In Saccharomyces cerevisiae, reduction of NAD+ to NADH occurs in dissimilatory as well as in assimilatory reactions. This review discusses mechanisms for reoxidation of NADH in this yeast, with special emphasis on the metabolic compartmentation that occurs as a consequence of the impermeability of the mitochondrial inner membrane for NADH and NAD+. At least five mechanisms of NADH reoxidation exist in S. cerevisiae. These are: (1) alcoholic fermentation; (2) glycerol production; (3) respiration of cytosolic NADH via external mitochondrial NADH dehydrogenases; (4) respiration of cytosolic NADH via the glycerol-3-phosphate shuttle; and (5) oxidation of intramitochondrial NADH via a mitochondrial ‘internal’ NADH dehydrogenase. Furthermore, in vivo evidence indicates that NADH redox equivalents can be shuttled across the mitochondrial inner membrane by an ethanol–acetaldehyde shuttle. Several other redox-shuttle mechanisms might occur in S. cerevisiae, including a malate–oxaloacetate shuttle, a malate–aspartate shuttle and a malate–pyruvate shuttle. Although key enzymes and transporters for these shuttles are present, there is as yet no consistent evidence for their in vivo activity. Activity of several other shuttles, including the malate–citrate and fatty acid shuttles, can be ruled out based on the absence of key enzymes or transporters. Quantitative physiological analysis of defined mutants has been important in identifying several parallel pathways for reoxidation of cytosolic and intramitochondrial NADH. The major challenge that lies ahead is to elucidate the physiological function of parallel pathways for NADH oxidation in wild-type cells, both under steady-state and transient-state conditions. This requires the development of techniques for accurate measurement of intracellular metabolite concentrations in separate metabolic compartments.
Pyridine-nucleotide coenzymes in Saccharomyces cerevisiae
The yeast Saccharomyces cerevisiae played a central role in the discovery of the pyridine nucleotide coenzyme NAD+ by Harden and Young , who identified NAD+ as ‘cozymase’, a dialysable substance required for glycolytic activity of cell-free yeast juices. After the subsequent discovery of the related coenzyme NADP+, the crucial role of pyridine-nucleotide-dependent redox reactions in the metabolism of S. cerevisiae has been firmly established. The yeast genome databases (see: http://www.proteome.com) currently list approximately 100 open reading frames that encode known or putative pyridine-nucleotide-dependent oxidoreductases, a number that may still increase as functions are assigned to genes with hitherto unknown functions.
NADH and NADPH serve different functions in the S. cerevisiae metabolic network. NADPH is preferentially used in assimilatory pathways. This is not always the case, however. In fact, the distinction between assimilatory and dissimilatory reactions in heterotrophic organisms is to some extent artificial. For example, glycolysis plays an essential role in sugar dissimilation, but also generates building blocks for biosynthesis. Furthermore, although most biosynthetic reactions use NADPH as a reductant, some NADH-linked reductions occur in the conversion of central metabolites (pyruvate, oxaloacetate, acetyl CoA) to cellular monomers, for example in amino acid biosynthesis.
During growth of S. cerevisiae on sugars, the preference for NADH in dissimilatory reductions (i.e. in the reduction of acetaldehyde to ethanol and the reduction of the quinone pool of the respiratory chain) is very strong. In contrast to various other yeasts (e.g. Candida utilis, Candida parapsilosis and Kluyveromyces lactis), S. cerevisiae cannot directly couple the oxidation of NADPH to the respiratory chain [3–6]. Furthermore, the yeasts investigated thus far lack a transhydrogenase activity that might catalyze the conversion of NAD+ and NADPH to NADH and NADP+ [5, 7]. Consequently, the hexose-monophosphate pathway, which produces the NADPH required for biosynthesis, cannot function as a dissimilatory route in S. cerevisiae. Also in fermentative sugar dissimilation by S. cerevisiae, the role of NADPH is limited, as the major alcohol dehydrogenases are strictly NAD+-dependent . NADPH-dependent acetate production should not be neglected, however .
Scope of this review
The aim of the present paper is to review current knowledge on the physiological aspects of the NADH/NAD+ redox balance in S. cerevisiae. Van Dijken and Scheffers reviewed physiological aspects of redox metabolism in yeasts in 1985 . Since then, two important developments have occurred that justify a revisitation of this subject. First, the identification of several structural genes encoding enzymes involved in NADH oxidation has enabled the application of a quantitative physiological analysis of defined mutants affected in these enzymes. Secondly, the completion of the S. cerevisiae sequencing project  has eliminated several uncertainties about the presence or absence of potential key enzymes in this yeast.
Redox metabolism cannot be viewed as an isolated process: its regulation is closely connected to central and peripheral reactions in carbon and nitrogen metabolism. This review will focus on the way in which NADH generated in these reactions is reoxidized, with special emphasis on the implications of metabolic compartmentation during sugar metabolism. As will be discussed, the occurrence of separate mitochondrial and cytosolic NADH/NAD+ redox couples is not only relevant during aerobic, respiratory growth, but also during anaerobic, fermentative growth.
With respect to respiratory reoxidation of NADH, this review will focus on the mechanisms at the interface between the NADH-yielding reactions in assimilatory and dissimilatory sugar metabolism and the mitochondrial respiratory chain. For detailed information on the molecular composition, biosynthesis and regulation of the mitochondrial respiratory chain in S. cerevisiae, the reader is referred to the excellent reviews by De Vries and Marres  and De Winde and Grivell .
Turnover of NADH in respiring and fermenting cultures
The NADH/NAD+ couple can be considered as a conserved moiety. Only catalytic amounts of these pyridine nucleotides are present in yeast. Intracellular concentrations of NAD+ plus NADH of 1 mM have been reported [14, 15]. Consequently, reduction of NAD+ has to be matched by a continuous reoxidation of NADH. Below, a brief overview will be presented of the major ‘sources and sinks’ of NADH and their impact on growth energetics.
Dissimilatory sugar metabolism
Under strictly anaerobic growth conditions, substrate-level phosphorylation in glycolysis is the sole source of ATP in S. cerevisiae and enables a net yield of 2 ATP for each molecule of glucose converted to two molecules of pyruvate. Dissimilation of glucose to pyruvate via glycolysis is stoichiometrically linked to the reduction of NAD+ in the glyceraldehyde-3-phosphate-dehydrogenase reaction. A closed redox balance in dissimilation is achieved by decarboxylation of pyruvate to acetaldehyde, which subsequently acts as the electron acceptor for NADH reoxidation. In wild-type S. cerevisiae, this pathway of ethanol formation is the only mode of fermentative metabolism. In contrast to many other facultatively fermentative eukaryotes, lactate fermentation does not occur in wild-type S. cerevisiae, since the predominant lactate dehydrogenases are not NAD+-linked, but cytochrome c-linked in this organism [16–18]. However, it has been demonstrated that introduction of heterologous lactate dehydrogenases can lead to a mixed ethanol-lactate fermentation in engineered S. cerevisiae strains [19–21].
In S. cerevisiae, maintenance of aerobic cultivation conditions is not sufficient to achieve respiratory sugar metabolism. Even fully aerobic cultures exhibit a mixed respirofermentative metabolism, unless they are grown with a limited sugar supply at low specific growth rates [22–24]. This phenomenon of aerobic fermentation (often referred to as the ‘glucose’ or ‘Crabtree’ effect [10, 22, 25]) is at least partly due to a repression of the synthesis of respiratory enzymes by excess glucose [13, 26–28]. In industrial and laboratory practice, completely respiratory growth on glucose and other sugars can be achieved by sugar-limited, aerobic cultivation in fed-batch or chemostat cultures at low specific growth rates [29–31].
During respiratory growth on sugars, reoxidation of the NADH formed in glycolysis can occur via mitochondrial respiration, thus yielding additional ATP via oxidative phosphorylation. Moreover, respiratory reoxidation of glycolytic NADH implies that pyruvate does not have to be converted into the electron acceptor acetaldehyde, but can be further oxidized. Oxidation of pyruvate to acetyl-coenzyme A predominantly occurs via the mitochondrial pyruvate–dehydrogenase complex . Subsequently, acetyl-CoA is oxidized to carbon dioxide and water by the enzymes of the tricarboxylic acid cycle. This yields one additional ATP equivalent per mole of pyruvate via substrate level phosphorylation in the reaction catalyzed by succinyl-CoA synthetase. Furthermore, with the exception of the FAD-linked succinate dehydrogenase, the dehydrogenases involved in the mitochondrial oxidation of pyruvate to carbon dioxide may all yield NADH. Reoxidation of NADH and FADH by the mitochondrial respiratory chain yields ATP via oxidative phosphorylation. Based on quantitative analysis of respiratory cultures, it has been concluded that the in vivo efficiency of this process in S. cerevisiae is lower than in many other micro-organisms. In cells growing under carbon and free-energy limitation, the effective P/O ratio (i.e. the number of ADP molecules phosphorylated per electron pair transferred to oxygen) is probably close to 1 [33–35].
Notwithstanding this low stoichiometry of oxidative phosphorylation, which is related to the absence of a proton-translocating NADH dehydrogenase in S. cerevisiae (see Section 3), completely respiratory dissimilation of glucose yields approximately 16 ATP per glucose (4 ATP from substrate level phosphorylation and circa 12 from oxidative phosphorylation). This is 8-fold higher than the maximum ATP yield from glucose dissimilation via alcoholic fermentation. The much higher ATP yield from respiratory sugar dissimilation is reflected in the biomass yields of sugar-limited cultures: the typical biomass yield on glucose of respiratory cultures is 0.5 g biomass per g glucose, whereas the biomass yield of anaerobic, fermentative cultures is typically 5-fold lower .
Assimilatory sugar metabolism
Yeast biomass (a typical chemical composition is C3.75H6.6N0.63O2.10) is slightly more reduced than glucose. Intuitively, one would therefore expect that assimilation (the conversion of glucose, ammonia and other nutrients into biomass) should result in a net input of reducing equivalents. In practice, biomass formation is accompanied by a net consumption of NADPH and a net production of NADH. The total process yields reducing equivalents in the form of NADH, because assimilation involves production of carbon dioxide . For example, in glucose-grown cultures the formation of acetyl-CoA, an important precursor for lipid and amino acid synthesis, involves oxidative decarboxylation of pyruvate. The implication of this assimilatory CO2 production is that the formation of 1 g of yeast dry biomass from glucose and ammonia is accompanied by the net reduction of circa 10 mmol NAD+ to NADH . The exact magnitude of this assimilatory NADH formation depends on the composition of the biomass and the extent to which exogenous lipids (S. cerevisiae requires unsaturated lipids for anaerobic growth ) contribute to the overall lipid content of the yeast cells. In addition to the formation of biomass, also the excretion of oxidized low-molecular-mass metabolites (e.g. pyruvate, acetaldehyde or acetate) during growth on glucose may lead to a net production of NADH [10, 40, 41].
During respiratory growth, the ‘excess’ NADH produced in assimilation and metabolite production can be reoxidized by mitochondrial respiration and thus contributes to meeting the overall ATP requirement for growth (Fig. 1). Under anaerobic conditions, respiration is not possible. Moreover, since alcoholic fermentation of glucose is itself a redox-neutral process, ethanol formation cannot account for the reoxidation of assimilatory NADH. S. cerevisiae and other yeasts solve this redox problem by reducing glucose to glycerol (glucose+2 ATP+2 NADH+2 H+→2 glycerol+2 ADP+2 Pi+2 NAD+) (Fig. 1). Mechanistic aspects and metabolic engineering of glycerol formation will be addressed in Section 2.
Compartmentation of NADH metabolism
As in other eukaryotic cells, the metabolism of S. cerevisiae cannot be understood without taking into account metabolic compartmentation. Microbodies (peroxisomes, glyoxysomes) play an important role during growth of S. cerevisiae on oleic acid and some other non-fermentable carbon substrates . In the context of this review, however, which primarily addresses the role of the NADH/NAD+ redox couple in carbohydrate metabolism, we will focus on the role of the cytosolic and mitochondrial compartments.
The mitochondrial inner membrane is virtually impermeable to pyridine nucleotide coenzymes . Consequently, the cellular redox balance dictates that reduced coenzymes must be reoxidized in the compartment where they are generated. In contrast to NADPH turnover, which occurs predominantly in the cytosol [44, 45], NADH-turnover occurs at high rates in the cytosol as well as in the mitochondrial matrix. For example, during respiratory growth on sugars, NADH is generated in the cytosol by glycolysis, as well as in the mitochondrial matrix by the enzymes of the tricarboxylic acid cycle. The mechanisms that couple the oxidation of NADH in these two compartments to the respiratory chain will be discussed in detail in Sections 3 and 4.
The relative rates of NADH turnover in the cytosol and the mitochondrial matrix strongly depend on the carbon source. For example, lactate dissimilation by S. cerevisiae, which is initiated by the oxidation of lactate to pyruvate via a cytochrome c-dependent mitochondrial lactate dehydrogenase, does not yield cytosolic NADH [16–18]. During growth on ethanol the situation is ambiguous. S. cerevisiae contains at least two cytosolic alcohol dehydrogenases and at least one mitochondrial isoenzyme [8, 46]. Although one of the cytosolic alcohol dehydrogenases, Adh2p, is believed to be primarily responsible for the dissimilation of ethanol , we observed that a prototrophic adh3Δ deletion mutant, which had a reduced mitochondrial alcohol dehydrogenase activity, had a 30% lower maximum specific growth rate on ethanol than the corresponding wild-type CEN.PK 113-7D (B.M. Bakker, unpublished). This suggests that growth on ethanol may involve the simultaneous oxidation of ethanol to acetaldehyde in two different compartments.
As discussed above, assimilation of glucose into biomass involves a net reduction of NAD+ to NADH. Also, this assimilatory NADH production is subject to metabolic compartmentation. Based on the biomass composition of S. cerevisiae, it has been estimated that some 60–80% of the NADH generated in biosynthetic reactions originates from the synthesis of amino acids [44, 45]. Although the localization of the enzymes is not known in all cases, rough calculations show that 30–50% of the NADH produced by amino acid synthesis is generated in the mitochondrial matrix . Nissen et al.  performed a quantitative flux analysis for anaerobically growing S. cerevisiae and showed that a large part of the mitochondrial NADH is derived from synthesis of 2-oxoglutarate, which is a precursor of glutamate. The synthesis of other amino acids from glutamate or other precursors contributes to a lesser extent to mitochondrial NADH production, while in DNA and RNA synthesis only cytosolic NADH and NADPH are involved . Since the enzymes involved in glycerol formation are exclusively cytosolic (see Section 2), anaerobic reoxidation of intramitochondrial NADH requires a shuttle mechanism that exports redox equivalents to the cytosol (see Section 4).
The respiratory chain of S. cerevisiae
During respiratory growth, both cytosolic and mitochondrial NADH are reoxidized by the respiratory chain. A general scheme for the respiratory oxidation of NADH by S. cerevisiae mitochondria is depicted in Fig. 3. NADH generated in the mitochondrial matrix is oxidized by an internal NADH:ubiquinone oxidoreductase, also called internal NADH dehydrogenase. In contrast to ‘classical’ complex I-type NADH dehydrogenases, the S. cerevisiae internal NADH dehydrogenase is neither inhibited by rotenone nor coupled to the generation of a proton-motive force . This contributes to the low ATP stoichiometry of oxidative phosphorylation in S. cerevisiae.
Like plant mitochondria [47, 48], but unlike mammalian mitochondria, yeast mitochondria oxidize cytosolic NADH directly [49, 50]. The enzyme responsible for this reaction is an NADH:ubiquinone oxidoreductase, which we will refer to as the external NADH dehydrogenase. Like the internal NADH dehydrogenase, it is localized in the mitochondrial inner membrane, it is rotenone-insensitive and it does not pump protons. Its active site faces the mitochondrial intermembrane space . Alternatively, cytosolic NADH can be oxidized by the respiratory chain via the glycerol-3-phosphate shuttle, consisting of cytosolic NADH-linked glycerol-3-phosphate dehydrogenase and a membrane-bound glycerol-3-phosphate:ubiquinone oxidoreductase [51, 52].
All known pathways of respiratory NADH oxidation in S. cerevisiae converge at the ubiquinone pool. Ubiquinone donates its electrons to cytochrome c via the bc1 complex. Terminal oxidation of cytochrome c by molecular oxygen is catalyzed by cytochrome c oxidase . Whether and how NADH can be shuttled across the mitochondrial inner membrane in S. cerevisiae, has not been completely resolved yet and will be discussed below.
The respiratory chain of S. cerevisiae differs from that of other fungi and of plants in several respects. Despite many disputes about the possible occurrence of rotenone sensitivity of NADH oxidation in S. cerevisiae, it is clear that complex I is not functional in S. cerevisiae, since the genes encoding the subunits of this multi-enzyme complex are absent from the mitochondrial genome of this yeast . Furthermore, plants and various yeasts, filamentous fungi and parasites contain a cyanide-insensitive alternative oxidase which catalyzes the direct oxidation of ubiquinone by molecular oxygen without generating a proton-motive force . Alternative oxidase is absent from S. cerevisiae, since the genome of S. cerevisiae does not contain any putative alternative-oxidase genes. Finally, mitochondria of many plants and fungi possess an external NADPH:ubiquinone oxidoreductase [47, 48]. As mentioned above, this enzyme is also absent from S. cerevisiae.
Caution should be taken when S. cerevisiae is used as a model organism for qualitative and quantitative studies of redox metabolism. As discussed above, a large variation is found in respiratory chains of plants and fungi. Also quantitative differences between S. cerevisiae strains are found, for example concerning the quantitative importance of multiple routes of NADH oxidation and the maximum rate of respiratory glucose metabolism.
In Section 2, we will discuss the biochemistry and regulation of glycerol production in S. cerevisiae. It will be shown that glycerol production not only serves as a redox sink, but it is also a way to protect the yeast cell against osmotic stress. In Section 3, the biochemistry of the internal and external NADH dehydrogenases is described and their physiological roles are discussed on the basis of mutant phenotypes. Section 4 deals with NADH shuttles across the mitochondrial membrane. First, the biochemistry and physiological importance of two shuttles that are now known to be active in S. cerevisiae, are discussed. These are the glycerol-3-phosphate shuttle and the ethanol–acetaldehyde shuttle. Secondly, on the basis of the known localization of enzymes and transporters it will be investigated which other mitochondrial redox shuttles may play a role in S. cerevisiae. Section 5 contains conclusions and future prospects.
Mechanism and regulation of glycerol production
In addition to its role in redox metabolism , which has been introduced in the preceding paragraphs, glycerol formation in S. cerevisiae is important for osmoregulation. At high extracellular osmolarity, glycerol acts as a compatible solute and is retained intracellularly [55, 56].
Glycerol formation is initiated by the reduction of the glycolytic intermediate dihydroxyacetone phosphate to glycerol-3-phosphate. The S. cerevisiae genome harbors two structural genes, GPD1 [57, 58] and GPD2 that each encode an active isoenzyme of the responsible NAD+-dependent glycerol-3-phosphate dehydrogenase. The amino acid sequences of Gpd1p and Gpd2p are 69% identical  and the kinetic properties of the two isoenzymes are similar. The Km of Gpd2p for dihydroxyacetone phosphate (86 μM) is somewhat higher than that of Gpd1p (37–54 μM) , whereas both enzymes exhibit a Km of 18 μM for NADH .
The final reaction in glycerol synthesis by S. cerevisiae is the irreversible hydrolysis of the phosphate group from glycerol-3-phosphate, a reaction catalyzed by glycerol-3-phosphatase. As for the dehydrogenase, two structural genes encoding isoenzymes of the phosphatase have been identified, GPP1 and GPP2. The sequences of these proteins are 95% identical . Although the cellular localization of Gpp1p and Gpp2p has not been investigated, it seems probable that they reside in the cytosol, where glycerol 3-phosphate is synthesized. Both isoenzymes have a high specificity for dl-glycerol-3-phosphate, a pH optimum of 6.5 and a Km for glycerol 3-phosphate of 3–4 mM .
The two physiological roles of glycerol production in S. cerevisiae have inspired a number of detailed studies into the regulation of the GPD and GPP genes. Despite the similar physical and catalytic properties of their gene products, the GPD1 and GPD2 genes are differentially regulated at the transcriptional level. Also, the mutant phenotypes indicate that Gpd1p and Gpd2p fulfil different roles in metabolism. Gpd1p is predominantly responsible for adaptation of S. cerevisiae to osmotic stress [57, 60, 62], while Gpd2p is important for maintenance of the cellular redox balance under anaerobic conditions [60, 63]. The mRNA level of GPD2, but not that of GPD1, was 9-fold upregulated after a shift of S. cerevisiae from aerobic to anaerobic conditions . During prolonged anaerobic growth, expression of GPD2 decreased  and in steady-state anaerobic chemostat cultures the mRNA concentration stabilized at 1.5 times the aerobic level . The analysis of deletion mutants has provided further support for the notion that Gpd1p and Gpd2p fulfil different roles. Under anaerobic conditions, a gpd2Δ deletion mutant grew slower and produced less glycerol than the wild-type, whereas anaerobic growth of a gpd1Δ mutant was not affected [60, 63].
At increasing osmolarity the expression of GPD1, but not that of GPD2, was induced . This is consistent with the phenotype of null mutants: at high osmolarity (growth in the presence of 0.7 M NaCl), growth of gpd1Δ strains was strongly impaired, whereas a gpd2Δ mutants exhibited a wild-type growth rate [57, 60].
A gpd1Δ gpd2Δ double mutant cannot grow at all under anaerobic conditions, unless an NADH-oxidizing agent, such as acetaldehyde or acetoin is added [60, 63]. At high osmolarity, the double mutant gpd1Δ gpd2Δ does not grow either, nor does it produce glycerol . Anaerobic growth of the gpd1Δ gpd2Δ double mutant in the presence of acetaldehyde or acetoin is possible [60, 63], because the use of glycerol 3-phosphate as a precursor for triacylglycerol can be bypassed by the enzymes dihydroxyacetone-phosphate acyltransferase and acyl-dihydroxyacetone-phosphate reductase [65, 66].
Synthesis of the two glycerol-3-phosphatase isoenzymes is also differentially regulated. The cellular concentration of Gpp2p protein exhibits a strong positive correlation with the NaCl concentration in batch cultures. At an NaCl concentration of 1.4 M, the Gpp2p level was 3–4-fold higher than in reference cultures without added NaCl . In these cultures, Gpp1p was present at a constant, high level irrespective of the salt concentration . In contrast, transcriptional regulation of GPP1 appears to be redox-related, since 2.5-fold higher transcript levels were found in anaerobic glucose-limited chemostat cultures than in aerobic cultures. In the same cultures, the mRNA level of GPP2 was slightly lower under anaerobic conditions than under aerobic conditions . As such, the differential regulation of the GPP gene products closely resembles that of the GPD gene products.
In contrast to the role of glycerol formation in redox metabolism, the function of glycerol as a compatible solute requires that it be retained intracellularly. In S. cerevisiae, transport of glycerol across the plasma membrane is facilitated by the channel protein Fps1p . At high osmolarity, the channel is closed and the produced glycerol is retained inside the cells, where it acts as a compatible solute [67, 68]. After a shift from high to low osmotic strength, the cells released the accumulated glycerol within a few minutes. Apparently the Fps1p protein remains present in an inactive form during high osmolarity [67, 68]. An fps1Δ deletion mutant was extremely sensitive to such a shift from high to low osmolarity, indicating that Fps1p channel is important to protect the cells from the consequences of a hypo-osmotic shock [67, 68]. In anaerobic steady-state cultures the expression of FPS1 is 1.5-fold higher than under aerobic conditions . This induction is comparable to that of GPD2 in anaerobic steady-state chemostat cultures. It would be interesting to see whether the expression of FPS1 is also induced more strongly directly after the switch to anaerobic conditions, like that of GPD2.
Metabolic engineering of glycerol production
During both World Wars, S. cerevisiae was used on an industrial scale to produce glycerol from carbohydrate-containing feedstocks [69, 70]. These processes were based on the observation that sulfite addition to anaerobic, fermenting yeast cultures leads to extensive glycerol formation . Bisulfite readily forms an adduct with acetaldehyde, thus preventing it from acting as an electron acceptor for reoxidation of glycolytic NADH. This forces the cells to use glycerol formation as an alternative redox sink. Such redirection of carbon metabolism to increase production of an economically important metabolite may be considered as ‘metabolic engineering avant la lettre’ .
In the sulfite process, glycerol formation has to be matched by the formation of equimolar amounts of the acetaldehyde adduct (plus carbon dioxide). Furthermore, as stoichiometric conversion of glucose into acetaldehyde and glycerol by S. cerevisiae does not lead to net production of ATP, some alcoholic fermentation is needed to sustain glycolytic activity. As a result, the glycerol yields achieved in practice did not exceed ca. 0.35 mol glycerol per mol of glucose equivalents. These low yields, problems in product recovery and the requirement for large quantities of sulfite have led to the demise of the sulfite process [69, 70].
As expected from the biochemical explanation of the sulfite process, S. cerevisiae strains with a decreased expression of alcohol dehydrogenase  or pyruvate decarboxylase  exhibit increased levels of glycerol in glucose-grown batch cultures. However, glycerol levels observed in cultures of pyruvate-decarboxylase-deficient S. cerevisiae mutants were low in comparison to those observed in the sulfite process . This led to the proposal that the high glycerol yields in the sulfite process may not be solely due to trapping of acetaldehyde . In this respect, it is interesting to note that elevated levels of NAD+-dependent glycerol-3-phosphate dehydrogenase have been observed in the presence of sulfite . Indeed, even in the absence of sulfite, overexpression of the GPD1-encoded glycerol-3-phosphate dehydrogenase results in a diversion of the glycolytic flux towards glycerol formation [74, 78]. High glycerol yields (up to 0.9 mol mol glucose−1) were observed in bioconversions with a tpi1Δ mutant . In this strain, which completely lacks triose phosphate isomerase activity, glycerol formation is the predominant mechanism to prevent toxic accumulation of dihydroxyacetone phosphate. As will be discussed in Section 4, inactivation of mechanisms for mitochondrial oxidation of cytosolic NADH can also result in substantial accumulation of glycerol by respiring cultures . It will be of interest to see whether combination of the metabolic engineering strategies mentioned above may lead to a further improvement of glycerol production rates and yields.
In anaerobic cultures, the redox balance dictates that an increased specific rate of glycerol production be balanced by an increased conversion of glucose into more oxidized metabolites (e.g. pyruvate, acetaldehyde or acetate; the latter two metabolites in combination with carbon dioxide). As these compounds are likely to influence the sensory qualities of wine, this may complicate attempts to modify the glycerol-to-ethanol ratio in wine fermentations via metabolic engineering .
Glycerol formation is undesirable in the industrial production of ethanol from carbohydrate feedstocks. It has recently been estimated that elimination of glycerol production in industrial yeast fermentations aimed at the production of (fuel) alcohol might increase the annual world-wide production of ethanol by 1.25×109 liters . Glycerol production in fermentative cultures is strongly dependent on the medium composition. In particular, the nitrogen source used for growth on glucose strongly influences the glycerol yield in anaerobic S. cerevisiae cultures . When, instead of ammonium sulfate, a mixture of amino acids was used as the nitrogen source, the glycerol production per amount of biomass was reduced by over 3-fold . Another way to control glycerol production is by oxygen-limited cultivation. At growth-limiting oxygen feeds, glycerol formation only sets in when the specific oxygen uptake rate of the biomass decreases below the rate that is required to reoxidize the ‘assimilatory’ NADH formed in biosynthesis  (Fig. 2). This tight regulation of glycerol production by oxygen availability implies that, in laboratory fermentations, the oxygen feed can be accurately controlled in such a way that oxygen is exclusively used for reoxidation of ‘assimilatory’ NADH and that the amount of NADH produced in glycolysis is reoxidized by ethanol formation. It is unlikely that the extremely accurate dosage of oxygen required to attain this situation can be reproducibly achieved in large-scale industrial fermentations. Consequently, much effort is invested in the development of metabolic engineering strategies to minimize glycerol formation.
The most promising results to date are based on the fact that, during assimilation of biomass from glucose, S. cerevisiae reduces NAD+ to NADH, but at the same time oxidizes NADPH to NADP+ (for a review see ). One of the major NADPH-consuming reactions is the reductive amination of 2-oxoglutarate by the GDH1- and GDH3-encoded NADP-dependent glutamate dehydrogenases. In an elegant study, Nissen et al.  inactivated GDH1 and overexpressed the GLN1 and GLT1 genes, which encode glutamine synthetase and glutamate synthase, respectively [84, 85]. As glutamate synthase uses NADH as a reductant, these genetic modifications changed the coenzyme requirement of glutamate biosynthesis from NADPH to NADH. Indeed, the resulting strains exhibited a 38% reduction of the glycerol yield and a 10% increase of the ethanol yield on glucose in anaerobic batch cultures . Recently, it has been attempted to further reduce glycerol production by the functional expression of heterologous transhydrogenases in S. cerevisiae [86, 87]. The rationale for these studies is that introduction of a transhydrogenase (catalysing the reaction NADH+NADP+↔NAD++NADPH) might allow for a conversion of NADH into NADPH, thus reducing the net production of NADH in biosynthesis. Although transhydrogenases from Escherichia coli and Azotobacter vinelandii were functionally expressed in S. cerevisiae, this did not result in a decreased production of glycerol. Instead of reducing NADH production, the heterologous transhydrogenases catalyzed the reverse reaction, i.e. the transfer of electrons from NADPH to NAD+ in anaerobic cultures [86, 87].
Mitochondrial NADH dehydrogenases
Internal NADH dehydrogenase
Physical and catalytic properties
In contrast to many eukaryotic cells including other yeasts , S. cerevisiae lacks the multi-subunit complex I-type NADH dehydrogenase . Instead, S. cerevisiae contains a single-subunit NADH:ubiquinone oxidoreductase, which couples the oxidation of intramitochondrial NADH to the respiratory chain. This enzyme, referred to as the ‘internal NADH dehydrogenase’, catalyzes the transfer of two electrons from intramitochondrial NADH to ubiquinone [88, 89]. The unique nuclear gene NDI1 encodes a 57-kDa precursor protein , with a 26-amino acid N-terminal targeting sequence that is cleaved off upon import into the mitochondria . Ndi1p contains non-covalently bound FAD as the sole prosthetic group .
After its initial purification, Ndi1p was thought to be an external NADH dehydrogenase . Later it was established that mature Ndi1p is localized in the inner mitochondrial membrane, with its active site facing the mitochondrial matrix . In contrast to complex I-type NADH dehydrogenases, Ndi1p itself does not pump protons. Thus, transfer of electrons from NADH to the ubiquinone pool does not contribute to the generation of a proton-motive force across the mitochondrial inner membrane . This has interesting consequences for growth energetics. The in vivo P/O ratio of S. cerevisiae mitochondria has been estimated to be close to unity . Consequently, the ‘textbook value’ of the ATP yield of completely respiratory dissimilation of glucose (36 mol ATP per mol glucose) is not correct for S. cerevisiae. With an estimated in vivo P/O ratio of 1, a value of 16 ATP/glucose would be more realistic.
Ndi1p reacts specifically with NADH. Its activity with related compounds such as NADPH or deamino-NADH is at least 250-fold lower than its activity with NADH . The Km for NADH is approximately 30 μM . In vitro, Ndi1p can use various electron acceptors, such as ubiquinone-2, ubiquinone-6, dichloroindophenol or ferricyanide . When localized in the mitochondrial membrane, the activity of Ndi1p with ubiquinone-2 is five times higher than with the endogenous electron acceptor ubiquinone-6 . With ubiquinone-6 as a substrate, Ndi1p has a very broad pH optimum, ranging from pH 4.5 to 9.5 . The enzyme is inhibited by flavone, but it is insensitive to typical complex I inhibitors, such as piericidin or rotenone .
NDI1 is subject to glucose repression. When growing on a non-fermentable carbon source, such as lactate or ethanol, S. cerevisiae contains much more Ndi1p protein than on glucose as the carbon source . In glucose-grown batch cultures the mRNA level of NDI1 is induced up to 4-fold after the diauxic shift, when glucose has been depleted and the cells continue to grow on ethanol . NDI1 transcription is not shut down under anaerobic conditions: transcript levels in anaerobic, glucose-limited chemostat cultures are only 2-fold lower than in aerobic cultures .
Physiology of ndi1Δ mutants
Mitochondria isolated from S. cerevisiae ndi1Δ mutants do not oxidize substrates that generate NADH in the mitochondrial matrix (i.e. malate plus pyruvate or ethanol). This confirms that Ndi1p is the only internal NADH dehydrogenase in S. cerevisiae [46, 89]. In shake-flask cultures on mineral medium with glucose, galactose or ethanol as the sole carbon source, ndi1Δ mutants exhibit similar specific growth rates as wild-type cells ( and B.M. Bakker, unpublished results). Conversely, growth on acetate and pyruvate is completely abolished . The apparently normal growth of the ndi1Δ mutant on ethanol  was surprising, because complete dissimilation of ethanol is expected to require reoxidation of the intramitochondrial NADH generated in the citric acid cycle. The ndi1Δ strain also exhibited completely respiratory growth in aerobic glucose-limited chemostat cultures at a dilution rate of 0.10 h−1, with a negligible production of ethanol or acetate. The biomass yield of the ndi1Δ mutant on glucose (0.43 g g−1) was slightly lower than that of the wild-type strain (0.49 g g−1) . These results may suggest that, in the absence of an internal NADH dehydrogenase, a redox shuttle might transport the redox equivalents from the mitochondrial matrix to the cytosol, thus coupling the oxidation of intramitochondrial NADH to the external NADH dehydrogenases and/or the glycerol-3-phosphate shuttle (Fig. 3).
One possible mechanism to couple oxidation of intramitochondrial NADH to the external NADH dehydrogenase would be an ethanol–acetaldehyde shuttle. This shuttle, first proposed by Von Jagow en Klingenberg , consists of a mitochondrial and a cytosolic isoenzyme of NADH dehydrogenase. Since ethanol and acetaldehyde are generally assumed to diffuse freely across biological membranes, no transporters should be required for the net exchange of mitochondrial NADH and cytosolic NAD+ (Fig. 3). Furthermore, such a process does not require additional Gibbs energy. Experimental evidence that this shuttle indeed operates in S. cerevisiae ndi1Δ mutants, has recently been obtained. In contrast to single adh3Δ and ndi1Δ mutants, an adh3Δ ndi1Δ strain, which not only lacks Ndi1p, but also the mitochondrial alcohol dehydrogenase Adh3p , did not exhibit completely respiratory growth in aerobic, glucose-limited chemostat cultures. Instead, it exhibited a high rate of alcoholic fermentation (1.1 mmol ethanol (g dry weight)−1 h−1) and its biomass yield on glucose was only 0.29 g g−1.
The somewhat reduced biomass yield of the ndi1Δ mutant on glucose (0.43 g g−1 for the mutant versus 0.49 g g−1 for the wild-type) is consistent with the operation of a redox shuttle that does not require a net input of metabolic energy, combined with a redirection of pyruvate metabolism via the pyruvate dehydrogenase bypass. This bypass involves the cytosolic enzymes pyruvate decarboxylase, acetaldehyde dehydrogenase, and acetyl-coenzyme A synthetase . Formation of acetyl CoA by the latter enzyme requires the input of 2 ATP equivalents. Redirection of pyruvate dissimilation via the cytosolic pyruvate dehydrogenase bypass could reduce the required rate of intramitochondrial NADH turnover. Accordingly, the biomass yield of the ndi1Δ mutant is almost identical to that of pyruvate dehydrogenase-negative S. cerevisiae, in which all acetyl CoA is synthesized via the pyruvate dehydrogenase bypass .
Heterologous expression of yeast NDI1: gene therapy of mammalian complex I deficiency?
In mammalian cells, deficiency of the proton-pumping complex I-type NADH dehydrogenase causes severe cellular disorders, mainly due to impairment of respiratory NADH oxidation and production of superoxide radicals . S. cerevisiae NDI1 has been proposed to be a serious candidate for gene therapy of complex I-deficient patients. This proposal is based on two considerations (see  and citations therein). First and foremost, Ndi1p is much simpler than complex I-type NADH dehydrogenases. In contrast to complex I, which is composed of at least 43 subunits, Ndi1p consists of a single subunit and lacks iron–sulfur clusters. Secondly, Ndi1p is assumed to catalyze a two-electron reaction, in contrast to complex I, which catalyzes a one-electron reaction. This reduces the risk of free-radical production.
The S. cerevisiae NDI1 gene has been functionally expressed in Escherichia coli. In the transformed cells, Ndi1p enzyme acted as an integral part of the respiratory chain, as was evident from its localization in the membranes and from the inhibition of NADH-oxidase activity by flavone, an inhibitor of the internal NADH dehydrogenase from S. cerevisiae. Furthermore, heterologous expression of S. cerevisiae NDI1 in complex I-deficient Chinese hamster fibroblasts restored oxidation of intramitochondrial NADH. This activity was not inhibited by rotenone, an inhibitor of complex I, but it was inhibited by flavone . Expression of NDI1 also restored the ability of complex I-deficient Chinese hamster cells to grow on galactose, which requires a functional respiratory chain . Recently, S. cerevisiae NDI1 was expressed in human kidney cells . Unlike non-transfected kidney cells, the NDI1-transfected cells were able to grow in the presence of rotenone on a medium with 0.6 mM glucose or 0.6 mM glucose plus 5 mM galactose, indicating that also in these cells the S. cerevisiae protein was functional . These heterologous expression studies demonstrate that Ndi1p can function in various cell types, ranging from bacteria to mammalian cells. This feature makes the S. cerevisiae NDI1 gene an interesting candidate for gene therapy of complex I-deficient patients.
The inverse experiment, the functional expression of complex I in S. cerevisiae, might be an interesting way to increase the biomass yield of this yeast. This is much more complicated, of course, than heterologous expression of NDI1. Complex I consists of many different subunits of which some are encoded on the mitochondrial genome and others on the nuclear genome. Correct and balanced expression, assembly and mitochondrial targeting of this multi-subunit complex will give rise to many complications. Although the biotechnological aim of increasing the free-energy efficiency of respiration may be far away, attempts of expressing complex I in S. cerevisiae are likely to improve our understanding of targeting and assembly of multi-subunit complexes.
External NADH dehydrogenases
Physical, catalytic and regulatory properties
Yeast mitochondria, like those of plants , not only contain an internal mitochondrial NADH dehydrogenase, but also an external NADH dehydrogenase activity . Like the internal NADH dehydrogenase, the external isoenzymes do not pump protons . S. cerevisiae has two genes encoding external NADH dehydrogenase isoenzymes, NDE1 and NDE2 [98, 99]. Both genes were identified on the basis of their homology to NDI1. The amino acid sequence of Nde1p is 48% identical to that of Ndi1p and the Nde2p sequence is 46% identical to that of Ndi1p, over the whole length of the proteins [98, 99]. The identity of the peptide sequences of Nde1p and Nde2p to each other is 62%[98, 99]. Alignment of the three protein sequences reveals that Nde1p and Nde2p possess an N-terminal extension that is absent in Ndi1p . These extensions may contain signals for targeting to the mitochondrial inner membrane. The N-terminal amino acid sequence of a purified external NADH dehydrogenase (XXXXVILQKVAT) suggests that Nde1p is cleaved between residue 41 and 42 . To our knowledge, no experiments have been performed to demonstrate directly which parts of the N-terminal sequences of Nde1p and Nde2p are responsible for mitochondrial targeting.
The kinetic properties of the external NADH dehydrogenases have not been studied in detail, as far as we know. De Vries and Grivell  found that the Km for NADH of the NADH:ubiquinone-2 oxidoreductase activity in sonicated mitochondria was approximately 30 μM, like that of purified Ndi1p. Since they found that after sonication, the orientation of the mitochondria remained right-side-out, the measured Km probably represents a weighted average of the Km values of the two external NADH dehydrogenases .
Wild-type mitochondria oxidize exogenous NADH at a high rate and with a respiratory control value of 3.0, which is high for isolated yeast mitochondria . Luttik et al. showed that mitochondria isolated from an nde1Δ nde2Δ deletion mutant do not oxidize exogenous NADH anymore . Small and McAlister-Henn , however, reported some residual oxidation of exogenous NADH by nde1Δ nde2Δ mitochondria. This discrepancy may be explained by the fact that the latter authors used commercial NADH, which contains a substantial amount of ethanol (0.5 mol ethanol per mol NADH) [98, 100]. Ethanol is readily oxidized by isolated yeast mitochondria via the intramitochondrial alcohol dehydrogenase (Adh3p) and the internal NADH dehydrogenase. Luttik et al. used pure NADH, generated in situ by glucose dehydrogenase, glucose and NAD+. They showed that all residual ‘NADH’ oxidation by nde1Δ nde2Δ mitochondria, as measured with commercial NADH, was in fact due to contaminating ethanol . Under conditions when the cells exhibited respiratory growth, such as in glucose-limited chemostats or in shake-flasks on ethanol, deletion of NDE2 alone did not substantially decrease the rate of mitochondrial oxidation of exogenous NADH. In contrast, deletion of NDE1 caused a 3-fold decrease of the NADH respiration rate by mitochondria isolated from these cultures [98, 99]. This suggests that the physiological role of NDE1 is more important than that of NDE2, at least during respiratory growth on glucose or ethanol.
In glucose-grown shake-flask cultures of wild-type S. cerevisiae, transcript levels of both NDE1 and NDE2 increase after the diauxic shift, where glucose is depleted and the cells start to consume the ethanol produced in the first growth phase . NDE2 is induced much more strongly (7-fold) than NDE1 (two-fold) . The mRNA levels of NDE1 and NDE2 are approximately identical when cells grown on glucose are compared to galactose-grown cells [99, 101]. Both NDE1 and NDE2 are typical aerobic genes. In steady-state glucose-limited chemostat cultures the mRNA level of NDE1 increases more than 4-fold when going from anaerobic to aerobic conditions, while the mRNA level of NDE2 increases even 14-fold under these conditions .
Physiology of an S. cerevisiae nde1Δ nde2Δ mutant
In glucose-grown shake-flask cultures on a mineral medium, the maximum specific growth rate (μmax) of an nde1Δ nde2Δ deletion mutant was approximately the same as that of the wild-type reference strain (0.4 h−1 at 30°C) . This is not surprising, because growth of S. cerevisiae under these conditions is predominantly fermentative . On mineral medium with ethanol or galactose, the nde1Δ nde2Δ mutant grew more slowly than the wild-type strain [98, 99]. This decrease of the growth rate could be attributed solely to deletion of NDE1, as the additional deletion of NDE2 was without further effect [98, 99]. Again this illustrates that, under these conditions, NDE1 is physiologically more important than NDE2. Moreover, these results indicate that alternative mechanisms that might couple oxidation of cytosolic NADH to the mitochondrial respiratory chain (e.g. the glycerol-3-phosphate shuttle), cannot completely replace the external NADH dehydrogenases.
In aerobic glucose-limited chemostat cultures, growth of wild-type S. cerevisiae is completely respiratory at low dilution rates, whereas a respirofermentative metabolism occurs at high dilution rates. Completely respiratory growth on glucose implies that all NADH produced in glycolysis by glyceraldehyde-3-phosphate dehydrogenase is oxidized via mitochondrial respiration. In the prototrophic strain CEN.PK113-7D, deletion of both NDE genes caused a reduction of the critical dilution rate at which alcoholic fermentation set in (from 0.30 to 0.23 h−1 (Fig. 4) . At low dilution rates, growth of the nde1Δ nde2Δ mutant was almost completely respiratory, with only traces of glycerol being produced. This indicates that, at low specific growth rates, the role of the external dehydrogenases can be taken over by the glycerol-3-phosphate shuttle or other cytosolic NADH-oxidizing systems. The external NADH dehydrogenases are essential, however, to maintain completely respiratory growth up to the dilution rate at which aerobic alcoholic fermentation sets in in the wild-type strain .
Mitochondrial redox shuttles in S. cerevisiae
Some transport of NAD+ or NADH across the mitochondrial inner membrane is necessary to supply the mitochondrial enzymes with their coenzymes during growth. However, this transport proceeds at an extremely low rate compared to the rate of oxidation of intramitochondrial NADH [43, 103, 104]. Although physical transport of NAD+ or NADH across the mitochondrial inner membrane is therefore insignificant in redox cycling, many eukaryotic cells harbor redox shuttles, which transfer the redox equivalents of NADH across the mitochondrial membrane. Other shuttle mechanisms do not result in a translocation of redox equivalents across the membrane, but do bypass the NADH dehydrogenases on one side of the membrane.
Originally, two types of redox shuttles were distinguished . Type-I shuttles involve a large change in free energy. An example of such a shuttle is the glycerol-3-phosphate shuttle (Fig. 3), which does not result in a translocation of redox equivalents across the mitochondrial inner membrane, but does bypass the NADH dehydrogenases. Type II shuttles are symmetrical shuttles that follow the general scheme of Fig. 5. Originally it was thought that the standard free-energy differences of type II shuttles was zero. Later it became apparent that, in some cases, the symmetry of a type II shuttle was broken because an active transporter was involved. An example of such an asymmetrical type II shuttle is the malate–aspartate shuttle, which will be discussed below. In contrast to shuttle mechanisms that completely rely on passive or facilitated diffusion, mechanisms that involve active transport may allow transport of redox equivalents against a concentration gradient.
Mechanism of the glycerol-3-phosphate shuttle
The glycerol-3-phosphate shuttle (Fig. 3) is an indirect mechanism to oxidize cytosolic NADH and transfer the electrons to the respiratory chain. This shuttle consists of two components: cytosolic NAD+-linked glycerol-3-phosphate dehydrogenase and mitochondrial glycerol-3-phosphate:ubiquinone oxidoreductase, often referred to as mitochondrial glycerol-3-phosphate dehydrogenase . The two isoenzymes of NAD+-linked glycerol-3-phosphate dehydrogenase, Gpd1p and Gpd2p, have been discussed Section 2 on glycerol production. The mitochondrial FAD-linked glycerol-3-phosphate dehydrogenase, encoded by GUT2, is localized in the inner mitochondrial membrane and transfers electrons from cytosolic glycerol 3-phosphate to ubiquinone. The GUT2 gene was cloned by Rønnow and Kielland-Brandt  by complementation of a mutant isolated by Sprague and Cronan that lacked mitochondrial glycerol-3-phosphate dehydrogenase activity . An earlier report, which claimed the cloning of S. cerevisiae GUT2, turned out to be incorrect; instead, GPD1 was cloned and partially sequenced . Biochemical evidence that the GUT2 gene sequenced by Rønnow and Kielland-Brandt indeed encoded mitochondrial glycerol-3-phosphate dehydrogenase, came later when it was shown that a gut2Δ mutant lacked mitochondrial glycerol-3-phosphate:ubiquinone oxidoreductase activity  and that mitochondria isolated from such a mutant did not oxidize glycerol 3-phosphate .
Physiological role and energetics of the glycerol-3-phosphate shuttle
The physiological importance of the glycerol-3-phosphate shuttle has been studied during growth of S. cerevisiae on various carbon sources [52, 80]. In the interpretation of the results from these studies, it should be taken into account that dissimilation of various carbon sources is accompanied by production of different amounts of cytosolic NADH.
A gut2Δ mutant produced substantial amounts of glycerol in ethanol-grown batch cultures, while the wild-type strain did not produce glycerol during growth on ethanol . When ethanol is consumed, cytosolic NADH may be produced by the cytosolic alcohol dehydrogenases Adh1p and Adh2p . Since glycerol production can serve as a redox sink , the phenotype of the gut2Δ mutant is indicative for an involvement of the glycerol-3-phosphate shuttle in reoxidation of cytosolic NADH by wild-type S. cerevisiae.
Lactate is oxidized directly by a lactate:cytochrome c oxidoreductase in the mitochondrial inner membrane, without concomitant production of cytosolic NADH . Dissimilation of the resulting pyruvate occurs inside the mitochondria by the pyruvate dehydrogenase complex and the citric acid cycle. Consistent with this information, glycerol production was not observed in pyruvate- or lactate-grown batch cultures of wild-type and gut2Δ strains . Even if pyruvate is dissimilated via the pyruvate dehydrogenase bypass (see Section 3), this does not generate cytosolic NADH, since the cytosolic acetaldehyde dehydrogenase is NADP+-linked .
In glucose-grown batch cultures, metabolism is mainly fermentative and, moreover, Gut2p is repressed by glucose [56, 107]. It is therefore not surprising that glycerol production in exponentially growing batch cultures on glucose was essentially the same in wild-type and gut2Δ S. cerevisiae. In aerobic glucose-limited chemostat cultures, glucose repression of GUT2 is alleviated . Under these conditions, the physiology of a gut2Δ mutant was almost indistinguishable from that of the wild-type strain (Fig. 4) . The only difference between the two strains was a slightly lower maximum specific growth rate of the gut2Δ strain . At low dilution rates, glucose metabolism was completely respiratory in both strains, indicating that all cytosolic NADH produced in glycolysis was reoxidized via mitochondrial respiration. Alcoholic fermentation set in at the same critical dilution rate in the wild-type and the gut2Δ mutant and glycerol formation was negligible in both strains. Apparently, if the glycerol-3-phosphate shuttle is involved in the oxidation of cytosolic NADH generated by glycolysis in wild-type cultures, this role can be taken over by other mechanisms (e.g. external NADH dehydrogenases) in a gut2Δ mutant.
Based on experiments with isolated mitochondria, Larsson et al.  proposed that NADH oxidation via the glycerol-3-phosphate shuttle is thermodynamically more efficient than via the external NADH dehydrogenases. In these experiments, a higher P/O ratio was found for the oxidation of glycerol 3-phosphate (P/O=1.7) than for oxidation of NADH (P/O=1.2) . This is unexpected, because both the external NADH dehydrogenases and Gut2p are directly coupled to the quinone pool and do not pump protons themselves. It is doubtful whether measurements on mitochondria isolated from S. cerevisiae are sufficiently accurate to compare efficiencies of free-energy transduction. Mitochondria isolated from S. cerevisiae are usually not well coupled. The highest respiratory-control value (i.e. oxidation rate in the presence of ADP divided by that in the absence of ADP) as measured by Luttik et al.  was 3.0 for oxidation of pure NADH. Larsson et al.  did not use pure NADH, but commercial NADH, which is contaminated with ethanol. At 5 mM of NADH, which these authors used, a contamination of 2.5 mM ethanol is introduced [98, 100]. This has two implications. First, the measured ‘NADH’ oxidation in fact partially represents ethanol oxidation, which proceeds via mitochondrial alcohol dehydrogenase Adh3p  and the internal NADH dehydrogenase Ndi1p . Secondly, the respiratory control values for oxidation of ethanol by yeast mitochondria are very low [4, 49, 98], suggesting that there is a substantial leak of protons in the presence of ethanol. This probably leads to a substantial underestimation of the P/O ratio with commercial NADH as the substrate. We have not found other evidence that indicates a different energetic efficiency of the glycerol-3-phosphate shuttle and the external NADH dehydrogenases.
Physiology of an nde1Δ nde2Δ gut2Δ triple mutant in glucose-limited chemostat culture
As discussed in this and Section 3, S. cerevisiae strains lacking either both external NADH dehydrogenases or a functional glycerol-3-phosphate shuttle were still capable of completely respiratory growth on glucose at low specific growth rates. To investigate a possible involvement of alternative systems that couple oxidation of cytosolic NADH to the mitochondrial respiratory chain, the phenotype of an nde1Δ nde2Δ gut2Δ triple mutant was studied in aerobic, glucose-limited chemostat cultures (Fig. 4). In the absence of alternative mechanisms, one would expect such cultures to exhibit high rates of ethanol and/or glycerol production.
At low dilution rates, an nde1Δ nde2Δ gut2Δ triple mutant produced no ethanol, but a large amount of glycerol (Fig. 4). At 0.10 h−1 a glycerol yield on glucose of 0.6 mol (mol glucose−1) was reached , which is even higher than the glycerol yield obtained with the classical sulfite process (approximately 0.35 mol mol−1; see Section 2.2) [69, 70]. Three models have been proposed that may explain why, at low specific growth rates, nde1Δ nde2Δ gut2Δ cultures produce glycerol, but not ethanol (Fig. 6). One explanation is an equimolar formation of glycerol and pyruvate in glycolysis (Fig. 6A). This process is redox-neutral and does not yield ATP. Pyruvate can be further dissimilated via the citric acid cycle and the respiratory chain, which should lead to net ATP synthesis. The theoretical glycerol production rate required to obtain a closed redox balance, was calculated to be 2.2 mmol g dry weight−1 h−1, which is sufficient to oxidize the NADH produced in biosynthesis and that produced in glycolysis at a dilution rate of 0.10 h−1. The measured glycerol production rate of the nde1Δ nde2Δ gut2Δ mutant was only 1.2 mmol g dry weight−1 h−1 at 0.10 h−1, indicating that an equimolar formation of glycerol and pyruvate cannot explain the physiology of the nde1Δ nde2Δ gut2Δ mutant .
A second explanation for the absence of ethanol from nde1Δ nde2Δ gut2Δ cultures, is depicted in Fig. 6B. In the cytosol glucose may be fermented to ethanol, which is further respired via intramitochondrial dehydrogenases. This mechanism can only account for the reoxidation of glycolytic NADH, since alcoholic fermentation is redox neutral. Therefore, a sufficient amount of glycerol should be produced to reoxidize the biosynthetic NADH generated in the cytosol. At low dilution rates, this mechanism is in quantitative agreement with the measured rate of glycerol production. However, it cannot explain the reduced glycerol yield observed at higher dilution rates . Above a dilution rate of 0.15 h−1, biomass yields and product yields of the nde1Δ nde2Δ gut2Δ mutant were poorly reproducible. However, the general trend was a decrease of the specific rate of glycerol production with increasing specific growth rate, accompanied by the onset of alcoholic fermentation .
A third option to explain the phenotype of the nde1Δ nde2Δ gut2Δ mutant is the involvement of an alternative redox shuttle (Fig. 6C). This shuttle might transfer cytosolic NADH to the mitochondrial matrix, where it can be oxidized by Ndi1p . Such a mechanism has an extra degree of freedom compared to the second mechanism. Depending on its activity and its affinity for NADH, a shuttle can compete with the glycerol-production route for oxidation of NADH. This might explain why the glycerol yield of the nde1Δ nde2Δ gut2Δ mutant on glucose is not constant with increasing specific growth rates. Below, several possible NADH shuttle mechanisms will be discussed.
In Section 3 it has been shown that an ethanol–acetaldehyde shuttle can export redox equivalents from the mitochondrial matrix to the cytosol and thus allow respiratory growth of an ndi1Δ mutant . S. cerevisiae has at least one mitochondrial alcohol dehydrogenase, Adh3p , and two cytosolic alcohol dehydrogenases, Adh1p and Adh2p . Since this shuttle does not require any transporters – ethanol and acetaldehyde diffuse freely across biological membranes – it is completely symmetrical. This implies that it cannot exchange NADH and NAD+ against a concentration gradient.
Based on the absence of a phenotype of an adh3Δ mutant in aerobic glucose-limited chemostat cultures grown at a dilution rate of 0.10 h−1, there is no evidence for a physiological role of the ethanol–acetaldehyde shuttle in respiring wild-type cultures. However, this shuttle may play a key role in the reoxidation of mitochondrial NADH in anaerobic cultures. In anaerobic cultures, S. cerevisiae cells contain only a few, large, branched mitochondria. These do not play a role in free-energy metabolism, but they are essential since important assimilatory reactions take place in the mitochondria [109, 110]. By applying metabolic flux analysis to anaerobic glucose-limited cultures, Nissen et al. have shown that these reactions result in a net formation of NADH in the mitochondrial matrix . In particular, the synthesis of glutamate from pyruvate, which involves the mitochondrial pyruvate dehydrogenase complex and isocitrate dehydrogenase, is a major source of this biosynthetic NADH (Fig. 7). Experimental evidence for involvement of the ethanol–acetaldehyde shuttle in anaerobic redox metabolism is that the maximum specific growth rate of an adh3Δ mutant in anaerobic, glucose-grown cultures is approximately 30% lower than that of the wild-type . The significant anaerobic growth rate that was still observed with the adh3Δ mutant may be related to a residual mitochondrial alcohol dehydrogenase activity measured in this strain . It is not known which gene or genes encode this residual mitochondrial alcohol dehydrogenase activity. Alternatively, other shuttle mechanisms may be involved in the reoxidation of mitochondrial NADH in anaerobic cultures.
The question remains whether the ethanol–acetaldehyde shuttle plays a role in the nde1Δ nde2Δ gut2Δ mutant to shuttle redox equivalents from the cytosol to the mitochondrial matrix . This can only occur if the cytosolic [NADH]/[NAD+] ratio is higher than the mitochondrial [NADH]/[NAD+] ratio. These ratios are not known in S. cerevisiae. Even if in the wild-type the mitochondrial [NADH]/[NAD+] ratio is much higher than the cytosolic [NADH]/[NAD+] ratio, this may be reversed in the nde1Δ nde2Δ gut2Δ mutant.
Especially in plant mitochondria, the activity of a malate–oxaloacetate shuttle (Fig. 8) seems very important [111, 112]. This shuttle consists of a mitochondrial and a cytosolic NAD+-dependent malate dehydrogenase and a transporter, which catalyzes malate–oxaloacetate exchange across the mitochondrial inner membrane. Depending on the mechanism of transport this shuttle may be reversible . Since the malate–oxaloacetate shuttle was first proposed, the occurrence of oxaloacetate transport across the mitochondrial inner membrane has been a subject of much debate and experimentation. It turned out that rat liver mitochondria do transport oxaloacetate, while rat heart mitochondria do not transport this compound [114, 115]. Also mitochondria isolated from potato tuber and pea, transport oxaloacetate at high rates [112, 116]. In these cells, oxaloacetate is transported across the mitochondrial membrane in exchange for another compound, such as phosphate, malate or malonate [112, 115].
The question whether a malate-oxaloacetate shuttle may operate in S. cerevisiae has long been elusive. Malate can be transported by the dicarboxylate carrier Dic1p and mitochondrial and cytosolic malate dehydrogenase are encoded by MDH1 and MDH2 [118, 119], respectively. Only recently, it has been demonstrated that S. cerevisiae mitochondria transport oxaloacetate via Oac1p . When reconstituted in proteoliposomes, Oac1p transports oxaloacetate, malonate, malate, phosphate, sulfate and other compounds alternatively via a uniport or an exchange mechanism . The rates of exchange are much higher than the rates of uniport . Exchange of malate and oxaloacetate by Oac1p has not been demonstrated directly, but since both compounds are substrates of the carrier, it seems probable that they are exchanged in vivo. Under respiratory growth conditions, the malate–oxaloacetate shuttle can only work in the direction of import of NADH into the mitochondria, since mitochondrial malate dehydrogenase also fulfils a function in the citric acid cycle where it oxidizes malate to oxaloacetate. Under anaerobic conditions, it is possible that the shuttle works in the reverse direction, because mitochondrial malate dehydrogenase can work in the reverse direction then . It seems unlikely, however, that the malate–oxaloacetate shuttle is very important during anaerobic growth on sugars, since the anaerobic expression of MDH2, encoding cytosolic malate dehydrogenase, is very low . Furthermore the Mdh2p protein is rapidly degraded in the presence of excess glucose .
In cases where the mitochondrial membrane is impermeable to oxaloacetate, such as in rat heart mitochondria , the malate-oxaloacetate shuttle can be extended to a malate–aspartate shuttle, as first proposed by Borst (Fig. 9) . In this shuttle, mitochondrial oxaloacetate is converted to aspartate by aspartate aminotransferase. Aspartate leaves the mitochondria in exchange for glutamate and is reconverted to oxaloacetate as depicted in Fig. 9. Except for the malate dehydrogenase isoenzymes discussed above, the malate–aspartate shuttle requires a mitochondrial and a cytosolic aspartate aminotransferase.
It has long been recognized that the malate–aspartate shuttle cannot pump NADH against a concentration gradient, unless it is coupled to a free-energy-dissipating process . Mammalian cells tend to have a higher NADH/NAD+ ratio in the mitochondria than in the cytosol . This implies that, in order to couple oxidation of cytosolic NADH – originating from glycolysis or lactate oxidation – to mitochondrial complex I, mammalian redox shuttles should be capable of operating against a concentration gradient. Therefore, a symmetrical malate–aspartate shuttle seemed of limited physiological importance initially . This view drastically changed when the transporters were characterized. In mammalian mitochondria, exchange of malate and α-ketoglutarate is electroneutral at physiological pH and catalyzed by a specific, phthalonic acid-sensitive transporter [122, 123]. However, export of aspartate from the mitochondria in exchange for glutamate is electrogenic and depends on the electrical component of the proton-motive force [123, 124]. The latter transport process allows, therefore, import of NADH into the mitochondria against a concentration gradient, at the expense of dissipation of the proton-motive force [51, 125, 126].
At present, it is unclear whether the malate–aspartate shuttle plays a role in S. cerevisiae. Previously it was thought that this shuttle could not work in S. cerevisiae, since only a cytosolic aspartate aminotransferase was known to be present . This cytosolic aspartate aminotransferase is encoded by AAT2 [127–129] and, depending on the growth conditions, it remains in the cytosol or it is targeted to the peroxisomes . Later, a second gene, AAT1, was cloned and sequenced . The sequence of Aat1p is more homologous to other mitochondrial aspartate aminotransferases than to other cytosolic aspartate aminotransferases . Moreover, it contains a putative mitochondrial targeting sequence . Mitochondrial localization of Aat1p has, however, not been demonstrated directly. It is unknown whether the transporters that are required for the malate–aspartate shuttle, are present in S. cerevisiae. To our knowledge neither the neutral α-ketoglutarate–malate antiporter nor the electrogenic aspartate–glutamate antiporter have been characterized in yeast. Yeast mitochondria have been reported to transport α-ketoglutarate, but based on specific inhibitor sensitivity it was concluded that the yeast α-ketoglutarate carrier differs from the corresponding mammalian system . Furthermore, the two genes in the S. cerevisiae genome that cluster with the bovine α-ketoglutarate–malate carrier on a phylogenetic tree , have been identified as the dicarboxylate carrier DIC1 and the oxaloacetate carrier OAC1. When reconstituted in proteoliposomes the dicarboxylate carrier Dic1p and the succinate–fumarate transporter Arc1p catalyze the exchange of α-ketoglutarate and malate at a very low rate. However, the physiological significance of this reaction is uncertain as these transporters have a much higher affinity for other substrates [133, 134].
Two independent studies have addressed the question whether the malate–aspartate shuttle functions in vivo in S. cerevisiae [99, 135]. Wills and co-workers studied the effect of two transaminase inhibitors on growth of S. cerevisiae. Aminooxyacetate and cycloserine both inhibit the aspartate aminotransferases. In mammalian cells, interpretation of the physiological effects of these inhibitors is complicated by the fact that they also inhibit gluconeogenesis. This is due to the mitochondrial localization of pyruvate carboxylase in mammalian cells [136–139]. If oxaloacetate itself cannot be exported to the cytosol, it is first converted to aspartate, which subsequently leaves the mitochondria and enters gluconeogenesis . In S. cerevisiae, pyruvate carboxylase is cytosolic [93, 140, 141], so transaminase inhibitors should not interfere with gluconeogenesis. However, aspartate aminotransferase is expected to be essential for growth of S. cerevisiae on minimal medium with ammonium as the sole carbon source, because it is required to synthesize aspartate. Therefore, it is surprising that growth on minimal medium with ammonium and glucose or acetate was possible at all in the presence of aminooxyacetate or cycloserine . Growth on minimal medium with glycerol, ethanol or lactate was inhibited by these transaminase inhibitors . It is unclear, however, whether this is due to impaired import of NADH into the mitochondria by the malate–aspartate shuttle. First, dissimilation of lactate should not result in cytosolic NADH production (see Section 4.1.2). Secondly, there are alternative routes for oxidation of cytosolic NADH, as discussed above, which may be expected to bypass the malate–aspartate shuttle in the presence of inhibitors. Based on these considerations, these inhibitor studies with aminooxyacetate and cycloserine cannot be interpreted as proof for in vivo activity of a malate–aspartate shuttle in S. cerevisiae.
Recently, the significance of the malate–aspartate shuttle in S. cerevisiae has been investigated by deleting MDH2, the gene encoding cytosolic malate dehydrogenase . Mutants lacking either MDH2 or NDE1 (encoding external NADH dehydrogenase) still grew on ethanol, albeit much slower than the wild-type strain, while an mdh2Δ nde1Δ double mutant did not grow at all on ethanol. This was interpreted as evidence for the operation of a malate–aspartate shuttle in ethanol-grown cultures . However, the impaired growth of the mdh2Δ strain may also be indicative for other physiological roles of Mdh2p, for example an involvement in the glyoxylate cycle . If the primary physiological role of Mdh2p were in oxidation of cytosolic NADH via the malate–aspartate shuttle, the glycerol-3-phosphate shuttle (which is active in ethanol-grown cultures ) may be expected to take over this role in during growth of an mdh2Δ nde1Δ mutant. Therefore, also the phenotype of this mutant does not provide conclusive evidence on the operation of the malate–aspartate shuttle in S. cerevisiae.
A malate–pyruvate shuttle, based on malic enzyme, has been described in pancreatic islets of rat . In these cells, the shuttle oxidizes NADH in the mitochondrial matrix and releases NADPH in the cytosol. It is based on the action of mitochondrial malate dehydrogenase and pyruvate carboxylase and cytosolic malic enzyme. In S. cerevisiae, a single gene, MAE1 encodes malic enzyme, which is localized in the mitochondrial matrix . The yeast enzyme utilizes NADP+ as a coenzyme  and has also been reported to use NAD+. In contrast to the situation in mammalian cells [93, 140, 141], the two isoenzymes of pyruvate carboxylase in S. cerevisiae, encoded by PYC1 [146, 147] and PYC2 [141, 148], are exclusively localized in the cytosol. Therefore, if the malate–pyruvate shuttle is active in S. cerevisiae, it should transfer redox equivalents from the cytosol to the mitochondrial matrix, rather than in the reverse direction (Fig. 10).
If this shuttle indeed works as depicted in Fig. 10, malate has to enter the mitochondrion and pyruvate has to be transported from the mitochondrial matrix to the cytosol. Transport of malate into the mitochondrion is catalyzed by the dicarboxylate carrier  encoded by DIC1 [133, 150]. Mitochondrial pyruvate transporters have been purified from different organisms [151–153], but until now the gene encoding the S. cerevisiae mitochondrial pyruvate transporter has not been identified. At any rate, during respiratory growth on sugars the pyruvate flux is normally directed into the mitochondria rather than backward. If a malate–pyruvate ‘shuttle’ is active in S. cerevisiae it can, therefore, not work as a full shuttle. Rather, it may provide an alternative means of pyruvate transport into the mitochondria, with concomitant transport of NAD(P)H into the mitochondria and at the expense of one ATP per NAD(P)H transported (Fig. 10).
In spermatozoa of various organisms, redox equivalents are transported into the mitochondria by a lactate–pyruvate shuttle. This shuttle consists of mitochondrial and cytosolic isoenzymes of NAD+-linked lactate dehydrogenase and of carriers that transport lactate and pyruvate across the mitochondrial inner membrane [154, 155]. A lactate–pyruvate shuttle cannot operate in S. cerevisiae as this yeast only has cytochrome c-linked lactate dehydrogenases, which are directly coupled to the respiratory chain and are required for growth on lactate as the sole carbon source [16–18].
Two other well-known redox shuttles, which are thought to be functionally connected in mammalian cells, are the malate–citrate shuttle and the fatty acid shuttle [51, 156]. We will not discuss these shuttles in detail, since they cannot operate in S. cerevisiae. The malate–citrate shuttle requires an active ATP:citrate lyase, which is absent from S. cerevisiae. A mitochondrial fatty-acid shuttle, transferring redox equivalents from the cytosol to the mitochondrial matrix, requires that the β-oxidation of fatty acids to occur in the mitochondrial matrix . In S. cerevisiae, however, β-oxidation occurs exclusively in the peroxisomes .
Physiological significance of parallel pathways
This review demonstrates that quantitative analysis of defined mutants is a useful tool to identify components of the S. cerevisiae metabolic network. This approach has led to the conclusion that this yeast harbors various parallel pathways for reoxidation of the NADH generated in assimilatory and dissimilatory reactions. In studies with mutants, the physiological significance of a pathway in wild-type cells is inferred from the absence or presence of a phenotype in null mutants. Obviously, this is a less than subtle approach. For example, the absence of a phenotype in aerobic, glucose-limited cultures of a gut2Δ strain does not prove that the glycerol-3-phosphate shuttle is physiologically insignificant in wild-type cultures grown under these conditions. The next difficult step in research on redox metabolism will therefore have to be the direct measurement of the contribution of each of the parallel pathways in wild-type cells grown under various carefully defined conditions. With respect to cultivation conditions, it is relevant to note that most physiological data discussed in this review have been obtained either in exponentially growing batch cultures or steady-state chemostat cultures. It is well conceivable that pathways which are apparently redundant under these pseudo-steady state conditions may play a key role under more dynamic conditions, for instance in the physiological response to fluctuating oxygen- or electron-donor availability.
In vivo analysis of the contribution of parallel pathways under steady-state and dynamic cultivation conditions requires the accurate analysis of relevant intracellular fluxes, metabolites and coenzymes. Techniques for measurement of intracellular metabolites in S. cerevisiae cultures have become more and more reliable as a result of major improvements of rapid-sampling and quenching techniques, as well as the development of sensitive and specific analytical methods [14, 159–161]. As illustrated in the preceding sections, metabolic compartmentation is a key issue in the redox metabolism of eukaryotic cells. A quantitative understanding of redox metabolism in S. cerevisiae therefore not only requires measurements of the overall cellular content of coenzymes and metabolites, but also of their concentrations in the cytosol, mitochondria and any other relevant compartments. Unfortunately, reliable and generally applicable methods to determine metabolite concentrations in different subcellular compartments of S. cerevisiae are still lacking. For mammalian cells, local [NADH]/[NAD+] ratios have been estimated indirectly by measuring the product–substrate ratios of NAD+-dependent dehydrogenases confined to specific compartments . If these dehydrogenases work very close to equilibrium and if they are strictly confined to a certain compartment, the free [NADH]/[NAD+] ratio in this compartment can be calculated from the equilibrium constants. The existence of isoenzymes in different compartments and the fact that it is often unclear how far from equilibrium they are, precludes a general applicability of this method, however. The development of reliable methods to measure subcellular metabolite concentrations clearly is one of the most important challenges for quantitative physiological studies on eukaryotic cells.
Metabolic engineering of yeast redox metabolism
As discussed in Section 2.2, fundamental knowledge on the S. cerevisiae redox metabolism has been successfully used in metabolic engineering strategies for reducing as well as increasing glycerol production. These interventions, which were chiefly aimed at changing the stoichiometry of redox metabolism, involved only one or a few structural genes. A major challenge, primarily in biomass-directed applications of S. cerevisiae, such as the production of baker’s yeast and heterologous proteins, is to increase the respiratory capacity and thereby to minimize the production of low-molecular-mass metabolites, such as ethanol and acetate. The mitochondrial respiratory chain consists of many enzymes and respiratory complexes. Since control of flux in metabolic pathways is often distributed over many or all enzymes involved, it is highly unlikely that the respiratory flux can be modified by stimulating the synthesis of one or a few components of the respiratory chain [162–164]. Probably a more promising approach is modulation of the expression of regulatory genes that coordinately regulate expression of many genes involved in respiration.
The potential of modifying the expression of global regulators has recently been demonstrated. For glucose repression of genes involved in respiration, the regulatory proteins Mig1p and Mig2p are of central importance, while derepression of these respiratory genes is mediated by the Hap2/3/4/5p complex . The observation that HAP2 and HAP3 seem to be expressed constitutively and that expression of HAP4 is induced upon a shift from glucose to a non-fermentable carbon source, may suggest that Hap4p is the activator of the complex . Deletion of MIG1 and MIG2 had subtle, but significant, effects on the distribution of fluxes among fermentation and respiration. For example, in glucose-limited cultures, the critical dilution rate at which alcoholic fermentation sets in, was increased . Hap4p overproduction has been shown to decrease ethanol production rate and to increase biomass yields of S. cerevisiae grown on excess glucose by as much as 40%.
Although modification of one or more transcriptional regulators is likely to be a powerful and a widely applicable tool for redirection of metabolic fluxes, one has to be careful not to make the same mistake with regulatory networks as was previously made with metabolic pathways. Just as there is usually not a single rate-limiting step in metabolic pathways, control of regulatory networks does probably not reside in a single ‘master switch’ in most cases. Consistent with this line of thought, it has been found that the impact of HAP4 overexpression on flux distribution strongly depends on the growth conditions (A.J.A. van Maris et al., manuscript in preparation). A quantitative understanding of regulatory networks will, therefore, be of great importance for biotechnology. The experimental approaches developed for analysis of metabolic networks, such as the quantitative analysis of defined mutants under controlled cultivation conditions, will prove to be equally important in the unravelling of regulatory networks.
Research on redox metabolism in yeasts by our group is financially supported by the Dutch Ministry of Economic Affairs (EET program) and by the Delft DIOC-6 program ‘Mastering the Molecules of Manufacturing’. We thank Prof. J.J. Heijnen and our colleagues at DSM Bakery Ingredients for stimulating discussions. We would like to thank Dr. Luigi Palmieri for interesting discussions and for information about mitochondrial transporters.