Inverting the Stereoselectivity of an NADH‐Dependent Imine‐Reductase Variant

Abstract Imine reductases (IREDs) offer biocatalytic routes to chiral amines and have a natural preference for the NADPH cofactor. In previous work, we reported enzyme engineering of the (R)‐selective IRED from Myxococcus stipitatus (NADH‐IRED‐Ms) yielding a NADH‐dependent variant with high catalytic efficiency. However, no IRED with NADH specificity and (S)‐selectivity in asymmetric reductions has yet been reported. Herein, we applied semi‐rational enzyme engineering to switch the selectivity of NADH‐IRED‐Ms. The quintuple variant A241V/H242Y/N243D/V244Y/A245L showed reverse stereopreference in the reduction of the cyclic imine 2‐methylpyrroline compared to the wild‐type and afforded the (S)‐amine product with >99 % conversion and 91 % enantiomeric excess. We also report the crystal‐structures of the NADPH‐dependent (R)‐IRED‐Ms wild‐type enzyme and the NADH‐dependent NADH‐IRED‐Ms variant and molecular dynamics (MD) simulations to rationalize the inverted stereoselectivity of the quintuple variant.

The sequence of (R)-IRED_Ms is printed in bold. For orientation some amino acid positions are given (related to (R)-IRED_Ms). A special feature is the "D-type" (S)-selective IRED from Streptomyces rimosus. In contrast to most (S)-selective IREDs, it does not contain tyrosine (Y) but aspartic acid (D) at the catalytic position (marked yellow).
In (R)-IREDs, the equivalent position of residue V212 NADH-IRED-Ms (standard position 228) often displays a tryptophan, but in (S)-IREDs mostly small amino acids are present at this position. Interestingly, an aspartic acid at the equivalent position of Q234 (standard position 250) is conserved in (S)-IREDs. In contrast, no conservation was found in (R)-IREDs.
Additionally, position 242 (standard position 257.1) was targeted, as it displays methionines or glycines in (S)-IREDs and histidines or glutamines in (R)-IREDs. Initially, several mutations were introduced and tested at the positions mentioned above. In addition, a saturation or combined partial saturation mutagenesis was carried out at positions S96 and D171 (Table S1).
The reduction of 2-methylpyrroline to (R)-or (S)-2-methylpyrrolidine was selected as the model reaction. Generated enzyme variants and mutant libraries were analysed by plate screening (Section S2). However, no altered stereoselectivity was detected. Table S1. Library design of the first mutagenesis round inspired by previously published differences in conservation described for (R)-and (S)-IREDs.
In the next round, additional libraries were generated in respect to the structural information provided by the crystal structure (Section S3). All amino acid positions in proximity to the C4 atom of NADH (position of the transferred hydride) were considered ( Figure S2). Neighbouring positions were combined into pairs and exchanged simultaneously in a combined partial saturation mutagenesis thus partly considering combinatorial effects. A saturation mutagenesis was performed at the remaining positions (Table S2). Figure S2. Residues considered for mutagenesis in the design of the second (orange) and the third (green) library. Table S2. Library design of the second mutagenesis round focussing on residues derived using the crystal structure to identify residues in proximity to NADH C4 atom.
Interestingly, variant A241V/H242Y in library Lib2.9 resulted in a slightly decreased enantiomeric excess (ee from > 99% (R) to 95% (R)). Similarly did mutation A245C in library Lib2.10 (ee from > 99% (R) to 97% (R)). However, variant A245L in library Lib2.10 was most promising, which displayed a significantly lower conversion but generated an almost racemic product mixture (ee ≈ 14% (R)). Based on these results, positions 241, 242 and 245, as well as the surrounding amino acids were modified in another round of mutagenesis (Table S3). Table S3. Library design of the third mutagenesis round focussing on mutations in the helical region identified within the second mutagenesis round.
As described previously, 3 substrate binding site variants were generated from pBAD33_NADH-IRED-Ms plasmid according to chosen screening libraries and mutagenesis methods (Section S1).

Cloning, protein expression, and purification
The variant NADH-IRED-Ms was cloned into the pBAD33 plasmid using a Gibson Assembly and fused with an N-terminal his6-tag. For the expression, we used E. coli JW5510 cells harboring the vector and let them grow as preculture overnight at 37 °C. Next, we inoculated   All primers (Table S2) for the introduction of the mutations targeting the switch in stereoselectivity were obtained from Metabion International AG (Planegg, DE).  Section S2. Screening System.
The screening system for the identification of IRED variants with altered stereoselectivity in the 96-well format was performed using gas chromatography (GC) and a chiral column. One reaction plate was prepared during each centrifugation. After centrifugation, 150 μl of the supernatant was added to the reaction plate, which had previously been mixed with 50 μl reaction mix.

Biotransformation and sample preparation
The

Data Collection, Structure Solution and Refinement
The datasets described in this report were collected at the Diamond Light Source, Didcot, Oxfordshire, U.K. on beamlines I03 ((R)-IRED-Ms-NADP + ) and I04-1 (NADH-IRED-Ms NAD + ). Data were processed and integrated using XDS 8 and scaled using SCALA 9 included in the Xia2 processing system. 10 Data collection statistics are provided in Table S6. Crystals of wt-MsIRED-NADP + were obtained in space group P21, with four molecules in the asymmetric unit; crystals of the NADH-IRED-Ms NAD + complex were in space group P65. The solvent content in the crystals was 49% and 60% respectively. The structure of the wt-MsIRED-NADP + was solved by molecular replacement using MOLREP 11,12 with the monomer of AtRedAm (PDB code 6EOD 13 ) as the model. The structure was built and refined using iterative cycles in  Section S4. Molecular Modelling.

Homology Modelling
To provide a structure including the unresolved residues, a homology model was generated. For this purpose, the Modeller 16,17 PlugIn PyMod2.0 18 was used with PyMOL 1.8. 19 For this purpose, all chains from 6TOE (NADH-IRED-Ms NAD + complex) were separated and superposed. NAD + was extracted from chain A of 6TOE and was adapted via PyMOL builder to provide a functional NADH cofactor. The atom names were adapted manually.

Generation of Substrate Complex and Parametrization
2-Methyl-1-pyrroline was used as model substrate. The structure was gained from PubChem compound database 20 and it was manually docked next to the NADPH in three different starting conformations to reduce the initial bias. Parameters for 2-methylpyrroline and NADH were calculated with antechamber. 21 Using the parameters listed in MOL2 and FRCMOD files, parameter XML files were generated. The according coordinate and parameter files are provided on the data repository of the University of Stuttgart.

Molecular Dynamics Simulation
The pKa values of the models' side chains were calculated using PROPKA 22 27 General Amber force field (GAFF) and Amber14 force field were used. 29,29 The cubic box with a padding of 1.5 nm was solvated with water (tip4p-Ew water model), 30 the protein charge was neutralized, and an ionic strength of 0.1 M NaCl was applied, and the total charge was neutralized. Energy minimization was performed, until 10 kJ/mole tolerance energy. A reference temperature of 300 K and a pH of 7 were utilized and the Langevin integrator was used with a friction coefficient of 1/ps and a step size of 2 fs. 31 Additionally, the Particle Mesh Ewald method was used to compute long range Coulomb interactions with a 1 nm nonbonded cut-off for the direct space interactions. To equilibrate the solvent, a 5 ns pressure coupled equilibration with Monte Carlo barostat was performed at a pressure of 1 atm assuming the reference temperature of 300 K (temperature coupling provided by Langevin integrator). Thereby, the protein backbone and each ligand were restrained, with a force of 100 and 150 kJ/mole*Å2, respectively. To enable conformational rearrangements toward a proper substrate-binding conformation, a similar equilibration was performed for 50 ns, whereby the protein restraints were removed but the ligand restraints were kept.
Successively, substrate and cofactor restrained were removed, and the equilibration was continued for 0.5 ns and 5 ns, respectively. With the resulting system, a production of 10 replicates à 50 ns was performed under periodic boundary conditions. The trajectories were written every 1000 steps.

Analysis
The analysis was mainly performed utilizing MDTraj. 32 Thereby, the 10 replicates per system were considered separately. To identify residues interacting with the imine nitrogen, residues within a sphere with a radius of 10 Å was defined around the hydride donor hydrogen of the NADH in the respective crystal structure were considered. The contact frequency (< 2.5 Å) between hydrogens of the selected binding site residues (Table S7) and the imine nitrogen were computed with a modulated code of the MDTraj-based Contact Map Explorer (https://contactmap.readthedocs.io/). To estimate the stereoselectivity, only approximated productive substrate conformations with (distance of < 3 Å between imine nitrogen + the potentially proton donating hydrogens; < 4.5 Å between imine carbon + the closer hydride of NADH cofactor) were considered. These frames were superposed on energy minimized structure to provide comparable coordinates and the stereopreference was derived by computing the dihedral angle α formed by two planes ( Figure S5). The first plane is spanned by imine nitrogen, C2 and C5 carbon of 2-methylpyrroline. The second plane is spanned by C2 and C5 carbon of 2methylpyrroline and the closer hydride donating hydrogen of NADH. In the case of α > 0.1 rad, the conformation was classified as (S)-selective and in the case of α < -0.1 rad, as (R)-selective.
Frames in between this range were not considered for the evaluation of the stereoselectivity.   Figure S5. Scheme of the planes spanning defining dihedral angle α which serves as criterium to decide over the frames' stereopreference.

Detailed description of simulation results
While the simulation of systems NADH-IRED-Ms_1 and NADH-IRED-Ms_3 solely displayed contacts with S95 and W179, some frames of system NADH-IRED-Ms_2 did also display close contacts to the conventional proton donor D171. In the systems of (S)-NADH_V11, no close contacts were observed with S95. Indeed, the contact frequency with D171 and W179 was remarkably higher. Thereby, the simulations of system (S)-NADH_V11_1 displayed contact with both, D171 and W179, while system (S)-NADH_V11_2 and (S)-NADH_V11_3 solely showed contacts with D171 and W179, respectively.
The ratio of theoretical stereopreference was derived by trends in simulation states with close 2-methylpyrroline substrate distances towards NADH and S95, D171 and W179 side chains.
Assuming D171 as being actively or at least passively involved in the protonation of the imine moiety, ternary complexes of NADH-IRED-Ms and (S)-NADH_V11-NADH with 2methylpyrroline were derived from the states with the closest distances to the catalytic atoms.
As the detection of a 'real' near-attack complexes in MD simulations often affords the application of complex and time-consuming enhanced sampling methods, [33][34][35][36][37] this focused approach settles with approximated productive binding modes. Figure S6. The five additional mutants in (S)-NADH_V11 (green) lead to a shifted loop-helix element which seem to alter the nicotinamide orientation of the cofactor compared to NADH-IRED-Ms (orange).