Reaction of pyranose dehydrogenase from Agaricus meleagris with its carbohydrate substrates

Monomeric Agaricus meleagris pyranose dehydrogenase (Am PDH) belongs to the glucose–methanol–choline family of oxidoreductases. An FAD cofactor is covalently tethered to His103 of the enzyme. Am PDH can double oxidize various mono‐ and oligosaccharides at different positions (C1 to C4). To study the structure/function relationship of selected active‐site residues of Am PDH pertaining to substrate (carbohydrate) turnover in more detail, several active‐site variants were generated, heterologously expressed in Pichia pastoris, and characterized by biochemical, biophysical and computational means. The crystal structure of Am PDH shows two active‐site histidines, both of which could take on the role as the catalytic base in the reductive half‐reaction. Steady‐state kinetics revealed that His512 is the only catalytic base because H512A showed a reduction in (k cat/KM)glucose by a factor of 105, whereas this catalytic efficiency was reduced by two or three orders of magnitude for His556 variants (H556A, H556N). This was further corroborated by transient‐state kinetics, where a comparable decrease in the reductive rate constant was observed for H556A, whereas the rate constant for the oxidative half‐reaction (using benzoquinone as substrate) was increased for H556A compared to recombinant wild‐type Am PDH. Steady‐state kinetics furthermore indicated that Gln392, Tyr510, Val511 and His556 are important for the catalytic efficiency of PDH. Molecular dynamics (MD) simulations and free energy calculations were used to predict d‐glucose oxidation sites, which were validated by GC‐MS measurements. These simulations also suggest that van der Waals interactions are the main driving force for substrate recognition and binding.

Monomeric Agaricus meleagris pyranose dehydrogenase (AmPDH) belongs to the glucose-methanol-choline family of oxidoreductases. An FAD cofactor is covalently tethered to His103 of the enzyme. AmPDH can double oxidize various mono-and oligosaccharides at different positions (C1 to C4). To study the structure/function relationship of selected active-site residues of AmPDH pertaining to substrate (carbohydrate) turnover in more detail, several active-site variants were generated, heterologously expressed in Pichia pastoris, and characterized by biochemical, biophysical and computational means. The crystal structure of AmPDH shows two active-site histidines, both of which could take on the role as the catalytic base in the reductive half-reaction. Steady-state kinetics revealed that His512 is the only catalytic base because H512A showed a reduction in (k cat /K M ) glucose by a factor of 10 5 , whereas this catalytic efficiency was reduced by two or three orders of magnitude for His556 variants (H556A, H556N). This was further corroborated by transient-state kinetics, where a comparable decrease in the reductive rate constant was observed for H556A, whereas the rate constant for the oxidative half-reaction (using benzoquinone as substrate) was increased for H556A compared to recombinant wild-type AmPDH. Steady-state kinetics furthermore indicated that Gln392, Tyr510, Val511 and His556 are important for the catalytic efficiency of PDH. Molecular dynamics (MD) simulations and free energy calculations were used to predict D-glucose oxidation sites, which were

Introduction
The oxidoreductase pyranose dehydrogenase from the litter-decomposing fungus Agaricus meleagris (AmPDH; EC 1.1.99.29; PDB code: 4H7U) is a glycoprotein of approximately 65 kDa carrying a monocovalently linked FAD cofactor [1]. The level of glycosylation depends on the source of PDH, ranging from approximately 7% (natural source A. meleagris) [2] to 30% (overexpressed in Pichia pastoris) [3]. Together with aryl-alcohol oxidase [4], choline oxidase [5], glucose 1-oxidase [6] and pyranose 2-oxidase (POx) [7], amongst others, AmPDH belongs to the structural family of glucose-methanol-choline oxidoreductases [1,8,9]. PDH was initially identified and studied in Agaricus bisporus [10] and Macrolepiota rhacodes [11]; however, PDH from A. meleagris [12] is the best characterized to date, with its crystal structure solved to a resolution of 1. 6 A [13]. In accordance with other flavin-dependent oxidoreductases, PDH shows a reaction mechanism consisting of two half-reactions. In the reductive half-reaction, an electron donor substrate (a sugar substrate) is oxidized, whereas the flavin is reduced. In the ensuing oxidative half-reaction, the reduced flavin is re-oxidized by a second electron-acceptor substrate. AmPDH can oxidize a wide range of different carbohydrates, and D-glucose (GLC) is one of its favoured substrates as judged by the catalytic efficiency [2]. Studies on the oxidized sugar products revealed that PDH is able to double oxidize mono-and oligosaccharides at C1-C4 [12,[14][15][16]. Molecular dynamics (MD) simulations suggested preferred modes of productive binding for different monosaccharide substrates, as well as an oxidation mechanism for GLC [17,18]. PDH shows very poor activity with oxygen in its oxidative half-reaction but readily reduces electron acceptors such as various substituted quinones, redox dyes or chelated metal ions. Previous characterizations of AmPDH have either focused on wild-type AmPDH [2,13] or on identifying variants with increased oxygen reactivity employing a semi-rational approach coupled with high-throughput screening [19,20].
The present study represents the first comprehensive rational investigation of several active-site residues in AmPDH ( Fig. 1) [13,17,19]. Previous computational studies on AmPDH suggested that both active-site histidines (His512 and His556) potentially act as catalytic base, and thus are responsible for C2 and C3 oxidation [17,19], respectively. These studies also showed that residues Gln392, Tyr510 and Val511 might play a role in GLC binding and affect the site of GLC oxidation [17]. To test these hypotheses, active-site variants Q392A, Y510A, V511F, V511W, H512A, H556A and H556N were generated by sitedirected mutagenesis. Additionally, the covalent linkage between His103 and FAD was disrupted in variant H103A to probe the influence of the covalent bond on the enzyme's properties. These variants were characterized both biochemically and biophysically by application of a broad set of methods.

Cloning, expression and purification
We chose six active-site residues as targets for mutagenesis (Table S1) and structure/function studies based on the crystal structure of A. meleagris PDH [13] (AmPDH; PDB code: 4H7U) and on recent findings employing MD simulations [17], namely His103, Gln392, Tyr510, Val511, His512 and His556 (Fig. 1). The computational studies indicated that these residues might be involved in substrate binding (Gln392, Tyr510, Val511, His556) or catalysis (His512, His556). We generated the variants V511F and V511W (to probe whether the backbone oxygen atom of Val511 supports substrate binding via hydrogen-bond formation) [17], as well as H512A and H556A (for testing the function of these histidines as catalytic bases). Additionally, the variant H556N was produced because the catalytically related enzyme pyranose 2-oxidase (which uses O 2 as electron acceptor) from Trametes multicolor (TmPOx) carries an asparagine at position 593 [7,21] that corresponds to His556 in AmPDH. Other variants that were studied in an alanine-scanning approach included H103A (removal of the covalent attachment of FAD), Q392A (removal of a hydrogen bond with bound sugar substrate) and Y510A (removal of stacking interaction with bound sugar substrate). All proteins were heterologously expressed using P. pastoris, which proved to be superior for the production of recombinant AmPDH to other microbial expression systems [3]. A purification scheme based on cross-flow filtration, hydrophobic interaction and immobilized metal affinity chromatography (IMAC) was employed (data not shown).

Molecular properties
After chromatographic purification, fractions of highest purity were combined for each variant, resulting in a final pool with an apparent homogeneity of > 98% as judged by SDS/PAGE (data not shown). Recombinant, Pichia-expressed AmPDH and all variants showed a broad smear at approximately 90 kDa on SDS/PAGE. Deglycosylation with PNGase F under denaturing conditions resulted in one distinct band of approximately 64 kDa, indicating a degree of glycosylation of approximately 30%, which is in good agreement with previous studies [3,19]. This rather high degree of glycosylation is caused by glycan additions of the high-mannose type by P. pastoris [22] because AmPDH purified from its native source A. meleagris has a sugar content of approximately 7% [2].
To assess the overall fold of the variants, far-UV electronic circular dichroism (ECD) spectra were recorded and compared with wild-type recombinant AmPDH. All proteins showed identical far-UV ECD spectra (data not shown), indicating that they have very similar secondary structure composition [23]. Consequently, the mutations introduced did not affect proper protein folding. Distinct minima at 208 and 222 nm indicate a dominating a-helical content [23,24], which agrees well with secondary structure elements of the crystal structure of PDH [13].
To assess the effect of the introduced mutations on thermal stability of the AmPDH variants, the unfolding temperature (T m ) was determined by the Thermo-FAD method [25] in the pH range 2-9 (Fig. 2). All variants are most stable in the pH range 5-6. A T m value of 73.5°C was determined for recombinant wildtype AmPDH. Interestingly, variant H103A has a T m of 69.5°C, which is 5.7°C higher compared to the H103Y variant studied previously [20]. Apparently, the introduction of the bulkier tyrosine side chain at this position destabilizes the protein more significantly. The largest decrease in T m for all studied variants was observed for H512A (T m = 67.0°C), which is 6.5°C lower than the value for wild-type recombinant AmPDH. This suggests that His512 is important in maintaining the active-site architecture and the overall fold. T m values ranging from 68.8 to 71.8°C were determined for the other variants, indicating that the amino acid replacements chosen have no major effect on thermal stability and do not destabilize any of the variants to an extent that complicates further studies.
Studies on the related enzyme TmPOx showed that active-site variants may contain mixed populations of (non)covalently bound FAD [26,27]. This was also found for amino acid replacements not involved in the covalent linkage as a result from a decrease of the overall positive charge around the flavin N1 region. Consequently, covalent FAD attachment was tested for all generated variants using protein precipitation with 10% trichloroacetic acid (TCA) and 40% acetone. Spectra of the protein solution before and after precipitation were recorded (Fig. 3). For variant H103A, lacking the covalent linkage to FAD, 89% of the protein released the FAD cofactor into the supernatant after precipitation. For AmPDH and all other variants, < 4% FAD was found in the supernatant, indicating covalently bound FAD cofactor. The UVvisible spectra of the oxidized, nonprecipitated enzyme solutions (Fig. 3, black lines) show a hypsochromic (blue) shift of the flavin peak at approximately 370 nm for all variants with a covalently attached FAD. This shift indicates the presence of 8a-substituted flavin [28,29], which, in the case of PDH, is an 8a-N 3 -histidyl-FAD [13,30]. Free FAD has a characteristic absorption peak at 450 nm [29], which is red-shifted in Fig. 1. Active site of AmPDH (PDB code: 4H7U) with GLC bound according to pose A (i.e. oriented for oxidation at C2). GLC coordinates from the closely related TmPOx (PDB code: 3PL8) were grafted into AmPDH after superimposing the X-ray structures of both enzymes. Atom-colouring scheme: carbon (beige, protein; yellow, FAD; white, ligand), nitrogen (blue) and oxygen (red). Red dashes represent important hydrogen bonds between GLC and AmPDH found during molecular dynamics simulations [17] and docking [13] in previous studies. Image generated using PYMOL (http://www.pymol.org/).
AmPDH and all variants. The bathochromic shift of the 450 nm peak upon FAD integration into a protein has been described earlier (e.g. for monomeric sarcosine oxidase) [31]. It is caused by a more extended delocalization of the p-electron system of the aromatic ring system in FAD [32]. For H103A, the maximum of this absorption peak blue-shifts from 460 to 450 nm after protein precipitation (Fig. 3, dashed line), as expected for free FAD.
Apparent steady-state kinetics, MD simulations and free energy calculations Apparent steady-state kinetic constants were determined for both, the electron donor substrate GLC and for the one-electron acceptor ferrocenium ion (Fc + ) ( Table 1). For Fc + , variants can be classified in two groups: those with K M values comparable to the wild-type (H103A, V511W, H556A) and those with a K M value that is at least twice as high (Q392A, Y510A, V511F, H556N). For all variants (except Q392A), the k cat values are between 20 and 40 s À1 and thus significantly lower than for the wild-type (135 s À1 ). Because the binding site of Fc + , the electron transfer path between the reduced enzyme and the electron acceptor, and the rate-limiting step in the PDH-catalysed reaction are unknown, care should be taken not to over interpret the apparent steady-state kinetic data obtained for Fc + , and we therefore refrain from drawing any further mechanistic conclusions.
Studies employing MD simulations [17] and semi-rational protein engineering [19] suggested that His512 and His556 can both act as catalytic base during GLC oxidation. H512A shows an approximately 10-fold increase in K M (K M,Glc of 6.19 mM) and a 41 200-fold lower k cat (k cat,Glc of 0.001 s À1 ) compared to recombinant wild-type AmPDH, indicating its functional importance. These values are in agreement with studies on the H548N variant of TmPOx, where His548 corresponds to His512 in PDH. For H548N, K M,Glc changed from 0.72 to 1.10 mM, whereas k cat,Glc was reduced 46 000-fold compared to wild-type TmPOx [7]. In more detailed studies, His548 in TmPOx was identified as the general base abstracting the 2-hydroxyl proton of GLC and initiating the hydride transfer from the substrate to the flavin [27]. For AmPDH variants involving His556, k cat,Glc is diminished by a factor of 3 (H556A) or 7 (H556N) compared to AmPDH, indicating that the sugar is still efficiently oxidized. When using the catalytic efficiencies measured for the His556 variants and wild-type AmPDH to calculate the difference in transition state energies (ΔΔG ES ‡ ) according to Eqn (1), we obtained values of 16.8 and 14.9 kJÁmol À1 for H556N and H556A, respectively, which is consistent with the removal of a hydrogen bond when exchanging His556. These findings imply that only His512 takes on the role as the general base in AmPDH abstracting a proton from a sugar substrate, whereas His556 appears to be mainly involved in substrate interaction. Consequently, the substrate specificity of AmPDH cannot be explained by the presence of two active-site histidines that can both act as a general base but must be attributed to other factors, such as a diverse hydrogen-bonding pattern and the dominance of relatively unspecific interactions (e.g. van der Waals) for substrate binding in different poses as observed before [17]. This is in agreement with previous studies on the glucosemethanol-choline family member aryl-alcohol oxidase (PDB code: 3FIM) where active-site histidines His502 (corresponding to His512 in AmPDH) and His546 (corresponding to His556 in AmPDH) were shown to have identical roles as proposed for AmPDH; namely,  proton abstraction and substrate interaction, respectively [4]. For glucose 1-oxidase, similar observations were obtained by experimental and computational studies for active-site histidines His516 (corresponding to His512 in AmPDH) and His559 (corresponding to His556 in AmPDH) [33]. Compared to wild-type recombinant AmPDH, the lowest increase (3.3-fold) in K M,Glc was observed for variant H103A, which lacks the covalent bond to FAD, and a reduction in k cat,Glc by a factor of only 2.7, resulting in the highest catalytic efficiency obtained for any variant (k cat /K M = 5.7 mM À1 Ás À1 ). Disrupting the covalent FAD linkage in H103A causes a decrease in flavin oxidative power, which can explain this decrease in the turnover number for GLC from 42.2 to 15.6 s À1 . The variants V511F and V511W were selected based on an in silico study [17], which showed that the backbone atoms of Val511 are important for substrate binding through hydrogen bonds (Fig. 1), and that abolishing these bonds might affect the site of GLC oxidation. In the present study, we wanted to corroborate this hypothesis by introducing Phe and Trp at position 511, with both having bulky side chains that potentially disrupt the interaction between substrate and the backbone. For variant V511F, this worked well, as indicated by the difference in transition state energies ΔΔG ES ‡ for wild-type AmPDH and V511F, for which we calculated a value 14.1 kJÁmol À1 , which is consistent with the loss of a hydrogen bond. For variant V511W, a k cat of 18.6 s À1 shows that GLC turnover is not dramatically decreased for this variant. Surprisingly, the difference in transition state energies ΔΔG ES ‡ for this side chain exchange was only 5.88 kJÁmol À1 , which is counterintuitive because more interference would be expected in the substrate interaction by the bulkier Trp side chain. This result, however, could be explained by MD simulations. The N-atom of the Trp511 side chain forms two prominent hydrogen bonds with the backbone O-atom of Asp90 (33.8%) and Pro92 (41.5%), cumulating in an average value of 0.75 observed hydrogen bonds. This hydrogen bonds causes the Trp511 side chain to flip out of the active site and, apparently, it does not interfere significantly with the interaction of the main chain carbonyl oxygen and glucose. Tyr510 was suggested to form the floor of the sugar-binding pocket in PDH, and to bind a sugar substrate via a stacking interaction. Removal of the aromatic ring as in Y510A resulted in a difference in transition state energies (ΔΔG ES ‡ ) of 13.3 kJÁmol À1 compared to wild-type AmPDH, which agrees well to quantum mechanical calculations for the interaction of an aromatic ring with a monosaccharide [34].
To further support experimental results with MD simulations, relative binding free energies (DDG bind ) for glucose were approximated from the experimentally obtained apparent K M values ðDDG exp bind Þ according to Eqn (2), and calculated from MD simulations ðDDG sim bind Þ according to an empirical free-energy method (Eqn 3) for wild-type AmPDH and selected a AmPDH produced in the natural source A. meleagris [2]. b AmPDH recombinantly produced in P. pastoris [3]. c AmPDH with a His 6 -tag, recombinantly produced in P. pastoris (present study). variants. Because we mainly wanted to investigate the effects on GLC turnover, only variants in which the exchanged residues are expected to interact directly with GLC were studied by MD simulations. The best match between DDG exp bind and DDG sim bind was obtained for the parameter set of a = 0.503 and b = 0 in Eqn (3), which resulted in a root mean square error of only 3.9 kJÁmol À1 . This is in line with previous studies on promiscuous enzymes that identified van der Waals interactions as the main driving force for substrate binding [17,35,36], which is reflected in the present study by neglecting the electrostatic b-term and using the van der Waals dependent a-term only [36].
The difference between approximated DDG exp bind and DDG sim bind values is below the thermal noise k B T (2.5 kJÁmol À1 ) for Q392A, V511W, H556A and H556N (Table 2). For variants Y510A, V511F and H512A, the differences in DDG exp bind and DDG sim bind ranged from 4 to 7 kJÁmol À1 , which is slightly higher than the chemical accuracy [37,38] of 4 kJÁmol À1 . For Y510A, both DDG exp bind and DDG sim bind are very high, indicating a qualitative agreement. According to DDG exp bind for V511F, the Phe side chain in this variant appears to cause steric clashes that interfere with tight substrate binding, resulting in its faster dissociation. To simulate the unbinding, much longer simulation time scales would be necessary, which is beyond the scope of the present study. For variant H512A, which lacks the catalytic His512, the difference between DDG exp bind and DDG sim bind is possibly caused by a different reaction mechanism compared to AmPDH. The very low but measurable residual activity of H512A could probably be derived from rarely occurring solvent-mediated proton abstraction of the GLC hydroxyl group attached to C2 or C3. This would allow for a hydride transfer to occur between the GLC hydrogen atom attached to C2 or C3 and the flavin N5-atom. DDG exp bind would hence refer to a different species or catalytic binding pose. Therefore, we refrain from drawing any further conclusions based on the conducted MD simulations.

Product analyses via GC-MS verify the predictions from MD simulations for the glucose oxidation mode
We measured the distances between: (i) the GLC hydrogen atom attached to C2 or C3 (HC2 and HC3) and the flavin N5-atom and (ii) the hydrogen atom of the hydroxyl group of GLC attached to C2 or C3 (HO2 and HO3) and the His512 Nd-or Ne-atom in MD simulations. These distances were successfully employed to reproduce and predict C2 and C3 oxidation of various monosaccharides by AmPDH in recent studies [17,18], in which a 0.3 nm cut-off was used to discriminate between reactive (≤ 0.3 nm) and nonreactive distances (i.e. enabling proton abstraction and hydride transfer). A qualitative agreement between the occurrences of reactive distances of less than 0.3 nm and the experimentally determined k cat,Glc values is observed (Fig. 4). When multiplying the probability of occurrences of distances of ≤ 0.3 nm for HC-N5 with those for HO-H512 (for C2 or C3, respectively), the product always yields a higher value for C2 compared to C3, indicating that C2 is more likely to be positioned for proton abstraction and hydride transfer than C3. This suggests that oxidation at C2 is preferred over C3. To confirm this hypothesis, GC-MS measurements were conducted to quantify GLC and its oxidized reaction products. We used a novel approach of two-step derivatization (ethoxylation and trimethylsilylation), capillary gas chromatography and electron ionization mass spectrometry (GC-EI-MS) to quantify GLC and its (di-)keto reaction products [39,40]. Because authentic standards for 3-keto-GLC and 2,3-diketo-GLC are not commercially available, capillary gas chromatography combined with chemical ionization time of flight mass spectrometry (GC-CI-QToFMS) was employed to confirm their identity via accurate mass, isotopologue pattern and modified retention indices (a detailed description of the GC-MS measurements is provided in the Materials and methods, as well as in Fig. 5 and Table 3). A total ion chromatogram of a sample analysed by GC-EI-MS is presented in Fig. 5A. The standard electron ionization (70 eV) mass spectra of the two-step derivatization product of 2-keto-GLC was confirmed via standard EI mass spectra with low resolution, which gave an m/z of 552.4 (M+, < 1%), 537.4 ([M-CH 3 ]+, 5%), 417.2 . Mass spectra of the two-step derivatization product of 2-keto-GLC were successfully cross-checked with published data. In addition, the fragmentation pathway of the two-step derivatization products of 2-keto-GLC and 3-keto-GLC was predicted with MASS FRONTIER, version 7.0SP1 (Thermo Scientific, Waltham, MA, USA). For 3-keto-GLC, the mass spectrum of the two-step derivatization products was similar to 2-keto-GLC, except for one fragment derived from 3-keto-GLC with an m/z of 174.1, which differed significantly between two isomers. With this fragment and the protonated ion in methane, chemical ionization spectra were used to screen for this compound in the enzyme reactions. However, a high degree of fragmentation during electron ionization and unspecific fragments in standard electron ionization were observed, preventing the distinction between 2-keto-GLC and 3-keto-GLC by only EI or CI mass spectra. Because of the lack of commercially available standards for 3-keto-GLC and 2,3diketo-GLC, the discrimination between the different keto-GLC derivatives was only possible by combining chemical ionization mass spectra and electron ionization mass spectra with the fragmentation pathway. High-resolution spectra of exthoxymation-trimethylsilylation of monoand diketo-GLC employing methane gas ionization are shown in Fig. 5B. For the investigated mono-and diketo-GLC, the mass accuracy of the protonated molecular ion was below 15 p.p.m. and the experimental isotopologue fractions were in agreement with the theoretical mass ( Table 3). Identification of 2,3-diketo-GLC was accomplished by its retention time during chromatography. In addition, the molecular formula generation function in qualitative MASSHUNTER with atoms filter C, H, N, O and Si was set to 19-23, 47-55, 2-4, 5-8 and 2-5, respectively. The comparison of selected compounds with the molecular formula generation was carried out not only by m/z and isotopologue fraction, but also according to the spacer between the isotopic patterns. The match scores were 93.9 and 96.4 for mass and isotopologue fraction distribution, respectively. Figure 5C shows the theoretical and experimental isotopologue fractions for 2,3-diketo-GLC.
The concentrations and relative ratios of GLC and its (di-)keto reaction products obtained from GC-MS measurements for AmPDH and its variants after 3, 8 and 120 min of reaction time are listed in Table 4. Only the reaction products of variant H512A could not be identified because of its extremely slow GLC turnover (Table 1). These results match the predictions from MD simulations very well, showing that C2 oxidation is indeed preferred over C3 oxidation because only very little or no 3-keto-GLC was detected, whereas there were significant amounts of 2-keto-GLC and 2,3diketo-GLC for all tested variants. This indicates that 2-oxidation is significantly faster than 3-oxidation, for both GLC and a mono-oxidized intermediate.
Based on these results, we propose a scheme for the two-step oxidation of GLC by AmPDH that is also valid for all tested variants. This scheme is very similar to the one reported earlier by Volc et al. [41] for PDH from Agaricus bisporus (AbPDH). The main difference between AbPDH and AmPDH is that AmPDH almost exclusively double-oxidizes GLC via 2-keto-GLC, whereas AbPDH performs this reaction via both 2keto-GLC and 3-keto-GLC, with the latter being favoured to some extent, to yield the final product 2,3diketo-GLC.
In a previous in silico study, mutations at Gln392, Tyr510 and Val511 in AmPDH were suggested to modify the preferred GLC oxidation site because these amino acids are involved in prominent hydrogen bonds with the substrate bound in a position  Table 4. GC-MS analysis of GLC and its reaction products 2-keto-GLC, 3-keto-GLC and 2,3-diketo-GLC. The reaction mixture contained 1 mL of 10 mM potassium phosphate buffer (pH 7.0), 1 U of AmPDH or its respective variant, 45 mM 1,4-benzoquinone as electron acceptor and, initially, 16.7 mM GLC. Samples (200 lL) were taken after 3, 8 and 120 min of reaction time, respectively. After deproteinization through a 10 kDa ultrafiltration membrane, the filtrate was used for subsequent GC-MS analysis. For GLC and 2-keto-GLC, an external calibration was possible with commercially available standards, allowing for a quantitative determination of their concentrations (mM) with a detection limit of 0.2 lM. Because no standards were commercially available for 3-keto-GLC and 2,3-diketo-GLC, only semiquantitative results (relative concentrations in mM) could be calculated, based on the peak-area-ratios of the respective analyte to 2-keto-GLC at a given mass to charge ratio. To obtain the 'Ratio', the percentage of each compound with respect to the sum of (relative) concentrations was calculated for each variant.

Variant
Reaction time (min) relevant for either C2 or C3 oxidation [17]. Interestingly, neither explicit simulations of the mutants, nor the GC-MS analysis after glucose turnover indicated an altered preference for C2 or C3 oxidation compared to wild-type AmPDH. Consequently, substrate binding in an orientation specific for either C2 or C3 oxidation cannot be attributed to one specific active-site residue in AmPDH, although an interplay between several residues appears to be critical for this.

C2 and C3 oxidation of various sugar substrates proceed with comparable rates
The reductive half reaction (FAD ? FADH 2 ) of AmPDH was studied in detail with sugar substrates that are specifically and exclusively oxidized at either C2 or C3 to investigate whether AmPDH shows a preferred mode of sugar oxidation (i.e. whether the rate constants are significantly different for C2 or C3 oxidation). As reported previously, AmPDH is reduced rapidly by GLC at 30°C, with an apparent bimolecular rate constant (k app,30°C ) of 1.1 AE 0.03 9 10 5 M À1 Ás À1 [20]. Consequently, all stopped-flow experiments in the present study were conducted at 4°C to slow down the reaction and capture as much kinetic information as possible. All kinetic traces obtained (Fig. 6) were simulated with parameters listed in Table 5 according to Kinetic Model 1 red : According to previously published data [1,16], D-galactose (GAL) and L-arabinose (ARA) are exclusively oxidized at C2, whereas methyl-a-D-glucopyranoside (MGP) is only oxidized at C3, with no double oxidation occurring. AmPDH oxidizes GLC at both C2 and C3 but C3 oxidation (also of mono-oxidized 2-keto-GLC) is very slow and takes minutes to hours [13] (Table 4). Therefore, GLC can be treated as C2oxidized substrate in the time course of stopped-flow experiments that spans just a few seconds. The reactions of AmPDH for GLC, GAL, ARA and MGP monitored at 463 nm were biphasic. Their time traces were fitted by a double-exponential function to obtain the respective k obs values at distinct sugar concentrations. The first phase k obs(1) corresponded to FAD To evaluate the experimental traces (black), a double-exponential fit was used. For all sugars, the first phase k obs(1) was the FAD reduction with a large absorbance decrease at 463 nm, which was used to calculate k app,4°C . The experimental traces were simulated with BERKELEY MADONNA, version 8.3.14, according to Kinetic Model 1 red (red traces) with the parameter sets listed in Table 5. The inset shows the plot of the pseudo first-order rate constants k obs(1) versus the concentrations of the respective sugar. reduction with a large absorbance decrease at 463 nm. It was used for calculating k app,4°C ( Table 6). The second phase k obs (2) showed a small decrease in absorbance at the same wavelength and was independent of the GLC concentration (data not shown). The y-intercepts between 7.6 and 15.2 s À1 indicate a reversible step that cannot be significant because the magnitude of the y-intercept is rather small compared with the scale of the y-axis for all tested sugars (Fig. 6). Moreover, the simulated traces (Kinetic Model 1 red ) agree well with experimental traces (Fig. 6), indicating the validity of the model employed, which does not consider a reversible step. For all sugars, the obtained k app,4°C values were similar and in a range of 4.4 to 6.8 9 10 4 M À1 Ás À1 ( Table 6). The pentose sugar ARA showed the highest k app,4°C , which can possibly be attributed to its smaller size compared to the other tested sugars (hexoses) because this could ultimately facilitate its diffusion to and from the active site.
Effects of the His ? Ala replacement at position 556 on both half reactions Although His556 does not act as a catalytic base, we found that it is important for sugar substrate binding (see above). To identify the rate-limiting step in variant H556A and compare the results with recombinant wild-type AmPDH, the transient-state kinetics of their reductive (FAD ? FADH 2 ) and oxidative (FADH 2 ? FAD) half-reactions were investigated by stopped-flow spectroscopy, using GLC as electron donor and 1,4-benzoquinone (BQ) as electron acceptor, respectively. In a previous study, the ferrocenium ion (Fc + ) was used as electron acceptor, which has an absorbance at 330-350 nm and 625 nm [20]. Therefore, Fc + interferes to a greater extent with the flavin absorbance than BQ, which has an absorption maximum at 244 nm, and the corresponding hydroquinone at 222 nm and 296 nm, respectively [42]. The stoppedflow experiments were performed essentially as Table 5. Parameter set for simulating stopped-flow traces with BERKELEY MADONNA, version 8.3.14. The reductive half-reaction recorded for different electron donors (sugars) was simulated in agreement with Kinetic Model 1 red . The oxidative half-reaction for the electron acceptor BQ was simulated according to Kinetic Model 1 ox . Simulated traces of the reductive and oxidative half-reactions for AmPDH are shown in red dashed lines in Figs 6 and 7, respectively.

Variant
Substrate AmPDH GLC 6.9 9 10 4 5.  described above, and the kinetics were recorded at 465 nm for H556A and at 463 nm for AmPDH, respectively. For both GLC and BQ, the reactions were biphasic and their time traces were fitted by a double-exponential function to obtain the k obs,4°C values. As described above, a k app,4°C of 5.4 AE 0.11 9 10 4 M À1 Ás À1 was obtained for the reductive half-reaction of AmPDH with GLC (Table 6), corresponding to a two-fold decrease compared to measurements at 30°C [20]. For the oxidative half-reaction of AmPDH with BQ, the first phase k obs(1) of the double-exponential fit corresponds to FAD oxidation with a large absorbance increase at 463 nm (Fig. 7), which was used to calculate k app,4°C ( Table 6). The second phase k obs (2) of the double-exponential fit showed a significant absorbance decrease from the intermediate species E ox : BQH 2 to E ox and was independent of the BQ concentration (data not shown). For the first phase, the y-intercept is very small, indicating a negligible reversible step. Simulations according to Kinetic Model 1 ox: ðKinetic Model 1 ox Þ with parameter sets listed in Table 5 corroborate this observation because the simulated traces agree well with experimental traces (Fig. 7), and the model does not consider a reversible step as well. The second phase of simulated and experimental traces does not overlap as well as the first phase, which might be attributed to the complex quinone reduction process or to polymerizations during side reactions. However, the imperfect fit does not affect the first phase, which was used to calculate k obs(1) for AmPDH oxidation, and therefore the utilized kinetic model is valid. For the reductive half-reaction of H556A with GLC, the first phase k obs(1) of the double-exponential fit only showed a small absorbance decrease and was independent of the GLC concentration (data not shown). The second phase k obs(2) of the double-exponential fit corresponds to FAD reduction with a large absorbance decrease at 465 nm (data not shown), which was used to calculate k app,4°C (Table 6). For the second phase, the y-intercept is close to zero, indicating a negligible reversible step. The k app,4°C for the reductive half-reaction of variant H556A with GLC dramatically decreased 720fold compared to AmPDH ( Table 6), indicating that the two-electron transfer to FAD is significantly hampered by this amino acid exchange.
In the oxidative half reaction of H556A with BQ, the first phase k obs(1) of the double-exponential fit corresponds to FAD oxidation with a large absorbance increase at 465 nm, which was used to calculate k app,4°C ( Table 6). The second phase k obs(2) of the double-exponential fit only showed a small absorbance increase and was independent of the BQ concentration (data not shown). For the first phase, the y-intercept is again close to zero, which indicates no reversible step. Similar to AmPDH, the second phase of simulated and experimental traces for the oxidative half-reaction in H556A do not perfectly overlap. For variant H556A, the k app,4°C of the biphasic oxidative half-reaction with BQ was approximately 6.2-fold higher compared to AmPDH (Table 6), which can possibly be attributed to the lower redox potential of H556A (see below) or to a more accessible active-site in the variant. However, because the binding site of BQ in AmPDH and the electron transfer path from AmPDH to BQ remain elusive, further studies will be necessary before any final conclusions can be drawn.
The presented pre-steady-state experiments clearly demonstrate that the impairment of the reductive halfreaction of variant H556A is the main reason for its at 4°C under anaerobic conditions. All given concentrations are reported after mixing. The reaction was monitored at 463 nm. The kinetic traces from right to left correspond to increasing BQ concentrations. To evaluate the experimental traces (black), a double-exponential fit was used. The first phase k obs(1) was the FAD oxidation with a large absorbance increase at 463 nm, which was used to calculate k app,4°C . The experimental traces were simulated with BERKELEY MADONNA, version 8.3.14, according to Kinetic Model 1 ox (red traces) with the parameter sets listed in Table 5. The inset shows the plot of the pseudo first-order rate constants k obs(1) versus the BQ concentrations. considerably lower catalytic efficiency (Table 1) compared to AmPDH. The oxidative half-reaction of variant H556A is significantly faster compared to AmPDH.

Disruption of the covalent FAD linkage decreases the redox potential
The covalent FAD attachment in flavin-dependent proteins, and especially the attachment of electron-withdrawing substituents at position 8, has been associated with the increase in redox potential of the enzymes [30,43], allowing them to use high-potential electron acceptors such as molecular oxygen as redox partners. Consequently, a covalently attached FAD was mainly found in oxidases. This makes AmPDH a rather unusual example for a dehydrogenase in carrying an FAD that is monocovalently tethered via position 8. Therefore, we probed the effect of the missing covalent FAD linkage in H103A on its redox potential. Furthermore, the decrease in the reductive half-reaction of variant H556A could potentially be explained by a decrease in the redox potential of this variant, and hence we determined its redox potential as well.
The reduction of AmPDH ( Fig. 8A and Table 7) followed a two-electron reduction process as judged by a slope of 1.0 when plotting log(E red /E ox ) versus log (D red /D ox ). The midpoint potential ðE o0 Þ for AmPDH was +92 AE 3 mV at pH 7.0 and 25°C. For H103A (Fig. 8B and Table 7), the fully oxidized enzyme  Table 7. Table 7. Redox potential values ðE o0 Þ for PDH from A. meleagris and selected variants. The corresponding reduction spectra are plotted in Fig. 8. A slope close to 1.0 indicates a two-electron reduction step, which explains why no one-electron reduction steps ðE o0 1 andE o0 2 Þ could be determined for AmPDH. A slope close to 0.5 indicates a one-electron reduction step, which was observed for variant H103A and H556A. Hence, both one-electron reduction steps could be determined for these variants. M, maximum fraction of thermodynamically stable semiquinone formed, which was necessary to calculate E o0 for both variants.  Fig. 8B). This does not perfectly fit to the characteristic absorbance of the anionic semiquinone radical at approximately 380 nm and 400 nm [29]. However, it is known that the microenvironment around the flavin (e.g. the polypeptide matrix, pH, ligands) can significantly affect the occurrence of semiquinone radicals [29] and can cause bathochromic shifts of, for example, the 450 nm flavin absorption band [32]. Consequently, the red-shifted semiquinone absorbance in variant H103A can possibly be attributed to similar factors. For variant H103A, E o0 was +63 AE 3 mV at pH 7.0 and 25°C (Table 7). Therefore, disruption of the covalent FAD linkage decreases the redox potential by 29 mV. For H556A (Fig. 8C and Table 7), similar observations as for H103A were made. The fully oxidized enzyme was reduced by two one-electron reduction steps, as indicated by a slope of 0.50 when plotting log (E red /E ox ) versus log(D red /D ox ). This again points towards a putative semiquinone intermediate on the way to the fully reduced enzyme. The observed intermediate for variant H556A had an absorption maximum at 368 nm (Fig. 8C, dashed line) and was detected previously in the resting state of AmPDH, which could be attributed to an equilibrium of the oxidized form of the enzyme with its reduced forms [20]. For variant H556A, E o0 was +84 AE 4 mV at pH 7.0 and 25°C (Table 7). Therefore, the redox potential of variant H556A is only decreased by 8 mV compared to AmPDH, indicating that the difference in redox potential between H556A and AmPDH is not responsible for the 720-fold decrease of the reductive half-reaction with GLC (see above).
It is not exactly known why flavoenzymes do or do not react with oxygen [19,44]. Factors such as a positive charge in the active site [45], its accessibility via channels and tunnels [46] and the microenvironment around the flavin N5-C(4a) [47,48] have been identified as important features for oxygen reactivity. Therefore, all generated variants were routinely probed for their steadystate oxygen reactivity, with variant H512A being the only exception because it reacted too poorly with the electron donor GLC. Results for relative oxygen activities are given in Fig. 9. Wild-type AmPDH had a very low oxygen reactivity of 0.104 lMÁmin À1 Ámg À1 . Several of the variants showed an increase in oxygen reactivity, yet, despite this increase, the reaction with oxygen is still poor. The highest oxygen reactivity amongst all tested variants (0.430 lMÁmin À1 Ámg À1 , corresponding to 414% of wild-type AmPDH) was found for H103A, which lacks the covalent bond to FAD and showed a significantly reduced redox potential. This is in line with previously published work on variant H103Y, which was found to have a 5.3-fold higher oxygen reactivity compared to AmPDH [19]. The oxygen reactivity for variant H556A was increased to a similar level as for H103A; however, its redox potential was only slightly decreased compared to AmPDH. Consequently, a direct link between lower redox potential and higher oxygen reactivity cannot be drawn in AmPDH. Moreover, one would expect rather the opposite correlation, namely a high redox potential that is linked to increased oxygen reactivity, because a higher redox potential is usually associated with electron accepting redox-partners that have high redox potentials, such as oxygen.

Conclusions
We investigated AmPDH and eight variants thereof by biochemical, biophysical and computational means. The secondary structures and thermostability were not affected by the newly introduced amino acids as judged by ECD spectroscopy and ThermoFAD experiments, Fig. 9. Relative oxygen reactivities of wild-type recombinant pyranose dehydrogenase from A. meleagris and active-site variants. Steady-state oxygen reactivity was determined using an Amplex Red/horseradish peroxidase based assay [19]. The assay contained 0.5 mgÁmL À1 of the purified enzyme, 50 lM Amplex Red, 25 mM GLC, 0.1 UÁmL À1 horseradish peroxidase and 50 mM sodium phosphate buffer (pH 7.4). Error bars represent the SD of four repeats. The specific activity of wild-type recombinant AmPDH with oxygen is 0.104 lMÁmin À1 Ámg À1 (green bar). respectively. All enzymes were monocovalently flavinylated, and only variant H103A had a noncovalently but tightly bound FAD cofactor. The redox potential for this variant was decreased by 29 mV, which is in line with other flavin-dependent oxidoreductases [30]. Apparent steady-state kinetics demonstrated that His512 is the sole catalytic base in the reductive half-reaction of AmPDH, whereas the second active-site histidine, His556, is mainly involved in substrate interaction. Consequently, the substrate specificity of AmPDH cannot be explained by the presence of two active-site histidines acting as a general base and must be attributed to other factors. Predictions from MD simulations, which were verified by GC-MS measurements, unequivocally demonstrated that AmPDH and its variants preferentially oxidize GLC via 2-keto-GLC to the final product 2,3-diketo-GLC.

Chemicals and vectors
All chemicals were of the highest available purity and purchased from Sigma-Aldrich (St Louis, MO, USA), VWR (Radnor, PA, USA) and Roth (Karlsruhe, Germany). Primers were obtained from LGC Genomics (Vienna, Austria) or Sigma-Aldrich. Restriction endonucleases, T4 DNA ligase and Phusion polymerase were obtained from Thermo Fisher Scientific Biosciences (St Leon-Rot, Germany). GoTaq polymerase was purchased from Promega (Madison, WI, USA). Zeocin and the pPICZB vector were obtained from Invitrogen (Carlsbad, CA, USA). The 2keto-GLC standard, meso-erythritol, LC-MS grade distilled water, ethoxylamine hydrochloride, water-free pyridine and N-methyl-N-(trimethylsilyl)trifluoroacetamide with 1% trimethylchlorosilane in water-free pyridine were purchased from Sigma-Aldrich.

Plasmid construction for expression in P. pastoris
When expressing a protein of interest with the original pPICZB plasmid from Invitrogen, a myc-epitope followed by a His 6 -tag is added to the C-terminus of the protein. To alter the protein of interest as little as possible, the pPICZB plasmid was slightly modified for the present study. When using the primer pair pPICZB-6His-fw and pPICZB-6His-XbaI-rv (Table S1) the myc-epitope was removed using the DpnImethod [49], creating pPICZB-His 6 . The PCR-product was purified from an agarose gel, and digested with DpnI for 2 h at 37°C for degradation of methylated template DNA. After another purification step, the pPICZB vector void of the myc-epitope was transformed into chemically competent E. coli NEB5a. Positive transformants were selected on LB-LS plates containing Zeocin and subsequently sequenced. The confirmed plasmid was amplified in E. coli NEB5a, purified, and stored at -20°C. The A. meleagris pdh1 (ampdh) gene in the pPICZB vector as used by Krondorfer et al. [19] was amplified with the primer pair AmPDH-NotI-fw and AmPDH-XbaI-rv (Table S1), thereby introducing a 3 0 -NotI-and a 5 0 -XbaI-restriction site. The purified product was digested with NotI and XbaI, and ligated into the equally treated pPICZB-His 6 vector, yielding pPICZB-His 6 -AmPDH, which was transformed into chemically competent E. coli NEB5a. Positive transformants were selected on LB-LS plates containing Zeocin and subsequently sequenced. An E. coli NEB5a colony containing the verified plasmid was proliferated, pPICZB-His 6 -AmPDH purified and finally stored at -20°C.

Protein expression and purification
Recombinant wild-type A. meleagris PDH1 (AmPDH) and the H556A variant were produced as described previously [3] but in a 42 L computer-controlled stirred tank reactor (Applikon, Schiedam, The Netherlands) with an initial volume of 20 L of basal salts fermentation medium. The variants H103A, Q392A, Y510A, V511F, V511W, H512A and H556N were produced as described previously [19] in a 7 L computer-controlled fermenter (MBR, Wetzikon, Switzerland) with an initial volume of 4 L of basal salts fermentation medium.
AmPDH and H556A were purified at room temperature according to a four-step purification scheme. First, the volume of the cultivation supernatant was reduced by a factor of approximately 10 employing cross-flow filtration (Microza ultrafiltration module; Pall, Vienna, Austria) using a membrane with a 10 kDa cut-off. The concentrate was subsequently purified by IMAC with a Ni 2+ -charged Chelating Sepharose Fast Flow column (65 mL; flow rate 10 mLÁmin À1 ; GE Healthcare, Little Chalfont, UK), equilibrated with 100 mM potassium phosphate buffer (pH 7.0), 1 M NaCl and 5 mM imidazole. After a washing step of 3 column volumes (CV), AmPDH or H556A were eluted with a linear gradient from 5 to 500 mM imidazole. Fractions with PDH activity were pooled. (NH 4 ) 2 SO 4 was added to the pooled IMAC I-fractions to 40% saturation and loaded onto a Phenyl Sepharose Fast Flow column (200 mL; flow rate 10 mLÁmin À1 ; GE Healthcare) equilibrated with 50 mM potassium phosphate buffer [pH 6.5, 40% saturation (NH 4 ) 2 SO 4 ]. After washing the column with 3 CV of the same buffer, proteins were eluted with a linear gradient of starting buffer to 50 mM potassium phosphate buffer (pH 6.5) in 1 CV. Fractions with PDH activity were pooled. As a polishing step, a second IMAC purification was conducted. The pooled hydrophobic interaction chromatography fractions were loaded onto four in-series connected HisTrap HP columns (5 mL each; flow rate 2 mLÁmin À1 ; GE Healthcare) equilibrated as for IMAC I. For washing and elution, conditions the same as those for IMAC I were applied. Fractions with PDH activity were pooled. The variants H103A, Q392A, Y510A, V511F, V511W, H512A and H556N were purified using a combination of hydrophobic interaction chromatography and IMAC II as described above.
Imidazole from the last IMAC step was removed for all purified enzymes by ultrafiltration (10 kDa cut-off membrane; Amicon Ultra Centrifugal Filter Device; Millipore, Billerica, MA, USA). The purified and concentrated enzymes were washed thrice with 10 mL of 50 mM potassium phosphate buffer (pH 7.0). AmPDH is mainly isolated in its reduced form [13,19]. To fully oxidize AmPDH and its variants, the enzymes were re-oxidized with 5 mM 2,6dichlorophenol-indophenol (DCIP). The solution was subsequently passed through a PD-10 desalting column (GE Healthcare) equilibrated with 50 mM potassium phosphate buffer pH 7.0 to remove DCIP. Then, the solution was sterile-filtered by passing through a membrane of 0.22 lm cut-off (Merck Millipore, Darmstadt, Germany), and finally diluted in sterile potassium phosphate buffer pH 7.0 to a protein concentration of approximately 20-50 mgÁmL À1 . Aliquots of 100 lL were shock-frozen in liquid nitrogen and stored at -80°C. All variants were stored at 4°C after sterile filtration.
Enzyme activity assay and molecular properties PDH activity was determined by following the reduction of the ferrocenium ion (Fc + ) at 300 nm and 30°C for 3 min as described previously [50] with certain adaptions: the standard 1 mL of reaction mixture contained 50 lmol sodium phosphate buffer (pH 7.5), 25 lmol GLC and 0.2 lmol of ferrocenium hexafluorophosphate. Protein concentrations were determined according to Bradford with a pre-fabricated assay (Bio-Rad, Hercules, CA, USA). SDS/ PAGE and enzymatic deglycosylation with PNGase F were conducted as described previously [2,19].
Determination of molar absorption coefficients for AmPDH, H103A, and H556A were performed using a protocol adapted from a previous study [27]. In brief, the enzyme solution was diluted in 50 mM potassium phosphate buffer (pH 7.0) to an absorbance at 450 nm of approximately 2.5. To 100 lL of that solution, either 900 lL of the same buffer (for spectra of intact enzyme) or 900 lL of 6 M guanidine HCl (for spectra of denatured enzyme) was added. Concentrations of free FAD released from the denatured enzymes were determined in triplicate based on the free FAD molar absorption coefficient (e 450 = 1.1 9 10 4 M À1 Ácm À1 ). The extinction coefficients of FAD bound to the polypeptides were e 463 = 9.5 9 10 3 M À1 Ácm À1 for AmPDH, e 460 = 1.2 9 10 4 M À1 Ácm À1 for H103A and e 465 = 1.1 9 10 4 M À1 Ácm À1 for H556A. The low e 463 for AmPDH suggests a deprotonated flavin N3-atom [29,51].

ECD spectroscopy
To compare the overall fold and secondary structure elements of AmPDH and its variants, far UV ECD spectra were recorded as described previously [20] from 180 to 260 nm on a Chirascan CD Spectrophotometer (Applied Photophysics, Leatherhead, UK), which was flushed with nitrogen. The cuvette pathlength was 1 mm, the spectral bandwidth was set to 3 nm and the scan time per point to 10 s. The protein concentration was adjusted to approximately 4 lM with 50 mM potassium phosphate buffer (pH 7.0).

ThermoFAD
ThermoFAD measurements were conducted to elucidate the thermal unfolding temperature T m [25], utilizing the increased intrinsic fluorescence of the FAD cofactor upon thermal protein denaturation. Two replica were measured, each containing 25 lL with 65 lM AmPDH or the respective variant in 40 mM Britton-Robinson buffer (pH 2-9, in steps of 1 pH units). Heating of samples was conducted from 20-95°C in steps of 0.5°CÁs À1 with an iCycler Thermal Cycler equipped with an MyiQ Real-Time PCR Optical Module (Bio-Rad), and the corresponding fluorescence was recorded. T m was determined from the maximum of the first derivative of the obtained sigmoidal curve [25,52].

TCA/acetone precipitation
To determine whether FAD is covalently attached to the polypeptide, a solution of oxidized enzyme with an absorbance (450 nm) between 0.1-0.3 was prepared in 50 mM potassium phosphate buffer (pH 7.0) and the oxidized spectrum was subsequently recorded from 250 to 650 nm (U-3000 spectrophotometer; Hitachi, Tokyo, Japan). Precipitation was conducted as described previously [19] by mixing the double-concentrated protein solution with 10% (v/v) TCA and 40% (v/v) acetone and incubation for 10 min on ice. After centrifugation at 17 400 g for 5 min at 4°C, the spectra of the supernatants were recorded to identify the presence of noncovalently bound FAD.

Steady-state kinetics
Apparent kinetic constants for the electron donor GLC were measured utilizing the standard Fc + assay. Data for Fc + were collected in 100 mM borate buffer (pH 8.5) and the concentration of the electron donor GLC was adjusted according to the K M value of the corresponding variant: 25 mM (nsAmPDH, recAmPDH and AmPDH), 50 mM (H103A, Q392A and V511W), 250 mM (Y510A) and 500 mM (H556A, H556N and V511F). The data obtained were fitted to the Michaelis-Menten equation by nonlinear least squares regression with SIGMA PLOT, version 11 (Systat Software, Chicago, IL, USA), from which the kinetic constants were derived. Catalytic efficiencies were used to calculate difference in transition state energies ΔΔG ES ‡ between variants and wild-type AmPDH [53,54] according to:

MD simulations and free energy calculations
The structure preparations were essentially performed as described previously [17]. In brief, two different binding poses (pose A and B) of GLC were considered, differing in a rotation of 180°about the axis defined by a line running through a point midway between the GLC atoms C5 and O5, and a point midway between atoms C2 and C3, allowing for GLC oxidation at C2 or C3. Only two protonation states of the active-site histidines were considered, which had the most favourable binding energies and highest ligand stabilities of GLC in AmPDH in a previously reported comparison of the protonation states [17]. For pose A (Fig. 1), His512 and His556 were both fully protonated (protonation state PP); for pose B, His512 was fully protonated and His556 was in its neutral state with a proton at Ne (protonation state PN). For in silico mutations to more bulky side chains (V511F and V511W), rotamers most similar to the wild-type were selected. During the 10 ns simulations, the side chains of these variants could rotate and adopt new conformations. For in silico mutations to less bulky residues, initial atom positions of the side chains were derived from the wild-type residues. The simulation set-up was carried out as described previously [17]. The MD simulations were conducted with GRO-MOS, version 11 [55] employing the 53A6 force field [56] in explicit solvent within a rectangular, periodic and pre-equilibrated box of SPC water [57]. After energy minimization and equilibration [58], two independent 10 ns production runs (md1 and md2) were performed at constant pressure (1 atm) and temperature (300 K) using the weak-coupling scheme [59] with coupling times of 0.5 and 0.1 ps, respectively. The isothermal compressibility was set to 4.575 9 10 -4 kJ À1 ÁmolÁnm À3 , and two separate temperature baths were used for solute and solvent. The SHAKE algorithm was applied to constrain bond lengths [60] for solute and solvent allowing for 2 fs time-steps. Nonbonded interactions were calculated using a triple range scheme. Interactions within a short-range cut-off of 0.8 nm were calculated at every time step from a pair list that was updated every fifth step. At these points, interactions between 0.8 nm and 1.4 nm were also calculated explicitly and kept constant between updates. A reaction field [61] contribution was added to the electrostatic interactions and forces to account for a homogenous medium outside the long-range cut-off using a relative dielectric constant of 61 as appropriate for the SPC water model [62]. Coordinate and energy trajectories were stored every 0.5 ps for subsequent analysis.
Binding free energies (DG bind ) relative to AmPDH (DDG bind ) were estimated from experimental K M values ðDDG exp bind Þ according to: where k B T is the Boltzmann constant multiplied with the absolute temperature, or calculated from MD simulations ðDDG sim bind Þ by employing the linear interaction energy method [63] according to the equation [64]: where angular brackets indicate ensemble averages, calculated over a simulation of the substrate bound to the protein (protein) or free in solution (free). V EL ligÀsur and V VdW ligÀsur represent the ligand-surrounding electrostatic and van der Waals interactions, respectively. a and b are parameters of the linear interaction energy equation. DDG sim bind was obtained by subtracting DG sim bind of wild-type AmPDH from DG sim bind of the respective variant. The different poses (pose A and B) and runs (md1 and md2) were averaged as described previously [65].

Analysis of GLC reaction products by GC-MS
Stock solutions of glucose, 2-keto-GLC and meso-erythritol (internal standard) were prepared by dissolving appropriate amounts of solid standard in LC-MS grade water in amber LC vials and were kept at À40°C. Working solutions were prepared daily by diluting the stock solution in LC-MS grade water. The ethoxymation solution was prepared daily by dissolving 18.7 mg of ethoxylamine hydrochloride in 1 mL of water-free pyridine. A nonpolar HP-1MS column (60 m length 9 250 µm inner diameter 9 0.25 µm thickness film; 100% dimethylpolysiloxane) and a guard column (5 m length 9 320 µm innner diametter without stationary phase) from Agilent Technologies were used for separation. EZ-2 Envi SpeedVAC from Genevac Inc (Stone Ridge, NY, USA) was used for drying the standards and samples. GC-MS was carried out with an Agilent Technologies GC-EI-MS machine consisting of an Agilent Technologies 7820A gas chromatograph and an Agilent Technologies 5975 mass selective detector. GC-CI-ToFMS was conducted on an Agilent Technologies 7200 GC-CI/EI-QToFMS. Sample preparation and derivatization was performed as described previously [66].
The derivatization product solution was kept at 4°C in an auto-sampler, of which 1 µL was injected into a nonpolar HP-1MS column with an Agilent Technologies guard column with pulsed injection mode at 30 psi. The temperature of the injector and the transfer line were kept constant at 250°C and 280°C, respectively. Helium (99.9999% purity grade) was used as carrier gas with a flow rate of 1.2 mLÁmin À1 in constant flow mode. The column oven was kept at 70°C for 1 min and heated up to 200°C (20°CÁmin À1 ), then increased to 240°C (2.5°CÁmin À1 ) and then finally to 310°C (20°CÁmin À1 ), where it was kept for 1 min to re-condition the column. For GC-CI-QToFMS, temperature, pressure of methane gas, emission current, electron energy and QToFMS parameters were optimized as reported previously [67]. The mass range in both cases was set to 70-700 amu. An n-alkane (C8-C40) retention index in ethoxylamine and N-methyl-N-(trimethylsilyl)trifluoroacetamide solution was injected into GC-EI-MS to calculate the modified retention indices according to [68]. Cross-check mass fragmentation, determiniation of the electron ionization mass spectra pathway and chemical ionization were carried out using MASS FRONTIER, version 7.0 SP1 (Thermo Scientific). Quantification of 2-keto-GLC was carried out with external standards. For 3-keto-GLC and 2,3-diketo-GLC no standards were commercially available. Therefore, semiquantitative analysis was performed by comparing the peak areas of the corresponding derivatization products with those of 2-keto-GLC at the same mass-to-charge ratio.

Rapid reaction experiments
Reactions were carried out as described previously [69]. In brief, 50 mM potassium phosphate buffer (pH 7.0) at 4°C, a TgK Scientific model SF-61DX or a TgK Scientific model SHU-61SX2 (TgK Scientific, Bradford-on-Avon, UK) stopped-flow spectrophotometer in single-mixing mode and a dead-time of 2 ms were used. The stopped-flow instrument was made anaerobic by flushing the flow system with an anaerobic buffer solution containing 0.5 mgÁmL À1 dithionite in 50 mM potassium phosphate buffer (pH 7.0) and equilibrated in that solution overnight. Before starting the experiments, the flow system of the instrument was washed three times with anaerobic 50 mM potassium phosphate buffer (pH 7.0).
The enzymes obtained from purification contained partially reduced populations. To prepare fully oxidized enzymes, 5 mM DCIP was used as described above. To prepare a reduced enzyme solution, an anaerobic oxidized enzyme solution was stoichiometrically reduced in an anaerobic glove box with a 300 lM GLC solution in potassium phosphate buffer (pH 7.0), which was monitored spectrophotometrically. All substrate concentrations used were more than five-fold excess of the enzyme concentration (30 lM) to ensure pseudo first-order conditions.
To study the kinetic reduction of wild-type AmPDH by sugar substrates, the reaction was performed aerobically because the enzyme was shown before to be reoxidized very slowly with oxygen [13]. Anaerobic conditions were used for stopped-flow experiments of variant H556A, which was shown to have a 3.7-fold elevated oxygen reactivity compared to AmPDH. Reoxidation experiments with 1,4-benzoquinone were also performed anaerobically to prevent side-reactions of 1,4-benzoquinone with oxygen.
Observed rate constants (k obs ) for the fast and slow phase were calculated from double-exponential fits to kinetic traces using the KINETIC STUDIO (TgK Scientific) and

Redox potential determination
The redox potentials of AmPDH, H103A, and H556A were measured at 25°C according to the method of Massey [70] using xanthine and xanthine oxidase as the reduction system [7,71]. The enzymes were fully oxidized with 5 mM DCIP as described above. Solutions of fully oxidized enzyme (AmPDH, H103A or H556A), the standard dye benzyl viologen, xanthine and xanthine oxidase (side arm) were pipetted together to a final volume of 1 mL in a specially designed cuvette equipped with two side arms and a stopcock. Anaerobiosis was established by repeated cycles of evacuation and flushing with oxygen-free nitrogen. Subsequently, the reaction was started by adding xanthine oxidase from the side arm. The reduction was monitored by recording spectra with an Agilent Technologies 8453 diode array spectrophotometer. Phenazine methosulfate (E o0 ¼ 80mV) was used as standard dye for AmPDH. Enzyme and dye were monitored at 463 nm (AmPDH) and 387 nm (phenazine methosulfate) during their slow reduction over a period of approximately 8 h. AmPDH has no isosbestic point at 387 nm but a significant absorbance, which changes with altering redox states of the enzyme. Phenazine methosulfate has a small absorbance at 463 nm, which hardly changes during the reduction process of the dye. Consequently, the absorbance at 463 nm and 387 nm had to be analysed with respect to a mixed population of enzyme and dye. For this, the equations used were: where superscript E represents 'Enzyme' and D represents 'Dye' at the wavelength indicated in the subscript, e represents the extinction coefficient, and c is the (unknown) concentration. This system of two equations was solved to yield the concentrations of the enzyme or dye. These concentrations were subsequently used to calculate the ratio of reduced and oxidized species at various stages during the reduction process. The midpoint potential ðE o0 Þ of AmPDH was determined using the standard Nernst equation as described previously [72]. For variants H103A and H556A, methylene blue ðE o0 ¼ 11mVÞ was used as standard dye. The variants and dye were monitored at 460 nm (H103A), 465 nm (H556A) and 664 nm (methylene blue), where the monitored absorbance of variant H103A or H556A did not interfere with the absorbance of the dye and vice versa. Consequently, the concentrations of dye and variant could be readily obtained by the absorbance at the indicated wavelength. For both variants, these concentrations were used to calculate the ratio of reduced and oxidized species at various stages during the reduction process. Because an intermediate species with an absorbance maximum of 422 nm (H103A) or 368 nm (H556A) was observed for both variants, subsequent calculations employing the standard Nernst equation [72] yielded the semiquinone/reduced enzyme half potential (E o0 2 ). E o0 2 was used to calculate E o0 for both variants according to the equations [71,73,74]: where M is the maximum fraction of thermodynamically stable semiquinone formed, T is 298 K, E o0 is the twoelectron reduction potential, E o0 1 is the oxidized enzyme/ semiquinone half potential and E o0 2 is the semiquinone/ reduced enzyme half potential. The amount of thermodynamically stable semiquinone (M) was quantified in mixtures with compositions similar to those described above. In these mixtures, the standard dye was replaced by 50 lL of a 2 mM allopurinol solution, a potent xanthine oxidase inhibitor [75], which was pipetted into the second side arm of the cuvette. The xanthine and xanthine oxidase reduction system was stopped after semiquinone absorption reached its maximum (SQ MAX ) at 422 nm (H103A) or 368 nm (H556A) by adding the allopurinol solution from the side arm of the cuvette. The absorbance at 422 nm (H103A) or 368 nm (H556A) was recorded again after letting the reaction mixture stand at room temperature in the dark overnight (SQ ON ). Finally, M was obtained by calculating SQ ON /SQ MAX .

O 2 reactivity
The steady-state oxygen reactivity for all purified enzymes was determined in quadruplicate with a fluorimetric Amplex Red/horseradish peroxidase assay using the calibration curve described previously [19]. In the present study, 25 and 100 lM of the electron donor GLC were used to compensate for rather high apparent K M,GLC values for some variants. The obtained oxygen reactivities were essentially the same for both GLC concentrations. Therefore, only oxygen reactivities for 25 lM GLC are reported. The measured H 2 O 2 generation (min À1 Ámg À1 enzyme) was converted via the calibration curve into the oxygen reactivity, which is given as lM O 2 min À1 Ámg À1 enzyme.

Supporting information
Additional supporting information may be found in the online version of this article at the publisher's web site: Table S1. Nucleotide sequences of the primers used.