NADH-dependent biosensor in Saccharomyces cerevisiae: principle and validation at the single cell level

The relative concentrations of NADH/NAD⁺ redox couple determine the thermodynamic feasibility of more than 100 reactions in cellular metabolism which places it at the core of the metabolic network of every organism including Saccharomyces cerevisiae (Förster et al. [2003]). Efficient tools to measure the impact of redox enzymes or environmental perturbations on the cellular NADH levels are therefore needed (Murray et al. [2011]). However quantification of redox couples in cells is complicated by several factors: the NADH and NAD⁺ are compartmentalized in S. cerevisiae and the cofactors are to a large extent in a protein bound form (Veech et al. [1969]; Bücher et al. [1972]). Fluorescence imaging enables the distinction between free and bound redox couples, but not between NADH and NADPH (Zhang et al. [2002]). In mammalian cells, coupled enzymatic reactions have been used to determine the ratio between the cofactors NADH/NAD⁺ (Williamson et al. [1967]; Theobald et al. [1997]). A similar approach has been applied for S. cerevisiae where mannitol-1-phosphate 5-dehydrogenase (M1PDH) was introduced and the cytosolic free NADH/NAD⁺ ratio was determined by the ratio between fructose-6-phosphate (F6P) and mannitol-1-phosphate (M1P) and the equilibrium constant for the reaction M1P + NAD ↔ F6P + NADH + H⁺ (Canelas et al. [2008]). The approaches are however often tedious and not applicable for high-throughput screening.

In S. cerevisiae, the cytosolic enzyme glycerol-3-phosphate dehydrogenase (GPD) 2 encoded by the GPD2 gene plays a central role in redox metabolism. Under anaerobiosis, GPD2 expression increases as the need for cytosolic reoxidation of NADH increases because Gpd2p acts as redox sink via the NADH-coupled reduction of dihydroxyacetone phosphate (DHAP) to glycerol 3-phosphate (G3P). G3P is further converted to glycerol (Eriksson et al. [1995]; Ansell et al. [1997]). The need for NADH reoxidation is thought to be linked, indirectly or indirectly, to the NADH/NAD⁺ ratio (Ansell et al. [1997]). Glycerol production can also be initiated by the isoenzyme Gpd1p encoded by GPD1. However Gpd1p and Gpd2p are expressed under different growth conditions, as illustrated by their different cellular localization (Valadi et al. [2004]). The physiological role of Gpd1p is also different from that Gpd2p as GPD1 is induced under osmotic stress (Blomberg and Adler [1989]). Although the genes can complement each other's functions, a double deletion renders the yeast cell osmosensitive as well as incapable of anaerobic growth, due to the inability to reoxidize cytosolic NADH arising from biomass formation (van Dijken and Scheffers [1986]; Nissen et al. [1997]).

We used this knowledge to design a GFP-based biosensor for cytosolic NADH/NAD⁺ levels by coupling the GPD2 promoter (GPD2 p) to the yEGFP3 gene encoding yeast enhanced green fluorescent protein, an established yeast reporter of the GFP class 2 (Cormack et al. [1996]; Cormack et al. [1997]; Zacharias and Tsien [2005]). The applicability of fluorescent proteins as reporters has previously been demonstrated through the measurement of various user-specified cell properties. For example, GFP has been used for protein localization (Valadi et al. [2004]), intracellular pH measurement (Valkonen et al. [2013]), for monitoring cell growth mode and cell membrane robustness (Carlquist et al. [2012]). Dardalhon et al. ([2012]) also implemented a fluorescent glutathione redox reporter in S. cerevisiae using a yellow fluorescent protein sensor that reports on the nuclear and cytosolic glutathione redox state i.e the glutathione/glutathione disulfide (GSH/GSSG) state, which is NADPH dependent.

The functionality of the NADH/NAD⁺-ratio biosensor developed in this work was first evaluated by introducing it into S. cerevisiae strains with reduced NADH reoxidation capacity due to the deletion of Gpd1,2p encoding genes. The biosensor potential was then validated by evaluating the impact on cell fluorescence of acetoin addition and of the introduction of xylose reductases with different selectivities for NADH.

Microbial populations exist in dynamic equilibrium with constantly changing conditions, which requires fast and fluid regulatory networks for optimizing cell functions. Among these, an accurate metabolic response is fundamental for the cell to maintain redox homeostasis. In S. cerevisiae, it is well known that a skewed ratio of cytosolic NADH/NAD⁺ can be compensated by glycerol production regulated by a fine-tuned expression of the GPD2 gene (Ansell et al. [1997]; Björkqvist et al. [1997]). This served as basis for the construction of our in vivo redox state reporter that enables green fluorescence to be used as an indicator of the cells ability to reoxidize NADH, i.e. to the NADH/NAD⁺ ratio of the cell. This is, to our knowledge, the first biosensor to use the GPD2 promoter as sensor for the cytosolic NADH/NAD⁺ cofactor balance.

By placing a GFP gene downstream of the GPD2 promoter, a direct relationship between green fluorescence intensity (FI) and glycerol formation or NADH accumulation was observed. In strains carrying double deletion of the GPD genes, the FI was significantly higher (21.5-fold) than for the control strain, indicating a strong activation of the GPD2 promoter, and thereby verifying the lack of other enzymes to counteract the lacking NADH oxidation functionality. In parallel, FIs were in the same order of magnitude in single GPD deletion strain and in the control strain, also confirming the previously observed compensatory pattern between Gpdp isoenzymes in order to maintain NAD⁺ regeneration via glycerol production (Ansell et al. [1997]). The FI signal of the gpd1Δ strain was also slightly higher than the FI of the gpd2Δ strain and the control strain: 25% and 50%, respectively, as expression of the GDP2, gene whose promoter is used in the constructed system, should be higher in the gpd1Δ strain, as a consequence of GPD2 compensating for the lacking GPD1 functionality. Finally the high responsiveness of the constructed reporter supported the hypothesis that GPD2 expression was directly connected to the cell need to oxidize NADH via glycerol formation. It cannot be excluded that FI may be correlated to more than the need for NADH oxidation via glycerol formation. Growth rate and the capacity to re-oxidise NADH are closely related and thus FI will always be indirectly correlated to growth rate. As an example of this, the rescued growth rate of the gpd1Δgpd2Δ deletion strain by increased NADH oxidation led to significantly lower FI. However the oxygen concentration, which reflects the cells ability to replace the need for glycerol formation by aerobic respiration, was probably a more important parameter in that strain. Cultivating cells under oxygen-limited conditions could also have had implications for the FI measurements as fluorescent proteins require oxygen to mature. However, GFP expression in S. cerevisiae could previously be measured in cells that were cultivated under strict anaerobic conditions, most probably due to exposure to oxygen during the aerobic handling of samples. (Hebisch et al. [2013]) previously demonstrated that the average maturation time in E. coli for GFP is around 6 min, which is enough time for GFP to mature during the preparation for the FI measurement in the present study.

There was a clear difference between placing the reporter cassette on a multicopy plasmid or on a plasmid that gets integrated in the chromosome. Copy number per cell varies between 50 and 200 when using YEp vectors as opposed to one when using YIp vectors (Romanos et al. [1992]). So placing the GFP-based reporter construct on YEp plasmids led to a significantly higher FI than when using YIp-based construct. However, more defined fluorescence peaks were obtained when using the YIp system as opposed to the YEp system. This is in accordance with previous findings illustrating the stability issues of YEp-based systems compared to YIp-based systems (Ishii et al. [2009]) and indicates lower degree of cell population heterogeneity in chromosomally integrated biosensors as compared to the YEp-based systems, which may later facilitate the distinction between any underlying subpopulations. Promoters can also have different levels of noise in the respective gene expression (Blake et al. [2003]), which can lead to a higher heterogeneity of the signal. The low degree of heterogeneity in the YIp-based system further stresses the suitability of the GPD2 promoter for analysis of populations at the single cell level.

A 2-fold lowering of the GFP levels was observed when acetoin was added to GPD double deletion strains, which can be attributed to the induced NADH-dependent reduction of acetoin (Gonzalez et al. [2000]). Acetoin or acetaldehyde, that is reduced to ethanol with the concurrent oxidation of NADH, have previously been used to demonstrate the lack of alternative cytosolic reductive reactions in double deletion strain (Lidén et al. [1996]; Ansell et al. [1997]). In these studies, the involved oxidation reactions were efficient enough to recover sufficient redox balance for the cells to start growing. By monitoring FI instead, we wanted to investigate whether the constructed biosensor could monitor less drastic changes in the redox balance. To validate this hypothesis, we introduced XR enzymes with slightly different catalytic activity for NADH oxidation, namely two xylose reductases having NADH/NADPH selectivity ratios of 0.04 and 0.80 respectively (Runquist et al. [2010]). As expected, the biggest difference in growth was observed in the double deletion strains between the control and the strain carrying the mutXR, since more carbon was made available for growth when xylitol formation from xylose was used as redox sink instead of glycerol formation from glucose. This is also part of the explanation for the lower glycerol formation in the TMB4141 and TMB4143 as opposed to the TMB4140, as the cells can produce xylitol as a redox sink in addition to glycerol. A higher xylitol production could have been expected for the mutXR than wtXR in the gpd1Δgpd2Δ strain, if only considering the NADH-dependent activity. However wtXR also has a significant activity with NADPH as a cofactor (Rizzi et al. [1988]), which led to a non-negligible NADH-independent xylitol production. The maximum difference in OD was 2.5-fold after 48 hours of growth whereas the difference between the control strain and the strain carrying the wtXR was only 1.5 fold. So, in theory, the reductases could be distinguishable by repeating batch cultivations and using growth as selection. Assuming that after 48 hours one cell of TMB4144 (double deletion mutXR) grows to 2.5 times the cell density of the TMB4140 (double deletion no XR), it would take 27 batch cultivations of 48 hours each before one TMB4144 cell would outgrow 10⁹ TMB4140 cells by a factor of 1:100 (according to the following calculation: (1/2.5)²⁷∙10⁹ ≈ 0.01). Using fluorescence instead, the difference was clearly visible after a few hours. Increased xylitol formation and lower fluorescence were observed in strains carrying the wtXR and the mutXR. However, a clear difference in FIs was observed between the two strains whereas differences in xylitol levels were not that significant. The results, although qualitative, indicate that fluorescence was the more sensitive and faster method to discriminate between the two XR enzymes and their difference in NADH-catalyzed xylitol production and consequently their ability to affect the cytosolic NADH/NAD⁺ ratio. It also opens the possibility for high-throughput screening of enzymes with increased selectivity for NADH. In our cases, similar XR levels were measured (data not shown). However, in future screening, an enzyme with lower specificity for NADH, but expressed at a higher level and giving a higher activity/per cell for NADH oxidation will lead to a lower FI. This limitation in the screening will, however, occur in any method based on the NADH pool and not only in the FI-based one. Another element that can be further considered when using a GFP-based reporter system for screening or analysis of the NAD(H) cofactor imbalance is the half-live of the chosen protein. The yEGFP3 protein used in the present study has been reported to have a half-life of about 7.5 hours (Mateus and Avery [2000]), and indeed the intracellular GFP level did not respond rapidly to dynamic changes in external conditions. This implies that rapid changes in NAD(H) cofactor balancing may not be monitored with the developed system. However alternative GFP variants with shorter half-life, such as the GFP fused with a fragment of the CLN2 (yeast G1 cyclin) protein having a half-life of 30 minutes (Mateus and Avery [2000]), could then be used instead.

Lately, much attention has been drawn in using FACS for screening large genetic space (Dietrich et al. [2010]) and the technique has already been applied for the screening of NADPH-dependent alcohol dehydrogenases in E. coli (Siedler et al. [2014]). However, the combination of a NADH biosensor and single-cell analysis has, to our knowledge, not yet been implemented in S. cerevisiae. Our tool offers the first single-cell high-throughput NAD(H) cofactor balance sensor in S. cerevisiae. In contrast to previously reported methods where extraction of intracellular NAD(H) metabolites was performed (Canelas et al. [2008]), the method presented in this work is faster and allows for analysis with single-cell level resolution. Therefore the constructed tool offers possible applications in the selection of new NADH-dependent enzymes based on single cell sorting. Additionally it could be applied in monitoring the impact of various metabolic engineering strategies on the cytosolic NAD(H) cofactor balance of the cell.