Structure of the imine reductase from Ajellomyces dermatitidis in three crystal forms

The structure of the imine reductase from A. dermatitidis is presented in three crystal forms, each of which provides information on conformational dynamics and cofactor and substrate binding within the enzyme.

These and other RedAms from fungi have been applied in a number of preparative imine reductions and reductive aminations (Mangas-Sanchez et al., 2020;Ramsden et al., 2019; Gonza ´lez-Martı ´nez et al., 2020; Zhang et al., 2021).One of them, AdRedAm from Ajellomyces dermatitidis, has been reported to be more stable than AspRedAm (Zachos et al., 2021) and thus perhaps more suitable for process applications.Indeed, AdRedAm has already been applied to the asymmetric reduction of imine substrates, including dibenzazepines such as 4 (Fig. 1b; France et al., 2017) and pyrrolidines such as 6 (Fig. 1c; Costa et al., 2018), and also the reductive amination of cyclohexanone (1) with various amines (Aleku et al., 2017).In addition, AdRedAm has been used in a biocatalytic flow system for the reductive amination of hydrocinnamaldehyde (8) with allylamine (9) (Fig. 1d; Finnigan et al., 2020).Given the enduring interest in AdRedAm as a catalyst, we have determined its structure using X-ray crystallography to assist in the interpretation of experimental biotransformation results and also for structure-guided engineering.The structure, which has been obtained in three crystal forms, sheds light on the conformational dynamics of the enzyme, cofactor binding, and also the molecular determinants of stereoselectivity in the transformations of the substrate 2,2-difluoroacetophenone.

Macromolecule production
Cloning and expression of the gene encoding AdRedAm and purification of the protein have been reported previously (Aleku et al., 2017).The protein, which was purified using nickel-affinity chromatography followed by size-exclusion chromatography, was concentrated to 40 mg ml À1 using centrifugal concentrators with a molecular-weight cutoff of 10 kDa.

Crystallization
Concentrated protein at 40 mg ml À1 was subjected to crystallization trials in a range of commercial screens in 96-well plate format, using 150 nl protein solution and 150 nl precipitant solution and a Mosquito robot (SPT Labtech).Crystals for data set 1 were harvested from conditions consisting of 0.1 M PCTP buffer (sodium propionate, sodium cacodylate trihydrate, bis-Tris propane) pH 4.0, 25% PEG 1500 with the protein pre-complexed with 2 mM NADPH 4 .Crystals for data set 2 were recovered from conditions consisting of 0.1 M MES buffer pH 6.0, 0.2 M MgCl 2 , 20% PEG 6000 with the protein pre-complexed with 2 mM NADP + .Crystals for data set 3 were grown in 100 mM phosphate buffer pH 8.6, 0.2 M MgCl 2 , 20%(w/v) PEG 3350 with the protein pre-complexed with 2 mM NADP + and 10 mM 2,2-difluoroacetophenone.Crystallization information is summarized in Table 1.

Data collection and processing
Data were collected on beamlines I03 and I04-1 at the Diamond Light Source (DLS) and were processed and integrated using XDS (Kabsch, 2010) and scaled using SCALA (Evans, 2006) within the xia2 (Winter, 2010) processing system.Data-collection statistics can be found in Table 2.

Structure solution and refinement
Crystals furnishing data set 1 were obtained in space group P3 1 21 with two molecules in the asymmetric unit constituting one dimer.The structure was solved with MOLREP (Vagin & Teplyakov, 2010) using the structure of the imine reductase from A. oryzae (54% sequence identity; PDB entry 5g6r; Aleku et al., 2017) as the molecular-replacement model.The structure was solved using iterative cycles of Coot (Emsley et al., 2010) and REFMAC (Murshudov et al., 2011).After building the protein backbone, side chains and water molecules, residual density was present in the omit map in one active site.This could be modelled and refined as the inactive cofactor molecule NADP 4 .Data sets 2 and 3, in space groups C2 1 and P3 1 21, respectively, were processed, built and refined in a similar fashion to data set 1, yielding structures with nine and one monomers in the asymmetric unit, respectively.Data set 2 featured density consistent with nine molecules of NADP + in nine active sites; data set 3 featured density consistent with one molecule of NADP + in its active site, and also residual density adjacent to the nicotinamide ring of the cofactor that was successfully modelled and refined as 2,2-difluoroacetophenone (15).The Ramachandran plot for the structure from data set 1 revealed 99.6% of residues in favoured regions, with 0.4% outliers.The corresponding figures for data sets 2 and 3 were also 99.6% of residues in favoured regions with 0.4% outliers.Refinement statistics for the structures can be found in Table 3. Coordinates and structure-factor files have been deposited in the Protein Data Bank (PDB) for data sets 1, 2 and 3 with accession codes 8ozw, 8p2j and 8ozv, respectively.

Results and discussion
The gene encoding AdRedAm was codon-optimized for expression in Escherichia coli and was expressed in E. coli BL21 (DE3) cells.The enzyme was purified by nickel-affinity (Ni-NTA) chromatography and size-exclusion chromatography (SEC) using previously described methods (Aleku et al., 2017), and the pure protein was concentrated to 40 mg ml À1 for crystallization.The enzyme crystallized in three forms.The first form (data set 1), which belonged to space group P3 1 21 and was refined to 2.01 A ˚resolution, contained two molecules in the asymmetric unit representing one dimer.The second form (data set 2), which belonged to space group C2 1 and was refined to 1.73 A ˚resolution, had nine molecules in the asymmetric unit, representing four and a half dimers.The third form (data set 3) belonged to space group P3 1 21 and was refined to 1.52 A ˚resoution.This structure featured only one molecule as a half-dimer in the asymmetric  unit.Data-collection and refinement statistics can be found in Tables 2 and 3. Using data set 1, the structure of AdRedAm was solved using the structure of AspRedAm (sequence identity of 54%) as a model.The structure was built and refined using iterative cycles of building in Coot and refinement in REFMAC.This data set yielded two molecules in the asymmetric unit as a dimer, which is the canonical form of previously described imine reductases (Aleku et al., 2016(Aleku et al., , 2017;;Rodrı ´guez-Mata et al., 2013;Huber et al., 2014;Man et al., 2015).Electron density for the backbone atoms was largely complete throughout the length of both chains from Ala2 to Lys288.The monomer of AdRedAm was compared with existing structures using the DALI server (Holm, 2022) Lenz et al., 2018).AdRedAm adopts the known IRED fold, with an N-terminal NADP + -binding domain (Ala2-Val162) connected by a long inter-domain helix (Gly163-Ser192) to a C-terminal helical bundle (Ala193-Lys288) (Fig. 2a).
The two AdRedAm monomers associate to form a dimer in which reciprocal domain sharing results in the formation of a large active-site cleft between the N-terminal domain of one monomer and the C-terminal domain of its partner.In the structure from data set 1, following building of the protein and water molecules one of the active sites featured clear density in the omit map that could be modelled as the redox-inactive cofactor NADPH 4 , with which the protein had been complexed (Fig. 2b).Interestingly, the other active site featured no cofactor density.A comparison of monomers with and without NADPH 4 showed that the side chain of Asn94 was rotated approximately 180 to accommodate the ribose of the cofactor, but that the orientation of the other side chains was largely conserved.Despite a sequence homology of only 54%, the active site of AdRedAm is highly conserved compared with that of AspRedAm (for example PDB entry 5g6r; Aleku et al., 2017), with Asp169 and Asn94, which are thought to have roles in amine activation in the reductive aminase mechanism (Sharma et al., 2018), and Tyr177, which is implicated in ketone activation, superimposing well with the equivalent residues in AspRedAm (Fig. 2b).In addition, hydrophobic residues in AspRedAm that were shown to form a binding pocket for ketone substrates (Aleku et al., 2017), including Leu173 and Met176 from the interdomain helix and Trp208 and Met212 from the C-terminal domain of the partner monomer (monomer B in Fig. 2b), are also conserved, with Met237 and Gln238 at the front of the active site.
The sequences and structures are less conserved in other regions.In the N-terminal domain of AdRedAm there are several differences between Ser60 and His110 when compared with AspRedAm, including Lys70 (Asn69 in AspRedAm), which forms an ionic interaction with Glu102 (Lys101) in AdRedAm.There are also differences at Trp84 (Leu83) and Trp106 (Phe105), residues that both project into the hydrophobic core of the N-terminal domain that also includes Leu77 (Leu76) and Ile89 (Ile88), which are both conserved between the enzymes.In addition, the hydroxyl group of Thr103 in AdRedAm, which is replaced by Leu102 in AspRedAm, forms a new hydrogen bond to the backbone carbonyl group of His99 (Gln98).
However, the major difference in tertiary structure between the AdRedAm and AspRedAm monomers is a shorter loop of 11 residues between Leu189 and Gly199 (LVQSANIPAAG) in the C-terminal helical bundle in AdRedAm, which was 14 residues (LIKSGQDTSTTATG) between Leu189 and Gly202 in AspRedAm.This loop, which is at the dimer interface, positions Ala193 and Ile195 in AdRedAm for hydrophobic interactions with Leu164 and Leu167 in the partner monomer.Just downstream of this loop, in the region between Val190 and Thr210, Val190 (Ile190), Ile195 (Asp195), Leu189 (Leu189) and Phe200 (Leu203) also make hydrophobic interactions with Ala171 and Leu172 of the neighbouring monomer at the dimer interface.Recent studies of the engineering of IREDs for improved stability using random mutagenesis suggest that mutations that enhance interactions, including hydrophobic forces, at the dimer interface were significant in producing variants with greater process stability (Schober et al., 2019;Kumar et al., 2021;Ma et al., 2021) and, indeed, previous research has suggested that AdRedAm is more stable than AspRedAm (Zachos et al., 2021).A comparative analysis of AspRedAm and AdRedAm using PISA (Krissinel & Henrick, 2007) suggests that AdRedAm should be more stable, with a monomer-monomer interfacial interaction of 4107 A ˚2 versus 3918 A ˚2.This would suggest a free energy of dissociation of À84.8 kJ mol À1 for AdRedAm versus À78.3 kJ mol À1 for AspRedAm and thus greater stability, as observed experimentally.
The AdRedAm dimer observed in the structure from data set 1 was already instructive in showing two possible states of the monomer in which the non-natural cofactor molecule was either absent or bound.Variations were also readily observed amongst the four dimers present in the structure from data set 2, which was obtained from crystals that grew in space group C2 1 and featured nine molecules in the asymmetric unit.In this structure, once again, the vast majority of backbone atoms could be modelled in subunits A-I from, in some cases, the leucine and phenylalanine residues in the purification tag at positions À5 and À4 through to Lys288.The exception was chain H, in which electron density for Gly229-Gly232 was poor and could not be modelled.In the case of data set 2, all monomers featured electron density in the omit maps that could be modelled as the cofactor NADP + .Despite the presence of the cofactor in all monomers, the difference in the conformation of some monomers was pronounced.In the most divergent examples, monomers E and I exhibited a hinge movement between the N-terminal and C-terminal domains of 14.8 as calculated using the DynDom server (Fig. 2c; Lee et al., 2003), with the hinge movement centred around the pendant aspartate residue Asp169.This was comparable to the most pronounced difference in conformation observed in multiple dimer structures of AtRedAm (Sharma et al., 2018).The overall effect of the hinge movement is to close the active site with respect to the cofactor, presumably to provide the hydrophobic environment that is required to favour greater stability of the transient imine intermediate.
The structure that resulted from data set 3 was more unusual, although not unique amongst IRED structures, in featuring only one monomer (for example PDB entry 6skx; Mangas-Sanchez et al., 2020) or a half-dimer within the asymmetric unit.However, this data set provides significant further information on ligand recognition within AdRedAm.This structure was obtained from crystals that had been co-crystallized with the ligand 2,2-difluoroacetophenone (15; Fig. 3) in an effort to shed light on the mixed selectivity of AdRedAm towards fluorinated acetophenones, for which both alcohol products and reductive amination products are observed (Gonza ´lez-Martı ´nez et al., 2020).Fluorinated acetophenones are unusual amongst IRED substrates as, with the exception of some examples using engineered enzymes (Jia et al., 2021), these ketones are the only examples of carbonyl compounds that undergo significant reduction to the alcohol (Gonza ´lez-Martı ´nez et al., 2020; Lenz et al., 2017).
A previous structure of an IRED from S. roseum reported by our group (SrIRED; PDB entry 5ocm Lenz et al., 2018) was presented in complex with the hydrate 20 of ketone 18 and suggested that the significant disposition towards ketone reduction in the case of 18 may be due not only to the extra activation of the carbonyl C atom, but also to aspects of specific substrate recognition within the active site.The complex showed that the fluorine substituents of 20 made hydrogen-bonding interactions with the O2D hydroxyl of the NADP + ribose, thus drawing the electrophilic C atom of the substrate C O group sufficiently close to the C4 atom of the cofactor for hydride exchange to occur.
For the structure from data set 3, once the protein, water and cofactor atoms had been modelled, significant omit density persisted within the active site that could be modelled as the added ketone ligand 2,2-difluoroacetophenone (15; Fig. 2d) and not as the hydrate form observed for SrIRED with 20.In the complex of AdRedAm with 15, the phenyl ring of the substrate is stacked against the side chain of Met176; the carbonyl group of the ketone is coordinated to the phenolic hydroxyl of the side chain of Tyr177 at a distance of 3.0 A ˚.The F atoms are again positioned 3.2 and 3.3 A ˚from the O2D atom of the ribose sugar, suggesting again that these interactions may be significant in permitting the reduction of the ketone by the cofactor to some extent, but only in the case where F atoms are present.However, the electrophilic C atom is not as ideally placed for hydride exchange as observed for 20 in PDB entry 5ocm, as the distance of the electrophilic C atom from the NADP + atom is suboptimal at 4.5 A ˚.In addition, the ketone 15 presents its re face to the cofactor, which would result in the experimentally observed (S)-alcohol product.The complex illustrates the imperfect binding of the fluorinated ketone 15 for either amine or alcohol production, and also provides a basis for understanding the stereoselectivity of AdRedAm for transformations of substrates in this series.
AdRedAm is a useful biocatalyst for a number of iminereduction and reductive amination reactions.The presented structures of AdRedAm provide new insights into conformational dynamics in this IRED, interactions that confer superior stability and also the basis for steroselectivity and chemoselectivity in the transformation of fluorinated ketones.In addition to these insights, these and other structures of IREDs will serve as valuable platforms for the structure-guided engineering of IREDs for improved process stability.
. Its closest structural homologs were AspRedAm (PDB entry 5g6r; Z-score 34.6; 54% sequence identity; r.m.s.d. of 1.0 A ˚over 289 C atoms; Aleku et al., 2017), Nf RedAm from Neosartorya fumigata (PDB entry 6sle; Z-score 33.8; 50% sequence identity; r.m.s.d. of 1.2 A ˚over 278 C atoms; Mangas-Sanchez et al., 2020) and the imine reductase from Streptosporangium roseum (PDB entry 5ocm; 39% sequence identity; r.m.s.d. of 1.2 A ˚over 282 C atoms; Figure 2 (a) Structure of the AdRedAm dimer from data set 1 in ribbon format, with monomers A and B shown in blue and brown, respectively.(b) Active site of AdRedAm showing the active-site residues, each of which is conserved from AspRedAm (PDB entry 5g6r; Aleku et al., 2017).NADPH 4 is shown in cylinder format with C atoms in grey.(c) Superimposition of monomers E and I of AdRedAm from data set 2 in blue and brown, respectively.(d) Structure of AdRedAm from data set 3. The symmetry neighbour has been incorporated (C atoms in brown) to show the contribution of both monomers in the active site.The ligand 2,2-difluoroacetophenone (15) is shown with C atoms in yellow.Electron density in blue corresponds to the unbiased F o À F c maps at a level of 2.5 obtained before refinement of the ligand atoms, which have been added for clarity.