Crystal Structure of Glucooligosaccharide Oxidase from Acremonium strictum

Glucooligosaccharide oxidase from Acremonium strictum has been screened for potential applications in oligosaccharide acid production and alternative carbohydrate detection, because it catalyzes the oxidation of glucose, maltose, lactose, cellobiose and cello- and maltooligosaccharides. We report the crystal structures of the enzyme and of its complex with an inhibitor, 5-amino-5-deoxy- cellobiono-1,5-lactam at 1.55- and 1.98-Å resolution, respectively. Unexpectedly, the protein structure demonstrates the first known double attachment flavinylation, 6-S-cysteinyl, 8α-N1-histidyl FAD. The FAD cofactor is cross-linked to the enzyme via the C6 atom and the 8α-methyl group of the isoalloxazine ring with Cys130 and His70, respectively. This sugar oxidase possesses an open carbohydrate-binding groove, allowing the accommodation of higher oligosaccharides. The complex structure suggests that this enzyme may prefer a β-d-glucosyl residue at the reducing end with the conserved Tyr429 acting as a general base to abstract the OH1 proton in concert with the H1 hydride transfer to the flavin N5. Finally, a detailed comparison illustrates the structural conservation as well as the divergence between this protein and its related flavoenzymes.

Sugar oxidases and dehydrogenases that catalyze carbohydrate oxidation to the corresponding lactones are of considerable commercial importance as potential diagnostic reagents and industrial biocatalysts. For example, glucose oxidase (GOX) 4 is widely used in analytical bio-chemistry and in the food industry (1). A search of the BRENDA enzyme data base (2) reveals that most of these enzymes are specific for a variety of mono-and disaccharides, and only a few enzymes are highly selective for oligosaccharides. These include galactose oxidase, cellobiose dehydrogenase (CDH), and glucooligosaccharide oxidase (GOOX).
GOOX from Acremonium strictum was screened with the aim of identifying enzymes with potential applications in oligosaccharide acid production and alternative carbohydrate assays (3). It is a monomeric glycoprotein with a covalently linked FAD and catalyzes the oxidization of a variety of carbohydrates with the concomitant reduction of molecular oxygen to hydrogen peroxide. A screening of more than 50 carbohydrates and derivatives showed that D-glucose, maltose, lactose, and cellobiose are good substrates. In addition, this enzyme can react with malto-and cellooligosaccharides, and hence the name of this novel oxidase. The broad substrate specificity of GOOX, particularly toward oligosaccharides, suggests that it may have great potential applicability.
To facilitate further characterization and potential industrial use of A. strictum GOOX, we have cloned the encoding gene, which is composed of a 25-residue signal peptide and a 474-residue mature protein (4). Although GOOX shows a similar substrate specificity as GOX, CDH, and pyranose oxidase (POX), it shares no sequence similarity with them. However, GOOX displays significant sequence homology to the FADbinding domain (F domain) of berberine bridge enzyme-like proteins including three other sugar oxidases, a red alga hexose oxidase (CcHEOX), a tobacco nectar protein (nectarin V, NlNEC5), and a sunflower defense protein (HaCHOX) (5)(6)(7). Interestingly, although NlNEC5 catalyzes oxidation of glucose, and GOOX, CcHEOX and HaCHOX catalyze the oxidation of glucose, maltose, lactose, and cellobiose, the protein sequences of their substrate-binding domains (S domains) are quite diverse and apparently lack conserved carbohydrateinteracting residues. Moreover, structural and mutational studies demonstrated a consensus histidine for flavinylation in the F domain of berberine bridge enzyme-like proteins. Unexpectedly, mutation of the putative FAD-linking residue in GOOX, His 70 , to alanine, serine, cysteine, or tyrosine does not abolish the covalent FAD attachment (4). To gain structural insight into the FAD incorporation, substrate specificity, and catalytic mechanism, we have determined the GOOX structure in the absence, or presence of a product analog, 5-amino-5-deoxy-cellobiono-1,5-lactam (ABL).

MATERIALS AND METHODS
Recombinant protein was expressed using the vector pPICZ␣A in Pichia pastoris KM71, and then isolated by a Toyopearyl Phenyl-650 column with elution of a gradient of 2.5-0 M ammonium sulfate (4). Yellow protein crystals were grown in 25% polyethylene glycol mono-methyl ether 550, 10 mM ZnSO 4 , and 0.1 M MES (pH 6.5), with a combination of 2 l of reservoir solution and 2 l of protein solution (35 mg/ml). Crystals appeared and reached their final dimensions in a month at 295 K. Only very few batches of recombinant proteins produce crystals of high quality. The high-resolution datasets were collected using the synchrotron radiation at the beamlines BL12B2 at SPring-8 (Harima, Japan) and NW12 at Photon Factory (Tsukuba, Japan). The data were processed using the HKL2000 program (8). The crystals belong to the P2 1 2 1 2 1 space group with 1 monomer/asymmetric unit. Because of a lack of fresh crystals, the frozen native crystals were used for preparation of the ABL derivative by soaking crystals for 30 min in a reservoir solution containing 60 mM ABL. The x-ray data were collected on a RAXIS IV area detector with a Rigaku RU-300 rotating anode (MSC).
Heavy atom screening yielded one platinum derivative (K 2 PtI 4 ) and sufficient phase information was derived from single-wavelength anomalous dispersion. Four Pt positions were identified with SHLEXD (9) and refined with SOLVE (10). The initial phase was improved significantly by direct method phasing refinement using OASIS (11) before density modification. An initial model with 248 polyalanine residues was built by RESOLVE (10). The electron density was improved greatly by fragment extension with OASIS, so that a complete model with side chains was built by ARP/wARP (12). The structure then underwent straightforward refinement against native data to 1.55 Å resolution using CNS (13). The structure of the ABL derivative was solved by molecular replacement using the refined coordinates mentioned above as the initial model. Statistics for data collection and refinement are summarized in TABLE ONE. More than 90% of the residues are in the most favored regions of Ramachandran plot, with the remaining in the additional allowed regions except for Thr 307 and Arg 445 , because of hydrogen bond interactions. Crystals with better quality were grown in the presence of 10 mM ZnSO 4 , and four zinc ions bound to the enzyme were identified by the zinc anomalous data. Three zinc ions were replaced by platinum ions in the K 2 PtI 4 derivative. Figs. 1, 3B, and 4A were generated by MolScript (14) and Raster3D (15), Figs

RESULTS AND DISCUSSION
The Overall Structure-The current model contains five linker residues and the 474-residue mature GOOX. The protein is composed of two distinct domains (Fig. 1). The F domain recruits the N-terminal residues 1-206 and the C-terminal residues 421-474 and folds into two (␣ϩ␤) subdomains. A small subdomain, residues 1-90, comprises a central four ␤-strands (␤1-␤4) with the strand order 1, 2, 4, and 3, and with ␤1 running antiparallel to the other strands. The ␤-sheet is sandwiched by three helices (␣A, ␣AЈ, and ␣B), in which there is one disulfide bond between Cys 6 of the ␣A helix and Cys 55 of the ␣B helix. The second subdomain contains five antiparallel ␤-strands (␤5-␤9) with the strand order 5, 6, 9, 7, and 8 and is surrounded by five ␣-helices (␣C, ␣D, ␣DЈ, ␣J, and ␣K). These two subdomains are packed against each other and accommodate the FAD cofactor between them.
Furthermore, the S domain (residues 207-420) is composed of a large seven-strand antiparallel ␤-sheet (␤10 -␤16) with the strand order 13, 10, 12, 11, 15, 16, and 14 and is flanked by five helices (␣E-␣I). This domain is positioned over the isoalloxazine ring of the FAD cofactor and constitutes most of the carbohydrate-binding groove. Two N-glycosylation modifications were detected in the S domain. The electron density showed clearly one ordered sugar residue each linked to Asn 305 and Asn 341 , which was assigned as N-acetyl-D-glucosamine based solely on fitting to the electron density.
A Novel Flavinylation-Unexpectedly, the FAD cofactor is crosslinked to the enzyme at two attachment sites ( Fig. 2A). One is the S ␥ atom of Cys 130 bound to the C 6 atom of the isoalloxazine ring, whereas the other is the N ␦1 atom of His 70 bound to the 8␣-methyl group (6-Scysteinyl, 8␣-N1-histidyl FAD). The ADP-ribityl moiety is embedded in the F domain and is completely solvent-inaccessible. The adenine and ribose groups form eight hydrogen bonds with two water molecules and the backbone atoms of Ala 65 , Gly 106 , and Val 195 (Fig. 2B). The negatively charged pyrophosphate group interacts with a constellation of backbone NH groups of Gly 67 , Gly 68 , Gly 69 , His 70 , Ser 71 , Gly 134 , and Gly 137 . In addition, the ribityl group makes four hydrogen bonds with one water molecule, the backbone of His 70 , and the side chains of His 138 and Asn 428 . The isoalloxazine ring lies at the juncture of the F and S domains. In addition to the covalent linking to His 70 and Cys 130 , the ring forms five hydrogen bonds with two water molecules, the backbones of Thr 129 and Tyr 144 , and the side chain of Tyr 426 . Five types of covalent flavinylation have been identified up to the present, and flavinylation has been shown to be an autocatalytic process (18). The four types of flavinylation at the 8␣-methyl group cross-linking to His (N ␦1 and N ⑀2 ), Tyr (O ), and Cys (S ␥ ), were shown to be involved in modulation of the redox potential of the isoalloxazine ring. Notably and in contradistinction, an unusual 6-S-cysteinyl modification was only observed in the FMN cofactor of trimethylamine dehydrogenase and its homologues, dimethylamine dehydrogenase and histamine dehydrogenase (19,20). This unusual 6-S-cysteinyl attachment has been suggested to prevent enzyme inactivation by hydroxylation at the C 6 atom. Thus, GOOX possesses a novel form of covalent flavinylation; it is the first example of 6-S-cysteinyl FAD and the first double covalent linkage identified to date. Replacement of His 70 with alanine, serine, cysteine, or tyrosine decreased the k cat value 50 -600-fold but had little effect on the K m (4). These four His 70 mutants still contain a covalently linked FAD. These indicate that the covalent attachment via the 8␣-methyl group with His 70 is able to enhance the oxidative power of the flavin, as has been shown in other covalent flavoenzymes (18). Nevertheless, the functional role of the 6-S-cysteinyl flavinylation in GOOX is still under investigation.
Substrate Specificity-ABL resembles cellobionolactone, the oxidation product of cellobiose, with replacement of the endocyclic O 5 with an NH group (Fig. 3). This replacement suppresses the spontaneous hydrolysis of ABL to cellobionic acid. ABL is the only inhibitor against GOOX activity found to date. Approximately 50% of glucose oxidation activity was lost in the presence of 25 mM ABL. In addition, ABL has been shown to be an inhibitor for CDH. The structure of CDH in complex with this inhibitor has been solved and used to delineate a reaction mechanism for CDH (21). In the complex structure presented here, ABL is identified by its strong electron density that correlates well with the shape of D-glucopyranosyl rings, and is firmly embedded on the si face of the isoalloxazine ring (Fig. 3A). The lactam moiety of ABL is bound in the Ϫ1 subsite. The lactam C 1 atom, which corresponds to the site of oxidation in cellobiose, binds in front of the flavin N 5 with a distance of 3.35 Å and an angle with the N 5 -N 10 atoms of 105°. It should be noted that like for the CDH complex (21), the lactam C 1 and O 1 atoms are almost perfectly aligned with the flavin N 5 and C 4␣ , respectively.
The lactam O 1 atom interacts with Tyr 429 O (2.78 Å) and Gln 384 N ⑀2 (3.14 Å), and the endocyclic NH makes close contacts with the Tyr 72 O (3.34 Å) (Fig. 3B) On the basis of the bound ABL molecule described above, a variety of carbohydrate molecules were modeled manually in the carbohydrate-binding groove, and the model was subjected to energy minimization with CNS. D-Glucose is the only monosaccharide substrate for GOOX (3). Simulation of the complexes shows that other hexoses and derivatives form either fewer hydrogen bonds or unfavorable contacts with the residues surrounding the substrate-binding groove. The axial OH 3 group in allose and the absence of an OH 2 in 2-deoxy-D-glucose and of the exocyclic CH 2 OH in D-xylose would entail a small number of hydrogen bonds. The axial OH 2 in mannose, the axial OH 4 in galactose, and the equatorial NH 2 in glucosamine would cause unfavorable contacts with Gln 353 , Trp 351 , and Arg 245 , respectively. Modification at the C 1 , C 2 , and C 6 positions such as phosphorylation and N-acetylation would cause steric hindrance because of lack of appropriate space. Like CDH and galactose oxidase (21,22), GOOX possesses an open carbohydrate-binding groove, so that the non-reducing end of the glucose residue can stick out into the solvent and be exposed on the protein surface (Fig. 3C). This explains why the enzyme is able to utilize oligosaccharides as good substrates (4). In contradistinction, the crystal structures of GOX  and POX have revealed a "size-exclusion mechanism," in which the shape of the active site cavity is such that the enzyme can only accept monosaccharides (23,24).
Catalytic Mechanism-Most of the FAD-dependent oxidases and dehydrogenases are proposed to function via a hydride transfer rather than by a carbanion mechanism (25,26). The pH optimum of 10 for GOOX implies that a tyrosine residue may serve as a general base (3). Based on the ABL binding, the substrate cellobiose was modeled into the active site. The reducing end C 1 atom of the Ϫ1 sugar binds in front of the flavin N 5 at a distance of 2.86 Å and an angle with the N 5 -N 10 atoms of 113°. These values are in agreement with those typically observed for flavoenzymes, and appropriate for a direct hydride transfer (25,26). However, no appropriate general base for carbanion formation can be identified. The hemiacetal OH 1 group of the ␤-anomer at the Ϫ1 subsite could interact with Tyr 429 O (3.11 Å), Gln 384 N ⑀2 (3.69 Å), and the isoalloxazine O 4 (3.11 Å), whereas that of the ␣ anomer may make close contacts with Thr 129 O ␥1 (3.17 Å) and the isoalloxazine O 4 (3.25 Å). The distance between the reducing end H 1 of the ␤-anomer and the flavin N 5 is 2.13 Å, whereas that for C 1 H of the ␣ to the flavin N 5 is 2.64 Å. Therefore, GOOX may preferentially oxidize the ␤-anomer possessing an equatorial hydroxyl group, with the conserved Tyr 429 acting as a general base.
As is common for flavoenzymes, the reaction mechanism of GOOX consists of two half-reactions (Scheme 1). The reductive half-reaction is involved in the oxidation of the free reducing end ␤-D-glucosyl residue to glucono-1,5-lactone by hydride transfer to the N 5 atom, probably initiated by proton abstraction from the OH 1 group by Tyr 429 . The lactone product is spontaneously hydrolyzed to gluconic acid. In the oxidative half-reaction, regeneration of the oxidized FAD by molecular oxygen yields hydrogen peroxide. Water-mediated hydrogen bonds between Asp 355 and Tyr 429 suggest that Asp 355 may assist in proton transfer by lowering the pK a value of Tyr 429 through a water molecule. Interactions between Gln 384 and the reducing end OH 1 group suggest that Gln 384 may contribute to position the substrate and facilitate proton abstraction by Tyr 429 . His 138 , Tyr 426 , and the backbone NH of Tyr 144 may be involved in stabilization of the anionic form of the reduced flavin. A large water-filled channel above the carbohydrate-binding groove and toward the flavin moiety may act as an entry point for the second substrate, molecular oxygen (Fig. 3C).
The FAD binding modes are one of the least conserved properties among flavoproteins, and the flavinylation sites are often located in poorly conserved regions (18,37). Although these PCMH members share a similar F domain, they contain different FAD binding types. In addition to the novel flavinylation in GOOX (6-S-cysteinyl, 8␣-N1-histidyl FAD), the FAD cofactor utilizes its 8␣-methyl group cross-linking to His 105 in ZmCKX and His 121 in BsCOX2 (8␣-N1-histidyl FAD), to His 422 in PsVAO (8␣-N3-histidyl FAD), and to Tyr 384 in PpPCMH (8␣-O-tyrosyl FAD). The remaining members bind to FAD non-covalently.
The F domain is much more superimposable than the S domain, perhaps because of distinct substrate binding but a similar FAD recognition mechanism across the enzymes. The eight ␤-strands (␤2-␤9, 60 structurally equivalent residues) of the F domain, overlay within 0.8 -1.4 Å root mean square deviations (Fig. 4A). Interestingly, these members utilize many main-chain atoms for the FAD binding, and hence the sequences of the FAD-interacting residues are divergent, whereas their spatial positions are convergent in four regions (Figs. 1 and 4B). The first region is located in the loop between the ␤3 and ␤4 strands and shares similar features with the "P-loop," which is highly conserved in nearly all nucleotide-binding proteins and responsible for interaction with the pyrophosphate group of the cofactor NDP moiety (38). The second region involves the short ␣D helix, and the third region is at located the N terminus of the ␤9 strand. It is worth noting that the hydrogen bonds between the SCHEME 1. The proposed catalytic mechanism for GOOX.
pocket and thus result in a "size-exclusion mechanism" (28 -31). In contradistinction, GOOX has an open carbohydrate-binding groove to accommodate larger oligosaccharides.
Convergent and Divergent Evolution in FAD-utilizing Sugar Oxidases-According to the structural fold, flavoenzymes have been classified into many superfamilies (37). Interestingly, folding topology does not correlate with enzyme function. For example, GOOX and BsCOX2 belong to the PCMH superfamily, whereas GOX, POX, CDH, and BsCOX1 belong to the glutathione reductase superfamily. Location of the active center at the re face of the isoalloxazine ring is generally conserved within the glutathione reductase superfamily, whereas that of the PCMH members is on the si side. Therefore, these structures provide elegant examples of convergent evolution, where starting from different ancestral folds, the same FAD-assisted glucose (or cholesterol) oxidation is achieved through opposite flavin faces within distinct substrate-binding sites. In addition, members of the PCMH superfamily have a remarkable tendency toward covalent flavinylation at the 8␣-methyl group even for different types of flavinylation (18), whereas those of the glutathione reductase superfamily prefer a non-covalent binding mode. This may imply that the former FAD-binding fold facilitates flavinylation to optimize the active site for enhancement of its oxidative power.
Interestingly, even starting from a similar structural fold, sugar oxidases evolve some dissimilar residues for interaction with the common substrates. Within the glutathione reductase superfamily, GOX, POX, and CDH catalyze the oxidation of glucose, and share a similar F domain; however, their S domains lack conserved carbohydrate-interacting residues except for the conserved histidine serving as a general base (21,23,24). Similarly, within the PCMH superfamily, although GOOX resembles NlNEC5, CcHEOX, and HaCHOX in substrate specificity and F domain, the protein sequences of their S domains are quite diverse (Fig. 4B). The sugar-binding resides Tyr 72 , Arg 245 , Gln 384 , and Tyr 429 in GOOX, all at the Ϫ1 subsite, may also be conserved in NlNEC5, CcHEOX, and HaCHOX. However, the other residues including Thr 129 , Tyr 300 , Trp 351 , and Gln 353 , apparently are not conserved. Therefore, these sugar oxidases have evolved a similar FAD-assisted oxidation mechanism but different substrate recognition, resulting in distinct binding affinities to the mono-and disaccharides, with K m values ranging from 50 to 30 mM.