DNA oligonucleotides, plasmids, and bacterial strains used for cloning procedures are given in the Supporting Information (Supplementary Table S1). Synechocystis sp. PCC6803 (Stanier et al., 1971) was cultivated in 20 or 50 ml YBG11 medium (Shcolnick et al., 2007) with 50 mM 4-(2-hydroxyethyl)-1-piperazinethanesulfonic acid, pH 7.2, using 100 or 250 ml baffled shaking flasks with cotton stoppers, respectively. Cultivation conditions were 30°C, 150 rpm [2.5 cm amplitude], 50 µmol_photons m⁻² s⁻¹, ambient CO₂ [0.04%], 75% humidity, and start OD₇₅₀ = 0.06–0.08. Plate cultivation was conducted on BG11 agar plates with 1.5–2.0% agar at 30°C and 25 µmol_photons m⁻² s⁻¹. If required, kanamycin (Km) was used at a final concentration of 50 μg ml⁻¹. Cyclohexanone, ≥ 99.5% purity, was purchased from Sigma-Aldrich (Steinheim, Germany). Cyclohexanol, ≥ 99% purity, was purchased from Merck (Darmstadt, Germany). All other chemicals were purchased from Carl Roth GmbH (Karlsruhe, Germany), Merck, or Sigma-Aldrich at the highest purity available.

Standard molecular biology procedures were applied as described by Sambrook and Russell (2001). For construction of pAH059, a fragment coding for the C-terminal Strep-tagged bvmo gene (Schäfer et al., 2020) was amplified from pCom10_Capro (Karande et al., 2018) via Phusion PCR using the primer pair PAH096/097 and cloned via Gibson assembly (Gibson et al., 2009) into pEERM3_Km, linearized by XbaI/PstI double digestion. For construction of pAH063, the expression cassette of PrnsB_bvmo_Strep-tag_terminator was amplified via Phusion PCR using the primer pair BVMO1/2 and cloned via Gibson assembly into pPMQAK1, linearized with EcoRI/PstI double digestion. Escherichia coli DH5α (Hanahan, 1983) was transformed with Gibson products via electroporation and correct assembly of plasmids was confirmed via colony PCR of transformants and subsequent sequencing. Commercial kits from Macherey-Nagel (Düren, Germany) were used for plasmid DNA extraction and DNA purification.

For electroporation of Syn6803, cells were grown under standard conditions to an OD₇₅₀ = 0.5–0.7 determined with a LibraS11 spectrophotometer (Biochrom, Cambridge, United Kingdom), harvested by centrifugation at 5,000 g and 4 °C, washed and 50 x concentrated in 1 mM 4-(2-hydroxyethyl)-1-piperazinethanesulfonic acid, pH 7.5 [according to (Ludwig et al., 2008)]. In total, 60 µl of cell suspension was supplied with 0.5–1 µg of plasmid DNA and electroporated (2.5 kV; 5 ms). After 24 h recovery in YBG11 under standard conditions, selection was performed at 25 µmol_photons m⁻² s⁻¹ on BG11 agar plates containing Km. Transformants grown within 7–10 days were confirmed via colony PCR and sequencing. For chromosomal integration, complete segregation was achieved by iterative plating on selective agar plates with increasing Km concentration.

Exponentially growing Syn6803_Ni_cBVMO, Syn6803_Ni_pBVMO, and Syn6803_Cu_pBVMO cultures (see Supplementary Table S1 for strain designations) were harvested by centrifugation at 5,000 g and 4°C 24 h after induction with defined concentrations of NiSO₄ and concentrated to an OD₇₅₀ = 20 in 1 x TBS (1 mM phenyl-methyl-sulfonyl-fluoride, pH 7.5). Cells were disrupted in a Precellys^® homogenizer for 3 cycles á 3 × 30″ with 30″ pause. After centrifugation at 13.300 g and 4°C for 15 min, the supernatant was used for protein concentration determination via the Bradford analysis (Bradford, 1976). Protein abundance was visualized using SDS-PAGE according to Laemmli (1970), with 3.6 and 12% acrylamide in the stacking and in the separation gels, respectively.

Syn6803_Cu_pBVMO, Syn6803_Ni_cBVMO, and 6803_Ni_pBVMO cultures (Supplementary Table S1) were harvested by centrifugation at 5,000 g and 4°C 24 h after induction with defined concentrations of NiSO₄ and adjusted to a defined cell concentration by re-suspending in YBG 11 containing 50 mM 4-(2-hydroxyethyl)-1-piperazinethanesulfonic acid and 10 µM NiSO₄, pH 7.2, if not stated otherwise. Cells were equilibrated for 10 min to assay conditions. Standard assay conditions were 30°C, 200 rpm, 150 µmol_photons m⁻² s⁻¹, ambient CO₂ (0.04%). The conversion of cyclohexanone (C-one) to ε-caprolactone (ε-Cl) was started by adding C-one to a final concentration of 5 mM and was stopped after 30 min by either quenching with 1 vol diethylether containing 0.2 mM n-decane as an internal standard (for GC analysis) or by adding 0.1 vol of acetonitrile (for HPLC analysis).

For short-term assays (30 min), a cell concentration of 0.7–1 g_CDW L⁻¹ was used in a total volume of 1 ml in 10 ml Pyrex tubes. For the investigation of whole-cell kinetics, the cell concentration was reduced to 0.25–0.5 g_CDW L⁻¹ and the reaction time to 5 min to record initial activities.

Kinetic parameters (V_max, K_S  and K_i) were obtained by fitting according to Michaelis–Menten kinetics with substrate inhibition using OriginPro2019 following Eq. 1, where V is the specific activity in U g_CDW ⁻¹ and [S] is the substrate concentration.   V = V m a x ∗ [ S ] K S + [ S ]   ( 1 +   [ S ] K i )   [ U g CDW − 1 ] (1)

For competitive and non-competitive inhibitors, K_s and V_max in Eq. 1 were replaced by K_s’ according to Eq. 2 or V_max’ according to Eq. 3, respectively, where [I] and K_i,I are the concentration and the inhibition constant of the respective inhibitor. K S ′ = K S ∗ ( 1 + [ I ] K i , I )   [ µM ] , (2) V m a x ′ = V m a x ( 1 + [ I ] K i , I )   [ U g CDW − 1 ] (3)

Whole-cell biotransformations were conducted in a stirred tank photo-bioreactor Labfors 5 Lux (Infors AG, Bottmingen, Switzerland) with 2 L culture volume. Cells were grown for 3–4 days at 30°C, 2 L min⁻¹ aeration with (high carbon, HC, additional 20 ml min⁻¹ CO₂) or without (low carbon, LC, no additional CO₂), 300 rpm stirrer speed, and specific light intensities (for details, see Table 1). Induction with 10 µM NiSO₄ was performed 24 h prior to biotransformation start. During biotransformation, aeration was reduced to 0.2 L min⁻¹ and light intensity was set to low light (LL, 150 µmol_photons m⁻² s⁻¹) or high light (HL, 700 µmol_photons m⁻² s⁻¹). After pre-conditioning for 30 min, biotransformation was initiated by starting a continuous feed of 0.25 M C-one in the YBG11 medium. Samples were taken at regular time intervals and quenched immediately by adding either 1 vol diethylether or 0.1 vol acetonitrile.

The cell concentration was assessed by determining OD₇₅₀ in a LibraS11 spectrophotometer (Biochrom, Cambridge, United Kingdom). The previously determined correlation factor of 0.225 g_CDW L⁻¹ per OD₇₅₀ unit was used to calculate cell dry weight (CDW) concentrations (Hoschek et al., 2017). Chl_a concentration was determined as described by Grund et al. (2019). Concentrations of cyclohexanol (C-ol), C-one, and ε-Cl were determined by gas chromatography (GC) as described before (Schäfer et al., 2020). See Supplementary Figure S1 for exemplary calibration curves.

Concentrations of 6-hydroxyhexanoic acid (6-HA) were determined by HPLC on a Dionex Ultimate 3,000 system equipped with an Acclaim^® OA column (both Thermo Fisher Scientific, Waltham, MA). The sample was acidified with 1 M HCl to pH 3.0. The mobile phase A consisted of 100 mM sodium sulfate, adjusted with methanesulfonic acid to pH 3. Acetonitrile (>99.95% purity, ProtoChem) was used as mobile phase B. Sample volumes of 10–20 μL were injected, while the flow and the column temperature were kept constant at 0.4 ml min⁻¹ and 60°C, respectively. The flow profile was as follows: 5% B for 2 min, 5–30% B in 6 min, 30–80% B in 1 min, 80% B for 1 min, 80–5% B in 2 min, and 5% B for 5.5 min. Detection was accomplished via a UV detector at 210 nm.

For kinetic bioprocess modeling, the Berkeley Madonna software (version 8.3.18) was used. The biotransformation converting cyclohexanone (C-one) to ε-caprolactone (ε-Cl) as a product and cylohexanol (C-ol) as a by-product was described in a kinetic model. The concentration courses for substrate (S′, C-one), product (p′, ε-Cl), and by-product (B′, C-ol) were calculated using Eqs 4–6: S ′ =   F − k B − k P   [ mM min - 1 ] , (4) B ′ = k B   [ mM min - 1 ] , (5) and P ′ = k P   [ mM min - 1 ] , (6) where the C-one feed rate F in mM min⁻¹, the reaction rate of the keto reduction kB (C-ol formation), and of ε-Cl formation kP, described in Eqs 7, 8. The reaction rates kB and kP were calculated based on determined kinetic parameters (K_s and V_max for keto reduction and K_s, V_max, and K_i,C-one for ε-Cl formation); kP was extended by the competitive inhibition by C-ol and the non-competitive inhibition by ε-Cl as follows: k P = V m a x ∗ [ S ] K S ( 1 + [ B ] K i , C − o l ) + [ S ] ( 1 + [ S ] K i , C − o n e ) ∗ 1 ( 1 + [ P ] K i , ε − C l ) ∗ X     [ mM min - 1 ] (7) and k B = V m a x ∗ [ S ] K i + [ S ] ∗ X     [ mM min - 1 ] , (8) where the biomass concentration X is in g_CDW L⁻¹ and the inhibition constants K_i,C-ol, K_i,C-one, and K_i,ɛ-Cl for C-ol (B), C-one (S), and ɛ-Cl (p), respectively. See section 4.4 for the exact values chosen for the kinetic parameters.

S, P, and B were initially set to 0, and the parameters X (biomass concentration in g_CDW L⁻¹) and F (C-one feed rate in mM min⁻¹) were fed into the model according to the respective experiment. The kinetic process model can be found in the supplemental material.

In order to systematically analyze different expression strategies in Syn6803, three different strains were constructed: Syn6803_Cu_pBVMO for plasmid-based BVMO gene expression under control of the Cu²⁺-inducible PpetE promoter and Syn6803_Ni_pBVMO and Syn6803_Ni_cBVMO for plasmid- and genome-based BVMO gene expression, respectively, under control of the Ni²⁺-inducible PnrsB promoter. All designed strains were compared in the detail regarding protein level and specific C-one to ε-CL conversion activity in short-term assays. Activities of the Ni²⁺-inducible strains Syn6803_Ni_pBVMO and Syn6803_Ni_cBVMO were found to increase with increasing Ni²⁺ concentration yielding maximum specific activities of 52.2 ± 0.3 (10 µM Ni²⁺) and 23.8 ± 4.0 U g_CDW ⁻¹ (15 µM Ni²⁺), respectively (Figure 1A). The keto reduction of C-one to cyclohexanol (C-ol) was found to occur at a low rate of 0.5 ± 0.2 U g_CDW ⁻¹ (for details, see Inhibitory Effects of Substrate and By-Product Necessitate Precise Reaction Control section). It is important to note that expression induced by 15 µM Ni²⁺ did not significantly affect growth of Syn6803_Ni_pBVMO, whereas this Ni²⁺ concentration slightly affected growth independently of BVMO gene expression (Supplementary Figure S2). Cu²⁺-induced Syn6803_Cu_pBVMO yielded a maximal specific activity of 29.7 ± 0.7 U g_CDW ⁻¹ (with 0.5 µM Cu²⁺, Figure 1B). Ni²⁺-inducible systems resulted in tightly regulated expression, whereas the Cu²⁺-inducible system showed slight leakiness and poor titratability by the inducer concentration. Due to these reasons and the lower activity obtained upon plasmid-based expression, we focused on the Ni²⁺-inducible system and investigated plasmid- and genome-based expression in more detail. SDS-PAGE analyses of protein extracts of Syn6803_Ni_pBVMO and Syn6803_Ni_cBVMO showed a distinct band at ∼60 kDa, coinciding with the BVMO size (57.8 kDa) (Figures 1C,D). This band intensified with increasing Ni²⁺-concentration and was absent in un-induced cells and cells carrying an empty vector. Densitometric analysis indicated that BVMO accounted for 1.1 ± 0.2% of total soluble protein (TSP) in Syn6803_Ni_pBVMO induced with 10 µM Ni²⁺, a remarkably high value for heterologous expression in Syn6803. Assuming a total protein content of 0.32–0.57 g g_CDW ⁻¹ (Zavrel et al., 2017), the measured specific activity of 52.2 ± 0.3 U g_CDW ⁻¹ translates into an in vivo activity of 7–18 U mg_BVMO ⁻¹ or a k_cat of 6.9–17.6 s⁻¹ as a rough estimation.

In summary, we were able to construct a highly active BVMO-containing Syn6803 strain. Our data show that the PnrsB-based expression system proved to be a useful tool allowing tight, tunable, and high-level expression. Syn6803_Ni_pBVMO was thus chosen for further characterization of photosynthesis-driven BVMO catalysis.

The BVMO reaction depends on NADPH as electron donor and thus is expected to constitute an artificial electron sink and thus to depend on the major electron source in phototrophs—the photosynthetic water oxidation. To test if and to what extent the BVMO reaction in Syn6803 depends on incident light, short-term activity assays at light intensities ranging from 0 to 250 µmol_photons m⁻² s⁻¹ were performed in medium deficient in organic compounds as potential electron donors. Clearly, specific activities increased with increasing light intensity with the maximal activity of 60.9 ± 1.0 U g_CDW ⁻¹ measured at 150 µmol_photons m⁻² s⁻¹ (Figure 2A). With chlorophyll a content of 11.8 ± 0.8 mg_Chla g_CDW ⁻¹, this activity translates into 5.2 ± 0.8 U mg_Chla ⁻¹. Higher light intensities (such as 250 µmol_photons m⁻² s⁻¹) led to a slight activity decrease, most likely due to photoinhibitory effects. When blocking electron transfer from PS II to the PQ pool by means of the respective inhibitor 3-(3,4-dichlorphenyl)-1,1-dimethylurea (DCMU) or upon incubation in the dark, residual activities of 13.6 ± 0.3 or 19.0 ± 0.8 U g_CDW ⁻¹ were observed, respectively. These typical background/dark activities are most likely fueled by reducing equivalents derived from storage compounds. We could show that extended periods (up to 16 h) of incubation in the dark or with DCMU, which involves the consumption of storage compounds and thus a reduction of their availability, decreases the activities to 4.7 ± 0.1 and 1.3 ± 0.2 U g_CDW ⁻¹, respectively (Figure 2B). These results clearly indicate a close coupling of the BVMO reaction to the photosynthetic electron transport chain.

To specify the optimal substrate concentration, the reaction kinetics were analyzed for Syn6803_Ni_pBVMO by varying the C-one concentration between 50 μM and 3 mM in 5 min assays. The cells showed Michaelis–Menten-type kinetics with slight substrate inhibition (Figure 3A). Fitting according to Michaelis–Menten kinetics with substrate inhibition gave a reasonable fit and the following kinetic parameters: K _S = 79 ± 23 μM, V _max = 83.7 ± 8.0 U g_CDW ⁻¹, and K _i = 4.4 ± 1.5 mM. This indicates that the C-one concentration is a critical reaction parameter requiring careful control.

C-one oxidation applying whole cells of both heterotrophic (Schäfer et al., 2020) and phototrophic hosts (Böhmer et al., 2017) has been reported to compete with keto reduction catalyzed by host-intrinsic enzymes, giving rise to C-ol (Figure 3C). This also was true for Syn6803_Ni_pBVMO, which was used for kinetic experiments investigating C-ol formation. Fitting according to Michaelis–Menten kinetics yielded a high K_S of 7.4 ± 0.6 mM and a comparably low V_max of 2.75 ± 0.15 U g_CDW ⁻¹ (Supplementary Figure S3). Additionally, 6-hydroxyhexanoic acid (6-HA) was detected as by-product, which can be ascribed to abiotic hydrolysis of ε-CL (Supplementary Figure S4). Potential inhibition by these by-products also was tested in activity assays. Whereas 6-HA did not significantly affect C-one oxidation (Supplementary Figure S6), C-ol was found to be a strong inhibitor with a K_i of 0.03 mM, determined via kinetic fittings as described in Whole-cell activity assays and determination of reaction kinetics section (Figure 3B), corresponding to a K_i,C-ol/K_s,C-one ratio of 0.38. Figure 3C depicts a reaction scheme based on the findings described in this paragraph. Since intrinsic alcohol dehydrogenase activity can barely be avoided without drastically interfering with host metabolism, high BVMO levels can be considered pivotal to promote ε-CL formation. In this way, low yields due to by-product formation and—even more relevant—inhibition by C-ol can largely be avoided.

To test the long-term stability of C-one oxidation by Syn6803_Ni_pBVMO, we recorded the biotransformation time course in shaking flasks initially containing 5 mM C-one and fed by pulsing the same amount after 1 and 3 h. The 15 mM C-one supplied were completely converted to 14.2 ± 1.5 mM products (ε-Cl and 6-HA, Supplementary Figure S5). Thereby, the initial specific activity of 33 U g_CDW ⁻¹ determined for the first hour was found to decrease continuously over time. This decrease can be attributed to C-ol accumulation, reaching a final titer of 0.51 ± 0.07 mM, and intermittently high C-one concentrations leading to enforced substrate inhibition.

In order to test the feasibility of Syn6803_Ni_pBVMO to produce ε-Cl in the gram scale, this strain was applied in a 2-L stirred-tank photo-bioreactor (photo-STR) providing an efficient O₂ and CO₂ mass transfer and a controlled environment. The latter included the application of a limiting continuous substrate feed to minimize both substrate and by-product inhibition. To establish different source/sink conditions, light intensity (low light, LL: 150 µmol_photons m⁻² s⁻¹ or high light, HL: 700 µmol_photons m⁻² s⁻¹) and CO₂ supply via aeration (low CO₂, LC: ambient CO₂ or high CO₂, HC, 1 and 10% CO₂ during growth and biotransformation, respectively) were varied to establish the three source/sink regimes LLLC, HLLC, and HLHC (for details, see Table 1).

After a growth phase of 3–4 days, including induction by adding 10 µM NiSO₄ 24 h before biotransformation start, C-one feeding was initiated at a fixed rate matching a specifically chosen activity. The first reactor was run with 0.55 ± 0.09 g_CDW L⁻¹ under LLLC and a feed rate set to 25 U g_CDW ⁻¹. A final product titer (sum of ε-Cl and 6-HA) of 11.6 ± 1.0 mM was reached after 28 h (see Figure 4A). However, 5.4 ± 0.1 mM C-one remained unconverted, as the specific activity strongly decreased within the first hours. A second reactor was set up for HLLC conditions to rule out light limitation as a reason for the decreasing activity. To avoid substrate accumulation, the initial substrate feed rate was adjusted to match an activity of 15 U g_CDW ⁻¹ using 0.61 ± 0.04 g_CDW L⁻¹. This lead to a prolonged product formation phase, yielding however a slightly lower final product titer of 9.8 ± 1.6 mM after 25 h (Figure 4B). To test, if product formation benefits from higher biomass concentrations, the third photo-STR was conducted using HLHC conditions. These conditions allowed for a 70% higher initial biomass concentration (1.02 ± 0.01 g_CDW L⁻¹, see Table 1). To increase the C-one conversion rate, substrate supply was set to match 30 U g_CDW ⁻¹. Comparable to the previous reactor runs, this preset activity of 30 U g_CDW ⁻¹ only was met initially and was decreasing in time, resulting in C-one accumulation, requiring adjustment of the C-one feed rate (decrease to 70% after 4 h and stop of C-one feed after 8 h, Figure 4C). The final product titer was 11.5 ± 0.2 mM after 27 h, corresponding to a volumetric productivity of 48.6 ± 0.8 mg L⁻¹ h⁻¹ and a product yield of 0.88 (mol_ε-Cl+6-HA mol_C-one ⁻¹). Thus, all experiments led to similar product titers (9.8–11.6 mM corresponding to 1.12–1.32 g L⁻¹).

Irrespective of the applied source/sink conditions, the specific activity decreased in all setups at substrate (C-one) and by-product (C-ol) levels not expected to lead to significant inhibition (Figures 4A–C). The independency of initial specific activities on the applied biomass concentration (varied between 0.55 and 1.02 g_CDW L⁻¹) indicates that light availability did not limit the specific activity (Table 1 and Figures 4A–C). This was further confirmed by light intensity variation (250 µmol_photons m⁻² s⁻¹ vs. 700 µ µmol_photons m⁻² s⁻¹) under otherwise identical conditions as in the HLHC experiment (Figure 4C). This low-light high-carbon (LLHC) experiment resulted in a similar initial volumetric productivity as in the HLHC experiment corroborating that light did not limit the biocatalytic activity (Supplementary Table S2). These results indicate that neither inhibition by C-one or C-ol nor photosynthesis limited the biocatalytic activity. Instead, product inhibition may have prevented a more effective and stable conversion. Indeed, ε-Cl was found to compromise BVMO activity in short term assays (Supplementary Figure S6). As no product inhibition has been observed on the enzyme level (Schäfer et al., 2020), ε-Cl rather inhibits on a physiological level, which can best be described as a non-competitive type of inhibition. With the respective fit (Eqs 1, 3), the inhibitory studies yielded a K_i,ε-Cl of 0.83 mM.

These findings prompted us to describe the biotransformations by means of a kinetic bioprocess model using Berkeley Madonna software. The model was parametrized using the kinetic data reported in the Potentials and Limitations of Photosynthesis-Driven Redox Biocatalysis section, which were derived from fitting according to Michaelis–Menten kinetics with substrate inhibition (Eq. 1). Taking into account that this approach neglects by-product and product inhibition, which occur, but were not considered for the determination of V_max and K_i,C-one, these parameters were set to the upper boundaries of the fittings, namely, 91 U g_CDW ⁻¹ and 5.9 mM. The inhibition by ε-Cl observed under assay conditions (Supplementary Figure S6, K_i,ε-Cl = 0.83 mM) appeared to be less pronounced in bioreactor experiments. Thus, K_i,ε-Cl was fitted to bioreactor data resulting in a value of 2.0 mM. This allowed to simulate concentration courses for substrate (C-one), product (ε-Cl + 6-HA), and by-product (C-ol) (depicted as lines in Figures 4A–C) with striking accuracy. This high simulation accuracy obtained with a mere kinetic model, not considering possible influences of light intensity and CO₂ supply variation in the experiments, indicates that C-one conversion was kinetically controlled under the conditions applied, including a physiological effect of ε-Cl, and was not influenced by the tested variation in source/sink regime.

Furthermore, the process model was used to identify optimal process parameters to handle the complex reaction kinetics, namely, the feed regime and the biocatalyst concentration. For this purpose, only reaction kinetics were considered, and avoidance of a possible light limitation was assumed. To explore the sensitivity regarding the biocatalyst concentration, values of 0.5, 1, 2, and 5 g_CDW L⁻¹ were chosen, being aware that high cell concentrations such as 5 g_CDW L⁻¹ would involve light limitation issues in the experimental setup (as further discussed in the Potentials and Limitations of Photosynthesis-Driven Redox Biocatalysis section). Thereby, the feed regime applied in the HLHC experiment (Figure 4C) was scaled to different biocatalyst concentrations (Figure 5A). The highest biomass concentration tested (5 g_CDW L⁻¹) indeed yielded the highest product titer (17.0 mM) but accumulated C-one (36.9 mM) and C-ol (11.4 mM) in high amounts. To avoid extreme substrate and by-product accumulation and thereby optimize product titer and yield, a more conservative C-one feed was targeted. To this end, an ideal C-one conversion rate was calculated by considering product inhibition only and neglecting substrate and by-product inhibition, reflecting the upper limit of C-one supply which can be handled by the catalyst. This continuously decreasing rate was translated into a feed rate followed by simulation using the kinetic model, again testing different biocatalyst concentrations. Irrespective of the initial feed rate (30, 20, or 10 U g_CDW L⁻¹, Figures 5B–D), the overall picture remained the same: Higher biocatalyst concentrations (2 and 5 g_CDW L⁻¹) heavily accumulated C-one and—to a minor extent—C-ol, limiting product formation. Although lower biocatalyst concentrations (0.75 and 1 g_CDW L⁻¹) suffered less C-one accumulation and consequential C-ol formation, the final product titer did not exceed 10.7 mM. To conclude, the “optimized” C-one feed regime led to only a moderate increase in final product titer compared to the HLHC experiment (15.5 vs. 11.5 mM), but only with 5 times higher biocatalyst concentration and thus on the cost of low biocatalyst-based yield (0.4 vs. 1.3 g _ε-Cl+6-HA g_CDW ⁻¹) and product yield (0.15 vs. 0.88 mol_ε-Cl+6-HA mol_C-one ⁻¹). With the same biocatalyst concentration as in the HLHC experiment, the achieved product titer was even lower (10.7 vs. 11.5 mM), which emphasizes the strong impact of the by-product C-ol demanding a more aggressive feed rate decrease. Overall, the reaction system remains intricate, as product titer and yield are not only restricted by product inhibition but also by the difficile interplay between competitive by-product inhibition and the formation of this by-product: A high substrate concentration can counteract competitive inhibition but simultaneously increases inhibitor formation. Importantly, assuming the absence of light limitation for higher cell densities did not enable significantly higher product titers than experimentally obtained, reemphasizing that kinetic restrictions are the primary factors limiting the performance of the investigated reaction system.

The development and application of photosynthetically active microorganisms as microbial hosts is still restricted by the limited availability of molecular biology tools and the comparably poor knowledge on cell metabolism and physiology. Here, we investigated different expression systems regarding expression levels and specific activities of BVMO from Acidovorax sp. CHX100 in Syn6803. The Ni²⁺-inducible PnrsB promoter system was found to enable high BVMO expression levels and be superior to the Cu²⁺-inducible PpetE promoter system in terms of inducibility and expression level, in accordance with previous studies (Englund et al., 2016; Liang et al., 2018). It is important to note that the expression level achieved with a promoter depends on the genetic context (Englund et al., 2016). The expression levels of up to 1.1 ± 0.2% of TSP reached in this study are to our best knowledge unprecedented with the PnrsB system in Syn6803. Numerous studies have characterized different controllable promoter systems in Syn6803: non-native ones such as the LacI-repressible Ptrc1O system (Huang et al., 2010), the TetR-regulated L03 promoter (Huang and Lindblad, 2013), and other Plac variants (Thiel et al., 2018) as well as native ones like Cu²⁺-inducible PpetE (Briggs et al., 1990) and Ni²⁺-inducible PnrsB promoters (Englund et al., 2016). However, most studies focused on the optimization of expression systems rather than on designing suitable biocatalysts and use fluorescent proteins as read-out for expression strength. Thus, not much is known about the cellular concentrations of heterologous protein. For catalytically active enzymes, such as dehydrogenases or monooxygenases (Lassen et al., 2014a), constitutive systems, such as the discovered “super strong” promoter Pcpc560 (Zhou et al., 2014) or the light-inducible promoter PpsbA, have been applied (Wang et al., 2018). They yield high expression levels up to 15% (Zhou et al., 2014) and 12.6% of TSP (Wang et al., 2018). However, for oxygenase biocatalysis, well-steerable promoter systems are more desirable to control metabolic burden such as ROS formation. They yield lower but still remarkable expression levels of 0.4% (Sebesta and Peebles, 2020) and 1.1% of TSP (Savakis et al., 2015), of which the latter was met by Syn6803_Ni_pBVMO generated in this study.

With PnrsB, plasmid-based expression yielded a 2.6 times higher BVMO level and activity than chromosome-based expression (Figure 1). This difference may be due to a higher copy number of the plasmid than the chromosome (Huang et al., 2010). It however has been reported that the copy number of pPMQAK1 is comparable to that of the chromosome (Jin et al., 2018). Such contradiction has been explained by the observation that both plasmid and chromosome copy numbers vary strongly in Syn6803 (Heidorn et al., 2011). This makes chromosomal integration the more laborious method due to the necessity for segregation, whereas the plasmid-based approach can suffer instability and the necessity for constant antibiotic selection (Kadisch et al., 2017).

The strong correlation of specific BVMO activity and the expression level (Figure 1) suggests that the achieved BVMO level still limits the biocatalyst activity. Future advances regarding molecular biology tools for cyanobacteria can be expected to enable even higher biocatalyst activities. Promising candidate expression systems include the recently characterized rhamnose-inducible promoter PrhaBAD (Kelly et al., 2018) or the vannilate-inducible promoter PvanCC (Behle et al., 2020).

BVMO containing Syn6803 was found to feature a 3.7 times lower apparent uptake constant K_S for C-one (80 ± 23 µM) than Pseudomonas taiwanensis VLB 120 containing the same enzyme (316 ± 21 µM) (Bretschneider et al., 2021a). This indicates a rather efficient substrate mass transfer into the cell. Furthermore, we observed moderate substrate inhibition, an often encountered effect with BVMOs (Scherkus et al., 2017; Fürst et al., 2019; Schmidt and Bornscheuer, 2020). The host metabolic background interferes with the biotransformation in terms of C-one reduction to C-ol as reported before (Böhmer et al., 2017), but with a rather low V_max (2.75 ± 0.15 mM) and a rather high K_S (7.4 ± 0.6 mM). More important was the inhibition of BVMO activity by C-ol already at low concentrations (K_i = 0.03 mM). This is in agreement with published data on heterotrophs containing the same enzyme (Schäfer et al., 2020), indicating an inhibitory effect on the enzyme level. To reduce formation of the (inhibiting) by-product, the introduction of an enzyme for re-conversion of C-ol to C-one could be an interesting solution (Bayer et al., 2017). Deletion of intrinsic alcohol dehydrogenases constitute another option but may harm cell viability and thus be tedious. Taken together, these data show that controlling substrate concentration is key for efficient photosynthesis-driven BVMO catalysis.

The optimized strain Syn6803_Ni_pBVMO reached a maximal specific activity of 60.9 ± 1.0 U g_CDW ⁻¹. This is around 4 to 5 times lower than published for the heterotrophic host Pseudomonas taiwanensis VLB120 (Schäfer et al., 2020), which showed significantly higher BVMO expression levels (Bretschneider et al., 2021a) and features a highly active energy and redox metabolism based on organic acids and sugars (Volmer et al., 2014). It is however noteworthy that the BVMO exhibited a much higher in vivo activity in Syn6803 than in P. taiwanensis, that is, a 4.6 to 11 times higher k_cat (6.9–17.6 s⁻¹ vs. 1.5 s⁻¹ (Bretschneider et al., 2021a)). It should be noted that BVMO catalysis in Syn6803 appeared not to be limited by the metabolic capacity to supply reduction equivalents as the photoproduction of NADPH in Syn6803 is in the range of 9–18 U mg Chl_a (Kauny and Setif, 2014), which translates to 106–212 U g_CDW ⁻¹. Presumably, this limitation was decisive for BVMO catalysis in P. taiwanensis. The activities achieved already indicate the great potential of Syn6803 as host for energy-dependent biotransfomations. Future research efforts to fill knowledge gaps regarding expression systems, other molecular biology tools, and cofactor regeneration can be expected to further improve photosynthesis-driven biotransformation efficiency (Hitchcock et al., 2020).

Despite keeping by-product and substrate strictly below inhibiting values, the biocatalyst performance was found to be compromised in the long term. Neither increased light nor CO₂ supply had a positive effect (Figure 4B). Instead, the product ɛ-Cl itself was found to affect biocatalyst performance (Supplementary Figure S6). Such product inhibition also was the decisive limitation in simulations based on kinetic data, which overall fit well the bioreactor data. Interestingly, product inhibition was not detected with heterotrophic hosts (Schäfer et al., 2020) and seems to be Syn6803 specific. Effects of ɛ-Cl on Syn6803 physiology may involve inhibition of metabolism leading to NADPH limitation as discussed below. To address product inhibition, in situ product removal technologies can be applied (Dafoe and Daugulis, 2014). Whereas two liquid-phase approaches are compromised by the intermediary polarity of the product, solid-phase extraction may be a promising but costly solution. Alternatively, BVMO catalysis can be amended by a lipase (Scherkus et al., 2017) or lactonase (Schäfer et al., 2020) reaction to obtain 6-HA which is less problematic in terms of inhibition but still feasible as a polymer precursor.

The biocatalyst-based yield (g_Product g_CDW ⁻¹) has been proposed as an important measure for catalyst performance (Fürst et al., 2019). Syn6803_Ni_pBVMO reached a yield of 1.3 and 1.6 g_Product g_CDW ⁻¹ in a 2-L photo-bioreactor and in shaking flasks, respectively. A sensitivity analysis based on the kinetic model affirmed a yield of 1.5 g_Product g_CDW ⁻¹ to be realistic for the used bioreactor setup. This is on the same level as an industry-relevant BVMO-based process with Escherichia coli featuring a yield of 1.6 g_Product g_CDW ⁻¹ for a however more complex product (Woo et al., 2018). However, phototrophic processes are not on the same readiness level as heterotrophic biocatalytic processes: It should be noted that the abovementioned metric neglects the immense differences in cell densities applied. The mentioned E. coli–based process uses around 25-times higher biomass concentrations than those applied in this study and thus features a higher volumetric productivity. On the other hand, Syn6803-based biocatalyst production, regeneration, and operation are based on CO₂ as carbon and water as electron source, making the addition of an organic carbon and energy source such as glucose dispensable. Future improvements of reactor design and cultivation strategies such as the use of capillary biofilm reactor concepts (Karande et al., 2014; Hoschek et al., 2019a) or immobilized cells (Vajravel et al., 2020) and internal illumination (Hobisch et al., 2021) are expected to push the boundaries of phototrophic cultivation/process operation. Already noteworthy are the initial and 24 h-average productivity (187 ± 1 mg L⁻¹ h⁻¹ and 59 ± 1 mg L⁻¹ h⁻¹, respectively), ranging among the highest productivities for biotransformations using cyanobacteria. Additionally, the product yield of 0.88 (mol_ε-Cl+6-HA mol_C-one ⁻¹) prefigures the high potential of phototrophic hosts for biocatalysis.

The comparably high in vivo BVMO activity obtained in this study indicates that—sufficient light supply provided—NADPH can efficiently be withdrawn from phototrophic metabolism to fuel productive reactions. As confirmed by process simulation merely based on whole-cell reaction kinetics, the photosynthesis-based NADPH supply capacity did not limit the biotransformation, with efficient illumination as a prerequisite. The product itself might interfere with NADPH supply at higher concentrations by affecting membrane integrity (Sikkema et al., 1995) and thus photosynthetic electron transfer. In contrast, electron consuming redox biotransformations catalyzed by the ene-reductase Yqjm were found to be limited by NADPH availability at activities above 150 U g_CDW ⁻¹ (Assil-Companioni et al., 2020). By inactivating the proteins Flv1/Flv3, an approach known to reduce electron loss to unproductive O₂ reductions (Thiel et al., 2019), Syn6803 reached specific activities up to 170 U g_CDW ⁻¹. It has to be investigated though, if this elevated electron supply can be sustained in the long term.

As oxygenases are prone to uncouple, that is, reduce O₂ without concomitant substrate oxidation, ROS formation represents another electron sink increasing the electron demand. Furthermore, ROS formation possibly affects photosynthetic metabolism (Nishiyama et al., 2005). Cyanobacteria have evolved different mechanisms to cope with and detoxify ROS (Latifi et al., 2009), which however involves additional electron/energy investment further increasing the electron demand.

Today, the disposability of electrons from water oxidation, for example, by water electrolysis, is a field of particular interest, especially for H₂ production. In this context, envisaged light-to-H₂ solutions using cyanobacterial photosynthesis as a transformation module are highly attractive (Appel et al., 2020). Success towards this goal relies, beside O₂ tolerance of biocatalysts and high cell-density cultivation concepts, on the electron supply capacity of photoautotrophic microbes. It is still unclear, if a high electron demand for technically feasible H₂ production can be met by cells still able to sustain themselves. Interestingly, additional electron sinks were shown to increase photosynthetic efficiency and thereby electron fluxes through the photosynthetic apparatus (Zhou et al., 2016; Grund et al., 2019). This observation reinforces the sink-limitation hypothesis, which states that photosynthesis is rather limited by the availability of electron sinks than by electron (energy) sources (Berepiki et al., 2016). This gives the possibility to tap unused electron supply potential for electron consuming reactions. Limits of and factors influencing electron supply via the photosynthetic light reaction for biotechnological application remain to be investigated. Reductases, such as Yqjm (Assil-Companioni et al., 2020), and oxygenases, such as the cycloalkane hydroxylating cytochrome P450 monoxygenase (Hoschek et al., 2019b) and the BVMO investigated in this study, have now been established as prominent electron sinks in cyanobacteria and constitute excellent blueprints to decipher the potential of the photosynthetic light reaction as electron source for biotechnological application.