Metabolic flux analysis of a glycerol-overproducing Saccharomyces cerevisiae strain based on GC-MS, LC-MS and NMR-derived 13C-labelling data


  • Editor: Barbara M. Bakker

Correspondence: Wouter A. van Winden, Department of Biotechnology, Delft University of Technology, Julianalaan 67, 2628 BC Delft, the Netherlands. Tel.: +31 15 278 6369; fax: +31 15 278 2355; e-mail:


This study focuses on unravelling the carbon and redox metabolism of a previously developed glycerol-overproducing Saccharomyces cerevisiae strain with deletions in the structural genes encoding triosephosphate isomerase (TPI1), the external mitochondrial NADH dehydrogenases (NDE1 and NDE2) and the respiratory chain-linked glycerol-3-phosphate dehydrogenase (GUT2). Two methods were used for analysis of metabolic fluxes: metabolite balancing and 13C-labelling-based metabolic flux analysis. The isotopic enrichment of intracellular primary metabolites was measured both directly (liquid chromatography-MS) and indirectly through proteinogenic amino acids (nuclear magnetic resonance and gas chromatography-MS). Because flux sensitivity around several important metabolic nodes proved to be dependent on the applied technique, the combination of the three 13C quantification techniques generated the most accurate overall flux pattern. When combined, the measured conversion rates and 13C-labelling data provided evidence that a combination of assimilatory metabolism and pentose phosphate pathway activity diverted some of the carbon away from glycerol formation. Metabolite balancing indicated that this results in excess cytosolic NADH, suggesting the presence of a cytosolic NADH sink in addition to those that were deleted. The exchange flux of four-carbon dicarboxylic acids across the mitochondrial membrane, as measured by the 13C-labelling data, supports a possible role of a malate/aspartate or malate/oxaloacetate redox shuttle in the transfer of these redox equivalents from the cytosol to the mitochondrial matrix.


Methods to enhance the production of glycerol from sugar using Saccharomyces cerevisiae have been developed since the start of the twentieth century. Traditionally, glycerol was produced in oxygen-limited fermentations of S. cerevisiae using the Neuberg process. In this process, bisulphite was used to trap acetaldehyde, the electron acceptor of alcoholic fermentation, thus yielding excess reduction power for the reduction of dihydroxyacetone phosphate (DHAP) to glycerol (Neuberg & Reinfurth, 1919a, b; Bakker et al., 2001). A more recent attempt, using metabolic engineering to improve glycerol production in S. cerevisiae, successfully redirected carbon flow by deletion of the structural gene (TPI1) encoding triosephosphate isomerase (Compagno et al., 1996, 2001). The strategy behind this deletion was to force equimolar formation of glyceraldehyde-3-phosphate (GAP) and DHAP. Although the yield obtained (0.90 mol glycerol per mol glucose) was indeed close to the maximum theoretical yield of 1.0 for a nongrowing tpi1ΔS. cerevisiae strain, the mutant did not grow on glucose as the sole carbon source. This was attributed to reactions that compete with glycerol-3-phosphate dehydrogenase (G3PDH) for the NADH that is generated in the lower branch of glycolysis, causing toxic accumulation of DHAP (Overkamp et al., 2002).

Indeed, additional deletion of the structural genes for the external mitochondrial NADH dehydrogenases (NDE1 and NDE2) and the respiratory chain-linked glycerol-3-phosphate dehydrogenase (GUT2) eliminated this growth defect (Overkamp et al., 2002). Quantitative analysis of the tpi1Δnde1,2Δgut2ΔS. cerevisiae strain displayed a difference in the glycerol yield on glucose between aerobic batch cultures (0.99 mol glycerol per mol glucose) and aerobic glucose-limited chemostat cultures (0.83 mol mol−1), which was attributed to the increased assimilatory metabolism in the latter cultivations (Overkamp et al., 2002). Not only do the biosynthesis of cell-wall constituents, storage carbohydrates, amino acids and fatty acids result in a decreased flux from glucose to DHAP, but so too does the generation of NADPH via the pentose phosphate pathway (PPP), thereby decreasing the glycerol production rate. Alternatively, the lower glycerol yield can also be explained by the breakdown of DHAP via the methylglyoxal pathway. This bypass converts DHAP to pyruvate and involves a direct transfer of reduction equivalents to the electron transport chain via flavine adenine dinucleotide (FAD)-linked, mitochondrial d-lactate dehydrogenase (Rojo et al., 1998; Grandier-Vazeille et al., 2001; Martins et al., 2001).

Both the PPP and the methylglyoxal pathway prevent equimolar fluxes through G3PDH and glyceraldehyde-3-phosphate dehydrogenase (GAPDH), causing a glycerol yield on glucose lower than 1.0. The PPP results in a higher flux through GAPDH, while the methylglyoxal pathway lowers the flux through G3PDH. Consequently, activity of either pathway would result in excess cytosolic NADH, requiring an alternative reoxidation mechanism, such as redox shuttles (Bakker et al., 2001). NADH shuttles enable the transfer of NADH equivalents from the cytosol to the mitochondrial matrix, where they are oxidized by the mitochondrial internal NADH dehydrogenase (Bakker et al., 2001). Notably, the difference in glycerol yield between batch and chemostat cultivation suggests the involvement of mechanisms sensitive to high glucose concentrations.

Metabolite balancing and 13C-labelling-based metabolic flux analysis (MFA) are established tools to study the distribution of metabolic fluxes. Metabolite balancing is based on the principle of mass conservation of the intracellular metabolites within a defined stoichiometric network operating under steady-state conditions (Wiechert, 2002). Intracellular fluxes can be calculated straightforwardly by combining the stoichiometric network with measured extracellular conversion rates. A downside of this method is that it fails to identify parallel metabolic pathways, metabolic cycles and bidirectional reaction steps. Furthermore, it often requires the inclusion of conserved moiety balances (i.e. ATP, NADH, NADPH), which are generally based on incomplete stoichiometric information (Bonarius et al., 1997; Wiechert, 2001). These problems can be (partly) overcome using 13C as a tracer. The 13C-labelling-based MFA method relies on feeding 13C-labelled substrate to a culture of the studied microorganism and measuring the 13C-label distribution in various intracellular compounds under isotopic (pseudo) steady state. The isotopic enrichment of the intracellular metabolites is measured directly using liquid chromatography (LC-MS) (van Winden et al., 2005) or indirectly by measuring the 13C-labelling of proteinogenic amino acids using nuclear magentic resonance (NMR) (Maaheimo et al., 2001) or gas chromatography (GC-MS) (Gombert et al., 2001; Fischer & Sauer, 2003b). The flux patterns within a metabolic network model are subsequently calculated by iteratively fitting simulated 13C-label distributions for a chosen set of metabolic fluxes to the measured 13C-label distributions (Wiechert, 2001). To date, all published 13C-labelling-based MFA studies have relied on one (Gombert et al., 2001; Maaheimo et al., 2001; Fischer & Sauer, 2003b; van Winden et al., 2005) or at most two (Yang et al., 2002; Fischer & Sauer, 2003a) independent measurement methods for determining the 13C-label distribution within the cell.

This study aims at mapping the distribution of carbon fluxes and reduction equivalents within a glycerol-overproducing tpi1Δnde1,2Δgut2ΔS. cerevisiae strain (CEN.PK 546-AHG) using both metabolite balancing and 13C-labelling-based MFA. 13C-labelling patterns are measured using three independent measurement techniques: NMR, GC-MS and LC-MS. In addition, the sensitivity of each of the three 13C-measurement techniques is evaluated for important metabolic nodes.

Materials and methods

Strain use and maintenance

All experiments described in this study were performed with an evolutionarily engineered glycerol-overproducing tpi1Δnde1,2Δgut2ΔS. cerevisiae strain (CEN.PK 546-AHG) (Overkamp et al., 2002). Stock cultures were grown to stationary phase in shake flask cultures on mineral medium supplemented with vitamins and trace elements (Verduyn et al., 1992), which was set to pH 6.0 using potassium hydroxide and contained 2% (w/v) glucose. After addition of sterile glycerol (20% v/v), 2-mL aliquots were stored at −80°C and subsequently used for inoculating precultures for chemostat experiments.

Chemostat cultivation and 13C-labelling

Two duplicate aerobic, glucose-limited chemostat cultivations were carried out in a 1-L laboratory fermentor (Applikon) that was temperature controlled at 30°C. The working volume was kept at 0.6 L by means of an overflow system. The pH was kept constant at 5.0 by an Applikon ADI 1030 biocontroller, by the addition of 2 M potassium hydroxide. Defined aerobic, continuous-culture media were prepared as described previously (Verduyn et al., 1992), containing 7.5 g L−1 glucose as the sole carbon source. The fermentor was sparged with air (0.33 vvm) and stirred at 600 r.p.m. An overpressure of 0.1 bar was applied to the system to facilitate the rapid sampling of broth. The dissolved oxygen tension was continuously monitored with an oxygen electrode (Ingold, model 34 100 3002, Mettler) and never dropped below 50%. Offgas was cooled in a condenser (2°C) and analysed for oxygen and carbon dioxide concentrations with a VC Prima600 mass spectrometer (Thermo Electron Corp.). The exact gas flow rate was determined using a Saga digital flow meter (Ion Science). The dilution rate was set to 0.05 h−1. Steady-state cultures did not exhibit detectable metabolic oscillations. Chemostat cultures were routinely checked for contaminations by phase-contrast microscopy.

A metabolic steady state was defined as the situation in which at least four volume changes had passed since the last change in culture parameters and in which the biomass concentration, as well as all other specific production or consumption rates, had remained constant (<2% variation) for 24 h. Upon reaching a metabolic steady state (at 85 h), the original medium containing 7.5 g L−1 naturally labelled glucose was replaced by chemically identical medium, but with 10% of the glucose replaced by [U-13C6]d-glucose (Isotec) and 90% replaced by [1-13C1] d-glucose (Campro Scientific). Grotkjaer et al. (2004) developed a detailed dynamic model describing carbon atom transitions in the central metabolism of S. cerevisiae to study the rate at which 13C is incorporated into biomass. The model showed that for the first three residence times the labelling of proteinogenic amino acids deviated significantly from the commonly assumed first-order washout kinetics, as a result of transamination reactions and protein turnover. To ignore the impact of these reactions and thus ensure full isotopic steady state for all biomass components, the chemostat was run for four residence times on the 13C-labelled medium before sampling.

Rate determination

Biomass dry weight determination and protein determination were performed as described previously in samples that were taken from the effluent and kept on ice (Geertman et al., 2006). Extracellular samples were acquired by rapidly sampling 2 mL of broth into a syringe containing cold steel beads (−18°C) as described previously (Mashego et al., 2003). Culture supernatants and culture medium were analysed for glucose, glycerol and ethanol by HPLC system mounted with a dual-wavelength absorbance detector (Waters 2487) and a refractive index detector (Waters 2410). An Aminex HPX-87 H (Bio-Rad) column was used and eluted with dilute sulphuric acid (5 mM; 0.6 mL min−1) at 60°C.

LC-MS: metabolite sampling, extraction and analysis

Samples (1 mL, ≈1.7 mg biomass dry weight) for LC-MS were taken using the rapid sampling set-up described by Lange et al. (2001). Immediate quenching of the metabolism, separation of the cells from the extracellular liquid and metabolite extraction were performed as described by Kleijn et al. (2006). The supernatant obtained was stored at −80°C prior to LC-MS analysis.

The mass isotopomer distributions of the intracellular metabolites were measured as described by van Winden et al. (2005). Briefly, metabolites were first separated by high-performance anion exchange chromatography (Waters) followed by MS analysis with a Quattro-LC triple quadrupole mass spectrometer (Mircomass Ltd) equipped with an electrospray ionization interface. LC-MS was used to analyse glucose-6-phosphate (G6P), fructose-6-phosphate (F6P), 6-phosphogluconate (6PG), mannose-6-phosphate (M6P), fructose-1, 6-bisphosphate (FBP), phosphoenol pyruvate (PEP), the combined pool of 2- and 3-phosphoglycerate (2/3PG), the combined pool of xylulose-5-phosphate, ribose-5-phosphate and ribulose-5-phosphate (P5P), and sedoheptulose-7-phosphate (S7P).

Mass fractions of CO2 in the offgas were determined using a VC Prima600 mass spectrometer. The mass spectrometer was calibrated using two 13CO2 calibration gases containing 2% CO2 and 98% N2 with a 13CO2/12CO2 ratio of 1:2 and 1:4. The calibration gases were prepared by, respectively, dissolving 1.0 g and 1.2 g of unlabelled K2CO3 with 0.5 g and 0.3 g of 13C-labelled K2CO3 (Cambridge Isotope Laboratories) in 30 mL of water. Through addition of 10 mL of 4 M HCl, the solution was brought to pH 1 and subsequently heated to 50°C. Then, 0.24 L of CO2 gas was captured in a 10-L gas bag (Alltech) and diluted with N2 gas.

GC-MS: biomass sampling and analysis

Biomass samples (115 mL, ≈200 mg biomass dry weight) for GC-MS analysis were taken from the chemostat, spun down directly for 5 min at 2000 g, washed with 0.9% NaCl solution and lyophilized. Aliquots (5 mg) of the lyophilized material were hydrolyzed in 1 mL 6 M HCl, dried and derivatized as described by Perrenoud & Sauer (2005). GC-MS analysis and raw data analysis were performed as described by Fischer & Sauer (2003b). Mass isotopomer distributions were measured for the following proteinogenic amino acids (fragments): alanine(all), (−1); aspartate(all), (−1), (1+2); glutamate(all), (−1); glycine(all), (−1); isoleucine(all), (−1); leucine(all), (−1); lysine(all); (−1), phenylalanine(all); (−1), (1+2); proline(all), (−1); serine(all), (−1), (1+2); threonine(all), (−1); tyrosine(all), (−1), (1+2); valine(all), (−1), (1+2); where (all) denotes the complete amino acid, (−1) is the amino acid minus the carboxyl group and (1+2) is the amino acid fragment consisting of only the first two carbon atoms (position 1 being the carboxyl group). The mass isotopomer fractions obtained were corrected for the occurrence of natural isotopes of N, H, O, S and Si in both the amino acid and the derivatizing agent and for the occurrence of natural isotopes of C in the derivatizing agent only. The applied correction was the inverse of the procedure proposed by van Winden et al. (2002).

2D [13C,1H] COSY NMR: biomass sampling and analysis

Together with the sample taken for GC-MS analysis, 385 mL of culture broth (≈665 mg biomass) was spun down directly for 5 min at 2000 g, washed with 0.9% NaCl solution and demineralized water and stored at −80°C. Before two-dimensional (2D) [13C,1H] correlation spectroscopy (COSY) NMR analysis the biomass was lyophilized and hydrolyzed. Subsequently, the amino acids were purified and prepared for measurement as described by van Winden et al. (2003). The NMR spectra were recorded and analysed by means of spectral fitting software as described by van Winden et al. (2001b). The resulting data are relative intensities of fine structures observed in the multiplets of the proteinogenic amino acids phenylalanine-α and -β, glycine-α, histidine-α and -β, serine-α and -β, tyrosine-α and -β, alanine-α and -β, aspartate-α and -β, glutamate-α, -β and -γ, isoleucine-β, -γ1 and -δ, lysine-β and -γ, leucine-α, -β, -γ, -δ1 and -δ2, methionine-α, proline-α and -δ, arginine-β and -δ, threonine-α, -β and -γ, valine-α and -γ1 and the storage carbohydrate trehalose-C1. Relative intensities for the symmetrical component glycerol-C1/C3 were measured in filtrate samples. Note that the nomenclature for numbering the carbon atoms of amino acids used is as follows: C is the C-terminus of the amino acid (carboxyl group); α is the carbon atom next to C; β is the carbon atom next to α, etc. Standard three-letter abbreviations were used for the amino acids.

Metabolite balancing

A previously developed compartmentalized stoichiometric model of S. cerevisiae (Lange, 2002) was used for metabolite balancing. Calculations were carried out with the software package MNAv3.0 (SpadIT). To describe the genetically engineered tpi1Δnde1,2Δgut2Δ strain, the reactions carried out by triosephosphate isomerase, the external NADH dehydrogenases and FAD-dependent glycerol-3-phosphate dehydrogenase were removed. Subsequently, eight versions of the stoichiometric model were constructed using different assumptions with respect to the existence of methylglyoxal bypass, the existence of a mitochondrial NADH shuttle and the cofactor specificity of acetaldehyde dehydrogenase (see Results). The measured biomass-specific conversion rates of glucose, glycerol, O2 and CO2 were used as input for metabolite balancing. These rates were calculated from the average steady-state values of the measured flows and concentrations. The flux balancing procedure was weighted using the full variance matrix associated with the calculated conversion rates.

13 C-labelling-based MFA

The metabolic network models described above formed the basis of the model used for the 13C-labelling-based MFA. The following adaptations were made: (1) mass balances for reduction equivalents were removed; (2) reaction reversibilities were included; (3) the conventional reactions of the nonoxidative branch of the PPP were replaced by metabolite-specific, reversible, C2 and C3 fragment-producing and -consuming half-reactions as proposed by Kleijn et al. (2005); (4) glycine synthesis via the enzyme threonine aldolase was included; (5) the glycine decarboxylase complex (GDC) catalysing the reversible cleavage of glycine into CO2 and 5,10-CH2-H4folate (E-C1) was included (Piper et al., 2000); (6) the transport of acetyl-coenzyme A (CoA) from the cytosol to the mitochondrion was included; (7) the gluconeogenic enzyme PEP carboxykinase catalysing the conversion of oxaloacetate into PEP was included; (8) malic enzyme catalysing the mitochondrial decarboxylation of malate into pyruvate was included; and (9) scrambling reactions for the symmetrical molecules glycerol and succinate (which forms part of the oxaloacetate pool) were included. Metabolite pools in the network that had only one influx were removed, while all metabolite pools in isotopic equilibrium due to fast exchange reactions were lumped into one single pool (van Winden et al., 2001a). Reversible reactions were modelled as separate forward and backward reactions and are referred to as net and exchange fluxes, where:


All major CO2-producing and CO2-consuming reactions were incorporated in the metabolic network model, making it possible to fit the measured mass fractions of CO2. To estimate the isotopic enrichment of the CO2 pool correctly, the inflow of naturally labelled CO2 as a result of aeration was also included in the metabolic network model. No distinction was made between CO2 produced and consumed in different compartments. A schematic representation of the metabolic network model is shown below in Fig. 3.

Figure 3.

 Intracellular metabolic fluxes in a tpi1Δnde1,2Δgut2ΔSaccharomyces cerevisiae strain determined by independently fitting the mass isotopomer fractions of the intracellular metabolites measured by LC-MS (top values, bold), by independently fitting the mass isotopomer fractions of the proteinogenic amino acids measured by GC-MS (top-middle values, italic), by independently fitting the relative intensities of the proteinogenic amino acids measured by NMR (bottom-middle values, bold italic) and by fitting the combined 13C-labelling dataset (NMR, LC-MS and GC-MS) (bottom values, normal). Fluxes are normalized for the glucose uptake rate. Values outside parentheses denote the net fluxes, while values in parentheses represent the exchange fluxes. Solid arrowheads denote the direction of the net flux. Abbreviations: FBP, fructose-1,6-bisphospate; P5P, pentose-5-phopshate; S7P, sedoheptulose-7-phosphate; E4P, erythrose-4-phosphate; E-C2, glycolaldehyde moiety covalently bound to the thiamine pyrophosphate/transketolase complex; E-C3, dihydroxyacetone moiety covalently bound to the enzyme transaldolase; ser, serine; gly, glycine; thr, threonine; other abbreviations as in Fig. 1.

The flux fitting procedure employed has been described in detail by van Winden et al. (2005). SDs for the measured 13C-labelling data were fixed at 1%. The variance-weighted sum of squared residuals (SSres) between the simulated and measured data was used as the target function in a minimization procedure based on sequential quadratic programming that was implemented in Matlab (The MathWorks Inc.).

Results and discussion

Macroscopic data

The biomass-specific consumption and production rates of the tpi1Δnde1,2Δgut2Δ strain are given in Table 1. As expected for a tpi1Δ strain the biomass yield [0.24±0.01 g biomass (g glucose)−1] on glucose was significantly lower than for a wild-type under the same conditions. The glycerol yield on glucose during these cultivations was 0.80±0.01 mol mol−1. In line with this high glycerol production, the respiratory quotient (RQ) was 1.24 mol CO2 (mol O2)−1. The carbon and the degree of reduction (γ) balances ended at 99.9±0.1% and 101.4±0.7%, respectively, indicating that apart from biomass, CO2 and glycerol, no significant amounts of other compounds were produced. In addition, black box balancing of the conversion rates (Table 1) and gross error detection of the measured conversion rates according to van der Heijden (1991) showed that the measurement set was statistically acceptable (P=0.21). The P-value represents the probability that the discrepancy between the balanced and measured conversion rates can be explained by measurement error. Typically, P-values of 0.05 or lower signify the presence of other errors (e.g. systematic errors, erroneous model assumptions). Measured biomass-specific consumption and production rates were, subsequently, used for metabolite balancing and 13C-labelling-based MFA.

Table 1.   Measured and reconciled conversion rates for an evolutionarily engineered glycerol-overproducing tpi1Δnde1,2Δgut2ΔSaccharomyces cerevisiae strain grown in duplicate aerobic, glucose-limited chemostat cultures at a dilution rate of 0.05 h−1
Conversion ratesUnitMeasured
Glucose consumptionmmol (CmolX h)−130.0 ± 1.030.3
Oxygen consumptionmmol (CmolX h)−147.7 ± 2.448.5
Carbon dioxide productionmmol (CmolX h)−160.3 ± 3.062.4
Glycerol productionmmol (CmolX h)−123.9 ± 0.823.8
Ethanol productionmmol (CmolX h)−1< 0.030.0

Metabolite balancing

To gain more insight into the primary metabolism of the adapted tpi1Δnde1,2Δgut2Δ strain, eight different variants of the stoichiometric model were constructed and used for metabolite balancing (see ‘Materials and methods’, and Table 2). The basic stoichiometric model (model 1) contained no methylglyoxal bypass, no mitochondrial NADH-shuttle and an NADP+-dependent acetaldehyde (Aald) dehydrogenase. It should be stressed here that metabolite balancing requires the inclusion of reduction-equivalent balances [NAD(H) and NADP(H)] in order to determine all fluxes. Statistical testing of the balanced conversion rates for this model gave a high χ2-test value and, subsequently, a low P-value (P<0.001, Table 2), indicating that the redundant measured conversion rates could not be reconciled by the stoichiometric model.

Table 2.   Metabolic fluxes for several important metabolic nodes in a tpi1Δnde1,2Δgut2ΔSaccharomyces cerevisiae strain as calculated via metabolite balancing for eight different stoichiometric models.
Specificity acetaldehyde
PP pathway
Flux through
NADH shuttle
Flux through
methylglyoxal bypass
  • The eight stoichiometric models differed with respect to the existence of a methylglyoxal bypass, the existence of a putative NADH shuttle and the cofactor specificity of the enzyme acetaldehyde dehydrogenase. Fluxes are normalized for the glucose uptake rate, which was set to 100.

  • *

    The P-value represents the probability that the discrepancy between the balanced and measured conversion rates can be explained by measurement error alone. Typically, P-values <0.05 signify an erroneous stoichiometric model.

  • PP pathway, pentose phosphate pathway.


The statistical rejection of model 1 is in part caused by an overestimation of the glycerol production rate [0.84 mol (mol glucose)−1, Fig. 1] compared with the measured rate of 0.80 mol (mol glucose)−1. In stoichiometric model 1 the glucose entering the cell can be metabolized towards GAP via two pathways: glycolysis and the PPP. As the PPP is the main pathway for cytosolic NADPH formation, the flux through the PPP is constrained by the mass balance for NADPH and thus depends directly on the biosynthetic demand for NADPH. In model 1, this maximum biosynthetic NADPH demand fixed the oxidative PPP flux at 0.15 mol (mol glucose)−1, which resulted in a glycerol production of 0.84 mol (mol glucose)−1 (see Fig. 1). A second reason for the statistical rejection of model 1 was the estimation of an ethanol production rate of 0.11 mol (mol glucose)−1, whereas the measured ethanol production rates were below detection level [<0.001 mol (mol glucose)−1]. Ethanol was formed in model 1 to remove the surplus of cytosolic NADH formed in anabolism [0.07 mol (mol glucose)−1] and from the GAP synthesized in the PPP [0.04 mol (mol glucose)−1]. Due to the exclusion from the model of the external mitochondrial NADH dehydrogenases and the glycerol-3-phosphate dehydrogenase, ethanol formation was the only remaining pathway to reoxidize the excess cytosolic NADH.

Figure 1.

 Intracellular metabolic fluxes in a tpi1Δnde1,2Δgut2ΔSaccharomyces cerevisiae strain quantified via metabolite balancing. The strain was grown in an aerobic, glucose-limited chemostat culture at D=0.05 h−1. Fluxes were determined by combining the measured conversion rates with a standard stoichiometric model (model 1, bold) and an extended stoichiometric model containing a putative NADH-shuttle and the methylglyoxal bypass (model 8, italic). Apart from the carbon flow through the primary metabolism, the NADH production or consumption is specified for each reaction. Fluxes are normalized for the glucose uptake rate. Abbreviations: glc, glucose; G6P, glucose-6-phosphate; F6P, fructose-6-phosphate; PPP, pentose phosphate pathway; DHAP, dihydroxyacetone phosphate; GAP, glyceraldehyde-3-phosphate; goh, glycerol; PEP, phosphoenol pyruvate; pyr, pyruvate; aald, acetaldehyde; etoh, ethanol; AcCoA, acetyl-CoA; OAA, oxaloacetate; citr, citrate; akg, α-ketogluterate.

To remedy the overestimation of the glycerol and ethanol production rates in model 1, additional reactions were included in the stoichiometric model. The first discrepancy, the overestimation of glycerol production, was addressed by incorporation of a methylglyoxal bypass (model 2). This reaction created an alternative mode of DHAP dissimilation. Addition of the methylglyoxal bypass increased the statistical acceptability of the model (lower χ2-test value), but still led to model rejection (P<0.001). In fact, a negative flux was fitted for the methylglyoxal bypass, signifying the conversion of pyruvate into DHAP. The reverse flux actually mimics the role of the enzyme triosephosphate isomerase by converting GAP into DHAP and thus indirectly removes part of the surplus NADH in the cytosol at the cost of an increased glycerol production rate. Note that in vivo the methylglyoxal bypass is an irreversible pathway, thereby rendering the proposed carbon flow infeasible. These results indicate that the cytosolic redox balance has a greater impact on the statistical acceptability of the model than the overestimation of the glycerol production rate.

The apparent cytosolic redox stress was addressed in model 3 by the introduction of an alternative pathway to reoxidize cytosolic reduction equivalents, in the form of a putative NADH shuttle between the cytosol and the mitochondrion. The putative NADH shuttle allowed for the net oxidation of cytosolic NADH and reduction of mitochondrial NAD+. It significantly increased the statistical acceptability of the model (χ2-test value=24.8) and abolished ethanol formation. The equivalent of 0.12 mol NADH (mol glucose)−1 was transported into the mitochondrial matrix. Nevertheless, the model was still statistically rejected (P=0.002).

As the separate inclusion of the methylglyoxal bypass and the putative redox shuttle had a positive but insufficient effect on the acceptability of the model, both were incorporated in stoichiometric model 4. This combination yielded statistically acceptable flux patterns when balancing the variance-weighted conversion rates (Fig. 1). The flux through the methylglyoxal bypass and the putative NADH shuttle were estimated at 0.045 and 0.161 mol (mol glucose)−1, respectively. These results suggest that the adapted tpi1Δnde1,2Δgut2Δ strain possesses a significant cytosolic NADH sink, most likely translocating excess NADH from the cytosol to the mitochondria. The estimated flux through the methylglyoxal bypass was 15-fold higher than the value for wild-type S. cerevisiae reported by Martins et al. (2001), who measured the methylglyoxal bypass flux to be 0.3% of the total glycolytic flux independent of its magnitude. The PPP split-ratio was estimated at 0.16 mol (mol glucose)−1. This value corresponds well with the value reported by Gombert et al. (2001) [0.42 mol (mol glucose)−1] for wild-type S. cerevisiae, which gives a PPP split-ratio of 0.20 mol (mol glucose)−1 for the tpi1Δnde1,2Δgut2Δ strain when corrected for its higher biomass yield [Yxs=0.52 g (g glucose)−1].

A higher flux through the PPP directs part of the carbon around FBP, resulting in reduced DHAP formation and thus less glycerol production. However, as described above the flux through the PPP is constrained by the NADPH balance. Therefore, in models 5–8 the cofactor specificity of cytosolic Aald dehydrogenase, required for the formation of cytosolic acetyl-CoA, was changed from NADP+ to NAD+. Even though most literature sources point to an NADP+ dependency of the cytosolic Aald dehydrogenase (ALD6) (Meaden et al., 1997; Eglinton et al., 2002), S. cerevisiae also contains two genes (ALD2 and ALD3) encoding for stress-induced, cytosolic NAD+-dependent Aald dehydrogenases (Navarro-Avino et al., 1999). As a result of the change in cofactor specificity, estimated PPP split-ratios for models 5–8 were on average 0.05 mol (mol glucose)−1 higher than those of models 1–4. The increased PPP split-ratio also lowered the methylglyoxal bypass flux (model 8), as less surplus carbon had to be withdrawn from the glycerol pathway. In contrast to the lower methylglyoxal bypass flux, the flux through the putative NADH shuttle increased as a result of the additional NADH produced by Aald dehydrogenase. Models 4 and 8 both had the same P-value (P<0.167), indicating that the cofactor specificity of the Aald dehydrogenase did not affect the statistical acceptability.

The above stoichiometric models indicate the presence of pathways that divert carbon away from glycerol formation (e.g. the PPP and the methylglyoxal bypass) and an additional sink for cytosolic NADH in the investigated tpi1Δnde1,2Δgut2ΔS. cerevisiae strain. In a next step, 13C-labelling experiments were performed to validate the presence of these pathways and to gain further insight into the actual mechanism of the cytosolic NADH sink.

Metabolic flux analysis using 13C-labelling

LC-MS, GC-MS and NMR datasets

The tpi1Δnde1,2Δgut2Δ strain was grown on a mixture of 10% [U-13C]glucose and 90% [1-13C]glucose. The inflow of labelled material was followed by online measurements of the 13CO2/12CO2 ratio in the offgas of the chemostat (Fig. 2). CO2 is primarily produced by the catabolic pathways of the cell [e.g. tricarboxylic acid (TCA) cycle]. The sharp increase in the 13CO2 concentration seen directly after switching to 13C-labelled glucose (at t=85 h) therefore indicates a rapid distribution of 13C-label throughout the primary metabolism. As CO2 is also produced during the synthesis and turnover of macromolecular biomass components (e.g. storage carbohydrates, proteins, and lipids) it takes several residence times before a constant ratio is reached (>3 residence times). Figure 2 shows that the four residence times of 13C-labelling applied in this study were sufficient to ensure full isotopic steady state for all biomass components (see also ‘Materials and methods’).

Figure 2.

 Online measurement of the 13CO2/12CO2 ratio in the offgas of a aerobic glucose-limited chemostat culture of a tpi1Δnde1,2Δgut2ΔSaccharomyces cerevisiae strain. After 4.25 residence times the naturally labelled feed was replaced by a chemically identical feed, but enriched in 13C. Samples to measure the isotopic enrichment of the intracellular components were harvested four residence times later.

The 13C-labelling patterns measured with LC-MS, GC-MS and NMR are given in supplementary Table S1. The LC-MS-derived mass isotopomer fractions of the hexose monophosphates are in accordance with the applied substrate labelling (10% [U-13C6] and 90% [1-13C]glucose). The observed decrease in the m+1, m+2 and m+6 mass fractions in favour of the unlabelled m+0 fraction is a result of the oxidative branch of the PPP, in which the first carbon atom (carboxyl group) of 6-phosphogluconate is split off to form CO2.

Analysis of the carbon transitions in the PPP and the upper glycolysis shows that the fragment consisting of the last three carbon atoms of the upper glycolytic and PPP metabolites remains intact during conversion and thus always ends up as GAP by the action of either FBP aldolase or transaldolase. In this specific strain the carbon transition is further constrained by the absence of the enzyme triosephosphate isomerase, preventing the exchange of label between DHAP and GAP. Consequently, the 13C-labelling of GAP is identical to the 13C-labelling of the bottom three carbon atoms of the glucose fed to the chemostat. This is confirmed by the LC-MS-derived mass isotopomer fractions of both PEP and 2/3PG (supplementary Table S1), which were measured to be 90% naturally labelled (m+0/m+1/m+2) and 10% uniformly labelled (m+3). Similar mass isotopomer distributions were measured using GC-MS for the amino acid fragments Phe/Thr(1+2) and Ser(1+2) (see supplementary Table S1), which correspond to the first two carbon atoms of PEP and 3PG, respectively. The top three carbon atoms of the glucose fed to the chemostat end up in DHAP and are thus primarily converted into glycerol. The exact labelling of glycerol depends on the extent of the carbon redistribution in the nonoxidative branch of the PPP, but will primarily be labelled at the first carbon position (<90%) and uniformly labelled (<10%). In accordance with this, NMR-derived relative intensities of the first carbon atom of glycerol contained a singlet peak of 85% and a doublet peak of 15%, indicating a ratio of 85 : 15 for the labelling patterns ‘10X : 11X’ of glycerol, where ‘0’ indicates a 12C-atom, ‘1’ indicates a 13C-atom and ‘X’ can denote either.

Metabolic flux analysis for the individual datasets

13C-labelling-based MFA was performed for each of three datasets presented in supplementary Table S1. GC-MS and NMR data reflect the 13C-labelling pattern of proteinogenic amino acids and storage carbohydrates. These components are synthesized from metabolic precursors that are localized in a specific cellular compartment. By contrast, the LC-MS-derived mass isotopomers contain 13C-labelling information about cell-averaged metabolite pools, irrespective of the compartments in which they are localized. Owing to insufficient knowledge of the distribution of metabolites localized in multiple compartments (such as pyruvate and the TCA cycle metabolites) only the mass isotopomers of the glycolytic and PPP metabolites were measured and the metabolic flux-estimates for the LC-MS dataset were limited to glycolysis and the PPP. Estimated flux patterns for the three datasets are shown in Fig. 3 (first three values).

The most prominent feature of Fig. 3 is the similarity in flux estimates for the three individually fitted datasets. Differences in flux patterns are seen for the reversibilities of the transketolase and transaldolase catalysed reactions in the nonoxidative branch of the PPP. The reversibilities in the nonoxidative branch of the PPP have always proven difficult to estimate accurately (Follstad & Stephanopoulos, 1998). However, this has no influence on the flux estimates in the other parts of the network. A similar observation was made by Christensen et al. (2002) when estimating fluxes in the central metabolism of S. cerevisiae using samples taken at different time points during the cultivation. A second part of the metabolism that shows different flux estimates for the GC-MS and NMR datasets in Fig. 3 is the transport of pyruvate and acetyl-CoA across the mitochondrial membrane. With the help of a sensitivity analysis, it can be shown that the 13C-subtrate labelling employed cannot distinguish the two alternative routes (see next section).

Metabolic flux analysis for the combined dataset

To investigate further the comparability among the three measurement sets, 13C-labelling-based MFA was performed on pairwise combinations of the three datasets as well as on the combination of the three datasets. Figure 4 compares the build-up of the SSres for the combined datasets with those of the individual datasets, where the SSres is a measure of the discrepancy between the measured and simulated 13C-labelling distribution. In general, combining the datasets did not lead to a major increase in SSres. The total SSres when combining all three datasets was only 20% higher than the summed SSres for each separately fitted dataset, thereby confirming the coherence between the three datasets. Estimated fluxes for the combination of all three datasets are represented by the bottom values in Fig. 3. In line with the minimal increase in SSres, similar fluxes were fitted for the combined datasets compared with those fitted for the three individual datasets (Fig. 3).

Figure 4.

 Contribution of the LC-MS, GC-MS and NMR datasets to the minimized sum of squared residuals (SSres) as calculated via 13C-labelling-based MFA. Minimized SSres values were first derived for each individual dataset (indv.), after which combinations (comb.) of the 13C datasets were used for flux estimation.

As an additional check of the accuracy of the estimated 13C-labelling-based flux patterns the combined dataset was refitted, but now with the triosephosphate isomerase catalysed reaction included in the stoichiometric network model of the tpi1Δnde1,2Δgut2Δ strain. It was determined whether the metabolic flux through the isomerase would be negligible. Indeed, a negligible flux of 0.0028 mol (mol glucose)−1 was fitted for the forward reaction from GAP to DHAP, while the flux through the backward reaction was completely absent.

Flux-sensitivity analysis

The sensitivity of the calculated 13C-based fluxes to measurement error was determined for several important metabolic nodes in the metabolism of the quadruple deletion mutant as described by Kleijn et al. (2006). The measured 13C-labelling distributions were independently refitted for a range of fixed flux values for the studied metabolic node and the fold-change in the SSres was used as a measure of the flux sensitivity. Flux sensitivities were determined for the following nodes: (1) the PPP split-ratio, (2) the pyruvate carboxylase catalysed reaction, (3) the reversibility of the oxaloacetate transport and (4) the pyruvate decarboxylase catalysed reaction. Figure 5 shows the fold-changes in the SSres for the different datasets.

Figure 5.

 Observed fold increase in SSres when independently refitting the LC-MS, GC-MS and NMR 13C-labelling data for fixed values of (a) the PPP split-ratio, (b) the pyruvate carboxylase catalysed flux, (c) the reversibility of the oxaloacetate transporter and (d) the pyruvate decarboxylase catalysed flux. Metabolic flux estimates for the LC-MS dataset were limited to glycolysis and the PPP. As a result, the sensitivities of nodes b, c and d were not analysed for the LC-MS dataset.

The PPP split-ratio can be estimated accurately with all three 13C-labelling measurement techniques (Fig. 5a). Nonetheless, the GC-MS measurements are most sensitive to changes in the oxidative PPP flux, as denoted by the large fold-change in the SSres when the PPP split-ratio was fixed at suboptimal values. The minimal flux needed for the synthesis of biosynthetic precursors (e.g. amino acids) determines the lower boundary for the PPP split-ratios at 0.04 mol (mol glucose)−1, while the fixed glycerol production rate and biosynthetic flux from the G6P pool determine the maximum PPP split-ratio at 0.25 mol (mol glucose)−1.

The joint action of the enzymes pyruvate kinase, pyruvate carboxylase and PEP carboxykinase creates a futile cycle within the metabolic network of Fig. 3. Owing to uncertainties concerning energy-consuming processes this cycle could not by examined via metabolite balancing. Omission of the ATP balance in the 13C-labelling-based MFA enabled the flux through the futile cycle to be quantified. Figure 5b shows that the anaplerotic flux from pyruvate to oxaloacetate is best estimated via NMR. Note that the same sensitivity holds for the fluxes catalysed by pyruvate kinase and PEP carboxykinase that take part in the same metabolic cycle. The anaplerotic flux has a lower limit of 0.09 mol (mol glucose)−1 owing to the fixed biomass synthesis fluxes for cytosolic oxaloacetate and mitochondrial α-ketoglutarate.

The reversibility of the oxaloacetate transporter was determined for both the NMR and the GC-MS dataset. The largest fold-change in the SSres when fixing the reversibility of the oxaloacetate transporter was observed for the NMR dataset (Fig. 5c). Assuming a two-fold increase in the SSres as significantly different, the reversibility of the transporter will be within the interval 0.60–0.80. In terms of relative flux size this means an exchange flux ranging from 0.12 to 0.40 mol (mol glucose)−1.

The distribution of the flux from pyruvate towards acetyl-CoA via the two possible routes (see Fig. 3) is insensitive to both the GC-MS and the NMR dataset (Fig. 5d). This insensitivity is caused by the similar labelling of cytosolic and mitochondrial pyruvate and hence cytosolic and mitochondrial acetyl-CoA for both datasets (Christensen et al., 2002). The similar labelling of cytosolic and mitochondrial pyruvate in the GC-MS dataset follows from the mass isotopomer fractions of Phe(1+2) and Tyr(1+2) (corresponding to the first two carbon atoms of cytosolic pyruvate) and those of Val(1+2) (corresponding to the first two carbon atoms of mitochondrial pyruvate). The similar labelling of cytosolic and mitochondrial pyruvate in the NMR dataset follows from the relative intensities of Tyr-α and Phe-α (corresponding to cytosolic pyruvate) and those of Ala-α (corresponding to mitochondrial pyruvate). Note that the flux for the malic enzyme catalysed reaction may cause a difference between cytosolic and mitochondrial pyruvate through an extra inflow of label into the mitochondrial pyruvate pool. The greater difference in 13C-labelling between cytosolic and mitochondrial pyruvate for the NMR data in comparison with the GC-MS data therefore results in a higher flux estimate for the malic enzyme catalysed reaction from the NMR dataset (Fig. 3).

In the flux sensitivity analysis of the PPP split-ratio (Fig. 5a), the fixed glycerol production rate and fixed biosynthetic efflux from the G6P pool determined the maximum split-ratios at 0.25 mol (mol glucose)−1. However, these fluxes derived from biomass synthesis also have a certain degree of error. Figure 6a shows the effect of varying the biosynthetic G6P efflux on the SSres of the flux fit. Note that the performed flux fits were based upon the combined dataset (LC-MS, GC-MS and NMR) and not on the individual datasets used for the flux sensitivity analyses of Fig. 5. The combined dataset gave the most accurate flux estimates for the PPP split-ratio, for reasons explained below. Figure 6a shows that the optimal biosynthetic G6P efflux [0.10 mol NADH (mol glucose)−1] closely resembled the calculated G6P efflux [0.11 mol NADH (mol glucose)−1 in Fig. 3]. More importantly, Fig. 6a shows that the optimal PPP split-ratio remained unchanged at 0.25 mol (mol glucose)−1, despite the possibility of also fitting higher PPP split-ratio values. In Fig. 6b the contribution of the three different datasets to the optimal SSres are plotted, showing that the LC-MS dataset is most sensitive to changes in the G6P efflux and hence the PPP split-ratio. In retrospect, the earlier observed sensitivity of the GC-MS and NMR dataset to changes in the PPP split-ratio (based on Fig. 5a) denotes in actuality the sensitivity of these datasets to the methylglyoxal bypass flux. By varying the biosynthetic G6P efflux, the PPP split-ratio was uncoupled from the methylglyoxal bypass flux, resulting in a lower fold-change in the SSres for the NMR and the GC-MS datasets. In accordance with this, a small methylglyoxal bypass flux was fitted, independent of the chosen G6P efflux (Fig. 6a).

Figure 6.

 Observed fold increase in total SSres (a) and SSres of the individual datasets (b) when refitting the combined LC-MS, GC-MS and NMR 13C-labelling data for fixed values of the biosynthetic efflux from the G6P pool. Estimated values for the PPP split-ratio and the methylglyoxal bypass flux for the different biosynthetic G6P effluxes are given in (a).

Metabolite balancing versus 13C-based MFA

NAD(P)H balances

The NADH and NADPH balance are used to constrain the fluxes derived via metabolite balancing, but not those derived via 13C-labelling-based MFA. The consistency of the NADH and NADPH balance for the 13C-labelling-derived fluxes (combined datasets, Fig. 3) was checked by calculating the total production of NADH and NADPH in the cytosol and comparing these values with those derived via metabolite balancing. In these calculations the enzyme Aald dehydrogenase was considered to be NADP+-dependent. In addition, the pyruvate decarboxylase bypass flux was set at 0.08 mol (mol glucose)−1, which corresponded to the minimal flux needed to cover the biosynthetic need for cytosolic acetyl-CoA. A more precise flux estimate could not be made owing to the insensitivity of this node (see Fig. 5d). Note that a minimal flux was chosen, as a high flux through the pyruvate decarboxylase bypass is energetically highly unfavourable to the cell. The conversion of acetate into acetyl-CoA by the pyrophosphate-generating enzyme acetyl-CoA synthetase costs two ATP equivalents.

The total NADPH production in the cytosol was calculated to be 0.56 mol NADPH (mol glucose)−1 for the 13C-derived flux pattern. This is substantially higher than the 0.41 mol NADPH (mol glucose)−1 needed for biomass synthesis as calculated from the stoichiometric model used for metabolite balancing. The total NADH production in the cytosol was calculated to be 0.94 mol NADH (mol glucose)−1. In total, 0.80 mol NADH (mol glucose)−1 is reduced in the cytosol via glycerol synthesis, resulting in an excess of 0.14 mol NADH (mol glucose)−1 for transportation via a redox shuttle. This value agrees with the 0.16 mol NADH (mol glucose)−1 calculated via metabolite balancing.

Upper glycolysis

The absence of the triosephosphate isomerase in the tpi1Δnde1,2Δgut2Δ strain assured an equal FBP aldolase catalysed flux towards DHAP and GAP. Given the measured glycerol production rate [0.80 mol (mol glucose)−1], this meant that 0.20 moles of glucose had to be metabolized via a route other than the upper glycolysis, for instance via assimilation or via the PPP. Alternatively, the surplus of carbon could be metabolized by an alternative DHAP dissimilation pathway, such as the methylglyoxal bypass. Comparison of the 13C-based flux pattern with the flux pattern derived from metabolite balancing shows a similar assimilation flux from the G6P pool. However, in the 13C-labelling-based MFA a higher flux through the oxidative branch of the PPP and, consequently, a lower flux through the methylglyoxal bypass is fitted compared with the metabolite balancing method. This discrepancy can to a certain extent be explained by the sensitivity of the PPP split-ratio to small deviations in glycerol formation. Every 0.01 mol (mol glucose)−1 lowering of the glycerol yield will result in a 0.03 mol (mol glucose)−1 increase in the PPP split-ratio. However, a second explanation for the discrepancy in PPP split-ratio between the metabolite balancing and the 13C-labelling method is the difference in the constraints imposed by the two MFA methods.

Metabolite balancing constrains the flux through the PPP based on assumptions on the cytosolic NADPH demand of the cell, which leads to a diversion of excess carbon through the methylglyoxal bypass. Uncertainty about cofactor specificities of enzymes, especially for multiple isoenzymes, the presence of transhydrogenation cycles and the occurrence of as yet unknown sinks of NAD(P)H (e.g. caused by oxidative stress) make predictions based upon balances of reduction equivalents prone to error.

By contrast, the 13C-labelling-based MFA primarily constrains the flux through the methylglyoxal bypass as a result of the labelling of alanine. The measured m+1 mass isotopomer fraction and singlet peak measured for ala-all and ala-α, respectively, are enough to account for the presence of naturally labelled 13C, but leave little room for the inflow of 1-[13C1] DHAP via the methylglyoxal bypass. The excess carbon is therefore shuttled to the other remaining option: the PPP. Generally, 13C-labelling-based MFA will lead to more reliable estimates for the methylglyoxal bypass flux and coupled to that to the PPP split-ratio, as these calculations are based upon actual (13C)-measurements. In accordance with this the 13C-labelling-based estimate of the methylglyoxal bypass flux (0.5% of the total glucose uptake flux) resembles the value reported by Martins et al. (2001) for wild-type S. cerevisiae much more closely than the metabolite balancing-based estimate of 5%. Seemingly, the methylglyoxal bypass only plays a minor role in diverting carbon away from glycerol formation.

Redox-shuttle mechanisms

Metabolite balancing showed the necessity of incorporating a putative NADH shuttle in the stoichiometric network model to allow for the reoxidation of the surplus of NADH formed in the cytosol. In the absence of the external mitochondrial NADH dehydrogenases and the glycerol-3-phosphate shuttle, S. cerevisiae apparently possesses other mechanisms to oxidize cytosolic NADH. Several redox-shuttle mechanisms that enable the transfer of reduction equivalents from the cytosol to the mitochondrion have been proposed, such as the ethanol – acetaldehyde shuttle, the malate – oxaloacetate (MAL/OAA) shuttle, the malate – aspartate (MAL/ASP) shuttle and the malate – pyruvate (MAL/PYR) shuttle (Bakker et al., 2001). All key enzymes needed to operate these shuttles are present in S. cerevisiae. However, no conclusive evidence for in vivo activity of these shuttles has been reported to date, except perhaps for the activity of the ethanol – acetaldehyde shuttle in an adh3Ä S. cerevisiae strain (Bakker et al., 2000). In this context, it is relevant to note that the MAL/ASP shuttle has recently been proposed to play an important role during growth of S. cerevisiae on acetate and fatty acids (Cavero et al., 2003).

The 13C-labelling MFA of this study provides additional insight into the redox-shuttle mechanism used by S. cerevisiae. 13C-based flux-patterns displayed in Fig. 3 show a considerable reversibility for the transport of oxaloacetate. In the stoichiometric model for the 13C-labelling-based MFA both the cytosolic and the mitochondrial pools of malate and oxaloacetate are grouped together, as no separate labelling information was available for malate. In principle, the transport of oxaloacetate across the mitochondrial membrane can form part of the MAL/OAA shuttle or the MAL/ASP shuttle. In the MAL/OAA shuttle cytosolic NADH is used to reduce oxaloacetate to malate, which in turn is reoxidized to oxaloacetate in the mitochondrion producing NADH. The MAL/ASP shuttle works similarly, except that oxaloacetate is first converted into aspartate, which is then transported in symport with glutamate. Interestingly, cytosolic malate dehydrogenase, a crucial enzyme in both shuttles, is deactivated in the presence of high glucose concentrations (Minard & Mcalisterhenn, 1992). This might explain why Overkamp et al. (2002) observed higher glycerol yields in batch cultivations of the tpi1Δnde1,2Δgut2Δ strain. Under glucose-excess conditions the MAL/ASP or MAL/OAA shuttles are inactive, leaving more NADH for the formation of glycerol.

Direct transport of oxaloacetate (MAL/OAA shuttle) or transport in the form of aspartate (MAL/ASP shuttle) could not be distinguished owing to the fact that GC-MS and NMR analysis only provide the labelling of aspartate and not that of its precursor oxaloacetate. If the observed transport of oxaloacetate is indeed part of the MAL/ASP or the MAL/OAA shuttle, the net NADH production in the mitochondrion [0.19 mol (mol glucose)−1 for the combined dataset, see Fig. 3] is enough to remove the surplus of NADH created in the cytosol as calculated via metabolic balancing [0.16 mol (mol glucose)−1, see Fig. 1]. As shown in Fig. 5, NMR-derived relative intensities allowed for the most accurate estimate of the reversibility of the oxaloacetate transporter. Based upon a two-fold increase in the SSres, the net flux through the oxaloacetate transport was estimated to be between 0.12 and 0.40 mol (mol glucose)−1.

It was shown for the metabolite balancing method that altering the cofactor specificity of the cytosolic Aald dehydrogenase from NADP+ to NAD+ increased the PPP split-ratio from 16 to 21%, making the oxidative branch of the PPP the sole producer of cytosolic NADPH. As a result of the changed cofactor specificity, additional NADH was produced in the cytosol, thereby increasing the flux for the putative NADH shuttle from 0.16 mol (mol glucose)−1 to 0.26 mol (mol glucose)−1. This value is slightly higher than the exchange flux estimated for the combined 13C-labelling dataset, but still falls within the range identified via the flux sensitivity analysis [0.12–0.40 mol (mol glucose)−1].

Activity of the MAL/PYR shuttle, in which pyruvate is reduced to malate in the cytosol via pyruvate carboxylase and malate dehydrogenase and again oxidized to pyruvate in the mitochondrion via malic enzyme, is improbable given the estimated low flux through malic enzyme. Activity of the ethanol – acetaldehyde shuttle could not be verified using the employed 13C-labelling experiment as no information on the 13C-label distribution of ethanol or acetaldehyde was available.


This study provides further insight into the carbon and redox metabolism within a glycerol-overproducing tpi1Δnde1,2Δgut2ΔS. cerevisiae strain. For the first time the isotopic enrichment of the intracellular compounds in a single 13C-labelling experiment was measured using GC-MS, NMR and LC-MS. The combination of the three datasets resulted in similar flux-patterns as with the individual datasets and only a minor increase (<20%) in the summed SSres, which shows that the three 13C measurement techniques yield consistent results. In addition, combining the different techniques led to a more accurate final flux pattern as the sensitivity of the fluxes around several important metabolic nodes in the primary metabolism of the tpi1Δnde1,2Δgut2Δ strain proved to be dependent on the method of analysis. For example, NMR-derived relative intensities enabled an accurate estimation of the reversibility of the oxaloacetate transporter, the PPP split-ratio was most accurately estimated by the LC-MS-derived mass fractions, while the methylglyoxal bypass flux was best estimated via the GC-MS-derived mass fractions. This observation calls for an approach with multiple 13C measurement tools in future 13C-labelling-based MFA experiments.

The combination of metabolite balancing and 13C-labelling-based MFA provided physiological evidence that three pathways were used to divert carbon away from glycerol formation (in order of flux size): (1) the PPP, (2) the assimilatory pathway towards storage carbohydrates and (3) the methylglyoxal bypass. The 13C-derived PPP split-ratio and methylglyoxal flux are believed to be more accurate than the fluxes derived via metabolite balancing, as metabolite balancing relies heavily on an uncertain NADPH balance. Based on the 13C-labelling data an almost negligible flux through the methylglyoxal bypass was fitted, accounting for 0.5% of the total glucose uptake flux. Most carbon was diverted away from glycerol formation through the PPP (24%). Metabolite balancing showed that an alternative mechanism for the oxidation of cytosolic NADH was required to maintain redox balance. The 13C-derived exchange flux of oxaloacetate (grouped together with malate) across the mitochondrial membrane provided evidence for the MAL/OAA or alternatively the MAL/ASP shuttle as the alternative redox mechanism.

Authors Contribution

R.J.K. and J.M.A.G. contributed equally to this work.


The PhD research of R.J.K. was financially supported by the Dutch EET programme (Project No. EETK20002) and DSM. The PhD research of J.M.A.G. was financed by Tate & Lyle Ingredients Americas, Inc. The research groups of J.T.P. and J.J.H. are part of the Kluyver Centre for Genomics of Industrial Fermentation, which is supported by the Netherlands Genomics Initiative. GC-MS measurements and data processing were made possible by financial support (travel grant R81-743) of the Netherlands Organization for Scientific Research (NWO). W.v.W. acknowledges the hospitality and support of the Sauer Group of the Institute of Biotechnology, ETH Zurich, Switzerland.