Escherichia coli was cultivated in modified M9^∗ minimal medium supplemented with 100 μg mL^-1 ampicillin (For medium composition see Supplementary Information). Precultures were inoculated from late exponential phase-harvested, permanent cultures (-80°C, 1:100), and were grown for 16 h in a volume of 30 mL in non-baffled 500 mL Erlenmeyer flasks at 37°C and 350 rpm in an orbital shaker (InforsHT Multitron 2 Standard). The fermentation was performed at 37°C in a 2 L RALF fermenter (Bioengineering, Inc.) with a filling volume of 1 L. The stirring speed of the impeller turbines was set to 1000 rpm and the aeration rate kept constant at 1.5 vvm with air control by MX4/4 (DASGIP). The pH value was adjusted to 6.5 with 3 M NaOH and 3 M H₃PO₄, and inoculation was done at a ratio of 1:50 with the preculture. The off-gas analysis was performed with a GA4 off-gas analyzer (DASGIP). When foaming occurred antifoaming agent Antifoam O-30 (100 μL, Sigma–Aldrich) was added. At OD = 1, the bioreactor was cooled down to 20°C and the CHMO expression was induced with IPTG (100 μM, final concentration).

The cell density of cultures was measured via absorbance at 590 nm in a spectrophotometric cell before the culture liquid was used for subsequent assays. The first sample was taken directly after inoculation. Therefore, 1 mL of the reaction culture was sampled every hour, and if necessary, diluted in a ratio of 1:10 with the medium to stay within the linear range (up to 0.7) of the apparatus (WPA colorwave CO7500 Colorimeter). We used the following relation: 1 OD = 0.43 g DCW^-1 to determine the dry cell weight (Walton and Stewart, 2004).

Samples for the analysis of the cell physiology were taken every hour. One mL of culture was transferred to a 1.5 mL Eppendorf reaction tube on ice and centrifuged (5000 × g, 10 min). The supernatant was filtered with a 0.2 μm syringe filter and stored at -20°C until HPLC analysis (Nexera Shimadzu). The supernatant was analyzed with a refractive index (RI) detector and photodiode array (PDA) detector for quantification of the analytes and an electrospray ionization (ESI) ion source with a quadrupole mass analyzer for additional confirmation of the substances (LC–MS 2020 Shimadzu). Separation was performed with an ROA-Organic Acid H+ (8%) column (300 mm × 7.8 mm, Phenomenex) with an isocratic flow of 0.4 mL min^-1 5 mM formic acid in water.

The quantification of soluble CHMO was performed as single measurements per time point. The volume of the sample was adjusted, so that after centrifugation and the washing steps, the re-suspended cells (0.5 mL) would have an OD at 590 nm of 7.0 to later on have a quantifiable amount of CHMO. The sample was at first centrifuged (5000 × g, 10 min) and the resulting pellet was washed with 1 mL 50 mM Tris-HCl buffer, pH 7.5, then centrifuged again (5000 × g, 10 min), and stored at -20°C until further workup.

For cell lysis, the pellets were re-suspended in 0.5 mL 50 mM Tris-HCl, pH 7.5, containing 0.1 mM phenylmethylsulfonyl fluoride (PMSF). Cells were placed on ice and sonicated in nine cycles (5 s pulse, 55 s break, amplitude 50%, Bandelin KE76 sonotrode connected to a Bandelin Sonopuls HD 3200 wave generator). Precipitates were removed by centrifugation (15000 × g, 15 min) and the clear supernatants containing the soluble proteome, including CHMO, were analyzed by SDS-PAGE (Laemmli, 1970). Every SDS-PAGE gel was additionally loaded with at least three, differently concentrated samples of pure CHMO (range: 0.4–2.5 μg) in order to quantify the soluble CHMO concentration in every sample (for preparation of purified enzyme: see Supplementary Information). The gels were stained with SimplyBlue SafeStain (Thermo Fisher) using the microwave method according to the manufacturer’s protocol, scanned with the Molecular Imager Gel Doc XR System (BioRad), and quantified with the Image Lab Software (BioRad).

The sampling procedure was repeated every hour as a single point, and every 2 h in triplicates unless stated otherwise. Metabolomics samples were taken using a fast filtration sampling procedure, as described elsewhere (Link et al., 2012) (for details see Supplementary Information).

Dry extracts were re-suspended in water (100 μL) and centrifuged (12000 × g, 3 min). The supernatants were used for subsequent analyses. Separation was achieved with a Luna-NH₂ HPLC column (150 mm × 2 mm, 3 μm particle size, 100 Å pore size, Phenomenex) using a binary gradient method (Solvent A: acetonitrile/Solvent B: 10 mM ammonium acetate, pH 9.9). Gradient parameters were as follows: 0–24 min: 20–100% B; 25–34 min: 100% B; 35 min: 20% B. Detection was performed with a tandem mass spectrometry detector with an ESI ion source (Shimadzu LCMS-8040) in multiple reaction monitoring (MRM) mode (Lorenz et al., 2011). Peak areas were normalized to fully ¹³C-labeled internal standards and absolute quantification of metabolites was achieved with linear calibration curves of the standards. Finally, concentrations were normalized to the amount of biomass. For MRM fragments see Supplementary Table S2.

Bicyclo[3.2.0]hept-2-en-6-one was purchased from Sigma–Aldrich and used as obtained. The activity measurements of CHMO in the batch culture were performed every 3 h. The first measurement was done 3 h post-induction with IPTG. For the activity measurement 5 mL of fermentation broth were transferred to a 250 mL non-baffled Erlenmeyer flask, and bicyclo[3.2.0]hept-2-en-6-one (8 mM, 1 M solution in dioxane) was added as substrate. The culture was placed on an orbital shaker at 20°C and 350 rpm to mimic the fermentation conditions in terms of aeration, mixing, and temperature. The first sample was taken immediately after substrate addition. Therefore, 100 μL of the culture broth were extracted with 500 μL ethyl acetate containing methyl benzoate (1 mM) as internal standard. The activity test was performed for 1 h; samples were taken every 15 min. The consumption of substrate as well as formation of the corresponding lactone was monitored by calibrated GC. The activity was normalized to the concentration of biomass, and to the soluble CHMO content to obtain the specific activity.

We investigated the levels of and changes in concentrations of the following metabolites in a whole-cell overexpression system, which had been identified as crucial for CHMO’s activity and stability in earlier studies in vitro and in silico: CHMO itself, NADPH, FAD (Goncalves et al., 2017; Milker et al., 2017). We based our study on the following hypotheses:

To test these hypotheses, we designed and conducted an experiment for the production of CHMO under controlled conditions in a bioreactor (pH, temperature, oxygen saturation) and performed the following analyses: we measured the intracellular concentration of CHMO (via SDS-PAGE and densitometry) and determined the activity of CHMO in a Baeyer–Villiger oxidation in a satellite culture to characterize the performance of the whole-cell biocatalyst. We determined the concentrations of the cofactors FAD and NADPH and their respective precursors (GXP, ATP), plus several indicative metabolites: amino acids (building blocks for the synthesis of CHMO) and compounds of the central carbon metabolism and tricarboxylic acid cycle (TCA) to monitor the physiology of E. coli. The sampling schedule is described in Supplementary Table S1.

We induced the production of CHMO with IPTG in the exponential phase, and held the culture stationary with >60 mM glucose for 15 h after reaching a limitation by the nitrogen (Supplementary Figure S1). The maximum concentration of soluble CHMO was measured after 19 h (20.9 μg mg DCW^-1, 11 h after induction). This value decreased toward the end of the experiment to 15.2 μg mg DCW^-1 (Figure 1) with a small exponential decay constant of approximately 0.04 (assuming first-order deactivation of the enzyme in the stationary phase). We thus observed a loss of 27% of the soluble CHMO. This observation led us to reject hypothesis A: CHMO was not stable over a time span common for a preparative experiment.

We also observed a fast decrease in the specific activity with rising concentration of soluble CHMO: the highest activity was observed at the first measurement (3 h after induction). The specific activity dropped by approximately 90% within the first 5 h, and then remained constant for the next 21 h (Figure 1). Despite the inverse relationship of the concentration of CHMO and its specific activity, E. coli produced enough CHMO to theoretically reach high specific activity, thus supporting hypothesis B.

Based on our analytical results from metabolomics (quantification by ¹³C-standards in MS analysis), we estimated the intracellular concentrations of NADPH and FAD, using a published value for the volume of 3.2 ± 1.2 × 10^-15 L per cell (Volkmer and Heinemann, 2011).

We estimated the cytosolic concentration of CHMO, but the propagated error from the estimated volume did not allow the identification of a significant trend (see Figure 2C); the mean value was still used for calculations. We observed stable intracellular concentrations of FAD (40 ± 16 μM, Figure 2A) and NADPH (63 ± 29 μM, Figure 2B) after adaption to 20°C (increase in the uptake of oxygen, Supplementary Figure S2). Since the cells were extracted vigorously in sampling procedure, the measurement could not differentiate between the free and bound form of FAD (or of any other metabolite that would form non-covalent complexes), preventing the exact calculation of equilibrium concentrations. That notwithstanding, and given the uncertainty in the estimated values, we made the following, drastically simplifying assumptions to obtain an extreme boundary condition: (i) the total concentration of FAD in the cell is equal to the concentration of unbound FAD, and (ii) CHMO is the only FAD-dependent enzyme in the cell. This way, a hypothetical, but optimum scenario was created for the population of the CHMO⋅FAD complex (i.e., loading 𝜃), calculated using a plain logistic function:

with [CHMO⋅FAD] as the concentration of the CHMO⋅FAD complex, [CHMO_apo] as the concentration of the CHMO apoenzyme, K_d as the dissociation constant of FAD (3.5 ± 0.6 μM; Goncalves et al., 2017), and [FAD] as the total intracellular concentration of FAD. Even under these theoretical conditions, the intracellular concentration of FAD was only sufficient to saturate on average 90–95% of CHMO with the coenzyme, far from thermodynamic saturation (>>99% at [FAD] > 100 × K_d; Figure 2D). Given the large compounded uncertainty, and our extreme assumptions, we speculated that most of CHMO was actually present in its apo form. The decay of specific activity to a residual non-zero value corroborated the insufficient loading, indicating a rising lack of FAD with rising CHMO concentrations in the cell.

The estimated intracellular concentration of NADPH was sufficiently high to largely saturate the enzyme (0.73 × v_max, K_M,NADPH = 18 μM; Figure 2B). Nevertheless, the concentration of NADPH was much lower than the level required for a strong stabilization, as earlier determined in vitro. There, a titer of >1 mM was found to have a large, beneficial effect on the thermodynamic and kinetic stability of CHMO (Goncalves et al., 2017). Together, these results could not clearly disprove hypothesis C on the basis of FAD; still, our data clearly showed that the cells could not provide enough of the coenzyme to saturate most CHMO molecules.

We then analyzed the metabolic network (Figure 3) to identify potential bottlenecks in the metabolic upkeep of active CHMO [specifically FAD, NADPH, and key metabolites that are required for their synthesis: guanosine and adenosine mono-, di-, and triphosphates (GMP, GDP, GTP, AMP, ADP, and ATP)]. We measured the concentrations of these biosynthetic precursors (Figure 4) to determine if a shortage thereof was limiting the availability of the cofactor to CHMO. GTP decreased already during the exponential growth phase of the cells (Figure 4A). After the culture had reached limitation by nitrogen, the concentration of the GXP pool was stable at approximately 0.300 ± 0.01 μmol gCDW^-1; the three compounds of this pool can be interconverted by the host metabolism. The concentration of ATP was stable over the entire observed period (approximately 2.0 ± 0.2 μmol gCDW^-1; Figure 4B); the concentrations of AMP and ADP decreased until the onset of the nitrogen limitation, and were then stable. We could not discern any apparent, meaningful correlation between the concentrations of ATP or GTP and FAD, or a bottleneck in their supply. The lack in FAD was thus not directly, or not obviously, caused by a poor supply of its precursors.

To assess the proper function of the host organism, we investigated indicative parameters for growth, physiology, and biocatalytic performance of E. coli. We found an overall decrease of the glycolytic intermediates fructose-1,6-bisphosphate and 3-phosphoglycerate (Figure 5A), and the TCA components fumarate and malate (Figure 5B) after the onset of the nitrogen limitation. These metabolites remained constant within experimental error after physiological adaption of the culture to 20°C. The heavy metabolic burden of heterologous overexpression became apparent in an unfavorable value for the adenylate energy charge (Figure 5C) of 0.65 at the beginning of protein overproduction. It later stabilized to 0.83 under nitrogen limitation. This value reflected a balanced pool of adenosine phosphates (Chapman et al., 1971) and cell viability until the end of the cultivation. The protein production stopped when the culture reached the nitrogen limitation, which was reflected in the concentration curves of most amino acids. Specifically, the low concentrations of glutamine, glutamate, and aspartate after the nitrogen limitation prevented the organism from further synthesizing CHMO (or other proteins, Supplementary Figures S4D–F).

In summary, the host organism E. coli was found capable of maintaining the NADPH requirements of CHMO, providing a stable environment with sufficiently high levels of precursors for the protein itself, as well as for essential cofactors. However, due to the low concentration of FAD, it failed to fully saturate CHMO in overexpression. Although the experimental evidence cannot lead us to directly reject hypothesis D — all necessary compounds were invariably present to some extent—, it does not support the assumption that E. coli is a suitable host for the efficient maintenance of CHMO.