NADPH-generating systems in bacteria and archaea

Reduced nicotinamide adenine dinucleotide phosphate (NADPH) is an essential electron donor in all organisms. It provides the reducing power that drives numerous anabolic reactions, including those responsible for the biosynthesis of all major cell components and many products in biotechnology. The efficient synthesis of many of these products, however, is limited by the rate of NADPH regeneration. Hence, a thorough understanding of the reactions involved in the generation of NADPH is required to increase its turnover through rational strain improvement. Traditionally, the main engineering targets for increasing NADPH availability have included the dehydrogenase reactions of the oxidative pentose phosphate pathway and the isocitrate dehydrogenase step of the tricarboxylic acid (TCA) cycle. However, the importance of alternative NADPH-generating reactions has recently become evident. In the current review, the major canonical and non-canonical reactions involved in the production and regeneration of NADPH in prokaryotes are described, and their key enzymes are discussed. In addition, an overview of how different enzymes have been applied to increase NADPH availability and thereby enhance productivity is provided.


Introduction
Reduced nicotinamide adenine dinucleotide phosphate (NADPH) is an essential electron donor in all eukaryotes, bacteria, and archaea. Not only is NADPH vital for the generation of reactive oxygen species (ROS) (Foreman et al., 2003;Brown and Griendling, 2009;Bylund et al., 2010;Nakamura et al., 2012) and the anti-oxidative defense mechanisms of most organisms (Nordberg and Arnér, 2001;Minard and Mcalister-Henn, 2005;Singh et al., 2008), most importantly, it is also the driving force of most biosynthetic enzymatic reactions, including those responsible for the biosynthesis of all major cell components, such as DNA and lipids (Arnér and Holmgren, 2000;Nordberg and Arnér, 2001;Hügler et al., 2002;Koh et al., 2004;Minard and Mcalister-Henn, 2005;Singh et al., 2008;Miller and Auchus, 2011). Owing to this essential role in biosynthesis, NADPH availability has been of major interest to industry (Papagianni, 2012;Lee et al., 2013b). Many natural products of industrial importance are complex secondary metabolites, the production of which often involves NADPH-dependent enzymes. To synthesize such products using purified enzyme systems in vitro would require the addition of huge amounts of NADPH in order to sustain production. From an industrial point of view, this would be too expensive. Hence, in situ NADPH regeneration from its oxidized counterpart (NADP + ) is required (Chenault et al., 1988;Wandrey, 2004;Wichmann and Vasic-Racki, 2005;Zhang and Xu, 2010;Uppada et al., 2014).
Regeneration of NADPH in situ can be achieved by various strategies, including chemical, electrochemical, photochemical, and enzyme-based methods (Chenault and Whitesides, 1987;Zhao and Van Der Donk, 2003;Wichmann and Vasic-Racki, 2005;Liu and Wang, 2007;Zhang and Xu, 2010;Uppada et al., 2014). NADPH can be regenerated enzymatically by complementing the in vitro system with additional enzymatic reactions or by using substrate-coupled reaction systems. The latter system employs enzymes that use both NADP + and NADPH and that are able to catalyze the synthesis of the desired product from one substrate and cofactor regeneration with another substrate (Chenault et al., 1988;Van Der Donk and Zhao, 2003;Liu and Wang, 2007). However, reduced productivity compared to systems without in situ regeneration and problems associated with enzyme stability make these options unattractive.
Microbial in vivo production systems also provide in situ NADPH regeneration and have several advantages when compared to in vitro systems. For example, microbes are able to grow on inexpensive renewable feedstocks that provide the organisms with reductant for the regeneration of NADPH. They also contain numerous pathways, involving stable and highly specific enzymes, thus obviating the need for expensive enzyme purification. In addition, our knowledge of natural metabolic pathways is rapidly advancing, allowing for rational design toward product formation (Chemler et al., 2010;Siedler et al., 2011;Papagianni, 2012;Lee et al., 2013b). Therefore, it is not surprising that microbial conversion is the preferred method for the synthesis of a range of products.
With the possibility of engineering microbial metabolism to facilitate product formation, it became clear that NADPH availability remains a major hurdle in the efficient generation of many products. These products range from medicinal compounds (Chemler et al., 2010;Siedler et al., 2011;Zhao et al., 2011) and (essential) amino acids (Becker et al., 2007;Papagianni, 2012) to molecules used as biofuels (Asadollahi et al., 2009;Kim et al., 2011;Peralta-Yahya et al., 2012) and building blocks for biodegradable plastic (Kabir and Shimizu, 2003). Given its involvement in a multitude of crucial biological functions and its importance in biosynthesis, NADPH is without question an essential molecule. Hence, a key question arises: what are the major NADPH-generating reactions and systems?
Traditionally, the dehydrogenase reactions of the oxidative pentose phosphate pathway (oxPPP), the Entner-Doudoroff (ED) pathway, and the isocitrate dehydrogenase step of the tricarboxylic acid (TCA) cycle have been considered the major sources of NADPH. However, the importance of other NADPHgenerating enzymes, such as transhydrogenases, glucose dehydrogenases, and non-phosphorylating glyceraldehyde 3phosphate dehydrogenase (GAPN), is becoming clear, indicating that the traditional view is over-simplistic Matsubara et al., 2011;Brasen et al., 2014).
In this review, we describe the major canonical and non-canonical biochemical mechanisms that are involved in the production and regeneration of NADPH in prokaryotes and discuss the key enzymes involved. We have divided the mechanisms into those that are directly coupled to central carbon metabolism and those that are not (Table 1). In addition, we briefly address how different enzymes have been applied to increase NADPH availability and thereby enhance NADPHdependent biotransformation processes.

Biosynthesis of NADP +
To maintain a sufficient quantity of NADP + for the generation of NADPH, NADP + biosynthesis is essential. While the necessity for NADP + synthesis during cell proliferation is clear, the requirement for NADP + synthesis in non-dividing cells might not be obvious. NADP(H) is generally thought of solely as a redox carrier that facilitates the transfer of electrons between two redox couples, a role that does not account for the need for constant resynthesis. The constant need to resynthesize NADP + arises from its participation in other crucial biological processes (Agledal et al., 2010). In eukaryotes, for example, NADP + serves as the substrate in the synthesis of nicotinic acid-adenine dinucleotide phosphate (NAADP), an important intracellular Ca 2+ -mobilizing messenger (Lee, 1997;Chini and De Toledo, 2002;Churamani et al., 2004;Patel et al., 2011). Although NAADP has also been found in bacteria (Churamani et al., 2004), its physiological relevance in these organisms remains to be established.
The enzymes responsible for NADP + resynthesis, best studied in eukaryotes, mainly include NAD(P) + nucleosidase (Vu et al., 1996;Ying, 2008), NADP + phosphatase, and NADPH phosphatase. The enzymes in prokaryotes are less well-studied, but similar activities and proteins have been found in this domain (Mather and Knight, 1972;Everse et al., 1975;Davis, 1980;Penyige et al., 1996;Pollak et al., 2007;Kawai and Murata, 2008;Ghosh et al., 2010). NADP + phosphatase and NADPH phosphatase, together with NAD(H) kinase, regulate the intracellular balance of NAD(H) and NADP(H) (Kawai and Murata, 2008). Like NADPH, NADH mainly serves to transfer electrons from one molecule to another. However, unlike NADPH, NADH is primarily involved in catabolic reactions (Ying, 2008). In addition to its redox function, NAD + serves as a substrate for mono-and poly-ADP ribosylation, participates in histone deacetylation, and contributes to the production of the signaling molecule cyclic ADP-ribose. Most of these reactions have been characterized in eukaryotes, but the ribosylation reactions also play a role in toxin production by pathogenic bacteria (Ziegler, 2000;Pollak et al., 2007). The consumption of NADP + is thus connected to the consumption of NAD + and to the regulation of various major biological activities such as DNA repair, gene expression, apoptosis, nitrogen fixation, and calcium homeostasis (Ziegler, 2000;Pollak et al., 2007;Kawai and Murata, 2008;Ying, 2008;Agledal et al., 2010).
Given that the intracellular balance between NAD(H) and NADP(H) is regulated via the addition or removal of a phosphate group by NAD(H) kinase or NADP(H) phosphatase, respectively, it is clear that the biosynthesis of NAD + plays a crucial role in the metabolism of NADP + . Two principal NAD + biosynthesis pathways have been characterized: (1) the de novo pathway and *Distribution of the enzyme in completely sequenced prokaryotic genomes. All available prokaryotic genomes were downloaded to Excel from the IMG database on 6/16/2015. The analysis included only finished genomes (198 archaeal and 3380 bacterial genomes) and only one representative genome from each genus (83 archaeal and 636 bacterial genomes), to correct for the many draft genomes and for the many sequences in closely related genomes, respectively. The obtained distribution is merely an indication of the true distribution of the respective enzymes among prokaryotes because the database is incomplete and gene functions are often automatically annotated. **Use of the enzyme in metabolic engineering to increase NADPH availability and enhance biotransformation processes. ***Gibbs free energies ( r G ′m ) calculated using the biochemical thermodynamics calculator eQuilibrator 2.0 Noor et al., 2012Noor et al., , 2013, which calculates r G ′m . This term describes the change in Gibbs free energy due to a chemical reaction at a particular pH and ionic strength (subscript r), using 1 mM concentrations for all reactants (superscript m). Default settings (pH 7 and ionic strength 0.1 M) were used for all reactions. n.a., not applicable; n.d., no data available in the IMG database; G6PDH, glucose-6-phosphate dehydrogenase; 6PGDH, 6-phosphogluconate dehydrogenase; IDH, isocitrate dehydrogenase; ME, malic enzyme; GAPN, non-phosphorylating glyceraldehyde 3-phosphate dehydrogenase; NADP + -GAPDH, NADP + -dependent phosphorylating glyceraldehyde 3phosphate dehydrogenase; GDH, glucose dehydrogenase; STH, energy-independent soluble transhydrogenase; H + -TH, energy-dependent or proton-translocating, membrane-bound transhydrogenase; FNR, ferredoxin:NADP + oxidoreductase; SH, cytosolic NADP + -reducing hydrogenase; NADK, NAD + kinase.
According to Kawai and Murata (2008), NADK orthologs can be classified into three types according to their substrate specificity: (I) NADKs that utilize both ATP and inorganic polyphosphate (poly(P)) as a phosphoryl donor and phosphorylate both NAD + and NADH. This type of NADK has been identified in gram-positive bacteria and archaea. (II) NADKs that utilize ATP but not poly(P) and phosphorylate both NAD + and NADH. This type has been identified in eukaryotes. (III) NADKs that utilize ATP but not poly(P) and phosphorylate NAD + but not NADH. The latter type of NADK has been identified in gram-negative bacteria. Because of its vital role in NADP + synthesis, NADK has received much attention since its discovery in 1950 (Kornberg, 1950). Hence, many papers about its structure, function, and application are available, including reviews by Kawai and Murata (2008), Shi et al. (2009), andAgledal et al. (2010).

Redox Potential of NADP + /NADPH
The redox potential of NADP + /NADPH at standard physiological conditions is identical to that of NAD + /NADH (E ′ 0 : −320 mV). However, the function of the cofactors is different. NADP + /NADPH is used for anabolic redox reactions, whereas NAD + /NADH is used for oxidation reactions. This is possible because NADP + /NADPH is generally maintained in a reduced state and NAD + /NADH in an oxidized state (Harold, 1986). The ratios for the electron carriers have been reported, and the values can differ by several orders of magnitude, depending on the organism and growth conditions. For example, reported NAD + /NADH values range from 3.74 to 31.3 for bacterial cells (Andersen and von Meyenburg, 1977;Thauer et al., 1977;Bennett et al., 2009;Amador-Noguez et al., 2011) and reach 1820 for mammalian cells (Veech et al., 1969), whereas NADP + /NADPH values range from 0.017 to 0.95 (Andersen and von Meyenburg, 1977;Thauer et al., 1977;Bennett et al., 2009;Amador-Noguez et al., 2011). This indicates that the actual redox potential of both redox couples can deviate significantly from the standard potential (i.e., generally more negative/positive than −320 mV for (NADP + /NADPH)/(NAD + /NADH), respectively). Such deviation affects the Gibbs energy of the NADPH-generating reactions and the feasibility of converting NADH into NADPH. To assess the feasibility of various NADPH-generating reactions, we calculated the Gibbs energies ( r G ′m ) using the biochemical thermodynamics calculator eQuilibrator 2.0 Noor et al., 2012Noor et al., , 2013, setting all reactants at a physiological concentration of 1 mM (Table 1). However, even under these conditions, several reactions appeared to be at equilibrium or even slightly Frontiers in Microbiology | www.frontiersin.org endergonic. At NADP + /NADPH ratios lower than 1, the reaction becomes even less feasible, indicating that the actual concentration of the reaction components must be substantially different for NADPH generation to proceed.

Systems for NADPH Generation
NADP + synthesized by NADK or generated by NADPHoxidizing reactions is eventually reduced to NADPH. To ensure that the cellular redox balance is maintained in the absence of de novo NADP + synthesis and NADP + consumption, the catabolic fluxes through the NADPH-regenerating reactions must be matched to the anabolic demand. However, this is generally not the case (Fuhrer and Sauer, 2009). In addition, the precise NADPH formation rate depends on fluxes through the generating pathways, which in turn vary with different growth conditions (Dauner et al., 2001;Zhao et al., 2004a;Kanai et al., 2011). Therefore, prokaryotes must have other network-wide biochemical mechanisms that maintain the cellular redox balance (Fuhrer and Sauer, 2009). The exact mechanisms are not fully understood and are beyond the scope of this review, but papers about the topic are available (Singh et al., 2008;Fuhrer and Sauer, 2009).
Although numerous reactions reduce NADP + to NADPH (at least 143 reactions according to the MetaCyc database, accessed on 03/30/15), only a few have been thought to contribute significantly. However, new studies have demonstrated the importance of other NADPH-generating reactions. The present review provides a general overview of the known major NADPH-generating reactions and discusses the key enzymes involved. We have divided the reactions into (I) those that are directly coupled to central carbon metabolism and (II) those that are not ( Table 1). The enzymes that comprise the first group are the oxPPP enzymes glucose-6-phosphate dehydrogenase and 6-phosphogluconate dehydrogenase; isocitrate dehydrogenase of the TCA cycle; malic enzyme; and three enzymes involved in non-canonical NADPH-generating reactions: phosphorylating GAP dehydrogenase (NADP + -GAPDH), non-phosphorylating GAP dehydrogenase (GAPN), and glucose dehydrogenase. The enzymes in the second group are transhydrogenases (NADH:NADP + ), ferredoxin:NADP + oxidoreductase, hydrogenases (H 2 :NADP + ), and NAD(H) kinase, although the latter is a de novo NADP + /NADPHsynthesizing enzyme rather than an NADPH-regenerating enzyme. NAD(H) kinase is included because it plays an essential role in NADP + biosynthesis and the regulation of the NAD(H)/NADP(H) balance offers a potential strategy for improving the biosynthesis of industrially valuable metabolites.

NADPH-Generating Reactions Coupled to Central Carbon Metabolism
Glucose-6-phosphate Dehydrogenase and 6-phosphogluconate Dehydrogenase The oxidative pentose phosphate pathway (oxPPP) enzymes glucose-6-phosphate dehydrogenase (G6PDH, EC:1.1.1.49) and 6-phosphogluconate dehydrogenase (6PGDH, EC:1.1.1.44) have a central role in metabolism in many microbes ( Table 1). The enzymes are involved in the conversion of glucose-6-phosphate into ribulose-5-phosphate (Figure 2), a precursor of important molecules such as nucleic acids, and are generally considered a major source of NADPH. The importance of the oxPPP enzymes as a source of NADPH has been demonstrated in various organisms (Summers et al., 1995;Christiansen et al., 2002;Sauer et al., 2004). Flux through these enzymes reportedly increases when NADPH requirements are high (Obanye et al., 1996;Christiansen et al., 2002;Bartek et al., 2011;Celton et al., 2012) and decreases when NADPH requirements are low (Marx et al., 1999;Jonsbu et al., 2001). This is consistent with the anticipated anabolic role of the oxPPP as a supplier of pentoses and NADPH for biosynthesis. The key enzyme with regard to the control of PPP flux is thought to be G6PDH (Kletzien et al., 1994;Moritz et al., 2000;Saliola et al., 2012). However, in addition to its role in the oxPPP, G6PDH can also participate in the classical phosphorylated version of the Entner-Doudoroff (ED) pathway (Figure 2). The ED pathway can be considered an alternative to the Embden-Meyerhof-Parnas (EMP) glycolytic pathway, because both catabolize glucose to pyruvate (Conway, 1992;Sato and Atomi, 2011). The ED pathway is mainly present in prokaryotes, although some eukaryotes possess a functional ED pathway as well (Fabris et al., 2012). A generally held view is that the ED pathway is less important than the EMP pathway with respect to glucose catabolism. However, Fuhrer et al. have shown that this might be a misconception and that the ED pathway might be a major pathway for glucose catabolism even in species that possess both pathways (Fuhrer et al., 2005).
or 6PGDH isozymes (Zamboni et al., 2004). In general, these isozymes are similar, with the same cofactor preferences, but exceptions exist. Mycobacterium smegmatis, for example, has two G6PDH isozymes, one of which is NADP + -dependent. The other is F 420 -dependent and does not utilize NAD + or NADP + (Purwantini and Daniels, 1996). In addition, cell-free extracts of several other Mycobacterium and Nocardia species contain both F 420 -dependent and NADP + -dependent G6PDH activity (Purwantini et al., 1997), indicating that the genomes of these species likely encode both variants. Similarly, exceptions among 6PGDH isozymes have been identified. The genome of Bacillus subtilis, for example, encodes three functional 6PGDH isozymes; two prefer NAD + , and one prefers NADP + (Zamboni et al., 2004). As discussed by Zamboni et al., the presence of isozymes with different cofactor preferences would, in principle, enable NADP + , NAD + , and F 420 reduction in the oxPPP to adjust to the overall metabolic requirements of the cell (Zamboni et al., 2004). However, the exact functional roles of these isozymes remains to be determined. The availability of NADPH is of principal importance for various industrially important classes of products, including amino acids, proteins, antibiotics, organic acids, and highvalue metabolites. The apparent relationship between the rate of NADPH generation and actual carbon fluxes through the oxPPP suggests that the oxPPP can be a target for metabolic engineering for overproduction of NADPH. The oxPPP has been engineered to increase the NADPH/NADP + ratio, through overexpression of oxPPP enzymes (Lim et al., 2002;Choi et al., 2003;Becker et al., 2007;Lee et al., 2007a) and through redirection of the carbon flux from glycolysis to the oxPPP, by disrupting phosphoglucose isomerase (Kabir and Shimizu, 2003;Marx et al., 2003;Chemler et al., 2010) or phosphofructokinase (Chin and Cirino, 2011;Wang et al., 2013d), overexpressing fructose 1,6bisphosphatase (Becker et al., 2005), or introducing glucose dehydrogenase . Both strategies have been applied successfully in various prokaryotes, but reduction in growth is a common side effect (Lim et al., 2002;Lee et al., 2003;Marx et al., 2003). However, the opposite effect was observed with an archaeal strain recently developed in our lab: introduction of the oxPPP increased the hydrogen yield and significantly improved the growth rate (unpublished data).
Although both strategies have been applied successfully in various prokaryotes, they are not effective in every organism (Poulsen et al., 2005;Smith et al., 2010). Moreover, as mentioned above, G6PDH is a key enzyme not only for control of the PPP flux but also for control of the ED pathway. Overexpressing G6PDH could therefore affect the flux of both pathways simultaneously. However, this is generally not observed. Cells apparently possess the ability to regulate fluxes to ensure a network-wide balancing of NADPH supply and demand (Nicolas et al., 2007;Fuhrer and Sauer, 2009).
The effects of G6PDH and 6PGDH disruption have also been investigated. The exact effects are strongly dependent on the organism, the genetic background of the parent strain, and the environmental growth conditions (Summers et al., 1995;Butler et al., 2002;Hua et al., 2003;Zamboni et al., 2004;Zhao et al., 2004b;Richhardt et al., 2013). However, in general, G6PDH-deficient and 6PGDH-deficient prokaryotes exhibit remarkably mild phenotypes. Their growth rates are similar to or somewhat lower than those of the parent strain, and they are more sensitive to oxidative stress. Flux through the oxPPP and ED pathways is lower in G6PDH-deficient strains. To compensate for this and to ensure a sufficient supply of anabolic precursors and NADPH, flux rerouting via the EMP pathway and the non-oxPPP is generally observed (Figure 2). In addition, glucose-grown mutants generally display enhanced TCA cycle activity. In contrast, acetate-or pyruvate-grown cells generally display decreased TCA cycle activity (Zhao et al., 2004a,b).
Like G6PDH-deficient strains, 6PGDH-deficient strains exhibit a strong reduction in oxPPP flux, but not in ED pathway flux. In contrast, prokaryotes with a functional ED pathway generally respond to 6PGDH knockout by rerouting through the ED pathway and by reversing the direction of the non-oxPPP, slightly increasing the flux through the EMP pathway, and activating malic enzyme (Jiao et al., 2003;Zhao et al., 2004b). Prokaryotes that lack a functional ED pathway display a similar response, but instead of rerouting the flux through the ED pathway, they increase flux through the EMP pathway (Zamboni et al., 2004). Thus, in general, prokaryotes can compensate for disruption of G6PDH or 6PGDH.

Isocitrate Dehydrogenase
Isocitrate dehydrogenase (IDH, EC:1.1.1.42) is a component of the TCA cycle that catalyzes the decarboxylation of isocitrate to 2-oxoglutarate (also known as α-ketoglutarate), with the release of CO 2 and NADPH (Figure 3). The IDH reaction is important for the generation of reducing power. Through the generation of 2-oxoglutarate, the reaction also links nitrogen and carbon metabolism and plays an important role in the cellular defense against oxidative damage and detoxification of ROS (Muro-Pastor et al., 2005;Mailloux et al., 2007;Paul et al., 2007). Together with isocitrate lyase, IDH is a branching point between the TCA cycle and the glyoxylate shunt (Figure 3), a pathway needed for growth on non-fermentative carbon sources such as acetate and ethanol. IDH activity is important in controlling the metabolic flux between both pathways and is affected by various regulatory factors, such as metal ions (Murakami et al., 1997;Yoon et al., 2003), PEP concentration (Ogawa et al., 2007), and phosphorylation/dephosphorylation (Walsh and Koshland, 1985;Cozzone, 1998). In addition, IDH is a component of the reductive tricarboxylic acid (RTCA) cycle. The RTCA cycle is a CO 2 fixation pathway, present in some bacteria and archaea, in which four molecules of CO 2 are fixed to produce one molecule of oxaloacetate (Shiba et al., 1985;Beh et al., 1993;Kanao et al., 2002). Obviously, IDH cannot act as a source of NADPH under these conditions.
In contrast to eukaryotes, prokaryotes are generally thought to possess only NADP + -dependent IDHs. Although this is true for most prokaryotes, a small but increasing number of reports have described bacterial and archaeal NAD + -dependent IDHs, such as the one found in Hydrogenobacter thermophilus (Lloyd and Weitzman, 1988;Chen and Jeong, 2000;Steen et al., 2001;Inoue et al., 2002;Aoshima et al., 2004;Stokke et al., 2007;Wang et al., 2011bWang et al., , 2012Wang et al., , 2013a. Moreover, a survey of sequenced prokaryotic genomes by Zhu et al. identified nine additional species that likely contain NAD + -dependent IDHs (Zhu et al., 2005). However, although the IDH of H. thermophilus was originally described as an NAD + -dependent IDH, Aoshima and Igarashi later showed it was not a conventional decarboxylating IDH but a novel non-decarboxylating enzyme that catalyzes the conversion between isocitrate and oxalosuccinate. They therefore proposed that the enzyme should not be categorized as an IDH (Aoshima and Igarashi, 2008).
With the exception of the few NAD + -preferring IDHs, prokaryotic IDHs generally catalyze the following FIGURE 3 | General overview of the TCA cycle and anaplerotic node. Solid and dashed arrows represent single and lumped enzymatic reactions, respectively. Malic enzyme is able to catalyze the decarboxylation of malate and OAA. Abbreviations for metabolites (normal text) and enzymes (bold text In addition to the cofactor NADP + , IDH also needs divalent cations, particularly Mg 2+ or Mn 2+ . Although most IDHs have a strong requirement for either Mg 2+ or Mn 2+ , substitution of other divalent cations, or even univalent ions such as Na + , is sometimes possible. However, in most cases, this leads to little or no activity (Murakami et al., 1997;Steen et al., 1997;Yoon et al., 2003;Banerjee et al., 2005). Like G6PDH and 6PGDH, IDH is an important source of NADPH (Hua et al., 2003;Sauer et al., 2004). In wildtype Escherichia coli aerobically grown in chemostats under glucose-or ammonia-limited conditions, the IDH reaction was the major producer of NADPH, accounting for more than 60% of the total NADPH production (Hua et al., 2003). Accordingly, overexpression of IDH resulted in a 31% increase in the production of GDP-L-fucose, whose formation is NADPH-dependent. While this is a significant improvement, the biosynthesis of GDP-L-fucose in the IDH overexpression strain was not better than that in a G6PDH overexpression strain (Lee et al., 2011). This suggests that, under the applied conditions, IDH is not a better option than G6PDH for NADPH regeneration in biotransformation reactions (Lee et al., 2013b). The ability of IDH overexpression to enhance glutamate formation in Corynebacterium glutamicum has also been tested (Eikmanns et al., 1995). C. glutamicum secretes glutamate under certain conditions. Because IDH supplies 2-oxoglutarate, the direct precursor of glutamate, overexpression of IDH was an obvious choice. However, overexpression of IDH, even in combination with glutamate dehydrogenase, did not enhance glutamate production.
Disruption of IDH has also been investigated. The exact effects of IDH disruption were dependent on the specific conditions and species used, but some general trends were apparent (Mcdermott and Kahn, 1992;Muro-Pastor and Florencio, 1994;Eikmanns et al., 1995;Kabir and Shimizu, 2004;Becker et al., 2009). First, all IDH mutants investigated were glutamate auxotrophs. Second, although glutamate-containing cultivation media sustained the growth of IDH mutants, their growth was slower than that of the parental strains. The reduced specific growth rate was most likely due to a lower ATP/ADP ratio and a lower NADPH/NADP + ratio, as shown in an E. coli IDH disruption strain grown on glucose under aerobic conditions (Kabir and Shimizu, 2004). Moreover, other significant effects observed in the mutant E. coli strain included upregulation of oxPPP enzymes and the anaplerotic glyoxylate pathway. Similarly, redirection from the TCA cycle to anaplerosis has also been reported for an IDH knockdown of C. glutamicum (Becker et al., 2009). However, with respect to citrate synthase, contradictory results have been obtained. Kabir and Shimizu (2004) found upregulation of citrate synthase, but an earlier study of E. coli IDH mutants (Lakshmi and Helling, 1976) and a study of Rhizobium meliloti (Mcdermott and Kahn, 1992) found that IDH mutants spontaneously gave rise to IDH-citrate synthase double mutants.

Malic Enzyme
The TCA cycle-associated malic enzyme (ME, EC:1.1.1.40) catalyzes the NAD(P) + -dependent oxidative decarboxylation of malate to pyruvate (Figure 3). Confusingly, malic enzyme is also known as "malate dehydrogenase (oxaloacetatedecarboxylating), " while true malate dehydrogenases, enzymes that catalyze the non-decarboxylating reaction between malate and oxaloacetate (Figure 3), are sometimes referred to as malic enzyme. In addition, various classes of ME that differ in their preference for NAD + or NADP + and in their ability to decarboxylate oxaloacetate have also been characterized (Frenkel, 1975). Enzymes in the first ME class (EC:1.1.1.38) use NAD + and can decarboxylate oxaloacetate as well as malate. Enzymes in the second class (EC:1.1.1.39) also prefer NAD + , but they are unable to decarboxylate oxaloacetate. Enzymes in the third class (EC:1.1.1.40) are NADP + -dependent; they can catalyze the decarboxylation of malate and oxaloacetate. Moreover, MEs that do not fit the general classification scheme exactly, such as those from the gram-negative bacterium Rhizobium meliloti and the hyperthermophilic archaeon Thermococcus kodakarensis, have been described as well (Voegele et al., 1999;Fukuda et al., 2005). The latter, for example, prefers NADP + but, unlike normal NADP + -dependent MEs, is not able to decarboxylate oxaloacetate.
In many aerobic and facultative anaerobic organisms, the TCA cycle-associated ME is part of the PEP-pyruvateoxaloacetate node, also referred to as the anaplerotic node (Figure 3). This node represents the metabolic link between glycolysis/gluconeogenesis and the TCA cycle (Owen et al., 2002;Sauer and Eikmanns, 2005). As such, ME is involved in the interconversion of C4 and C3 compounds, which is important for maintaining the levels of TCA cycle intermediates (anaplerotic reactions) and for growth on C4 and C3 compounds and substrates that enter central metabolism via acetyl-CoA, such as acetate, fatty acids, and ethanol. Although ME can catalyze C3-carboxylation and C4-decarboxylation reactions, it is generally involved in the latter (decarboxylation of malate to pyruvate with the concomitant formation of NADPH) (Voegele et al., 1999;Gourdon et al., 2000;Sauer and Eikmanns, 2005;Lerondel et al., 2006;Wang et al., 2011a). However, examples of prokaryotic MEs without a preference for either  and MEs with a preference for the C3-carboxylation reaction have been identified (Matula et al., 1969). To determine the exact contribution of ME to the cellular NADPH pool, the physiological direction and flux through the enzyme needs to be established. However, regulation of the carbon flux at the anaplerotic node involves a complex interplay between the enzymes involved and depends on the growth conditions and species Dauner et al., 2001;Fischer and Sauer, 2003;Netzer et al., 2004). A detailed discussion of this complex regulation is beyond the scope of this review, but comprehensive reviews, such as those by Owen et al. (2002) and Sauer and Eikmanns (2005), are available.
To understand the physiological role of ME in the metabolism of prokaryotes, the disruption of ME and alternative pathways has been studied. The significance of the latter is clearly demonstrated by the absence of a detectable phenotype in Campylobacter jejuni and E. coli ME disruption strains (Velayudhan and Kelly, 2002;Kao et al., 2005). The lack of a clear phenotype can be explained by the presence of alternative pathways and enzymes, such as the malate dehydrogenasephosphoenolpyruvate carboxykinase pathway (mdh-PEPCK) and isocitrate dehydrogenase, that can fulfill the pyruvate/PEP and NADPH requirements, respectively (Figure 3) (Hansen and Juni, 1975;Oh et al., 2002;Velayudhan and Kelly, 2002;Wang et al., 2011a). Because of the wide distribution of such alternatives in prokaryotes, it is likely that ME gene disruption in other species will yield similar results. Nevertheless, species in which ME gene disruption yields a phenotype, such as Corynebacterium glutamicum, Bacillus subtilis, and various lactic acid bacteria, have been identified (Gourdon et al., 2000;Lerondel et al., 2006;Landete et al., 2010;Espariz et al., 2011).
Analysis of various disruption strains created to understand the physiological role of ME in prokaryotes has shown that it is crucial to take the genetic background of a strain into consideration because the exact physiological role of ME differs from species to species (Riedel et al., 2001;Velayudhan and Kelly, 2002;Kao et al., 2005;Landete et al., 2010;Meyer and Stulke, 2013). Analysis has also shown that the effect of ME disruption can be subtle and/or heavily dependent on the cultivation conditions (Gourdon et al., 2000;Lerondel et al., 2006). A C. glutamicum ME disruption strain, for example, showed no detectable phenotype during growth on either acetate or glucose, but showed a significantly modified growth behavior during lactate metabolism (Gourdon et al., 2000). Moreover, the physiological role of NAD + -and NADP + -dependent MEs appears to be different (Lerondel et al., 2006;Bologna et al., 2007;Landete et al., 2010;Espariz et al., 2011;Wang et al., 2011a;Meyer and Stulke, 2013). Whereas NAD + -dependent MEs generally play a role in malate catabolism, NADP +dependent MEs function either as gluconeogenic enzymes, by supplying pyruvate from C 4 -dicarboxylic acids, or as NADPHgenerating systems needed for various biosynthetic purposes. It is interesting in this respect that the concerted action of ME, MDH, and pyruvate carboxylase (PCx) can act as a kind of transhydrogenase. The combined action of these enzymes enables the ATP-dependent conversion of pyruvate to OAA and malate and back to pyruvate, while converting NADH into NADPH (Figure 3) (Sauer and Eikmanns, 2005). A similar reaction sequence can involve PEP carboxylase (PEPCx) instead of pyruvate carboxylase, which also enables conversion of NADH into NADPH. This so-called malate shunt was recently described in Clostridium thermocellum (Zhou et al., 2013).
NAD + -and NADP + -dependent MEs have also been used in metabolic engineering approaches to improve specific biotechnological applications. However, to our knowledge, no prokaryotic ME has been used to increase intracellular NADPH levels, as has been done in the yeast Saccharomyces cerevisiae (Moreira Dos Santos et al., 2004;Matsuda et al., 2013). Instead, prokaryotic MEs have been overexpressed for the formation of C 4 -dicarboxylic acids from glucose or pyruvate (Stols and Donnelly, 1997;Hong and Lee, 2001;Shin et al., 2007;Zheng et al., 2009). These acids are important intermediates in the production of tetrapyrroles (Shin et al., 2007) and amino acids , for example, and they can be chemically converted into common intermediates in the petrochemical industry. Hence, C 4 -dicarboxylic acids are important intermediates in the production of commodity chemicals from renewable carbohydrate feedstocks (Jain et al., 1989;Szmant, 1989). However, to produce these acids from glucose or pyruvate, one has to reverse the physiological direction of ME. This has been accomplished in different ways (Stols and Donnelly, 1997;Hong and Lee, 2001;Zheng et al., 2009).
phosphorylating or non-phosphorylating type or both types, thus generating confusion.
GAPN was originally thought to be present only in photosynthetic eukaryotes (Kelly and Gibbs, 1973;Mateos and Serrano, 1992). However, early reports of GAPN activity in Streptococcus mutans were later confirmed by the characterization of the first prokaryotic GAPN (Crow and Wittenberger, 1979;Boyd et al., 1995). Since then, GAPN has also been found in various (hyper)thermophilic archaea, such as Thermococcus kodakarensis, Thermoproteus tenax, and Sulfolobus solfataricus (Brunner et al., 1998;Ettema et al., 2008;Matsubara et al., 2011), and several gram-positive bacteria, such as Clostridium acetobutylicum, Streptococcus pyogenes, and Bacillus halodurans ( Table 1) (Iddar et al., 2002(Iddar et al., , 2003(Iddar et al., , 2005Sophos and Vasiliou, 2003;Agledal et al., 2010). GAPN catalyzes the irreversible, non-phosphorylating oxidation of GAP to 3-PG with the concurrent reduction of NADP + to NADPH: In this reaction, 1,3-BPG is not formed, and the oxidation of GAP is not coupled to the generation of ATP (Figure 4). Such modifications of the classical EMP or ED pathway are regularly found in archaea (Ahmed et al., 2005;Siebers and Schonheit, 2005;Ettema et al., 2008;Matsubara et al., 2011;Brasen et al., 2014). Interestingly, GAPN has been reported to have a key function in the regulation of these modified pathways (Brunner et al., 1998;Lorentzen et al., 2004;Ettema et al., 2008). With the exception of the GAPN in T. tenax, which has a preference for NAD + , prokaryotic GAPNs, in general, have a clear preference for NADP + . However, although the GAPN of T. tenax was shown to be NAD + -dependent, it was later shown to be allosterically regulated, shifting its cofactor preference from NAD + to NADP + in the presence of activators such as F6P and G1P (Lorentzen et al., 2004). Based on sequence similarity, similar allosteric regulation might also exist for GAPNs in a number of other hyperthermophiles (Lorentzen et al., 2004;Ito et al., 2012). Aside from NADP + , GAPNs do not require other cofactors or metals.
While the physiological significance of GAPN might vary between organisms, the enzyme is an important and essential source of NADPH in the bacterium S. mutans and the archaeon T. kodakarensis, respectively (Boyd et al., 1995;Matsubara et al., 2011;Arutyunov et al., 2013). Moreover, the ability to use GAPN as a source of NADPH has been demonstrated in a mutant E. coli strain in which the endogenous NAD + -dependent glyceraldehyde-3-phosphate dehydrogenase (NAD + -GAPDH) was replaced by the GAPN of S. mutans. Transcriptional analysis of this strain revealed upregulation of the transhydrogenase and downregulation of the PPP and TCA cycle, which together indicate a response to avoid NADPH excess (Centeno-Leija et al., 2013). In addition, GAPN has been used successfully in metabolic engineering efforts intended to improve the production of metabolites whose biosynthesis depends on NADPH, such as Llysine or poly-3-hydroxybutyrate, or to decrease the formation of undesired by-products such as glycerol in industrial ethanol production strains (Takeno et al., 2010;Guo et al., 2011;Centeno-Leija et al., 2014).
GAP + NADP + + Pi → 1, 3 − bisphosphoglycerate + NADPH + H + r G ′m = 25.9 ± 1.0 kJ/mol (6) The generation of 1,3-BPG is an important difference between the reactions catalyzed by NADP + -GAPDH and GAPN, because further oxidation of 1,3-BPG to 3-PG generates ATP. Moreover, while the oxidation of GAP by NADP + -GAPDH is reversible, the oxidation by GAPN is not. Hence, NADP + -GAPDH, in contrast to GAPN, is active in both glycolysis and gluconeogenesis. However, in prokaryotes that also contain alternative GAP oxidation enzymes, such as NAD + -GAPDH, GAPN, or GAP:ferredoxin oxidoreductase (GAPOR) (Figure 4), NADP + -GAPDH is primarily involved in gluconeogenesis and does not appear to play a role in generating NADPH (Schäfer and Schönheit, 1993;Koksharova et al., 1998;Fillinger et al., 2000;Matsubara et al., 2011;Ito et al., 2012). As with GAPN, the significance of NADP + -GAPDH as a source of NADPH has been demonstrated. Overexpression of NADP + -GAPDH or replacement of the native NAD + -GAPDH with the NADP-variant improved the yield of NADPH-dependent products such as lycopene, ǫ-caprolactone, L-ornithine, or coenzyme Q 10 (Martinez et al., 2008;Huang et al., 2011;Jiang et al., 2013;Wang et al., 2013c). Moreover, metabolic flux analysis of the exchange mutant revealed that the oxPPP branch and TCA fluxes were significantly reduced, presumably to avoid NADPH excess (Martinez et al., 2008). In addition, an attempt in C. glutamicum to alter the cofactor specificity of native NAD + -GAPDH to NADP + through protein engineering resulted in a 60% increase in lysine production (Bommareddy et al., 2014).
GDHs are responsible for a main difference between modified ED pathways and the classical ED pathway. GDHs catalyze the direct oxidation of glucose into gluconolactone. In contrast, in the classical ED, glucose is phosphorylated before its oxidation into 6-P-gluconolactone (Brasen et al., 2014). L-glucose + NADP + → L-glucono-1,5-lactone + NADPH + H + r G ′m = −2.4 ± 2.2 kJ/mol (7) Two different classes of GDHs are capable of using NADP + as cofactor: EC:1.1.1.47, which shows dual cosubstrate specificity with a preference for NADP + , and EC:1.1.1.119, which is strictly dependent on NADP + . GDHs belonging to these groups have been found in extremely halophilic and thermoacidophilic archaea, as well as in bacteria such as Bacillus subtilis, Bacillus megaterium, and Gluconobacter oxydans (Pauly and Pfleiderer, 1975;Adachi and Ameyama, 1981;Lampel et al., 1986;Brasen et al., 2014). NADP + -GDHs have been used for NADPH regeneration in various production systems, such as the in vitro synthesis of poly(3-hydroxybutyrate) or L-leucovorin (Eguchi et al., 1992;Satoh et al., 2003). In addition, NADP + -GDHs are important FIGURE 5 | Classical ED pathway compared to modified ED pathways. Solid and dashed arrows represent single and lumped enzymatic reactions, respectively. Shown are the classical ED pathway (in black), the semiphosphorylated ED pathway (in blue), the non-phosphorylated ED pathway (in red), and the branched ED pathway that includes the semi-and non-phosphorylated ED pathways. Only the glucose degradation directions are shown. Abbreviated enzymes (in bold text): G6PDH, glucose-6-phosphate dehydrogenase; GAPN, non-phosphorylating glyceraldehyde-3-phosphate dehydrogenase; GDH, glucose dehydrogenase. components of many commercial glucose-sensing assays. As such, they have been utilized for blood glucose monitoring (Ferri et al., 2011). Moreover, NADP + -GDHs have also been used as NADPH regenerators in various whole-cell, or in vivo, production systems. The GDHs of B. megaterium and B. subtilis, for example, have been expressed in several E. coli strains for the synthesis of various products, including indigo, S-sulfoxide, and ethyl (R/S)-4-chloro-3-hydroxybutanoate (Kataoka et al., 1998(Kataoka et al., , 1999Kizaki et al., 2001;Yun et al., 2005;Lu and Mei, 2007;Xu et al., 2007;Park et al., 2010;Zhang et al., 2011).
The two transhydrogenase isoforms transfer electrons between NAD(H) and NADP(H) in the following reaction: However, in contrast to STHs, H + -THs couple the electron transfer reaction to the translocation of a proton across the membrane, according to the reaction: Where "out" and "in" denote the periplasmic space and cytosol, respectively. Although the reaction is in principle reversible, the electrochemical proton gradient ( p) across the membrane under most physiological conditions strongly favors NADPH formation (Hoek and Rydstrom, 1988;Bizouarn et al., 2000;Jackson, 2003;Pedersen et al., 2008).
Since the discovery of transhydrogenases, there has been much debate about their physiological role (Bragg et al., 1972;Voordouw et al., 1983;Harold, 1986;Hoek and Rydstrom, 1988;Jackson et al., 1998;Pedersen et al., 2008). Given the type of reaction they catalyze, transhydrogenases might play an important role in maintaining the redox balance within microorganisms. Various studies have shown that several bacterial species require STH for growth under metabolic conditions with excess NADPH formation (Canonaco et al., 2001;Hua et al., 2003;Sauer et al., 2004;Zhao et al., 2008;Fuhrer and Sauer, 2009). Moreover, mutant bacteria suffering from a disturbed redox balance due to an increased NADPH concentration often benefit from the heterologous expression of STH (Boonstra et al., 2000;Angermayr et al., 2012;Qi et al., 2014;Reddy et al., 2015). These observations support the idea that the physiological role of STH is to convert NADPH into NADH to prevent an excess of NADPH (Voordouw et al., 1983), a notion that is generally accepted and is supported by actual NAD + /NADH and NADP + /NADPH ratios (see above), which indicate that the formation of NADPH from NADH will generally require energy input.
However, Sanchez et al. have reported that the NADPHrequiring production of poly(3-hydroxybutyrate) in E. coli benefitted from the overexpression of STH (Sanchez et al., 2006). This result is contrary to the generally accepted physiological role of STH and suggests that the enzyme catalyzes the opposite reaction. Moreover, recent reports have shown that STH is able to catalyze the opposite reaction and that overexpression of the enzyme can improve NADPH-dependent production yields (Lee et al., 2010;Decorosi et al., 2011;Jan et al., 2013;He et al., 2014). Additionally, although STH disruption and overexpression mutants often have disturbed or improved redox balances, respectively, similar mutants that do not show any effect have also been described (Chin et al., 2009;Yamauchi et al., 2014). Together, these examples show that we have yet to achieve a thorough understanding of the physiological role of STH and that its role is strongly dependent on species, culture conditions, and genotypes.
Similarly, the physiological role of H + -TH has been debated in the literature. It has often been reported that the H + -TH isoform is a major source of NADPH, needed for the biosynthesis of amino acids, for example, or the reduction of glutathione (required to minimize oxidative damage caused by free radicals generated in the respiratory chain) (Bragg et al., 1972;Voordouw et al., 1983;Hoek and Rydstrom, 1988;Hickman et al., 2002;Hua et al., 2003;Sauer et al., 2004;Fuhrer and Sauer, 2009;Imam et al., 2013;He et al., 2014). In E. coli, H + -TH has been shown to provide about 40% of the total NADPH during growth on glucose in batch cultures . Moreover, its expression is induced when there is demand for NADPH, and it is required for optimal growth on carbon sources whose metabolism does not directly generate NADPH Imam et al., 2013). Additionally, Fuhrer and Sauer calculated that it is very unlikely, from a thermodynamic point of view, that H + -THs catalyze NADPH oxidation. They concluded that the NADP + reduction catalyzed by H + -THs is irreversible under physiological conditions in various bacterial species (Fuhrer and Sauer, 2009). Reports such as these contribute to the general view that the physiological role of H + -TH is reduction of NADP + at the expense of NADH. However, other reports suggest that H + -TH catalyzes the opposite reaction or show that disruption or overexpression has no measurable effect (Chin et al., 2009;Jan et al., 2013;Yamauchi et al., 2014).
THs require the cofactors NAD(H) and NADP(H) for activity. In addition, and in contrast to H + -THs, STHs also require a flavin (flavin adenine dinucleotide, FAD) cofactor (Middleditch et al., 1972;Boonstra et al., 1999;Cao et al., 2011). THs generally do no required other cofactors or metal ions. In fact, metal ions have even been reported to inhibit TH activity (Whitehead et al., 2005(Whitehead et al., , 2009Cao et al., 2011). Nevertheless, submillimolar concentrations of Ca 2+ and Mg 2+ have been reported to stimulate the H + -THs of Rhodobacter capsulatus and E. coli (Cotton et al., 1989;Lever et al., 1991;Chang et al., 1992). Similarly, saturating concentrations of Ca 2+ have been reported to release the phosphate-induced inhibition of the NADP + reducing reaction and to cause full activation of the enzyme in Azotobacter vinelandii and Pseudomonas aeruginosa, respectively (Hojeberg and Rydstrom, 1977;Voordouw et al., 1980). Because of their ability to generate NADPH, STHs and H + -THs have been overexpressed to improve the production yields of NADPH-dependent metabolites. For example, overexpression of STH increased NADPH-dependent poly(3-hydroxybutyrate) production by 82% in an engineered E. coli strain (Sanchez et al., 2006). Moreover, the overexpression of STH in two other engineered E. coli strains increased (S)-2-chloropropionate and thymidine production 150 and 100%, respectively (Lee et al., 2010;Jan et al., 2013). Similarly, overexpression of H + -TH has significantly improved the yields of various NADPHdependent products, such as chiral alcohols, isobutanol, and 3hydroxypropionic acid in E. coli (Weckbecker and Hummel, 2004;Bastian et al., 2011;Rathnasingh et al., 2012;Shi et al., 2013) and L-valine and L-lysine in C. glutamicum (Kabus et al., 2007;Bartek et al., 2011).

Ferredoxin:NADP + Oxidoreductase
Ferredoxin:NADP + oxidoreductases (FNRs, EC:1.18.1.2) are ubiquitous flavoenzymes ( Table 1) that catalyze the reversible transfer of reducing equivalents between the one-electron carrier ferredoxin (Fd) and the two-electron carrying NADP(H), according to the following reaction (Figure 6): 2Fd red + NADP + + H + ↔ 2Fd ox + NADPH r G ′m = −15.6 ± 11.7 kJ/mol (10) FNRs are present in all three domains of life (Ma and Adams, 2001a;Ceccarelli et al., 2004;Santangelo et al., 2011;Yan et al., 2014). They are mainly known for their essential role in photosynthetic organisms (plants, algae, and cyanobacteria), in which they connect the Fd red -generating reactions of photosynthesis to NADPH-requiring carbon assimilation (Shin and Arnon, 1965;Goss and Hanke, 2014). In addition to enabling the flux of electrons from the photosystem into other metabolic pathways, FNRs might also avert this efflux of electrons and mediate the return of Fd-bound electrons to the plastoquinone pool of the photosynthetic system. By doing so, FNRs contribute to the so-called cyclic electron flow, which is essential for balancing the ATP/NADPH ratio generated by the photosystems (Munekage et al., 2004;Shikanai, 2007;Kramer and Evans, 2011;Mulo, 2011;Goss and Hanke, 2014). In non-photosynthetic tissues and organisms, FNRs participate in the electron transfer chains of various other metabolic processes such as nitrogen fixation, sulfate assimilation, isoprenoid biosynthesis, osmotic and oxidative stress responses, and iron-sulfur cluster biogenesis (Arakaki et al., 1997;Krapp et al., 2002;Seeber et al., 2005;Giro et al., 2006;Lee et al., 2007b;Balconi et al., 2009;Tang et al., 2010;Lewis et al., 2013). However, in contrast to the FNRs involved in photosynthesis, the physiological direction of FNRs involved in these processes is toward the production of Fd red . As a result, FNRs are sometimes classified as autotrophic (all photosynthetic FNRs) and heterotrophic (all other FNRs) (Arakaki et al., 1997), which, from a protein-family point of view, is not entirely correct.
FNRs are generally classified into two phylogenetically and structurally unrelated protein families that are subdivided into four subclasses: the plant-type FNRs comprising the plastidictype and bacterial-type subclasses and the glutathione (GR)-type FNRs comprising the mitochondrial adrenodoxin reductases (AdR-like) and oxygenase-coupled NADH:ferredoxin reductases (ONFR-like) (Carrillo and Ceccarelli, 2003;Ceccarelli et al., 2004;Karplus and Faber, 2004;Aliverti et al., 2008;Musumeci et al., 2012). In addition, a third type of FNR has been reported, the so-called thioredoxin reductase-like FNR (TRLF), which functions as an FNR but has structural homology to NADPHdependent thioredoxin reductases (Seo and Sakurai, 2002;Seo et al., 2004;Komori et al., 2010;Muraki et al., 2010;Yan et al., 2014). Interestingly, a novel type of FNR has recently been reported, the electron-bifurcating NADH-dependent reduced ferredoxin:NADP + oxidoreductase (NfnAB), which couples the exergonic reduction of NADP + with Fd red and the endergonic reduction of NADP + with NADH in a reversible reaction (reaction 11 and Figure 6) (Wang et al., 2010;Huang et al., 2012;Rydzak et al., 2014). Genome analyses have revealed the NfnAB genes are present in many anaerobic bacteria and archaea (Huang et al., 2012;Buckel and Thauer, 2013), a finding that might reflect the importance of a postulated NfnAB function: NfnAB and other electron-bifurcating reactions have been implicated as a third mode of energy conservation, in addition to substrate level phosphorylation and electron transport phosphorylation (Buckel and Thauer, 2013). (11) Despite what the names imply, both subdivisions of the planttype FNRs (plastidic-type and bacterial-type) as well as FNRs of the GR-type, TRLF-type, and "NfnAB-type" are found in prokaryotes (Ceccarelli et al., 2004;Aliverti et al., 2008;Komori et al., 2010). Unfortunately, the literature does not always make clear the class to which an FNR belongs, generating confusion and making it difficult to judge the true NADPH-generating potential of FNRs found in different prokaryotes. Nevertheless, under normal physiological conditions, it appears that only the plastidic-type FNRs, found in cyanobacteria, for example, and the novel "NfnAB-type" have significant effects on NADPH availability in prokaryotes.
The reaction catalyzed by the different FNR types is in principle reversible, as demonstrated by various in vitro studies, as well as by in vivo studies in cyanobacteria and the thermophilic bacterium Hydrogenobacter thermophilus TK-6. Cyanobacteria contain a single FNR that is responsible for NADP + reduction in vegetative cells as well as for Fd reduction in heterocysts (Razquin et al., 1996). A similar situation exists in H. thermophilus TK-6, where a single FNR catalyzes either the forward or reverse reaction depending on the specific type of Fd involved in the reaction. A [4Fe-4S]-containing Fd is involved in the forward (NADP + reducing) reaction, whereas a [2Fe-2S]-containing Fd is involved in the reverse reaction (Ikeda et al., 2009). Such insights might facilitate the development of rational engineering approaches aimed at gaining directional control of the reaction. Elucidating the reaction mechanisms of various FNR-types is a topic of many studies and has been reviewed extensively (Arakaki et al., 1997;Onda et al., 2000;Aliverti et al., 2001Aliverti et al., , 2008Carrillo and Ceccarelli, 2003;Medina and Gomez-Moreno, 2004;Martinez-Julvez et al., 2009;Yeom et al., 2009;Musumeci et al., 2012).
FNRs, in general, contain a non-covalently bound flavin (FAD) cofactor as a prosthetic group, but a flavin mononucleotide (FMN)-containing FNR has also been characterized (Ikeda et al., 2009). The FAD group can alternate between an oxidized, one-electron-reduced semiquinone and a fully reduced hydroquinone state, thereby mediating the reversible electron change between the one-electron carrier Fd and the two-electron carrying NADP(H) (Carrillo and Ceccarelli, 2003). The iron-sulfur-containing Fd can sometimes be substituted by the flavin mononucleotide (FMN)-containing flavodoxin (under iron-depleted conditions in some bacteria) or, in the case of the GR-type FNRs, by the mitochondrial ironsulfur-containing adrenodoxin (Yamazaki and Ichikawa, 1990;Ceccarelli et al., 2004;Sancho, 2006;Ewen et al., 2011). Moreover, a novel rubrerythrin-like protein, named ferriperoxin (Fpx), has been described that exhibits NADPH-and FNR-dependent peroxidase activity in the thermophilic hydrogen-oxidizing bacterium H. thermophilus (Sato et al., 2012). Whereas most FNRs are NADP(H)-dependent, an NAD(H)-dependent variant also exists, the ONFR-like subtype of the GR-type FNRs (EC:1.18.1.3) (Aliverti et al., 2008). However, because this group of FNRs is not involved in NADPH availability, it will not be discussed further in this review.

NADP + -dependent Hydrogenase
Hydrogenases (H 2 ases) are metalloenzymes that catalyze the reversible oxidation of hydrogen gas into two protons and two electrons: H 2 ases enable organisms to utilize H 2 as a source of reducing power or to use protons as terminal electron acceptors, generating H 2 . According to the metal atoms at their active site, H 2 ases can be classified into three types: [NiFe]-H 2 ases, [FeFe]-H 2 ases, and [Fe]-H 2 ases. The latter were formerly known as metal-free or iron-sulfur-free H 2 ases; they are involved in CO 2 reduction with H 2 to methane and are only found in methanogenic archaea (Lyon et al., 2004;Vignais and Colbeau, 2004;Shima and Thauer, 2007;Trchounian et al., 2012;Trchounian and Gary Sawers, 2014;Peters et al., 2015).
[NiFe]-H 2 ases are generally responsible for H 2 oxidation and are widely distributed among both aerobic and anaerobic bacteria and archaea; they have been identified in 28.9 and 57.8% of the available complete bacterial and archaeal genomes, respectively.
[FeFe]-H 2 ases, in contrast, are generally responsible for H 2 evolution and are only found in anaerobic bacteria; they have been identified in 9.8% of the available complete bacterial genomes (Peters et al., 2015). Alignment of the sequence motifs that coordinate the active site, separates [NiFe]-H 2 ases into at least four, and possibly five, functional groups that are, in general, consistent with the physiological roles of the enzymes (Vignais et al., 2001;Vignais and Colbeau, 2004;Vignais and Billoud, 2007;Peters et al., 2015). Thus, H 2 ases are a diverse group of enzymes. They are cytosolic or membrane bound, involved in H 2 oxidation and/or evolution, and able to use a variety of electron acceptors and donors. It is beyond the scope of this review to address all of the different types of H 2 ases. We will focus on those involved in NADPH generation in prokaryotes. For more information about H 2 ases in general, the reader is referred to various excellent reviews, such as those by Vignais and Colbeau (2004), Shima and Thauer (2007), and Peters et al. (2015).
The requirement for these different ligands, metals, and cofactors makes the biosynthesis of multisubunit H 2 ases a complex and dynamic process, involving various maturation proteins (Bock et al., 2006;Forzi and Sawers, 2007;Watanabe et al., 2012;Peters et al., 2015). It is thus no surprise that reports about the practical applications of NADP + -reducing H 2 ases for NADPH generation are limited. Nevertheless, the potential of these enzymes to improve the production yields of NADPH-dependent metabolites has been demonstrated. For example, two prochiral model substrates, acetophenone and (2S)-hydroxy-1-phenyl-propanone, were quantitatively reduced to the corresponding (S)-alcohol and (1R,2S)-diol by an NADPH-dependent alcohol dehydrogenase in an in vitro system containing H 2 and the partially purified NADP + -reducing H 2 ase of P. furiosus (Mertens et al., 2003). Although in vitro examples like this demonstrate the potential of NADP + -H 2 ases, expression of heterologous [NiFe]-H 2 ases outside of a closely related host is still challenging (English et al., 2009). It often results in an inactive enzyme or activity that is detectable only in vitro (Voordouw et al., 1987;Grzeszik et al., 1997;Sun et al., 2010). Nevertheless, successful expression of heterologous [NiFe]-H 2 ase has been reported (Wells et al., 2011). NAD + Kinase and NADH Kinase NAD + kinase (NADK, EC:2.7.1.23) catalyzes the conversion of NAD + to NADP + . It is the sole enzyme leading to de novo NADP + biosynthesis (Figures 1, 6). As such, NADK is essential for NADP + and NADPH availability. Together with NADP + phosphatase (NADPase), which is thought to catalyze the opposite reaction (the conversion of NADP + to NAD + ), NADK is crucial for the regulation of the intracellular balance of NAD(H) and NADP(H) (Kawai and Murata, 2008;Shi et al., 2009). Although NADPase activity has been detected in various organisms (Reidl et al., 2000;Kawai et al., 2004), to our knowledge, no enzyme that functions solely as an NADPase has been isolated to date. However, some enzymes that exhibit NADPase activity in addition to another activity have been isolated, such as the bifunctional NADP + phosphatase/NAD + kinase, discovered in the archaeon Methanococcus jannaschii Fukuda et al., 2007).
NADK activity was enriched in extracts from the yeast Saccharomyces cerevisiae for the first time in 1950 (Kornberg, 1950) and has since been purified from various organisms (Mcguinness and Butler, 1985;Magni et al., 1999). However, it was not until 2000 that the gene encoding NADK was identified by Kawai et al. in Micrococcus flavus and Mycobacterium tuberculosis H37Rv (Kawai et al., 2000). Since then, NADKs from various organisms have been studied in detail, and although the exact physiological functions, related pathways, and regulatory mechanisms have not always clearly been elucidated, some reviews of NADK structure, function, and potential applications are available (Magni et al., 2006;Pollak et al., 2007;Kawai and Murata, 2008;Shi et al., 2009;Agledal et al., 2010).
NADKs are ubiquitous enzymes (Table 1). Homologs, according to literature, can be found in almost all eukaryotes and prokaryotes, with the exception of the intracellular parasite Chlamydia trachomatis (Grose et al., 2006;Kawai and Murata, 2008;Szaszák et al., 2011). The discrepancy between the distribution shown in Table 1 and that described in the literature might reflect the rapidly increasing availability of prokaryotic genomes and errors inherent to automated genome annotation (Bakke et al., 2009). Functionally and structurally, NADKs are weakly related to other kinases, suggesting that they belong to a kinase superfamily (Labesse et al., 2002). With the exception of the NADKs of gram-negative bacteria, the enzymes accept NAD + and its reduced form, NADH (Kawai and Murata, 2008). Nevertheless, they are generally referred to as NAD + kinases (EC:2.7.1.23) because they usually have a strong preference for the oxidized form. NADKs with a preference for the reduced form exist as well (EC:2.7.1.86) (Outten and Culotta, 2003;Strand et al., 2003;Turner et al., 2005), but have not to our knowledge been found in prokaryotes.
Aside from NAD(H), NADKs also need a phosphoryl donor for catalysis. The preferred phosphoryl donor often is a nucleoside triphosphate, particularly ATP, but many bacterial and archaeal NADKs can also utilize the ancient inorganic energy carrier poly(P) (Kawai et al., 2000Garavaglia et al., 2003;Sakuraba et al., 2005). In addition, some bacterial enzymes can utilize ADP or glucose-6-phosphate (Bark et al., 1993;Lindner et al., 2010). However, the use of glucose-6-phosphate as a phosphoryl donor has not been investigated further. Finally, NADKs often also require divalent cations such as Mg 2+ , Mn 2+ , Zn 2+ , or Ca 2+ for their activity (Kawai et al., 2000(Kawai et al., , 2001Garavaglia et al., 2003;Sakuraba et al., 2005). The NADK of Corynebacterium glutamicum, however, does not require divalent cations when using ATP as phosphoryl donor, but the enzyme is stimulated in their presence. When using poly(P) as phosphoryl donor, divalent cations are required for activity (Lindner et al., 2010).
According to the phosphoryl donor specificity, NADK orthologs are subdivided into several types. The classification typically applied divides the NADK orthologs into: (I) poly(P)/ATP NADKs, which can use poly(P) and ATP as phosphoryl donors and (II) ATP NADKs, which are specific for ATP (Mori et al., 2005;Nakamichi et al., 2013). Poly(P)/ATP NADKs have been identified in gram-positive bacteria such as Mycobacterium tuberculosis, Micrococcus flatus, and Bacillus subtilis (Kawai et al., 2000;Garavaglia et al., 2003) and in archaea such as Methanococcus jannaschii and Pyrococcus horikoshii Sakuraba et al., 2005). ATP NADKs have been found in gram-negative α-and γ-Proteobacteria such as Sphingomonas sp. A1, Salmonella enterica, and E. coli and in eukaryotes (Kawai et al., 2001;Ochiai et al., 2004;Grose et al., 2006;Nakamichi et al., 2013). A slightly different classification system is sometimes applied, dividing the NADK orthologs into: (I) ATP NADKs, which are strictly dependent on ATP, (II) NTP NADKs, which use ATP in addition to other nucleoside triphosphates, and (III) poly(P)/NTP NADKs, which use nucleoside triphosphates and poly(P) (Shi et al., 2009). The existence of these slightly different classification systems can make it difficult to ascertain the type of NADK being discussed. However, because no strict ATP NADKs (as defined in the latter system) have been identified in prokaryotes to date, it is not an issue in this review. The reaction catalyzed by NADKs can be described as either: NAD + + nucleoside triphosphate (NTP, preferably ATP) → NADP + + NDP or NAD + + NTP/poly(P) → NADP + + NDP/poly(P) n−1 (14) NADK is the sole enzyme leading to de novo NADP + biosynthesis. It has a central role in the regulation of the NAD(H)/NADP(H) balance because it can directly phosphorylate NADH to form NADPH or cooperate with various NADP + reducing enzymes to form NAD(H)/NADP(H) metabolic networks that ultimately convert NADH into NADPH (Singh et al., , 2008Shi et al., 2009). In view of the importance of NADK, it is not surprising that disruption of NADK is lethal in various prokaryotes (Gerdes et al., 2002;Thanassi et al., 2002;Kobayashi et al., 2003;Sassetti et al., 2003;Zalacain et al., 2003;Grose et al., 2006;Poncet-Montange et al., 2007). Because of its essential role in NADP + biosynthesis and the regulation of the NAD(H)/NADP(H) balance, NADKs have attracted attention as a potential strategy by which to improve the biosynthesis of industrially valuable metabolites. Several studies have shown that overexpression of NADKencoding genes can increase the production yields of various NADPH-dependent biosynthetic pathways, such as poly(3hydroxybutyrate) production and thymidine production in E. coli (Lee et al., 2009(Lee et al., , 2010Li et al., 2009), lysine and isoleucine production in C. glutamicum (Lindner et al., 2010;Shi et al., 2012), and arginine production in Corynebacterium crenatum (Rahman et al., 2012). These results suggest that increased NADP + availability, resulting from NADK overexpression, disturbs the intracellular redox balance, subsequently affecting the expression of NADP + -reducing enzymes (Lee et al., 2009;Li et al., 2009;Lee et al., 2010). However, a comparable attempt to improve isobutanol production in E. coli showed that NADK overexpression alone is not always sufficient and that the co-(over)expression of an NADP + reducing enzyme might be required to increase NADPH availability (Shi et al., 2013). Similarly, combined overexpression of the NADK gene from E. coli and the NADP + -GAPDH gene from Bacillus subtilis in E. coli increased 2-chloropropionic acid production yield (Wang et al., 2013c). In addition to increasing NADPH availability indirectly by overexpressing NADK, a more direct approach has also been applied. For example, the production yields of GDP-L-fucose and ε-caprolactone in E. coli were increased by introducing the NADH kinase (EC:2.7.1.86) gene from the yeast S. cerevisiae (Lee et al., 2013a).

Final Remarks and Conclusion
NADPH is an essential reducing agent in all organisms. It plays an important role in various biological processes and is the driving force behind numerous biosynthetic enzymatic reactions of importance to industry. However, efficient NADPH regeneration is often a bottleneck that limits the productivity of such biotransformation processes. A thorough understanding of the reactions involved in the generation of NADPH is thus essential to increase the availability of NADPH and enhance productivity in engineered production strains.
This review has provided an overview of the major canonical NADPH-generating reactions as well as recently described NADPH-generating reactions (Table 1), and it has summarized how these different reactions have been applied to improve NADPH availability. However, additional NADPHgenerating enzymes, such as gluconate 5-dehydrogenase or the recently discovered NADP + -specific electron-bifurcating [FeFe]hydrogenase might also contribute to the overall NADPH supply (Klasen et al., 1995;Wang et al., 2013b). Similarly, enzymes generally considered to consume NADPH or enzymes newly developed to regenerate NADPH, such as glutamate dehydrogenase and F420-dependent NADP + reductases or phosphite dehydrogenase, respectively, can occasionally become important NADPH suppliers (Berk and Thauer, 1997;Lee et al., 2013b;Yokooji et al., 2013).
Although various NADPH-generating enzymes have been overexpressed successfully, no general strategy that guarantees improved NADPH supply exists. Cells usually contain various integrated, network-wide, redox balancing systems, such that engineering of NADPH-generating pathways can result in unexpected metabolic perturbations. Thus, strategies often need to be optimized individually, taking the culture conditions and genetic background of the strains into consideration. Future research involving a systems biology approach is thus required to improve our understanding of the individual NADPH-generating enzymes and their effect on various metabolic backgrounds, in order to predict which approaches will increase NADPH availability in various biotransformation processes.