC free article] [PubMed]
15. Kanehisa, M. 1997. A database for post-genome analysis. Trends Genet. 13:375-376. [PubMed]
16. Kaneko, M., E. I. Hwang, Y. Ohnishi, and S. Horinouchi. 2003. Heterologous production of flavanones in Escherichia coli: potential for combinatorial biosynthesis of flavonoids in bacteria. J. Ind. Microbiol. Biotechnol. 30:456-461. [PubMed]
17. Kauffman, K. J., P. Prakash, and J. S. Edwards. 2003. Advances in flux balance analysis. Curr. Opin. Biotechnol. 14:491-496. [PubMed]
18. Keseler, I. M., J. Collado-Vides, S. Gama-Castro, J. Ingraham, S. Paley, I. T. Paulsen, M. Peralta-Gil, and P. D. Karp. 2005. EcoCyc: a comprehensive database resource for Escherichia coli. Nucleic Acids Res. 33:D334-D337. [PMC free article] [PubMed]
19. Kim, T. Y., S. B. Sohn, H. U. Kim, and S. Y. Lee. 2008. Strategies for systems-level metabolic engineering. Biotechnol. J. 3:612-623. [PubMed]
20. Koffas, M., C. Roberge, K. Lee, and G. Stephanopoulos. 1999. Metabolic engineering. Annu. Rev. Biomed. Eng. 1:535-557. [PubMed]
21. Lee, S. J., D. Y. Lee, T. Y. Kim, B. H. Kim, J. Lee, and S. Y. Lee. 2005. Metabolic engineering of Escherichia coli for enhanced production of succinic acid, based on genome comparison and in silico gene knockout simulation. Appl. Environ. Microbiol. 71:7880-7887. [PMC free article] [PubMed]
22. Leonard, E., K. H. Lim, P. N. Saw, and M. A. Koffas. 2007. Engineering central metabolic pathways for high-level flavonoid production in Escherichia coli. Appl. Environ. Microbiol. 73:3877-3886. [PMC free article] [PubMed]
23. Leonard, E., Y. Yan, Z. L. Fowler, Z. Li, C. G. Lim, K. H. Lim, and M. A. Koffas. 2008. Strain improvement of recombinant Escherichia coli for efficient production of plant flavonoids. Mol. Pharmacol. 5:257-265. [PubMed]
24. Miyahisa, I., N. Funa, Y. Ohnishi, S. Martens, T. Moriguchi, and S. Horinouchi. 2006. Combinatorial biosynthesis of flavones and flavonols in Escherichia coli. Appl. Microbiol. Biotechnol. 71:53-58. [PubMed]
25. Park, J. H., S. Y. Lee, T. Y. Kim, and H. U. Kim. 2008. Application of systems biology for bioprocess development. Trends Biotechnol. 26:404-412. [PubMed]
26. Patil, K. R., I. Rocha, J. Forster, and J. Nielsen. 2005. Evolutionary programming as a platform for in silico metabolic engineering. BMC Bioinformatics 6:308. [PMC free article] [PubMed]
27. Pharkya, P., A. P. Burgard, and C. D. Maranas. 2004. OptStrain: a computational framework for redesign of microbial production systems. Genome Res. 14:2367-2376. [PMC free article] [PubMed]
28. Reed, J. L., T. D. Vo, C. H. Schilling, and B. O. Palsson. 2003. An expanded genome-scale model of Escherichia coli K-12 (iJR904 GSM/GPR). Genome Biol. 4:R54. [PMC free article] [PubMed]
29. Sambrook, J., E. F. Fritsch, and T. Maniatis. 1989. Molecular cloning: a laboratory manual, 2nd ed. Cold Spring Harbor Laboratory Press, New York, NY.
30. Schuetz, R., L. Kuepfer, and U. Sauer. 2007. Systematic evaluation of objective functions for predicting intracellular fluxes in Escherichia coli. Mol. Syst. Biol. 3:119. [PMC free article] [PubMed]
31. Segre, D., D. Vitkup, and G. M. Church. 2002. Analysis of optimality in natural and perturbed metabolic networks. Proc. Natl. Acad. Sci. USA 99:15112-15117. [PMC free article] [PubMed]
32. Shlomi, T., O. Berkman, and E. Ruppin. 2005. Regulatory on/off minimization of metabolic flux changes after genetic perturbations. Proc. Natl. Acad. Sci. USA 102:7695-7700. [PMC free article] [PubMed]
33. Stephanopoulos, G., A. A. Aristidou, and J. Nielsen. 1998. Metabolic engineering: principles and methodologies, 1st ed. Academic Press, San Diego, CA.
34. Takamura, Y., and G. Nomura. 1988. Changes in the intracellular concentration of acetyl-CoA and malonyl-CoA in relation to the carbon and energy metabolism of Escherichia coli K12. J. Gen. Microbiol. 134:2249-2253. [PubMed]
35. Tyo, K. E., H. S. Alper, and G. N. Stephanopoulos. 2007. Expanding the metabolic engineering toolbox: more options to engineer cells. Trends Biotechnol. 25:132-137. [PubMed]
36. Vadali, R. V., G. N. Bennett, and K. Y. San. 2004. Cofactor engineering of intracellular CoA/acetyl-CoA and its effect on metabolic flux redistribution in Escherichia coli. Metab. Eng. 6:133-139. [PubMed]
37. Varma, A., and B. O. Palsson. 1993. Metabolic capabilities of Escherichia coli. 1. Synthesis of biosynthetic precursors and cofactors. J. Theor. Biol. 165:477-502. [PubMed]
38. Varma, A., and B. O. Palsson. 1993. Metabolic capabilities of Escherichia coli. 2. Optimal growth patterns. J. Theor. Biol. 165:503-522.
39. Varma, A., and B. O. Palsson. 1994. Stoichiometric flux balance models quantitatively predict growth and metabolic by-product secretion in wild-type Escherichia coli W3110. Appl. Environ. Microbiol. 60:3724-3731. [PMC free article] [PubMed]
40. Watts, K. T., P. C. Lee, and C. Schmidt-Dannert. 2004. Exploring recombinant flavonoid biosynthesis in metabolically engineered Escherichia coli. Chembiochem 5:500-507. [PubMed]

Articles from Applied and Environmental Microbiology are provided here courtesy of American Society for Microbiology (ASM)
PubReader format: click here to try


Save items

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


  • Compound
    PubChem Compound links
  • MedGen
    Related information in MedGen
  • PubMed
    PubMed citations for these articles
  • Substance
    PubChem Substance links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...