An “energy‐auxotroph” Escherichia coli provides an in vivo platform for assessing NADH regeneration systems

An efficient in vivo regeneration of the primary cellular resources NADH and ATP is vital for optimizing the production of value‐added chemicals and enabling the activity of synthetic pathways. Currently, such regeneration routes are tested and characterized mainly in vitro before being introduced into the cell. However, in vitro measurements could be misleading as they do not reflect enzyme activity under physiological conditions. Here, we construct an in vivo platform to test and compare NADH regeneration systems. By deleting dihydrolipoyl dehydrogenase in Escherichia coli, we abolish the activity of pyruvate dehydrogenase and 2‐ketoglutarate dehydrogenase. When cultivated on acetate, the resulting strain is auxotrophic to NADH and ATP: acetate can be assimilated via the glyoxylate shunt but cannot be oxidized to provide the cell with reducing power and energy. This strain can, therefore, serve to select for and test different NADH regeneration routes. We exemplify this by comparing several NAD‐dependent formate dehydrogenases and methanol dehydrogenases. We identify the most efficient enzyme variants under in vivo conditions and pinpoint optimal feedstock concentrations that maximize NADH biosynthesis while avoiding cellular toxicity. Our strain thus provides a useful platform for comparing and optimizing enzymatic systems for cofactor regeneration under physiological conditions.

regeneration of NAD(P)H and ATP. Another example is the use of auxiliary substrates that are not converted to biomass but rather fully oxidized to provide the cell with more reducing power and energy, thus enabling a higher fraction of the primary substrate to provide carbon units for biosynthesis (Babel, 2009). This approach is also commonly used in cell-free systems, where the bioconversion process is supported by parallel systems for NAD(P)H and ATP regeneration (Claassens, Burgener, Vogeli, Erb, & Bar-Even, 2019).
To optimize the in vitro regeneration of NAD(P)H and ATP it usually suffices to choose an enzymatic system with a high V max , which can be measured and compared under the specific relevant conditions. However, optimization of in vivo regeneration of NAD(P) H and ATP is more delicate. First, the enzyme kinetic parameters can substantially change within the cellular environment, which is difficult to mimic in in vitro measurements (van Eunen & Bakker, 2014;van Eunen, Kiewiet, Westerhoff, & Bakker, 2012). Moreover, the physiological steady-state concentrations of the substrates and products affect the reaction thermodynamics and kinetics . Finally, overexpressed enzymes, native and heterologous alike, tend to misfold and degrade, thus lowering, sometimes dramatically, the amount of functional protein in the cell (Roodveldt, Aharoni, & Tawfik, 2005). Hence, there is a growing need to establish an in vivo platform to test and compare cofactor regeneration routes.
In this study, we construct an Escherichia coli strain which, when fed on acetate as a carbon source, cannot generate reducing power and energy for cell growth. Under these conditions, the strain depends on the introduction of an independent enzymatic system for NADH regeneration, which could further produce ATP via oxidative phosphorylation. We exemplify the use of this strain as a platform for testing and comparing the ability of different enzymes to produce NADH via the oxidation of either formate or methanol, two commonly used auxiliary substrates (Babel, 2009). We identify the most efficient enzymes under in vivo conditions and pinpoint optimal feedstock concentrations to maximize NADH regeneration rate while minimizing substrate toxicity. The results of this study pave the way for improved utilization of formate and methanol as energy sources for engineered autotrophic and methylotrophic growth of E. coli (Gleizer et al., 2019;Kim et al., 2020) and further provide a useful platform to compare other cofactor regeneration systems within the cellular environment.
PCR reactions for cloning were performed using PrimeSTAR MAX DNA Polymerase (Takara). Restrictions were carried out using Fas-tDigest enzymes and ligations using T4 DNA ligase, all purchased from Thermo Fisher Scientific. Sodium acetate, sodium formate, sodium formate-13 C, methanol, methanol-13 C, and sodium succinate were ordered from Sigma-Aldrich (Steinheim, Germany). D-Glucose, gluconate, xylose, and glycerol were ordered from Carl Roth (Karlsruhe, Germany).

| Bacterial strains
All strains used in this study are listed in Table 1. E. coli SIJ488 (Jensen, Lennen, Herrgard, & Nielsen, 2015), derived from E. coli K12 MG1655, was used as the parental strain for the Δlpd and Δlpd ΔfrmRAB strains. The lpd gene and the frmRAB operon were first deleted by λ-Red recombineering (Jensen et al., 2015) in individual strains. Then, lpd was deleted in the ΔfrmRAB strain thus generating the Δlpd ΔfrmRAB genotype. E. coli K12 MG1655, ΔserA ΔltaE Δkbl ΔaceA (Kim et al., 2020;Yishai et al., 2018) was used for engineering the formate assimilating strain used to test the activity of the methanol dehydrogenases. For all cloning procedures, the E. coli strain DH5α was used.

| Genome engineering
Gene deletions were performed as described before (Wenk, Yishai, Lindner, & Bar-Even, 2018). For λ-Red recombineering in SIJ488, ∼300 ng of a PCR amplified chloramphenicol resistance cassette (Datsenko & Wanner, 2000) with 50 bp homology arms on each side (targeting the up-and downstream region of the gene of interest) were electroporated into the SIJ488 target strain with arabinoseinduced recombineering genes to replace the target gene with the resistance gene. After chloramphenicol selection, gene deletions were confirmed by PCR using verification (ver) primers that bind outside the target locus. The antibiotic marker was removed by inducing flippase expression in the SIJ488 strain. Marker removal was verified by taking colonies that grew only on plates without antibiotics and further confirmed by PCR using ver_primers.

| Gene expression
Expression units (individual genes or synthetic operons) were designed using Geneious 8 (Biomatters, New Zealand) and constructed as described before (Wenk et al., 2018). Genes were expressed with the ribosome binding site C under the control of a constitutive strong pgi promoter and a medium copy number origin of replication (p15A; Wenk et al., 2018). Genes not native to E. coli were codon optimized using JCat (PMID: 15980527) and de novo synthesized. and Candida boidinii (CbFDH; UniProt: O13437) were obtained from collaborators at the Weizmann Institute of Science (Gleizer et al., 2019).
Native E. coli genes-ethanol dehydrogenase (adhP) and acetaldehyde dehydrogenase (mhpF)-were PCR-amplified from the E. coli MG1655 genome. The individual genes were integrated into a high copy number cloning vector pNiv to construct a synthetic operon using the method described previously (Wenk et al., 2018  tion of C≥2 compounds was blocked as we did not observe it experimentally. Based on the WT strain growing solely on acetate, the maximum acetate uptake rate was estimated to be 13 mmol·gCDW −1 ·hr −1 , which is within the range previously reported (Edwards, Ibarra, & Palsson, 2001). When glucose was used as a carbon source, its maximum uptake rate was set to be 10.5 mmol·gCDW −1 ·hr −1 (Varma & Palsson, 1994); this value also matches the growth rate of the Δlpd strain cultivated on glucose and acetate. For the gC1M-gC2M strain, we added to the model the reactions producing serine, glycine and C 1 -tetrahydrofolate and amended the biomass function such that serine, glycine, and C 1 -tetrahydrofolate are consumed instead of 3phosphoglycerate (He et al., 2018). Biomass yields were calculated as the ratio between the growth rate and the uptake/utilization rate of the limiting substrate; the resulting number was converted to units of OD 600 /mM by dividing with 0.39 gCDW/L/OD 600 (Milo, Jorgensen, Moran, Weber, & Springer, 2010). The optimal stoichiometric ratio of substrate coutilization was calculated as the ratio between their uptake/utilization rates. The full code, including changes to the model, reactions specific to each energy module, and the calculations can be found at https://github.com/he-hai/PubSuppl within the "2020_En-ergy_Auxotroph" directory.

| Construction and validation of an E. coli strain auxotrophic to reducing power and energy
Dihydrolipoyl dehydrogenase, encoded by lpd, catalyzes the oxidative regeneration of lipoic acid, as part of three enzyme complexes: pyruvate dehydrogenase, 2-ketoglutarate dehydrogenase, and the glycine cleavage system (Guest & Creaghan, 1972;Pettit & Reed, 1967;Steiert, Stauffer, & Stauffer, 1990). Deletion of lpd is expected to render these enzyme complexes inactive, effectively blocking the known routes for the complete oxidation of acetyl-CoA ( Figure 1): the canonical TCA cycle that relies on the activity of 2-ketoglutarate dehydrogenase, and the alternative PEP-glyoxylate cycle (Fischer & Sauer, 2003) that relies on the activity of pyruvate dehydrogenase. On the other hand, the glyoxylate shunt, which serves to assimilate acetyl-CoA into biomass, is not blocked by this deletion. Hence, the Δlpd strain is expected to be auxotrophic to reducing power and energy when cultivated on acetate: acetate can be activated to acetyl-CoA and assimilated via the glyoxylate shunt but cannot be oxidized to provide the cell with reducing power and energy. Conversely, when provided with carbon sources that enter upper metabolism, the Δlpd strain is expected to behave as an acetyl-CoA auxotroph: reducing power and energy can be obtained by the oxidation of these carbon sources, but the biosynthesis of acetyl-CoA is blocked due to the disruption of pyruvate dehydrogenase.
To test these hypotheses, we constructed the Δlpd strain and characterized its growth on various carbon sources ( Figure 2). As expected, growth on glucose, glycerol, and gluconate-all entering upper metabolism-was made possible only upon addition of acetate. (We note that this finding contrasts a previous study, using a different E. coli strain-BW25113 instead of K-12 MG1655-in which the deletion of lpd was compensated with the activation of pyruvate oxidase, thus providing an alternative route for acetyl-CoA biosynthesis (Li, Ho, Yao, & Shimizu, 2006). On the other hand, growth on carbon sources that enter lower metabolism-pyruvate and succinate-was not observed even upon addition of acetate, as the cells cannot extract reducing power from these feedstocks ( Figure 2). Furthermore, as originally anticipated, growth on acetate was not possible.
To validate that the inability to grow on acetate stems from lack of reducing power, we aimed to cultivate the Δlpd strain on ethanol.
Similarly to acetate, ethanol is assimilated via its conversion to acetyl-CoA. However, in contrast to acetate, metabolism of ethanol to acetyl-CoA produces two NADH molecules that could provide the cell with sufficient reducing power. Yet, E. coli cannot grow on ethanol, due to the low expression levels of adhE, which encodes a bifunctional ethanol dehydrogenase/acetylating aldehyde dehydrogenase, as well as the oxygen sensitivity of the enzyme (Membrillo-Hernandez et al., 2000). We, therefore, chose to overexpress the oxygentolerant enzymes ethanol dehydrogenase AdhP (Thomas et al., 2013) and acetylating aldehyde dehydrogenase MhpF (Fischer et al., 2013), which together are expected to convert ethanol to acetyl-CoA. The Δlpd strain expressing the genes encoding for these two enzymes was able to grow on ethanol as a sole carbon source. (We note that we WENK ET AL. . After a few cultivation cycles in test-tubes, the growth rate and final OD 600 improved considerably. We sequenced the genome and plasmid of the adapted strain and found a single mutation in the mhpF gene on the plasmid: Asp209Asn. Notably, this residue was previously found to be involved in the activation of the active site cysteine in an ortholog enzyme (Manjasetty, Powlowski, & Vrielink, 2003) and hence probably also participates in the catalytic cycle of MhpF. When the mutated plasmid was transferred to a native (nonevolved) Δlpd strain, efficient growth on ethanol was directly observed (Figure 3).  Figure 4a). Addition of formate without FDH expression did not enable growth (gray line in Figure 4a). PsFDH supported a higher growth rate (doubling time of~5 hr) than CbFDH (doubling time of~15 hr). As mentioned above, rather than reflecting only the in vitro measured kinetics of the enzymes, this finding corresponds to F I G U R E 1 Central carbon metabolism of the Δlpd strain. The deletion of the lpd gene renders the pyruvate dehydrogenase complex and the 2-ketoglutarate dehydrogenase complex inactive, interrupting lower (EMP) glycolysis, and the TCA cycle (indicated with red bars). When acetate is provided as a feedstock (dark blue), the strain can assimilate acetyl-CoA into biomass via the glyoxylate shunt (indicated in green) but is expected to be auxotrophic to reducing power and energy since acetyl-CoA oxidation is blocked. On the other hand, the assimilation of ethanol (light blue) produces two molecules of NADH, which should be sufficient to provide the cell with reducing power and energy. Next, we wanted to determine the optimal concentration of formate supporting NADH regeneration. We focused on PsFDH due to its apparent superiority. We cultivated the Δlpd strain overexpressing PsFDH on 20 mM acetate and varying concentrations of formate: from 1.7 mM, increasing stepwise by a factor of 1.5, up to 100 mM. We found that the final OD 600 of the culture increased monotonically with increasing formate concentration (Figure 4b).
This indicates that even at the highest formate concentration tested, the availability of reducing power limits bacterial growth, rather than the supply of carbon from acetate. The doubling time was rather constant (5.2-5.8 hr) at the formate concentration range 10−70 mM (at lower concentrations the doubling time was difficult to define as the culture did not spend sufficient time in exponential phase but rather reached saturation quite fast). A higher formate concentration, 100 mM, lowered the growth rate, probably due to the toxicity of formate (Nicholls, 1975;Warnecke & Gill, 2005). Our results thus indicate an optimal range within which the concentration of formate should be kept to enable E. coli to use this C 1 compound efficiently as a source of reducing power without adverse effects.
We performed FBA using a model of E. coli core metabolism (Orth, Fleming, & Palsson, 2010) to estimate the utilization rates of acetate and formate as well as the expected biomass yields (i.e., final OD 600 per substrate consumed) of the Δlpd strain expressing PsFDH (Section 2). For the observed doubling times (5.2-5.8 hr; Figure 4b), we calculated that the oxidation (i.e., utilization) rate of formate needs to be approximately fivefold higher than the utilization rate of acetate (assuming optimal coutilization of the substrates). Hence, at 20 mM acetate, the tested formate concentrations, 1.7-100 mM, should limit biomass production, as confirmed by the growth experiments ( Figure 4b).
As compared to a WT strain growing on acetate, the Δlpd strain growing with acetate and formate displayed a lower growth rate ( Figure 4), suggesting that the formate oxidation rate is limiting. Indeed, the estimated utilization rate of acetate in a WT strain was

| Testing methanol as a reducing power source
Methanol can be either derived cheaply from fossil methane (Pfeifenschneider, Brautaset, & Wendisch, 2017) or renewably produced via the hydrogenation of CO 2 (Szima & Cormos, 2018). As such, methanol has gained a lot of attention as a microbial feedstock (Antoniewicz, 2019;Schrader et al., 2009) and is actively researched as a source of reducing power and energy (Guo et al., 2019 as well as the endogenous enzymes of the glycine cleavage system ( Figure 6a; Kim et al., 2020). We previously demonstrated that this strain, termed gC 1 M-gC 2 M (Kim et al., 2020), is capable of producing glycine and serine via the assimilation of formate and CO 2 . We also showed that expression of MDH enabled the replacement of formate with methanol in the medium (Kim et al., 2020): methanol is oxidized to formaldehyde via MDH activity, while the endogenous glutathione system metabolizes this highly reactive intermediate to formate (Gutheil, Kasimoglu, & Nicholson, 1997; Figure 6a). Using FBA, we confirmed that the gC 1 M-gC 2 M strain is indeed expected to be considerably more sensitive to methanol than the Δlpd strain, that is requiring substantially lower oxidation rates for growth: while a methanol oxidation rate as low as 1 mmol·gCDW −1 ·hr −1 is expected to support a doubling time of approximately 2 hr of the former strain, an oxidation rate of as high as 5 mmol·gCDW −1 ·hr −1 supports a doubling time of more than 30 hr of the Δlpd strain. Overall, it is clear that most MDH variants are indeed active in vivo. Still, only BsMDH can support a sufficiently high NADH regeneration rate to enable the cell to gain all of its reducing power and energy from methanol (Figure 5a, where the required methanol oxidation rate is more than sixfold higher than in the gC 1 M-gC 2 M strain). This demonstrates that predicting in vivo activity from in vitro kinetic measurements is inaccurate: while BsMDH does not have the best kinetic parameters among the MDH variants tested (Whitaker et al., 2017), it is the only enzyme that could support sufficient NADH regeneration within the cellular context.
Next, we tested the dependency of growth on methanol concentration. We cultivated the Δlpd strain expressing BsMDH on F I G U R E 3 Ethanol metabolism provides both acetyl-CoA and reducing power to the Δlpd strain. The Δlpd strain could grow on 50 mM ethanol, but not on acetate, upon expression of the oxygen-tolerant enzymes ethanol dehydrogenase AdhP and the acetylating aldehyde dehydrogenase MhpF. Shown is the growth of a strain adapted to grow on ethanol (see main text). Experiments were conducted within 96-well plates and were performed in triplicates, which displayed identical growth curves (±5%), and hence were averaged. All experiments were repeated three times, which showed highly similar growth behavior. OD, optical density [Color figure can be viewed at wileyonlinelibrary.com] 20 mM acetate and varying concentrations of methanol: from 26 mM, increasing stepwise by a factor of 1.5, up to 1,500 mM. We found that that the final OD 600 was monotonically increasing with methanol concentrations up to 667 mM. Addition of 1,000 mM methanol resulted in a lower OD 600 and 1,500 mM methanol completely abolished growth (Figure 5b), presumably due to cell dehydration and protein precipitation (Hobro & Smith, 2017) as well as increased levels of formaldehyde, the concentration of which is (thermodynamically) coupled to that of methanol (Kim, Kim, & Oh, 2003). The growth rates associated with the different methanol concentrations were difficult to define as they were not constant but rather seem to decrease during the growth phase. This can be attributed to evaporation of methanol from the culture and to the unfavorable thermodynamics of methanol oxidation (Δ r G' m = +31 kJ/mol [Flamholz et al., 2012]) such that the oxidation rate decreases with the consumption of the substrate. Still, it seems that the highest growth rates were attained at methanol concentrations of 444 and 667 mM, indicating an optimal concentration range in which methanol should be added as a source of reducing power and energy.
Notably, the observed increase in OD 600 with increasing methanol concentration up to 667 mM contradicts the results of a FBA which indicated that the concentration of methanol needs to be only ninefold higher than that of acetate (assuming a representative doubling time of 30-40 hr). This can be explained by noting, as mentioned above, that methanol oxidation is thermodynamically unfavorable, such that very high concentrations, well beyond the stoichiometric requirements, are needed to push the reaction forward. Therefore, growth with methanol is expected to slow down and stop well before it is being depleted. Yet, it could still be the case, just as with formate, that the allocation of methanol-derived NADH is not optimal in the Δlpd strain (i.e., overrespiration at the expense of acetate assimilation), such that methanol remains the limiting substrate even at concentrations well above ninefold that of acetate (>200 mM).
Due to the unfavorable thermodynamics, methanol oxidation depends on the presence of a strong metabolic sink for formaldehyde. As mentioned above, in E. coli, the glutathione system serves to metabolize formaldehyde via the activity of S-(hydroxymethyl)glutathione dehydrogenase and S-formylglutathione hydrolase, encoded by frmA and frmB, respectively (Gutheil et al., 1997). We speculated that deletion of these two genes would abolish growth with methanol due to the accumulation of formaldehyde, which would halt methanol oxidation and, by reacting and modifying macromolecules, would reduce cell viability (Metz et al., 2004). Indeed, a Δlpd ΔfrmRAB strain overexpressing BsMDH was not able to grow on acetate and methanol (Figure 5a), indicating that the metabolic sink for formaldehyde is required for methanol to serve as an electron source.  (Vrtis, White, Metcalf, & van der Donk, 2001).
The Δlpd strain should be considered as auxotrophic to reducing power rather than to NADH alone since the strain can be regarded also as auxotrophic to NADPH, the regeneration of which ultimately relies on the external electron donor. Biosynthesis of NADPH in the Δlpd strain using formate or methanol as electron donors probably takes place mainly via the membrane-bound transhydrogenase (PntAB), which transfers electrons from NADH to NADP + at the expense of a proton imported from the periplasm to the cytosol (Sauer, Canonaco, Heri, Perrenoud, & Fischer, 2004). A reverse electron transfer could also take place, that is, electrons derived from the external electron donor to regenerate NADPH could be transferred to produce NADH via the soluble transhydrogenase (SthA; Sauer et al., 2004). Hence, our "biosensor" strain cannot effectively differentiate between enzymes that regenerate NADH and those producing NADPH, as both could relieve its auxotrophy to reducing power. If such distinction is required, further deletion of sthA would be necessary, such that enzymes that transfer electrons primarily to NADP + could not support the growth of the Δlpd strain.
Identifying and utilizing efficient enzymatic systems for the in vivo regeneration of NAD(P)H can improve microbial growth and bioproduction. This is in line with the auxiliary substrate concept (Babel, 2009). For example, addition of formate was found to improve the growth of Hansenula polymorpha on glucose (Babel, Müller, & Markuske, 1983), Vibrio natriegens on glucose (Linton, Griffiths, & Gregory, 1981)  F I G U R E 6 Most MDH variants are active in vivo. All methanol dehydrogenase variants were expressed and tested in a strain that requires a lower methanol oxidation flux to enable its growth. (a) The gC 1 M-gC 2 M strain (Kim et al., 2020), deleted in the endogenous glycine and serine biosynthesis routes (ΔserA ΔltaE Δkbl ΔaceA), overexpresses Methlybacterium extorquens enzymes that convert formate to methylene-THF (FtfL, Fch, and MtdA) as well as the endogenous enzymes of the glycine cleavage system. Upon methanol oxidation by MDH and conversion of formaldehyde to formate via the glutathione system (FrmAB), the strain could assimilate formate and CO 2 to produce glycine and serine (Kim et al., 2020). Therefore, an active MDH is expected to enable the growth of the strain with methanol. (b) The strain expressing the different MDH variants was tested for growth on 10 mM glucose and 600 mM methanol in a 96-well plate at 10% CO 2 . CnMDH, BsMDH, BmMDH*, and CgMDH restored growth of the strain, although the later variant supported a low growth rate. Upon further addition of 30 mM formate, all strains showed comparable growth, indicating that the failure of BmMDH to support growth is not related to protein burden or accumulation of formaldehyde. Experiments were conducted within 96-well plates and were performed in triplicates, which displayed identical growth curves (±5%), and hence were averaged. Doubling times (DT) are shown in the figure. All experiments (in triplicates) were repeated five times, which showed highly similar growth behavior.