Multiple implications of an active site phenylalanine in the catalysis of aryl-alcohol oxidase

Aryl-alcohol oxidase (AAO) has demonstrated to be an enzyme with a bright future ahead due to its biotechnological potential in deracemisation of chiral compounds, production of bioplastic precursors and other reactions of interest. Expanding our understanding on the AAO reaction mechanisms, through the investigation of its structure-function relationships, is crucial for its exploitation as an industrial biocatalyst. In this regard, previous computational studies suggested an active role for AAO Phe397 at the active-site entrance. This residue is located in a loop that partially covers the access to the cofactor forming a bottleneck together with two other aromatic residues. Kinetic and affinity spectroscopic studies, complemented with computational simulations using the recently developed adaptive-PELE technology, reveal that the Phe397 residue is important for product release and to help the substrates attain a catalytically relevant position within the active-site cavity. Moreover, removal of aromaticity at the 397 position impairs the oxygen-reduction activity of the enzyme. Experimental and computational findings agree very well in the timing of product release from AAO, and the simulations help to understand the experimental results. This highlights the potential of adaptive-PELE to provide answers to the questions raised by the empirical results in the study of enzyme mechanisms.

Thus, F397Y, F397L and F397W were 1.4-, 1.6-, and 2-fold more efficient using O 2 as an electron acceptor than the native protein. Finally, steady-state constants calculated from anisaldehyde release (as in previous experiments) were compared with those obtained from H 2 O 2 release (in additional kinetic analyses under atmospheric O 2 saturation). In all variants, k cat and K m(Al) tended to be almost identical using both approaches (Supplementary  Table S1), as it had been previously reported for native AAO 23 .
Rapid kinetics of the two half-reactions for the Phe397 variants. In the light of the above results, the reductive and oxidative half-reactions of the F397 variants were analyzed to unveil the rate-limiting step during catalysis. The spectra collected during the reductive half-reactions of all the variants indicated an essentially irreversible two-electron reduction of the flavin, in agreement with the previously reported hydride transfer reaction for the native AAO 16 . Global analyses of the spectral evolution were fitted to a one-step model (A → B) in all cases (Fig. 3). The values of the observed rate constants (k obs ) at different substrate concentrations exhibited a hyperbolic dependence on the alcohol concentration ( Supplementary Fig. S2) that allowed the determination of the reduction rate constant (k red ) and the dissociation constant (K d(Al) ) upon fitting to equation (3). For native AAO (spectral changes not shown in Fig. 3) and the F397A and F397L variants, the k red values (Table 2) were of the same range of the previously determined k cat values (Table 1) indicating that the reductive half-reaction is the rate-limiting step in catalysis. Nevertheless, variants F397Y and F397W showed k red values 3-and 2-fold higher than the respective turnover rates, suggesting that reductive half-reaction is not the limiting step in these variants. For all the F397 variants, K d(Al) values were similar to the K m(Al) estimated under steady-state conditions.
The oxidative half-reactions of the four Phe397 variants, and native AAO (not shown), fitted two-step model equations (A → B → C) describing a biphasic pattern (Fig. 4) where the first phase accounts for more than      Table 2. Transient-state kinetic constants for the reductive and oxidative half-reactions of AAO and its Phe397 variants. The constants were measured using stopped-flow rapid spectrophotometry in 50 mM sodium phosphate (pH 6.0) at 12 °C under anaerobic conditions. 1 The F397W constants for the first phase of the oxidative halfreaction show a hyperbolic dependence on O 2 concentration (in contrast to the other variants) with k ox /K d(ox) and k ox values of 689 ± 92 mM −1 s −1 and 156 ± 12 s −1 respectively, estimated from fit to equation (5). Means and standard deviations estimated from the fits to equations (3), (4) and (5). All kinetics were measured by triplicates. The second phase for all variants was too slow to be relevant for catalysis (k obs2 < 7 s −1 ), although it was independent of O 2 concentration for the F397Y and F397W variants and dependent on it in the case of the F397L and F397A variants ( Supplementary Fig. S3A, inset). The O 2 -independent slow phase in native AAO reoxidation had been previously attributed to the presence of damaged protein ensuing the stopped-flow experiments 24 .
Studies on AAO:p-anisic acid complex formation and dissociation. The differences between the k cat and k red values for both F397Y and F397W encouraged us to investigate whether product release has an effect on turnover. The study could not be carried out with p-anisaldehyde because the formation of the enzyme-aldehyde complex is too rapid, which prevents its detection by the stopped flow equipment. Therefore, this study ( Fig. 5) was performed with the final product, p-anisic acid, whose crystallographic complex with AAO was recently solved (PDB 5OC1) 17 . The rate constants for complex formation (k for ) and dissociation (k dis ) of native AAO and the F397Y and F397W variants are shown in Table 3. The k for for F397Y and F397W were 2-fold and 2 orders of magnitude lower than that of the native enzyme, respectively. Differences in k dis were also observed, since the k dis values for these variants were at least 10-fold slower than that of the native AAO. The fast diffusion of the ligand in and out the active site of the F397L and F397A variants most likely impeded estimation of these parameters. The K d(Ac) values calculated from the corresponding k dis/ k for ratios for native AAO and the F397Y and F397W variants (150, 31 and 271 µM respectively) agree well with the values obtained by differential spectrophotometry (Supplementary Fig. S4 and Table 3). F397Y showed the smallest K d(Ac) , which means that it is the variant that more tightly binds p-anisic acid. On the contrary, the K d(Ac) for the F397L and F397A variants could not be estimated by differential spectrophotometry since saturation could not be attained in agreement with the above results.
AAO molecular dynamics and ligand diffusion simulations. MD simulations were carried out to study the dynamical behaviour of the five protein systems (native enzyme and its Phe397 variants). Loops are the secondary structures that most likely adopt different conformational structures. Hence, the motions of the loops Gln395-Thr406 and Ser89-Met95, where two of the main gate residues -Phe397 and Tyr92-are located, were investigated. However, the root-mean-square-deviation values obtained, as a function of time, were low (data not shown) indicating that there are only moderate displacements of these secondary structures along the dynamics of the protein. Despite the small loop movements, a closer look at the distances between the gate residues indicates that the F397Y and F397W substitutions induce significantly shorter values ( Supplementary Fig. S5A). Moreover, inspection of the protein structures shows that the side chains of the F397Y and F397W variants establish hydrogen bonds with Tyr92 with a frequency of 20% and 5%, respectively ( Supplementary Fig. S5B), while the aliphatic variants and the native enzyme cannot develop such hydrogen bonds resulting in larger distances between gate residues. Altogether, the MD data suggests local constraints induced by the tyrosine and tryptophan substitutions in the gate region.
The diffusion of products -p-anisaldehyde and p-anisic acid-and reactants -p-methoxybenzyl alcohol and O 2 -was also investigated with adaptive-PELE to explain the changes in kinetics observed for the mutants. The trajectory analysis of the acidic product in the aromatic variants clearly indicates a stabilization of the local   Table 3. Dissociation constants, and formation/dissociation rates of p-anisic acid complexes. The dissociation constants (K d(Ac) ) of p-anisic acid complexes with AAO and its Phe397 variants were measured by differential spectroscopy, while the complex formation (k for ) and dissociation (k dis ) rates were measured by transient-state kinetics spectroscopy. All experiments were performed in 50 mM sodium phosphate (pH 6.0) at 12 °C. nd, not determined (too fast complex formation and dissociation). Means and standard deviations from the fit to equations (7) or (6). All data were measured by triplicates.
Scientific REPORTS | (2018) 8:8121 | DOI:10.1038/s41598-018-26445-x minima at the gate, before ligand release, which does not take place during release of the aldehyde products. This energy minimum involves interaction between the carboxylic group and three amino acids, Ser393, Gln395 and Ser411, placed inside the catalytic cavity. The aromatic variants required in average ~3-fold and ~7-fold more simulation steps to release the aldehyde and acid, respectively, than the aliphatic ones (Fig. 6). Moreover, the number of PELE steps required to release both products also tends to be higher in the case of the aromatic mutants than for the native enzyme, especially for anisic acid due to the interactions of the carboxylic group with the three residues mentioned above. Furthermore, the distances among any of the atoms of the O 2 substrate and both the C4a locus of reduced FAD and the Nε 2 of His502 (acting as a catalytic base in the reductive half reaction) were calculated for all the variants. Results suggest that the catalytic distances are similar in the native enzyme and variants, except for F397W that shows higher population at favourable -that is, shorter-distances (Fig. 7).
Finally, the energetic profiles of the p-methoxybenzyl alcohol catalytic distances (hydroxyl-to-His502 and proR-H-to-flavin) in all protein systems were studied, as a measure of the binding effectiveness of the alcohol in the active centre. The energy landscape shows an additional energetic favourable region only for the F397W. In this new minimum, the side chain of the tryptophan interacts via hydrogen bond to the hydroxyl group of the alcohol substrate, resulting in an inefficient catalytic pose ( Supplementary Fig. S6). This is in agreement with the experimental values, where the K m(Al) and K d(Al) are almost 11-fold higher for the F397W variant compared to the native protein.

Discussion
Phe397 is situated at a loop characteristic of the AAO family, in a region prone to harbouring insertions and deletions amongst the members of the GMC superfamily. Representatives of this superfamily are monomeric or multimeric proteins, with the entrance to their active site partially covered by the adjacent monomer in the second case. Given their monomeric nature, AAOs and a few other GMCs have developed loop structures to control diffusion of molecules into their active sites. Among them, choline oxidase 25 , cholesterol oxidase 26 and cellobiose dehydrogenase 27 are notable examples, whose insertions are even longer than the AAO insertion. The latter insertion partially forms the 395-406 loop in P. eryngii AAO and encloses the active-site cavity from the outer environment (Fig. 1) 18 . Previous computational studies hinted that Phe397 oscillates with the substrate as a mechanism of gating 19 , similarly to what has been reported for phenylalanine residues in P450 enzymes 28 . A homologous phenylalanine residue is conserved in 50% of all the putative AAO sequences from basidiomycete genomes available at JGI (https://jgi.doe.gov). The multiple roles of this phenylalanine residue in catalysis, as revealed by studies on the P. eryngii model AAO, are discussed below.
Phe397 favours the correct positioning of the alcohol and its oxidation at the active site, as evidenced by the steep decrease of catalytic efficiency in the mutated variants. Such effect is most noticeable in F397W, which shows the lowest k cat /K m(Al) value due to: i) the constrained distance between gate residues, and ii) the formation of a H bond with its reactive hydroxyl group that disrupts the substrate's catalytic poses. The introduction of a bulkier residue at the Phe501 and Tyr92 positions of AAO also produced similar steric hindrances for catalysis 13,14,17 . Remarkably, the drop of the catalytic efficiency in F397Y is due to its small turnover rates, as affinity for alcohol (K m(Al) and K d(Al) ) remains invariable in this case. Furthermore, the reduction rates being higher than turnover in F397Y and F397W, indicates that, unlike native enzyme, the reductive half-reaction is not the rate-determining step in these two variants, as discussed below.
Phe397 also plays a role in flavin reoxidation by either compressing the active site to reduce the free diffusion of O 2 , or altering its redox environment. Evidence comes from the low oxidation rates estimated for the aliphatic mutants, in contrast to the ones obtained for the aromatic variants and the native AAO, which possess bulkier b a residues that constrain the catalytic pocket. However, the oxidative half-reaction does not limit the catalytic cycle in the aliphatic variants, as k cat and k red values are similar and the constants measured as H 2 O 2 release are identical as those measured by aldehyde formation. Alteration of the FAD environment and electrochemistry might be involved in the decreased efficiency for oxidation of the F397A cofactor. Regarding the F397Y and F397W variants, they show high app k ox , similar to the native protein. In the case of F397W variant, the obtained k ox and k red are similar indicating that the reductive and oxidative half-reactions are almost balanced, which agrees with its redox state during turnover. Computational simulations of F397W reoxidation revealed that withdrawal of His502 due to the presence of the tryptophan increases the ability of this variant for properly positioning the oxygen molecule during flavin reoxidation, bringing O 2 and C4a closer than in the native enzyme (Fig. 7). However, the introduction of this bulkier residue could be hindering the oxygen diffusion into the active-site 19 and thus causing the unusual saturation behaviour observed for this variant with increasing oxygen concentration. This saturation profile has been observed in cholesterol oxidase from Brevibacterium sterolicum related to the interconversion between open and closed channel enzyme conformations. This event regulated the oxygen diffusion and constituted the rate-limiting step preceding flavin reoxidation 29,30 . Regarding F397A and F397L, they display slower reoxidation rates probably due to the absence of an aromatic residue in that position. Moreover, both variants display slow phases that are dependent on O 2 concentration, that could reveal two different kinetic processes, the H and H + transfers recently reported to take place in separate kinetic steps in native AAO 24 .
Finally, Phe397 also plays a role in controlling the product release from the active site, with this catalytic step limiting the turnover of the F397Y and F397W variants, while the F397A and F397L variants (and native AAO) let the substrate and product diffuse easily in and out of the active site. Furthermore, ligand migration studies with both p-anisic acid and p-anisaldehyde indicate that product release is delayed due to the reduced gated space -which forces the aromatic side chains to be displaced-and the hydrogen bond between the tyrosine or tryptophan and Tyr92 in the aromatic variants. The native enzyme also requires longer times for product release because its side chain must move to let it out, although there is no hydrogen bonding between the Phe397 and Tyr92. This is in agreement with the 10-fold lower k dis for the aromatic variants than for native AAO. Flavoenzymes with catalytic cycles limited or partially limited by the product release have been reported, such as D-amino acid oxidase 31 , cyclohexanone monooxygenase 32 or nitroalkane oxidase 33 . The case of amadoriase I is also relevant, since its catalysis is partially limited by the product release, although the oxidation of the enzyme takes place through an ordered ternary complex of enzyme, product and O 2 34 .
In conclusion, Phe397 plays a central role in the catalysis of P. eryngii AAO, which is reinforced by its high conservation among other sequences of characterised and putative AAOs from fungal genomes. As drawn from the experimental and computational results, this residue supports both half-reactions of the enzyme. During the reductive reaction, it helps the substrate attain a catalytically relevant position and facilitates product release from the active site. Analysis of the oxidative reaction suggests that the presence of an aromatic residue at this position is important to modulate the flavin environment and to reduce the space in the active site enabling the enzyme to reduce O 2 to H 2 O 2 . Experimental and computational results complement each other well and, thus, highlight the suitability of the recently developed adaptive-PELE 22 for the study of protein binding to different ligand types.

Methods
Reagents. Glucose oxidase type VII from Aspergillus niger, glucose, p-methoxybenzyl alcohol and p-anisic acid were purchased from Sigma-Aldrich.
Enzyme expression and purification. Native AAO from P. eryngii (GenBank accession number AF064069) and its mutated variants F397A, F397Y, F397W and F397L were heterologously expressed in E. coli W3110 as recombinant proteins using the pFLAG1 vector. The above-mentioned AAO variants were prepared using the QuickChange ® site-directed mutagenesis kit.
Molar absorbance coefficients of the variants were calculated by heat denaturation of the proteins and determination of the FAD released. The coefficients for native AAO and the F397Y, Initial rates were obtained from the linear phase of the aldehyde production and were calculated as the change in absorbance over time. Kinetic constants were estimated by fitting the k obs to equation (1) describing a ping-pong mechanism: where ν stands for the initial velocity, e is the enzyme concentration, k cat is the catalytic constant, A stands for the alcohol concentration, B represents O 2 concentration and K m(Al) and K m(ox) are the Michaelis constants for p-methoxybenzyl alcohol and O 2 , respectively. Kinetic constants estimated as p-anisaldehyde and H 2 O 2 release were compared under atmospheric O 2 saturation conditions, at 25 °C. p-Anisaldehyde was measured as detailed above, while H 2 O 2 release was estimated by coupling the reaction of a horseradish peroxidase (6 U·mL −1 ) and AmplexRed ® (60 nM), which uses the H 2 O 2 produced by AAO to give coloured resorufin (Δε 563 = 52000 M −1 ·cm −1 ). Catalytic constants were estimated by fitting to a Michaelis-Menten equation for one substrate in both cases (equation (2)): Studies on the reductive half-reaction were performed upon mixing AAO with increasing concentrations of p-methoxybenzyl alcohol (0.018-0.6 mM) under anaerobic conditions. The stopped-flow apparatus was made anaerobic by flushing sodium dithionite through the system, which was then rinsed out with O 2 -free buffer. All buffers, substrates and the enzyme were poured into glass tonometers that were subsequently subjected to 20-25 cycles To ensure anaerobiosis, glucose (10 mM) and glucose oxidase (10 U·mL −1 ) were added after some vacuum-Ar cycles. Measurements were recorded using the PDA detector, in 50 mM sodium phosphate (pH 6.0) at 12 °C. Observed rate constants (k obs ) were calculated by global fitting of the spectra with Pro-K software to a single exponential equation. Those averaged k obs at each substrate concentration were then non-linearly fitted to equation (3), describing hyperbolic substrate dependence of k obs : where k red and k rev represent the reduction rate constant at infinite substrate concentration and its reverse reaction, respectively; A stands for the alcohol concentration; and K d is the dissociation constant. The oxidative half-reaction was investigated by mixing reduced AAO with increasing O 2 concentrations. Procedures were as explained above for the reductive half-reaction, except that, in this case, AAO and glucose were put into a tonometer bearing a side-arm, where p-methoxybenzyl alcohol (1.3-fold the concentration of AAO) was poured along with glucose oxidase. After the required vacuum-Ar cycles, enzyme and substrate were mixed before being mounted onto the stopped-flow equipment. Reactions were measured with both the PDA and the monochromator detectors at 12 °C. k obs were obtained by either global fitting of the spectra or fitting the monochromator traces to exponential equations describing two-step and three-step processes. Fitting averaged k obs either to equation (4) that describes a linear dependence on O 2 concentration or equation (5) that describes hyperbolic saturation with increasing O 2 concentration allowed the estimation of the apparent second-order rate constant for reoxidation ( app k ox ) and the first-order rate (k ox ) and second-order constants for reoxidation (k ox /K d(ox) ), respectively: Estimation of the rates of the AAO:p-anisic acid complex formation and dissociation were performed by analyzing spectral changes upon mixing enzyme (~10 µM) with different concentrations of the ligand (0.04-2 mM) at 12 °C. Data were globally fitted to an equation describing a one-step process. The obtained k obs were linearly depended on the ligand concentration and were, thus, fitted to equation (6): obs f or dis in which k for stands for the second-order rate constant for the complex formation; L represents the ligand concentration, and k dis is the rate constant for the complex dissociation.
Spectral characterization of the AAO-p-anisic acid complex. The affinity of the native AAO and its variants for p-anisic acid was assessed by titrating the enzyme with increasing concentrations of the ligand in 50 mM sodium phosphate (pH 6.0) at 12 °C. Spectral changes were recorded using a spectrophotometer and their magnitude upon complex formation was fitted to equation (7), which accounts for a 1:1 stoichiometry, as a function of p-anisic acid concentration: in which ΔA accounts for the observed change in absorbance; Δε represents the maximal absorption difference in each of the spectra; K d is the dissociation constant; and E and L, the enzyme and p-anisic acid concentrations.

Molecular dynamics (MD) studies.
The protonation states of the titratable amino acids present at the AAO crystal structure (PDB 3FIM) were set up using PROPKA 36 of Protein Preparation Wizard 37 and checked with the H++ server 38 . The selected mutations (F397W, F397Y, F397A and F397L) were introduced and all 3D structures were minimized with Prime 39,40 . The overall systems were solvated and neutralized using Desmond 41 with the SPC water model and 0.15 M of NaCl. 50 ns of MD simulation were performed with Desmond 41 for each protein structure using the default Desmond's protocol. The temperature was regulated with the Nosé-Hoover chain thermostat 42-44 with a relaxation time of 1 ps. The pressure was controlled with the Martyna-Tobias-Klein barostat 45 with isotropic coupling and a relaxation time of 2 ps. RESPA integrator 45,46 was used with bonded, near and far time steps of 2.0, 2.0 and 6.0 ps, respectively. Furthermore, a 9 Å cut-off was employed for nonbonded interactions with the smooth particle mesh Ewald method 47 . Ligand diffusion. The minimized models described above were also used to study the ligand diffusion. The acid product was manually docked inside the cavity, next to Phe397 and Tyr92, while reactants were docked in their catalytic position. The FAD cofactor was protonated to its semiquinone state for the O 2 simulations, while it remained in its quinone state for the acid, aldehyde and alcohol diffusion studies.
The FAD cofactor was optimized with QM/MM at the M06/6-31G* level of theory using Qsite 48 and the atomic charges were obtained from the electrostatic potential (ESP). Ligand geometries were optimized at the Scientific REPORTS | (2018) 8:8121 | DOI:10.1038/s41598-018-26445-x same level of theory in gas phase and re-optimized using the PBF implicit solvent with the Jaguar software 49 . The ligands were parameterized according to the OPLS 2005 force field, maintaining the ESP charges, and a rotamer library was build using Macromodel 37 .
Ligand diffusion was modelled with the recent adaptive-PELE technique 22 , using an anisotropic network model (ANM) backbone perturbation 50 applied to all Cα atoms, side chains re-sampling within 6 Å of the ligands, and a full energy minimization in each PELE step. The products were randomly translated and rotated within a spherical box of 22 Å of its initial position. The O 2 and the alcohol reactants were randomly roto-translated within a spherical box of 3 Å and 7 Å of its centre of mass initial position, respectively, to inquire about their poses in the active centre. Adaptive-PELE simulations involved ~50 epochs of 4 PELE steps. All simulations were performed using 32 processors. The O 2 simulations were executed during 72 h, while the other simulations were performed during 23 hours.
Data availability statement. All data generated or analysed during this study are included in this published article (and its Supplementary Information files).