Structure-guided design of Serratia marcescens short-chain dehydrogenase/reductase for stereoselective synthesis of (R)-phenylephrine

Bioconversion is useful to produce optically pure enantiomers in the pharmaceutical industry, thereby avoiding problems with side reactions during organic synthesis processes. A short-chain dehydrogenase/reductase from Serratia marcescens BCRC 10948 (SmSDR) can stereoselectively convert 1-(3-hydroxyphenyl)-2-(methylamino) ethanone (HPMAE) into (R)-phenylephrine [(R)-PE], which is marketed medically as a nasal decongestant agent. The whole-cell conversion process for the synthesis of (R)-PE using SmSDR was reported to have an unexpectedly low conversion rate. We reported the crystal structure of the SmSDR and designed profitable variants to improve the enzymatic activity by structure-guided approach. Several important residues in the structure were observed to form hydrophobic clusters that stabilize the mobile loops surrounding the pocket. Of these, Phe98 and Phe202 face toward each other and connect the upper curvature from the two arms (i.e., the α7 helix and loopβ4–α4). The mutant structure of the double substitutions (F98YF202Y) exhibited a hydrogen bond between the curvatures that stabilizes the flexible arms. Site-directed mutagenesis characterization revealed that the mutations (F98Y, F98YF202Y, and F98YF202L) of the flexible loops that stabilize the region exhibited a higher transformation activity toward HPMAE. Together, our results suggest a robust structure-guided approach that can be used to generate a valuable engineered variant for pharmaceutical applications.


Results
Structures of the apo-form SmSDR, F98AF202L SmSDR, F98LF202L·NADPH, and F98YF202Y variants. The structure of the apo form of SmSDR was solved to 1.47 Å (SmSDR), revealing a homo-dimer (AB) in the crystal. Weak or negative density was seen at two segments of subunit B (residues 40-50 and 191-208), indicating a flexible region. Subunit A comprises residues 5-249, whereas subunit B comprises residues 2-40, 51-190, and 209-377. Subunit A contacts with subunit B through α1, α6 helices, loop α8-β7 , and a region after the β7 sheet ( Fig. 1A and Fig. S1). A tight homotetramer is created with two homodimers related by twofold crystallographic symmetry (Fig. S2A). Furthermore, size-exclusion chromatography of the purified SmSDR demonstrated a single peak that corresponds to the size of a tetramer (Fig. S2B), confirming that SmSDR is assembled into a tetramer, in accord with the tetrameric structure seen for homologues 28 .
Each subunit is intertwined to form an α/β/α-fold barrel, of which the seven-β-stranded central sheet is flanked by six α-helices on either side, showing a nucleotide-binding Rossmann scaffold. The presumed active-site pocket is surrounded by five loops (β2-α2, β3-α4, β4-α4, α5-α6, and β6-α7) and an α7 helix (acting as a lid) (Fig. 1A). A presumed NADPH-binding cleft is surrounded by β2-α2, β4-α4, β3-α4, α5-α6, β6-α7, and an α7 helix. The α-helix, β-sheet, and loop are colored cyan, magenta, and brown, respectively. Phe98 and Phe202 are depicted as green stick models. The residues of the hydrophobic cluster are colored as purple stick models. The carbon, nitrogen, and oxygen atoms are shown as green, blue, and red, respectively. (B) Close-up view of the hydrophobic core on the presumed binding pocket. Phe98 (green) from the β4-α4 loop and Phe202 (green) from the α7 helix face toward each other and contact with the L100, Y158, N196, and I206 residues (purple). (C) Superposition of subunit A and B reveals flexible side chain of the Phe98.
SCIeNTIfIC RepoRTS | (2018) 8:2316 | DOI: 10.1038/s41598-018-19235-y Two phenylalanines, Phe98 from the β4-α4 loop (residues 94-107) and Phe202 from the α7 helix, face toward each other, thus constituting a bridge structure over the top of the binding pocket ( Fig. 1A and B). Notably, they contact with L100, Y158, N196, and I206 to form a hydrophobic cluster on the peripheral region of the pocket. Superposition of subunit A and subunit B of apo-SmSDR reveals a slight shift near Phe98 at the β4-α4 loop. Of note, there is 1.85-Å deviation between the phenyl Cζ atoms of Phe98 [Cζ(Phe98:A)-Cζ (Phe98:B)] of superimposed subunits, suggesting that Phe98 resides at a flexible region (Fig. 1C).
We therefore prepared the SmSDR-NADPH-HPMAE complex model using an in silico docking method 33,34 . The top-ranking docked pose of HPMAE shows HPMAE-interacting environment ( Fig. S3): (1) The C4 atom of the nicotinamide ring from NADPH contacts with the C3 atom of HPMAE; (2) the carbonyl O atom interacts with the catalytic residue Tyr158; (3) the aromatic ring of HPMAE aligns with that of Phe98 to some extent, yielding displaced π stacking force; and (4) the hydroxyl group of the HPMAE contacts with Ser144 and Asn146. On the basis of this model, we propose a catalytic mechanism that converts HPMAE into (R)-PE through SmSDR at the expense of NADPH to yield NADP + (Fig. 3): Tyr158 acts as a key catalytic player, the nearby Lys162 interacts with the hydroxyl of the tyrosine that lowers the pKa, Ser144 stabilizes the substrate, and Asn118 maintains the active-site framework to build up a proton relay environment as suggested by Filling et al. 31 .
Structure-guided mutagenesis on Phe98 and Phe202. Given that Phe98 and Phe202 are situated at a relatively hydrophobic, flexible region over the top of the binding pocket, we sought to evaluate whether engineering of these sites affected enzymatic activity. We have therefore generated and expressed point-mutation variants using the Escherichia coli expression system: F98Y, F98YF202L and F98YF202Y. We were able to obtain good-diffracting crystals of F98AF202L, F98YF202Y, as well as F98LF202L that grew in the condition consisting of NADPH. Given an atomic resolution diffraction of apo-form (1.47 Å, Fig. S4A), the structure of the mutant F98AF202L was determined to 1.87 Å, showing a clear electron density map of F98A and F202L (Fig. S4B). The structure of the double-tyrosine-substituted mutant F98YF202Y was solved to 1.47 Å, revealing a clear electron density map of the mutated tyrosines (Fig. S4C). In addition, we obtained the F98LF202L·NADPH complex crystal using the soaking method. The 1.90-Å SmSDR mutant structure (F98LF202L·NADPH) was determined, showing a clear density map of the mutated residues F98L and F202L (Fig. S4D). Structural comparison of the mutant structures (F98AF202L, F98LF202L, and F98YF202Y) demonstrated an overall homologous fold (Fig. 4A). A close-up view of the region between residues 98 and 202 showed that there is significant conformational Thus, the replacement of the bulky aromatic chain with a shorter side chain at the region of residue 98 greatly led to the loss of the hydrophobic contacts between two arms. Conversely, F98YF202Y that introduces an additional hydroxyl group on the phenyl side chain creates a strong H bond.
In the F98LF202L·NADPH structure, a clear density map near the bottom of the pocket in Subunit A was observed, which could be modeled as NADPH. Superposition of the apo form of SmSDR, F98AF202L, F98LF202L·NADPH, and F98YF202Y·SmSDR demonstrated an essentially identical fold. Structural comparison indicated relatively low RMSD values among these structures [apo vs. F98AF202L: 0.11 Å (223 Cα atoms); apo vs. F98YF202Y: 0.14 Å (234 Cα atoms); and apo vs. F98LF202L·NADPH: 0.14 Å (226 Cα atoms)] (Fig. 4A). Unlike the conformational change altered in the β-ketoacyl-ACP reductase from E. coli 35 , SmSDR reveals hardly any conformational change upon NADPH binding, providing a stable configuration of the cofactor-binding-site framework for catalysis.
We next evaluated the mobility of these variants based on a thermal stability shift assay that provides a quantitative measure of the enzymatic thermostability 36 . As compared with the wild-type enzyme, F98Y and F98YF202Y had an increased thermal shift (2.08 and 1.28 °C) (Fig. 5). F98YF202L mutant exhibited a more flat peak with biphasic melting temperatures (45.59 and 62.91 °C), implicating a two-state denaturation process for F98YF202L. These results suggested that the introduction of a hydroxyl moiety on the aromatic side chain at residues 98 could enhance the thermostability.
WT, F98Y, F98YF202L and F98YF202Y were subjected for enzymatic kinetic analysis. As shown in Table 1, there was a lower K m value for F98Y, F98YF202L and F98YF202Y, indicating an increased binding affinity with substrate. A slightly lower k cat value was derived for each variant, which led to a comparable k cat /K m value (Table 1). We next evaluated the transformation yield of PE from HPAME using E. coli-expressing variants. Notably, F98Y, F98YF202L and F98YF202Y all exhibited a higher bioconversion yield than did the wild-type enzyme. Of those, F98YF202L had the highest transformation activity as compared with wild-type (1.57-fold). These results suggest that F98YF202L with a biphasic feature maintains its flexibility and stability, hence higher transformation power.
Comparison with NADPH-preferring SDR structures. We compared SDR structures that prefer NADPH as a cofactor involved in the hydride transfer of NADPH 37 . On the basis of Dali analysis 38 , with the F98LF202L·NADPH structure serving as the query, we searched for similar structures with high sequence identity. Of those having a high Z score (>10), three enzymes were observed to share high sequence identity (>40%) with SmSDR: Sinorhizobium meliloti 1021 3-oxoacyl-[ACP] reductase (SmOACPR; 57% identity; pdb code: 3V2G) 39 , Gluconobacter oxydan putative reductase (GoPOR; 43% identity; pdb code: 3WTB) 40 , and Synechococcus elongatus PCC 7942 OACPR, (SeOACPR; 42% identity; pdb code: 4DMM) (Fig. S5) 41 . Of these, only SeOACPR that behaves as an NADPH-preferring enzyme has an available NADPH-liganded structure. The other crucial H-bond with the 2′-phosphate group of the nicotinamide ribose structure is from the conserved basic arginine located at the end of the glycine motif (Ser18Arg19 in SmSDR; Ser17Arg18 in SeOACPR) (Fig. S5B). Thus, SmSDR and SeOACPR belong to a subgroup (cP1) of classical SDRs that acquire a conserved basic residue (Lys/Arg) to interact with NADPH 8 . The 2′-phosphate acts not only as the preference of the coenzyme but also stabilizes the loop β2−α2 .
We also evaluated the effect of A42S, which was previously assumed to introduce an additional contact between the 2′-phosphate group of NADPH involving the loops β2-α2 (Fig. S5B), mimicking that in SeOACPR 42 . A42S had a comparable k cat /K m value as compared with the wild-type enzyme. A higher PE conversion yield was obtained in the A42S variant as compared with wild-type (1.34-fold increase) (Table 1), indicating a slightly beneficial effect in biotransformation. Thus, engineering of these sites (residues 98 and 202) at the α7 helix and loop β4-α4 to enhance the stability, as well as of those residues assumed to interact with the 2′-phosphate group of NADPH to augment the contacts with NADPH 30 , offers a useful strategy for improving the conversion rate of HPMAE.

Discussion
Of the cP1-type classical SDRs, SmSDR and SeOACPR are two members that catalyze the stereoselective reduction of a ketone to its corresponding chiral alcohol 37,43 . Both SmSDR and SeOACPR structures comprise a strictly conserved Asn-Ser-Tyr-Lys tetrad and a highly conserved Rossmann-fold N-terminal lobe that binds to the NADH/NADPH cofactor 10,14,44 . Based on the SmSDRs' structures analyses, the most sensitive sites of SmSDR to temperature are identified on the surface region (α4, α7 and loop β4-α4 at the C-terminal segment; Fig. S6A-D) of the C-terminal lobe. Above all, the α7 helix and loop β4-α4 with high B-factor values extends from the compact N-lobe Rossmann domain, suggesting a flexible region to accommodate various substrates.
Targeted mutagenesis guided by structural analysis to engineer enzymes of acquired features has been emerged as a useful strategy 45 . F98 and F202 that protrudes from two arms contribute to hydrophobic contacts thus provide a potential site for engineering to enhance the enzymatic thermostability (Fig. 4) since they are less  likely to disturb the catalytic framework and increase the likelihood for obtaining active variants with increased thermostability. Thermostability analysis revealed that the F98Y mutation with an additional hydroxyl group could effectively enhance the thermostability (∆Tm: 2.08 °C), whereas the F98YF202L showed a biphasic melting pattern (Tm: 45.59 and 62.91 °C). These variants exhibited relatively comparable k cat /K m values, suggesting that the overall active-site framework is reserved. Interestingly, the whole-cell transformation showed that all three variants had a higher PE biotransformation yield, in which F98YF202L had the best performance. From the structural point of view, the phenyl moiety of Phe98 contacts with the aromatic ring of Tyr158 (Phe98(Cδ1)-Phe158 (Cγ), 3.70 Å), yielding an edge-to-face aryl-aryl interaction importantly to orient the precise position of the catalytic Tyr158 46 . F98Y is likely to introduce an additional H contact between the O atom of Tyr98 and N atom of Asn146, hence increasing the stability based on the F98Y structure model. The F98YF202Y mutation that introduces an extra H bond seems to strengthen the conformation rigidity, hence resulting in an increased HPMAE affinity. Yet, a lower k cat value of F98YF202Y implicated that this mutation might restrict subsequent catalytic events possibly due to its inflexibility. On the other hand, F98YF202L that gains the interaction with Asn146 at residue 98 while introduces more flexibility at residue 202 as shown by its biphasic melting feature had the highest transformation yield. These results together suggest that engineering the thermally sensitive loops to strengthen protein stability while maintain its precise catalytic architecture and flexibility represents a robust structure-guided approach for pharmaceutical applications and that F98LF202L and F98YF202Y are profitable mutants 47 . Engineering the cofactor affinity is an additionally useful strategy to improve the production efficiency 48,49 . From the practical point of view, NADH is a much economical cofactor (15 times price lower than NADPH) for production of (R)-PE by a cell-free method. It is even more desirable by the use of a whole-cell biocatalysis transformation process that directly produces NADH/NADPH using the media carbon source; log-phase E. coli culture generates three times higher concentrations of NADH than NADPH [48][49][50] . Analysis of the NADPH-liganded structures between SmSDR and SeOACPR shows that the 2′-phosphate group of the nicotinamide-ribose points toward loop β2-α2 . Substitution of alanine by serine at residue 42 (SmSDR A42S) on this loop slightly enhanced biotransformation activity, despite comparable enzymatic activity. This might have implication that this mutation had a switch in cofactor specificity reported by Huang et al. and Lerchner et al. 30,51 . Further investigation would be needed to derive an NADH-preferring variant by targeting loop β2-α2 to increase whole-cell biocatalysis efficiency. In support of this notion, a complete switch of the cofactor specificity (from NADP + to NAD + ) has been reported by engineering this loop in Ralstonia sp. alcohol dehydrogenase, realizing a complete change in NADH/NADPH specificity (a factor of ~3.6 million) 30 . This also sheds lights into the future engineering of SmSDR.
In summary, the apo-form and NADPH-liganded SmSDR structures determined in this investigation reveal an unexpectedly mobile region near the entrance of the substrate-binding region. Together with the simulated HPMAE complex model, the catalytic mechanism of (R)-PE conversion is proposed. Rationally designed mutants of the mobile loops significantly improved the synthesis of (R)-PE. These results together suggest that stabilization of this flexible region (i.e. α7 helix and loop β4-α4) provides a robust means to engineer SmSDR with higher transformation activity and a valuable variant F98YF202L.

Materials and Methods
Cloning, expression and purification of SmSDRs. The S. marcescens sdr gene encoding SmSDR protein was amplified by PCR from S. marcescens BCRC 10948 genomic DNA (GenBank: CP003959.1) using gene specific primers (Table S1). Chromosomal S. marcescens DNA was isolated as described previously 21 and used as the template. The amplified product was inserted into the pET30a vector (Novagen, Inc., USA) to generate pET30a·smsdr plasmid, followed by transformation using E. coli BL21 (DE3) strain. The cells were cultivated in Luria-Bertani medium at 37 °C for 16 h until the OD 600 reached 0.8. The overexpression SmSDR protein was induced at 28 °C for 16 h by the addition of isopropyl-β-D-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM. The cells were collected by centrifugation at 6,000 × g at 4 °C for 15 min and resuspended in 30 ml of 50 mM sodium phosphate buffer (pH 7.0) containing 300 mM NaCl and followed by sonication. The cell debris was pelleted at 10,000 × g at 4 °C for 20 minutes and the resulting supernatant was loaded into a nickel affinity column (Novagen, Madison, WI, USA). The His6-tagged proteins were eluted using the buffer containing 200 mM imidazole. The purified protein was verified by SDS-PAGE analysis. Protein concentration was assayed by the Bradford method using bovine serum albumin as the standard 52 .
Site-directed Mutagenesis. Mutated SDR genes were amplified by overlap extension PCR amplification 53 using the site-directed mutagenesis primers listed in Table S1. The pET30a-Smsdr plasmid was used as the template. Each amplified gene was cloned into the KpnI/SalI site in pET30a vector and the sdr-inserted plasmid carrying requisite point mutation was individually introduced into E. coli BL21 (DE3). The mutated plasmid was confirmed by DNA sequencing. Diffractions were collected at the BL15A1 using Rayonix MX300HE CCD area detector, National Synchrotron Radiation Research Center (NSRRC), Taiwan. All datasets were collected at 123 K and FOMBLIN was used as the cyroprotectant. Collected images were indexed and integrated using the HKL 2000 software 54 . SmSDR crystals were characterized as belonging to the space P4 1 2 1 2, with the unit cell of a = b = 83.87, and c = 115.05 Å; α = β = γ = 90°. The SmSDR structures were solved by the molecular replacement methods by AMoRe module of CCP4 software 55,56 using the Sinorhizobium meliloti 1021 3-oxoacyl-[acyl-carrier-protein] reductase structure (PDB code: 3V2G) as the search template. Rotation and translation functions followed by the rigid body refinement procedure were carried out using data from 8-to 3-Å. Crystallographic refinement was carried out using the REFMAC5 module 57 and coupled to ARP/wARP 58 to add solvent molecules. 2Fo-Fc and Fo-Fc maps were produced and inspected after refinement and revised the model manually with the program 'Coot' 59 . The overall quality of the final model was assessed by the program PROCHECK 60 . Table 2 shows the crystallographic statistics.

Sequence and Structural Comparisons.
SmSDR structures were compared with all protein structures in the DALI server (http://ekhidna.biocenter.helsinki.fi/dali_server/start). The secondary protein elements alignment was performed by ESPript tool version 3.0 61 to generate the figures of the secondary elements alignment profiles. Structural comparison between the SmSDR and its homologous structures was carried out using the program Lsqkab module 62 in CCP4 program to superimpose Cα atoms. Structural figures were prepared with the program PyMOL (www.pymol.org).
Virtual Docking. The model containing NADPH and HPMAE was prepared using the SmSDR structure as the template. The NADPH binding residues within the 3.8 Å of the NADPH of the F98L·F202L·NADPH·SmSDR were chosen as the binding pocket for docking NADPH using the GOLD module of Discovery Studio 2016 63,64 . Parameter settings were used as same as the described previously 33 . Docking procedures were terminated while three best solutions were obtained within a root mean square (r.m.s.) tolerance of 1.5 Å. Top ranking of the docked pose of the NADPH were chosen for superposing the orientation of the liganded NADPH. The NADPH  pose with the lowest r.m.s of the superposition was selected as the second template for the HPMAE docking. The docked method of HPMAE was performed as well as that of NADPH with modification. The binding site was defined within 5 Å radius around the hydroxyl O atom the Tyr158 residue. Crucial residues for contacting the HPMAE are added to the criteria of the protein H-bond constraints. Top ranking of the docked pose of the HPMAE with the lowest energies was chosen for analysis.

Enzyme activity assay. 1-(3-hydroxylphenyl)-2-(methylamino) ethanone (HPMAE) was gifted from
Industrial technology Research Institute, Hsinchu, Taiwan. The SmSDR enzyme activity was measured by determining the consumption of NADPH using a 0.2 ml reaction mixture containing 12 mM HPMAE, 5 µM proteins with 50 mM sodium phosphate (pH 7.0), 300 mM NaCl. Each reaction was initialed by adding 0.4 mM NADPH, and subsequently monitored the extinction of NADPH fluorescence (λ ex 340 nm, λ em 460 nm) by CLARIO star ELISA reader. The steady-state kinetics analysis of SmSDR was performed at 30 °C with the reaction mixtures as described above. The K m and k cat values were determined by plotting the initial velocity as a function of substrate concentration and fitting the plots to the Michaelis-Menten equation. E. coli BL21 (DE3). Conversion of HPMAE to (R)-PE was calculated according to the previous method 21 . The cell pellet was washed with 10% glycerol and resuspended in 10 ml of reaction mixture containing 5% (wet weight) cells, 10 mM HPMAE, 100 mM sodium phosphate buffer (pH 7.0) and 2% fructose. The bioconversion reaction of the whole cell was performed in a shaking flask at 30 °C to convert HPMAE to (R)-PE in one hours. The cells were then removed by centrifugation, and the supernatant was subjected to HPLC for the analysis of substrate consumption and product formation.

Conversion of HPMAE to (R)-PE by
Thermal shift assay. Thermal shift assays were performed according to Huynh and Partch 36 . The SmSDR enzyme (5 μM) was incubated with 1000× dilution of SYPRO Orange in 50 mM sodium phosphate (pH 7.0) solution containing 300 mM NaCl. The mixture was transferred to a 96-well plate and spectra were collected at 0.3 °C intervals per minute from 25 °C to 95 °C on the Step One Plus Real-Time PCR System (Applied Biosystems ™ ).
The acquisition results were analyzed using GraphPad Prism. A melting temperature (Tm) is calculated from the denaturing curve.
Ethical approval. This article does not contain any studies with human participants or animals performed by any of the authors.