Systems Metabolic Engineering for Bioenergy and...

Systems Metabolic Engineering for Bioenergy and Biorefinery Sang Yup Lee

Department of Chemical and Biomolecular Engineering (BK21 program)

KAIST

335 Gwahangno, Yuseong-gu, Daejeon 305-701, Korea

Topics to be covered in the lecture

1. Why bioenergy and biorefinery

2. Metabolic engineering

3. Succinic acid as an example of biorefinery

4. Examples of systems metabolic engineering

5. Conclusions

REPORT

Systems metabolic engineering of Escherichia coli forL-threonine production

Kwang Ho Lee1,2,3,4,5, Jin Hwan Park1,2,3,4, Tae Yong Kim1,2,3,4, Hyun Uk Kim1,2,3,4 and Sang Yup Lee1,2,3,4,6,*

1 Metabolic and Biomolecular Engineering National Research Laboratory, Department of Chemical and Biomolecular Engineering (BK21 program), BioProcessEngineering Research Center, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, 2 Center for Ultramicrochemical Process Systems,Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, 3 Center for Systems and Synthetic Biotechnology, Korea Advanced Instituteof Science and Technology (KAIST), Daejeon, Korea, 4 Institute for the BioCentury, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea,5 R&D Center for Bioproducts, CJ Corp, Seoul, Korea and 6 Department of Bio and Brain Engineering, and Bioinformatics Research Center, KAIST, Daejeon, Korea* Corresponding author. Metabolic and Biomolecular Engineering National Research Laboratory, Department of Chemical and Biomolecular Engineering, BioProcessEngineering Research Center, Korea Advanced Institute of Science and Technology (KAIST), 373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Republic of Korea.Tel.: þ 82 42 869 3930; Fax: þ 82 42 869 8800; E-mail: [email protected]

Received 9.7.07; accepted 18.10.07

Amino-acid producers have traditionally been developed by repeated random mutagenesis owing tothe difficulty in rationally engineering the complex and highly regulated metabolic network. Here,we report the development of the genetically defined L-threonine overproducing Escherichia colistrain by systems metabolic engineering. Feedback inhibitions of aspartokinase I and III (encodedby thrA and lysC, respectively) and transcriptional attenuation regulations (located in thrL) wereremoved. Pathways for Thr degradation were removed by deleting tdh and mutating ilvA. The metAand lysA genes were deleted to make more precursors available for Thr biosynthesis. Further targetgenes to be engineered were identified by transcriptome profiling combined with in silico fluxresponse analysis, and their expression levels were manipulated accordingly. The final engineeredE. coli strain was able to produce Thr with a high yield of 0.393 g per gram of glucose, and 82.4 g/lThr by fed-batch culture. The systems metabolic engineering strategy reported here may be broadlyemployed for developing genetically defined organisms for the efficient production of variousbioproducts.Molecular Systems Biology 4 December 2007; doi:10.1038/msb4100196Subject Categories: metabolic and regulatory networks; synthetic biologyKeywords: amino-acid production; flux response analysis; metabolic engineering; systems biology;transcriptome analysis

This is an open-access article distributed under the terms of the Creative Commons Attribution Licence,which permits distribution and reproduction in any medium, provided the original author and source arecredited. This licence does not permit commercial exploitation or the creation of derivative works withoutspecific permission.

Introduction

Amino acids are important fermentation products thathave been used in food, animal feed, pharmaceutical andcosmetic industries with annual growth rate of 5–7%(Leuchtenberger et al, 2005). All of the traditional amino-acidproducing strains have been constructed by multiple rounds ofrandom mutation and selection. Despite the successfuldevelopment of industrial amino-acid producers by suchapproaches, they have several disadvantages. Further straindevelopment is often required to improve the productivity andyield, and to cope with changing fermentation such as thenecessity for utilizing alternative raw materials. However,undesirable random mutations may cause growth retardationand by-product formation. Furthermore, unidentified muta-

tions make it difficult to understand the production mechan-ism, which hinders subsequent strain improvement. Otherindustrial microorganisms producing various metabolites thathave been developed by random mutagenesis share the sameproblems.

Recent advances in genomics and other omics technologycombined with computational analysis are now opening anew avenue toward the development of genetically definedproduction strains (Lee et al, 2005a; Park et al, 2007). AlthoughThr is one of the three major amino acids produced byfermentation processes (Takors et al, 2007), there has been noreport on the construction of genetically defined Thr-over-producing strain. It was thus our goal to engineer Escherichiacoli to make a Thr overproducer with completely knowngenotype by systems-level metabolic engineering.

& 2007 EMBO and Nature Publishing Group Molecular Systems Biology 2007 1

Molecular Systems Biology 3; Article number 149; doi:10.1038/msb4100196Citation: Molecular Systems Biology 3:149& 2007 EMBO and Nature Publishing Group All rights reserved 1744-4292/07www.molecularsystemsbiology.com

leesycopyright

Results and discussion

Metabolic engineering of E. coli for Thr production

We first constructed a Thr-producing E. coli base strain fromWL3110, a lacI-mutant strain of W3110, using the metabolicand regulatory information available in the literature (Patte,1996). Thr belongs to the aspartate family of amino acids andits biosynthetic pathway consists of five enzymatic steps fromL-aspartate (Figure 1). Aspartokinase isozymes are keyenzymes. Aspartokinase I encoded by thrA is inhibited byThr, and its synthesis is repressed by Thr and L-isoleucine.Aspartokinase II is encoded by metL and its synthesis isrepressed by L-methionine. In E. coli K-12, the amount ofaspartokinase II is much less than that of isozyme I (Patte,1996). Aspartokinase III encoded by lysC is inhibited byL-lysine and also inhibits its synthesis. Based on the previously

published data (Ogawa-Miyata et al, 2001; Lee et al, 2003),feedback inhibitions of aspartokinase I and III were removedby replacing the 1034th base C with T (Ser 345-Phe) in thethrA gene and the 1055th base C with T (Thr 342-Ile) inthe lysC gene. Transcriptional attenuation was removedby replacing the native promoter of the thrABC operonwith the tac promoter, which allows constitutive expressionin lacI-mutant strain. Next, the pathways that compete withThr formation and degrade Thr were removed by deletingthe lysA (encoding diaminopimelate decarboxylase), metA(encoding homoserine succinyltransferase), and tdh (encodingthreonine dehydrogenase) genes. Furthermore, a point muta-tion replacing the 290th C with T (Ser 97-Phe) wasintroduced into the ilvA gene to decrease the activity ofthreonine dehydratase (Lee et al, 2003). This base strain wasnamed TH07 strain (Figure 1).

Figure 1 Overall systems metabolic engineering strategies employed for the development of a genetically defined Thr-overproducing E. coli strain. Central metabolicpathways that lead to biosynthesis of Thr together with the regulatory circuits and competing pathways are shown. The shaded boxes represent the targeted mutationsintroduced into the genome. The gray Xs indicate that the genes are knocked out or the inhibition/repression is removed. Thick red arrows indicate the increased flux oractivity by directly overexpressing the corresponding genes. Dashed line indicates repression regulation. Dotted lines indicate feedback inhibition.

Strain improvement for L-threonine productionKH Lee et al

2 Molecular Systems Biology 2007 & 2007 EMBO and Nature Publishing Group

Plasmid pBRThrABC was constructed to increase the carbonflux from L-aspartate to Thr by the amplification of the mutatedthrAC1034TBC operon. When recombinant TH07 (pBRThrABC)strain was flask cultured in TPM1 medium containing 50 g/lglucose for 48 h, the final Thr concentration of 10.1 g/l (a yieldof 0.202 g Thr per gram of glucose) was obtained. After thisinitial success in metabolic engineering of E. coli for Thrproduction, further stepwise improvements were made basedon combined transcriptome analysis and in silico metabolicsimulation.

Transcriptome analysis of the Thr productionstrain

To identify the target genes to be manipulated, transcriptomeprofiling was carried out during Thr production. Transcrip-tome profiles of TH07 (pBRThrABC) were compared withthose of the WL3110 (pBR322) control strain (Supplementaryinformation; Supplementary Figure 3; Supplementary Tables V,VI and VII). In the TH07 (pBRThrABC) strain, the expression ofthe thrABC genes was significantly upregulated by 43.60(thrA), 39.05 (thrB) and 23.55-fold (thrC), reflecting theirsuccessful overexpression.

Effect of phosphoenolpyruvate carboxylase on Thrproduction

Among the genes that were downregulated during Thrproduction (Supplementary Table VI), the ppc gene (down-regulated to 0.43-fold) encoding the phosphoenolpyruvatecarboxylase (PPC) was selected as a target gene to bemanipulated. The ppc gene was cloned into pACYC177 tomake pACYCppc, which was transformed into the TH07(pBRThrABC) strain. To examine Thr production, batchcultures were carried out in TPM2 medium containing 10 g/lglucose. The specific activity of PPC in the ppc-amplifiedstrain was 9.3-fold higher than that of the control strain(Supplementary Table IV). To our surprise, however, Thrproduction was rather decreased by 17.1% compared to thecontrol strain TH07 (pBRThrABC, pACYC177). When the ppc-deleted strain TH11C (pBRThrABC, pACYC177) was culturedunder the same condition, Thr production significantlydecreased by 87%, which suggested that PPC is necessaryfor Thr production (Supplementary Figure 4).

Thus, it was reasoned that there is an optimal level of PPCfor efficient production of Thr. To find the relationship betweenthe PPC flux and Thr production rate, we performed fluxresponse analysis; the response of Thr production rate to thevarying PPC flux was examined in TH07 (pBRThrABC) strain.As shown in Figure 2A, Thr production rate increased as thePPC flux increased to 12.2 mmol/gDCW/h (gram dry cellweight (gDCW)), but decreased with further increase of thePPC flux. Instead, increased PPC flux was predicted to enhancebiomass formation. The in vivo PPC flux experimentallymeasured for E. coli growing in a glucose minimal medium wasreported to be 4 mmol/gDCW/h (Fischer et al, 2004). Thus,these simulation results suggested that the expression of theppc gene should be increased but not to a too high level.To implement this, we replaced the native promoter of the

ppc gene with the trc promoter in the chromosome to makeTH19C strain, which should provide higher PPC flux than thewild type but lower than the plasmid-based overexpression.The TH19C (pBRThrABC) strain showed a 27.7% increase inThr production compared with the control TH07 (pBRThrABC)strain (Supplementary Figure 4F).

Effect of glyoxylate shunt on Thr production

To find next modification targets, we went through the list ofsignificantly upregulated genes from the transcriptome analy-sis. It was notable that the aceBA genes involved in glyoxylateshunt were upregulated by 2.23- and 1.80-fold, respectively(Supplementary Table V), which is contradictory to ourknowledge that they are not activated in aerobic conditionusing glucose as carbon source (Yamamoto and Ishihama,2003). It was reasoned that this might be due to the increaseddemand for oxaloacetate in Thr-producing strain, which madethe glyoxylate shunt operate under this condition. Thus, tofurther increase the carbon flux through the glyoxylate shunt,the iclR-deleted TH09C strain was constructed; the transcrip-tional regulator IclR negatively regulates the expression of theaceBA genes encoding isocitrate lyase (ICL) and malatesynthase (Yamamoto and Ishihama, 2003). Batch culture ofTH09C (pBRThrABC) resulted in the production of 1.93 g/l Thr(a yield of 0.186 g Thr per gram of glucose), which is 30.4%higher than that obtained with the control TH07C(pBRThrABC) strain (Supplementary Figure 4). This resultindicates that the increased expression of the aceBA genesindeed improves Thr production (see Supplementary informa-tion). To combine the positive effects of the ppc overexpressionand iclR deletion, the iclR gene in the chromosome of theTH19C strain was deleted to make TH20C strain. Batch cultureof TH20C (pBRThrABC) resulted in the production of 2.24 g/lThr (a yield of 0.213 g Thr per gram of glucose), which is 51.4%higher than that obtained with the control TH07 (pBRThrABC)strain (Supplementary Figure 4; Supplementary Table III).

Enhanced production of Thr by engineeringtransport system

Efficient export of amino acid is also important to furtherenhance its production. Transcriptome profiling suggestedseveral transporter targets to be engineered. The expression ofthe tdcC gene encoding a Thr transporter (Sumantran et al,1990) was found to be upregulated by 1.7-fold (SupplementaryTable V). Since this transporter uptakes extracellular Thr intothe cell, we deleted the tdcC gene to prevent it. When theresulting TH27C (pBRThrABC) strain was batch cultured inTPM2 medium containing 30 g/l glucose, 7.4 g/l of Thr (a yieldof 0.246 g Thr per gram of glucose) was produced. This is15.6% higher than that obtained with TH20C (pBRThrABC)under the same culture condition. This positive effect was notexpected, as the tdcC gene has been known to be inducedunder anaerobic conditions (Sumantran et al, 1990). Thus, itwould have not been possible to select the tdcC gene fordeletion if transcriptome analysis was not performed.

Next, we searched for the Thr exporters to engineer. Thetranscriptome data showed that the rhtC gene encoding a



threonine exporter (Kruse et al, 2002) was upregulated by2.99-fold (Supplementary Table V). It is likely that the rhtCgene was induced to protect cells from toxic effects ofintracellular Thr accumulation by exporting Thr out of thecell. In order to further increase the Thr export capacity, arecombinant TH27C (pBRThrABCR) strain, which overexpresses

the rhtC gene, was constructed (see Supplementary informa-tion). Batch culture of this strain in TPM2 medium containing30 g/l glucose resulted in the production of 11.1 g/l Thr (a yieldof 0.370 g Thr per gram of glucose), which is 50.2% higherthan that obtained without rhtC amplification (SupplementaryFigure 5). Further amplification of the rhtA and rhtB genes,

Figure 2 Flux response analysis during production of Thr using the in silico genome-scale metabolic model. (A) The response of Thr production rate to varying PPCflux. (B) The response of Thr production rate to varying ICL flux. The predicted flux distributions belonging to the region i, ii, iii, and iv are described with colored arrows;red, blue, and black arrows indicate the fluxes that are predicted to be increased, decreased, and remain unchanged, respectively, in each region. (C) The response ofacetic acid formation rate to the varying flux of the individual central metabolic reaction. Those reactions that upon increasing their fluxes result in decreasing the aceticacid formation rate are shown in blue. In contrast, those reactions that upon increasing their fluxes result in increasing the acetic acid formation rate are shown in red. Inthe two graphs inside, the x-axis denotes the normalized flux of one of the color-coded reactions, whereas the y-axis denotes the normalized acetic acid formation rate.Abbreviations are as follows: G6P, glucose-6-phosphate; RL5P, ribulose-5-phosphate; X5P, xylulose-5-phosphate; R5P, ribose-5-phosphate; E4P, erythrose-4-phosphate; F6P, fructose-6-phosphate; F1,6dP, fructose-1, 6-bisphosphate; G3P, glyceraldehyde-3-phosphate; DHAP, dihydroxyacetonephosphate; PEP,phosphoenolpyruvate; PYR, pyruvate; ICT, isocitrate; a-KG, a-ketoglutarate; SUC, succinate; MAL, malate; OAA, oxaloacetate; THR, L-threonine; ACE, acetate.



which are involved in the export of Thr and homoserine(Zakataeva et al, 1999), resulted in the small increase in Thrproduction. Batch culture of the final engineered strain TH27C(pBRThrABCR3) allowed production of 11.8 g/l Thr with ahigh yield of 0.393 g Thr per gram of glucose (SupplementaryTable III). Thus, systems metabolic engineering of E. coli bytargeted genome engineering, manipulation of target genesbased on transcriptome profiling and in silico simulation, andtransporter engineering successfully yielded a genetically-defined Thr overproducer.

Metabolic engineering based on in silico fluxresponse analysis

Fed-batch fermentation is the cultivation method of choice inindustry for the high level production of amino acids. We thusperformed fed-batch culture of TH27C (pBRThrABCR3) strainwith intermittent glucose feeding (see Materials and methods),which allowed production of 77.1 g/l Thr in 56.1 h, resulting ina volumetric Thr productivity of 1.37 g/l/h (Figure 3A;Supplementary Table III). Acetic acid started to accumulateconsiderably after consuming 180 g/l of glucose, and itretarded cell growth and lowered the rates of glucose uptakeand Thr production in the later stages of fermentation. Tosolve this problem, the flux that can most effectively reduceacetic acid production was examined by in silico flux responseanalysis; the response of the acetic acid production rate tothe varying individual flux of central metabolic pathwaywas evaluated (Figure 2C). Knocking out the genes in theacetate pathway (pta-ack or poxB) was not considered becauseit can retard growth and/or increase pyruvate excretion(Causey et al, 2004). The strategy of amplifying the acs geneencoding acetyl-CoA synthetase was selected to reduce aceticacid production (Supplementary information). The resulting

TH28C strain, which constitutively expresses the acs gene, wastransformed with pBRThrABCR3 and fed-batch fermentationwas performed (Figure 3B; Supplementary Figure 6). Asshown in Figure 3B, acetic acid production decreasedconsiderably to 2.35 g/l. The final Thr concentration reached82.4 g/l in 50 h, with a volumetric productivity of 1.65 g/l/h(Supplementary Table III), which is 20.4% higher thanthat obtained with TH27C (pBRThrABCR3) strain, and iscomparable to a value typically obtained with a randomlymutagenized industrial strain.

In this paper, we reported the strategies for systemsmetabolic engineering of E. coli to develop a geneticallydefined Thr overproducer. Regulatory and metabolic circuitswere rewired to develop an initial Thr producer, which wasfurther engineered based on transcriptome profiling and insilico flux response analysis. Additional metabolic engineeringwas performed to enhance Thr export and to reduce acetic acidformation. The final engineered strain was able to produceThr with high efficiency. As this engineered strain iscompletely defined with respect to its genome traits, it ispossible to further engineer the metabolic and regulatorycircuits according to the changing demands. The maximumachievable theoretical yield (Hong et al, 2004) of L-threoninewas calculated to be 0.81 g Thr per gram of glucose, which canbe achieved if there is no biomass formation. In the case ofbatch culture of TH27C (pBRThrABCR3), the biomass yieldand Thr yield were 0.366 and 0.393 g per gram, respectively.These two yields add to 0.759 g per gram, which is close to themaximum achievable yield of Thr if there was no biomassformation. This analysis also suggests that there is still a roomfor further improvement in Thr production. Taken together, thesystems metabolic engineering strategy reported here can bebroadly employed for developing the genetically definedstrains for the efficient production of other amino acids andvarious bioproducts.

Figure 3 Time profiles of cell growth, Thr production, acetic acid accumulation during the fed-batch culture of (A) TH27C (pBRThrABCR3), and (B) TH28C(pBRThrABCR3). The arrows indicate the sampling points for real time RT–PCR analysis. Symbols are as follows: J, cell growth (OD600); ’, L-threonine (g/l);m, glucose (g/l); &, acetic acid (g/l); }, lactic acid (g/l).



Materials and methods

Strains and plasmids used in this study

The strains and plasmids constructed and used in this study and theprimers used for gene cloning are listed in Supplementary Table I.Detailed procedures for the construction of plasmids are shown inSupplementary information and Supplementary Figure 1. All restric-tion enzymes and pfu DNA polymerase for PCR used in this work werepurchased from New England Biolabs (Ipswich, MA) and Solgent(Daejeon, Korea), respectively.

Genome engineering: promoter replacement andsite-directed mutagenesis

The primers used for chromosomal manipulation are listed inSupplementary Table I. Replacement of native promoter in threonineoperon with the strong tac promoter, removal of feedback inhibitionsof aspartokinase I and III, and reduction of threonine dehydrataseactivity were performed using sacB homologous recombinationsystem (Schweizer, 1992). Substitution of the endogenous promotersof the ppc gene and that of the acs gene by the trc promoterwere performed by PCR-mediated l-Red recombination. Detailedprocedures for genome engineering are provided in Supplementaryinformation and Supplementary Figure 1.

Knocking out chromosomal genes

Deletion of the lysA, metA, tdh, ppc, and iclR genes was performedusing one-step inactivation method (Datsenko and Wanner, 2000). Theschematic procedure of gene inactivation is shown in SupplementaryFigure 1F. The l-Red recombinase expression plasmid pKD46 was usedto disrupt the genes in the chromosome of E. coli W3110 withappropriate antibiotic markers. Detailed procedures for gene inactiva-tion are provided in Supplementary information.

Batch and fed-batch cultivation

Batch and fed-batch cultures were carried out using the TPM2 mediumcontaining the indicated amount of glucose at 311C. The TPM2medium contains the following per liter: yeast extract, 2 g; MgSO4

. 7-H2O, 2 g; KH2PO4, 2 g; (NH4)2SO4, 10 g; L-methionine, 0.2 g; L-lysine,0.2 g; L-isoleucine, 0.05 g; and trace metal solution, 10 ml. Glucose wasadded to give the final concentration of 10 or 30 g/l depending on theexperiments described in the text. Chloramphenicol (35mg/ml),kanamycin (40mg/ml), and ampicillin (50 mg/ml) were added to themedium when necessary. Seed cultures were prepared by transferring500 ml of 10 ml overnight cultures prepared in Luria–Bertani (LB)medium into 250 ml Erlenmeyer flasks containing 50 ml LB mediumand cultured at 311C with 250 r.p.m. Cultured cells were used toinoculate the fermentor containing 2 l of TPM2 medium. Batchfermentations were carried out in a 6.6-l Bioflo 3000 fermentor (NewBrunswick Scientific Co., Edison, NJ). The pH was controlled at 6.5 byautomatic feeding of 25% (v/v) NH4OH. The dissolved oxygenconcentration was maintained above 40% of air saturation bysupplying air at 1 vvm (air volume/working volume/minute) and byautomatically controlling the agitation speed up to 1000 rpm.

Fed-batch cultures were carried out in a 6.6-l Bioflo 3000 fermentorcontaining 2 l of TPM2 medium containing 20 g/l glucose at 311C.Feeding solution was made of 500 g/l glucose, 12.5 g/l KH2PO4, 3 g/lL-methionine, and 4.357 g/l L-lysine. When the glucose concentrationin the culture broth fell to 1 g/l, 80 ml of feeding solution (thus,equivalent to 40 g glucose, 1 g KH2PO4, 0.24 g L-methionine, and 0.35 gL-lysine) was added. The dissolved oxygen concentration wascontrolled at 40% of air saturation by automatically increasing theagitation speed up to 1000 r.p.m. and by changing the percentage ofpure oxygen added. The pH was controlled at 6.5 by automatic feedingof 25% (v/v) NH4OH. The pH-stat fed-batch fermentation was carriedout similarly. When glucose becomes exhausted, the pH increases.When the pH becomes higher than 6.59, 6 ml of feeding solution

composed of 565 g/l glucose, 12.9 g/l KH2PO4, 3.59 g/l L-methionine,and 5.27 g/l L-lysine was automatically added.

Enzyme assays

To determine enzyme activities, exponentially growing cells wereharvested by centrifugation at 5000 g and 41C for 10 min. Aftercentrifugation, the supernatant was decanted and the cell pellet waswashed twice with 50 mM potassium phosphate buffer containing1 mM dithiothreitol (DTT, pH 7). The washed cell pellet wasresuspended in the same buffer, rapidly frozen, and stored at �701Cuntil required. Cell disruption was performed using an ultrasonichomogenizer (VCX-600; Sonics and Materials Inc., Newtown, CT),equipped with a titanium probe 40 T (4-mm diameter and 127-mmlength; Sonics and Materials Inc.) in a conical tube in ice water.Sonication was carried out at 20 kHz for five cycles with 30 s of restingtime between each cycle; each cycle was composed of 10 rounds of 3-ssonication and 5 s rest. The disrupted cells were centrifuged for 10 minat 13 000 g and 41C. The supernatant was immediately used or stored at�701C for future use. Protein concentration in the cell-free extracts wasdetermined by Bradford assay using bovine serum albumin as astandard. Enzyme activities were determined spectrophotometricallyin a temperature-controlled spectrophotometer (SpectraMax M2;Molecular Devices Co., Sunnyvale, CA) in triplicates. Reactions weremonitored by following the NADH at 340 nm. The extinctioncoefficient for NADH of 6.23/mM � cm at 340 nm was used. One unit(U) of enzyme activity was defined as the amount of enzyme necessaryto catalyze the conversion of 1 nmol of substrate per minute intospecific products. The specific activity was defined as units per mg ofprotein. The activity of PPC was measured in a reaction mixturecontaining 66 mM Tris–HCl (pH 9), 5 mM PEP, 10 mM MgCl2, 10 mMNaHCO3, 0.15 mM NADH, 2 U malate dehydrogenase, and the cell-freeextract (Terada et al, 1991). The activity of ICL was measured asdescribed by Van der Werf et al (1997) in a reaction mixture containing0.1 M potassium phosphate (pH 7), 5 mM MgCl2, 1 mM DTT, 0.3 mMNADH, 5 mM isocitrate, 20 U rabbit muscle L-lactate dehydrogenase,and the cell-free extract.

RNA purification and labeling of cDNA

Samples for transcriptome analysis were taken at the exponentialphase (Supplementary Figure 2). Total RNA was purified using theRNeasy Mini kit (Qiagen, Chatsworth, CA) following the manufac-turer’s protocol. Fluorescent labeled cDNAwas prepared by an indirectlabeling method. For this purpose, aminoallyl-modified nucleotides(aa-dUTP; Sigma) were incorporated during the reverse transcriptionof the first strand. The aminoallyl-modified first-strand cDNA wassynthesized from 20 to 50 mg of total RNA by reverse transcriptionusing random hexamer primers, SuperScriptII reverse transcriptase(Invitrogen Corp., Carlsbad, CA) and a dNTP plus aa-dUTP mixture(25 mM dATP, dCTP, dGTP, and 7.5 mM dTTP, aa-dUTP). Afterhydrolysis and cleanup using Microcon YM-30 filters (Millipore,Bedford, MA), the N-hydroxysuccinimidyl ester dyes (Cy3- and Cy5-NHS esters; Amersham Pharmacia, Piscataway, NJ) were coupled tothe aminoallyl-labeled first-strand cDNA. Uncoupled dyes wereremoved using Qiaquick PCR purification columns (Qiagen).

Transcriptome profiling

Transcriptome profiling was performed using the TwinChip E. coli-6Koligo chip (Digital Genomics, Seoul, Korea), following the manufac-turer’s protocol. In this oligo chip, 5978 of 70-mer probes that cover thegenomes of E. coli K12, O157:H7 (EDL933) and O157:H7 (Sakai) arespotted in duplicates. Cy3- and Cy5-labelled cDNA probes wereconcentrated by vacuum drying, dissolved in 20ml of hybridizationsolution (30% formamide, 5� SSC, 0.1% SDS, 0.1 mg/ml salmonsperm DNA), and combined. The microarray was incubated withprehybridization solution at 421C for 1 h, washed in water for 5 min,and finally centrifuged at 650 r.p.m. for 5 min. The combinedhybridization probes were denatured by incubation at 951C for5 min. Hybridization was performed under a coverslip, with a final



volume of 20 ml inside a hybridization chamber (Genomic Tree,Daejeon, Korea). After hybridization at 421C for 16 h, the microarraywas washed for 5 min in 2� SSC and 0.1% SDS at 421C, 10 min in0.1� SSC and 0.1% SDS at room temperature, and four times for 1 minin 0.1� SSC at room temperature. Finally, the microarray was dried bycentrifugation for 5 min at 650 r.p.m. The mean signal intensity valuesof the duplicate spots were averaged and then normalized by the globalnormalization method. The genes showing the P-values lower than0.01 were considered significant.

The gene expression data from this study have been submitted toGene Expression Omnibus database under accession numbersGSE9434.

Real-time RT–PCR analysis

The expression level of acetyl-CoA synthetase was determined by real-time RT–PCR analysis, because enzyme assay of acetyl-CoA synthetasecan be interfered by activities of acetate kinase and phosphotransa-cetylase. Samples were taken during fed-batch fermentations at OD600values of 34.9 and 70.8 for TH27C (pBRThrABCR3), and at OD600values of 32.5 and 71.2 for TH28C (pBRThrABCR3). Total RNA waspurified using the RNeasy Mini kit (Qiagen, Chatsworth, CA) followingthe manufacturer’s protocol. Real-time RT–PCR was carried out withthe iCycler iQ (Bio-Rad, Hercules, CA), using the iQ SYBR GreenSupermix (Bio-Rad), following the manufacturer’s protocol. PCRconditions were as follows: 3 min at 951C, and three-step cycle at951C for 30 s, 551C for 30 s, and 721C for 30 s for a total of 40cycles followed by melting curve analysis. PCR was performedwith 1 ng of cDNA template and 0.1 pM of primers. The rrsB genewas used as a reference gene for normalization. The data wereanalyzed using the 2�DDCt method (Livak and Schmittgen, 2001). Theprimers used were as follows: acsf (50-CATTCCTGCCAACATCGCA-30),acsr (50-TTCGCCCCAGAAGGTATCA-30), rrsBf (50-GTGGCGGACGGGTGAGTAAT-30), and rrsBr (50-CCTCTTTGGTCTTGCGACG-30). The primerpairs were designed to yield PCR products of 100 bp in length.

In silico flux response analysis

The in silico flux response analyses were performed using the genome-scale metabolic model E. coli MBEL979 consisting of 979 metabolicreactions and 814 metabolites (144 extracellular metabolites and 670intermediates), which is a slightly modified network of iJR904reported by Palsson and co-workers (Reed et al, 2003; Lee et al,2005b). The steady-state flux values for in silico mutants werecalculated by linear programming-based optimization of an objectivefunction as follows.

Step 1: To construct the in silico mutant strains, the knockoutof the genes was reflected in the model by setting the correspondingfluxes to zero.

Step 2: In order to examine the effect of PPC or ICL flux on threonineproduction, the PPC or ICL flux was perturbed from the minimumvalue to the maximum value, with maximization of the Thr productionrate as an objective function. To investigate the effect of the centralmetabolic flux on acetic acid formation rate, each flux was perturbedfrom the minimum value to the maximum value, with the maximiza-tion of the acetic acid production rate as an objective function.

Step 3: The flux profile graph is generated by plotting and comparingthe threonine production rate or acetic acid production rate on y-axis atdifferent rates of the corresponding flux perturbed.

In order to investigate the effect of PPC or ICL flux on Thrproduction, constraint-based flux analysis was performed on therationally engineered TH07C (pBRThrABC) strain. The glucose uptakerate was set to be 3.457 mmol/gDCW/h, which is the experimentallymeasured value at the exponential growth phase (7.5B10.75 h). Inorder to investigate the most effective flux change for reducing theacetic acid production rate, constraint-based flux analysis wasperformed on the rationally engineered TH27C (pBRThrABCR3) strain.The glucose uptake rate was set to be 0.9 mmol/gDCW/h, which isthe experimentally measured value at 51.9B56.1 h. During simulation,the uptake rates of L-methionine and L-lysine were set at 5 mmol/gDCW/h, which is large enough to ensure cell growth.

Analytical procedures

Cell growth was monitored by measuring absorbance at 600 nm(OD600) using an Ultrospec3000 spectrophotometer (PharmaciaBiotech, Uppsala, Sweden). Cell concentration defined as gram drycell weight (gDCW) per liter was calculated from the predeterminedstandard curve relating OD600 to dry weight (1 OD600¼0.431 gDCW/l).Glucose concentration was measured using a glucose analyzer(model 2700 STAT; Yellow Springs Instrument, Yellow Springs, OH).For analysis of amino acids, the supernatant obtained by centrifugationand filtration of the sampled culture broth (5 ml) was analyzed atScience Lab Center Co. (Daejeon, Korea). The appropriately dilutedsupernatant (100B500-fold diluted) was injected into the cationseperation column (LCA K06/Na 1.6�150 mm; Sykam GmbH, Eresing,Germany) and analyzed with the Sykam S433 amino-acid analyzer.

Supplementary information

Supplementary information is available at the Molecular SystemsBiology website (www.nature.com/msb).

AcknowledgementsWe thank BL Wanner for providing plasmid pKD46. This work wassupported by the Korean Systems Biology Research Program(M10309020000-03B5002-00000) of the Ministry of Science andTechnology through Korea Science and Engineering Foundation.Further supports by the IBM-SUR program, BK21 program, and LGChem Chair Professorship are appreciated.

The gene expression data from this study have been submitted toGene Expression Omnibus database under accession number GSE9434.

References

Causey TB, Shanmugam KT, Yomano LP, Ingram LO (2004)Engineering Escherichia coli for efficient conversion of glucose topyruvate. Proc Natl Acad Sci USA 101: 2235–2240

Datsenko KA, Wanner BL (2000) One-step inactivation ofchromosomal genes in Escherichia coli K-12 using PCR products.Proc Natl Acad Sci USA 97: 6640–6645

Fischer E, Zamboni N, Sauer U (2004) High-throughput metabolic fluxanalysis based on gas chromatography-mass spectrometry derived13C constraints. Anal Biochem 325: 308–316

Hong SH, Kim JS, Lee SY, In YH, Choi SS, Rih JK, Kim CH, Jeong H, HurCG, Kim JJ (2004) The genome sequence of the capnophilic rumenbacterium Mannheimia succiniciproducens. Nat Biotechnol 22:1275–1281

Kruse D, Kramer R, Eggeling L, Rieping M, Pfefferle W, Tchieu JH,Chung YJ, Jr Saier MH, Burkovski A (2002) Influence of threonineexporters on threonine production in Escherichia coli. ApplMicrobiol Biotechnol 59: 205–210

Lee JH, Lee DE, Lee BU, Kim HS (2003) Global analyses of transcrip-tomes and proteomes of a parent strain and an L-threonine-overproducing mutant strain. J Bacteriol 185: 5442–5451

Lee SY, Lee DY, Kim TY (2005a) Systems biotechnology for strainimprovement. Trends Biotechnol 23: 349–358

Lee SY, Woo HM, Lee DY, Choi HS, Kim TY, Yun H (2005b) Systems-level analysis of genome-scale in silico metabolic models usingMetaFluxNet. Biotechnol Bioproc Eng 10: 425–431

Leuchtenberger W, Huthmacher K, Drauz K (2005) Biotechnologicalproduction of amino acids and derivatives: current status andprospects. Appl Microbiol Biotechnol 69: 1–8

Livak KJ, Schmittgen TD (2001) Analysis of relative gene expressiondata using real-time quantitative PCR and the 2�DDCt method.Methods 25: 402–408

Ogawa-Miyata Y, Kojima H, Sano K (2001) Mutation analysis of thefeedback inhibition site of aspartokinase III of Escherichia coli K-12and its use in L-threonine production. Biosci Biotechnol Biochem 65:1149–1154



Park JH, Lee KH, Kim TY, Lee SY (2007) Metabolic engineering ofEscherichia coli for the production of L-valine based ontranscriptome analysis and in silico gene knockout simulation.Proc Natl Acad Sci USA 104: 7797–7802

Patte JC (1996) Biosynthesis of threonine and lysine. In EscherichiaColi and Salmonella: Cellular and Molecular Biology, Neidhardt FC(eds), 2nd edn, pp 528–541. Washington, DC: ASM Press

Reed JL, Vo TD, Schilling CH, Palsson BO (2003) An expanded genome-scale model of Escherichia coli K-12 (iJR904 GSM/GPR). GenomeBiol 4: R54. 1

Schweizer HP (1992) Allelic exchange in Pseudomonas aeruginosausing novel ColE1-type vectors and a family of cassettes containinga portable oriT and the counter-selectable Bacillus subtilis sacBmarker. Mol Microbiol 6: 1195–1204

Sumantran VN, Schweizer HP, Datta P (1990) A novel membrane-associated threonine permease encoded by the tdcC gene ofEscherichia coli. J Bacteriol 172: 4288–4294

Takors R, Bathe B, Rieping M, Hans S, Kelle R, Huthmacher K (2007)Systems biology for industrial strains and fermentation processes–example: amino acids. J Biotechnol 129: 181–190

Terada K, Murata T, Izui K (1991) Site-directed mutagenesis ofphosphoenolpyruvate carboxylase from E. coli: the role of His579 inthe catalytic and regulatory functions. J Biochem (Tokyo) 109: 49–54

Van der Werf MJ, Guettler MV, Jain MK, Zeikus JG (1997)Environmental and physiological factors affecting the succinateproduct ratio during carbohydrate fermentation by Actinobacillussp. 130Z. Arch Microbiol 167: 332–342

Yamamoto K, Ishihama A (2003) Two different modes of transcriptionrepression of the Escherichia coli acetate operon by IclR. MolMicrobiol 47: 183–194

Zakataeva NP, Aleshin VV, Tokmakova IL, Troshin PV, Livshits VA(1999) The novel transmembrane Escherichia coli proteins involvedin the amino acid efflux. FEBS Lett 452: 228–232

Molecular Systems Biology is an open-access journalpublished by European Molecular Biology Organiza-

tion and Nature Publishing Group.This article is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 Licence.



Metabolic engineering of Escherichia coli for theproduction of L-valine based on transcriptomeanalysis and in silico gene knockout simulationJin Hwan Park*†, Kwang Ho Lee*†‡, Tae Yong Kim*†, and Sang Yup Lee*†§¶

*Metabolic and Biomolecular Engineering National Research Laboratory, Department of Chemical and Biomolecular Engineering (BK21 program),BioProcess Engineering Research Center, †Center for Systems and Synthetic Biotechnology, Institute for the Biocentury, and §Department ofBioSystems and Bioinformatics Research Center, Korea Advanced Institute of Science and Technology (KAIST), Daejeon 305-701, Korea;and ‡R & D Center for Bioproducts, CJ Corporation, Seoul 157-724, Korea

Communicated by Lonnie O. Ingram, University of Florida, Gainesville, FL, March 20, 2007 (received for review January 3, 2007)

The L-valine production strain of Escherichia coli was constructedby rational metabolic engineering and stepwise improvementbased on transcriptome analysis and gene knockout simulation ofthe in silico genome-scale metabolic network. Feedback inhibitionof acetohydroxy acid synthase isoenzyme III by L-valine was re-moved by site-directed mutagenesis, and the native promotercontaining the transcriptional attenuator leader regions of theilvGMEDA and ilvBN operon was replaced with the tac promoter.The ilvA, leuA, and panB genes were deleted to make moreprecursors available for L-valine biosynthesis. This engineered Valstrain harboring a plasmid overexpressing the ilvBN genes pro-duced 1.31 g/liter L-valine. Comparative transcriptome profilingwas performed during batch fermentation of the engineered andcontrol strains. Among the down-regulated genes, the lrp andygaZH genes, which encode a global regulator Lrp and L-valineexporter, respectively, were overexpressed. Amplification of thelrp, ygaZH, and lrp-ygaZH genes led to the enhanced production ofL-valine by 21.6%, 47.1%, and 113%, respectively. Further improve-ment was achieved by using in silico gene knockout simulation,which identified the aceF, mdh, and pfkA genes as knockouttargets. The VAMF strain (Val �aceF �mdh �pfkA) overexpressingthe ilvBN, ilvCED, ygaZH, and lrp genes was able to produce 7.55g/liter L-valine from 20 g/liter glucose in batch culture, resulting ina high yield of 0.378 g of L-valine per gram of glucose. These resultssuggest that an industrially competitive strain can be efficientlydeveloped by metabolic engineering based on combined rationalmodification, transcriptome profiling, and systems-level in silicoanalysis.

systems biology � global regulator � exporter � in silico prediction

Most amino acid-producing bacterial strains have been con-structed by random mutagenesis. A significant disadvantageof this approach is the possibility that the random distribution ofmutations in regions not directly related to amino acid biosynthesiscan cause unwanted changes in physiology and growth retardation.Rational metabolic engineering by specific targeted modificationscan overcome this disadvantage. The recent development of omicstechnology, combined with computational analysis, now provides anew avenue for strain improvement (1–4) by providing new infor-mation extracted from a large number of data, which is termed‘‘systems biotechnology’’ (5).

L-valine, an essential hydrophobic and branched-chain aminoacid, is used as a component of cosmetics and pharmaceuticals aswell as animal feed additives. L-valine has been produced byemploying bacteria belonging to the genera Brevibacterium, Coryne-bacterium, and Serratia, which have been improved by randommutation and selection (6, 7). A recent report describes the pro-duction of L-valine by rationally constructed Corynebacterium glu-tamicum, in which a feedback inhibition-resistant small subunit ofacetohydroxy acid synthase (AHAS; encoded by ilvN) was gener-ated by site-directed mutagenesis; the panB (encoding pantothe-

nate synthase) and ilvA (encoding threonine dehydratase) geneswere deleted; and the ilvBNC operon, encoding acetohydroxy acidsynthase (ilvBN), and acetohydroxy acid isomeroreductase (ilvC)were overexpressed (8).

Escherichia coli has been widely used for amino acid production.For example, randomly mutated and/or metabolically engineeredE. coli strains have been used for the production of L-threonine (2),L-phenylalanine (9), and L-alanine (10). However, there has beenno report on the production of L-valine by E. coli constructed byrational metabolic engineering. This situation may be because E.coli possesses more complex regulatory mechanisms for L-valinebiosynthesis than other microorganisms, which makes it difficult toredirect the flux toward enhanced L-valine production. E. coli hasthree AHAS isoenzymes, key enzymes in the biosynthesis ofbranched-chain amino acids. Three E. coli AHAS isoenzymes differin their biochemical properties and regulatory mechanisms (11).The isoenzymes AHAS I, II, and III are encoded by the ilvBN,ilvGM, and ilvIH genes, respectively. Expression of the ilvGMEDAoperon is controlled by transcriptional attenuation mediated by allthree amino acids, L-leucine, L-valine, and L-isoleucine, whereas theexpression of the ilvBN operon is controlled by attenuation medi-ated only by L-valine and L-leucine. The sites responsible forfeedback inhibition are located in the small subunits of AHAS I andIII encoded by ilvN (12) and ilvH (13, 14), respectively. Theisoenzyme AHAS II encoded by ilvGM is insensitive to L-valine(11). It has been known that in the E. coli K-12 strain, AHAS IIisoenzyme is not expressed because of frameshift mutation inilvG (15).

The global regulator leucine responsive protein (Lrp) is involvedin the regulation of L-valine biosynthesis; it activates the expressionof the ilvIH genes encoding the AHAS III isoenzyme (16, 17) andrepresses the expression of the ilvGMEDA operon encoding theAHAS II isoenzyme (18). Lrp is also known to repress theexpression of the livJ and livK genes, encoding the branched-chainamino acid transporter (19). These regulatory roles of Lrp corre-lated well with the previous results from the global gene expressionprofiling in E. coli K12, which revealed that in an lrp-deleted strain,the expression levels of ilvG and ilvM genes were up-regulated by2.17- and 3.07-fold, respectively, and that of the livJ gene by2.32-fold (20). However, its role in L-valine production in E. coli hasnot been elucidated.

Author contributions: J.H.P. and S.Y.L. designed research; J.H.P., K.H.L., and T.Y.K. per-formed research; K.H.L. and T.Y.K. contributed new reagents/analytic tools; J.H.P., T.Y.K.,and S.Y.L. analyzed data; and J.H.P. and S.Y.L. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

¶To whom correspondence should be addressed. E-mail: [email protected].

This article contains supporting information online at www.pnas.org/cgi/content/full/0702609104/DC1.

© 2007 by The National Academy of Sciences of the USA

www.pnas.org�cgi�doi�10.1073�pnas.0702609104 PNAS � May 8, 2007 � vol. 104 � no. 19 � 7797–7802

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1leesycopyright

In this study, rational metabolic engineering based on knownregulatory and metabolic information and new knowledge gener-ated by transcriptome analysis, combined with in silico simulation,was carried out to develop an E. coli strain capable of overproducingL-valine.

ResultsMetabolic Engineering of E. coli for L-Valine Production. The E. coliW3110 strain was rationally engineered to produce L-valine by thefollowing targeted genetic modifications. The biosynthetic path-ways of branched-chain amino acids in E. coli, the regulationsinvolved, and the strategies for constructing L-valine productionstrain are shown in Fig. 1. First, feedback inhibition and transcrip-tional attenuation were removed by site-directed mutagenesis. Thesequences involved in these regulatory mechanisms were identifiedfrom previously published data or patents (13, 14, 21–23). Feedbackinhibition in the small subunit of AHAS III encoded by the ilvHgene was removed by displacing the 41st base G with A and the 50thbase C with T. Transcriptional attenuation sites were removed bydisplacing the attenuator leader region with the tac promoter byhomologous recombination. The successfully mutated clone wasidentified by sacB-positive selection (24). Although the frameshiftmutation in ilvG causes defects in the expression of ilvG, thetranscriptional attenuator needs to be replaced with the tac pro-moter for constitutive expression of the downstream ilvED genes.The WL3110 strain, a lacI-deleted W3110 strain, was constructedto allow constitutive expression of genes under the tac promoterwithout adding an inducer.

Second, L-valine production was enhanced by increasing thecarbon flux toward L-valine formation. The availability of pyruvateand 2-ketoisovalerate is important because it can influence the fluxtoward L-valine formation. Pyruvate is used as a common substratefor biosynthesis of L-valine, L-leucine, and L-isoleucine. To elimi-nate the formation of 2-ketobutyrate, the intermediate precursor of

L-isoleucine biosynthesis competing with L-valine biosynthesis, theilvA gene (encoding L-threonine dehydratase), was knocked out.The panB (encoding 3-methyl-2-oxobutanoate hydroxymethyl-transferase) and leuA (encoding 2-isopropylmalate synthase) geneswere also knocked out to increase 2-ketoisovalerate availabilitytoward L-valine synthesis. This base strain was named the Val strain,of which genotype was described in supporting information (SI)Table 2.

After removing the feedback inhibition and attenuation followedby chromosomal inactivation of competing pathways, we amplifiedthe AHAS I isoenzymes, which catalyze the first reaction of L-valinebiosynthetic pathway. It is known that the AHAS I, encoded by theilvBN genes, has a much higher affinity for pyruvate than for2-ketobutyrate, compared with the isoenzymes AHAS II andAHAS III (25). Hence, the ilvBN operon was selected for ampli-fication. The operon was cloned into pKK223–3 to make pKKil-vBN. The Val strain was transformed with pKKilvBN and culturedin a bioreactor. The time profiles of cell growth and L-valineproduction during the batch fermentation of Val (pKKilvBN) areshown in SI Fig. 4. The growth rate of the recombinant Val strainwas slightly lower than that of a control strain, WL3110 harboringpKK223–3 (SI Fig. 4). The Val (pKKilvBN) strain successfullyproduced 1.31 g/liter of L-valine. After this initial success inmetabolic engineering of E. coli for L-valine production, furtherstepwise improvements were attempted by using this Val strain.

Transcriptome Analysis of the L-Valine Production Strain. To under-stand the altered global gene expression levels in the recombinantVal strain producing L-valine, transcriptome profiling was per-formed. Transcriptome profiles of Val (pKKilvBN) and WL3110(pKK223–3) were compared by using the samples taken at theOD600 of 6.55 (the former) and 6.77 (the latter). They were both atthe exponential growth phase, and the specific growth rates were0.14 h�1 for the former and 0.19 h�1 for the latter. At the time ofsampling, the Val (pKKilvBN) strain produced 0.5 g/liter of L-valine. Genes that were found to be significantly up- or down-regulated are listed in SI Tables 3–5. The expression ratios of genesinvolved in central metabolic pathways and L-valine biosynthesis areshown in Fig. 2. The expression levels of genes involved in glycolysisand pentose phosphate pathway were not significantly changed,whereas those of most of the genes in the tricarboxylic cycledecreased. The increased expression of AHAS isoenzymes, whichare directing pyruvate toward L-valine biosynthesis, may causedepletion of acetyl-CoA and thereby reduce the expression oftricarboxylic cycle genes.

As expected, the ilvBN genes, which were amplified both chro-mosomally (attenuation removal) and extrachromosomally (plas-mid-born amplification), were significantly up-regulated by 54.46-(ilvB) and 32.50-fold (ilvN), respectively. The ilvGM genes, in whichthe native transcriptional attenuator leader region was replaced bythe tac promoter, were also up-regulated by 7.41- (ilvG) and2.21-fold (ilvM), respectively. Similarly, the ilvCDE genes that areinvolved in L-valine biosynthetic pathway were up-regulated by3.74-, 4.28-, and 1.32-fold, respectively. However, the leuABCDgenes involved in L-leucine biosynthesis were significantly down-regulated to 0.17-, 0.14-, 0.31-, and 0.17-fold, respectively. Theseresults revealed that the targeted genetic modifications were suc-cessfully reflected at the transcript level. Other notable changesfound from transcriptome profiling are described in SI Text.

Enhanced Production of L-Valine by Coamplification of the ilvCEDGenes. Transcriptome profiling revealed that the expression levelsof the ilvCED genes encoding acetohydroxy acid isomeroreductase(ilvC), branched-chain amino acid aminotransferase (ilvE), anddihydroxy acid dehydratase (ilvD) involved in the L-valine biosyn-thetic pathway (Fig. 1) were increased in the recombinant Val strain(Fig. 2). However, the extents of the increase were much less thanthose of the ilvBN genes (Fig. 2). This result is easily understandable

Fig. 1. The biosynthetic pathways of branched-chain amino acids in E. coli,regulations involved, and the strategies for constructing the L-valine-producingbase strain. The shaded boxes represent mutations introduced into the genome,and the gray bars indicate the genes that were knocked out. Thin red arrowsindicate the increased flux by knocking out the genes suggested by in silicosimulation. Thick red arrows indicate the increased flux or activity by directlyoverexpressingthecorrespondinggenes.Thickbluearrows indicaterepressionofgene expression by Lrp. Thin blue arrows indicate the decreased flux by knockingoutthecorrespondinggenes.Dotted lines indicatefeedback inhibition.TheblackX indicates that the inhibition or repression is removed. The plus (�) and minus(�) symbols indicate activation and repression of gene expression, respectively.

7798 � www.pnas.org�cgi�doi�10.1073�pnas.0702609104 Park et al.

http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1

because the ilvBN genes were amplified by plasmid-based overex-pression, whereas the ilvCED genes were not; only the attenuationcontrol was removed for the ilvED genes. Thus, we decided tocoamplify the ilvCED genes to possibly increase L-valine produc-tion. The ilvC and livED genes were cloned sequentially followingthe ilvBN operon in pKKilvBN to make pKKilvBNCED. The Valstrain was transformed with pKKilvBNCED and cultured in NM2medium for 48 h. The final OD600 and L-valine concentration at theend of flask cultivation were 23.8 and 3.43 g/liter, respectively. Thus,the coamplification of the ilvCED genes was found to be beneficialfor increasing L-valine production.

Influence of Lrp on L-Valine Production. According to the transcrip-tome data, the expression of the lrp gene was down-regulated to0.52-fold (SI Table 4). This value was also validated by real-timeRT-PCR analysis, which showed its down-regulation to 0.45-fold. Ithas been known that Lrp activates the expression of the ilvIHoperon, which encodes the AHAS III isoenzyme (16). Thus,down-regulation of lrp expression might have caused insufficientexpression of the ilvIH operon. Indeed, the expression level of theilvH gene decreased to 0.58-fold, whereas that of the ilvI gene stayedat the similar level. Furthermore, it has been known that exogenousL-leucine antagonizes the positive effect of Lrp on the expression ofthe ilvIH operon and certain Lrp-regulated genes (26, 27). BecauseL-leucine is added during the cultivation to complement the aux-otrophy, this also might have happened in our case. In anotherstudy, it has been reported that when the intracellular concentrationof Lrp is sufficiently high, all positively regulated genes become

insensitive to L-leucine (28). Taken together, it was assumed thatthe L-leucine requirement in our system may cause insufficientactivation of ilvIH expression by Lrp. For this reason, the lrp genewas selected for amplification to overcome the negative effect ofL-leucine on ilvIH expression. However, we thought that overex-pression of too many genes at the same time by using a high-copy-number plasmid may be detrimental to cell growth. Thus, the originof replication of pKKilvBNCED was replaced with that of amedium-copy-number plasmid pBR322 to make pKBRilvBNCED(SI Fig. 5A).

To examine the ability to produce L-valine, the Val strainharboring pKBRilvBNCED was cultured in NM2 medium for 48 h.The final OD600 and L-valine concentrations were 25.65 and 3.73g/liter, respectively. These values are slightly better than thoseobtained with the Val strain harboring pKKilvBNCED (see above).For the amplification of the lrp gene, plasmid pTrc184lrp, which iscompatible with pKBRilvBNCED, was constructed (SI Fig. 4B).When the Val strain harboring pKBRilvBNCED and pTrc184lrpwas cultured under the same condition, the final L-valine concen-tration obtained was 4.34 g/liter (Table 1), which is 21.6% higherthan that obtained without lrp amplification. To see whether thisincrease is truly due to the availability of Lrp, the Vlrp strain, whichis the lrp-deleted Val strain, was transformed with pKBRilvBNCEDand cultured. The final L-valine concentration obtained was 2.38g/liter, which is 36.2% lower than the control strain (Table 1). Theseresults suggest that amplification of the lrp gene is beneficial forL-valine production in E. coli.

Enhanced Production of L-Valine by Overexpressing the ygaZH Genes.In C. glutamicum, which is the most widely used bacterium foramino acid production, the brnFE genes encode the exporter ofbranched-chain amino acids (29). The E. coli genes homologous tobrnFE were suggested to be the ygaZH genes by homology search(29). The amino acid sequence of YgaZ showed 27% identity to thatof BrnF (SI Fig. 6). Thus, we examined whether the ygaZH genes,which form an operon, encode an L-valine exporter. The ygaZHgenes were knocked out in W3110 to make the WygaZH strain (SITable 2). Then the minimal inhibitory concentration (MIC) test wasperformed for the W3110 and WygaZH strains by using an L-valineanalogue, DL-norvaline, which is not metabolized by the cell (SITable 6). The ygaZH-deficient WygaZH strain did not grow in thepresence of 9 mM DL-norvaline, whereas the wild-type W3110strain survived in the presence of 15 mM DL-norvaline. Further-more, the W3110 strain harboring pTrc184ygaZH, which overpro-duces YgaZH, showed normal growth even in the presence of 50mM DL-norvaline (SI Table 6). This result is the first experimentalproof that YgaZH functions as an L-valine exporter.

According to the transcriptome data, the expression levels of theygaZ and ygaH genes decreased to 0.75- and 0.61-fold, respectively,in the recombinant Val strain (SI Table 4). We thus reasoned thatthe low-level expression of the ygaZH genes might have beenlimiting the L-valine production. Hence, the ygaZH genes wereoverexpressed by using pTrc184ygaZH plasmid in the Val (pK-

Fig. 2. Results of comparative transcriptome profiling. The numbers are theratios of the expression levels in Val (pKKilvBN) vs. WL3110 (pKK223–3). Theshadedandboxednumbers indicate significantlyup-anddown-regulatedgenes,respectively, in L-valine-producing strain Val (pKKilvBN). GLC, glucose; G6P, glu-cose-6-phosphate; 6PGL, gluconolactone-6-phosphate; 6PGC, 6-phospho glu-conate; RL5P, ribulose-5-phosphate; X5P, xylurose-5-phosphate; R5P, ribose-5-phosphate; S7P, sedoheptulose-7-phosphate; E4P, erythrose-4-phosphate; F6P,fructose-6-phosphate; FBP, fructose-1, 6-bisphosphate; GAP, glyceraldehyde-3-phosphate; DHAP, dihydroxyacetonephosphate; GBP, 1,3-bisphosphoglycerate;PEP, phosphoenolpyruvate; PYR, pyruvate; ACA, acetyl-CoA; ACP, acetyl-phosphate; Ace, acetate; ACL, 2-acetolactate; DHV, 2,3-dihydroxyisovalerate;KIV, 2-ketoisovalerate; Val, L-valine; Leu, L-leucine; Pan, pantothenate; Ala, L-alanine; CIT, citrate; ICIT, isocitrate; AKG, �-ketoglutarate; SUCOA, succinyl-CoA;SUC, succinate; FUM, fumarate; MAL, malate; OAA, oxaloacetate.

Table 1. Effects of overexpressing or deleting the lrp and/orygaZH genes on L-valine production

Strain OD600 L-valine (g/liter)*

Val (pKBRilvBNCED) 25.65 � 0.35 3.73 � 0.16Val (pKBRilvBNCED, pTrc184) 23.97 � 0.15 3.57 � 0.06Val (pKBRilvBNCED, pTrc184lrp) 23.42 � 0.67 4.34 � 0.03Vlrp (pKBRilvBNCED) 22.05 � 0.39 2.38 � 0.03Val (pKBRilvBNCED, pTrc184ygaZH) 24.59 � 1.58 5.25 � 0.14VygaZH (pKBRilvBNCED) 24.15 � 0.62 3.73 � 0.09Val (pKBRilvBNCED, pTrc184ygaZHlrp) 24.85 � 0.07 7.61 � 0.24

*The L-valine concentration was determined at 48 h of flask cultivation in NM2medium containing 50 g/liter glucose.

Park et al. PNAS � May 8, 2007 � vol. 104 � no. 19 � 7799

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1

BRilvBNCED) strain. This recombinant Val strain produced 5.25g/liter L-valine, which is 47.1% higher than that obtained withoutygaZH amplification (Table 1). This result further suggests that theygaZH genes encode an exporter of L-valine in E. coli. We thenexamined L-valine production in the VygaZH strain harboringpKBRilvBNCED to see whether L-valine production is hampered.Surprisingly, L-valine production was not affected in the VygaZHstrain. This result suggests that other L-valine exporters exist and donot (or less efficiently) export DL-norvaline. Taken together, we canconclude that the YgaZH is one of the L-valine exporters in E. coli,which, on amplification, is beneficial for the enhanced productionof L-valine.

Synergistic Effect of Lrp and YgaZH on L-Valine Production. Becausethe overexpression of the lrp and ygaZH genes resulted in theincreased production of L-valine by 21.6% and 47.1%, respec-tively, we tried to overexpress both genes at the same time. Therecombinant Val strain harboring pKBRilvBNCED andpTrc184ygaZHlrp produced 7.61 g/liter L-valine, which is 113%higher than that produced by the control strain. Thus, the effect ofLrp and YgaZH is more than additive, and the positive effect of Lrpseems to be further enhanced when accumulated L-valine is morerapidly exported by the overproduced YgaZH.

To understand the reason for this positive effect of coexpress-ing the lrp and ygaZH genes, the regulation of ygaZH expressionby Lrp was investigated by real-time RT-PCR analysis by usingthe W3110 and Wlrp strains in response to 5 �M L-valine. InW3110, the expression of the lrp, ygaZ, and ygaH genes increasedby 6.96-, 4.0-, and 3.2-fold, respectively, in response to 5 �ML-valine. However, the expression of the ygaZ and ygaH genes inthe Wlrp strain decreased to 0.7- and 0.5-fold, respectively.These results suggest that the induction of the ygaZH expressionby L-valine is likely to be mediated by Lrp, which functions as anactivator of ygaZH expression.

Metabolic Engineering Based on in Silico Gene Knockout Studies. Tofurther improve the Val strain, in silico gene knockout simulationstudies were performed (SI Table 7). Fig. 3 presents the results ofthe in silico gene knockout simulation with respect to L-valineproduction and growth rates. It was straightforward to select thefirst knockout target as the aceF (or aceE or lpdA) gene (encodingpyruvate dehydrogenase); the L-valine production rate was in-creased by 2-fold to 0.377 mmol�gDW�1�h�1 in this pyruvatedehydrogenase knockout mutant strain (Fig. 3A) by increasing thepyruvate pool. Consequently, we selected the aceF gene as the first

knockout candidate because it affects the pyruvate dehydrogenaseactivity most significantly (10). After deleting the aceF gene, thesecond knockout candidate was identified by simulating the addi-tional gene knockout effects on metabolic flux distribution by usingthe Val (�aceF) strain. Knocking out the tpiA gene was predictedto increase the L-valine production rate most significantly, but thegrowth rate was significantly reduced as well (Fig. 3B). Thus, thenext best target, the pfkA/B gene, was selected as the second geneknockout candidate because it allows a considerable increase in theL-valine production rate with less reduction in the growth rate. ThepfkA/B knockout seems to increase the availability of NADPH, animportant cofactor for L-valine production, by pushing more carbonflux toward the pentose phosphate pathway. For the actual inacti-vation of phosphofructokinase, the pfkA gene was deleted becauseit encodes 6-phosphofructokinase-1, which possesses 90% of theactivity (30). Using the Val (�aceF �pfkA) strain, the third geneknockout simulation was performed (Fig. 3C). There were severalgene knockout candidates that were able to increase the L-valineproduction rate further. Among these, knocking out the mdh generesulted in the highest increase of the L-valine production rate.Thus, the mdh gene was chosen as the third knockout target toinactivate malate dehydrogenase (31). By doing so, the pyruvatepool is expected to be indirectly increased by the enhanced flux ofmalate to pyruvate by a malic enzyme. Further knockout simulationwas not attempted because the triple knockout mutant strainalready showed significantly lower growth rate than the wild-typestrain. These three predicted genes were knocked out by chromo-somal gene inactivation by using the � red operon (SI Text).Consequently, the VAMF strain, which is Val (�aceF �mdh�pfkA), was constructed. For the better growth of the VAMF strainlacking in pyruvate dehydrogenase, sodium acetate (3 g/liter) wassupplemented.

To examine the performance of the VAMF strain, batchcultures of the following four recombinant strains were carriedout in NM1 medium containing 20 g/liter glucose: Val (pKBRil-vBNCED), VAMF (pKBRilvBNCED), Val (pKBRilvBNCED,pTrc184ygaZHlrp), and VAMF (pKBRilvBNCED,pTrc184ygaZHlrp). The final L-valine concentrations obtainedwith these four strains were 2.75, 4.00, 3.33, and 7.55 g/liter,respectively. The triple knockout mutation allowed a 45.5%increase in L-valine production over an engineered Val strain inthe presence of plasmid pKBRilvBNCED. Furthermore, theL-valine concentration obtained with the VAMF (pKBRil-vBNCED, pTrc184ygaZHlrp) was 2.27-fold higher than that

Fig. 3. Results of in silico gene knockout simulations by using the genome-scale metabolic model of E. coli MBEL979. The results of single (A), double (B), andtriple (C) gene knockout simulations with respect to L-valine production and growth rates are shown. Only the five best candidates with respect to the L-valineproduction rate are shown for each stage of knockout simulation. Slashes indicate isoenzymes or subunits of the enzyme complex. The L-valine production andgrowth rates of the control Val strain harboring pKKilvBN are also indicated for comparison.


http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1

obtained with the corresponding recombinant Val strain. TheL-valine yield achieved with VAMF (pKBRilvBNCED,pTrc184ygaZHlrp) was as high as 0.378 g of L-valine per gram ofglucose. These results suggest that metabolic engineering basedon the results of an in silico genome-scale prediction of knockouttargets is an efficient approach toward strain improvement.

DiscussionTraditionally, amino acid-producing strains have been developed byrandom mutagenesis and selection. Although various large-scaleanalytical techniques such as transcriptome and proteome analysisare now available, it is difficult to apply these techniques on therandomly mutated industrial strains for further strain improvementbecause of unknown mutations in their genome. Thus, the aim ofthis study was to construct a genetically well defined engineered E.coli strain based on known metabolic and regulatory information,transcriptome analysis, and in silico genome-scale knockout simu-lation. Although E. coli has more complex regulatory mechanismsfor L-valine biosynthesis than any other microorganisms, manyknown regulatory mechanisms could be engineered toward en-hanced L-valine production. After the systematic metabolic engi-neering of E. coli, we were able to achieve a high yield of 0.378 gof L-valine per gram of glucose, which is even higher than that ofindustrial C. glutamicum or E. coli strains developed by randommutation and selection. More importantly, the engineered straindeveloped in this study can be further improved because all of themodifications are clearly defined, which is not easy for thoserandomly mutagenized strains.

We first constructed the L-valine-producing E. coli base strain byrational metabolic engineering by using the metabolic and regula-tory information available in the literature. Then this base strainwas improved stepwise by metabolic engineering based on newinformation obtained from transcriptome analysis and in silico geneknockout simulation. First, based on the published metabolic andregulatory information, we removed feedback inhibition and at-tenuation by chromosomal replacement and knocked out the genesresponsible for the major competing pathways. It should be men-tioned that removal of another feedback inhibition may be neces-sary to further increase L-valine production; the feedback inhibitionin the small subunit of AHAS I encoded by ilvN was not removedin this study because the site responsible for inhibition has not beenreported yet. This Val strain was used as a base strain for L-valineproduction. Then several genes encoding those enzymes directlyinvolved in L-valine biosynthesis were amplified. After the manip-ulation of all these known and rationally selected genes, no furtherengineering targets could be easily identified. Thus, we performedcomparative transcriptome profiling on the control and L-valine-producing strains to examine the differences in global transcriptlevels between the two strains.

Among the up-regulated genes was the ilvC gene encodingacetohydroxy acid isomeroreductase. It has been known that theexpression of the ilvC gene is induced by acetolactate. This induc-tion is mediated by a positive transcriptional activator encoded bythe ilvY gene (32). In the production strain, acetolactate wasaccumulated by amplified AHAS I encoded by the ilvBN genes.Thus, the accumulated acetolactate seems to result in IlvY-mediated induction of the ilvC gene expression. Among the 35genes that showed significant down-regulation, we could easilyremove the leuABD genes from the possible gene amplificationtargets; their down-regulation is beneficial for increased L-valineproduction as we originally designed. We went through the list of32 other down-regulated genes to select those that may positivelyaffect L-valine production. It should be admitted that there is nogeneral method to select the amplification target genes among thedown-regulated genes. First, the lrp gene, which encodes a globalregulator Lrp, was selected as a target; this was because Lrppositively regulates the expression of genes involved in L-valinebiosynthesis, such as ilvIH. Because the Val strain is auxotrophic to

L-leucine, we need to add L-leucine exogenously. However, L-leucine binds to Lrp and prevents it from activating L-valinebiosynthesis genes. Thus, it was thought that a sufficient level of Lrpis required to activate the expression of L-valine biosynthesis genesin the presence of exogenous L-leucine. It should be noted that E.coli has three acetohydroxyacid synthase isoenzymes (AHAS I, II,III), which are encoded by the ilvBN, ilvGM, and ilvIH genes,respectively. Among them, AHAS II is not expressed in E. coli K-12due to the frameshift mutation in the ilvG gene. Because theplasmid-born amplification of many genes can be a burden to cellgrowth, we decided to overexpress only one of them. Thus, weoverexpressed the ilvBN genes encoding AHAS I, which has ahigher affinity for pyruvate, the most important precursor forL-valine formation. Additionally, by using the Lrp, which up-regulates the expression of the ilvIH gene (encoding AHAS III),further amplification of AHAS was possible without any detrimen-tal effect of plasmid-born overexpression. Furthermore, becausethe expression of the ilvGM operon is repressed by Lrp (18), thenative promoter of the ilvGM operon, to which Lrp binds, wasreplaced with the tac promoter to abolish the repression. As a result,we used only the positive effect of Lrp on L-valine production by theoverexpression of the lrp gene. As expected, the overexpression ofthe lrp gene enhanced L-valine production by 21.6% (Table 1).Opposite results were obtained in the lrp knockout mutant, whichfurther proves that Lrp is necessary for efficient L-valine productionin E. coli.

Another amplification target gene, the ygaZH genes encoding aprotein showing significant homology to the known Corynebacte-rium L-valine exporter, could be selected from the results oftranscriptome profiling. Recently, YgaZH was suggested to be anL-valine exporter in the patent (33). We independently identifiedthis gene by BLAST search and proved by knocking-out andamplification studies that it is truly involved in the export ofL-valine. Interestingly, the level of L-valine production did notchange in the ygaZH-deleted L-valine production strain (Table 1).This result suggests that YgaZH may not be the major L-valineexporter in E. coli. Nonetheless, the overexpression of the ygaZHgenes allowed enhanced production of L-valine. Also, we were ableto show that the expression of the ygaZH genes is activated by Lrp.The cooverexpression of the lrp and ygaZH genes was synergisticand led to a 2.13-fold increase in L-valine production up to a levelof 7.61 g/liter from 50 g/liter glucose in flask culture. Recently, theL-leucine exporter in E. coli encoded by the yeaS gene has beenpublished (34). It was shown that the MICs of L-valine in E. coliTG1, the yeaS-amplified TG1, and the yeaS-deleted TG1 strainswere 0.002, 0.005, and 0.001 mg/ml, respectively. This result suggeststhat the yeaS gene product may also function as an L-valine exporterin E. coli. However, the small difference in MIC also suggests thatthe yeaS gene product may not be a major L-valine exporter in E.coli either. Identifying the major L-valine exporter in E. coli will bea good strategy for further improvement of L-valine production.

Further improvement of the L-valine-producing strain wasachieved by using the results obtained from in silico gene knockoutsimulation. There has been no report on the successful applicationof in silico prediction strategy to improving amino acid-producingstrains. By sequential in silico gene knockout studies, we were ableto identify a triple mutant strain that allowed a higher L-valineproduction rate with a reasonable growth rate. In the VAMF strain,which is a base Val strain with additional knockouts in the aceF,mdh, and pfkA genes, the flux through pyruvate was increased bydirectly blocking the pyruvate and malate dehydrogenases. Dele-tion of the pfkA gene increases the flux through the pentosephosphate pathway and thus produces more NADPH, which isrequired for L-valine biosynthesis. The VAMF strain harboringpKBRilvBNCED and pTrc184ygaZHlrp showed a 126.7% increasein L-valine production, compared with the Val strain harboring thesame plasmids. These results suggest that the beneficial effects offlux redistribution on L-valine production achieved in VAMF strain

Park et al. PNAS � May 8, 2007 � vol. 104 � no. 19 � 7801

APP

LIED

BIO

LOG

ICA

LSC

IEN

CES

could be further strengthened by implementing regulatory andexport engineering. It should be noted that the in silico simulationcurrently used does not account for any regulatory mechanisms andexporting effects. Hence, it is important to combine the results oftranscriptome profiling and in silico metabolic flux analysis, whichare complementary to each other.

In summary, rational metabolic engineering based on knownmetabolic and regulatory information, and new information de-rived from transcriptome analysis and in silico simulation, allowedfor the development of an E. coli strain capable of efficientlyproducing L-valine. An impressively high yield of 0.378 g of L-valineper gram of glucose could be achieved by using this metabolicallyengineered strain in batch culture. The approaches described in thisarticle are a good example of systematically engineering strains forthe enhanced production of amino acids, and they can also beapplied in general for developing strains for the efficient productionof other metabolites.

Materials and MethodsBacterial Strains, Plasmids, and Chromosome Manipulation. Thestrains and plasmids constructed and used in this study are listed inSI Table 2. The primers used for gene cloning and deletion are listedin SI Table 8. Detailed procedures for the construction of plasmidsare shown in SI Text and SI Figs. 5 and 7. All DNA manipulationswere carried out by following standard protocols (35). Detailedprocedures for site-directed mutagenesis and gene inactivation aredescribed in SI Text. All the genes described in this paper are listedin SI Table 9.

Cultivation and Transcriptome Profiling. Batch fermentation of wild-type and engineered E. coli strains was performed in semidefinedNM1 medium at 31°C (pH 6.0). Flask cultures were carried out in250-ml baffled flasks containing NM2 medium at 31°C and 250 rpmwith shaking incubator (SI-900; Jeio Tech, Daejeon, Korea). Cul-ture media, culture conditions, and the procedure for determininga minimal inhibitory concentration of DL-norvaline are described indetail in SI Text. Transcriptome profiling was performed by usingthe TwinChip E. coli-6K oligo chip (Digital Genomics, Seoul,Korea) following the manufacturer’s protocol. Detailed procedurescan be found in SI Text. The mean signal intensity values of theduplicate spots were averaged and then normalized by the globalnormalization method. The genes showing P values �0.05 wereconsidered to be significant.

In Silico Analysis. The in silico gene-perturbation studies wereperformed by using the genome-scale metabolic model E. coli

MBEL979 (SI Table 7 and ref. 36) consisting of 979 metabolicreactions and 814 metabolites (144 extracellular metabolites and670 intermediates), which is a slightly modified network of iJR904reported by Palsson et al. (37). Constraints-based flux analysis wasperformed on the rationally engineered Val strain, in which theilvA, leuA, and panB genes were knocked out. To implement thegene deletion during the in silico metabolic flux analysis, weprovided a constraint that sets the flux through a correspondingreaction to zero. Further gene knockout targets were identified byemploying the minimization of metabolic adjustment (MOMA)algorithm by using the linear and quadratic programming methodsby using a GAMS script (38). The base steady-state solutionrequired for the quadratic programming for MOMA was calculatedfor each simulation by first solving the linear programming problemof maximizing the growth rate. For constraints-based flux analysisof the Val strain, the glucose-uptake and L-valine productionrates of 4.37 and 0.17 mmol�gDW�1�h�1, which are the experimen-tally measured values at the exponential growth phase (�10–11 h),were used as constraints. The uptake rates of L-isoleucine, L-leucine, and D-pantothenate, which were provided in the medium,were set to 5 mmol�gDW�1�h�1, which were large enough to ensurecell growth. The simulation of metabolic flux distribution in mul-tiple gene knockout mutants was performed by a sequential ap-proach (39). A gene was first identified whose deletion resulted inthe maximum improvement in L-valine production, and double andtriple mutants were subsequently sought by searching for the genesthat upon knockout resulted in further improvement in L-valineproduction without too significant a reduction in the growth rate.To investigate the effects of sodium acetate addition on metabolicflux distribution, the acetic acid-uptake rate of 5 mmol�gDW�1�h�1was arbitrarily set, which is larger than the actual acetic acid-uptakerate.

Analytical Procedure. Cell growth was monitored by measuring theabsorbance at 600 nm (OD600) by using an Ultrospec3000 spectro-photometer (Amersham Pharmacia Biotech, Uppsala, Sweden).Real-time RT-PCR was performed as described in SI Text, and theresults were analyzed by using the 2��Ct method (40). Glucoseconcentration was measured by using a glucose analyzer (model2700 STAT; Yellow Springs Instrument, Yellow Springs, OH). Theconcentrations of amino acids were analyzed by an amino acidanalyzer (Sykam S433; Sykam, Eresing, Germany).

We thank B. L. Wanner (Purdue University, West Lafayette, IN) forproviding the plasmid pKD46. This work was supported by the KoreanSystems Biology Project of the Ministry of Science and Technology(M10309020000-03B5002-00000) and the LG Chem Chair Professorship.

1. Kromer JO, Sorgenfrei O, Klopprogge K, Heinzle E, Wittmann C (2004) J Bacteriol 186:1769–1784.

2. Lee JH, Lee DE, Lee BU, Kim HS (2003) J Bacteriol 185:5442–5451.3. Lange C, Rittmann D, Wendisch VF, Bott M, Sahm H (2003) Appl Environ Microbiol

69:2521–2532.4. Huser AT, Chassagnole C, Lindley ND, Merkamm M, Guyonvarch A, Elisakova V, Patek M,

Kalinowski J, Brune I, Puhler A, Tauch A (2005) Appl Environ Microbiol 71:3255–3268.5. Lee SY, Lee DY, Kim TY (2005) Trends Biotechnol 23:349–358.6. Nakayama K, Kitada S, Kinoshita S (1961) J Gen Appl Microbiol Tokyo 7:52–69.7. Plachy J (1975) Folia Microbiol 20:346–350.8. Elisakova V, Patek M, Holatko J, Nesvera J, Leyval D, Goergen J, Delaunay S (2005) Appl

Environ Microbiol 71:207–213.9. Gerigk M, Bujnicki R, Ganpo-Nkwenkwa E, Bongaerts J, Sprenger G, Takors R (2002)

Biotechnol Bioeng 80:746–754.10. Lee M, Smith GM, Eiteman MA, Altman E (2004) Appl Microbiol Biotechnol 65:56–60.11. Umbarger HE (1996) in Escherichia Coli and Salmonella: Cellular and Molecular Biology, eds

Neidhardt FC, Curtiss III R, Ingraham JL, Lin ECC, Low KB, Magasanik B, Reznikoff WS,Riley M, Schaechter M, Umbarger HE (ASM Press, Washington, DC), 2nd Ed, pp 442–457.

12. Eoyang L, Silverman PM (1986) J Bacteriol 185:5442–5451.13. Livshits VA (2004) US Patent 6,737,255; Chem Abstr 135:367699.14. Vyazmensky M, Sella C, Barak Z, Chipman DM (1996) Biochemistry 35:10339–10346.15. Lawther RP, Calhoun DH, Adams CW, Hauser CA, Gray J, Hatfield GW (1981) Proc Natl Acad

Sci USA 78:922–925.16. Platko JV, Willins DA, Calvo JM (1990) J Bacteriol 172:4563–4570.17. Wang Q, Calvo JM (1993) J Mol Biol 229:306–318.18. Rhee KY, Parekh BS, Hatfield GW (1996) J Biol Chem 271:26499–26507.19. Haney SA, Plakto JV, Oxender DL, Calvo JM (1992) J Bacteriol 174:108–115.20. Hung SP, Baldi P, Hatfield GW (2002) J Biol Chem 277:40309–40323.21. Lawther RP, Hatfield GW (1980) Proc Natl Acad Sci USA 77:1862–1866.

22. Lawther RP, Wek RC, Lopes JM, Pereira R, Taillon BE, Hatfield GW (1987) Nucleic Acids Res15:2137–2155.

23. Wek RC, Hauser CA, Hatfield GW (1985) Nucleic Acids Res 13:3995–4010.24. Schweizer HP (1992) Mol Microbiol 6:1195–1204.25. Barak Z, Chipman DM, Gollop N (1987) J Bacteriol 169:3750–3756.26. Calvo JM, Matthews RG (1994) Microbiol Rev 58:466–490.27. Marasco R, Varcamonti M, Cara L, Ricca E, DeFelice M, Sacco M (1994) J Bacteriol

176:5197–5201.28. Ernsting BR, Denninger JW, Blumenthal RM, Matthews RG (1993) J Bacteriol 175:7160–7169.29. Kennerknecht N, Sahm H, Yen MR, Patek M, Saier MH, Jr, Eggeling L (2002) J Bacteriol

184:3947–3956.30. Daldal F (1983) J Mol Biol 168:285–305.31. Cronan JE, LaPorte DC (1996) in Escherichia Coli and Salmonella: Cellular and Molecular

Biology, eds Neidhardt FC, Curtiss III R, Ingraham JL, Lin ECC, Low KB, Magasanik B,Reznikoff WS, Riley M, Schaechter M, Umbarger HE (ASM Press, Washington, DC), 2nd Ed,pp 206–216.

32. Wek RC, Hatfield GW (1988) J Mol Biol 203:643–663.33. Tabolina EA, Rybak KV, Khourges EM, Voroshilova EB, Gusyatiner MM (2005) US Patent

2005/0239175; Chem Abstr 137:231479.34. Kutukova EA, Livshits VA, Altman IP, Ptitsyn LR, Zyiatdinov MH, Tokmakova IL, Zakataeva

NP (2005) FEBS Lett 579:4629–4634.35. Sambrook J, Russell DW (2001) Molecular Cloning: A Laboratory Manual (Cold Spring Harbor

Lab Press, Cold Spring Harbor, NY), 3rd Ed.36. Lee SY, Woo HM, Lee D-Y, Choi HS, Kim TY, Yun H (2005) Biotechnol Bioproc Eng

10:425–431.37. Reed JL, Vo TD, Schilling CH, Palsson BO (2003) Genome Biol 4:R54.38. Segre D, Vitkup D, Church GM (2002) Proc Natl Acad Sci USA 99:15112–15117.39. Alper H, Jin YS, Moxley JF, Stephanopoulos G (2005) Metab Eng 7:155–164.40. Livak K, Shmittgen TD (2001) Methods 25:402–408.


http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1http://www.pnas.org/cgi/content/full/0702609104/DC1

Systems biotechnology for strainimprovementSang Yup Lee1,2, Dong-Yup Lee1,2 and Tae Yong Kim1

1Metabolic and Biomolecular Engineering National Research Laboratory and Department of Chemical and BiomolecularEngineering, Korea Advanced Institute of Science and Technology, 373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Korea2Department of BioSystems and Bioinformatics Research Center, Korea Advanced Institute of Science and Technology,373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Korea

Glossary

Bilevel optimization: A traditional mathematical programming problem

maximizes a single objective function over a set of feasible solutions. Bilevel

optimization seeks to maximize two objective functions simultaneously over a

set of feasible solutions.

Flux: The production or consumption of mass (metabolite) per unit area per

unit time. It is, however, often used on the basis of unit cell mass rather than

unit area in metabolic flux analysis.

High cell density culture: High cell density culture is used to increase the

cell concentration usually by well-controlled fed-batch culture. Because the

volumetric productivity (product formed per unit volume per unit time) is a

multiplication of the specific productivity (product formed per unit cell mass per

unit time) by cell concentration, high cell density culture generally increases the

volumetric productivity by the increase in the cell concentration.

IGF-I: IGF-I, insulin-like growth factor I, is a polypeptide produced primarily in

the liver that causes hypoglycemia and a transient decrease in free serum fatty

acids. It influences cellular differentiation and stimulates collagen and matrix

synthesis by bone cells.

Isotopomer: An isotopomer is specified by the number of 13C atoms in specific

positions of the molecule composed of n carbon atoms.

Leptin: A protein produced in adipocyte (fat cell) tissue, which acts as a signal

to the brain. It is being considered for treating obesity.

Linear programming: A mathematical technique that finds the maximum or

minimum of linear functions in many variables subject to constraints.

Metabolic control analysis: A phenomenological quantitative sensitivity

analysis of fluxes and metabolite concentrations.

Metabolic flux analysis: The calculation and analysis of the flux distribution of

the biochemical reaction network.

Mixed-integer optimization: Used to solve global optimization problems with

piece-wise linear objective and constraints, using special ordered sets.

Quadratic programming: Used for optimization problems in which the

objective function is a convex quadratic and the constraints are linear.

X-omics: Any omics studies currently being carried out, including genomics,

transcriptomics, proteomics,metabolomics, fluxomics, regulomics, signalomics,

physiomics and so on.

Genome: All of the genetic information and material possessed by an

organism; the entire genetic complement of an organism.

Transcriptome: The full complement of mRNAs transcribed from a cell’s

genome.

Proteome: The complete profile of proteins expressed in a given tissue, cell or

biological system at a given time.

Metabolome: The whole set of metabolites in a cell, tissue, organ, organism

and species.

Various high-throughput experimental techniques areroutinely used for generating large amou

Systems Metabolic Engineering for Bioenergy and...

Documents

Transcript of Systems Metabolic Engineering for Bioenergy and...