Identification and circumvention of bottlenecks in CYP21A2‐mediated premedrol production using recombinant Escherichia coli

Synthetic glucocorticoids such as methylprednisolone are compounds of fundamental interest to the pharmaceutical industry as their modifications within the sterane scaffold lead to higher inflammatory potency and reduced side effects compared with their parent compound cortisol. In methylprednisolone production, the complex chemical hydroxylation of its precursor medrane in position C21 exhibits poor stereo‐ and regioselectivity making the process unprofitable and unsustainable. By contrast, the use of a recombinant E. coli system has recently shown high suitability and efficiency. In this study, we aim to overcome limitations in this biotechnological medrane conversion yielding the essential methylprednisolone‐precursor premedrol by optimizing the CYP21A2‐based whole‐cell system on a laboratory scale. We successfully improved the whole‐cell process in terms of premedrol production by (a) improving the electron supply to CYP21A2; here we use the N‐terminally truncated version of the bovine NADPH‐dependent cytochrome P450 reductase (bCPR−27) and coexpression of microsomal cytochrome b5; (b) enhancing substrate access to the heme by modification of the CYP21A2 substrate access channel; and (c) circumventing substrate inhibition which is presumed to be the main limiting factor of the presented system by developing an improved fed‐batch protocol. By overcoming the presented limitations in whole‐cell biotransformation, we were able to achieve a more than 100% improvement over the next best system under equal conditions resulting in 691 mg·L−1·d−1 premedrol.

high coaffinity of these glucocorticoidal ligands to the mineralocorticoid receptor (MR) causing an aldosterone-like response (Farman & Rafestin-Oblin, 2017). Avoiding such side effects while increasing the therapeutic effect of steroidal drugs is a realistic goal and is often achieved by steroid modification. This can be illustrated particularly well using the example of the synthetic glucocorticoid methylprednisolone, which exhibits reduced MR affinity and multiplied glucocorticoid potency compared with its unmethylated derivative, prednisolone. Methylprednisolone, also known as medrol, is applied for the treatment of arthritis and other inflammatory diseases and thus, highly appreciated by the pharmaceutical industry (Bavetta, Bekhor, Shah, O'Day, & Nimni, 1962). The increasing demand of this compound requires the development of production strategies, that are aimed at achieving high-product yields while maintaining the cost-effectiveness of the process. Besides that, there is a common interest in reducing the ecological footprint to a minimum. The synthesis of this compound is challenging since, in particular, the introduction of a hydroxyl group in position C21 is a demanding step and difficult to achieve by chemical methods. Such reactions often require multistep processes involving the introduction of protective groups and the use of toxic reagents such as halogens. In addition, by-products increase the total expenditure and thus reduce production efficiency due to required purification by chromatographic methods (Koechlin, Kritchevsky, & Gallagher, 1951). The replacement of chemical by biotechnological processes has therefore aroused interest in recent decades, as the latter are highly suitable for the environmentally sustainable and economic production of pharmaceuticals. Bacterial fermentation for the production of pharmaceuticals offers (a) the possibility to accumulate high amounts of biomass by the use of simple and nontoxic raw materials, (b) a high-yield production of catalytically active enzymes and (c) the production of whole cells acting as biocatalysts. The latter can be used for the execution of regio-and stereospecific reactions under mild conditions while the cells provide and regenerate expensive cofactors such as NADPH (Bureik & Bernhardt, 2007). In particular, the use of cytochrome P450 enzymes (P450) in optimized bacterial whole-cell systems shows high potential and suitability for pharmaceutical production (Bernhardt, 2006;Urlacher & Girhard, 2012). Recently, the C21 hydroxylation of medrane has been established in a recombinant Escherichia coli (E. coli) whole-cell system that enables the CYP21A2-mediated production of premedrol, the precursor of methylprednisolone (Brixius-Anderko, Schiffer, Hannemann, Janocha, & Bernhardt, 2015). Although this system allows the production of considerable quantities of this compound, there is still potential for improvement by circumventing existing limitations of the CYP21A2mediated conversion. Frequently observed bottlenecks of P450 systems include an insufficient electron delivery (Bernhardt & Urlacher, 2014;Sagadin, Riehm, Milhim, Hutter, & Bernhardt, 2018), whose efficiency is dependent on the choice of redox partners for bioconversion. Their suitability varies dependent on the respective P450 isoform. Additionally, there is evidence that the ratelimiting transfer of the second electron can be supported by an alternative bypass-pathway via cytochrome b 5 , affecting product formation velocity and efficiency (Vergères & Waskell, 1995).
Another possible limitation that can occur in recombinant wholecell biocatalysis is the molecular obstruction of substrate access to the active site (Wade, Winn, Schlichting, & Sudarko, 2004), which can be modulated by substitution of the relevant residues that cause steric or polar constraints. There is also the possibility of the substrate and product inhibition (Edwards, 1970;Lin et al., 2001), caused by exposure of toxic substrates or complex-formation of the enzyme with the substrate or product at excessively high concentrations, respectively. Here, we addressed all of these possible obstacles and, thus, (a) improved the efficiency of electron supply by using the modified natural redox partner bCPR −27 and the coexpression of

| Bacterial strains and cultivation
All cloning experiments were performed with E. coli strain TOP10 (Invitrogen). Protein synthesis and whole-cell biotransformation were carried out in the E. coli strain C43(DE3) as described in (Miroux & Walker, 1996). Transformed cells were stored as glycerol stock with a 1:1 mixture of an overnight culture and glycerol (80%) at −80°C. latter is an N-terminally truncated version with a C-terminal 3-glycine-6-histidine tag as described (Neunzig et al., 2014). The truncation of 27 amino acids at the N-terminus enables high-yield expression of a soluble catalytically active CPR. The cloning and generation of the vector p21b_bRED(−27) was performed as described in (Brixius-Anderko et al., 2015;Neunzig et al., 2017) ( Figure 1a).

| Cytochrome b 5 and GroEL, GroES chaperonins
The vector pREP4-groESL_cytb5 (Figure 1b) was generated using the pREP4-groESL backbone (Cole, 1996). For cloning of the bovine cytochrome b 5 [UniProt ID: P00171], the expression cassette SmaI-P T7mod -cytochrome b 5 -T T7mod -SalI was amplified via PCR using the template previously described (Klymiuk et al., 2017) and primers listed in Table S1. The T7 promoter sequence was modified (P T7mod ), to reduce the cytochrome b 5 expression level. Hence, adenine was substituted into thymine in position −6 of the T7 sequence as proposed by Ikeda, Ligman, and Warshamana (1992). Following, the digested amplicon was inserted into SmaI and SalI treated pREP4-groESL via ligation.

| Site-directed mutagenesis in M210 of CYP21A2
p21b_bRED(−27) served as a template for a Quik change ® mutagenesis changing position M210 of the bovine CYP21A2 sequence. For this, Pfu polymerase was used and mutagenesis was performed according to manual instructions from Agilent Technologies (Santa Clara, CA). All utilized primers are listed in Table S1. Aliquots were stored at −80°C until used. Protein concentration of bCPR −27 was calculated using a molar extinction coefficient ɛ 585 = 2.4 mM −1 ·cm −1 (Vermillion, Janice, & Coon, 1978). The concentration of bovine cytochrome b 5 was determined using a difference extinction coefficient of 185 mM −1 ·cm −1 for the absorbance change at 424-409 nm (Mulrooney & Waskell, 2000). CYP21A2 concentration was determined by performing difference spectroscopy according to Omura and Sato using a molar extinction coefficient of ɛ 448 = 91 mM −1 ·cm −1 (Omura & Sato, 1964). The determination of reaction velocity dependent on enzyme ratio were performed using the purified enzymes in a reconstituted in vitro system. The reaction was carried out in 250 µl reaction buffer containing potassium phosphate buffer (KP i buffer) [50 mM K 2 HPO 4 / KH 2 PO 4 ; pH 7.4], 1 mM MgCl 2 , 5 mM glucose-6-phosphate and 1 U glucose-6-phosphate dehydrogenase. Purified bCYP21 A2 (0.5 µM) and varying concentrations of the CPR −27 , cytochrome b 5 and medrane were added to the reaction buffer. The reaction was started by application of 500 µM NADPH into the prewarmed reaction mixture, followed by incubation at 37°C for a defined time.
The reaction was stopped by addition of 250 µl of chloroform.

| Steroid quantification by reverse-phase high-performance liquid chromatography (RP-HPLC)
For product quantification via RP-HPLC, the samples were extracted twice with double volumes of chloroform. The organic solvent was evaporated and the remaining steroids were suspended in 50% acetonitrile and separated on a Jasco reverse phase HPLC system of the LV2000 and LV900 series using a reverse-phase ec MN NucleoDur C18 (4.0 × 125 mm; Macherey-Nagel, Betlehem, PA). The column was kept at an oven temperature of 40°C. For measurements of the samples F I G U R E 1 Vector maps containing genes for CYP21A2, CPR, cytochrome b 5 , and chaperonins. Restriction enzyme cleaving sites are depicted in red with corresponding positions in the respective plasmid. (a) Vector map of bicistronic p21b_bRED(− 27) containing the genes of the bovine CYP21A2 and N-terminally truncated bovine CPR (CPR −27 ) with the corresponding ribosome binding sites (RBS). Expression of these proteins is regulated by the T7 promoter. The plasmid contains the gene for ampicillin resistance (AmpR) and a pBR322 origin of replication (ori). (b) Vector map of the plasmid pREP-groESL_cytb5, a modified derivative of pREP4-groESL (Cole, 1996) containing the additional gene of cytochrome b 5 with corresponding RBS under control of a modified T7 promoter (T7*) with decreased induction strength (Ikeda et al., 1992) and the gene for kanamycin resistance (KanR) an acetonitrile/water gradient was applied with a flow rate of 0.8 ml/ min and absorbance of the substances was detected at a wavelength of 240 nm (Solution A: 50% ACN; Solution B: 100% ACN; 0-8 min: 100% A; 8-10 min: 10% A; 10-10.1 min 100% A, 10.1-15 min 100% A). The retention times of medrane (3.2 min) and premedrol (2.2 min) were verified with the respective standards. The relative peak area of each substance was used for calculation of the conversion rates and product concentration. The seed culture was incubated at 37°C overnight. The main culture was generated in 300 ml baffled flasks by inoculation of 30 ml of terrific broth (TB) medium containing the respective antibiotics (100 µg/ml ampicillin and 50 µg/ml kanamycin) inoculated with the seed culture using a ratio of 1:100. When the OD 600 nm reached 0.5, the protein production was induced by the addition of 4 g L -1 arabinose, 1 mM isopropyl-β-D-thiogalactoside (IPTG) and 1 mM δ-aminolevulinic acid. The temperature was reduced to 30°C and the expression was performed for 24 hr at 150 rpm. After the expression period, the cells were harvested by centrifugation (2500 g, RT, 15 min) and washed with 50 mM KP i . After the second centrifugation, the pellet was suspended in 50 mM KP i adjusting cell densities of 24-96 g wet cell weight (g wcw )/L. The biotransformations were performed in 100 ml baffled flasks containing 10 ml of the culture unless otherwise stated. Polymyxin B (32.5 µg/ml) was added for cell permeabilization providing an improved substrate uptake (Janocha & Bernhardt, 2013). Biotransformations using batch cultivation were performed by the addition of 1.2 or 4.8 mM medrane and 2% or 8% (vol/vol) glycerol. Fed-batch fermentation was carried out by adding 1.2 mM medrane and 2% (vol/vol) glycerol, which was repeated for three times in 2 hr periods. Medrane conversions were performed at 30°C and 150 rpm for a defined time. The steroid substrate was dissolved in dimethyl sulfoxide, which did not exceed 4.8% in conversion cultures. The reaction was stopped by the addition of the double volume of chloroform.

| Computational methods
The crystal structure of the bovine CYP21 (PDB 3QZ1) was used as a receptor for docking into the wild-type, whereas the M210V mutant was constructed using the SWISS pdb viewer (version 4.0.1) (Guex & Peitsch, 1997). For the necessary preparation steps before docking AutoDock Tools (Windows version 1.5.6r3) was used (Sanner, 1999).
AMBER charges were assigned to the amino acids of the protein part and Gasteiger-Marsili charges were computed for the heme cofactor.
To account for its oxidation state, the partial charges were adjusted to 0.400e on the iron atom and -0.348e to each of its ligating nitrogen atoms, respectively. The rectangular grid box (56 × 50 × 64) was centered around residue 210 To capture all conceivable binding positions of the ligands in this region. The grid spacing was set to the default value of 0.375 Å. Medrane and 17α-hydroxyprogesterone (17OHP) were constructed manually and energetically optimized using the MM+ force field parameters as implemented in HYPERCHEM (HYPERCHEM, Version 6.02, Hypercube Inc., Gainsville, FL). Atom types and Gasteiger-Marsili charges for use in AutoDock (version 4.2) (Huey, Morris, Olson, & Goodsell, 2007;Morris et al., 1998) were computed with in-house PERL scripts. During docking all rotatable bonds of the ligands were treated as flexible, whereas the protein part was kept rigid. A total of 250 runs of the Lamarckian Genetic Algorithm for each ligand were carried out, otherwise default parameters of AutoDock were applied. Hydrophobicity profiles of the full-length and truncated bCPR were determined and compared using the ExPASy online tool Protscale (Gasteiger et al., 2005) based on the amino acid scale of Kyte and Doolittle (1982).

| RESULTS AND DISCUSSION
3.1 | In vivo effect of the N-terminal CPR truncation on premedrol formation in the E. coli whole-cell system To enhance the previously established whole-cell system in E. coli resulting in premedrol. Since an optimal electron supply is allocated to be an important factor in limiting the conversion yield of P450 mediated systems, we sought to optimize this by substitution of the coexpressed redox partner in the E. coli whole-cell system. Therefore, we modified the membrane-bound microsomal bovine CPR, which is the naturally occurring redox partner of bovine CYP21A2. According to a previous study, the N-terminal truncation of 27 amino acids (−27) of the human CPR led to an improved solubility and higher protein concentrations without affecting the enzyme's activity in a recombinant E. coli system (Sandee & Miller, 2011). For electron transfer, positively charged residues on the proximal site of CYP21A2 are interacting with acidic residues on the CPR surface (Hlavica, Schulze, & Lewis, 2003;Shen, Porter, Wilson, & Kasper, 1989). Since the removal of 27 N-terminal amino acids did not affect catalytic activity (Sandee & Miller, 2011), the binding sites should be unaffected. To this end, we generated a modified version of the plasmid p21b_bRED (Brixius-Anderko et al., 2015), harboring the encoding sequence of bCPR −27 . The truncated protein-enhanced medrane conversion in whole-cell biotransformations by about 19% compared with the reported arh1/etp fd based redox system resulting in 196.8 ± 24.9 mg·L −1 ·d −1 although statistical significance cannot be assured (Figure 2). The presented premedrol production here did not exceed 320 mg L -1 as observed by Brixius-Anderko et al. using arh1/ etp fd , since the latter yield was obtained by multiple addition of lower substrate concentrations summed up to 1 mM. However, comparing premedrol production obtained by non-successive substrate addition, using the bCPR −27 increased the premedrol production 2.4-fold compared with premedrol production supported by WT CPR (Brixius-Anderko et al., 2015).
This difference in efficiency is due to the membrane-bound nature of the WT microsomal CPR which makes it difficult to be overexpressed in a recombinant bacterial system. The removal of the N-terminal membrane-binding domain putatively increased the protein's solubility, as this region is described to be hydrophobic in rat (Gilep, Guryev, Usanov, & Estabrook, 2001), yeast (Venkateswarlu, Lamb, Kelly, Manning, & Kelly, 1998), human (Migita, Togashi, Minakawa, Zhang, & Yoshida, 2005) and rabbit (Masters, 2005)

| In vitro effect of cytochrome b 5 on premedrol formation
As shown before by using a truncated CPR version, the efficiency of electron transfer is one of several limiting factors on whole-cell medrane conversions in recombinant E. coli (Figure 2). Thus, we sought to introduce a putatively supporting protein that is able to interact with the CPR and/or CYP21A2. In microsomes, cytochrome b 5 is colocated with CYP21A2 and CPR and has previously been shown to support several individual P450s in their reaction efficiency, for instance, CYP3A4, CYP2B4 and CYP17A1 (Bart & Scott, 2017;Bridges et al., 1998;Gilep et al., 2001;Katagiri, Kagawa, & Waterman, 1995). However, the supportive effect is dependent on the individual P450 isoform as well as the corresponding redox partner and the molar ratio of the present enzymes (Imai & Sato, 1977;Lu, West, Vore, Ryan, & Levin, 1974;Morgan & Coon, 1984;Schenkman, Jansson, & Robie-Suh, 1976).
The mechanism of cytochrome b 5 action is hypothesized as follows: (a) allosteric activation of P450 supporting the correct positioning towards CPR (Miller, 2005), (b) cytochrome b 5 acts as an electron transfer intermediate between the reductase and P450s.
Whereas the CPR transfers the first electron with high efficiency, cytochrome b 5 provides the rate-limiting input of the second electron to P450 (Schenkman & Jansson, 2003), (c) complexation of cytochrome b 5 to P450 leading to a stabilization of the oxycytochrome P450 complex. In this way, cytochrome b 5 can provide a faster transfer of the second electron than the formation and release of a superoxide anion. The three suggested mechanisms allow the F I G U R E 2 CYP21A2-mediated whole-cell conversion of medrane in dependence on the redox partners (arh1/etp fd ; bCPR −27 ). The 24 hr expression took place in TB medium after induction with 1 mM IPTG, δ-ALA, and 0.4% arabinose. The reactions were carried out in 30 ml of KPi (pH 7.4), containing 30 µg/ml polymyxin B, 2% glycerol, 1 mM medrane, and 24 g wcw /L of recombinant Escherichia coli cells for 24 hr. Extracted steroids were quantified via RP-HPLC. Values represent the mean of three conversion experiments. Error bars indicate respective standard deviations. IPTG, isopropyl-β-Dthiogalactoside; RP-HPLC, reverse-phase high-performance liquid chromatography; δ-ALA, δ-aminolevulinic acid F I G U R E 3 Time-dependent in vitro conversion of medrane to its 21-hydroxylated product, premedrol, in dependence on the CYP21A2:CPR:cytochrome b 5 ratio. The in vitro conversions of 100 µM medrane were performed utilizing purified bCYP21A2, bCPR −27 and cytochrome b 5 in KP i buffer. Independent samples were measured within the linear range (3, 5, 10, and 30 min) via RP-HPLC. Error bars indicate the respective standard deviation. b5: cytochrome b 5 ; CPR, CPR −27 ; P450, CYP21A2; RP-HPLC, reversephase high-performance liquid chromatography increase of product formation without influencing NADPH consumption (Gorsky & Coon, 1986;Gruenke, Konopka, Cadieu, & Waskell, 1995). To investigate the effect of cytochrome b 5 on CYP21A2 activity, we performed in vitro conversions of medrane using the purified bovine enzymes of CYP21A2, bCPR −27 and cytochrome b 5 (Table S2) at varying ratios from 1:4:0 to 1:4:8 (Figure 3). The optimal P450:CPR ratio was previously determined to be 1:4, as shown in Figure S1. At a P450:CPR:cytochrome b 5 ratio of 1:0:1 no product was formed highlighting the mandatory role of CPR for CYP21A2 activity which increased up to 1.25 µM premedrol/min at a ratio of 1:4:0 ( Figure 3). We observed that cytochrome b 5 has a supporting effect on product formation, which is increased at least by 90% at a ratio of 1:4:0.5 compared with 1:4:0 and did not alter considerably at higher P450:bCPR −27 :cytochrome b 5 ratios (Figure 3). The lack of premedrol production at ratio 1:0:1 confirmed the above-stated hypothesis that the electron transfer, at least the transfer of the first electron, is exclusively mediated by CPR and cannot be replaced by cytochrome b 5 (Miller, 2005;Schenkman & Jansson, 2003). The enhancing effect of the cytochrome b 5 at ratios of 1:4:0.5 and above shows that despite the removal of the first 27 amino acids of the CPR protein sequence, interaction between both, CPR and cytochrome b 5 , still is possible resulting in a higher reaction velocity. With regard to the elevating effect of cytochrome b 5 on the CYP21A2 related biotransformation, hypotheses (a)-(c) are conceivable in this context.
By contrast, a previous study described an inhibitory effect of cytochrome b 5 on the CYP21A2 activity in vitro (Wang et al., 2017).
The opposite effect demonstrated there may, on one hand, be due to the use of enzymes from different sources. Whereas in our experiments all enzymes originate exclusively from the bovine species, in those experiments human CYP21A2, cytochrome b 5 and rat CPR were used. Moreover, different substrates were investigated (medrane and progesterone, respectively), which could affect the influence of cytochrome b 5 on the performance of P450 (Bart & Scott, 2017). Finally, in the experiments described by Wang et al. (2017), excessive concentrations of cytochrome b 5 (1:300) were used, which may lead to the contrary effect as observed in our studies, since we did not exceed a P450 to cytochrome b 5 ratio of 1:8. conversion. According to Zhao et al. (2012), the methionine in position 210 of bCYP21A2 is part of the distal 17OHP-binding region which interacts with the substrate within 4 Å and is located at the substrate access channel (Figure 4c). To enhance substrate access to the active site, we thus examined the impact of steric effects in product formation in vivo. To this end, we performed site-directed mutagenesis of M210 by introducing other hydrophobic residues of different size (valine, glycine, alanine, isoleucine and phenylalanine).

| Protein engineering of bCYP21A2 via site-directed mutagenesis and in vivo screening
Indeed, substitution with the smaller valine yielded a 1.3-fold increase in premedrol formation (Figure 5a), resulting in 255 ± 20 mg·L −1 ·48 hr −1 (Figure 5b), whereas the introduction of smaller residues into position M210 produced premedrol levels not different from that of the WT (Figure 5a). We assume that steric and hydrophobic effects play a role in this context. Since valine is a more hydrophobic residue than methionine (Eisenberg, 1984;Kyte and Doolittle, 1982), hydrophobic interactions could be implicated to allow correct substrate orientation and faster guidance towards the catalytically active heme. Our docking results showed energetically preferred binding positions of both medrane and 17OHP in the same region as the ligand in the crystal structure (3QZ1), considering its limited resolution of 3 Å (see Figures 4c and 4d). Interestingly, the obtained docking conformations in the M210V mutant were identical to those in the wild type (data not shown). However, since residue 210 is not in direct contact with the ligands, the introduction of a smaller side chain (valine) obviously does not affect their binding position. Nevertheless, it is likely that conformational changes of the protein go along with the diffusion of the ligands further to the active site, which, however, cannot be elucidated by docking.
Therefore, we assume that the smaller side chain in position 210 affects the dynamics of substrate access indirectly, for example by altered enzyme stability or expression level. However, it cannot be ruled out that the mutation may facilitates the substrate's access by conformational changes of the protein. Unfortunately, experiments analyzing modifications of the active site (M197 with saturation mutagenesis) or the surface (L442A and L224R) of CYP21A2 for altered solubility did not produce any improved variants for premedrol production (data not shown).

| Combinatorial application of developed strategies on whole-cell biotransformation of medrane
Although the effect of CPR truncation on premedrol production and CYP21A2 mutagenesis has been demonstrated in the E. coli wholecell system, the effects of cytochrome b 5 copresence were so far shown only in vitro. This encouraged us to create a functional in vivo whole-cell system that combines the three optimization strategies.
To keep the number of expression vectors to a limit of two, we created a vector containing the genes for the chaperonins GroES/EL and cytochrome b 5 . For this purpose, we modified the vector pREP4-groESL (Cole, 1996) and introduced the gene for cytochrome b 5 under the control of a modified T7 promoter. As the P450:CPR: b 5 ratio 1:4:0.5 is sufficient to have an elevating impact on product formation in vitro (Figure 3) we introduced a point mutation into the T7 promoter according to Ikeda et al. (1992). This modification leads to a residual promoter activity of 13%, which prevents excessive showed an elevating effect on premedrol production by 15% compared with cultures solely expressing WT CYP21A2. The substitution M210V in CYP21A2 yielded 404 ± 28 mg·L −1 ·d −1 , corresponding to a 25% increased product formation compared with WT CYP21A2, which is in good agreement with the observations of the mutant screening shown above ( Figure 5). The additional coexpression of cytochrome b 5 slightly increased premedrol formation by further 13%, summing up to 41% compared with WT CYP21A2 without cytochrome b 5 under batch-cultivation conditions. To investigate whether a possible substrate inhibition could be a bottleneck of the process encountered with high initial substrate concentrations, we changed the protocol from batch-into fed-batch cultivation. To this end, we applied 1.2 mM of the steroidal substrate medrane together with 2% glycerol in four consecutive feeding steps, every 2 hr ( Figure 6, dark columns). Indeed, the successive addition of the substrates, medrane and glycerol, had an increasing effect on final premedrol concentrations as well as on the respective conversion yields. The latter was enhanced by a maximum of 16% (WT CYP21A2 with coexpressed cytochrome b 5 ) when using fedbatch cultivation in comparison to the batch method.
The fed-batch method led to substantially higher product concentrations for both, the CYP21A2 WT and M210 mutant showing an enhancement by 46% and 43%, respectively, compared with respective premedrol concentrations resulted from batch cultivation ( Figure 6, bright columns). This resulted in concentrations of 472 ± 34 (WT) and 578 ± 59 mg·L −1 ·d −1 (M210V) (Figure 6, dark columns). The enhancing effect of the amino acid replacement M210V in CYP21A2 is reflected by an improvement of 22% in premedrol productivity in fed-batch cultivation, compared with the WT CYP21A2. This improvement corresponds to a final product concentration of 578 ± 59 mg·L −1 ·d −1 . However, the enhancing effect of the mutation M210V seemed to be limited to 4% (691 ± 104 mg·L −1 ·d −1 ), when cytochrome b 5 is coexpressed compared with WT CYP21A2 coexpressed with cytochrome b 5 (664 ± 35 mg·L −1 ·d −1 ). The premedrol production is leveled off at 691 ± 104 mg·L −1 ·d −1 . This indicates that there are still other aspects, limiting the process such as restricted substrate uptake or product release. Cultures coexpressing cytochrome b 5 show highest divergence between batch-and fed-batch induced premedrol productivity which correspond to an increase by 79% and 51% of the WT and M210V variant, respectively, compared with the respective batch induced premedrol productivity. The improved method led to final space-time yields of 665 ± 35 and 691 ± 104 mg·L −1 ·d −1 , respectively.
In fed-batch cultivation the coexpression of cytochrome b 5 results in an elevation of final premedrol concentrations by 40% when WT CYP21A2 is used and by 19% when mutant M210V is applied (compared with the respective fed-batch cultivation values without cytochrome b 5 coexpression). The latter reflects the maximum productivity and conversion yield obtained here by the optimized process.
Taken together, we showed that by successive substrate feeding we reduced inhibition encountered with high initial substrate concentrations. This effect was also observed under conditions, where the conversion yield decreases asymptotically with increasing substrate concentrations indicating a substrate-mediated inhibition of CYP21A2 ( Figure S2). Lin et al. showed the same phenomenon for several P450 isoforms, and, thus, proposed a two-site model for understanding dose-depending substrate inhibition in P450catalyzed reactions. They hypothesized that the substrate has high affinity to the active site and lower affinity to a second binding site, which is occupied at extensive concentrations of the substrate (Lin et al., 2001). This model can be helpful to describe the higher productivity by feeding experiments, where substrate concentrations are kept low due to continuous substrate consumption. In vivo, this mechanism could be used for a natural regulation process, since steroid metabolism is a sensitive, fine-tuned physiological system.
Especially the activity of CYP21A2 is needed to be regulated for example by feedback-responses, as the C21 hydroxylation results in activation of the steroids to act as highly potent ligands of the glucocorticoid or MR (Attardi et al., 2007;Boland, 1961). Further, we F I G U R E 6 CYP21A2-mediated whole-cell conversion of medrane dependent on the CYP21A2 variant (WT or M210V) and the presence of cytochrome b 5 (Cyt b 5 ). The 24 hr expression period took place in TB medium after induction with 1 mM IPTG, δ-ALA and 0.4% arabinose. The reactions were carried out in 10 ml of KP i buffer (pH 7.4) for 24 hr, containing 32.5 µg/ml polymyxin B, 2% glycerol and 96 g wcw /L of recombinant Escherichia coli cells. 4.8 mM of medrane was either added at the beginning of the conversion period (no feeding, bright columns) or stepwise in 2-hr time slots each with 1.2 mM medrane (t0-t6) and 2% glycerol (t 2 -t 6 ) (feeding, dark columns). Extracted steroids were quantified via RP-HPLC. Values represent the mean of three conversion experiments. Error bars indicate respective standard deviations. IPTG, isopropyl-β-Dthiogalactoside; RP-HPLC, reverse-phase high-performance liquid chromatography; TB, terrific broth; δ-ALA, δ-aminolevulinic acid showed that the enhancing effects of the M210V mutation and cytochrome b 5 on CYP21A2 activity are recognizable more clearly under the conditions of fed-batch cultivation. Thus, we assume that the inhibitory effect, which is related to a high substrate concentration is a major limitation of the process. Further optimization to exploit the complete catalytic potential could be realized by controlled fermentation in an automated bioreactor, that allows continuous titration of the substrate, which is ideally matched on residual substrate concentration within the culture.

| CONCLUSION
We successfully optimized the CYP21A2-based whole-cell system in E.
coli. The application of a truncated CPR and coexpression of cytochrome b 5 led to increased efficiency of premedrol formation putatively based on an improved electron supply or allosteric interaction (Figure 7). The amino acid replacement M210V located in the substrate access channel caused a slightly increased product formation. The visibility of the supportive optimizations was limited by single-step substrate application, which we finally uncovered by successive substrate feeding. Applying the enhanced fed-batch protocol together with the use of bCPR −27 and CYP21A2_M210V as well as the coexpression of cytochrome b 5 enabled us to produce up to 691 mg·L −1 ·d −1 premedrol within a biotechnological E. coli based whole-cell system which is a more than a 100% improvement compared with our previously reported system under similar conditions (Brixius-Anderko et al., 2015). This is a big step forward towards application of this system in an industrial process for the sustainable production of the methylprednisolone-precursor premedrol.

ACKNOWLEDGMENTS
We would like to thank Dr. Bernd Janocha and Dr. Claus Lattemann for having introduced us to the needs of industrial steroid production.