Much less work has been performed with respect to investigating the control of the glycolytic flux in L. lactis. Andersen and colleagues used modulation of gene expression to show that lactate dehydrogenase has no control on the glycolytic flux in L. lactis MG1363 (1). Phosphofructokinase (PFK), on the other hand, appeared to be present in a very low excess, and even a small reduction of PFK activity resulted in an almost proportional decrease in the glycolytic flux (2). However, recent studies showed that PFK has no control over the glycolytic flux either (25; B. J. Koebmann et al., unpublished data). Poolman and coworkers used inhibition with iodoacetate to investigate the control by glyceraldehyde-3-phosphate dehydrogenase (GAPDH) in nongrowing cells of L. lactis subsp. cremoris strain Wg2, and their results suggested that GAPDH had almost full control (90%) over the glycolytic flux (33). Even and coworkers later showed that also in the prototrophic L. lactis strain NCDO2118, growing slowly on lactose, GAPDH was an important enzyme, with about 30% of the flux control (9). However, flux control can also reside outside the pathway itself, for instance, in all of the ATP-consuming processes. In Escherichia coli, the glycolytic flux appears to be almost exclusively controlled by the processes that consume ATP (23). A similar study with L. lactis MG1363 showed that ATP consumption controls the glycolytic flux in nongrowing cells but not during fast growth (24). This result implies that the control of glycolytic flux in fast-growing cells probably resides in the glycolytic reactions themselves, which would fit with the observations of Poolman et al. (33).

The implication of a high degree of control by GAPDH over the glycolytic flux is that a small change in GAPDH activity should result in an almost proportional change in the magnitude of the glycolytic flux, which could have important implications for the use of L. lactis as a starter culture. We therefore constructed a series of mutants which have altered activity of GAPDH, both below and above the activity in the wild-type strain MG1363. We found that changes in GAPDH activity around the normal activity had virtually no effect on the glycolytic flux in growing as well as nongrowing cells. We conclude that the control by GAPDH over the glycolytic flux in MG1363 is close to zero.

Bacterial strains and plasmids. The bacterial strains and plasmids used in this study are listed in Table 1.

TABLE 1. Bacterial strains and plasmids

Growth media and growth conditions. E. coli strains were grown aerobically at 37°C in Luria-Bertani (LB) broth (34). L. lactis was routinely cultivated at 30°C without aeration in M17 broth (37) or in chemically defined SA medium (19) supplemented with 10 g of glucose per liter and appropriate selective antibiotics.

Growth experiments with L. lactis were performed with batch cultures (flasks) at 30°C in 100 ml of SA medium (19) supplemented with 1 to 1.2 g of glucose per liter and appropriate selective antibiotics. Slow stirring with magnets was used to keep the culture homogenous. Regular measurements of optical density at 450 nm (OD₄₅₀) or of OD₆₀₀ were made, and samples were taken to measure the product formation and glucose consumption by high-pressure liquid chromatography (HPLC). The cell density was correlated to the cell mass of L. lactis as 0.36 g (dry weight)/liter of SA medium for an OD₆₀₀ of 1 and 0.19 g (dry weight)/liter SA medium for an OD₄₅₀ of 1 (2). The glycolytic fluxes were determined by HPLC.

Antibiotics. Erythromycin was used at 5 μg/ml for L. lactis and 200 μg/ml for E. coli, and tetracycline was used at 5 μg/ml for L. lactis and 8 μg/ml for E. coli.

DNA techniques and DNA isolation. Extraction of chromosomal DNA from L. lactis was performed as previously described (22) with the following modifications. A final concentration of 15.5 mg of lysozyme per ml was used for lysis at 37°C for 30 min. The sample was then incubated with sodium dodecyl sulfate for 10 min at 37°C followed by 10 min at 65°C. PCR amplification with Elongase enzyme (Invitrogen, Tåstrup, Denmark) was done as recommended by the manufacturer. Plasmid DNA from E. coli for analytic purposes was isolated by using an alkaline lysis method (34), and that for preparative purposes was isolated by using columns from Qiagen (Hilden, Germany). Plasmid DNA from L. lactis was isolated as described by Birnboim and Doly (4) with the following modification: a final concentration of 15.5 mg of lysozyme per ml for 20 min was used for lysis of the cell membrane. Digestion with restriction enzymes (Gibco BRL, Pharmacia, and New England Biolabs) and treatment with T4 DNA ligase (Gibco BRL) and calf intestinal alkaline phosphatase (Pharmacia) were carried out by standard recombinant DNA techniques as described by Sambrook et al. (34) and as prescribed by the manufacturers. DNA fragments were purified from agarose gels by using GFX columns (Pharmacia) or the High Pure PCR product purification kit (Boehringer Mannheim). Linearized cloning vectors were treated with calf intestinal alkaline phosphatase to avoid religation of the vector.

Transformation. E. coli strains were made competent by CaCl₂ treatment (34). After transformation, the cells were regenerated in LB medium (34) and subsequently transferred to LB agar plates supplemented with the appropriate antibiotic. Cells of L. lactis were made competent by growth in GM17 medium supplemented with 10 g of glycine per liter and resuspended in 100 g of glycerol per liter-0.5 M sucrose as described by Holo and Nes (17). Plasmid DNA was used to transform the cells by electroporation (17), and the cells were allowed to regenerate in Schmidt-Ruppin medium (32) for 2 h and then plated on GM17 agar plates supplemented with appropriate selective antibiotics. Histochemical screening of gusA was performed with 5-bromo-4-chloro-3-indolyl-β-d-glucuronidase (X-Gluc), at final concentration of 100 μg/ml.

Primers. The following primers were used in this study: CpgapA (5′-ACGACTAGTGGATCCATNNNNNAGTTTATTCTTGACANNNNNNNNNNNNNNTGRTATAATANNNNNNNAGAGATAAGAGATGTCTCC-3′), CpgapB (5′-ACGACTAGTGGATCCATNNNNNAGTTTATTCTTGACANNNNNNNNNNNNNNTGRTATAATANNNGAGTAAGTTAAATTGTTAACTTAG-3′), gapBback (5′-CTCTACATGCATGTTTTTATACCGTTAAAATCGG-3′), gapB′back (5′-CTCTACGGTACCACAAGACCGATAGCTTTAG-3′), gapAback (5′ GTATGTCGACGAATACTACAACAAAGAAGGC-3′), and T7 primer (5′-GTAATACGACTCACTATAGGGC-3′) (N = 25% each of A, C, G, and T; R = 50% each of A and G).

Construction of a tetR derivative of pLB85. In order to construct a Tet^r integration vector, the plasmid pLB85 (6) was digested with BamHI to remove the erythromycin resistance gene and blunted with Klenow polymerase. Subsequently the vector was ligated to a 1.8-kb SacI-XbaI DNA fragment blunted with Klenow polymerase, which encoded the tet gene from pG⁺host8 (29). The resulting vector, pCS574, which harbors two tet genes in tandem, was subsequently demonstrated to be fully capable of integration into the TP901-1 attachment site of MG1363.

Construction of strains with site-specific integration of an additional gapA or gapB allele on the chromosome. A fragment containing either the gapA or gapB gene of L. lactis MG1363 in which the native gapA-gapB promoter has been replaced by a range of synthetic constitutive promoters was obtained by PCR with primers CpgapA and gapAback (gapA) or CpgapB and gapBback (gapB), using the method described by Solem and Jensen (36). The resulting PCR fragments carrying gapB were digested with SpeI and NsiI and ligated to the plasmid vector pCS574 digested with XbaI and PstI, which are compatible with SpeI and NsiI, respectively. The PCR fragments carrying gapA were digested with SpeI and SalI and ligated to pCS574 digested with XbaI and SalI. The resulting ligation mixtures were introduced into L. lactis LB436, a derivative of MG1363 containing a helper plasmid with a gene coding for the phage TP901-1 integrase, which allows pLB85-type plasmids to site-specifically integrate on the chromosomal attachment site for TP901-1, attB (6). The cells were allowed to regenerate for 2 h and then plated on GM17 plates supplemented with 5 μg of tetracycline per ml and 100 μg of X-Gluc per ml. This resulted in libraries of L. lactis clones with different levels of overexpression of the gapA or gapB gene.

Construction of strains with the native gapB promoter replaced by synthetic promoters in the chromosomal gapB locus. A PCR fragment was amplified with primers CpgapB and gapB′ back, which amplify a truncated version of gapB transcribed from a library of randomized synthetic promoters. Primer gapB′back contains a KpnI restriction site that allowed for cloning of the truncated gapB product in the E. coli vector pRC1 (27), which cannot replicate in L. lactis. The resulting plasmid library was subsequently introduced into L. lactis MG1363 and plated on GM17 supplemented with 5 μg of erythromycin per ml. Chromosomal DNAs from a selection of clones were subsequently used as templates to screen for correct integration at the gapB locus by PCR (primers T7 and gapBback), and in the majority of clones the plasmids had integrated at the gapB locus.

Introduction of ATPase activity. A plasmid harboring the atpA, atpG, and atpD genes, encoding F₁-ATPase in L. lactis (24), was introduced into L. lactis CS811 by standard transformation procedures.

Measurement of GAPDH activity. The GAPDH activities were measured in cell extracts obtained by sonication. Fifty milliliters of cell culture at an OD₄₅₀ of approximately 0.7 was placed on ice and subsequently harvested by centrifugation (7000 rpm in a Sorvall centrifuge with an SLA1500 rotor) for 10 min. The cells were washed twice by using ice-cold sonication buffer (100 mM triethanolamine [pH 7.5], 2 mM dl-dithiothreitol) and then resuspended in 600 μl of the same buffer. The cell suspension was sonicated three times for 45 s each with intervals of 30 s. The preparation was kept on ice during the sonication. Following sonication, the cell debris and intact cells were removed by centrifugation (10 min at 20,000 × g) in a 4°C centrifuge. As a measure of the degree of cell disruption, the OD₂₈₀ was used. The measurement of GAPDH activity, based on the conversion of NAD⁺ to NADH, was by monitoring the change of absorbance at 340 nm over time at 30°C on a Zeiss M500 spectrophotometer. The assay mixture was composed of 100 mM triethanolamine, 2 mM dithiothreitol, 5 mM natriumarsenate, 2 mM NAD⁺, and 2 mM dl-glyceraldehyde-3-phosphate. The measured reaction rates were related to the OD₂₈₀ of the extract for the purpose of determining relative activities. The specific GAPDH activity was determined as units per milligram of protein, where 1 U is defined as the amount of enzyme producing 1 μmol of NADH per min.

Quantification of glucose and end fermentation products by HPLC. Quantification of glucose, pyruvate, lactate, formate, acetoin, acetate, and ethanol by HPLC was performed as described previously (1).

Flux determination in resting cells. Cultures were grown to an OD₄₅₀ of approximately 0.9, and the cells were placed on ice. When chilled, they were harvested by centrifugation (5,000 rpm in a Sorvall SLA1500 centrifuge for 10 min), washed once with 100 ml of ice-cold buffer (SA without vitamins or amino acids), and resuspended in buffer at 30°C with 2 g of glucose per liter. Two-milliliter samples of the cell suspension were filtered for determination of lactate and glucose by HPLC. The observed glycolytic flux remained constant for approximately 80 min after resuspension.

Curve fitting and control coefficients. To estimate the control of GAPDH over the glycolytic flux (J) for the entire range of GAPDH activity (a_GAPDH), we first fitted the experimental data points to the function J(a_GAPDH) = 6.86 × exp(−0.12 × a_GAPDH) − 22.46 × exp(−5.41 × a_GAPDH) + 17.35, using the software program GRAFIT (Erithacus Software Ltd., Harley Surrey, United Kingdom) (20). The control coefficient was then calculated from the equation C_GAPDH^J = [dJ(a_GAPDH)/J(a_GAPDH)]/[d(a_GAPDH)/a_GAPDH)] for the entire range of a_GAPDH. The growth rate data were fitted to the function μ(a_GAPDH) = −0.0094 + 0.778 × [1 − exp(−6.45 × a_GAPDH)]. To estimate the control of GAPDH over the glycolytic flux (J) in resting cells, we first fitted the experimental data points to the function J(a_GAPDH) = −1.84 × (a_GAPDH)² +3.87 × a_GAPDH + 6.81. The control coefficient was then calculated from the equation C_GAPDH^J = [dJ(a_GAPDH)/J(a_GAPDH)]/[d(a_GAPDH)/a_GAPDH)] for the entire range of a_GAPDH. Fits other than second-order polynomial were also used and also gave control coefficients of close to zero at the wild-type level.

Construction of strains with modulated GAPDH activities. A series of strains with altered GAPDH activities were constructed by two different strategies (36): (i) introducing an extra copy of the gapA or gapB allele into the chromosome or (ii) replacing the native promoter of gapB with a synthetic constitutive promoter. The technique to introduce an extra copy of the gapA or gapB allele into the chromosome was used in order to achieve strains with increased activities of GAPDH. Clones with expression of the extra gapB gene were identified as blue colonies with different intensities on plates containing the chromogenic substrate X-Gluc, a substrate for β-glucuronidase encoded by the reporter gene gusA. Several clones were picked for analysis of GAPDH activities. These strains had GAPDH activities ranging from slightly above the wild-type level to twofold the activity of MG1363 (Fig. 1). In order to obtain clones with GAPDH activities both above and below the wild-type level, the native gapB promoter was replaced by a range of synthetic promoters with different strengths. The resulting strains had GAPDH activities ranging from 14% to slightly above wild-type activity in MG1363 (Fig. 1).

FIG. 1. Relative activities of GAPDH for clones with altered expression of gapB. GAPDH activity was measured in extracts from strains with (i) an extra copy of gapB inserted in the L. lactis chromosome in the attachment site for phage TP901-1 and transcribed (more ...)

Growth of L. lactis mutants with altered GAPDH activity. We then analyzed how changes in GAPDH encoded by gapB affected the growth of L. lactis. Growth experiments were performed in defined SA medium supplemented with glucose, using a selection of strains with altered GAPDH activities. Figure 2A shows the growth rate as a function of GAPDH activities from 14 to 210% of the GAPDH activity in MG1363 (1.48 U/mg of protein). It is clear that there is virtually no effect on the specific growth rate when the activity is changed close to the wild-type activity. At highly reduced GAPDH activity, the growth rate decreased significantly, confirming that GAPDH is an essential enzyme for growth of L. lactis. The efficiency with which the cells convert glucose into biomass, i.e., the yield of biomass on glucose, remained constant for the whole range of GAPDH activities except for the lowest GAPDH activity (14%), where the growth yield was decreased by 20% (data not shown).

FIG. 2. Effect of GAPDH activity on growth (A) and glycolytic flux (B) of L. lactis, relative turnover number of GAPDH (C), and flux control by GAPDH (D). The relative turnover number of GAPDH was calculated as the ratio of the relative glycolytic flux and the (more ...)

GAPDH has no control on the glycolytic flux in growing wild-type cells. We then measured the glycolytic flux in steadily growing cells with different activities of GAPDH encoded by gapB. Changes in GAPDH of between 59 and 210% of the wild-type activity had no measurable effect on the glycolytic flux (Fig. 2B). At below 59%, a gradual drop in flux was observed, but even when the activity was reduced to 25% of the wild-type activity, the flux was still 80% of the wild-type flux. These data show that the turnover number of the individual GAPDH molecules can be increased threefold; i.e., there is at least a threefold excess capacity of GAPDH in wild-type cells of MG1363 (Fig. 2C). Based on the curve fit in Fig. 2B, we then calculated the flux control coefficient for GAPDH on the glycolytic flux at all GAPDH levels. Figure 2D shows that GAPDH has virtually zero control over the glycolytic flux at the normal enzyme activity level. With GAPDH activities much below 25%, the decrease in flux became almost proportional to the enzyme activity; i.e., the enzyme gains most of the control over the glycolytic flux with a flux control coefficient increasing towards 1, as expected for an essential enzyme.

Changes in GAPDH activity have no control over the glycolytic flux in resting cells. Poolman and coworkers showed in 1987 that GAPDH had a very high control over the glycolytic flux in resting cells of L. lactis subsp. cremoris Wg2 (33). Those experiments were performed by inhibitor titrations with iodoacetate, which should specifically inhibit GAPDH. Under such conditions, the cells have a significantly reduced glycolytic flux (24), which may strongly affect the extent to which GAPDH influences the glycolytic flux. We therefore measured the flux in a similar setup in which cells containing different GAPDH activities were resuspended in buffer. The flux in these resting cells was indeed much lower than that in steadily growing cells (Fig. 3). However, the response of the glycolytic flux to changes in GAPDH activity was even smaller than the response observed in growing cells, and we conclude that also under these conditions, GAPDH has no control over the glycolytic flux.

FIG. 3. Effect of changes in GAPDH activity on the glycolytic flux in resting cells. See Materials and Methods and Results for details. The curve fitted to the experimental data points for the flux in growing cells is shown for comparison.

Increased GAPDH activity encoded by gapA also has no control over the flux. Since two gap genes (gapA and gapB) have been identified in L. lactis, it was also of interest to study whether the GAPDH activity encoded by gapA had control over the glycolytic flux. The gapA gene was recently found to encode an auxiliary GAPDH which is expressed only under certain stress conditions in L. lactis (41). We analyzed two strains with increased expression of gapA, but the resulting increase in GAPDH activity had no effect on the glycolytic flux in growing cells of MG1363 (Fig. 4).

FIG. 4. Glycolytic flux in strains with increased GAPDH activity resulting from overexpression of gapA. The glycolytic fluxes are plotted as functions of the respective GAPDH activities of strains MG1363, CS267, and CS314.

The GAPDH activities in strains MG1363 and Wg2 differ significantly. A possible explanation for the differences in glycolytic flux control found in our experiments with L. lactis MG1363 and in the experiments of Poolman et al. (33) with L. lactis Wg2 could be related to differences in GAPDH activity between the two strains. Thus, we measured the specific GAPDH activities in extracts from cells of the two strains exponentially growing under identical growth conditions. We found that extracts of L. lactis Wg2 had only 52% of the GAPDH activity measured in extracts of MG1363.

Cells with altered GAPDH remain homolactic. When L. lactis is grown under conditions that lead to a low glycolytic flux, i.e., in sugar-limited chemostat or batch cultures supplemented with a less readily fermentable carbon source, the metabolism usually shifts into the mixed acid fermentation mode, where formate, acetate, and ethanol are formed along with lactate. It was thus of interest to analyze whether the low glycolytic flux resulting from decreased GAPDH activity would also cause such a shift in the fermentation mode. We therefore determined the composition of by-products formed by cultures with altered expression of gapB. However, changes in GAPDH activity had no influence on the pattern of product formation, and the cells remained essentially homolactic at all GAPDH levels, even for the mutants with only 14% of the GAPDH activity of the wild type (Fig. 5) We therefore conclude that GAPDH has no control on the formation of mixed acid fermentation products.

FIG. 5. Relative lactate and formate production of strains with altered GAPDH activity from an anaerobic batch experiment at 30°C in SA medium supplemented with 1.2 g of glucose per liter.

Introduction of ATPase activity in a modulated strain. In a recent publication we showed that increasing the ATP turnover in L. lactis, by expressing an uncoupled ATPase, did not result in an increased glycolytic flux (24), and we suggested that in order to increase the glycolytic flux, it might be necessary to increase the glycolytic capacity and the ATP turnover simultaneously. In order to study the effect of ATPase activity under conditions with a higher GAPDH activity, uncoupled ATPase activity was introduced into strain L. lactis CS811 (which has 2.1 times the wild-type GAPDH activity) by transformation with the plasmid pCPC7::atpAGD (24). Growth experiments were performed in chemically defined SA medium (19), and the specific growth rate, biomass yield, and glycolytic flux (J_g) were determined (Table 2). The introduction of ATPase activity resulted in a lower growth rate and decreased yield of biomass per mole of glucose, as observed previously for MG1363 (24), but no increase in glycolytic flux could be observed. These data show that GAPDH also has no control over the glycolytic flux under conditions in which ATP turnover is enhanced.

TABLE 2. Specific growth rates, yields, and glycolytic fluxes for strains overproducing GAPDH with and without ATPase activity

The activity of GAPDH in L. lactis subsp. cremoris MG1363 was modulated around the wild-type activity and found to have no control over the glycolytic flux, either in growing cells or in cells resuspended in buffer. Only after the GAPDH activity was reduced to below 25% of the normal activity did the enzyme become limiting for growth and glycolysis. Our calculations showed that there is at least a threefold excess capacity of GAPDH in L. lactis MG1363. The effect of reducing GAPDH was found to be smaller in resting cells than in growing cells. This is also in line with our expectations, since the excess capacity of GAPDH should be even higher in resting cells, which have a two- to threefold lower glycolytic flux (24).

Our results are therefore in sharp contrast to the earlier data obtained by Poolman and coworkers, who found that GAPDH has almost full control over the glycolytic flux in L. lactis (33). One explanation could be the choice of strain used in the experiments; for instance, different strains may have different activities of GAPDH, and this could then affect the outcome of the experiments. Another explanation could be the method used. Poolman et al. used the inhibitor titration method with iodoacetate, a specific inhibitor of GAPDH. This method allows for decreasing but not for increasing enzyme activities, which makes the subsequent determination of flux control more difficult. Our result is the opposite of the result by Poolman et al.: we find zero control (= 0.0), whereas they found almost full control (= 0.9). Another explanation could be that the activity was much higher in our strain than in the strain (Wg2) used by Poolman et al. Indeed, our measurement showed that the activity in Wg2 was almost twofold lower than that in MG1363. However, with a twofold change in GAPDH activity, the control by GAPDH in MG1363 would still be lower than 0.15. In addition, the results of Poolman et al. came from nongrowing cells, which usually have a two- to threefold-lower glycolytic flux than growing cells as a consequence of the lower requirements for ATP (24). Under these conditions, we expect the excess capacity of GAPDH to be even higher, which should decrease the flux control by this enzyme further. Similarly, results obtained with slow-growing L. lactis NCDO2118 (9) may also be an underestimate of the flux control exerted by GAPDH in fast-growing cells. An explanation for the different observations could be that iodoacetate inhibits another enzyme(s) in L. lactis besides GAPDH and that this inhibition is even stronger than the inhibition of GAPDH. However, Even and coworkers (9) tested this hypothesis and found that at least two of the candidates, pyruvate kinase and lactate dehydrogenase, were not significantly affected by iodoacetate.

Early experiments have shown that cells with a low glycolytic flux as a consequence of sugar limitation change their pyruvate metabolism to a mixed acid pattern, which then provides the cell with an extra ATP per glucose molecule (12, 38, 39, 40). It was therefore of interest to determine how changes in GAPDH affected the pyruvate metabolism in L. lactis. We found no change in the product formation, which may seem surprising at first. However, Even et al. (9) showed that inhibition of GAPDH activity with iodoacetate caused a shift towards homolactic fermentation in a mixed-acid-producing strain, which is in good agreement with our results, where the cells remain homolactic.

The analysis of the genome sequence of L. lactis IL-1403 (5) has revealed the presence of two putative gap genes, gapA and gapB. Bacillus subtilis also possess two GAPDH-encoding genes (gapA and gapB) (26), which were found to have opposite physiological roles (11), i.e., one enzyme for catabolic purposes and one for anabolic purposes. This does not appear to be the case for the two GAPDH proteins in L. lactis. Rather, it looks like gapA is an auxiliary GAPDH, expressed only under certain stress conditions, and that gapB is the main source of GAPDH activity in L. lactis growing under standard laboratory conditions (41). Our finding that lowering the expression of gapB almost abolishes GAPDH activity in L. lactis would fit this interpretation. We have also constructed mutants which have up to 30% increased GAPDH activity due to overexpression of gapA. In these mutants the glycolytic flux also was unaffected, which demonstrates that our result of zero control by GAPDH over glycolytic flux in L. lactis MG1363 is independent of which of the two gap genes is modulated.

Based on the present knowledge of flux control by the individual glycolytic enzymes over glycolysis in L. lactis, it is reasonable to believe that a simultaneous modulation of several steps might be necessary in order to increase the glycolytic flux. Since an increased glycolytic flux eventually will result in an increased ATP generation at the expense of ADP, it is likely that the increase in flux-controlling glycolytic enzymes must be accompanied by an increased ATP turnover. This would also be in good agreement with previous observations for tryptophan metabolism in yeast (31) and for glycolysis in Trypanosoma brucei (3), where experiments and computer simulations showed that the flux control is indeed shared by several steps in the pathway.

In addition to a concomitant increase of many enzyme activities, it may also be necessary to accurately adjust the individual activities to specific levels in order to maintain proper metabolite concentrations (2, 10) and to avoid a protein burden (35). A few years ago this would have seemed an almost impossible task, but the recent development of the synthetic promoter library technology (21, 36) used in this study has opened the door to such work.

We thank Pierre Renault for providing the sequence of the gapB prior to the publication of the IL-1403 genome sequence.

This work was supported by Chr. Hansen A/S and the Danish Research Agency.