Structure–function characterization of an aldo–keto reductase involved in detoxification of the mycotoxin, deoxynivalenol

Deoxynivalenol (DON) is a mycotoxin, produced by filamentous fungi such as Fusarium graminearum, that causes significant yield losses of cereal grain crops worldwide. One of the most promising methods to detoxify this mycotoxin involves its enzymatic epimerization to 3-epi-DON. DepB plays a critical role in this process by reducing 3-keto-DON, an intermediate in the epimerization process, to 3-epi-DON. DepBRleg from Rhizobium leguminosarum is a member of the new aldo–keto reductase family, AKR18, and it has the unusual ability to utilize both NADH and NADPH as coenzymes, albeit with a 40-fold higher catalytic efficiency with NADPH compared to NADH. Structural analysis of DepBRleg revealed the putative roles of Lys-217, Arg-290, and Gln-294 in NADPH specificity. Replacement of these residues by site-specific mutagenesis to negatively charged amino acids compromised NADPH binding with minimal effects on NADH binding. The substrate-binding site of DepBRleg is larger than its closest structural homolog, AKR6A2, likely contributing to its ability to utilize a wide range of aldehydes and ketones, including the mycotoxin, patulin, as substrates. The structure of DepBRleg also suggests that 3-keto-DON can adopt two binding modes to facilitate 4-pro-R hydride transfer to either the re- or si-face of the C3 ketone providing a possible explanation for the enzyme’s ability to convert 3-keto-DON to 3-epi-DON and DON in diastereomeric ratios of 67.2% and 32.8% respectively.


Results
SSN cluster analysis of DepB Rleg with AKRs. AKRs are traditionally classified into families based on sequence identity and phylogenetic analysis. The nomenclature system consists of the root symbol AKR, followed by an arabic numeral designating the family and an alphabet representing the subfamily 18 . The number of AKRs in the protein database continues to expand and to date 18 families have been identified. Protein sequences in each family share at least 40% sequence identity while sub-families share 60% sequence identity 19 . We used a protein sequence similarity network (SSN) 20 as an alternate, less computationally demanding method to determine sequence relationships of DepB Rleg with other AKR family members. For ease of computation, sequences were iteratively clustered to produce a 40% representative node network and filtered based on sequence length. The total number of sequences following the reduction of the initial dataset is 878. In this network, nodes represent protein sequences sharing 40% or more sequence identity and edges represent the pairwise alignments between protein sequences. Edges were drawn between nodes if they exceeded the prescribed stringency threshold or BLAST E-value of e −57 .
Previous phylogenetic analysis showed that AKRs can be divided into two large groups with AKR1-AKR5 forming one branch of the phylogenetic tree and the other AKRs forming a separate branch 21 . This is generally reflected in the SSN topology where AKR1 to AKR5 family members cluster together ( Fig. 2A). At this stringency threshold, there was good delineation of AKR7-11 and AKR 13-15 into distinct, isofunctional clusters. DepB Rleg and its homologs from D. mutans 17-2-E-8 10 and AKR18A1 from Sphingomonas S3-4 22 clustered with AKR6, AKR12, and AKR14 families. At higher stringency thresholds (e −67 ), DepB Rleg and its closest homologs resolve into a single isofunctional cluster (Fig. 2B). Curiously, in vitro assays previously revealed that AKR18A1 reduces 3-keto-DON to DON rather than 3-epi-DON in the presence of NADH 22 . The enzyme also reduces the C7 ketone group of the estrogenic mycotoxin, zearalenone (ZEN) to produce α-zearalenol (α-ZOL) and β-zearalenol (β-ZOL). Aside from this, AKR18A1's substrate specificity towards other endogenous aldehydes and ketones has not been examined, nor has its crystal structure been solved.
Besides AKR18A1, the next closest homologs of DepB Rleg identified from this SSN analysis were AKR12 family members based on the first stringency cut-off threshold (e −57 ). As per AKR classification rules, DepB Rleg is not an AKR12 family member as it shares less than 40% sequence identity with enzymes from this family. In addition, these AKR12 members are involved in the biosynthetic pathways of polyketide macrolide antibiotics such as tylosin, erythromycin, and avermectin [23][24][25] . A genome neighborhood diagram (GND) of DepB Rleg revealed the presence of a putative monoamine oxidase (MAO) upstream of the DepB Rleg gene along with several putative Figure 1. The Dep pathway. The C3 OH group of DON (shown in red) is stereochemically inverted to produce the diastereomer, 3-epi-DON. In the expanded image of this stereogenic center, R 1 and R 2 represent the priority groups attached to the sp 2 hybridized C3. Hydride attack on the re-face at the prochiral C3 center generates 3-epi-DON.
Crystallization and structure determination of DepB Rleg . DepB Rleg was successfully crystallized, and its structure was solved by molecular replacement using an AKR from Polaromonas sp. JS666 as a search model (PDB ID: 4XK2). Structural and refinement statistics are summarized in Table 2.
Four protomers of DepB Rleg are present in the asymmetric unit designated as chains A, B, C, and D. Overall, the three-dimensional structure of each protomer adopts the classic triose-phosphate isomerase (TIM) barrel fold (α/β) 8 , except that the presence of Pro-204 in the region that would otherwise form strand β7 disrupts the (A) Nodes within each cluster contain a representative protein sequence of a collection of sequences that share 40% or more sequence identity. Clusters have been annotated and color-coded based on curated protein sequences from the AKR Database 21 . Grey nodes represent AKRs that have no functional annotation. Metabolic pathways associated with the function of AKR enzymes have been color-coded to cross-reference with each cluster and numerically labeled to designate each AKR family number. The nodes of AKRs possessing biochemical data on coenzyme specificity and crystal structures have also been accordingly labeled as outlined in the legend. (B) SSN generated at a threshold of e −67 depicting complete segregation of the AKR18 family from AKR6, AKR12 and AKR14 members. (C) Genome Neighborhood Diagram of DepB Rleg. The GND was generated using the EFI-GNT server which depicts the coding sequence region of depB Rleg along with putative functions of upstream and downstream genes. www.nature.com/scientificreports/ A pair of protomers (A and C, B and D) associate together forming a dimer with an average interface area of 2362.9 Å 2 , stabilized by 24 hydrogen bonds and 13 salt bridges. Dimerization involves 3D domain swapping with the exchange of the α-helix of loop C which interacts with loop A on the associating protomer. These dimers interact around the crystallographic four-fold symmetry axis to form a D 4 symmetric octamer (Fig. 3B), with key interactions mediated by the N-terminal β-hairpin, α3, α4, and the β4α4 loop. PISA 34 estimates that these interactions bury 748.6 Å 2 and result in a solvation-free energy gain of -809.8 kcal/mol. The absolute molecular weight of DepB Rleg was determined experimentally by SEC-MALS to be 325.6 ± 12.1 kDa which is consistent with the 317.5 kDa predicted for a DepB Rleg octamer. (See Supplementary Fig. S3).
Diversity in certain loop regions was observed among these structural homologs. Differences in the conformation of loop B and the auxiliary helices, H1 and H2 could be due to conformational changes upon binding of NADP(H) and/or substrates. Loop B in the apo-structures of DepB Rleg and Polaromonas sp. AKR is disordered, but clear electron densities for this region were observed for the NADP + binary complexes of the other structural homologs. This is in agreement with the finding that loop B becomes ordered upon coenzyme binding 38 . In addition to these conformational differences, variations in loop length are also evident, particularly for the E. coli Tas protein which possesses a long α4β4 loop (24 residues) compared with DepB Rleg (8 residues). The length and structure of DepB Rleg loop C (26 residues) also differed with AKR6A2 (6 residues), Tas (5 residues), AKR14A1 (10 residues), and mithramycin side chain reductase (6 residues). In the hydroxysteroid reductase subfamily of Table 2. X-ray diffraction data and refinement statistics. *Highest resolution shell is shown in parentheses. www.nature.com/scientificreports/ AKRs truncation of this loop led to reduced specific activity towards steroid substrates, therefore the diversity in loop lengths among these AKRs likely contributes to corresponding differences in substrate specificity 39 .
Coenzyme binding site. DepB Rleg reduced 3-keto-DON using NADH and NADPH as coenzymes, although the catalytic efficiency with NADH was 40-times lower than with NADPH. The dissociation constant (K d ) for NADH, as determined by tryptophan fluorescence quenching experiments, was about sixfold higher than NADPH. In contrast, most AKRs are NADPH specific with NADPH dissociation constants on average, 1000-fold lower than NADH 40 . DepB Rleg , therefore, shares the rare dual coenzyme specificity with a small number of enzymes from different AKR families, including the hydroxysteroid dehydrogenases (AKR1C) 41,42 , xylose reductases (AKR2B) [43][44][45] , beta subunit of voltage-gated potassium ion channels (AKR6A) 46 , aflatoxin reductases (AKR7A) 47 , and pyridoxal dehydrogenases (AKR15A) 48 . The structural basis for NADH specificity was studied for Candida tenuis xylose reductase (AKR2B5), revealing a key Glu-227 residue on loop B which forms bidentate hydrogen bonds with the 2' and 3' hydroxyl groups of the adenosine moiety of NAD +44 . In DepB Rleg, this residue is also conserved and corresponds to Glu-222. In the NADP + bound complex of AKR2B5, the 2' monophosphate of NADP + interacts with a positively charged Arg-280, the peptide backbone of Asn-276, and Ser-275. These residues are not strictly conserved in DepB Rleg and are instead replaced with Gln-294, Arg-290, and Ala-288, underscoring the general variability in NADP + interactions among the AKRs. Although NADP + was present at 400 µM during the crystallization of DepB Rleg, there was no electron density corresponding to this coenzyme. The putative coenzyme binding pocket was suitably defined and superimposed well with coenzyme binding sites of AKR binary complexes. AKRs bind NADPH in an extended anti-conformation to achieve the stereospecificity of hydride transfer 33 . An aromatic residue typically forms π-stacking interactions with the nicotinamide ring while the carboxamide moiety is oriented by highly conserved glutamine, serine, and asparagine residues. In DepB Rleg , these residues correspond to Trp-206, Gln-178, Ser-152, and Asn-153 respectively. In AKR6A2, the closest structural homolog of DepB Rleg, the 2'monophosphate of NADP + interacts with the side chains of Gln-62, Lys-254, Ser-325, and Gln-329, and (Fig. 4A). In DepB Rleg , these residues correspond to Glu-26, Lys-217, Arg-290, and Gln-294 respectively. To examine the conservation of these residues, a www.nature.com/scientificreports/ multiple sequence alignment was constructed with AKR family representatives. Lys-217 which is present on loop B is not strictly conserved and is often substituted for a basic residue or a smaller hydrophobic residue. Arg-290 is partially conserved across the AKR families and this position is often frequented by a basic residue or a polar hydrophilic residue. Gln-294 is strictly conserved across AKR6A2, AKR7A1, AKR9A1, AKR11A1, AKR14A1, AKR15A1, and AKR18A1 but for the other families, it is replaced with either a basic or aromatic residue (Fig. 4B). Lastly, Glu-26 was the least conserved and was therefore not selected for further analysis. All recombinant N-terminal His-tagged DepB Rleg coenzyme variants were purified by Ni-NTA chromatography with expected molecular weights of 39 kDa (See Supplementary Fig. S1). Replacements of Lys-217, Arg-290, and Gln-294 for negatively charged glutamate significantly altered the K d for NADPH relative to wild type DepB Rleg , with no significant effect on the K d of NADH. Among these variants, the R290E variant displayed an 11-fold increase, followed closely by Q294E with a ninefold increase and finally K217E with only a sixfold change (Table 3). Substitutions for neutral residues as in the case for R290N displayed a moderate change in the K d by fivefold, while for K217M, the K d change relative to the wild type was more pronounced with a nearly sevenfold change. Overall, for all the mutants, there was no appreciable change in K d for NADH.
Substrate-binding site. The catalytic mechanism of AKRs involves a stereospecific 4-pro-R hydride transfer from the nicotinamide ring of the coenzyme to the substrate carbonyl carbon, followed by protonation of the carbonyl oxygen by a tyrosyl residue 49 . This is facilitated by a proton relay with histidine or a lysine-aspartate pair 40 . In DepB Rleg , these catalytic residues are conserved: Asp-48, Tyr-53, Lys-81, and His-122. A water molecule occupies the position equivalent to the carbonyl oxygen group of the substrate and forms hydrogen bonds (2.6-2.9 Å) to the side chains of both His-122 and Tyr-53 49 . DepB Rleg 's substrate-binding pocket is lined with the following residues: Met-21, Asp-48, Val-52, Tyr-53, Lys-81, Arg-83, Phe-84, His-122, Ala-123, Ser-152, Asn-153 and Gln-180. Overall, DepB Rleg 's substrate-binding pocket is slightly larger and more hydrophobic in comparison with the NADP + -cortisone bound complex of AKR6A2. The physiological substrate of AKR6A2 is unknown, however, much like DepB Rleg it also displays broad substrate specificity with aldehyde and ketone substrates 50 .  Table S1 for accession codes). www.nature.com/scientificreports/ The stereospecificity of hydride transfer is strictly conserved in AKRs. In DepB Rleg, the residues Ser-152, Asn-153, and Gln-178 are critical for maintaining the anti-conformation of the nicotinamide ring to achieve this stereospecificity 51 . However, hydride attack may occur either on the re-face or si-face of the prochiral carbonyl group. We determined by HPLC analysis that DepB Rleg reduced 3-keto-DON to produce a diastereomeric ratio of 67.2% for 3-epi-DON and 32.8% for DON (See Supplementary Fig. S4 for HPLC chromatograms). To examine the molecular basis for this diastereoselectivity, we superimposed apo-DepB Rleg with the recently solved Debaryomyces nepalensis xylose reductase (PDB ID: 5ZCM) complexed with a DTT-NADP + adduct 52 . In this complex, the geometry of the C4N of NADP + strongly resembles the puckered ring conformation of reduced NADPH. 3-keto-DON was then modeled into this complex in an orientation poised for hydride transfer to the re-face to produce 3-epi-DON. The carbonyl oxygen of 3-keto-DON was placed within hydrogen-bonding distance of Tyr-53 and His-122, while Arg-343 from Loop C was re-positioned to provide hydrogen bond contacts with the C8 ketone of 3-keto-DON. Phe-84, which is present on Loop A could potentially stack with the cyclohexene ring of 3-keto-DON (Fig. 5A). This residue corresponds to Trp-86 in Thermotoga maritima AKR, an enzyme that catalyzes the reduction of ethyl 2-oxo-4-phenylbutyrate to the R-and S-enantiomer of ethyl-2-hydroxy-4-phenylbutyrate 53 . In that study, the replacement of Trp-86 with smaller amino acids alleviates space constraints and enabled the enzyme to increase the production of the R-alcohol 53 . To produce DON, 3-keto-DON would require a 180° flip in the substrate-binding pocket (Fig. 5B). An observation gleaned from sequence analysis with other AKRs, is the presence of a bulky residue following the catalytic histidine (Fig. 5C). However, in DepB Rleg this position is occupied by Ala-123 which eliminates any space constraints allowing for the accommodation of the C15 primary alcohol of 3-keto-DON. We note that in AKR18A1 which reduces 3-keto-DON to DON, a glycine residue occupies the positions corresponding to Phe-84 and Ala-123 providing a great deal of flexibility in its respective active site.

Discussion
A comprehensive biochemical and structural analysis was conducted for DepB Rleg , a close homolog of DepB from D. mutans 17-2-E-8. This AKR catalyzes the critical step during DON detoxification to afford the significantly less toxic diastereomer, 3-epi-DON. SSN cluster analysis revealed that DepB Rleg is a member of the recently discovered AKR18 family. Within this family, only preliminary biochemical characterization was conducted for AKR18A1 from Sphingomonas S3-4 suggesting its involvement in a DON detoxification pathway 22 . At a cutoff threshold of e −57 , DepB Rleg localizes to a functionally divergent cluster in the SSN comprised of AKR6, AKR12, and AKR14 families. AKR6 members comprise the cytosolic subunit of voltage-gated potassium ion channels and are proposed to modulate these channels via coenzyme binding 46 . AKR12 members are reductases involved in macrolide antibiotic synthesis [23][24][25] , while the AKR14A1 member identified in the SSN is proposed to be involved in methylglyoxal detoxification 54 . The gene organization as depicted in the GND indicates that depB Rleg is not located in a similar biosynthetic operon. Instead, it is flanked by a putative MAO and several putative putrescine/ spermidine transporter protein genes.
DepB Rleg 's substrate specificity was examined across a diverse range of model carbonyl substrates with its preferred coenzyme, NADPH. Amongst the ketone substrates, DepB Rleg possesses the highest specificity constant for 9,10-PQ while comparable specificity constants were observed for the other diketone, isatin and 3-keto-DON. 9,10-PQ is a xenobiotic air pollutant that triggers the generation of reactive oxygen species (ROS) in vivo 12 while isatin is an endogenous indole derived from tryptophan or phenylalanine metabolism by gut bacteria and has cytotoxic effects on microbes and various human cancer cell lines 55 . Overall, higher specificity constants were observed for ketones, versus aldehyde substrates. Among the lipid peroxidation substrates, DepB Rleg displayed the highest specificity constant for 4-oxo-2-nonenal, followed by methylglyoxal and acrolein. DepB Rleg had low specificity constants for 4-hydroxy-nonenal and DL-glyceraldehyde. The latter is typically a model substrate used for aldose reductases and aldehyde reductases from AKR1 families 17 . Surprisingly, DepB Rleg can also reduce the mycotoxin patulin to E-ascladiol, an activity that was previously described only for the short-chain www.nature.com/scientificreports/ dehydrogenase family of oxidoreductases 56,57 . In summary, the broad substrate specificity of this AKR, therefore, lends credence to its secondary role in general detoxification processes. A consensus among most AKRs is their preference for the coenzyme, NADPH over NADH 49 . Investigations of the coenzyme binding affinities of DepB Rleg revealed that it can utilize both NADPH and NADH, albeit with preference for NADPH over NADH. Given that NADPH is more expensive than NADH, the ability of DepB Rleg to utilize NADH as a coenzyme will be beneficial for its use as a biocatalyst to detoxify DON. Based on comparative structural analysis, the residues Lys-217, Arg-290, and Gln-294 are proposed as key contributing elements for DepB Rleg 's preference for NADPH. These structural elements likely contribute to a hydrogen bond network encompassing the 2'monophosphate of NADPH. Precedents for coenzyme specificity are well established for short-chain dehydrogenases (SDRs). In NADP + dependent SDRs, a basic residue in the glycine-rich motif of the Rossman fold lies governs NADP + specificity while for NAD + dependent SDRs, typically an aspartate or glutamate is present 58 . Similarly, in DepB Rleg , the basic residues, Arg-290 and Lys-217 contribute the necessary salt-bridge interactions with the 2'monophosphate, while Gln-294 also provides hydrogen bond interactions and provides planar stacking interactions between the amide group of its side chain and the adenosine ring. Not surprisingly, substitutions for negatively charged residues generated electrostatic repulsion with the 2'monophosphate group resulting in an increased NADPH K d value for R290E, Q294E, and K217E. While the change in K d for the R290N variant was not as dramatic, we rationalize that replacement of the arginine for a polar, shorter residue like asparagine likely altered the hydrogen bond network coordinating the 2'monophosphate which consequently affected the binding mode of NADPH. K217M and K217E displayed a similar increase in K d for NADPH, however, the effect was not as pronounced as in the Arg-290 variants. As Lys-217 is present on Loop B, which displays a great deal of sequence variance, it is plausible that a proximal, basic residue may have compensated for the absence of the lysine residue in the K217M and K217E variants.
Reduction of 3-keto-DON by DepB Rleg produced 3-epi-DON and DON in diastereomeric ratios of 67.2% and 32.8% respectively. Modeling suggests that the formation of the two diastereomers is due to alternate 3-keto-DON binding modes that enable the re or si-face to approach the 4-pro-R hydrogen of the nicotinamide cofactor. Although DepB Rleg does not show strict diastereoselectivity towards 3-epi-DON formation, coupling the enzyme with DepA which has an apparent K m of 32 ± 4 µM for DON 9 , would drive the equilibrium towards 3-epi-DON formation.
In conclusion, this comprehensive analysis of DepB Rleg provides extensive insights into the AKR18 family. As evidenced by this structure-function study, DepB Rleg presents as a versatile biocatalyst owing to its broad substrate specificity and a large active site which can accommodate a range of toxic endogenous aldehydes and ketones. The availability of the crystal structure of the enzyme should also facilitate future protein engineering work in targeting residues in the vicinity of the proposed substrate-binding site to improve the diastereoselectivity of the enzyme. While the feasibility of employing DepB in DON detoxification applications remains a contentious issue due to its NADPH dependence, this study provides a translational framework to modify its coenzyme specificity.
Protein expression and purification. His-tagged DepB Rleg and variants were recombinantly expressed in E.coli BL21 LOBSTR (low background strain). This engineered strain minimizes background contamination by histidine-rich E. coli proteins such as SlyD and ArnA during Ni-NTA chromatography 59 . An overnight starter culture was used to inoculate 4L of LB media and cultures were grown at 37 °C with shaking speed at 200 RPM. Recombinant protein expression was induced with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) and incubated at 15 °C overnight with shaking. Cells were harvested by centrifugation and washed with 20 mM HEPES pH 8.0. The pellet was resuspended in 20 mM HEPES pH 8.0 buffer containing up to 1 mg/mL DNase I and lysed by 7-8 passages through a French press at 15,000 lb/in 2 . Cell debris was removed through centrifugation at 4 °C and the clarified lysate filtered through a 0.45 µm filter before incubation for 1 h at 4 °C with Ni 2+ -NTA resin in buffer (50 mM sodium phosphate buffer pH 8.0, 300 mM NaCl) with binding buffer (20 mM imidazole pH 8, 150 mM NaCl). The mixture was loaded onto a gravity column and washed with the same binding buffer. His-tagged proteins were eluted with 150 mM imidazole pH 8. Buffer exchange with 20 mM HEPES pH 7.5 with 10% glycerol and 150 mM NaCl was conducted in a stirred cell equipped with a YM10 filter (Amicon). Protein concentration was determined using a Bradford Assay 60 with Bovine Serum Albumin used as a standard. The purity of the recombinant enzyme and the molecular weight of recombinant DepB Rleg and coenzyme variants was estimated using Coomassie blue-stained SDS-PAGE.

Molecular weight determination. The absolute molecular weight of DepB Rleg was determined using Size
Exclusion Chromatography coupled with Multi-Angle Light Scattering (SEC-MALS). The column utilized for size exclusion was a P3000 single-pore GPC/SEC column (Malvern Panalytical). The system was equilibrated with the following buffer system: 20 mM HEPES, 150 mM NaCl, 0.02% NaN 3 at pH 7. The concentration of the protein samples was 0.5 mg/mL and samples were filtered through 0.2 µm filters prior to injection.
Crystallization and X-ray diffraction collection. Crystallization conditions for DepB Rleg were screened using the JCSG-plus™ kit (Molecular Dimensions) and an initial hit was obtained in condition B1 (0.1 M citrate pH 4, 0.8 M ammonium sulfate). Condition B1 was optimized and a crystal was obtained in the following www.nature.com/scientificreports/ optimized condition: 0.8 M ammonium sulfate, 20 mM ammonium acetate, and 0.1 M TRIS citrate. The protein was crystallized at 11 mg/mL in the presence of 400 µM NADP + via sitting drop vapor diffusion at 288 K. X-ray diffraction data were collected at the Canadian Macromolecular Crystallography Facility (CMCF-BM), processed using XDS, and scaled using XSCALE 61 . Crystals diffracted to 2.5 Å, and belonged to the space group P42 1 2, with unit-cell parameters a = b = 159.45, c = 163.9 Å. The crystal structure of DepB Rleg was solved using the automated molecular replacement pipeline, MrBUMP which generates an exhaustive list of search models. First, multiple sequence alignments are conducted between the query and homologous protein structures, second, the top hits are subjected to model preparation using a suite of CCP4 programs and finally, the models are submitted to Phaser for molecular replacement 62 . The best search model prepared from this process is the AKR from Polaromonas sp. JS666. AutoBuild and subsequent refinements were conducted in PHENIX 63 and final model building was conducted in COOT 64 .
Molecular determinants of coenzyme specificity. Structural alignments were conducted via the DALI server 65 . The ternary complex of the AKR6A2 member, Rattus norvegicus voltage-dependent K + channel beta subunit (Kvβ) (PDB ID: 3EAU) was superimposed on to the apo-DepB Rleg structure. Structures were visualized in PyMOL (Schrödinger, Inc.) and torsion angles of residues were adjusted in the DepB Rleg structure to prevent steric clash with the 2' monophosphate and the nicotinamide ring of NADP + . Putative residues involved in NADPH specificity were selected within a 4-5 Å distance from the 2' monophosphate attached to the adenosine ribose. Multiple sequence alignments were constructed in UGENE 66 using representative sequences from each AKR family and aligned using the MUSCLE algorithm 67 Based on this analysis, the following residues were selected as candidates for mutagenesis: Lys-217, Arg-290 and Gln-294.

Mutagenesis.
Site-specific mutagenesis and mutagenic primer design were conducted as per the Single-Primer Reactions IN Parallel (SPRINP) protocol previously described 68 . The wild type DepB Rleg plasmid construct (pET28aDepB Rleg ) was used as a template with single point mutation oligonucleotide primers (Integrated DNA Technologies) and Q5 High-Fidelity DNA Polymerase (New England Biolabs). All variant constructs were confirmed by a double restriction digest (NdeI/HindIII) followed by sequence analysis conducted at (AFL) Laboratory Services Division at the University of Guelph. All primers used for mutagenesis are listed in Supplementary Table S2.

SSN construction and cluster analysis.
The AKR SSN was constructed as previously described 69 . The master list of AKR sequences was downloaded from UniProt using the PFAM identifier (PF00248). In total, 318,574 sequences were retrieved, and the dataset was reduced by iteratively clustering the sequences using the H-CD-HIT server 70 . Sequences were then filtered based on an amino acid length cutoff between 320-350 to remove fragments and AKRs part of large multi-domain complexes. Finally, curated sequences from the AKR Database (https:// hosti ng. med. upenn. edu/ akr/) and the PDB were also added to the list of sequences, and any duplicates were removed by submission to CD-HIT with the sequence identity cut-off parameter set at 100%. An in-house BLAST search against this finalized database was then conducted at varying E-values and the resulting network was visualized using Cytoscape 71 . An optimal stringency threshold was selected to minimize the extensive fragmentation of the network.
Differential scanning fluorimetry. The thermal stability of wild-type DepB Rleg and variants were assessed using SYPRO Orange while fluorescence was monitored with a StepOnePlus Real-Time PCR system (Applied Biosystems, Foster City, U.S.A.). Experiments were carried out in triplicate in 100 mM HEPES pH 7.5 with a final protein concentration of 0.3 mg/mL and 1X stock of 5000X Sypro Orange as previously described 72 (See  Supplementary Table 3).
Tryptophan quenching assays. Dissociation constants (K d ) of DepB Rleg and each coenzyme was determined using a tryptophan fluorescence-quenching assay with the PTI Fluorimeter (with FelixGX software). Excitation and emission wavelengths of 295 nm and 335 nm respectively were used while slit widths were fixed at 4 nm. Titrations were performed in 20 mM HEPES pH 7.5 with 120 μg of protein. The absorbance of only protein and buffer was measured before the assay to minimize any inner filter effects. All assays were conducted in triplicate. The change in fluorescence (ΔF) relative to the control (protein and buffer) was determined and then plotted against the concentration of each respective coenzyme. The data were fitted to the following equation 73 by nonlinear regression using GraphPad Prism Version 8 software to determine the K d : F is the protein fluorescence intensity at varying concentrations of either NADPH or NADH represented by [L], and F 0 is the fluorescence intensity of the protein in the absence of coenzyme.
In vitro enzyme assays. For  www.nature.com/scientificreports/ ing conditions, for this reason, reactions were monitored at 380 nm due to the high absorbance of NADPH at 340 nm. Reactions were conducted in triplicate and initiated upon the addition of the enzyme ranging between 6 and 787 µg. One unit of enzyme is defined as the amount of enzyme required to oxidize 1 µmol of NADPH per minute. Each assay was monitored for a time period of 10 min. Stock solutions of substrates were prepared in dimethyl sulfoxide (DMSO) and the amount of substrate added to the cuvette was adjusted to ensure that DMSO concentrations were about 10% of the reaction volume to minimize inhibition of enzyme activity (no more than 9% reduction in enzyme activity).
HPLC analysis. Samples and standards were analyzed using an HPLC system (Agilent Technology 1200 Series, Palo Alto, CA, USA) equipped with a quaternary pump, an inline degasser, and a diode array detector set at 218 nm. A Phenomenex® 4 µ Jupiter Proteo 90A (250 × 4.6 mm) with a C18 guard column (Torrance, CA, USA) was used for the separation. DON, 3-keto-DON, and 3-epi-DON were eluted using a binary mobile phase set at a flow rate of 1.0 mL/min. The composition of the mobile phase was acetonitrile: water (10:90) for DON/3epi-DON and acetonitrile:water (20:80) for 3-keto-DON. The injection volume was 10 μL.

Data availability
Coordinates for the DepB Rleg structure have been deposited in the Worldwide Protein Data Bank (wwPDB) with the PDB accession code: 7UTF. Data from this study can be requested from the corresponding author Stephen YK Seah.