Assessing the stereoselectivity of Serratia marcescens CECT 977 2,3-butanediol dehydrogenase

α -Hydroxy ketones and vicinal diols constitute well-known building blocks in organic synthesis. Here we describe one enzyme that enables the enantioselective synthesis of both building blocks starting from diketones. The enzyme 2,3-butanediol dehydrogenase (BudC) from S. marcescens CECT 977 belongs to the NADH-dependent metal-independent short-chain dehydrogenases/reductases family (SDR) and catalyses the selective asymmetric reductions of prochiral α -diketones to the corresponding α -hydroxy ketones and diols. BudC is highly active towards structurally diverse diketones in combination with nicotinamide cofactor regeneration systems. Aliphatic diketones, cyclic diketones and alkyl phenyl diketones are well accepted, whereas their derivatives possessing two bulky groups are not converted. In the reverse reaction vicinal diols are preferred over other substrates with hydroxy/keto groups in non-vicinal positions.


Introduction
Enantiomerically pure α-hydroxy ketones and vicinal diols constitute a valuable group of molecules useful directly as fine chemicals or as building blocks for asymmetric synthesis of bioactive compounds and agrochemicals. 1,2 Numerous chemical and enzymatic strategies have been devised to address this challenge. 2 To date all chemical approaches [3][4][5] and most of the biochemical approaches [6][7][8] only allow the formation of one of the two chiral products: either the α-hydroxy ketones or vicinal diols. A particular challenge is the development of methodologies allowing for the targeted selective synthesis of both α-hydroxy ketones and vicinal diols, in particular with aliphatic side chains (Scheme 1).
Acetoin reductases/2,3-butanediol dehydrogenases (EC 1.1.1.4 and 1.1.1.76) constitute a less explored group of enzymes. They belong to the family of NADH-dependent metalindependent short-chain dehydrogenases/reductases and are responsible for the accumulation of 2,3-butanediols in high titers (>100 g L −1 ) during the cultivations of species such as Klebsiella, Enterobacter, Serratia and Bacillus. 9 In general, 2,3butanediol is not produced as a single isomer, but as a mixture of meso-2,3-butanediol with (S,S)-butanediol (L-(+)-form) or (R,R)-2,3-butanediol (D-(−)-form). Microorganisms that do not produce the meso diol, generate optically enriched (R,R)butanediol. The variations in optical purity might be due to a single unselective enzyme or due to mixtures of selective enzymes. In particular, the Bacillus stearothermophilus 10 enzyme (2S,3S)-2,3-butanediol dehydrogenase (BDH, EC 1.1.1.76) gave excellent enantioselectivities for a range of substrates towards the S,S-diols. R,R-Diols could be obtained employing for instance Saccharomyces cerevisiae BDH. 11 This indicates that more enzymes with similarly high specificities for other substrates and isomers including the meso isomer should be available. The oxidation of the meso isomer would be of particular interest. In principle, this enzymatic transformation could afford enantio-enriched α-hydroxy ketones with up to 100% yield rather than the 50% obtained through a kinetic resolution approach.

Cloning and heterologous expression of BudC
In the present work, the budC gene encoding to a 2,3-BDH from S. marcescens CECT 977 (756 bp) was amplified from genomic DNA and cloned into pET-28a in frame with the N-terminal His 6 -tag to facilitate purification of the recombinant enzyme. The DNA sequence exhibited a 91.8% identity with S. marcescens H30 strain 25 and both proteins were identical at protein level. This means that the production and purification described earlier 25 is also valid for the enzyme described here. We therefore concentrated on substrate range and enantioselectivity.

Enzyme kinetics
The budC gene was successfully expressed in E. coli BL21 (DE3) star cells and the enzyme was purified from E. coli cellfree extracts by Ni-NTA affinity chromatography. Activity of crude extracts reached 14.4 U mg −1 using rac-acetoin (50 mM) as substrate at pH 7.0 in a reductive assay (ESI ‡ Fig.  S1). SDS-PAGE gels revealed the production of a protein of approximately 28 kDa in agreement with the calculated molecular weight (28.4 kDa). Native gels confirmed the tetrameric oligomerization state, yielding a band between 120-140 KDa (ESI ‡ Fig. S2) in line with the earlier results. 25 BudC catalyses the reduction of rac-acetoin preferably at weakly acidic pH 5.0 rather than neutral pH (975 U mg −1 vs. 250 U mg −1 ). However, kinetic parameters at pH 5.0 were difficult to determine, yielding inconsistent data due to the low stability of the enzyme at this pH (t 1/2 : 2.5 h, Fig. 1).
In line with these observations, only kinetic parameters at pH 7.0 were determined (Table 1, ESI ‡ Fig. S4, A-D). In the reductive reaction diacetyl is the preferred substrate of BudC over acetoin, while only meso-2,3-butanediol oxidation was catalysed by the enzyme under the conditions assessed. Despite its protein sequence identity with the enzyme from S. marcescens H30, the specific activities obtained in our study with BudC were significant higher than those reported earlier (up to 4.5-fold at pH 5.0) 25 under identical reaction conditions.

Bioconversions
When reduction reactions containing stoichiometric amounts of substrate (diacetyl or acetoin), NADH (5 mM) and BudC (0.5 μM) were performed, no product formation was observed in any case. Variation of the NADH concentration revealed this to be due to NADH inhibition. Although not previously reported for this enzyme, NADH inhibition was observed at concentrations higher than 0.3 mM; 1 mM being sufficient for complete inhibition (ESI ‡ Fig. S5).
Consistent with the previous results, nicotinamide cofactor regeneration strategies are essential to achieve significant product yields. Ethanol, 2-propanol, and acetone were tested as co-substrates for substrate-coupled cofactor regeneration. However, none of them was accepted by BudC (Table 2). An enzyme-coupled approach was accomplished employing formate dehydrogenase and glucose dehydrogenase (both as crude cell extracts) to recycle NADH, and NADH oxidase (as crude cell extracts) to regenerate NAD + . With these systems at hand, both the BudC catalysed oxidation and reduction could be investigated.
Overall, BudC is S-selective for the reduction reaction of diacetyl to (2S,3S)-2,3-BDO but displays a preference for (R)acetoin as a substrate for the second reduction step, which is also reduced with S-selectivity. In the reverse oxidation reaction meso-2,3-BDO was converted under the tested conditions, while the other diols and acetoin are poor substrates for the oxidation reaction (Scheme 2).

Substrate scope
In order to broaden the applicability of BudC, we extended our investigation to structurally diverse α-diketones and vic-diols. Enzymatic activity was determined spectrophotometrically and the results are presented in Table 2. In the reduction reaction 2,3-diketones were good substrates of BudC, while the enzyme activity decreases with the increase of alkyl chain length. Sterically more demanding ketone moieties, such as 1,2-cyclohexanedione and 1-phenyl-1,2-propanedione were also converted by BudC, while bulky-bulky α-diketones, such as benzil, or α-hydroxy ketones (i.e. benzoin) were not converted. Thus, BudC catalyses the reduction of aliphatic symmetric and non-symmetric α-diketones, while its activity is severely affected by the presence of bulky moieties next to the carbonyl groups either due to steric hindrance (i.e. 1c, 1d and 1e) or conformational effects (1c and 1f).
In the oxidation reaction non-vicinal diols such as 2,4pentanediol and 1,3-butanediol were no substrates of BudC. These results are in agreement with those previously reported by Zhang et al. 25 indicating that vicinal diols (1,2-, 2,3-or 3,4) are required for catalysis.

Analysis of BudC stereoselectivity
The reduction of compounds 1a-1f was further investigated employing both NADH regeneration systems using formate dehydrogenase or glucose dehydrogenase, depending the reaction scale ( Fig. 2 and Scheme 3).
Reaction products were isolated by column chromatography and characterised by 1 H and 13 C NMR, optical rotation, and by comparison with the products obtained under the same conditions with a well characterised commercial enzyme ADH-380 (Evoxx, Düsseldorf, Germany), which is known to yield the corresponding S-stereoisomers. Moreover, ADH-380 catalysed reactions were used as positive control. 26 Slight differences in reaction yields and stereoselectivities were observed depending on the regeneration system employed (crude extracts of formate or glucose dehydrogenase).
2,3-Pentanedione (1a) was reduced to the diol (2S,3S)-4a in good yield and selectivity after 1 h. After 24 h the yield was improved further but a slight loss of selectivity was observed ( Table 3, entry 1-2, ESI ‡ Fig. S7 and S11). The results are in line with those for diacetyl (Scheme 2).
2,3-Hexanedione (1b) was also rapidly reduced to the corresponding α-hydroxy ketone, with (S)-3-hydroxy-2-hexanone (3b) as the main reaction product (80% yield, 99% ee, Table 3, entry 4), while 2,3-hexanediol was produced only in low yields. High yields of 3b were also achieved at semipreparative scale with a minor loss of enantioselectivity (Table 3, entry 5, ESI ‡ Fig. S8 and S12). It should be noted that the less accessible 3-ketogroup was selectively reduced and the reaction almost stopped completely after a single reduction. The second reduction step was much slower, allowing to selective production of the acyloin. A modified reaction profile was observed with 2,3heptanedione (1d). A mixture of 2-hydroxy-3-heptanone (2d) and 3-hydroxy-2-heptanone (3d) was produced, the latter again as the main reaction product ( Table 3, entries 9-10). As with 1b the less accessible 3-ketogroup was thus reduced preferably. Despite extensive efforts, the regio-isomers could not be separated by column chromatography; therefore the absolute stereochemistry of 3d could not be determined (ESI ‡ Fig. S10 and S15-S17). Based on the other results it can tentatively be assigned as (S). (2S,3R)-2,3-Heptanediol (4d) was obtained in low yields (<5%). Neither longer reaction times nor fresh pulses of the enzyme improved the yield of the diol.
In contrast to other dehydrogenases (i.e., ADH-A from Rhodococcus ruber or ADH-T from Thermoanaerobacter sp.), 19,20 where the extension of α-diketone alkyl chain (n = 1 to n = 3) increased the selectivity of the enzymes, a lower regio-selectivity was observed in case of budC, indicating its preference for small substrates.
1,2-Cyclohexanedione was completely reduced to (S,S)-1,2cyclohexanediol (4f) in good yields (66%) and with excellent ee (99%) although poorer de (67%), as cis-1,2-cyclohexanediol was also detected as a reaction product. It could not be distinguished whether the low de was due to poor selectivity in the first or second reduction step.

Homology modelling of BudC and substrate docking
An homology model of BudC was made based solely on the structure of the meso-2,3-BDH from Klebsiella pneumoniae (KpBDH, PDB 1GEG). 27 This resulted in a model in which the positioning of several active site residues is in line with another meso-BDH, which seems important for the enantioselectivity of the enzyme. 28 The model shows the enzyme as a Scheme 3 Stereoselective reduction of 1,2-diketones using BudC from S. marcescens CECT 977. ND stands for not determined. a Reaction conditions: the corresponding diketone (1a-1f, 10 or 50 mM) was added to 50 mM potassium phosphate buffer pH 7.0 with the nicotinamide cofactor (0.16 mM), BudC (2 U ml −1 ) and regeneration system (FDH or GDH, 2 U mL −1 ). b Yields, ees and des were determined by chiral GC analysis (when possible). c Small peaks could not be assigned due to missing standards, therefore it was impossible to distinguish between ee and de.  S18 and S19). The important active site residues for SDRs (Asn-Ser-Tyr-Lys) are in the exact same position as in other BDHs, the only difference is Asp111 in the place were normally an Asn is located. This residue plays an essential role in keeping the conformation of the active site and allowing a proton relay system, 29 which is still there involving the Asp residue. Docking different substrates and products did not result in good binding poses. This is probably due to the small nature of the substrates. This was confirmed by the fact that docking these small compounds into the crystal structure of the KpBDH also did not result in proper binding modes. Furthermore, the loop just outside the binding pocket is probably moving upon binding of the substrate which cannot be mimicked by the docking in silico.
Normally the reduction of the proximal keto group results in the S-enantiomer of the corresponding α-hydroxy ketone. The pro-R NADH hydride transfer from NADH to the re-face of diketones yields (S)-hydroxy ketones, indicating that the enzyme obeys canonical Prelog addition (Fig. 3). Small diketones are most probably fixed at the correct distance to NADH with hydrogen bonds, similar to KpBDH. Since the active site easily can accommodate larger substrates, the larger and more aliphatic penta-, hexa-and heptadione can bind into the active site in an alternative orientation. In this way the 3-keto group may be positioned properly towards the NADH, resulting in the respective 3S-hydroxyketone. Reduced affinity of this alternative binding mode can explain the decreased activity for the larger substrates.

Conclusions
The BudC gene was successfully cloned and recombinantly expressed in E. coli. The enzyme is (S)-selective for all tested substrates and has a broad substrate scope. Unlike other ADHs it displays a better stereoselectivity for smaller substrates. Although the enzyme is strictly S-specific for all small substrates, meso-2,3-BDO is produced with excellent selectivity using Serratia marcescens strains. The production of this compound can be explained by the accumulation of R-acetoin in these strains, which is a very good substrate for the S-specific BudC.
By varying reaction conditions, single or double reductions as well as oxidations could be achieved in one step. This makes this enzyme an outstanding biocatalyst for the synthesis of different chiral compounds from prochiral diketones and diols.

Strains, plasmids and culture conditions
All standard recombinant DNA procedures were performed as described by Sambrook et al. 30 The host strains Escherichia coli DH5α (Thermo Fisher Scientific, Breda, The Netherlands) and Escherichia coli BL21 star (DE3) (Thermo Fisher Scientific) were used in sub-cloning and expression experiments, respectively.
Serratia marcescens CECT 977 was supplied by the Spanish Type Culture Collection (CECT, University of Valencia, Spain) and as DNA source for budC cloning. pGEMT-easy (Promega, Promega Benelux B.V., Leiden, The Netherlands) was used as cloning vector and pET-28a (Merck Millipore, Amsterdam, The Netherlands) as expression vector.

Construction of pET-28a budC expression vector
S. marcescens genomic DNA was extracted following a standard phenol/chloroform procedure as reported by Sambrook et al. 30 A budC consensus sequence was generated The PCR product was first purified using a Gel purification kit (Qiagen, Venlo, The Netherlands), followed by an A-tailing procedure with Taq DNA polymerase (Qiagen).
BudC gene was ligated into pGEM® T-Easy vector system (Promega, Leiden, The Netherlands), and the resulting plasmids were produced in E. coli DH5α. The resulting constructs were confirmed by DNA sequencing, restricted with NdeI and XhoI, and the coding gene was ligated into the corresponding sites of pET-28a vector (Merck Millipore, Amsterdam, The Netherlands), introducing a (His) 6 -tag at the N-terminus of budC gene product.

BudC purification
Cell-free extracts (CFE) were obtained incubating first the cells in binding buffer (sodium phosphate 50 mM, 300 mM NaCl, 20 mM imidazole, pH 8.0) containing egg white lysozyme (1 mg ml −1 ) and DNAse I (0.1 mg ml −1 ) for 1 h at 4°C, followed by 2 cycles of cell disruption (French Press, 1.5 kBar) and finally centrifuged at 12 500 × g and 4°C for 45 min. CFE was filtered through 0.45 μm filter and applied onto a 5 ml Ni-NTA HisTrap HP column (GE Healthcare Life Sciences, Eindhoven, The Netherlands) previously equilibrated with binding buffer. An imidazole gradient was then applied (20 mM-500 mM imidazole, 3 ml min −1 , 60 minutes) and BudC was eluted with about 350 mM of imidazole.
Enzyme fractions were analysed by SDS-PAGE (12% Bis-Tris gels, Bio-Rad, Munich, Germany). Fractions containing BudC were combined, concentrated by centrifugation using an Amicon filter (10 kDa, Merck Millipore) and applied onto a PD-10 desalting column (Thermo Fischer Scientific) previously equilibrated with 50 mM potassium phosphate, pH 7.0. Enzyme aliquots were frozen in liquid nitrogen and stored at −80°C. Protein concentration was determined by the bicinchoninic acid assay (Interchim Uptima BC assay, Interchim, Montluçon France) using bovine serum albumin (Bio-Rad, Veenendaal, The Netherlands) as standard.

Enzyme kinetics
Enzymatic activity was determined spectrophotometrically by monitoring NADH consumption (oxidation) or formation (reduction) at 340 nm and 37°C using an extinction coefficient of 6.22 mM −1 cm −1 . One unit of activity was defined as the amount of enzyme catalysing the oxidation/reduction of 1 μmol of nicotinamide cofactor per minute under standard conditions (pH 7.0, 37°C).
Specific activities were measured in potassium phosphate buffer (50 mM, pH 7.0, 1 ml) in presence of 50 mM of substrate and 0.16 mM NADH or 0.32 mM NAD + . DMSO (5% (v/v)) was added as a co-solvent to improve the solubility of benzil and benzoin. Reactions were started by addition of the enzyme solution and measured for 3 min. Activities of the CFE was determined using E. coli BL21 (DE3) star pET28Ĳa)-buC or E. coli BL21 (DE3) star pET28(a) as negative control.
Kinetics parameters were investigated using 0.27 nM enzyme in presence of varying substrate (0.01-200 mM) or nicotinamide cofactor concentrations (12.5 μM to 5 mM) at 37°C . Data were fitted to the Michaelis-Menten equation (nonlinear regression, SigmaPlot 8.0) to estimates of K m and k cat .
Reactions were incubated at 37°C and 600 rpm during 24 h. Aliquots (1 ml) were taken at different reaction times, centrifuged at 13 000 rpm for 2 minutes and supernatants were first saturated with a NaCl saturated solution (0.1 ml) followed by extraction with ethyl acetate (containing isoamyl alcohol as internal standard) (0.5 mL × 2). The combined organic layer was dried over Na 2 SO 4 and analysed with chiral GC with respect to yield, chemoselectivity and ee (see ESI ‡).