Accelerated Electro-Fermentation of Acetoin in Escherichia coli by Identifying Physiological Limitations of the Electron Transfer Kinetics and the Central Metabolism

Anode-assisted fermentations offer the benefit of an anoxic fermentation routine that can be applied to produce end-products with an oxidation state independent from the substrate. The whole cell biocatalyst transfers the surplus of electrons to an electrode that can be used as a non-depletable electron acceptor. So far, anode-assisted fermentations were shown to provide high carbon efficiencies but low space-time yields. This study aimed at increasing space-time yields of an Escherichia coli-based anode-assisted fermentation of glucose to acetoin. The experiments build on an obligate respiratory strain, that was advanced using selective adaptation and targeted strain development. Several transfers under respiratory conditions led to point mutations in the pfl, aceF and rpoC gene. These mutations increased anoxic growth by three-fold. Furthermore, overexpression of genes encoding a synthetic electron transport chain to methylene blue increased the electron transfer rate by 2.45-fold. Overall, these measures and a medium optimization increased the space-time yield in an electrode-assisted fermentation by 3.6-fold.


Introduction
Sustainable bioproduction will be the key to reaching the goal of a socio-economic transition into a bioeconomy. A low energy gain for the organisms and consequently a high catabolic production rate accompanied by low anabolic substrate conversion is favorable for fermentation-based biotechnological processes. This can be achieved under anoxic conditions. However, anoxic fermentations for instance the conversion of glucose to lactic acid or ethanol and carbon dioxide come with the boundary condition that the products must have the same oxidation state as the substrate. A more oxidized end-product demands the addition of an electron acceptor as oxygen. An elegant trade-off between both strategies was presented by Causey and colleagues. In this study, they disconnected oxygen reduction from energy generation by deleting parts of the membrane-bound ATPase (∆atpFH) [1]. Although this concept solves the problem of an increasing biomass production rate, it cannot prevent the dissipation of a not negligible amount of energy in the form of heat.  [22]. Green shapes represent the heterologously expressed enzymes AlsS and AlsD, grey shapes are native enzymes of E. coli and hollow shapes and faded reactions indicate pathways that are blocked by gene deletions.
This study reports on several measures that were taken to increase the fitness of the strain under anoxic conditions and to increase the methylene blue reduction rates. We discovered-using adaptation experiments and resequencing of the resulting strain-mutations that increased the fitness of the strain under anoxic respiratory conditions. The genes identified in this experiment hint towards an anabolically constraint metabolism. Moreover, overexpressing key-genes for the electron transfer process we increased the methylene blue reduction rate of the organisms. In bioelectrochemical experiments, we demonstrate the success of the approach by increased production rates.

Vector and Strain Construction
The copy number of the cymA and stc gene in the E. coli strain JG1397 (Table 1) was increased by cloning the two genes behind the alsSD genes in the pMAL plasmid derived from Förster et al. [22] ( Table 2). The cymA and stc genes were first codon-optimized for use in E. coli and amplified using Primer 1 and 2 ( Table 3, Appendix A Sequence A4 and A5). Amplification also led to the elongation of the PCR fragment by homologous regions to the pMAL_alsSD plasmid. After cleaving this plasmid  [22]. Green shapes represent the heterologously expressed enzymes AlsS and AlsD, grey shapes are native enzymes of E. coli and hollow shapes and faded reactions indicate pathways that are blocked by gene deletions.

Vector and Strain Construction
The copy number of the cymA and stc gene in the E. coli strain JG1397 (Table 1) was increased by cloning the two genes behind the alsSD genes in the pMAL plasmid derived from Förster et al. [22] ( Table 2). The cymA and stc genes were first codon-optimized for use in E. coli and amplified using Primer 1 and 2 ( Table 3, Appendix A Sequence A4 and A5). Amplification also led to the elongation of the PCR fragment by homologous regions to the pMAL_alsSD plasmid. After cleaving this plasmid with SalI (New England Biolabs) the PCR fragment was inserted via isothermal DNA assembly [23]. The resulting plasmid was sequenced and transformed into JG1352 together with pEC86.

Adaptive Laboratory Evolution (ALE)
ALE selection for an accelerated anaerobic metabolism with DMSO was performed in Hungate tubes containing 10 mL MOPS-buffered medium (137 mmol/L NaCl, 2.7 mmol/L KCl, 15 mmol/L MOPS, 9 mmol/L (NH 4 ) 2 SO 4 , 1 mmol/L MgSO 4 , 0.1 mmol/L CaCl 2 , 10 mL/L trace element solution, 1 g/L casein hydrolysate or yeast extract). The medium was supplemented with 20 mmol/L Glucose, 40 mmol/L DMSO, 100 µmol/L methylene blue, and 14.8 µmol/L thiamine. The tubes were then inoculated with JG991 to an initial OD 600 = 0.05. These cultures were incubated at 30 • C and transferred into new tubes when the suspension became visibly turbid (OD 600 ≈ 0.3). For this purpose, the optical density was determined in all approaches at the beginning and end of each transfer and the number of generations was calculated (n = log 2 N − log 2 N 0 , n = number of generations, N = cell number). The next set of culture tubes was then inoculated with 1 mL culture of the tube with the highest number of cell divisions.

Preparation of Nucleic Acids
The preparation of plasmids and chromosomal DNA was carried out using 2 mL of an aerobic culture in LB medium. The plasmids were isolated using the Wizard Plus Miniprep DNA Purification System (Promega, Madison, WI, USA). The Wizard Genomic DNA Purification Kit (Promega, Madison, WI, USA) was used to isolate chromosomal DNA. Deviating from the protocol, the isolated DNA was rehydrated in the last step in 75 µL Tris/HCl buffer (15 mmol/L). Nucleic acids were quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Chromosomal DNA was further quantified using the QBit dsDNA BR Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA).

Sequencing
Plasmid DNA was sequenced using the Mix2SeqKit (Eurofins Genomics, Ebersberg, Germany). The genomes of JG1295 and JG991 were resequenced using an Illumina MiSeq with v2 chemistry (Illumina, San Diego, CA, USA) with 2 × 250 bp reads for the variant analysis. The library was created using the NEBNext ® Ultra™ II FS DNA kit. Illumina sequencing was carried out by IMGM Laboratories GmbH (Martinsried, Germany).

Bioinformatic Analysis
The analysis of sequencing data was conducted using the CLC Genomics Workbench 11 (Qiagen, Venlo, The Netherlands). The workflow for NGS data contained: merging forward and reverse reads, a quality trim (p = 0.05), and mapping reads to the reference genome. The consensus sequence of the reference genome resequencing has been extracted by majority vote. This consensus sequence has been used as the reference for a variant analysis of the strain obtained from the ALE experiment. Variants with a fidelity >90% and a coverage of 20× were regarded as significant.

Electro-Fermentation
In brief, the bioelectrochemical system had a volume of 23 mL. The working electrode was a graphite felt (40 cm 2 surface, GFD 2.5, Sigracell, Germany), and the counter electrode a platinum mesh (1.25 cm 2 , chemPUR, Karlsruhe, Germany). An Ag/AgCl electrode (Xylem group, Germany) was used as a reference. The anode potential was poised to 0 mV against Ag/AgCl. The detailed procedure including schematics of the reactor system can be found in Förster et al. [22]. Differing from that, a MOPS buffered medium was used supplemented with 0.1% yeast extract and 10 or 50 mmol/L Glucose. The pH value of 7.2 was maintained using a pH-controller (Hach Lange, Düsseldorf, Germany) feeding a 0.5 mmol/L NaOH solution. The electrical current was recorded for 8 h. Of note, the doubling time of the organisms. under these conditions will be at least 20 h ( Figure 2). transformation of methylene blue reduction kinetics into current densities does not account for a loss of mediator over time in the BES (bioelectrochemical system), it also neglects possible mass transport limitations with regards to substrate uptake. Moreover, the rather fast initial methylene blue reduction kinetics will be hampered by the growth inhibiting effect of this electron shuttle that can also pass the inner membrane [26,27]. Still, the comparison of the data achieved from electrofermentation experiments with these theoretical values reveals to which account the mentioned factors indeed limit the overall process and give a clear maximum for the achievable current density with the initial strain's physiology.

Adaptation to Improved Anoxic Respiration
Previous experiments revealed that the deletion of all fermentation pathways in the developed production strain led to a severe limitation of growth also under anoxic respiratory conditions. The physiological reason for this remained at least partly enigmatic. The deletion of ack/pta will lead to a reduced production of ATP that will influence the growth rate and yield. Interestingly, when strains with an increasing number of deletions in fermentation pathways were compared, it was observed that the growth rate was already decreased to 17% of the initial level after the first two deletions steps (frd and adh gene). Further, the deletion of ldh lowered the growth rate to 8% of the initial level and the loss of ack/pta to 5%. In contrast, the deletions of frd, adh, and ldh combined reduced the biomass yield to 50% of the initial level, while the additional deletion of the pta/ack locus led to 15% of the initial value [22]. Albeit, the growth and yield data suggest that the deletion of NADH reoxidation pathways seems to have a general dysfunctional central metabolism as consequence. Hence, we aimed for an adaptation experiment and hypothesized that consecutive transfers on the anoxic medium would select for mutations that lead to a more balanced central metabolism and consequently faster growth.
The medium for the selection experiment was altered compared to our previous study since growth experiments revealed that the addition of yeast extract instead of casamino acids can increase anoxic growth by the strain (Appendix A Figure A1). Moreover, we aimed for optimizing the medium concerning the later application in a bioelectrochemical system. Hence, we determined the effect of either phosphate-, HEPES-or MOPS-buffer on the electrochemical interaction of methylene blue with

Analytical Methods
The chemical analysis of the growth experiments and cell suspension assays was conducted via HPLC (Ultimate 3000 DAD and RI-Detector, Thermo Scientific, Waltham, MA, USA) with an Aminex HPX 87-H column (Bio-Rad, Hercules, CA, USA) at 60 • C with 5 mmol/L H 2 SO 4 as the mobile phase. Acetoin was also quantified using the Voges-Proskauer reaction as described by Förster et al. [22].

Calculation of Evaluation Parameters
The electro-fermentation experiments have been evaluated according to the average current density within the 8 h time frame of the experiment (j), coulombic efficiency (CE), yield (Y), and productivity (Q). The parameters have been calculated following these formulas: I: electrical current, A: electrode surface, t: time, V r : reactor volume, z e : electrons per reaction, c: concentration, : elementary charge, M: Molar Mass, Ace: Acetoin, Gluc: Glucose, N A : Avogadro constant.

Results and Discussion
In a previous study, we established the production of acetoin from glucose in an electrode-assisted fermentation [22]. The addition of methylene blue was necessary to allow electron transfer from the periplasm to the anode surface. While the carbon efficiency set a new benchmark, the space-time yield was rather low compared to oxic processes. Hence, the initial aim of this study was to elucidate the theoretically achievable conversion rate based solely on the kinetics of methylene blue reduction by the strain. Therefore, K M and V max of methylene blue reduction were determined in a microtiter-based reduction assay. The measured methylene blue reduction rate for the K M value of 273 µmol/L was 1.4 µmol/L/s. In comparison, the rate for 50 µmol/L was 0.45 µmol/L/s ( Figure 2). This value is equivalent to a theoretical current density of 498.35 mA/m 2 under conditions applied in the experiment of Förster et al. [22]. In said experiment, however, only 115.8 ± 6.7 mA/m 2 were measured. Hence, if the reduction rate of the production strain in an electro-fermentation setup could converge to the rates of this idealized reduction assay, an increase in the metabolic activity of up to 4.3-fold could be achieved. For a methylene blue concentration of 273 µmol/L which equals the K M value of the cellular system an increase of up to 13.9-fold might be possible. Still, this theoretical transformation of methylene blue reduction kinetics into current densities does not account for a loss of mediator over time in the BES (bioelectrochemical system), it also neglects possible mass transport limitations with regards to substrate uptake. Moreover, the rather fast initial methylene blue reduction kinetics will be hampered by the growth inhibiting effect of this electron shuttle that can also pass the inner membrane [26,27]. Still, the comparison of the data achieved from electro-fermentation experiments with these theoretical values reveals to which account the mentioned factors indeed limit the overall process and give a clear maximum for the achievable current density with the initial strain's physiology.

Adaptation to Improved Anoxic Respiration
Previous experiments revealed that the deletion of all fermentation pathways in the developed production strain led to a severe limitation of growth also under anoxic respiratory conditions. The physiological reason for this remained at least partly enigmatic. The deletion of ack/pta will lead to a reduced production of ATP that will influence the growth rate and yield. Interestingly, when strains with an increasing number of deletions in fermentation pathways were compared, it was observed that the growth rate was already decreased to 17% of the initial level after the first two deletions steps (frd and adh gene). Further, the deletion of ldh lowered the growth rate to 8% of the initial level and the loss of ack/pta to 5%. In contrast, the deletions of frd, adh, and ldh combined reduced the biomass yield to 50% of the initial level, while the additional deletion of the pta/ack locus led to 15% of the initial value [22]. Albeit, the growth and yield data suggest that the deletion of NADH reoxidation pathways seems to have a general dysfunctional central metabolism as consequence. Hence, we aimed for an adaptation experiment and hypothesized that consecutive transfers on the anoxic medium would select for mutations that lead to a more balanced central metabolism and consequently faster growth.
The medium for the selection experiment was altered compared to our previous study since growth experiments revealed that the addition of yeast extract instead of casamino acids can increase anoxic growth by the strain (Appendix A Figure A1). Moreover, we aimed for optimizing the medium concerning the later application in a bioelectrochemical system. Hence, we determined the effect of either phosphate-, HEPES-or MOPS-buffer on the electrochemical interaction of methylene blue with the electrode using cyclovoltammetry. The experiments revealed that the highest current could be observed using MOPS-buffer, while the application of HEPES and phosphate-buffer led to 25% and 12% reduced peak currents (Appendix A Figure A2a). Moreover, the addition of yeast extract to the medium did not influence the methylene blue redox reaction in the conducted cyclovoltammetry experiments (Appendix A Figure A2b).
With this optimized medium, we started the selection experiment using DMSO as the electron acceptor and glucose as the electron donor. DMSO was chosen as its reduction is dependent on menaquinol which is also the electron donor for the heterologously produced CymA from S. oneidensis. Production of the latter is key for the introduced synthetic electron transport chain to methylene blue. After 16 transfers, the growth rate increased 3-fold from 0.016 ± 0.003 for the progenitor strain JG991 to 0.050 ± 0.001 for the developed strain, which will be referred to as JG1295 (Figure 3a,b). Of note, we continued the experiment for 7 further transfers but could not observe an increase in growth past transfer 16. It is further notable, that the lag phase has been reduced from approximately 100 h (JG991) to less than 24 h (JG1295). Still, an E. coli control strain that is not fermentation deficient shows no apparent lag phase and grows twice as fast as JG1295. Overall, the conducted adaptation reduced the gap in growth rate between control and adapted strain from 6-fold to 2-fold. Of note, the observed effect is specific for anaerobic growth since a control experiment under oxic conditions did not reveal a statistically significant difference between the growth rates of strain JG991 and JG1295 (Figure 4). With this optimized medium, we started the selection experiment using DMSO as the electron acceptor and glucose as the electron donor. DMSO was chosen as its reduction is dependent on menaquinol which is also the electron donor for the heterologously produced CymA from S. oneidensis. Production of the latter is key for the introduced synthetic electron transport chain to methylene blue. After 16 transfers, the growth rate increased 3-fold from 0.016 ± 0.003 for the progenitor strain JG991 to 0.050 ± 0.001 for the developed strain, which will be referred to as JG1295 (Figure 3a,b). Of note, we continued the experiment for 7 further transfers but could not observe an increase in growth past transfer 16. It is further notable, that the lag phase has been reduced from approximately 100 h (JG991) to less than 24 h (JG1295). Still, an E. coli control strain that is not fermentation deficient shows no apparent lag phase and grows twice as fast as JG1295. Overall, the conducted adaptation reduced the gap in growth rate between control and adapted strain from 6fold to 2-fold. Of note, the observed effect is specific for anaerobic growth since a control experiment under oxic conditions did not reveal a statistically significant difference between the growth rates of strain JG991 and JG1295 (Figure 4).   With this optimized medium, we started the selection experiment using DMSO as the electron acceptor and glucose as the electron donor. DMSO was chosen as its reduction is dependent on menaquinol which is also the electron donor for the heterologously produced CymA from S. oneidensis. Production of the latter is key for the introduced synthetic electron transport chain to methylene blue. After 16 transfers, the growth rate increased 3-fold from 0.016 ± 0.003 for the progenitor strain JG991 to 0.050 ± 0.001 for the developed strain, which will be referred to as JG1295 (Figure 3a,b). Of note, we continued the experiment for 7 further transfers but could not observe an increase in growth past transfer 16. It is further notable, that the lag phase has been reduced from approximately 100 h (JG991) to less than 24 h (JG1295). Still, an E. coli control strain that is not fermentation deficient shows no apparent lag phase and grows twice as fast as JG1295. Overall, the conducted adaptation reduced the gap in growth rate between control and adapted strain from 6fold to 2-fold. Of note, the observed effect is specific for anaerobic growth since a control experiment under oxic conditions did not reveal a statistically significant difference between the growth rates of strain JG991 and JG1295 (Figure 4).

The Adaptation to Growth on DMSO Is Based on Three Point Mutations
We resequenced the genome of the strain to correlate the phenotype of the organism to specific changes in the genome sequence. The analysis revealed three variants with frequencies above 90% and a coverage of at least 20× (Table 4). Two mutations were found in genes encoding enzymes of the central metabolism. A stop codon was introduced in the gene for the pyruvate-formate-lyase (pfl) by replacing amino acid (AA) 431 (glutamate). A second point mutation in the gene of the dihydrolipoamide acetyltransferase subunit (aceF) of the pyruvate dehydrogenase replaces aspartate 430 by glycine. Moreover, the gene for the RNA polymerase rpoC was modified by replacing proline 359 by leucine (AA359). As the mutation in pflB leads to an interruption of the translation after approximately half the gene, we have to assume that pyruvate will not be converted to acetyl-CoA and formate via this enzyme anymore. Hence, it is interesting to observe that a second mutation occurred in an enzyme that can also catalyze the conversion of pyruvate to acetyl-CoA. The point mutation in aceF cannot lead to an inactive enzyme as this would cause no or at least very slow growth under oxic conditions, which could not be observed in this study (Figure 4) [28]. The aceF gene encodes subunit E2 of the pyruvate dehydrogenase to which E1 and E3 bind non-covalently. E2 consists of three lipoyl-domains at the N-terminus, a peripheral subunit-binding domain, and a large C-terminal core-forming acetyl-transferase domain to which the here observed mutation is localized [29,30]. Although the pyruvate-formate-lyase is the major pyruvate converting enzyme under anoxic conditions, the expression levels of the pyruvate dehydrogenase do not seem to be negatively influenced by the absence of oxygen [31]. Still, the activity of the enzyme is regulated by the NADH sensitivity of the dihydrolipoamide acetyltransferase subunit and the prevailing higher NADH/NAD ratios under anoxic conditions [22][23][24][25][26][27][28][29][30][31][32][33][34]. In fact, screening for mutants with higher ethanol productivity revealed clones with point mutations in the aceF gene, which rendered the enzyme insensitive to NADH-inhibition. Still, these mutations (H322Y or E354K) occurred at different positions [31]. Nevertheless, the situation is different in the here studied strain and a selection for an NADH-insensitive enzyme is unlikely. The genes for the majority of enzymes that would consume acetyl-CoA in the wild type under anoxic conditions were deleted. As the citric acid cycle is also downregulated, it would be detrimental for the strain to produce higher concentrations of acetyl-CoA, as accumulating CoA-esters can be toxic to the cell [35]. Moreover, we know that the carbon yield under acetoin-producing conditions is not lower (see below) compared to the strain before ALE which suggests that the available pyruvate pool is at least stable. Additionally, the mutation in the aceF-gene does not lead to a growth disadvantage under oxic conditions, which would indicate a generally reduced enzymatic activity. Hence, we hypothesize that the mutations render the enzyme even more stringent regarding reduced activity under the cellular conditions during anoxic growth. The NADH to NAD ratio as well as the cellular pyruvate concentrations could be used as triggers to reduce the enzyme's activity. The TCA cycle could still be delivered by necessary intermediates via the reduced acetyl-CoA pool and-probably more important-via the routes from pyruvate or phosphoenolpyruvate to oxaloacetate [36]. These three delivery routes will most likely be sufficient to support building block formation for biomass production under DMSO reducing conditions. Following this line of thought, the observed truncation and most probable deactivation of the pyruvate formate lyase would be a selective advantage as it simply stops the major acetyl-CoA producing reaction under anoxic conditions. The high pyruvate flux establishes in wild type E. coli an acetyl-CoA pool that is mainly deployed for energy production and redox balancing reactions [37,38]. As the mutant cannot catalyze these reactions anymore the high flux will not be relevant to sustain growth and-as described above-detrimental due to CoA-ester accumulation. While it is from a physiological viewpoint straight forward to hypothesize on the function of the pfl and aceF mutations it sems impossible to easily explain the potential impact of the rpoC mutation. RNA polymerase genes are often found to be targets for mutation in a multitude of screens and it is not always obvious how the particular mutation leads to the observed change in phenotype [39]. Nevertheless, we are not aware of a screen that would have revealed particularly this mutation before [40]. Overall, it is difficult to assess the influence of the individual point mutations as the two mutations in pyruvate converting enzymes will likely be interdependent regarding the growth stimulating effect. Moreover, it is not possible to easily assess the function of the rpoC mutation. The original strain grows poorly under anoxic respiratory conditions and partly forms flocks during growth. Hence, the mass limitations due to limited diffusion into the flocks will likely lead to an inhomogeneous transcriptome, which hampers comparison to the transcriptome of the strain after ALE.

The Selected Mutations Increase Productivity and Current Production in Bioelectrochemical Systems
Next, we sought to transfer the experiments from the cell suspension assays to bioelectrochemical systems. Strain JG1295 was introduced into the systems with an initial OD 600 = 6. The chronoamperometric measurement of the electro-fermentation at OD 600 = 6 revealed an initial current density maximum of 667.7 ± 83.6 mA/m 2 after 4 min followed by a slow decrease in current density ( Figure 5). Overall, the process and strain optimization conducted in this study led to a 3.6-fold increase in productivity compared to the previous report by Förster et al. [22]. The coulombic efficiency remained with 88.1 ± 16.3% in the range on the previously observed high value of 84.4 ± 4.1%.
Microorganisms 2020, 8, x FOR PEER REVIEW 10 of 16 above-detrimental due to CoA-ester accumulation. While it is from a physiological viewpoint straight forward to hypothesize on the function of the pfl and aceF mutations it sems impossible to easily explain the potential impact of the rpoC mutation. RNA polymerase genes are often found to be targets for mutation in a multitude of screens and it is not always obvious how the particular mutation leads to the observed change in phenotype [39]. Nevertheless, we are not aware of a screen that would have revealed particularly this mutation before [40]. Overall, it is difficult to assess the influence of the individual point mutations as the two mutations in pyruvate converting enzymes will likely be interdependent regarding the growth stimulating effect. Moreover, it is not possible to easily assess the function of the rpoC mutation. The original strain grows poorly under anoxic respiratory conditions and partly forms flocks during growth. Hence, the mass limitations due to limited diffusion into the flocks will likely lead to an inhomogeneous transcriptome, which hampers comparison to the transcriptome of the strain after ALE.

The Selected Mutations Increase Productivity and Current Production in Bioelectrochemical Systems
Next, we sought to transfer the experiments from the cell suspension assays to bioelectrochemical systems. Strain JG1295 was introduced into the systems with an initial OD600 = 6. The chronoamperometric measurement of the electro-fermentation at OD600 = 6 revealed an initial current density maximum of 667.7 ± 83.6 mA/m ² after 4 min followed by a slow decrease in current density ( Figure 5). Overall, the process and strain optimization conducted in this study led to a 3.6fold increase in productivity compared to the previous report by Förster et al. [22]. The coulombic efficiency remained with 88.1 ± 16.3% in the range on the previously observed high value of 84.4 ± 4.1%. Figure 5. Chronoamperometric detection. Electro-fermentation in MOPS buffered medium supplemented with 10 mmol/L glucose and 0.1% yeast extract. Displayed is the current density over time for the previous production strain (JG991, yellow) and the newly adapted strain (JG1295, teal). * data originally published in [22] and added here as a reference for the performance of the newly adapted strain.
The increase in current density agrees well with the observed acceleration in the growth rate on DMSO. Hence, it seems as if central metabolism and not the electron transfer kinetics limit the process since the conducted methylene blue reduction assays suggest that higher current densities could be Figure 5. Chronoamperometric detection. Electro-fermentation in MOPS buffered medium supplemented with 10 mmol/L glucose and 0.1% yeast extract. Displayed is the current density over time for the previous production strain (JG991, yellow) and the newly adapted strain (JG1295, teal). * data originally published in [22] and added here as a reference for the performance of the newly adapted strain.
The increase in current density agrees well with the observed acceleration in the growth rate on DMSO. Hence, it seems as if central metabolism and not the electron transfer kinetics limit the process since the conducted methylene blue reduction assays suggest that higher current densities could be achieved considering the maximum reduction rates obtained in cell suspension assays. To confirm that the limitation of the strain is most probably in central metabolism and not in electron transfer kinetics, we increased the expression rate of cymA and stc by introducing an additional plasmid-encoded copy of the two genes (strain JG1397). The plasmid also contained the alsSD genes for acetoin production. The introduction of the plasmid and expression of the genes increased the methylene blue reduction rate by 2.45-fold (Appendix A Figure A3). The experiments are consistent with previous results that reveal that cymA expression levels are constraining for extracellular electron transfer in E. coli and that the expression strength of genes encoding for cytochromes can have a crucial impact on electron transfer rates [41,42]. Still, the introduction of this strain to bioelectrochemical systems led to an insignificant increase in current density and acetoin yield compared to the progenitor strain JG1295 ( Figure 6). This result bolsters the hypothesis that current density is limited by central metabolism rather than electron transfer kinetics. achieved considering the maximum reduction rates obtained in cell suspension assays. To confirm that the limitation of the strain is most probably in central metabolism and not in electron transfer kinetics, we increased the expression rate of cymA and stc by introducing an additional plasmidencoded copy of the two genes (strain JG1397). The plasmid also contained the alsSD genes for acetoin production. The introduction of the plasmid and expression of the genes increased the methylene blue reduction rate by 2.45-fold (Appendix A Figure A3). The experiments are consistent with previous results that reveal that cymA expression levels are constraining for extracellular electron transfer in E. coli and that the expression strength of genes encoding for cytochromes can have a crucial impact on electron transfer rates [41,42]. Still, the introduction of this strain to bioelectrochemical systems led to an insignificant increase in current density and acetoin yield compared to the progenitor strain JG1295 ( Figure 6). This result bolsters the hypothesis that current density is limited by central metabolism rather than electron transfer kinetics.

Conclusions
Using a targeted and non-targeted evolution of a production strain for anode-assisted acetoin production from glucose, it was possible to raise the current density (3.6-fold) and consequently the productivity (3.6-fold) that can be achieved with this strain.
This study started with establishing the characteristics of methylene blue reduction kinetics and a KM value of 273 µmol/L was established for the whole-cell biocatalyst. Using this data and converting the measured reduction rates to a potentially achievable current density leads to a value of approximately 1.55 A/m 2 . While the initial current densities observed in the electro-fermentation experiment are only 2.3-fold lower as this value, the average current density throughout 8 h is 4.8fold lower. This has the following implications: First, the development of the strain was successful as it is easily imaginable that the initially discussed mass transfer limitations of the BES cause the 2.3fold difference in electron transfer rate. Second, this difference becomes more pronounced over longer periods is most probably due to the growth-inhibiting effect of methylene blue that is not interfering with the fast kinetics measured in cell suspension experiments. Hence, a further step towards a more stable current production would be to establish an E. coli strain that is more resilient towards higher methylene blue concentrations. Previous studies already revealed potential targets for a directed strain evolution approach. Apparently, the efflux pump AcrAB is necessary for E. coli

Conclusions
Using a targeted and non-targeted evolution of a production strain for anode-assisted acetoin production from glucose, it was possible to raise the current density (3.6-fold) and consequently the productivity (3.6-fold) that can be achieved with this strain.
This study started with establishing the characteristics of methylene blue reduction kinetics and a K M value of 273 µmol/L was established for the whole-cell biocatalyst. Using this data and converting the measured reduction rates to a potentially achievable current density leads to a value of approximately 1.55 A/m 2 . While the initial current densities observed in the electro-fermentation experiment are only 2.3-fold lower as this value, the average current density throughout 8 h is 4.8-fold lower. This has the following implications: First, the development of the strain was successful as it is easily imaginable that the initially discussed mass transfer limitations of the BES cause the 2.3-fold difference in electron transfer rate. Second, this difference becomes more pronounced over longer periods is most probably due to the growth-inhibiting effect of methylene blue that is not interfering with the fast kinetics measured in cell suspension experiments. Hence, a further step towards a more stable current production would be to establish an E. coli strain that is more resilient towards higher methylene blue concentrations. Previous studies already revealed potential targets for a directed strain evolution approach. Apparently, the efflux pump AcrAB is necessary for E. coli to thrive in medium with higher methylene blue concentrations [43,44]. A defect in this efflux pump led to lacking growth already with 30 µmol/L methylene blue. The gene cluster is negatively regulated by AcrR. Nevertheless, our initial attempts to delete this repressor were so far not successful possibly due to a toxic effect of this mutation.
Increasing space-time yields of electrode-assisted fermentations will be key for their future biotechnological implementation. The here conducted study provides a workflow for strain development and characterization of the capabilities of the applied BES. Further research will focus on accelerating glucose metabolism in the strain and aim at increasing the amount of active biocatalyst within the bioreactors by advancing reactor technology.