Identifying gene targets for the metabolic engineering of lycopene biosynthesis in Escherichia coli
Introduction
The central goal of metabolic engineering is the improvement of cellular phenotype, such as metabolite overproduction, by the introduction of genetic controls. To this end, metabolic engineering efforts have considered the properties of the overall metabolic network, in sharp contrast to the single-gene focus that characterizes typical applications of genetic engineering. This is a formidable task considering the fact that molecular and genetic interactions are complex, non-linear and rather poorly understood. Owing to the lack of extensive knowledge about molecular interactions and their kinetics, the dissection and optimization of metabolic pathways is an outstanding issue of central importance to metabolic engineering (Stephanopoulos et al., 2004). A notable exception to the dearth of kinetic information is the constraints imposed upon metabolic function by the stoichiometry of the reaction network.
During the past decade, metabolic engineering has produced an impressive portfolio of results that were guided by rational analysis of well understood systems from a kinetic and regulatory standpoint (Koffas et al., 2003; Padilla et al., 2004; Stafford et al., 2002; Stephanopoulos et al., 1998). Significantly, fewer applications resulted from lesser-understood systems, or from the manipulation of genes not directly connected with the product-synthesizing pathway (Hemmi et al., 1998). These lesser-understood systems have traditionally been handled by sequential approaches whereby a single gene is found (usually from a screen of a combinatorial search) to affect the phenotype of interest, and subsequent searches are conducted in a genetic background defined by the deletion or overexpression of said gene. Despite successes using this approach, there is no evidence that such combinatorial searches are complete or that subsequent sequential modifications would lead to the global phenotype optimum. Furthermore, it is unclear how to systematically identify gene targets when considering the entire bioreaction network.
We address these issues here computationally and experimentally in the context of lycopene synthesis in Escherichia coli. Our computational search makes use of a stoichiometrically balanced, genome-wide bioreaction network of E. coli metabolism whose fluxes are computed such as to maximize cell growth yield in the framework of flux balance analysis (FBA) (Edwards and Palsson, 2000; Segre et al., 2002). Yields and rates of product synthesis (such as lycopene) can be obtained from the calculated fluxes. Although this model is genome-wide and global for most metabolic reactions, it is important to note that it is a strictly stoichiometric model, totally devoid of any kinetic or regulatory information. Consequently, targets identified by this model improve product synthesis solely on the basis of increased availability of metabolic precursors and cofactor balancing. This beneficial effect may be negatively impacted by non-predictive, adverse kinetic and/or regulatory effects.
We employed this formalism to investigate the effect of gene deletions, the most common means of introducing genetic perturbations, on lycopene production. As such, we sought to identify genes whose elimination might lead to an increase in product yield. A major limitation of the FBA approach is that fluxes so calculated are actually those that support maximum growth for the particular genotype. As such, these calculated, in silico, fluxes are not necessarily the same as the actual, in vivo, fluxes in the organism and this is especially so for genetically perturbed systems, (gene knockouts or over-expressions), where the resulting phenotype is often suboptimal in growth and metabolite levels. To correct for this shortcoming, flux profiles were computed for recombinant suboptimal systems by optimizing a different objective function, the minimization of metabolic adjustment (MOMA) requirement between the wild type and a potential single gene knockout mutant (Segre et al., 2002). This calculation yielded flux profiles that are intermediate between the wild-type optimal and gene knockout mutant optimal. While single gene knockout phenotypes have been studied using this method (Edwards and Palsson, 2000), a systematic analysis of multiple gene knockout systems has received less attention. Furthermore, often target identification of multiple knockouts is performed through search algorithms (Burgard et al., 2003), rather than a comprehensive search of all possible multi-gene targets. Nevertheless, these predicted gene targets and identification methods have not been verified with experimental results.
We investigated the issues of gene target identification in the context of heterologous lycopene production in E. coli using the non-mevalonate pathway (Adam et al., 2002). Production of secondary metabolites is typically an expensive and complex cellular process; therefore it provides a good platform for testing concepts of pathway optimization (Mijts and Schmidt-Dannert, 2003). Lycopene production in E. coli utilizes glycolytic intermediates to form precursor monomers, which subsequently undergo polymerization to form the 40 carbon biopolymer (Fig. 1). The isoprenoid pathway and downstream reactions to create a diverse library of carotenoids have received significant attention recently (Mathews and Wurtzel, 2000; Misawa and Shimada, 1998; Sandmann, 2002; Smolke et al., 2001; Wang et al., 1999). Initial attempts for improving carotenoid production in E. coli targeted the expression of genes coding for enzymes that catalyze pathways upstream of the enzymes coded by the crtEBI operon (Farmer and Liao, 2000, Farmer and Liao, 2001; Lee and Schmidt-Dannert, 2002). Even with the over-expression of dxs and idi genes (Kim and Keasling, 2001; Kajiwara et al., 1997; Mathews and Wurtzel, 2000), cellular production and accumulation of carotenoids were limited by regulatory networks and precursor supply (Farmer and Liao, 2000, Farmer and Liao, 2001; Jones et al., 2000; Lee and Schmidt-Dannert, 2002; Wang et al., 1999). Therefore, these studies suggest a need to generate an enhanced production phenotype. Cellular optimization and metabolic engineering are necessary to obtain maximal production rates since lycopene is not an endogenous product and the cell lacks necessary control mechanisms and the infrastructure to produce these molecules.
The computational search identified several putative single- and multiple-gene knockouts yielding increased lycopene production. While most gene knockout mutants exhibited experimental behavior in qualitative agreement with the model predictions, a notable exception illustrates the limitations of using a purely stoichiometric model of metabolism. We also compared an iterative, sequential approach, whereby multiple gene targets are determined by following a path of maximal phenotype increase resulting from sequential single gene modulation, with a simultaneous multiple-gene modulation that examines phenotypic response over the entire space of all possible single- and multiple-gene expression modifications. This comparison led to interesting observations regarding a potential mutant optimization strategy. Application of this strategy ultimately yielded an E. coli mutant strain that accumulated 6600 ppm (6.6 mg/g dcw) of lycopene, an increase of approximately 40% relative to the engineered parental strain.
Section snippets
Strains and media
E. coli K12 PT5-dxs, PT5-idi, PT5-ispFD, provided by DuPont, was used as the lycopene expression strain when harboring the pAC-LYC (Cunningham et al., 1994) plasmid containing the crtEBI operon. Over-expressions of dxs, idi, and ispFD were chromosomally incorporated without an antibiotic marker through promoter delivery (Suh et al., 2003) of the PT5 promoter. Gene deletions were conducted using PCR product recombination (Datsenko and Wanner, 2000) using the pKD46 plasmid expressing the lambda
Gene knockout simulations
The E. coli iJE660a GSM model (Reed et al., 2003) served as the basis for this stoichiometric network. Furthermore, the crtEBI operon was added to the model along with updated isoprenoid synthesis reaction details discovered after the formulation of this model (Adam et al., 2002; Hecht et al., 2001) (see supplemental information for details). Using this updated model, a total of 965 metabolic fluxes (included exchange fluxes) were calculated such as to: (a) balance the rates of synthesis and
Discussion
A systematic, exhaustive computational or experimental search of all feasible gene knockouts in E. coli to determine the genotype yielding the optimal lycopene production phenotype is a tedious and often difficult process. The effects of individual gene knockouts are not necessarily additive in determining the effect(s) of combinations of knockouts, which successfully enhance a phenotype. As the search space increases to include all possible double or higher (triple) knockouts, systematic and
Acknowledgments
We acknowledge financial support of this work by the DuPont-MIT Alliance. In particular, we would like to thank Dr. Wonchul Suh for providing the parental E. coli strain. Dr. Daniel Segre provided the software used (with some modifications) for flux determination. Materials for constructing E. coli knockout strains were obtained from the E. coli Stock Center.
References (32)
- et al.
The two analogous phosphoglycerate mutases of Escherichia coli
FEBS Lett.
(1999) - et al.
Low-copy plasmids can perform as well as or better than high-copy plasmids for metabolic engineering of bacteria
Metab. Eng.
(2000) - et al.
Engineering metabolism and product formation in Corynebacterium glutamicum by coordinated gene overexpression
Metab. Eng.
(2003) - et al.
Engineering of secondary metabolite pathways
Curr. Opin. Biotechnol.
(2003) - et al.
Metabolic engineering for the production of carotenoids in non-carotenogenic bacteria and yeasts
J. Biotechnol.
(1998) - et al.
Controlling the metabolic flux through the carotenoid pathway using directed mRNA processing and stabilization
Metab. Eng.
(2001) Biosynthesis of terpenes: Studies on 1-hydroxy-2-methyl-2-(E)-butenyl 4-diphosphate reductase
Proc. Natl. Acad. Sci. USA
(2002)- et al.
Isolation and characterization of an Escherichia coli mutant lacking the cytochrome o terminal oxidase
J. Bacteriol.
(1985) Selection analyses of insertional mutants using subgenic-resolution arrays
Nat. Biotechnol.
(2001)- et al.
Optknock: a bilevel programming framework for identifying gene knockout strategies for microbial strain optimization
Biotechnol. Bioeng.
(2003)
Molecular structure and enzymatic function of lycopene cyclase from the cyanobacterium Synechococcus sp strain PCC7942
Plant Cell
One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products
Proc. Natl. Acad. Sci. USA
The Escherichia coli MG1655 in silico metabolic genotype: Its definition, characteristics, and capabilities
Proc. Natl. Acad. Sci. USA
Improving lycopene production in Escherichia coli by engineering metabolic control
Nat. Biotechnol.
Precursor balancing for metabolic engineering of lycopene production in Escherichia coli
Biotechnol. Prog.
Studies on the nonmevalonate pathway to terpenes: the role of the GcpE (IspG) protein
Proc. Natl. Acad. Sci. USA
Cited by (0)
- 1
These authors contributed equally to this work.