NADPH-related processes studied with a SoxR-based biosensor in Escherichia coli.

NADPH plays a crucial role in cellular metabolism for biosynthesis and oxidative stress responses. We previously developed the genetically encoded NADPH biosensor pSenSox based on the transcriptional regulator SoxR of Escherichia coli, its target promoter PsoxS and eYFP as fluorescent reporter. Here, we used pSenSox to study the influence of various parameters on the sensor output in E. coliduring reductive biotransformation of methyl acetoacetate (MAA) to (R)-methyl 3-hydroxybutyrate (MHB) by the strictly NADPH-dependent alcohol dehydrogenase of Lactobacillus brevis (LbAdh). Redox-cycling drugs such as paraquat and menadione strongly activated the NADPH biosensor and mechanisms responsible for this effect are discussed. Absence of the RsxABCDGE complex and/or RseC caused an enhanced biosensor response, supporting a function as SoxR-reducing system. Absence of the membrane-bound transhydrogenase PntAB caused an increased biosensor response, whereas the lack of the soluble transhydrogenase SthA or of SthA and PntAB was associated with a strongly decreased response. These data support the opposing functions of PntAB in NADP+ reduction and of SthA in NADPH oxidation. In summary, the NADPH biosensor pSenSox proved to be a useful tool to study NADPH-related processes in E. coli.

divergently to the soxS gene. Whereas initial studies indicated that the promoter of soxR is located within the soxS coding region (Wu & Weiss, 1991), a subsequent analysis revealed that the promoters of the two genes overlap (Hidalgo, Leautaud, & Demple, 1998). The transcription factor SoxR is a homodimer with each subunit containing an [2Fe-2S] cluster (Hidalgo & Demple, 1994;Watanabe, Kita, Kobayashi, & Miki, 2008). SoxR activity is controlled by a change of the redox state of its [2Fe−2S] clusters, which is associated with conformational changes: only in the oxidized [2Fe−2S] 2+ state, but not in the reduced [2Fe−2S] + state, SoxR activates soxS expression (Ding, Hidalgo, & Demple, 1996;Gaudu & Weiss, 1996). SoxR binds to its target site, which is located between the −10 and −35 regions of the soxS promoter and downstream of the −10 region of the soxR promoter, both in the oxidized and in the reduced state with high affinity (Gaudu & Weiss, 1996;Hidalgo & Demple, 1994). Besides activating soxS expression in the oxidized state, SoxR simultaneously represses expression of its own gene, both in the oxidized and the reduced state (Hidalgo et al., 1998). SoxR was previously considered to activate expression of soxS only, but recent studies uncovered further direct SoxR target genes (Seo, Kim, Szubin, & Palsson, 2015).
SoxS functions as a transcriptional activator of genes, many but not all of which are responsible for coping with damage caused by oxygen radicals, such as sodA for superoxide dismutase, zwf for the NADPH-generating glucose 6-phosphate dehydrogenase, or fumC for fumarase C (Blanchard, Wholey, Conlon, & Pomposiello, 2007;Seo et al., 2015). It has been shown that the intrinsic instability of SoxS (t 1/2 ~ 2 min) and the degradation of SoxS, primarily through the Lon protease, are responsible for the rapid return of the SoxRS system to the inactive state when the stimuli activating the system are no longer present (Griffith, Shah, & Wolf, 2004).
Current evidence indicates that there are multiple ways how the conversion of inactive, reduced SoxR into active, oxidized SoxR can be triggered. These include direct oxidation of SoxR by superoxide (Fujikawa, Kobayashi, & Kozawa, 2012;Liochev & Fridovich, 2011) and by redox-cycling drugs (Gu & Imlay, 2011), nitrosylation of SoxR (Ding & Demple, 2000), and conditions leading to a diminished NADPH/NADP + ratio within cells (Krapp, Humbert, & Carrillo, 2011;Liochev & Fridovich, 1992). The responsiveness to the NADPH availability is presumably due to the fact that SoxR is subject to permanent autoxidation under aerobic conditions, but is kept in the reduced state by NADPH-dependent reductases (Koo et al., 2003).
In a previous study, we made use of the NADPH-responsiveness of the SoxRS system to construct the biosensor pSenSox, in which the SoxR-activated soxS promoter controls expression of the eyfp gene, allowing detection of SoxR activation at the single-cell level (Siedler et al., 2014). Using the reduction of methyl acetoacetate (MAA) to (R)-methyl 3-hydroxybutyrate (MHB) by the strictly NADPH-dependent alcohol dehydrogenase of Lactobacillus brevis (LbAdh) as model reaction, we could show that the specific eYFP fluorescence of E. coli cells correlated not only with the MAA concentration added to the cells, but also with the specific LbAdh activity when a fixed MAA concentration was provided. The latter property enabled high-throughput screening of an LbAdh mutant library by fluorescence-activated cell sorting (FACS) for variants with improved activity for the alternative substrate 4-methyl-2-pentanone (Siedler et al., 2014).
In this study, we employed pSenSox to test various conditions, including growth media, redox-cycling drugs, mutants lacking SoxR reductases, and mutants lacking transhydrogenases for their influence on SoxR activity.

| Influence of different media on the NADPH biosensor response
To test the influence of different media on the response of the pSen-Sox-based NADPH biosensor, the biotransformation of MAA to MHB, catalyzed by the NADPH-dependent LbAdh, was performed in three complex media (TB,2xTY,or LB) and in a defined minimal medium (M9) with glucose as carbon source using the experimental setup shown in Figure 1 and as described in the methods section.
The experiments with the different media, including control cultures in which MAA was omitted or in which pSenNeg, encoding a defective LbAdh, was used, are shown in Figure 2. MAA itself had a negative influence on growth, even in the absence of LbAdh, and this negative influence was further enhanced in the presence of LbAdh activity, when MAA was reduced to MHB with NADPH as reductant. Regarding the response of the SoxRS-based NADPH biosensor, the experiments shown in Figure 2 confirmed that expression of the eyfp gene is dependent on the biotransformation of MAA to MHB by the LbAdh. In the absence of either MAA or LbAdh activity, eYFP synthesis was not induced.
When comparing the different media, it became obvious that TB allowed by far the best growth, followed by 2xTY and LB medium, in which the cells grew comparably, and M9 glucose medium, in which almost no growth occurred ( Figure A1). When comparing the different media with respect to eYFP synthesis, the highest fluorescence after 24 hr was obtained in 2xTY medium and LB medium, whereas it was much lower and comparable for TB and M9-glucose medium ( Figure A1). The almost complete lack of growth in M9-glucose was due to the biotransformation of MAA to MHB, as growth was observed in the absence of MAA or of LbAdh activity ( Figure 2). In this medium, cells have to synthesize all cellular components, in particular amino acids, from glucose, whereas in the other media the presence of yeast extract and tryptone provides amino acids and other cellular components that do not need to be synthesized by the cell but can be imported from the medium. Nevertheless, also in these media the NADPH-dependent reduction of MAA to MHB had a strong negative effect on growth, presumably due to a lack of NADPH for biosynthetic purposes. An interesting case is TB medium. Although this medium allowed the best growth, the biosensor response was much lower compared to 2xTY or LB and similar to that in M9-glucose medium. Besides a higher concentration of yeast extract and phosphate buffering, the major difference of TB medium to 2xTY and LB is the presence of glycerol as additional carbon source.
In conclusion, media without a separately added carbohydrate as carbon source, such as 2xTY and LB, led to a higher biosensor response than media containing an added carbohydrate, such as M9-glucose or TB, which contains 4 ml/L glycerol. This is probably due to a higher NADPH availability by carbohydrate catabolism. M9 glucose medium can in principle be used to monitor the SoxR-based NADPH biosensor response, which can be necessary or advantageous for experiments in which components of yeast extract or tryptone are disturbing. Overall, the strongest biosensor signal was observed in 2xTY medium, which was therefore chosen for the following experiments.

| Influence of redox-cycling drugs on the NADPH biosensor response
Paraquat (1,1′-dimethyl-4,4′-bipyridinium dichloride) and menadione (2-methyl-1,4-naphthoquinone) have been reported to induce the soxRS regulon in E. coli (Greenberg et al., 1990;Seo et al., 2015;Wu & Weiss, 1991). We therefore monitored the response of E. coli BL21(DE3)/pSenSox to different paraquat concentrations (0, 1, 5 µM) and different menadione concentrations (0, 5, 10 µM) using the experimental setup shown in Figure 1, except that paraquat and menadione were added instead of MAA. At the F I G U R E 1 (a) NADPH-dependent reduction of MAA to MHB by the LbAdh. (b) Experimental setup used in this work to study the responses of the SoxR-based NADPH biosensor encoded by plasmid pSenSox during whole-cell biotransformation of MAA to MHB. (1) Escherichia coli BL21(DE3) carrying pSenSox was cultivated in shake flasks with a starting OD 600 of 0.05. (2) If desired, IPTG was added to the cultures at an OD 600 between 0.6 and 0.8. In the absence of IPTG, basal expression of the Lbadh gene by the tac promoter is sufficient for LbAdh synthesis and biotransformation of MAA to MHB. However, IPTG can be added to maximize Lbadh expression. (3) To ensure that enough biomass is formed for the biotransformation, the cultures were grown for at least 5 hr until an OD 600 of 5 or higher was reached.
(4) For the biotransformation, 800 µl of the culture with an OD 600 adjusted to 5 were transferred into a Flowerplate (m2p-labs, Baesweiler, Germany) and the biotransformation was started by the addition of 100 µl MAA at the desired concentration to these cultures. To study the effect of redox-cycling drugs, either 100 µl paraquat or 100 µl menadione were added to the cultures at the desired concentration.
(5) The change in eYFP fluorescence and biomass was monitored for around 24 hr with a BioLector microcultivation system that enables online recording of eYFP fluorescence (excitation at 485 nm, emission at 520 nm) and biomass gain (change in cell density measured as backscattered light at 620 nm (Kensy, Zang, Faulhammer, Tan, & Büchs, 2009  (hr) (hr) lead to a decreased NADPH/NADP + ratio, thereby interfering with the NADPH-dependent reduction of SoxR and causing an increased level of oxidized, active SoxR; (d) the redox-cycling drugs might be directly reduced by the SoxR-reducing system(s) of the cell, thereby inhibiting SoxR reduction and causing increased levels of oxidized active SoxR. It is possible that several of these mechanisms contribute to the activation of SoxR by paraquat and menadione.

| Influence of rseC and rsxABCDGE deletion on the NADPH biosensor response
By screening an E. coli mutant library, mutations in the rseC gene and in the rsxABCDGE operon were found to cause constitutive expression of a P soxS -lacZ reporter gene in a SoxR-dependent manner (Koo et al., 2003). Further studies led to the conclusion that the membrane-integral RsxABCDGE complex and the membrane protein RseC constitute a SoxR-reducing system (Koo et al., 2003). The statement that purified RsxC exhibits NADPH-dependent cytochrome c reduction activity (Koo et al., 2003) suggests that NADPH serves as electron donor of the Rsx complex. To test the influence of this To confirm that the observed effects were due to the gene deletions, plasmids pACYC-rseC, pACYC-rsx, and pACYC-rseC-rsx were constructed and transferred into the corresponding deletion mutants. The parent vector pACYCDuet-1 served as control. Basal expression of rseC and/or rsxABCDGE in the respective deletion mutants without addition of IPTG resulted in decreased biosensor signals compared to the ones obtained in the mutants carrying the control plasmid pACYCDuet-1, but the response was still higher than in the parental strain ( Figure A3). Although complementation was only partial, which could be due to an inadequate expression level of the plasmid-encoded genes, it confirmed that the rseC and rsx deletions were responsible for the increased biosensor response.
The results described above are in agreement with previous data showing a function of RsxABCDGE and RseC in SoxR reduction (Koo et al., 2003). An interesting observation made by Koo and coworkers and by us was that the deletion of both rseC and the rsx cluster had no additive effect and expression of the reporter gene was even somewhat lower in the ΔrseCΔrsx mutant than in the ΔrseC single mutant. This indicates that the Rsx complex and RseC do not function independently to reduce SoxR, but work together, as proposed previously (Koo et al., 2003). The RsxABCDGE complex belongs to the family of Rnf complexes, enzymes that drive the endergonic reduction of ferredoxin (E 0 ′ = −420 mV) with NAD(P)H (E 0 ′ = −320 mV) by the proton-or sodium-motive force via import of H + or Na + , or, in the reverse reaction, the exergonic reduction of NAD(P) + with reduced ferredoxin coupled to the export of H + or Na + (Biegel, Schmidt, Gonzalez, & Müller, 2011). The redox potential of SoxR in its DNA-free and its DNA-bound state was reported to be −293 and −320 mV (Kobayashi, Fujikawa, & Kozawa, 2015), respectively, that is, in the same range as the one of NAD(P)H. Consumption  of the protein was present in the reduced state, but in rsxC and rseC mutants only about 60% and 56% (Koo et al., 2003). The final steps of electron transfer to SoxR are unknown at present. In the Rnf complexes, RnfB is suggested as electron donor for ferredoxin (Biegel et al., 2011). Therefore, the homologous RsxB protein could serve to reduce SoxR, or alternatively, RseC might transfer electrons from RsxB to SoxR, as a conserved cysteine motif in the N-terminal region of RseC could be part of an iron-sulfur cluster. Further studies are required to solve this issue. The observation that in rsxC and rseC mutants still about 60% and 56% of SoxR was in the reduced state (Koo et al., 2003) suggests that further enzymes for SoxR reduction exist, which need to be identified.
We also tested the response of the rsx and rseC mutants in the presence of 5 µM paraquat instead of MAA. Although the parental strain already showed a strong response to paraquat, that of the mutants was still further increased. Again, the ΔrseC mutant showed the highest specific fluorescence followed by the ΔrseCΔrsx mutant and the ΔrsxABCDGE mutant (Figure 4c,d). The observation that paraquat and rsx and/or rseC deletion showed an additive effect on the biosensor response confirms that a fraction of SoxR must still be in the reduced state in the mutants and suggests that paraquatbased activation of SoxR is not due to interference with SoxR reduction by the Rsx/RseC system. Due to the relevance of transhydrogenases in the regulation of cellular NADPH levels, we studied the influence of these enzymes on the NADPH biosensor response by constructing deletion mutants of E. coli BL21(DE3) lacking either sthA, or pntAB, or both.

| Influence of the transhydrogenase deletions
The growth behavior of the transhydrogenase mutants was tested in shake flask experiments using 2xTY medium. Whereas the ΔpntAB mutant grew like the parental strain, the ΔsthA mutant and the ΔsthAΔpntAB double mutant showed a growth defect that became apparent during the exponential growth phase ( Figure A4).
Presumably, an excess of NADPH is formed in this growth phase, which cannot be readily diminished in the absence of SthA. The defect could be largely abolished by plasmid-based expression of sthA ( Figure A5), confirming that it was caused by the sthA deletion.
The influence of SthA and PntAB on the NADPH biosensor signal was analyzed according to the standard experimental setup shown in Figure 1 with 30 mM MAA as substrate for the NADPH-dependent LbAdh. Whereas the ΔpntAB mutant displayed a slightly increased biosensor signal, it was decreased by more than 60% in the ΔsthA mutant and also in the ΔsthAΔpntAB double mutant ( Figure 5). The latter result showed that the sthA deletion was dominant over the pntAB deletion.
To confirm that the observed effects were due to the gene deletions, plasmids pACYC-pntAB and pACYC-sthA were constructed and transferred into the corresponding deletion mutants. The results obtained with the ΔpntAB mutant indicate that, under the conditions used, PntAB catalyzes NADP + reduction by NADH, leading to an increased NADPH availability. Absence of pntAB therefore results in a lowered NADPH availability and thus an increased biosensor signal. This conclusion is in agreement with previous studies showing that PntAB is involved in NADPH formation in E. coli (Sauer et al., 2004) and that overexpression of pntAB enhanced conversion of acetophenone to (R)-phenylethanol by the NADPH-dependent alcohol dehydrogenase of Lactobacillus kefir (Weckbecker & Hummel, 2004) and improved the biosynthesis of 3-hydroxypropionic acid from its precursor malonyl-CoA by an NADPH-dependent malonyl-CoA reductase (Rathnasingh et al., 2012). Moreover, heterologous overexpression of the E. coli pntAB genes in Corynebacterium glutamicum was shown to enhance production of l-lysine, whose biosynthesis is strongly NADPH-dependent (Kabus, Georgi, Wendisch, & Bott, 2007). availability, which is reflected by a decreased biosensor signal. These data are in agreement with previous studies showing that SthA is required under conditions leading to excess NADPH formation (Sauer et al., 2004). The observation that pntAB deletion in the ΔsthA mutant did not reverse the decrease in the biosensor signal suggests that SthA activity is much higher than PntAB activity, which can be expected based on the fact that PntAB catalyzes a reaction coupled to proton transfer across the membrane. In conclusion, our data confirm that PntAB and SthA play important and opposite functions for NADPH availability in E. coli. PntAB and SthA had opposite effects on the biosensor response, in agreement with PntAB being involved in NADP + reduction and SthA catalyzing NADPH oxidation. In conclusion, the pSenSox-based NADPH biosensor is a useful tool not only for HT-screening of the activity of NADPH-dependent alcohol dehydrogenases (Siedler et al., 2014), but also for analyzing conditions and proteins influencing NADPH availability or SoxR reduction. Recently, another NADPH biosensor was described, which is based on the specific oxygen-independent amplification of the intrinsic fluorescence of NADPH by the mBFP protein (Hwang, Choi, Han, & Kim, 2012). The mBFP protein was shown to be well suited to study the dynamics of intracellular NADPH availability with a resolution of seconds and to allow the quantitation of NADPH (Goldbeck, Eck, & Seibold, 2018). This is not possible with the pSenSox NADPH biosensor, which requires transcription, translation, and oxygen-dependent maturation of eYFP. However, pSenSox allows to preserve changed NADPH levels as a stable fluorescence signal, which is a prerequisite, for example, for FACS-based screening of mutant libraries of NADPH-dependent enzymes, and it allows to specifically analyze the effects of enzymes that are involved in SoxR reduction or oxidation, which is presumably not possible with the mBFP sensor. Thus, mBFP and pSenSox represent two different types of NADPH biosensors, which both have their specific advantages and application fields.

| Recombinant DNA work and construction of deletion mutants
Standard methods such as PCR and DNA restriction enzyme digestion were carried out according to established protocols (Sambrook & Russell, 2001). Oligonucleotides were synthesized by Eurofins Genomics (Ebersberg, Germany) and are listed in Table   A1. Construction of pACYC-rseC, pACYC-rsx, pACYC-rseC-rsx, pACYC-sthA, and pACYC-pntAB was performed by Gibson assembly (Gibson et al., 2009). All plasmids were sequenced by Eurofins Genomics, Ebersberg, Germany. The construction of E.
coli deletion mutants was performed with the lambda Red recombinase method (Datsenko & Wanner, 2000) using a recent protocol (Jensen, Lennen, Herrgard, & Nielsen, 2015). For the markerless deletion of genes, the arabinose-inducible lambda Red recombineering genes (exo, bet, and gam) and the rhamnose-inducible flippase (FLP) recombinase were introduced into E. coli BL21(DE3) using the temperature-sensitive plasmid pSIJ8 (Jensen et al., 2015). The FRT-flanked kanamycin cassette, which was integrated into the E. coli genome at the locus of the gene to be deleted, was encoded by plasmid pKD4. For amplification of the FRT-flanked kanamycin cassette of pKD4, oligonucleotides that carried homology regions to the up-and downstream regions of the genes to be deleted were used. All gene deletions were verified by colony PCR using DreamTaq Master Mix 2X (Thermo Scientific, Schwerte, Germany) and the oligonucleotides listed in Table A1.

| Monitoring the NADPH biosensor response
The NADPH biosensor response during the whole-cell biotransformation of MAA to MHB by the strictly NADPH-dependent LbAdh was measured as described (Siedler et al., 2014).

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

E TH I C S S TATEM ENT
None required.

DATA ACCE SS I B I LIT Y
All data are included in the article.