Metabolic engineering of lactic acid bacteria, the combined approach: kinetic modelling, metabolic control and experimental analysis

Marcel H. N. Hoefnagel1,2, Marjo J. C. Starrenburg1,3, Dirk E. Martens1,2, Jeroen Hugenholtz1,3, Michiel Kleerebezem1,3, Iris I. Van Swam1,3, Roger Bongers1,3, Hans V. Westerhoff4 and Jacky L. Snoep4,5

Wageningen Centre for Food Sciences1 and Food and Bioprocess Engineering Group,2 Wageningen University, PO Box 8129, 6700 EV Wageningen, The Netherlands
NIZO Food Research, PO Box 20, 6710 BA, Ede, The Netherlands3
BioCentrum Amsterdam, Dept of Molecular Cell Physiology, Free University, De Boelelaan 1087, NL-1081 HV Amsterdam, The Netherlands4
Dept of Biochemistry, University of Stellenbosch, Private bag X1, Matieland 7602, Stellenbosch, South Africa5

Author for correspondence: Marcel H. N. Hoefnagel. Tel: +31 317 483435. Fax: +31 317 482237. e-mail: marcel.hoefnagel{at}algemeen.pk.wau.nl


   ABSTRACT
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
Everyone who has ever tried to radically change metabolic fluxes knows that it is often harder to determine which enzymes have to be modified than it is to actually implement these changes. In the more traditional genetic engineering approaches ’bottle-necks’ are pinpointed using qualitative, intuitive approaches, but the alleviation of suspected ’rate-limiting’ steps has not often been successful. Here the authors demonstrate that a model of pyruvate distribution in Lactococcus lactis based on enzyme kinetics in combination with metabolic control analysis clearly indicates the key control points in the flux to acetoin and diacetyl, important flavour compounds. The model presented here (available at http://jjj.biochem.sun.ac.za/wcfs.html) showed that the enzymes with the greatest effect on this flux resided outside the acetolactate synthase branch itself. Experiments confirmed the predictions of the model, i.e. knocking out lactate dehydrogenase and overexpressing NADH oxidase increased the flux through the acetolactate synthase branch from 0 to 75% of measured product formation rates.

Keywords: Metabolic control analysis, in silico modelling, Lactococcus lactis, pyruvate distribution

Abbreviations: ALS, acetolactate synthase; LDH, L-lactate dehydrogenase; MCA, metabolic control analysis; NOX, NADH oxidase; (abbreviations used in rate reactions and equations are defined in Table 1)

The GenBank accession number for the sequence reported in this paper is AY046926.


   INTRODUCTION
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
Micro-organisms are not naturally optimized for maximal production rates of biotechnologically important compounds. Flavour compounds, such as diacetyl and acetaldehyde, are often secondary metabolites and are produced in insignificant amounts with respect to total carbon metabolism (Hugenholtz & Starrenburg, 1992 ). Redirecting the carbon flux towards such products without disturbing the overall cell physiology is a complex task. Metabolism is a highly branched system connected not only via carbon metabolites but also via redox [NAD(P)H] and Gibbs free energy (ATP) carriers. An intuitive approach to the optimization of fluxes towards the desired end products is not infallible in such a complex network and such an approach alone is unable to come with unexpected strategies. Several theoretical frameworks, such as biological system theory (Savageau, 1991 ), metabolic control analysis (MCA) (Kacser & Burns, 1973 ) and metabolic design (Kholodenko et al., 1998 ), have been developed to analyse multi-enzyme systems predictively and quantitatively. In MCA, so-called flux-control coefficients, which quantify the importance of an enzyme for the magnitude of a flux, are defined as the percentage change in the flux caused by a 1% modulation of the enzyme activity. Consequently, MCA points at the enzymes that will have the largest effect on the desired flux upon a very small change in any enzyme activity, but it also has the disadvantage that it does not deal with the more relevant larger changes. Upon modulation of an enzyme with a high control coefficient, kinetic modelling can be used to integrate the system and to study the effect of a more substantial perturbation in silico. Because larger increases in flux are important in bioengineering, we propose to use an integrated approach between kinetic modelling, MCA and experimentation to come to a rational strategy for genetic engineering. Here we illustrate this approach for the optimization of the metabolic flux through the acetolactate synthase (ALS) branch in lactic acid bacteria.

Lactic acid bacteria are used in milk fermentation and the major product from this process is lactate. For some dairy products, like butter, diacetyl produced by Lactococcus lactis is an important flavour component. Diacetyl is produced in only small amounts and various groups have tried to optimize its production (Monnet et al., 1994b ; Platteeuw et al., 1995 ; Swindell et al., 1996 ; Lopez de Felipe & Hugenholtz, 1999 ). Their experimental strategies have involved the overexpression and deletion of genes intuited to be important for the regulation of the carbon fluxes in L. lactis. Here, we use a more rational combination strategy consisting of (i) a detailed kinetic model of the branches around pyruvate metabolism, (ii) MCA and (iii) experiments to describe the various mutations and to illustrate the use of (i) and (ii) in a more successful metabolic engineering strategy.


   METHODS
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
Construction of an ldh deficient mutant of L. lactis NZ9000.
To construct an ldh::ery replacement variant of L. lactis NZ9000 (Kuipers et al., 1997 ), a plasmid was constructed that allowed the direct selection of the desired mutant obtained by the occurrence of a double cross-over event. The upstream flanking region of the ldh gene of the las operon was amplified by PCR using MG1363 genomic DNA as a template and the primers PYKECO-5 (5'-CTCGTGAATTCGAAGTTGAAGC-3') and LDHREV-6 (5'-ATCAGCCATGGTTTTCTTTAATTCC-3') (EcoRI and NcoI sites in bold, respectively). Similarly, the downstream flanking region of the ldh gene of the las operon was amplified by PCR using the primers LDHDOW-4 (5'-GACACCATGGCCGCGGATCCAAATAAAAAAGAGTTAGTAGAAAATTTCAAC-3') and 317REV-12 (5'-CGCATCTAGACAACACCTCTAATCTTTAC-3') (NcoI, BamHI and XbaI sites in bold, respectively). Both the up- and downstream fragments, of approximately 1·1 kb each, were cloned as EcoRI–NcoI and NcoI–XbaI fragments, respectively, into similarly digested pUCNCO. The sequence of the cloned fragment was verified by automated sequence analysis with an ALF DNA sequencer (Pharmacia Biotech). Sequence reactions were performed according to the manufacturer’s protocols, using the autoread sequencing kit and fluorescein-labelled universal M13 primers. Both of the ldh flanking regions were re-isolated as EcoRI–NcoI (5') and NcoI–XbaI (3') fragments and cloned into EcoRI–XbaI-digested pUC19 in a three-point ligation. A BamHI–XbaI fragment of pUC19ERY (Kuipers et al., 1995 ), containing the erythromycin-resistance gene, was introduced into the BamHI restriction site of the resulting pUCAB plasmid after the cohesive ends had been filled using the Klenow fragment of DNA polymerase I of Escherichia coli. The resulting plasmid, in which the orientation of the erythromycin gene was the same as the preceding pyk gene, was designated pUCAeryB. Finally, a SwaI–Ecl136II fragment of pGhost8 (Chopin et al., 1984 ), containing the tetracycline-resistance gene, was cloned into the XbaI site of pUCAeryB after filling in the XbaI cohesive ends using Klenow. The resulting plasmid, designated pUCAeryBTc, was transformed into competent L. lactis NZ9000 cells and erythromycin-resistant colonies were selected; the resulting transformants were analysed for tetracycline resistance by replica plating. Colonies displaying erythromycin resistance and tetracycline sensitivity should have been the result of a one-step double-cross-over event on both flanking regions of the ldh gene of the las operon and should have contained the desired ldh::ery replacement mutation. The anticipated genetic organization of the ldh::ery mutant candidates was verified by PCR and Southern analysis; the resulting ldh::ery variant of L. lactis NZ9000 was designated NZ9010.

Cloning of the lactococcal nox gene and construction of the nox overexpression plasmid.
The nox gene of Streptococcus mutans, derived from pNZ2600 (Lopez de Felipe et al., 1998 ), was used as a heterologous probe in Southern blotting experiments to clone the NADH-oxidase(NOX)-encoding gene of L. lactis MG1363. A 1·8 kb fragment of EcoRI-digested chromosomal DNA of L. lactis MG1363 hybridized with this probe. The 1·8 kb fragment was cloned into EcoRI-digested pUC19 and colonies containing the hybridizing fragment were selected by colony blotting, again using the nox gene of S. mutans as a probe. Sequence analysis of the 1·8 kb fragment revealed that it contained the 3' end of a gene that had high similarity with the nox gene of S. mutans, putatively representing the 3' end of the lactococcal nox gene. Moreover, downstream of the 3' end of this putative nox gene, ORFs were found that shared a high level of similarity (43·8% at the protein sequence level) with the ssb gene of B. subtilis, encoding a putative single-stranded DNA-binding protein (Meyer & Laine, 1990 ). Following the ssb gene, the 5' end of the previously cloned groES operon of L. lactis was found (Kim & Batt, 1993 ).

The 5' end of the lactococcal nox gene was amplified by PCR using L. lactis MG1363 chromosomal DNA as a template and the fully degenerated primer NOX2F (5'-ACNGGNACYGAYCANGCNGCNGGYATHGC-3'; N=A/C/G/T, Y=C/T and H=A/C/T), based on the previously determined N-terminal sequence of the lactococcal NOX protein (Lopez de Felipe & Hugenholtz, 2001 ), combined with primer NOXR (5'-TGACCTGCAGTTCTGCGTCAATTGCTTGACC-3'), which was based on the 3' region of the cloned nox sequence, and was extended with a PstI-generating clamp sequence (indicated in bold). The approximately 400 bp fragment obtained was cloned into pGEMT (Promega) and sequenced; the sequence revealed that the fragment contained the 5' region of nox. This fragment (the NcoI–PstI fragment from the pGEMT clone) was used as a probe in the cloning of a hybridizing 1·4 kb EcoRI L. lactis MG1363 chromosomal fragment. Sequence analysis of this fragment revealed that it contained the entire 5' region of the lactococcal nox gene. Upstream of nox, an ORF encoding a hypothetical protein displaying local similarity with a phenylalanine tRNA ligase (78% identity with the YdjD protein of L. lactis IL1403) was found. The sequence of the combined EcoRI fragments, containing the entire noxE sequence (93% identity with the NoxE protein of L. lactis IL1403), has been deposited in the GenBank database under accession number AY046926.

The water-forming NOX-encoding nox gene of L. lactis was amplified by PCR using L. lactis MG1363 chromosomal DNA as a template and the primers NOXF (5'-CGTACCATGGAAATCGTAGTTATCGGTAC-3') and NOXR (5'-CGTATCTAGATTCAAAAGCCTGCCTACTGTGC-3'). The PCR fragment obtained was cloned as a NcoI–XbaI fragment (restriction sites were introduced into the primers and are indicated in bold) into similarly digested pNZ8048 (de Ruyter et al., 1996 ; Kuipers et al., 1997 ). The resulting plasmid was designated pNZ2610 and contained the lactococcal nox gene translationally fused to the nisA promoter. Transcription of nox in this construct is dependent on the activity of the nisin-inducible nisA promoter, whereas translation of the nox transcript depends on the nisA-derived RBS (de Ruyter et al., 1996 ).

Control strain.
L. lactis NZ9000(pNZ8048) was used as a control during this study.

Fermentation.
Cultures were grown at 30 °C in M17 medium (Merck) supplemented with 1% (w/v) glucose. Chloramphenicol and erythromycin were used at 10 and 5 µg ml-1, respectively. Nisin was used at 1 ng ml-1. A 1 l bioreactor (Applikon Dependable Instruments) was inoculated with cells from an overnight culture to an initial OD600 of about 0·1 in 700 ml medium. A pH of 6·5 was maintained by the addition of 2 M NaOH and the stirrer speed was set at 500 r.p.m. Air was bubbled through with a flow rate of 520 ml min-1. The batch cultures used showed exponential growth between about 2 and 5·5 h post-inoculation. The flux distribution was calculated during this exponential (pseudo steady-state) phase.

Analysis of fermentation products.
Glucose, lactate, acetate, formate, ethanol, acetoin and 2,3-butanediol were analysed by HPLC, as described previously (Starrenburg & Hugenholtz, 1991 ). In contrast to the wild-type, the carbon recovery for the mutant strains was incomplete (between 70 and 85% recovery). This may have been due to activity of the pentose phosphate pathway, which leads to CO2 release. However, this can not be confirmed as this pathway was not monitored in this study.

Growth and enzyme assays.
The OD600 was determined and corrected for the optical density of the growth medium. When the OD600 value was too high (>0·75) the sample was diluted with medium.

L-Lactate dehydrogenase (LDH) activity was determined by the method of Hillier & Jago (1982) and the NOX and ALS activities were determined according to Lopez de Felipe et al. (1998) and Platteeuw et al. (1995) , respectively. Protein concentrations were determined according to the Bradford method (Bradford, 1976 ), with BSA as a standard. Preliminary experiments along these lines have been published previously (Platteeuw et al., 1995 ; Lopez de Felipe et al., 1998 ), but they were not performed in an isogenic background and did not contain precise measurements of catabolic fluxes. Instead, they consisted merely of end-point determinations of product concentrations. This made them unsuitable for a proper demonstration of the rational engineering procedure we propose here.

The kinetic model.
A set of ordinary differential equations was used to describe the time dependence of the metabolite concentrations. Table 1 gives definitions of the mathematical symbols and abbreviations used in the equations and kinetic models in this study. Please note that the flux of glycolysis is given in C3 units.


View this table:
[in this window]
[in a new window]
 
Table 1. Mathematical symbols and abbreviations for equations and kinetic models used in this study

 
Most of the enzymes were modelled using a reversible Michaelis–Menten equation with noncompeting substrate–product couples. When the specific enzyme has been characterized with respect to the kinetic type, then this type was used. For vPDH (reaction 3) an inhibition by a high NADH:NAD ratio has been described and these kinetics have been included in the equation. For vPTA (reaction 4) the equation was derived from Henkin & Abeles (1976) . For vALS (reaction 8) the reversible Hill equation, as derived by Hofmeyr & Cornish-Bowden (1997) , was used. For vATPase (reaction 12) an irreversible Hill equation, sensitive to the ATP:ADP ratio, was used. For reaction 9 and 12 kinetics for an irreversible reacion were used. Simple Michaelis–Menten kinetics were used for reaction 10. Reaction 14 represents non-enzymic decarboxylation of acetolactate. The rate equations shown in Table 2 were used.


View this table:
[in this window]
[in a new window]
 
Table 2. Rate equations used in this study

 
The concentrations of the following metabolites were assumed to be constant: glucose, 15 mM; lactate, 1 mM; inorganic phosphate, 10 mM; ethanol, 1 mM; butanediol, 0·01 mM; oxygen, 0·2 mM; and acetate, 1 mM. The sums of [ATP]+[ADP], [NAD]+[NADH] and [coenzyme A]+[acetyl coenzyme A] were assumed to be constant at the respective values of 5, 10 and 1 mM.

Kashket & Wilson (1973) found that in L. lactis the internal cell volume is 1·5 µl (mg dry weight)-1. We assumed that protein makes up 50% of the dry weight, as it does in E. coli (Neidhardt & Umbarger, 1996 ), and that an OD600 of 1 is equal to 0·445 mg dry weight ml-1.

Kinetic parameters.
The kinetic parameters used for the enzymes in reactions 1–15 (Fig. 1) and the rate equations in Table 2 can be found in Table 3



View larger version (26K):
[in this window]
[in a new window]
 
Fig. 1. Reactions included in the model to describe the distribution of carbon from pyruvate in L. lactis. Numbers in circles indicate the following enzymes or steps: 1, ‘lumped’ glycolysis; 2, LDH; 3, pyruvate dehydrogenase; 4, phosphotransacetylase; 5, acetate kinase; 6, acetaldehyde dehydrogenase; 7, alcohol dehydrogenase; 8, ALS; 9, acetolactate decarboxylase; 10, acetoin efflux; 11, acetoin dehydrogenase; 12, ATPase; 13, NOX; 14, non-enzymic acetolactate decarboxylation; 15, pyruvate formate lyase, which is considered not to be active under aerobic conditions (see Table 3); 16, chemical conversion to diacetyl, not included in the model. Substrates and products that were clamped in the model are indicated in italics.

 

View this table:
[in this window]
[in a new window]
 
Table 3. Kinetic parameters used for the enzymes in reactions 1–15

 

   RESULTS
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
Modelling the pyruvate branches in L. lactis
We first constructed a kinetic model describing the flux distribution over the pyruvate branches in L. lactis (Fig. 1). Glycolysis was modelled as a single irreversible step (Fig. 1, 1) with product inhibition; the ATP-consuming reactions (Fig. 1, 12) were grouped in a single module. The kinetic parameters of these two modules were estimated on the basis of ‘fits’ to in vivo measurements of fluxes (see Methods and Table 3). In contrast to these unstructured modules, all of the kinetic steps in the pyruvate branches were modelled in detail. In vitro kinetic data from the literature were used as input. A detailed description of the model is given in Methods and Table 3. The model prediction of the flux distribution in the wild-type strain (white lettering) as compared to the experimental data (black lettering) is shown in Fig. 2(a). The model predicted an essentially (97%) homolactic fermentation, which is in agreement with experimental observations (Fig. 2), where between 90 and 100% of the glucose is converted to lactate (Hugenholtz & Starrenburg, 1992 ; Platteeuw et al., 1995 ). The flux through the acetolactate branch was predicted to be less than 0·1% of the glucose influx, in agreement with what was observed experimentally (Fig. 2a). Thus, in the wild-type strain only a negligible flux was directed towards the branch of interest and this can be understood on the basis of the known enzyme kinetics (i.e. the model gives a good description of this condition). What strategy can now be followed to optimize the flux through this branch?



View larger version (16K):
[in this window]
[in a new window]
 
Fig. 2. Flux distribution over the pyruvate branches in L. lactis. Experimental results are shown in black letters on a white background, whereas model predictions are shown in white letters on a black background. The fluxes are given in mmol (l internal vol.)-1 min-1 [multiply by 290 to obtain µmol (mg protein)-1 min-1]. (a) Wild-type; (b) Ldh- mutant; (c) NOX overexpression mutant; (d) Ldh- and NOX overexpression mutant. As can be seen from the sum of the fluxes, the carbon balance is not complete in all strains. In (b, d) pyruvate that accumulated in the medium accounted for up to 15% of the carbon flux.

 
MCA quantifies the importance of each of the enzymes in controlling the flux through the acetolactate branch. Thus, the enzyme with the highest flux-control coefficient is the enzyme that will have the biggest effect upon a small modulation around its wild-type level, suggesting that MCA is a good approach for genetic engineering. But is it?

MCA versus the intuitive approach
In an intuitive approach ALS would be the enzyme of choice for genetic engineering, as it is the first dedicated enzyme in the branch and it is far from equilibrium. But what does MCA say? At first sight it might seem to say the same, i.e. in the wild-type situation the ALS has a flux-control coefficient of 1·0 for the flux to acetoin (Table 4. This qualifies as a high flux-control coefficient, as flux-control coefficients always sum up to 1. However, flux-control coefficients can be negative and the complete MCA analysis shows how relevant this is for this branched pathway – the highest flux-control coefficients for the acetolactate branch reside in enzymes outside this branch! (Namely LDH and NOX; Table 4). This corresponds to an early MCA observation for branched pathways, in which the fluxes over the different branches differ by orders of magnitude (Kacser, 1983 ). With hindsight, this can be understood also intuitively: a minor decrease in any branch carrying a major flux from pyruvate might increase the pyruvate concentration more than proportionately provided that the concentration exceeded the apparent Michaelis constant for that major pathway. This could then result in a more than proportional increase in flux through the ALS branch.


View this table:
[in this window]
[in a new window]
 
Table 4. Control coefficients of the different steps/enzymes on the flux through ALS as predicted by the model for the wild-type and three mutants

 
MCA suggested that ALS should have a substantial control on the ALS flux, implying that a doubling of the activity of the enzyme could at most double the very small flux towards acetoin. However, we were seeking much more substantial increases. Moreover, the overexpression of enzymes with high control coefficients is often not successful because the control coefficient tends to decrease strongly as the enzyme is overexpressed (i.e. the enzyme stops being limiting). A calculation with the kinetic model (not shown) showed that increasing the amount of ALS by a factor of 20 [the overexpression level found in Platteeuw et al. (1995) ] increased the ALS flux to only 0·02% of the glucose flux. Indeed, Platteeuw et al. (1995) demonstrated that in anaerobically cultured L. lactis (MG5267) overexpression of ALS did not increase the production of acetoin and 2,3-butanediol. Therefore, we decided that overexpression of ALS was not going to be a successful strategy to increase the flux through ALS.

LDH knockout
LDH has the highest (negative) flux-control coefficient (-2·3; Table 4). Thus, a 1% reduction in LDH activity should lead to a 2·3% increase in the flux through the acetolactate branch. After MCA has indicated the best candidate for genetic engineering, modulations substantially exceeding 1% can be test-run in the kinetic model. Such a test-run is an important part of the genetic engineering strategy, since MCA is defined at the original steady state and the size of the optimal perturbation might be limited due to kinetic restrictions in the system. For example, if the acetolactate dehydrogenase and pyruvate dehydrogenase complex branches have only a low capacity these branches might not be able to absorb all of the carbon flow from lactate upon a knockout mutation in LDH. Such an effect was observed in the model calculations, where pyruvate and NADH accumulated upon a deletion in LDH and the glycolytic flux was severely inhibited (Fig. 2b). Such a major perturbation of primary metabolism is bound to lead to a strong reduction in growth rate and possibly to deleterious effects in Ldh- strains. Therefore, even though all of the remaining flux was directed into the ALS branch, this modulation seemed ill-advised (Fig. 2b).

Experimentally, the overall carbon flux was reduced by 11% but this reduction was not as severe as the model had predicted. The major product formed was acetoin (50% of the measured product formation rates, Fig. 2), which was in agreement with the model prediction.

Overexpression of NOX
In addition to LDH, NOX also had a high (positive) control coefficient (Table 4). The model predicted that about 20% of the glucose would be converted via the ALS branch if NOX was 40-fold overexpressed, whereas in the experimental set-up 13% was converted to acetoin (Fig. 2c).

The strongest effect was expected when the LDH knockout mutant was combined with NOX overexpression. The model predicted that, under these conditions, 92% of the pyruvate would be converted via the acetolactate branch, but with a much higher glycolytic flux when compared to the situation in which only LDH was deleted (Fig. 2d). Indeed, in the experimental situation acetoin was the main product (75%) of the recovered carbon.


   DISCUSSION
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
In this study we have illustrated the combined use of MCA, kinetic modelling and experimentation to increase dramatically the flux through a pathway that is of biotechnological relevance. The predicted increase in flux was substantial enough to be of great industrial interest and was confirmed experimentally.

The model
L. lactis is an ideal organism for modelling purposes, as it is well-studied, genetically accessible and can be cultured readily. Furthermore, its anabolic reactions can be simulated as being separate from the catabolic reactions with respect to carbon metabolism: when L. lactis is cultured on rich medium almost all the carbon (95%) from the free-energy source (e.g. glucose) is recovered as external products (e.g. lactate) (Novak et al., 1998 ). This simplifies the modelling process considerably. The only link between anabolism and catabolism that was considered in this study was ATP and this could be modelled as a single step; all ATP-consuming reactions in anabolism were grouped into a single module. Biomass formation from glucose was assumed to be, overall, a redox-neutral process; the homolactic fermentation in the wild-type indicates that catabolism itself is redox neutral.

The accuracy of the model depends largely on the kinetic data on which it was built. These data were taken from the literature and were determined by several different groups working with different organisms and they were not always determined under physiological conditions. Also, sometimes no data were available for L. lactis. In these instances, data either from another streptococcal species or from other bacteria were used (see Table 3). Another limitation of the model is that a constant level of gene expression was assumed, i.e. the enzyme levels that were measured in the wild-type strain were assumed not to change during the genetic manipulation steps.

The model is available online at http://jjj.biochem.sun.ac.za/wcfs.html and it can be run from this site. As well as the standard simulations, the site also allows the user to set various parameters and to test all sorts of mutants.

Designing the optimal strain: reducing trial and error
Indeed, if the kinetic parameters for the system under study are available then the construction of a kinetic model is relatively simple and quick. Even without all of the kinetic parameters available, one can sometimes make simplifications that allow for the construction of a model that can be useful. For instance, here we have focused on the flux distribution over the pyruvate branches and have grouped all of the reactions outside these branches, thereby avoiding modelling all of the glycolytic steps in detail.

Once a kinetic model has been constructed it is simple to calculate the control coefficients on the flux (or on the concentration of a metabolite) that one wishes to optimize. On the basis of the control coefficients one can then design a genetic engineering strategy, which can again be tested in the kinetic model. This latter testing is important, as one will want to make large changes in enzyme concentrations in order to attain biotechnologically interesting improvements. If indeed the perturbation leads to an appreciable improvement of the production strain in silico one can then move to the experimental process of genetic engineering.

All in all, the model we used was not perfect. Therefore, our finding that it was very useful in the engineering strategy, and that it even came close to predicting its results quantitatively, may reflect the phenomenon that metabolic fluxes have limited sensitivities to many kinetic parameters. More generally, it suggests that the frequent adage that biotechnology is not yet ready for the application of mathematical models may be much too pessimistic.


   ACKNOWLEDGEMENTS
 
For the genetic engineering of the various L. lactis strains the financial support of the EU-GEMOLAB project (BIO4-CT98-0118) is gratefully acknowledged.


   REFERENCES
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
 
Abbe, K., Takahashi, S. & Yamada, T. (1982). Involvement of oxygen-sensitive pyruvate formate-lyase in mixed-acid fermentation by Streptococcus mutans under strictly anaerobic conditions. J Bacteriol 152, 175-182.[Medline]

Bradford, M. M. (1976). A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein–dye binding. Anal Biochem 72, 248-254.[Medline]

Bresters, T. W., De Kok, A. & Veeger, C. (1975). The pyruvate-dehydrogenase complex from Azotobacter vinelandii. 2. Regulation of the activity. Eur J Biochem 59, 347-353.[Abstract]

Cachon, R. & Divies, C. (1993). Localization of Lactococcus lactis ssp. lactis bv. diacetylactis in alginate gel beads affects biomass density and synthesis of several enzymes involved in lactose and citrate metabolism. Biotechnol Tech 7, 453-456.

Carballo, J., Martin, R., Bernardo, A. & Gonzalez, J. (1991). Purification, characterization and some properties of diacetyl (acetoin) reductase from Enterobacter aerogenes. Eur J Biochem 198, 327-332.[Abstract]

Chopin, A., Chopin, M. C., Moillo-Batt, A. & Langella, P. (1984). Two plasmid-determined restriction and modification systems in Streptococcus lactis. Plasmid 11, 260-263.[Medline]

Crow, V. L. & Pritchard, G. G. (1977). Fructose-1,6-diphosphate-activated L-lactate dehydrogenase from Streptococcus lactis: kinetic properties and factors affecting activation. J Bacteriol 131, 82-91.[Medline]

de Ruyter, P. G., Kuipers, O. P. & de Vos, W. M. (1996). Controlled gene expression systems for Lactococcus lactis with the food-grade inducer nisin. Appl Environ Microbiol 62, 3662-3667.[Abstract/Free Full Text]

Fox, D. K. & Roseman, S. (1986). Isolation and characterization of homogeneous acetate kinase from Salmonella typhimurium and Escherichia coli. J Biol Chem 261, 13487-13497.[Abstract/Free Full Text]

Gibson, T. D., Parker, S. M. & Woodward, J. R. (1991). Purification and characterization of diacetyl reductase from chicken liver and Streptococcus lactis and enzymic determination of diacetyl and diketones. Enzyme Microb Technol 13, 171-178.

Henkin, J. & Abeles, R. H. (1976). Evidence against an acyl-enzyme intermediate in the reaction catalyzed by clostridial phosphotransacetylase. Biochemistry 15, 3472-3479.[Medline]

Hillier, A. J. & Jago, G. R. (1982). L-Lactate dehydrogenase, FDP-activated, from Streptococcus cremoris. Methods Enzymol 89 Pt D, 362–367.[Medline]

Hofmeyr, J. H. & Cornish-Bowden, A. (1997). The reversible Hill equation: how to incorporate cooperative enzymes into metabolic models. Comput Appl Biosci 13, 377-385.[Abstract]

Hugenholtz, J. & Starrenburg, M. J. C. (1992). Diacetyl production by different strains of Lactococcus lactis ssp. lactis var. diacetylactis and Leuconostoc spp. Appl Microbiol Biotechnol 38, 17-22.

Kacser, H. (1983). The control of enzyme systems in vivo: elasticity analysis of the steady state. Biochem Soc Trans 11, 35-40.[Medline]

Kacser, H. & Burns, J. A. (1973). The control of flux. Symp Soc Exp Biol 27, 65-104.[Medline]

Kashket, E. R. & Wilson, T. H. (1973). Proton-coupled accumulation of galactoside in Streptococcus lactis 7962. Proc Natl Acad Sci USA 70, 2866-2869.[Abstract]

Kholodenko, B. N., Cascante, M., Hoek, J. B., Westerhoff, H. V. & Schwaber, J. (1998). Metabolic design: how to engineer a living cell to desired metabolite concentrations and fluxes. Biotechnol Bioeng 59, 239-247.[Medline]

Kim, S. G. & Batt, C. A. (1993). Cloning and sequencing of the Lactococcus lactis subsp. lactis groESL operon. Gene 127, 121-126.[Medline]

Kuipers, O. P., Beerthuyzen, M. M., de Ruyter, P. G., Luesink, E. J. & de Vos, W. M. (1995). Autoregulation of nisin biosynthesis in Lactococcus lactis by signal transduction. J Biol Chem 270, 27299-27304.[Abstract/Free Full Text]

Kuipers, O. P., de Ruyter, P. G. G. A., Kleerebezem, M. & De Vos, W. M. (1997). Controlled overproduction of proteins by lactic acid bacteria. Trends Biotechnol 15, 135-140.[Medline]

Lopez de Felipe, F. & Hugenholtz, J. (1999). Pyruvate flux distribution in NADH-oxidase-overproducing Lactococcus lactis strain as a function of culture conditions. FEMS Microbiol Lett 179, 461-466.[Medline]

Lopez de Felipe, F. & Hugenholtz, J. (2001). Purification and characterization of the water forming NADH oxidase of Lactococcus lactis. Int Dairy J 11, 37-44.

Lopez de Felipe, F., Kleerebezem, M., De Vos, W. M. & Hugenholtz, J. (1998). Cofactor engineering: a novel approach to metabolic engineering in Lactococcus lactis by controlled expression of NADH oxidase. J Bacteriol 180, 3804-3808.[Abstract/Free Full Text]

Mendes, P. (1993). GEPASI: a software package for modelling the dynamics, steady states and control of biochemical and other systems. Comput Appl Biosci 9, 563-571.[Abstract]

Meyer, R. R. & Laine, P. S. (1990). The single-stranded DNA-binding protein of Escherichia coli. Microbiol Rev 54, 342-380.

Monnet, C., Phalip, V., Schmitt, P. & Divies, C. (1994a). Comparison of alpha-acetolactate synthase and alpha-acetolactate decarboxylase in Lactococcus spp. and Leuconostoc spp. Biotechnol Lett 16, 257-262.

Monnet, C., Schmitt, P. & Divies, C. (1994b). Diacetyl production in milk by an alpha-acetolactic acid accumulating strain of Lactococcus lactis ssp. lactis biovar diacetylactis. J Dairy Sci 77, 2916-2924.[Abstract/Free Full Text]

Neidhardt, F. C. & Umbarger, H. E. (1996). Chemical composition of Escherichia coli. In Escherichia coli and Salmonella, Cellular and Molecular Biology, 2nd edn, pp. 13–16. Edited by F. C. Neidhardt and others. Washington, DC: American Society for Microbiology.

Novak, L., Cocaign-Bousquet, M., Lindley, N. D. & Loubiere, P. (1998). Cometabolism sugar–amino acids in Lactococcus lactis. Lait 78, 17-22.

Petit, C., Vilchez, F. & Marczak, R. (1989). Influence of citrate on the diacetyl and acetoin production by fully grown cells of Streptococcus lactis ssp. diacetylactis. Curr Microbiol 19, 319-324.

Platteeuw, C., Hugenholtz, J., Starrenburg, M. J. C., Van Alen-Boerrigter, I. & De Vos, W. M. (1995). Metabolic engineering of Lactococcus lactis: influence of the overproduction of alpha-acetolactate synthase in strains deficient in lactate dehydrogenase as a function of culture conditions. Appl Environ Microbiol 61, 3967-3971.[Abstract]

Savageau, M. A. (1991). Biochemical systems theory: operational differences among variant representations and their significance. J Theor Biol 151, 509-530.[Medline]

Schmidt, H. L., Stocklein, W., Danzer, J., Kirch, P. & Limbach, B. (1986). Isolation and properties of an H2O-forming NADH oxidase from Streptococcus faecalis. Eur J Biochem 156, 149-155.[Abstract]

Shone, C. C. & Fromm, H. J. (1981). Steady-state and pre-steady-state kinetics of coenzyme A linked aldehyde dehydrogenase from Escherichia coli. Biochemistry 20, 7494-7501.[Medline]

Snoep, J. L., Teixeira De Mattos, M. J., Starrenburg, M. J. C. & Hugenholtz, J. (1992a). Isolation, characterization, and physiological role of the pyruvate dehydrogenase complex and alpha-acetolactate synthase of Lactococcus lactis ssp. lactis bv. diacetylactis. J Bacteriol 174, 4838-4841.[Abstract]

Snoep, J. L., Westphal, A. H., Benen, J.-A. E., Teixeira De Mattos, M. J., Neijssel, O. M. & De Kok, A. (1992b). Isolation and characterization of the pyruvate dehydrogenase complex of anaerobically grown Enterococcus faecalis NCTC 775. Eur J Biochem 203, 245-250.[Abstract]

Snoep, J. L., De Graef, M. R., Westphal, A. H., De Kok, A., Teixeira De Mattos, M. J. & Neijssel, O. M. (1993). Differences in sensitivity to NADH of purified pyruvate dehydrogenase complexes of Enterococcus faecalis, Lactococcus lactis and Escherichia coli: implications for their activity in vivo. FEMS Microbiol Lett 114, 279-283.[Medline]

Starrenburg, M. J. C. & Hugenholtz, J. (1991). Citrate fermentation by Lactococcus and Leuconostoc spp. Appl Environ Microbiol 57, 3535-3540.

Strecker, H. J. & Harary, I. (1954). Bacterial butylene glycol dehydrogenase and diacetyl reductase. J Biol Chem 211, 263-270.[Free Full Text]

Swindell, S. R., Benson, K. H., Griffin, H. G., Renault, P., Ehrlich, S. D. & Gasson, M. J. (1996). Genetic manipulation of the pathway for diacetyl metabolism in Lactococcus lactis. Appl Environ Microbiol 62, 2641-2643.[Abstract]

Thauer, R. K., Jungermann, K. & Decker, K. (1977). Energy conservation in chemotrophic anaerobic bacteria. Bacteriol Rev 41, 100-180.[Medline]

Wills, C., Kratofil, P., Londo, D. & Martin, T. (1981). Characterization of the two alcohol dehydrogenases of Zymomonas mobilis. Arch Biochem Biophys 210, 775-785.[Medline]

Received 5 September 2001; revised 5 November 2001; accepted 20 November 2001.