Loop 6 and the β‐hairpin flap are structural hotspots that determine cofactor specificity in the FMN‐dependent family of ene‐reductases

Flavin mononucleotide (FMN)‐dependent ene‐reductases constitute a large family of oxidoreductases that catalyze the enantiospecific reduction of carbon–carbon double bonds. The reducing equivalents required for substrate reduction are obtained from reduced nicotinamide by hydride transfer. Most ene‐reductases significantly prefer, or exclusively accept, either NADPH or NADH. Despite their usefulness in biocatalytic applications, the structural determinants for cofactor preference remain elusive. We employed the NADPH‐preferring 12‐oxophytodienoic acid reductase 3 from Solanum lycopersicum (SlOPR3) as a model enzyme of the ene‐reductase family and applied computational and structural methods to investigate the binding specificity of the reducing coenzymes. Initial docking results indicated that the arginine triad R283, R343, and R366 residing on and close to a critical loop at the active site (loop 6) are the main contributors to NADPH binding. In contrast, NADH binds unfavorably in the opposite direction toward the β‐hairpin flap within a largely hydrophobic region. Notably, the crystal structures of SlOPR3 in complex with either NADPH4 or NADH4 corroborated these different binding modes. Molecular dynamics simulations confirmed NADH binding near the β‐hairpin flap and provided structural explanations for the low binding affinity of NADH to SlOPR3. We postulate that cofactor specificity is determined by the arginine triad/loop 6 and the residue(s) controlling access to a hydrophobic cleft formed by the β‐hairpin flap. Thus, NADPH preference depends on a properly positioned arginine triad, whereas granting access to the hydrophobic cleft at the β‐hairpin flap favors NADH binding.


Introduction
Ene-reductases (ERs) represent a large group of oxidoreductases that catalyze the stereospecific reduction of carbon-carbon double bonds adjacent to an electron-withdrawing group.Among ERs, flavin mononucleotide (FMN)-dependent enzymes of the Old Yellow Enzyme (OYE) family (EC1.6.99.1) have been studied extensively since the isolation and characterization of the first member from Saccharomyces pastorianus (OYE1) by Warburg and Christian in 1932 [1,2].The last 90 years have witnessed the discovery of many more homologs in bacteria, fungi, and plants.Today, the initial classification of OYEs, which included two main groups based on their structural and functional characteristics-the classical OYEs (class I) and thermophilic OYEs (class II) [3]-was extended by class III, IV, and V [4][5][6].Although the sequence similarity across the five classes of OYEs is relatively low (< 15% [7]), they share the same typical (βα) 8 -barrel (TIM barrel) topology, which is characterized by eight alternating β-strands and α-helices where the latter generally form the outer part of the barrel while the βstrands constitute the inner part.The redox-active flavin cofactor is non-covalently bound at the C-terminal ends of the β-strands [8][9][10], where it forms the active site together with two histidine residues (or a histidine-asparagine pair) and a tyrosine residue.Based on structural and biochemical studies, the mechanism of substrate reduction could be elucidated in detail and, consequently, was utilized in several seminal contributions to exploit members of this family in biocatalytic applications [6].
The reduction of the FMN cofactor occurs by hydride transfer at the expense of a reduced nicotinamide cofactor, with most enzymes exhibiting a clear preference for either NADH or NADPH [3,[11][12][13][14][15][16][17].In contrast to the extensive research efforts toward biocatalytic applications, for example, in cascade reactions [18,19], whole-cell biotransformations [19][20][21], and coenzyme recycling systems [22][23][24][25], very little research has aimed at a thorough understanding of the coenzyme preference in flavin-dependent ERs.Since NADH is more stable and significantly cheaper (about 70× [26]) than NADPH, a deeper understanding of the binding mode of the nicotinamide coenzymes in ERs is not only of pivotal importance in the field of fundamental research but also in the industrial environment.Crystallographic studies involving a non-reactive analog of NAD(P)H (i.e., NAD(P)H 4 ) have demonstrated that precise alignment of the nicotinamide segment of either NADPH or NADH above the isoalloxazine ring of FMN is essential for the hydride transfer to the N (5)-position [27][28][29].On the other hand, the orientation of the AMP moiety of the nicotinamide coenzymes appears to exhibit considerable variability and constitutes the molecular rationale for coenzyme selectivity due to the absence (NADH) or presence (NADPH) of an additional phosphate group at the 2 0 -position of the adenosine moiety.Structural comparison led to the identification of several structural elements that were postulated to interact with the coenzyme and/or substrate.In this context, the highly adaptable loop 3 (connecting β-strand 3 and α-helix 3), which is also referred to as the β-hairpin flap, has been targeted in several studies and was identified as a crucial element to affect the preference for either NADH or NADPH [3,7,16,30,31].This loop exhibits significant diversity in length, sequence, and structure, spanning from short α-helices to β-hairpins and even highly unstructured turns.Additionally, it has been observed that these structural elements frequently appear in parallel or in a complementary fashion with the conformation of loop 6 (connecting β-strand 6 and α-helix 6).Consequently, loop 6 also considerably varies in length, sequence, and structure, ranging from as short as 7 amino acids (NemA, PDB: 3GKA) to up to 18 amino acids (OYE2.6,PDB: 3TJL) [32].Despite this trend, the relevance of loop 6 for nicotinamide cofactor specificity has surprisingly not been addressed thus far.
In order to obtain more information on the role and cooperation of the β-hairpin flap and loop 6, we investigated 12-oxophytodienoic acid reductase 3 from Solanum lycopersicum (SlOPR3), which catalyzes the conversion of 12-oxophytodienate to the corresponding cyclopentanone (Fig. 1), as a model enzyme for the ER family.Based on a combination of computational (i.e., molecular docking of NAD(P)H into the active site of SlOPR3 and molecular dynamics simulations) and structural methods (i.e., elucidation of the crystal structures of SlOPR3 in complex with either NADPH 4 or NADH 4 ), we propose that the β-hairpin flap and Fig. 1.Reaction catalyzed by SlOPR3.12-oxophytodienoate is reduced to the corresponding cyclopentanone at the expense of NADPH.The figure was prepared using ChemDraw.loop 6 play a complementary role by providing anchor sites for the tail of the nicotinamide cofactors.

Initial kinetic characterization of SlOPR3
To assess the specificity of SlOPR3 for NADH and NADPH, we determined the rate of reduction as a function of the concentration of the reducing cofactor.In both cases, the data could be fit with a hyperbolic equation (Fig. 2), yielding k red values of 0.99 AE 0.01 and 17.0 AE 0.4 s À1 for NADH and NADPH, respectively.The dissociation constants for NADH and NADPH were 1400 AE 60 and 28 AE 4 μM, respectively.Thus, the catalytic efficiency with NADPH (k red /K D ) is 850 times higher than that with NADH, clearly demonstrating a substantial preference for NADPH over NADH in the reductive half-reaction of SlOPR3.

Molecular docking
Molecular docking was performed to predict potential binding modes for NAD(P)H to SlOPR3 and identify key structural elements responsible for the observed coenzyme specificity in SlOPR3.Positional restraints were applied to the nicotinamide moiety of the coenzyme to fix its position above the isoalloxazine ring of FMN, which is a prerequisite for the hydride transfer reaction.Therefore, the binding mode of the nicotinamide moiety was the same in all calculated binding poses, reflecting that observed in the crystal structures of other nicotinamide nucleotide-binding flavoenzymes such as morphinone reductase (MR) (Fig. 3) [31].The 10 top-ranked poses of NADPH were almost identical, indicating a unique binding mode and the formation of a stable complex.In this binding mode, the 2 0 -phosphate-AMP moiety of NADPH is oriented toward loop 6 and located at a highly positively charged region, where it mainly interacts with the three arginine residues R283, R343, and R366 via strong ionic interactions (Fig. 3, left panel).Furthermore, the adenine ring of NADPH is stacked between R283, which is located on loop 6, and R343 and thereby stabilized via cation-π-cation interactions.Thus, the docking results suggest that loop 6 and the arginine triad R283, R343, and R366 are the main contributors to NADPH binding in SlOPR3.
In contrast to the calculated NADPH poses, the 10 top-ranked NADH poses differ substantially from each other, indicating a less specific binding behavior than in the case of NADPH (Fig. 3, middle panel).Interestingly, in all NADH poses, the AMP moiety is oriented toward the β-hairpin flap at a predominantly hydrophobic region.There, the AMP tail of NADH exhibits multiple conformations in which its AMP moiety is either highly solvent-exposed or located within a hydrophobic cavity, where it is engaged in mostly hydrophobic interactions but, in some poses, interacts with Y149 and Y190 via hydrogen bonding (Fig. 3, right panel).Despite pointing away from the loop-6 site and the arginine triad, NADH interacts electrostatically with R283 and R366 via its pyrophosphate backbone.Nevertheless, the AMP moiety of NADH favors the rather hydrophobic region at the βhairpin flap, resembling the NADH binding mode found in the crystal structure of MR in complex with NADH 4 (PDB: 2R14) [31].In MR, the AMP moiety of NADH is accommodated within a narrow hydrophobic cleft formed by the β-hairpin flap and the Cterminal loop, where it is stabilized by π-stacking interactions with a phenylalanine (F137).SlOPR3 possesses a similar cleft, but instead of phenylalanine, it has a leucine (L140) at the corresponding position that sterically blocks the entrance to the cleft.This may explain the unspecific binding poses of NADH, as the AMP moiety fails to establish a stable conformation deeper inside the hydrophobic cavity.
The docking results suggest that NADPH and NADH exhibit different binding modes in which the 2 0phosphate-AMP tail of NADPH binds electrostatically at a positively charged region located at loop 6, whereas the AMP tail of NADH binds primarily via van der Waals and H-bonding interactions at a hydrophobic region situated at the β-hairpin flap.Furthermore, NADH docking showed that the structure and high solvent accessibility of the active site of SlOPR3 can be unfavorable for binding elongated ligands, as a large portion will inevitably protrude into the bulk solvent and, thus, destabilize the protein-cofactor complex.

Crystal structure of SlOPR3-FMN-NADPH 4
To validate our docking results, we crystallized SlOPR3 in complex with NADPH 4 .The crystal structure of SlOPR3 in complex with NADPH 4 (PDB: 8QMX) was determined at 1.4 Å resolution using molecular replacement.The complex crystallized in the monoclinic space group P 1 2 1 1 with two molecules (chains A and B) per asymmetric unit.Both chains are essentially identical (RMSD Cα = 0.22 Å) and have NADPH 4 bound to their active sites.As predicted by our molecular docking experiment, NADPH 4 binds with its 2 0 -phosphate-AMP moiety near loop 6 in both chains.However, the structure and conformation of loop 6 differ substantially between the chains, affecting NADPH 4 binding (Fig. 4).
The binding mode of the nicotinamide ring of NADPH 4 is the same in both chains and in other nucleotide-binding flavoenzymes such as MR.The nicotinamide ring is located directly above the isoalloxazine ring of FMN so that the distance between its C(4) atom, which carries the hydride to be transferred, and the N(5) atom of FMN is 3.6 and 3.5 Å, respectively (Fig. 4A,D).It is stabilized by hydrogen bonding to T33, H185, and H188.However, the chains differ in the binding of the NADPH 4 tail.
In chain A, the pyrophosphate backbone of NADPH 4 interacts electrostatically with R366, and the 2 0 -phosphate group is engaged in strong ionic interactions with R366 and R343 and a hydrogen bond with Y364 (Fig. 4A,B).The adenine ring, which is not fully resolved in the electron density map, is located next to loop 6 and highly solvent-exposed but interacts with R343 via weak cation-π interactions.Loop 6 is highly disordered in chain A, as evidenced by the lack of electron density for 10 amino acid residues (A286-L295); however, residue R283, which according to molecular docking is a key interacting residue, is still resolved but points away from the active site, not interacting with the coenzyme (Fig. 4A).There is a second NADPH 4 molecule (occupancy = 0.71) in chain A located above Y364 and close to the first one, which most likely is a crystallographic artifact attributed to the high NADPH 4 concentration used during crystallization (Fig. 4A).Nevertheless, the position of this second NADPH 4 molecule confirms that NADPH enters the active site from the loop 6 side.
In contrast to that in chain A, loop 6 in chain B is almost entirely resolved (only three residues are missing; G286-E298) and forms a hairpin-like structure in which R283 points toward the active site, interacting with the coenzyme (Fig. 4C).This loop 6 conformation also differs substantially from that observed in the previously solved SlOPR3 wildtype structure (PDB: 2HSA) in which loop 6 adopts a stretched conformation and protrudes into the active site of another SlOPR3 molecule to form a self-inhibitory dimer [27].In chain B, the pyrophosphate group of NADPH 4 is slightly shifted toward Y370 compared with its position in chain A and interacts with R283, R366, and Y370 (Fig. 4D,E).The 2 0 -phosphate group is located at the same position as in chain A and thus interacts also with the same set of residues: R343, Y364, and R366 (Fig. 4C).However, the adenine ring of NADPH 4 is rotated by approximately 66°compared with that in chain A so that it is stacked between R283 and R343, exhibiting cation-π-cation interactions (Fig. 4C).Due to these additional R283-mediated interactions, the 2 0 -phosphate-AMP tail of NADPH 4 is substantially better stabilized in chain B than in chain A. This observation is also supported by the electron density map in which markedly more and less diffuse electron density was observed for the coenzyme tail in chain B than in chain A.
Considering the conformations of loop 6 and the different positions of R283 in both chains, SlOPR3 seems to adopt an open (chain A) and closed (chain B) conformation.Thus, we postulate that in the open conformation, loop 6 and R283 move away from the active site to facilitate the access of the coenzyme to the active site, whereas, in the closed conformation, loop 6 and R283 move toward the active site to stabilize the tail of the coenzyme during hydride transfer to FMN.This is in accordance with the findings of Horita et al., who also observed a closed and open conformation for loop 6 in OYE from Candida macedoniensis depending on whether phydroxybenzaldehyde, which forms a charge-transfer complex with the FMN, was present or not [33].
Crystal structure of SlOPR3 in complex with NADH 4 and comparison to the complex with NADPH 4 The crystal structure of SlOPR3 in complex with NADH 4 (PDB: 8QN3) was determined at 1.75 Å resolution using molecular replacement.The complex also crystallized in the monoclinic space group P 1 2 1 1 with two molecules (chains A and B) per asymmetric unit.Both chains are identical (RMSD Cα = 0.23 Å), but chain B is highly disordered, as evidenced by its poor electron density and high average B-factor (chain A = 34.7 Å2 vs. chain B = 68.9Å2 ).Therefore, the following analysis is solely based on chain A.
NADH 4 is bound to the active site, but the coenzyme was only partially resolved, as we observed electron density only for the nicotinamide-pyrophosphate moiety (Fig. 5A,B).Like in the complex with NADPH 4 , the nicotinamide moiety of NADH 4 forms hydrogen bonds to T33, H185, and H188; however, its position is slightly shifted compared with that of NADPH 4 , leading to an increase of the C(4) coenzyme -N(5) FMN distance from 3.5-3.6Å (NADPH 4 ) to 4.0-4.1 Å (NADH 4 ).The larger distance between donor and acceptor may contribute to NADH's 17-fold slower rate of reduction compared with that of NADPH observed in our presteady-state analysis (see Fig. 2).The pyrophosphate group of NADH 4 interacts with R366 and R283 via ion-ion interactions and with H244 via H-bonding (Fig. 5A,B).As observed in the closed conformation of SlOPR3 in complex with NADPH 4 , R283 points to the enzyme's active site and thus interacts with the coenzyme (Fig. 5C).However, the superposition of the structure of the closed conformation of the complex with NADPH 4 and NADH 4 shows that the tail of NADPH 4 and the pyrophosphate group of NADH 4 point in opposite directions (Fig. 5C,D).The β-phosphate of the pyrophosphate group of NADH 4 is oriented toward Y370 and thus points toward the βhairpin flap, suggesting that the tail of NADH 4 is located at the β-hairpin flap region, adopting multiple conformations (Fig. 5C,D).Furthermore, the superposition revealed that a sulfate ion in the complex structure with NADH 4 is bound at the same position as the 2 0 -phosphate group of NADPH 4 in the SlOPR3-NADPH 4 structure, confirming that this site, including the interacting residues R343, R366, and Y364, represents a high-affinity binding site for phosphate in SlOPR3.
As in the case of the NADPH 4 -bound SlOPR3 structures, loop 6 is not fully resolved in the complex structure with NADH 4 , which misses 10 residues (E291-E301) due to the lack of electron density.However, the residues of loop 6 (Q281-T290) that form the hairpinlike structure in the closed conformation of the SlOPR3-NADPH 4 structure are resolved and fold into a small helical segment.It is currently unclear if and how exactly these different loop 6 conformations (i.e., hairpin-like, stretched, and helical conformations), apart from providing and withdrawing R283 to and from the active site, affect coenzyme binding in SlOPR3.
Our crystal structures confirm our initial dockingbased hypothesis that NADPH and NADH bind at different sites within the active site.The present results extend those of Barna et al. [31] and Iorgu et al. [7], who have found that the β-hairpin flap governs the selectivity and binding affinity of NAD(P)H in MR and pentaerythritol tetranitrate reductase (PETNR), which are both members of the OYE family.PETNR contains two arginine residues, R130 and R142, on its β-hairpin flap that facilitate NADPH binding, while MR contains a glutamic acid, E135, instead, which promotes NADH binding [7,31,34].Here, we have shown that in SlOPR3, loop 6 and the arginine triad are responsible for NADPH binding, while the β-hairpin flap seems responsible only for NADH binding.Although SlOPR3 also bears an arginine residue (R138) on its β-hairpin flap at the corresponding position of R130 in PETNR, the β-hairpin flap is not involved in NADPH binding.Moreover, Shewanella yellow enzyme 1 also carries a positively charged residue at this position (K130) but prefers NADH over NADPH as a reducing coenzyme [35].Thus, the presence of positively charged residues on the β-hairpin flap does not necessarily confer NADPH preference to ERs.
The low affinity of NADH to SlOPR3 may be attributed to the inability of the β-hairpin flap region to stably accommodate the adenosine tail of NADH like in MR [31].In MR, the adenosine tail is located within a narrow hydrophobic cleft in which it is stabilized via strong π-π stacking interactions between its adenine ring and F137.The corresponding hydrophobic cleft in SlOPR3 is differently shaped due to differences (i.e., length and structure) in the β-hairpin flap between SlOPR3 and MR and is sterically blocked by L140, which is located at the corresponding position of F137 in MR.Therefore, the adenosine tail of NADH fails to firmly bind to SlOPR3 and instead exhibits multiple unspecific binding conformations, as evidenced by the multiple and substantially different docking poses for NADH and the lack of electron density for its adenosine tail in our diffraction data.
In summary, two structural hotspots were identified that influence coenzyme binding in OYEs: loop 6 (and the arginine triad) and the β-hairpin flap.According to crystallographic data, these structural elements differ significantly in structure, length, and flexibility among OYEs [3,7,16,30,31].Their size and flexibility determine the size and ligand accessibility of the active site, suggesting that NAD(P)H binding follows an inducedfit mechanism in which either loop 6 or the β-hairpin flap undergo conformational changes to accommodate NADPH or NADH.Thus, we propose that the architecture of these structural elements is the key behind the NADPH-NADH discrimination in OYEs.

MD simulations of SlOPR3 in complex with NADH 4
We performed a 250-ns MD simulation for SlOPR3 in complex with NADH to further analyze the binding behavior of NADH, as the coenzyme was not fully resolved in our crystal structure.Toward this aim, we generated a complex of SlOPR3 with NADH in which the coenzyme's AMP tail was positioned between loop 6 and the β-hairpin flap.The RMSD plot for the backbone atoms shows that after approximately 50 ns, the system reaches equilibrium and remains stable throughout the simulation (Fig. 6A).The position of the nicotinamide group of NADH also remains stable, with the C(4) atom of NADH and N(5) atom of FMN staying close together (< 4 Å), which is a prerequisite to hydride transfer.During the simulation, NADH is mainly stabilized by electrostatic interactions between its pyrophosphate group and R283 and R366 (as observed in our crystal structure), while its adenosine tail moves toward the enzyme's β-hairpin flap, π-stacking with Y364 for most of the simulation time (Fig. 6B).In this conformation, the adenine group is highly solvent-exposed and surrounded by the side chains of Y364, R366, F369, and Y370.These bulky residues, especially Y364 and Y370, prevent the adenosine tail from approaching the βhairpin flap even closer, which is why it stays locked in this conformation during the simulation.
Therefore, we performed a second 250-ns MD simulation at a higher temperature (310 K) to escape from the above conformation and sample a larger conformational space.Despite the higher temperature, the system reaches equilibrium after approximately 50 ns and remains stable for most of the simulation time (Fig. 6A).Only at the end of the simulation (last 30 ns), the RMSD shows a jump due to loop 6 movements, which do not affect NADH binding.Again, R283 and R366 are the main binding contributors by electrostatically interacting with the coenzyme's pyrophosphate group (Fig. 6C).However, in contrast to the first MD simulation at 298 K, the adenosine tail moves much further toward the β-hairpin flap, trying to adapt a conformation similar to that found in the crystal structure of MR in complex with NADH 4 (PDB: 2R14), where the adenosine moiety of NADH 4 is buried within a narrow cleft formed by the β-hairpin flap and the C-terminal loop of MR (Fig. 6D).As mentioned before, in this cleft, the adenine ring is stabilized by π-stacking to F137; however, SlOPR3 has a leucine (L140) instead at this position, which sterically blocks the corresponding cleft in SlOPR3 (Fig. 6D, left insert).Thus, the adenine ring fails to enter this cleft in SlOPR3 and moves around the β-hairpin flap most of the simulation time, trying to enter the cleft.These tail conformations are stabilized mainly by hydrophobic interactions between the adenine ring and surrounding hydrophobic residues and electrostatic interactions between the pyrophosphate group and R138 (Fig. 6C).Moreover, compared with the βhairpin flap of MR, that of SlOPR3 moves closer to the active site region, rendering the hydrophobic cleft narrower and shorter than in MR, which further complicates the binding of NADH to this site (Fig. 6D).This observation may be due to the size difference in the β-hairpin flap between MR and SlOPR3, as the loop connecting the two β-strands is longer in MR by six amino acids.Thus, the β-hairpin flap in MR may have greater freedom of movement than in SlOPR3, facilitating the induced-fit formation of the NADHbinding cleft, as proposed above.This assumption is supported by the fact that in the crystal structures of OYEs preferring NADH as the reducing cofactor, the β-hairpin flap region is more disordered and thus more flexible than in the crystal structures of OYEs preferring NADPH, such as SlOPR3 [36].Conversely, loop 6 is generally more flexible in the crystal structures of OYEs preferring NADPH than in those preferring NADH, which, according to our results, indicates that loop 6 can participate in an induced-fit ligand binding event.
Our MD simulations suggest that the presence of L140 and the absence of a properly folded hydrophobic cleft seem to be the main reasons for the low binding of NADH to SlOPR3, as the coenzyme's tail cannot be stably accommodated.NADH prefers the more hydrophobic β-hairpin flap region over the positively charged loop 6 region in SlOPR3.

Conclusions
A major bottleneck of the industrial application of OYEs as biocatalyst is their dependence on the reduced coenzymes NADPH and NADH.In this regard, NADH is preferred over NADPH because it is more stable and about 70 times less expensive than NADPH [26].Therefore, efforts have been made to switch the cofactor preference in these nicotinamide nucleotide-dependent enzymes toward NADH.However, the molecular basis of cofactor specificity in OYEs remains poorly understood.
Using X-ray crystallography and computational approaches, we demonstrated that SlOPR3-an OYE that prefers NADPH over NADH-binds NADPH and NADH at different sites in the active site.NADPH binds at a positively charged region at loop 6, while NADH binds at a more hydrophobic region at the βhairpin flap.This contrasts with other NADPHdependent OYEs like PETNR in which two arginine residues located on the β-hairpin flap are responsible for NADPH binding.Thus, we conclude that OYEs exhibit different NADPH binding mechanisms.
Regarding the NADH acceptance of SlOPR3, we established that SlOPR3 and maybe other NADPHdependent OYEs exhibit low binding affinity toward NADH because they lack an appropriate NADHbinding site at their β-hairpin flap region that can stably accommodate the AMP moiety as is the case in the NADH-dependent MR.
Our results suggest that in order to switch the coenzyme preference in NADPH-dependent OYEs, the β-hairpin flap region needs to be targeted by structurebased engineering approaches to create a high-affinity binding site for NADH, for example, by replacing L140 for an adenine ring-stabilizing phenylalanine and adjusting the length of the hairpin so that it resembles more that of NADH-preferring ERs.Conversely, loop 6 should be the engineering target to switch the nicotinamide affinity in NADH-dependent OYEs.Generation of such SlOPR3 variants with switched coenzyme preference is currently underway.

Materials
Chemicals were from Carl Roth (Karlsruhe, Germany) or Merck (Darmstadt, Germany) if not stated otherwise.

Molecular docking
We used Schr ödinger's Glide module (version 2021-4) [37] for docking NAD(P)H to SlOPR3.Despite the crystal structures of SlOPR3 wildtype and some variants being available in the PDB, we needed to generate a suitable receptor file for docking because, in the PDB models, loop 6 is either not completely resolved or adopts an unusually stretched conformation required to form a self-inhibitory complex (PDB: 2HSA [38]).Thus, we used AlphaFold2 [39] to predict the structure of SlOPR3 by providing PDB: 2HS8 as a template (after undoing the Y364F amino acid exchange).From the top five AlphaFold models, we chose loop 6 from that model in which R283-a residue located on loop 6 that we anticipated to be involved in NADPH binding-was pointing toward the enzyme's active site and merged it into the structure of PDB: 2HS8 [38], after deleting the original loop 6.This final structure was used as receptor file for SlOPR3 docking.
Before docking, all receptor files were processed with Glide's protein preparation tool, which adds hydrogen atoms and formal charges, optimizes the protein structure, and performs an energy minimization with the POLS-2005 force field.The coenzyme structures were downloaded from the PDB and prepared with the program's LigPrep module using default parameters (i.e., hydrogens were added, and the structures were minimized using the same force field).The search space for the docking was set with Glide's grid preparation module by superimposing the MR-NADH 4 crystal structure (PDB: 2R14) onto our receptor structures and generating a grid box based on the centroid of the nicotinamide moiety of the bound NADH 4 molecule in 2R14.The box dimensions of the inner and outer grid boxes were 20 Å × 20 Å × 20 Å and 38 Å × 38 Å × 38 Å, respectively.We also applied positional core constraints to constrain the coenzyme's nicotinamide group (via SMARTS strings) to a position similar to that observed in 2R14 (i.e., above the isoalloxazine ring).The coenzymes were then docked into the SlOPR3 structure employing the extra precision (XP) docking feature in Glide with default parameters.Interactions between the coenzymes and SlOPR3 were visualized and analyzed with PYMOL (The PyMOL Molecular Graphics System, Version 2.0 Schr ödinger, LLC) and LIGPLOT  (v4.5.3, [40]).
Escherichia coli main cultures were inoculated with fresh overnight cultures to an optical density of OD 600 = 0.05-0.075 in 800 mL LB medium supplemented with 100 μgÁmL À1 ampicillin and 34 μgÁmL À1 chloramphenicol and grown at 37 °C and 140 rpm.At an OD 600 of 0.7-0.8, the temperature was lowered to 20 °C, and gene expression was induced by adding 0.1 mM IPTG (isopropyl-β-Dthiogalactopyranosid).After 18-20 h, cells were harvested by centrifugation for 15 min at 4 °C and 5000 g.Pellets were washed with a cold 0.9% NaCl solution, centrifuged at 4000 g, and stored at À20 °C until further use.
Pellets were thawed at room temperature, resuspended in lysis buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 10 mM Imidazole, pH 8.0) supplemented with a spatula tip of FMN, and vortexed until completely dissolved.Cells were disrupted by sonication on ice.Cell debris was pelleted by centrifugation at 38 500 g for 45 min at 4 °C.The supernatant was filtered using 0.45-μm PVDF syringe filters and applied onto an equilibrated Ni-NTA column.The column was washed with wash buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 20 mM Imidazole, pH 8.0) and the target protein was then eluted in elution buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 250 mM Imidazole, pH 8.0).Before pooling, all fractions were checked for purity using SDS/PAGE.Fractions containing SlOPR3 were pooled and concentrated using 30-kDa cut-off Amicon ® Ultra Centrifugal Filters (Merck).Afterward, the buffer was exchanged for storage buffer (25 mM NaH 2 PO 4 , 150 mM NaCl, pH 8.0) using a PD-10 column (Cytiva, Freiburg, Germany).SlOPR3 was flashfrozen to droplets using liquid nitrogen and stored at À80 °C.

Determination of the extinction coefficient
Spectra of the native and denatured protein were recorded in the range of 200-800 nm on a double-beam UV/Vis spectrophotometer (Analytik Jena, Jena, Germany) at 20 °C and a final protein concentration of ≈ 1 mgÁmL À1 .For the native spectrum, 900 μL protein in storage buffer was mixed with 100 μL dH 2 O.For the denatured spectrum, 900 μL protein in storage buffer was mixed with 100 μL 20% SDS solution and incubated for at least 20 min at 22 °C.To determine the specific extinction coefficient for SlOPR3, the known extinction coefficient of free FMN at 450 nm (ε 450 = 12 500 M À1 cm À1 [42]) and the following equation were used: A max value native spectrum A 450 value denatured spectrum : Unspecific signals due to light scattering artifacts were corrected by normalizing all absorbance values according to the 600 nm value and shown as a function of the respective extinction coefficient.

Determination of pre-steady-state kinetics: Stopped-flow measurements
The pre-steady-state kinetic parameters of the reductive half-reaction were determined using a stopped-flow device (SF-61DX2, TgK Scientific, Bath, UK) under anoxic conditions (O 2 = 7-9 ppm) in a glove box (Belle Technology, Weymouth, UK) at 25 °C.To provide anoxic conditions, we flushed and incubated the reaction buffer with nitrogen in the glove box overnight.SlOPR3 and pre-weighed coenzymes were incubated in the glove box for at least 30 min or until the O 2 level reached a maximum of 9 ppm and diluted to the desired concentration directly in the glove box.A KinetaScanT diode array detector (MG-6560, Hi-Tech Scientific, Wiltshire, UK) was used to monitor spectral changes of the flavin cofactor at a wavelength of 450 nm, and collected data were analyzed using the Kinetic Studio software (version 4.01, TgK Scientific).To reduce the effect of photoreduction on the flavin cofactor, measurements that exceeded 20 s per shot were recorded with an active auto-shutter.For the reductive rates, 50 μM of enzyme were shot against increasing NAD(P)H concentrations, and the decreasing absorbance at 450 nm was used to determine the kinetic parameters.All measurements were performed in triplicates.Dissociation constants (K d ) and the limiting rate of reduction (k red ) values were calculated using the hyperbolic fit function in the GRAPHPAD PRISM software (v8.4.3,GraphPad Software, San Diego, CA, USA).

Preparation and purification of NADH 4 and NADPH 4 for crystallization
To obtain crystals of SlOPR3, the unreactive NADH and NADPH analogs 1,4,5,6-tetrahydro-NAD(P)H (NADH 4 and NADPH 4 ) were used, which, unlike NAD(P)H, are not oxidized and released from the enzyme during the crystallization experiment.NADPH 4 and NADH 4 were prepared by catalytic hydrogenation of NADPH and NADH, respectively.A solution of 185 mg NADPH or 250 mg NADH dissolved in 6 and 9 mL H 2 O, respectively, was pumped through an HCube ® system (Thales Nanotechnology Inc., Budapest, Hungary) with the following parameters: 1.2 mLÁmin À1 resp. 1 mLÁmin À1 , full H 2 -mode (1 bar), 30 °C.A CatCart ® cartridge containing 10% Pd on C was employed as catalyst.The obtained product solution was collected in a round bottom glass flask, frozen in liquid nitrogen, and the water was removed via lyophilization, yielding 186 mg NADPH 4 and 216 mg NAHDH 4 , respectively.

Crystallization
SlOPR3 crystals were grown by the vapor-diffusion method in hanging drops consisting of 2 μL protein solution (10 mgÁmL À1 ) and 1 μL reservoir solution (100 mM MES/-Tris, pH 6.5, 10 mM ammonium sulfate, and 12-16% PEG8000) at 293 K. Large but macroscopically twinned yellow crystals grew within 1-3 days.Crystals of SlOPR3 complexed with NADH 4 or NADPH 4 were obtained by crystal soaking into reservoir solutions containing NADH 4 or NADPH 4 for 1-60 min; longer soaking times cracked the crystals and substantially deteriorated their diffraction power.To achieve the highest NAD(P)H 4 concentration possible, we added NAD(P)H 4 directly as powder to the soaking solutions at amounts over their solubility limit.SlOPR3 crystals were also obtained via co-crystallization and direct crystallization of the ternary complex, but the resulting crystals were of substantially lower quality than those obtained via soaking.All crystals were cryo-protected in reservoir solution supplemented with a spatula tip of NAD(P)H 4 and 20% MPD and then vitrified in liquid nitrogen before data collection.
Data collection, structure solution, and refinement X-ray diffraction data for SlOPR3 complexed with NAD(P)H 4 were collected at DESY (Hamburg, Germany) on beamline P11 at 100 K. Data sets were processed and scaled using the XDS package [43].Data collection information is summarized in Table 1.
Both structures were solved by molecular replacement using PHASER [44] from the PHENIX suite [45].A truncated version of PDB: 2HS6 [38] lacking loop 6 was used as search model.Model building and refinement were performed with Coot [46] and phenix.refine[47], respectively.The final models were validated using MolProbity [48], and refinement statistics are provided in Table 1.Both crystal structures were deposited in the Protein Data Bank (see Table 1 for PDB entry codes).

Molecular dynamics simulations
The receptor and coenzyme molecules were separately extracted from the PDB file containing the docking results using the EDITCONF tool from GROMACS 2022 [49].The receptor (i.e., SlOPR3) GRO coordinates and topology files were generated using the PDB2GMX tool with the amber99sb-ildn force field.The GMX ligand topologies were generated using acpype (alanwilter.github.io/acpype)by feeding it the PDB and charge values.The receptor and coenzyme coordinates and topology files were merged using the gmx editconf and gmx genconf tools to generate the complex (i.e., SlOPR3-NADH) GRO and topology files.The MD unit cell was defined as a dodecahedron with a minimum distance of 1.0 nm between the complex and the box edges using the gmx editconf tool.The system was solvated with the SPC/E water model using the gmx solvate tool and neutralized by adding Na + ions using the gmx grompp and gmx genion tools with a salt concentration of 150 mM.The system was then subjected to standard energy minimization and equilibration using the gmx grompp and gmx mdrun tools.Energy minimization was performed using the steepest descent algorithm with a maximum force of 1000 kJÁmol À1 Ánm À1 and a maximum of 5000 steps.The equilibration step consisted of two phases: NVT (constant number of particles, volume, and temperature) and NPT (constant number of particles, pressure, and temperature).The NVT phase was run for 100 ps with a temperature coupling of 298 K using the v-rescale thermostat.The NPT phase was run for 200 ps with a pressure coupling of 1 bar using the Parrinello-Rahman barostat.
The parameters for the 250-ns production run of the SlOPR3-NADH system were as follows: the temperature and pressure were set at 298 K and 1 bar by the v-rescale thermostat and Parrinello-Rahman barostat, respectively; bonds were constrained using the LINCS algorithm; the Verlet cut-off scheme was used to process intra-atomic interactions; the PME method was implemented to account for Coulombic and Lennard-Jones interactions; and a van der Waals cut-off radius of 1.0 was applied.
A second 250-ns run at a higher temperature was performed to sample a larger conformational space.We started from the equilibrated structure generated during the first run, regenerating the initial velocities.The running parameters were kept constant except for the temperature, which was set at 310 K.

Fig. 2 .
Fig. 2. Stopped-flow pre-steady-state kinetics of SlOPR3.The concentration dependence of FMN reduction in SlOPR3 with NADPH (left panel) is compared to that with NADH (right panel).All measurements were performed in triplicates.c NAD(P)H , NAD(P)H concentration; k, rate constant; K d , dissociation constant; k red , rate constant associated with the reductive half-reaction.The figure was prepared using GRAPH-PAD PRISM.

Fig. 3 .
Fig. 3. Illustrative summary of the docking results.The 10 top-ranked binding poses for both NADPH (blue sticks) and NADH (pink sticks) were superimposed to show differences in their binding modes.The left and right panels show the best binding pose (i.e., pose with the highest docking score) of NADPH and NADH, respectively.The enzyme is shown as a gray cartoon while interacting residues and ligands are shown as sticks.FMN is depicted as yellow sticks, and loop 6 and the β-hairpin flap are highlighted in red and green, respectively.Dashed black lines indicate cofactor-enzyme interactions.Red dashed lines highlight cation-π-cation interactions.The figure was prepared using PYMOL.

Fig. 4 .
Fig. 4. Crystal structures of the open and closed conformation of SlOPR3 in complex with NADPH 4 .(A) Detailed view of NADPH 4 (blue sticks) binding to the active site of the open conformation of SlOPR3.SlOPR3 is shown as a gray cartoon while interacting residues are illustrated as sticks.FMN is depicted as yellow sticks, and polar and electrostatic interactions are indicated by black dashes.For the sake of clarity, no interactions are depicted for the second bound NADPH 4 molecule (bottom).(B) LigPlot of the interactions between SlOPR3 and NADPH 4 in the open conformation.H-bonds and electrostatic interactions are indicated by green dashes (distances are in Å), while hydrophobic interactions are shown as red arcs.(C) Superimposition of loop 6 and NADPH 4 from the open (magenta cartoon) and closed (red cartoon) conformation (dashes indicate missing atoms).The NADPH 4 molecules from the open and closed conformation are depicted as light blue and blue sticks, respectively, while R343 and R366 of the arginine triad are shown as gray and cyan sticks, respectively.Structural elements from the open and closed conformation are denoted by (o) and (c), respectively.Black arrows indicate the movements of loop 6 and R283 to transit from the open to the closed conformation.(D) Detailed view of NADPH 4 (blue sticks) binding to the active site of the closed conformation of SlOPR3.Color code and presentation style are the same as for panel (A).Red dashes indicate the cation-π-cation interactions between the adenine ring and R283/R343.(E) LigPlot of the interactions between SlOPR3 and NADPH 4 in the closed conformation.Annotations are the same as in panel (B).The figure was prepared using LigPlot and PYMOL.

Fig. 5 .
Fig. 5. Crystal structure of the SlOPR3-NADH 4 complex and comparison to the SlOPR3-NADPH 4 structure.(A) Detailed view of NADH 4 (pink sticks) binding to the active site of SlOPR3.SlOPR3 is shown as a cyan cartoon while interacting residues are illustrated as sticks.FMN is depicted as yellow sticks, and polar and electrostatic interactions are indicated by black dashes.(B) LigPlot of the interactions between SlOPR3 and NADH 4 .H-bonds and electrostatic interactions are indicated by green dashes (distances are in Å), while hydrophobic interactions are shown as red arcs.(C) Superimposition of the active site of SlOPR3-NADH 4 (cyan cartoon and sticks) onto that of SlOPR3-NADPH 4 (gray cartoon and sticks).Sulfate ions are shown as yellow sticks while NADPH 4 and NADH 4 are depicted as transparent blue and solid pink sticks, respectively.The asterisks indicate that the sulfate ions are only present in the SlOPR3-NADH 4 complex structure (not in the SlOPR3-NADPH 4 complex structure).Note that the bottom sulfate ion, which is omitted in panel (A) for clarity, overlaps perfectly with the 2 0 -phosphate group of NADPH 4 .(D) Superimposition of the binding conformation of NADH 4 (pink sticks) onto that of NADPH 4 (blue sticks) to visualize their different binding modes.The arrows indicate toward which structural elements the coenzyme tails are pointing.Note that the tail of NADH 4 could not be fully modeled due to the lack of electron density; however, the β-phosphate clearly points in the direction of the β-hairpin flap.The figure was prepared using LigPlot and PYMOL.

Fig. 6 .
Fig. 6.Overview of the molecular dynamics simulations results.(A) Root-mean-square (RMSD) plots for the investigated systems.(B) Representative binding mode from the molecular dynamics (MD) simulation of the SlOPR3-NADH system at 298 K. SlOPR3 is shown as a gray cartoon while interacting residues are depicted as sticks.FMN and NADH are shown as yellow and pink sticks, respectively.Loop 6 and the β-hairpin flap are highlighted in red and green, respectively.Black dashes indicate hydrogen bonding and electrostatic interactions.(C) Representative binding mode from the MD simulation of the SlOPR3-NADH system at 310 K. Color code and presentation style are the same as for panel (B).(D) Comparison of NADH/NADH 4 binding between our MD simulation at 310 K and the crystal structure of morphinone reductase (MR) in complex with NADH 4 (PDB: 2R14).The middle panel shows the structural superimposition of a representative SlOPR3-NADH structure from our simulations onto the crystal structure of MR in complex with NADH 4 .The presentation style is the same as for panel (B).SlOPR3 and MR are shown as gray and cyan cartoons, respectively.The β-hairpin flap of SlOPR3 and MR are highlighted in green and dark green, respectively.NADH from our simulation and NADH 4 from the MR-NADH 4 crystal structure are depicted as pink and light pink sticks, respectively.The curved arrow indicates that the β-hairpin flap of SlOPR3 is located closer to the active site barrel than that of MR, thereby downsizing the hydrophobic cleft formed by the β-hairpin flap and the C-terminal loop compared with that in MR.The left and right panels show the hydrophobic binding cleft of SlOPR3 (gray) and MR (cyan) in surface representation, respectively.The surfaces of the β-hairpin flap in SlOPR3 and MR are highlighted in green and dark green, respectively.In SlOPR3 (left panel), the potential NADH-binding cleft is small (as it is closed by the β-hairpin flap from the right side) and blocked by L140 (dark green sticks), which prevents NADH binding.In MR (right panel), the β-hairpin flap points away from the active site so that a NADH-binding cleft is formed in which the adenosine tail of NADH 4 fits perfectly; in addition, the adenine ring is stabilized by π-stacking interactions with F137 (red arrow).The figure was prepared using PYMOL.

Table 1 .
Data collection and refinement statistics.