Laboratory-evolved Vanillyl-alcohol Oxidase Produces Natural Vanillin*

The flavoenzyme vanillyl-alcohol oxidase was sub-jected to random mutagenesis to generate mutants with enhanced reactivity to creosol (2-methoxy-4-methylphe-nol). The vanillyl-alcohol oxidase-mediated conversion of creosol proceeds via a two-step process in which the initially formed vanillyl alcohol (4-hydroxy-3-methoxyben-zyl alcohol) is oxidized to the widely used flavor compound vanillin (4-hydroxy-3-methoxybenzaldehyde). The first step of this reaction is extremely slow due to the formation of a covalent FAD N-5–creosol adduct. After a single round of error-prone PCR, seven mutants were generated with increased reactivity to creosol. The single-point mutants I238T, F454Y, E502G, and T505S showed an up to 40-fold increase in catalytic efficiency ( k cat / K m ) with creosol com- pared with the wild-type enzyme. This enhanced reactivity was due to a lower stability of the covalent flavin-substrate

The increased use of enzymes and other proteins in the pharmaceutical, chemical, and agricultural industry has generated considerable interest in the design of proteins with new or improved properties. Two different but complementary technologies have been applied to this goal: (i) rational design, which relies on the availability of the three-dimensional structure and knowledge about the relationship between sequence, structure, and mechanism, and (ii) directed evolution methods, which use random mutagenesis of the gene encoding the protein or recombination of gene fragments to create diversity and then experimental screening of the libraries generated for the desired properties. Rational design has been used to elucidate and change enzyme mechanism, substrate and product specificity, enantioselectivity, cofactor specificity, and protein stability (1)(2)(3). Directed evolution has been applied to increase catalytic activity; to invert or improve enantioselectivity; and to alter substrate and product specificity, protein stability, pH optimum, and tolerance against organic solvents (1, 4 -7). An obvious advantage of directed evolution methods over sitedirected mutagenesis is that enzymes can be tailored for the production of (intermediate) products without detailed knowledge of protein structure and structure-function relationships.
In this study, we engineered the flavoenzyme vanillyl-alcohol oxidase (VAO) 1 by random mutagenesis such that the evolved mutants are capable of producing natural vanillin (4hydroxy-3-methoxybenzaldehyde) from the precursor creosol (2-methoxy-4-methylphenol). Vanillin is a widely used flavor compound in food and personal products, with an estimated annual worldwide consumption of over 2000 tons (8,9). Moreover, vanillin displays antimicrobial and antioxidant properties and is used as a food preservative and for medicinal purposes (10 -15). Natural vanilla flavor from the orchid Vanilla planifolia (16,17) supplies Ͻ1% of the total demand for vanillin. Therefore, because of the increasing interest in natural products, alternative processes are being developed to produce natural vanillin (15, 18 -20). One possible approach involves the use of enzymes such as VAO (21). This flavoprotein is able to produce vanillin from creosol (see Fig. 1) and, if optimized for this purpose, would provide an attractive alternative for the production of the natural form of vanillin.
VAO is an oxidase from the ascomycete Penicillium simplicissimum, containing a covalently bound FAD cofactor. The enzyme is the prototype of a large family of structurally related oxidoreductases sharing a conserved FAD-binding domain (22,23). VAO catalyzes the oxidation of a wide range of phenolic compounds, including 4-alkylphenols (24,25). The catalytic cycle of VAO with 4-alkylphenols consists of two half-reactions (see Fig. 1) (26,27). In the reductive half-reaction, FAD is reduced by the aromatic substrate with the concomitant formation of a p-quinone methide product intermediate. In the oxidative half-reaction, FAD is reoxidized, and the p-quinone methide reacts with water to yield the corresponding 1-(4Јhydroxyphenyl) alcohol. The formed alcohol product can be further oxidized by VAO to 1-(4Ј-hydroxyphenyl)alkanone via a similar mechanism.
VAO has many properties in common with p-cresol methylhydroxylase (PCMH), a heterotetrameric flavocytochrome in-volved in the degradation of aromatic compounds in Pseudomonas species (28). The flavoprotein subunit of PCMH shares 32% sequence identity with VAO, and their active sites are remarkably conserved. Superposition of the crystallographic models of VAO (Protein Data Bank code 1VAO) (23) and PCMH (code 1DIQ) (29) has revealed a root mean square deviation of 1.0 Å for 470 C-␣ atoms. The two enzymes catalyze similar reactions, but p-cresol (4-methylphenol), the physiological substrate of PCMH (30), is an extremely poor substrate for VAO (27). Kinetic studies have shown that the low activity of VAO with p-cresol and also with creosol is due to the formation of an air-stable flavin N-5 substrate adduct (21,27,31). From the available crystallographic data, it is not obvious which structural features are involved in determining the different reactivity of both enzymes to 4-methylphenols. The only non-conservative amino acid substitution in the active site is found near the reactive carbon atom of the substrate. This position in VAO is occupied by Thr 457 , whereas the equivalent residue in PCMH is Glu 427 . Studies with site-directed mutants have shown that residue 457 is not involved in the reactivity with 4-methylphenols, 2 but is involved in determining the enantioselectivity of VAO (32).
In this study, we aimed to obtain, by directed evolution, VAO mutants with enhanced reactivity to creosol. To this end, we developed a library screening assay for the production of vanillin and used error-prone PCR to evolve VAO. Selected singlepoint mutant enzymes with improved reactivity to creosol were characterized by kinetic studies and x-ray crystallography.
Error-prone PCR Mutagenesis-Plasmid pBC11 contains the vao gene with a silent mutation at position 882, introducing a SalI restriction site (33). For random mutagenesis, the vao gene in pBC11 was amplified using manganese-based error-prone PCR mutagenesis (34,35). 25 l of a PCR mixture containing 10 pmol of primers 5Ј-GTA-AAACGACGGCCAGT-3Ј and 5Ј-CAGGAAACAGCTATGAC-3Ј, 90 ng of template DNA, and 250 or 500 pmol of MnCl 2 was heated for 3 min at 95°C and, after cooling down to 80°C, mixed with 25 l of a mixture containing 0.4 mM each dATP and dGTP, 2 mM each dCTP and dTTP, 2.5 units of Taq DNA polymerase, 20 mM Tris-HCl (pH 8.85), 4 mM MgCl 2 , 50 mM KCl, and 10 mM (NH 4 ) 2 SO 4 in a thin-wall PCR tube. 30 cycles were carried out using a thermal cycler with the following steps: 45 s at 94.5°C, 60 s at 48°C, and 120 s at 72°C. 2 l of this PCR mixture was used in a second PCR with the same setup, except that MnCl 2 was replaced with 2 nmol of dITP. The final PCR products were digested with the restriction enzymes PstI and KpnI, and the resulting fragments were subcloned into the corresponding sites of the pUC19 vector and transformed into E. coli TG2 cells by electroporation. The colonies were grown overnight on LB plates containing 100 g/ml ampicillin at 37°C.
Preparation of Library, Screening, and Enzyme Purification-Single colonies of transformed E. coli cells were transferred using a robotic system in 96-well plates (master plates) containing 200 l of LB medium and 100 g/ml ampicillin. After 48 h of growth at 37°C and the addition of 10% (v/v) glycerin, the master plates were stored at Ϫ80°C. For screening, the master plates were replicated onto 96-well plates containing 200 l of LB medium, 100 g/ml ampicillin, and 20 g/ml isopropyl-␤-D-thiogalactopyranoside. After 40 h of growth at 37°C, 50-l aliquots of the cells were screened for creosol conversion at pH 8 and 10 and for vanillyl alcohol conversion at pH 8. The activity of the cells was measured by following absorption spectral changes due to the conversion of creosol and vanillyl alcohol into vanillin at 340 nm at a substrate concentration of 750 M. Of 9600 clones, seven clones with the highest activity for creosol were selected and grown overnight at 37°C in 5 ml of LB medium supplemented with 100 g/ml ampicillin and 20 g/ml isopropyl-␤-D-thiogalactopyranoside. Cells from the 5-ml culture were spun down and resuspended in 50 mM potassium phosphate (pH 7.5). The cells were disrupted by sonication, and the cell extracts were collected after centrifugation. The activity of the cell extracts for creosol and vanillyl alcohol was measured as described below. Next, the cell extracts were analyzed by SDS-PAGE, and the amount of VAO in the cell extracts was estimated by comparing the fluorescence of the covalently bound FAD cofactor of VAO in the cell extracts with the fluorescence of known amounts of purified VAO in a gel incubated in 5% (v/v) acetic acid. The seven clones with the highest activity for creosol were sequenced. Wild-type VAO and its mutants were overexpressed and purified according to a previously established procedure (33,36). The enzyme purity was checked by SDS-PAGE and by size-exclusion chromatography.
Analytical Methods-All experiments were performed in 50 mM potassium phosphate (pH 7.5) or 50 mM glycine/potassium hydroxide (pH 10) at 25°C. SDS-PAGE was carried out on 12.5% slab gels essentially as reported previously (37). Coomassie Brilliant Blue R-250 was used for protein staining. Before protein staining, gels were incubated in 5% (v/v) acetic acid for fluorescence detection of covalently bound FAD (38). Analytical size-exclusion chromatography was performed with a Superdex 200 PG 10/30 column (Amersham Biosciences) (27). Absorption spectra were recorded using a Hewlett-Packard HP 8453 diode array spectrophotometer. Fluorescence spectra were recorded on a Cary Eclipse fluorescence spectrophotometer using an excitation wavelength of 360 nm. HPLC experiments were performed with an Applied Biosystems pump equipped with a Waters 996 photodiode array detector and an Alltech C 18 column (4.6 ϫ 150 mm) essentially as reported previously (25).
Stopped-flow kinetics was performed with a Hi-Tech SF-51 apparatus equipped with a Hi-Tech SU-40 spectrophotometer (26). In anaerobic FAD reduction experiments, glucose-containing enzyme solutions were flushed with oxygen-free argon gas, and glucose oxidase was added to eliminate final traces of oxygen.
For electrospray ionization mass spectrometry, VAO samples were prepared in 50 mM ammonium acetate (pH 6.8). Enzyme samples were introduced into the nanoflow electrospray ionization source of a Micromass LCT mass spectrometer (Waters Associates) operating in positive ion mode. Source pressure conditions and electrospray voltages were optimized for transmission of larger VAO assemblies (Pirani pressure, 6 millibars; capillary voltage, 1450 -1650 V; and cone voltage, 45-100 V) (39). Borosilicate glass capillaries (Kwik-Fil, World Precision Instruments, Inc.) were used on a P-97 puller (Sutter Instruments Co.) to prepare the nanoflow electrospray capillaries with an orifice of ϳ5 m. The capillaries were subsequently coated with a thin gold layer (ϳ500 Å) using an Edwards Scancoat 6 Pirani 501 sputter coater.
Crystallization and Structure Determination-Crystals of four VAO mutants were grown at 20°C by the hanging-drop vapor diffusion method. The drops contained 4 l of an equal mixture of protein solution (6.5-13 mg/ml) in 50 mM potassium phosphate (pH 7.5) and reservoir solution containing 100 mM sodium acetate/hydrochloride (pH 5.1) 2 R. H. H. van den Heuvel and W. J. H. van Berkel, unpublished data. and 4 -6% (w/v) polyethylene glycol 4000. Hanging drops were allowed to equilibrate against 1 ml of reservoir solution for 4 days until yellow crystals appeared. Enzyme-inhibitor complexes were prepared by soaking the enzyme crystals in the reservoir solution containing 5 mM isoeugenol for 1 h.
Diffraction data of the VAO mutant crystals were collected using a Rigaku R-AXIS IV image plate detector mounted on an RU-200B rotating anode CuK ␣ x-ray generator equipped with an MSC confocal optic system. VAO crystals were placed for a few seconds in a cryoprotectant solution containing sodium acetate/hydrochloride (pH 5.1), 4 -6% (w/v) polyethylene glycol 4000, and 20% (v/v) glycerol. VAO crystals had the symmetry space group I4 and two molecules in the asymmetric unit. The data were processed with MOSFLM (40) and further processed with programs from the CCP4 package (41). Data processing statistics are reported in Table I. The mutant crystals are isomorphous to those of wild-type crystals. The final model of free wild-type VAO (Protein Data Bank code 1VAO) (23) was used as the starting model for refinement of the evolved VAO mutants. The subsequent refinement of the initial model consisted of alternating rounds of manual fitting of the model to electron density maps using the program O (42) and maximum likelihood refinement with REFMAC (43). 5% of the data were set aside to compute R free (44). R free was used to monitor the progress of the refinement. Refinement statistics are shown in Table I. The final models of the VAO mutants consist of two polypeptide chains (A and B) with a total of 1098 residues (A6 -A41, A47-A560, B6 -B41, and B47-B560) and two FAD and two isoeugenol molecules.

Directed Evolution of VAO and Characterization of Mutant
Enzymes-Random mutations were introduced into the wildtype vao gene coding for 560 amino acids by error-prone PCR. We used manganese concentrations designed to generate one to four mutations per vao gene to produce an average of one amino acid substitution per VAO molecule (45). Sequence determination of 42 randomly picked colonies showed that the evolved mutations were divided equally over the vao gene; however, we found more A and T mutations (70% of the total) than G and C mutations (30% of the total). After one round of mutagenesis, the activity of the clones for creosol was determined at pH 8 and 10 at saturating substrate concentrations. The molar absorption coefficients of vanillin at pH 8 and 10 are 16.0 and 23.0 mM Ϫ1 cm Ϫ1 , respectively. From 9600 clones, we selected seven clones with an increased reactivity to creosol (Table II). Sequencing revealed that three mutant enzymes shared the amino acid substitution I238T, whereas all other mutations were found only once. The turnover rate at pH 10 in the cell extracts of five single mutants, one double mutant, and one triple mutant was 4 -7-fold higher compared with that of the wild-type enzyme (Table II).
To study the rationale for the increased reactivity to creosol, we proceeded with the four single-point mutants I238T, F454Y, E502G, and T505S. These mutants were overexpressed in E. coli, purified, and analyzed in detail. The evolved mutants were, like wild-type VAO, expressed to ϳ2.5% of the total protein in E. coli strain TG2. The purified enzymes were bright yellow, and enzyme purity was Ͼ95% as estimated by SDS-PAGE. Size-exclusion chromatography revealed that the mutant enzymes were, like the wild-type enzyme, in equilibrium between dimeric and octameric forms (38). Next, we studied the hydrodynamic behavior of the VAO variants at low enzyme concentrations (1 and 4 M) by electrospray mass spectrometry. The resulting data show that F454Y and T505S have similar hydrodynamic behavior compared with the wild-type enzyme (39). The most abundant species were found to have octameric assembly with lower amounts of dimeric assembly (octamer/ dimer ratio of ϳ1.5:1 for F454Y, T505S, and wild-type VAO). The quaternary structures of I238T and E502G were different from that of the wild-type enzyme. I238T appeared to be present mainly in the octameric form (octamer/dimer ratio of 4:1), and E502G was almost present completely as a dimer (octamer/ dimer ratio of 1:10). X-ray crystallography has shown that VAO His 422 is covalently linked to the FAD prosthetic group via a FAD C-8␣-  His 422 N-⑀3 bond (23). Fluorescence analysis of unstained SDSpolyacrylamide gels in 5% (v/v) acetic acid showed that the four mutants were as fluorescent as the wild-type enzyme, strongly indicating that the FAD is covalently bound to His 422 in all four mutants. The flavin absorption spectral properties of the VAO mutants were virtually identical to those of the wild-type enzyme, with maxima at 355 and 439 nm for I238T, F454Y, and T505S and at 357 and 442 nm for E502G (356 and 439 nm for the wild-type enzyme). Moreover, the absorption ratio between 280 and 439 nm for the mutants was similar to the reported value for wild-type VAO (12.0 -12.5).
Catalytic Properties of VAO Mutants-The four purified VAO mutants were selected on the basis of their reactivity to creosol. Tables III (pH 7.5) and IV (pH 10) show that the catalytic efficiencies (defined as k cat /K m ) of the four mutants for this substrate were indeed 4 -11-fold higher at pH 7.5 and 20 -40fold higher at pH 10 than those of wild-type VAO. The increased catalytic efficiency was due mainly to an increase in turnover rate and not to a lowered Michaelis constant. This is in agreement with the screening procedure in which we used saturating substrate conditions of 750 M creosol. Apart from the enhanced reactivity to the substrate for which we screened, the evolved mutant enzymes had remarkably different catalytic properties compared with wild-type VAO (Tables III and  IV). The ortho-substituted 4-methylphenol derivatives 2-methyl-p-cresol and 2-amino-p-cresol were converted with higher catalytic efficiencies compared with wild-type VAO, especially at pH 10. On the other hand, 2-chloro-p-cresol was a substrate neither for the mutants nor for wild-type VAO. p-Cresol, the physiological substrate for PCMH, is a very poor substrate for VAO (27). Intriguingly, the mutant enzymes performed worse than wild-type VAO with this substrate (turnover rates Ͻ 0.001 s Ϫ1 ). It should be stressed here that the investigated 4-methylphenol derivatives react, unlike other VAO substrates, via a two-step reaction to the corresponding aldehyde ( Fig. 1) and that the steady-state kinetic assay monitored only the production of the aldehyde.
Like wild-type VAO, the evolved mutants had a relaxed substrate specificity for other 4-hydroxybenzylic compounds (24,25). Since vanillyl alcohol is the intermediate product formed upon hydroxylation of creosol, it was of special interest to study the reactivity of the mutant enzymes with this substrate. The catalytic efficiencies at both pH 7.5 and 10 with vanillyl alcohol were similar compared with wild-type VAO, despite a considerably lower maximum turnover rate (Tables  III and IV). The supposed physiological substrate 4-(methoxymethyl)phenol (26) was oxidized efficiently by the mutant enzymes. In fact, the catalytic efficiencies at pH 10 with this substrate were 4 -9-fold higher compared with the wild-type enzyme (Table IV). As 4-allylphenols are among the best substrates for VAO (24), it was also of interest to study the reactivity of the mutant enzymes with eugenol. At pH 7.5, wild-type VAO converted eugenol to coniferyl alcohol (4-hydroxy-3-methoxycinnamyl alcohol) with a far higher catalytic efficiency than the mutants (10 -250-fold), which was caused mainly by dramatically decreased turnover rates (Table III). At pH 10, however, wild-type VAO and I238T showed similar catalytic efficiencies with eugenol, although the maximum turnover rate of I238T was 65-fold lower than that of the wild-type enzyme (Table IV). The other three mutants displayed lower catalytic efficiencies with eugenol, which was again caused by strongly decreased turnover rates. In summary, the catalytic efficiency of the evolved VAO mutants was significantly increased for ortho-substituted 4-methylphenols. However, the reactivity to p-cresol and eugenol was decreased by the mutations.
Reduction of wild-type VAO (VAO ox ) in an anaerobic environment by 4-(methoxymethyl)phenol (S) results in a stable complex between the reduced enzyme and the p-quinone methide of the substrate (VAO red ⅐Q), which is described by Equation 1.
Under anaerobic conditions at pH 7.5, wild-type VAO is reduced by 4-(methoxymethyl)phenol in a single irreversible step (k 2 ϭ 3.3 s Ϫ1 at saturating substrate conditions). The rate of reduction equals the maximum turnover rate of wild-type VAO with this substrate (26). Anaerobic stopped-flow experiments showed that 4-(methoxymethyl)phenol-mediated reduction of the four VAO mutants also occurred in a single exponential process with the formation of a stable complex between the reduced enzyme and the p-quinone methide intermediate. With all mutants, the rates of reduction at saturating substrate concentrations and pH 7.5 were in the same range as the turnover rates (k 2 ϭ 2.9 s Ϫ1 for I238T, k 2 ϭ 1.9 s Ϫ1 for F454Y, k 2 ϭ 2.0 s Ϫ1 for E502G, and k 2 ϭ 2.5 s Ϫ1 for T505S), suggesting that the mutant enzymes have the same mechanism of action with 4-(methoxymethyl)phenol as the wild-type enzyme. Reduction of VAO by creosol can also be described by Equation 1; however, with this substrate, the reductive half-reaction

Directed Evolution of Vanillyl-alcohol Oxidase
does not limit catalysis (21). Table V summarizes the data of the reductive half-reaction of wild-type VAO and its mutants with creosol at pH 7.5 and 10. When creosol was mixed with wild-type VAO, the FAD cofactor reduced in a single exponential step. However, when the kinetic data were analyzed, a best fit was obtained when an apparent "initial" reduction rate at infinite low substrate concentration was taken into account (46) (Equation 2).
This suggests that the reduction process is reversible (k Ϫ2 Ͼ 0) and that the creosol-mediated reduction of the FAD cofactor results in an equilibrium in which the flavin is partially in the oxidized state. The rate of enzyme reduction was dependent on the substrate concentration, with a dissociation constant (k Ϫ1 / k 1 ) of 188 M at pH 7.5. At low creosol concentrations, the reduction rate reached the finite value of 0.18 s Ϫ1 (k Ϫ2 ). The reduction rate at saturating substrate concentration was calculated to be 0.57 s Ϫ1 (k 2 ) and was 29-fold higher than the turnover rate, which indicates that the reduction rate does not determine the rate of overall catalysis. As with the wild-type enzyme, the creosol-mediated reduction of the mutants is best described by a reversible reduction process. However, the maximum reduction rates of F454Y (k 2 ϭ 0.25 s Ϫ1 ) and T505S (k 2 ϭ 0.36 s Ϫ1 ) with creosol at pH 7.5 were in the same order of magnitude as their turnover rates (kЈ cat ϭ 0.14 and 0.12 s Ϫ1 , respectively) (Fig. 2), suggesting that the reduction may partially limit catalysis at neutral pH. The calculated reverse reduction rates of the mutants were comparable with that of wild-type VAO (k Ϫ2 ϭ 0.24 and 0.18 s Ϫ1 for F454Y and T505S, respectively). The reduction rates of the mutant enzymes with creosol increased with pH and became far higher than the turnover rates (Table V). Thus, at basic pH values, the reduction rate does not limit catalysis for wild-type VAO, F454Y, and T505S. FAD Substrate Adduct in VAO Mutants-Enzyme-monitored turnover experiments and fluorescence spectroscopy have indi-cated that the FAD cofactor of wild-type VAO forms covalent flavin N-5 adducts with both p-cresol and creosol and that these adducts become more stable at higher pH (21,27). Because the evolved VAO mutants, especially at basic pH values, converted creosol to vanillin much better than the wild-type enzyme, it was of interest to investigate adduct formation with the mutant enzymes with these techniques. In enzyme-monitored turnover experiments, it was found that, at neutral pH, the FAD in wild-type VAO and E502G was for 47 and 45%, respectively, in the oxidized state (Table VI). On the other hand, the FAD in I238T, F454Y, and T505S was for Ͼ50% in the oxidized state. This suggests that, in the latter three mutants, the covalent adduct was less stabilized than in wild-type VAO and the E502G variant. At pH 10, the differences in the FAD redox state between wild-type VAO and the four mutants became more pronounced. Under these conditions, the relative amount of oxidized FAD in the four mutants was 2-3-fold larger than in the wild-type enzyme, indicative of a lower stability of the adduct (Table VI).
When adduct formation between protein-bound FAD and creosol was studied by fluorescence spectroscopy, mixing of (mutant) enzyme and substrate at pH 7.5 resulted in the initial formation of an emission spectrum with a maximum at 480 nm. This spectrum, indicative of the flavin N-5-creosol adduct (21), was stable for a few minutes, after which it changed to a spectrum with an emission maximum at 425 nm (Fig. 3). The latter spectrum represented vanillin, the fluorescent end product of the two-step oxidation of creosol. For both wild-type VAO  and its mutants, the covalent flavin N-5-substrate adduct became more stable at pH 10 ( Fig. 3), in good agreement with the apparent reduced state of the flavin cofactor observed during enzyme-monitored turnover experiments. However, as at pH 7.5, the adduct decay was much faster in the mutant enzymes. When adduct formation between the mutant enzymes and p-cresol was studied at pH 7.5 and 10, we observed fluorescence spectra with very stable emission maxima at 480 nm, indicat-ing the formation of abortive FAD N-5-p-cresol adducts. As for wild-type VAO, the FAD N-5-p-cresol adduct was stable for several minutes (27). In summary, enzyme-monitored turnover and fluorescence emission experiments showed that the four single-point mutations decreased the stability of the covalent adduct between the FAD cofactor and creosol; however, the four mutations did not have an effect on the stability of the flavin adduct formed in the reaction with p-cresol. These data agree well with the kinetic data (Tables III and IV).
Crystal Structures of VAO Mutants-The crystal structures of the evolved mutants I238T, F454Y, E502G, and T505S, refined to resolutions of 2.55, 2.7, 3.0, and 2.7 Å, respectively, are highly similar to that of wild-type VAO (Table I). However, the present medium resolutions do not allow the identification of very subtle structural perturbations due to the mutations. The root mean square deviations between the wild-type enzyme and I238T, F454Y, E502G, and T505S are 0.24, 0.26, 0.33, and 0.27 Å, respectively (Protein Data Bank codes 1w1k, 1w1l, 1w1m, and 1w1j). The electron densities of the structures of the mutant enzymes clearly confirmed the substitution of the different amino acids and the presence of the inhibitor isoeugenol. Fig. 4 displays the positions of the four amino acid replacements in the VAO monomer.
Ile 238 is positioned in a short ␣-helix in the FAD-binding domain at the interface between two VAO monomers. The introduction of Thr 238 did not have any effect on the conformation in the region around residue 238 in both monomers. Moreover, compared with the native model, no conformational shifts were found in the position of isoeugenol, FAD, or the residues lining the enzyme active site. As in wild-type VAO (23), the aromatic ring of the substrate analog isoeugenol stacks against the isoalloxazine ring of FAD at an angle of 14°with respect to the cofactor plane. The hydroxyl moiety of isoeugenol forms a hydrogen bond with the side chains of Tyr 108 , Tyr 503 , and Arg 504 , facilitating substrate deprotonation, and the propenyl group points toward Asp 170 . The C-␣ atom of the propenyl group is positioned at 3.4 Å from flavin N-5.
Phe 454 is situated in a loop, deeply buried within the cap domain. The F454Y mutation did not introduce any shifts  within the region around Tyr 454 , in the VAO active-site cavity, or in the orientation of bound isoeugenol (Fig. 5A). The extra hydroxyl group introduced by Tyr 454 potentially forms an ad-ditional hydrogen bond with the N-⑀2 atom of His 467 , which, together with its neighbor His 466 , forms part of the active site of VAO.
Glu 502 and Thr 505 are positioned in the FAD-binding domain in a loop close to the active-site cavity of VAO. In fact, the neighboring residues Tyr 503 and Arg 504 stabilize the phenolate form of the substrate by forming hydrogen bonds (23). The two mutations E502G and T505S did not induce any major conformational changes in this loop or in the side chains of Tyr 503 and Arg 504 (Fig. 5B). However, the E502G mutation had an effect on the hydrogen bond network, as two potential hydrogen bonds between the negatively charged side chain of Glu 502 and the carbonyl oxygens of Arg 504 and Ser 426 could not be formed. DISCUSSION The structurally related flavoenzymes VAO and PCMH have many catalytic properties in common, but show dramatic differences in their performance with 4-methylphenols. Of special interest is the production of the widely used flavor compound vanillin from the natural precursor creosol. Because PCMH is highly active with creosol and VAO is not, we undertook the enhancement of the reactivity of VAO with creosol by manganese-based error-prone PCR.
The key factor for a successful random mutagenesis experiment is a rapid and effective screening or selection procedure (47). Chromogenic substrates have been used successfully in a number of studies utilizing a 96-well plate reader or visual screening of the colored colonies (48,49). With VAO, the end product vanillin formed upon the two-step oxidation of creosol could be measured directly in whole cells by absorption spectroscopy at 340 nm. Both creosol and vanillin are able to diffuse efficiently through the E. coli membrane such that vanillin is produced in the intracellular environment and is monitored in the extracellular environment. The evolved VAO mutants were selected on the basis of their reactivity to creosol at both pH 8 and 10. With this screening procedure, we found seven clones with considerably increased reactivity to creosol. Interestingly, none of the mutations was positioned within a distance of direct interaction with the substrate. The I238T mutation was found three times, whereas the other mutations, F454Y, E502G, and T505S, were found only once. The high frequency of the I238T mutation was not due to sequence bias as proven by sequence determination of randomly picked colonies. On the basis of the mutational efficiency, this residue could have been replaced by several other residues; however, only the I238T mutation was detected. This indicates that position 238 is a hot spot in determining the reactivity to ortho-substituted 4-methylphenols.
Previous studies have shown that the poor reactivity of VAO to creosol is due to the formation of an abortive covalent FAD N-5-creosol adduct (21,27). Adduct formation was less favorable in the mutant enzymes presented here, resulting in considerably increased conversion rates, especially at basic pH values. Similar effects were observed with other ortho-substituted 4-methylphenol substrates, and rate enhancements of up to 80-fold occurred. The mutant enzymes had a relaxed specificity for other 4-hydroxybenzylic compounds, but the maximum turnover rates and the Michaelis constants for the various substrates varied among the different mutants and wild-type enzyme (Tables III and IV).
Intriguingly, p-cresol was, as for wild-type VAO, a very poor substrate for the generated mutants. Indeed, the mutants formed a very stable covalent flavin-p-cresol adduct, thereby preventing product formation. The x-ray model of wild-type VAO in complex with p-cresol has shown that formation of the flavin adduct is associated with distortion of the FAD ring, which deviates from planarity with an angle of 8°between the pyrimidine and dimethylbenzene rings. Moreover, p-cresol FIG. 5. Structural comparison of wild-type VAO with VAO mutants and PCMH. A, superposition of isoeugenol-complexed wild-type VAO (light gray) and isoeugenol-complexed F454Y (dark gray); B, superposition of isoeugenol-complexed wild-type VAO (light gray) and isoeugenol-complexed T505S (dark gray); C, superposition of p-cresolcomplexed wild-type VAO (light gray) and p-cresol-complexed PCMH (dark gray). The figures were prepared with MOLSCRIPT (56) and Raster3D (57). makes an angle of 34°with the flavin, whereas the noncovalently bound inhibitors isoeugenol and 2-nitro-p-cresol are less tilted with respect to the FAD. From this it was argued that steric restrictions imposed by the shape of the active-site cavity are a key factor in preventing enzyme inactivation through covalent adduct stabilization (23). The x-ray models of I238T, F454Y, E502G, and T505S did not reveal any conformational perturbations in the active-site cavities, and the orientation of the flavin and isoeugenol was highly conserved. In addition, the x-ray models did not suggest structural perturbations in regions far from the catalytic center. It should be stressed, however, that the medium resolution (2.55-3.0 Å) of the presented x-ray structures is not sufficient to detect very subtle perturbations. This conservation of structural features might explain why all mutant enzymes formed very stable flavin-p-cresol adducts. However, the structural data do not provide us with a clue at to why only the wild-type enzyme formed a stable adduct with creosol.
The active-site topology of VAO is highly similar to that of PCMH, with several conserved key residues (Fig. 5C) (29). The closest distance between p-cresol C-7 and FAD N-5 in the PCMH crystallographic model is 3.1 Å, and the orientation of the phenolic ring of the PCMH substrate is similar to that of the noncovalent phenolic inhibitors of VAO. Tyr 108 , Tyr 503 , and Arg 504 in VAO (Tyr 95 , Tyr 473 , and Arg 474 in PCMH) are involved in the deprotonation of the substrate phenol. Arg 504 / Arg 474 are also proposed to stabilize the anionic form of the reduced FAD. As suggested for His 436 , Glu 177 , and Asp 434 in PCMH (29), His 466 , Asp 192 , and Glu 464 in VAO may form a proton transfer pathway during the formation of the quinone methide intermediate product. The most significant difference between the active sites of VAO and PCMH is the arrangement of acidic residues. Asp 170 of VAO has no counterpart in PCMH, but its role is taken over by Glu 380 . Furthermore, PCMH has a second acidic residue (Glu 427 ) within the active-site cavity, which is not present in VAO (Thr 457 ). Previously, we created the VAO double mutant D170E/T457E to mimic the active site of PCMH; however, these mutations do not have an effect on the tendency of VAO to form adducts with 4-methylphenols (32). Thus, the comparison of VAO with PCMH does not rationalize the adduct formation of VAO with p-cresol and creosol.
The kinetic and structural data presented in this work clearly show that residues distal from the active site determine the reactivity to creosol. This is in line with recent studies (5, 50 -53) indicating that enzymes exert their functions not only through the chemical properties of the side chains of the amino acid residues that contact the cofactor or the substrate, but also through long-range effects such as hydrogen bond networks and electrostatic interactions.
The I238T mutation was found in three of seven mutants with enhanced reactivity to creosol. Ile 238 is positioned at the VAO monomer-monomer interface at a distance of 33 Å from FAD N-5. The I238T mutation did not induce any major structural change and did not prevent octamerization. This implies that a residue very far from the active-site cavity without any apparent catalytic function plays a role in VAO catalysis. Intriguingly, the counterpart of this residue in PCMH is Thr 207 (29), which is also positioned at the monomer-monomer interface.
Phe 454 is also positioned far (16 Å) from FAD N-5, but is localized within the cap domain. The F454Y mutation is conserved in PCMH (Tyr 424 ). The hydroxy moiety of the Tyr 454 side chain is hydrogen-bonded to the strictly conserved His 467 (His 437 in PCMH), which is pointing toward the active-site cavity and the FAD cofactor (Fig. 5A). In fact, the neighboring residue His 466 lines the active-site cavity of VAO. This suggests that the F454Y mutation might influence catalysis by introducing subtle changes in electrostatic interactions and possible backbone flexibility. We have no indication that Tyr 454 and His 467 are involved in an extended hydrogen bond network.
Both Glu 502 and Thr 505 are positioned in a loop lining the VAO catalytic center and are a relatively short distance from FAD N-5: 11.8 Å for Glu 502 O-⑀2 and 12.4 Å for Thr 505 C-␣, The generated mutations at these positions (E502G and T505S) are not conserved in PCMH (Val 472 and Val 475 ). Due to the E502G mutation, a negative charge near the enzyme catalytic center is removed as well as two possible hydrogen bonds between the side chain of Glu 502 and the carbonyl oxygens of Arg 504 and Ser 426 . The T505S mutation had no effect on neighboring hydrogen bond interactions. It should be mentioned here that the active-site cavity of VAO is completely solvent-inaccessible (23). Close inspection of the VAO x-ray model does not suggest an obvious structural element whose conformational change may allow substrate admission. This implies that the substrate enters the active-site cavity by "breathing" of the enzyme. We speculate that this property of VAO is somewhat changed in the evolved mutants.
In conclusion, by error-prone PCR, we created four singlepoint mutants distal from the VAO active-site cavity with enhanced reactivity to creosol and other ortho-substituted 4-methylphenols. Given the small differences in structural properties, we are not able to offer a clear explanation for the different activity changes encountered with different substrates. Nevertheless, our results demonstrate the power of directed evolution to generate mutations that would not readily be predicted by site-directed mutagenesis. Moreover, we have confirmed earlier findings that changing enzyme activity requires remodeling of the active-site cavity not only by mutations of residues within the active-site pocket, but also by distal mutations. The results obtained in this study with single-point mutations can serve as a starting point to investigate possible cumulative effects of residues remote from the active-site cavity by performing successive rounds of error-prone PCR. An interesting approach to investigate possible long-range structural perturbations in VAO (not detected in this study) is the characterization of single and double mutants of two mutations that are known to have a stimulating effect on creosol reactivity, e.g. T207S and E502G in VAO. In the case of structural perturbations caused by the single mutations, it is likely that the effect of the double mutation is non-additive due to different structural perturbations. This principle has been demonstrated for some residues of dihydrofolate reductase (53,54).