A β-l-Arabinopyranosidase from Streptomyces avermitilis Is a Novel Member of Glycoside Hydrolase Family 27*

Arabinogalactan proteins (AGPs) are a family of plant cell surface proteoglycans and are considered to be involved in plant growth and development. Because AGPs are very complex molecules, glycoside hydrolases capable of degrading AGPs are powerful tools for analyses of the AGPs. We previously reported such enzymes from Streptomyces avermitilis. Recently, a β-l-arabinopyranosidase was purified from the culture supernatant of the bacterium, and its corresponding gene was identified. The primary structure of the protein revealed that the catalytic module was highly similar to that of glycoside hydrolase family 27 (GH27) α-d-galactosidases. The recombinant protein was successfully expressed as a secreted 64-kDa protein using a Streptomyces expression system. The specific activity toward p-nitrophenyl-β-l-arabinopyranoside was 18 μmol of arabinose/min/mg, which was 67 times higher than that toward p- nitrophenyl-α-d-galactopyranoside. The enzyme could remove 0.1 and 45% l-arabinose from gum arabic or larch arabinogalactan, respectively. X-ray crystallographic analysis reveals that the protein had a GH27 catalytic domain, an antiparallel β-domain containing Greek key motifs, another antiparallel β-domain forming a jellyroll structure, and a carbohydrate-binding module family 13 domain. Comparison of the structure of this protein with that of α-d-galactosidase showed a single amino acid substitution (aspartic acid to glutamic acid) in the catalytic pocket of β-l-arabinopyranosidase, and a space for the hydroxymethyl group on the C-5 carbon of d-galactose bound to α-galactosidase was changed in β-l-arabinopyranosidase. Mutagenesis study revealed that the residue is critical for modulating the enzyme activity. This is the first report in which β-l-arabinopyranosidase is classified as a new member of the GH27 family.

Arabinogalactan proteins (AGPs) 3 are a family of complex proteoglycans widely distributed in plants (1,2). AGPs are also found in tree exudate gums and coniferous woods (3) and are characterized by the presence of large amounts of carbohydrate components rich in galactose (all the sugars in the present study are in the D-configuration unless otherwise specified) and L-arabinose and by protein components rich in hydroxyproline, serine, threonine, alanine, and glycine (4). Type II arabinogalactans and short oligosaccharides are the two types of carbohydrates attached to the AGP backbone. Type II arabinogalactans have ␤-1,3-linked galactosyl backbones in mono-or oligo-␤-1,6-galactosyl and/or L-arabinosyl side chains (2,5). L-Arabinose and lesser amounts of other auxiliary sugars such as glucuronic acid, L-rhamnose, and L-fucose are attached to the side chains primarily at nonreducing termini (2). Molecular and biochemical evidence indicates that AGPs have specific functions during root formation, promotion of somatic embryogenesis, and attraction of pollen tubes to the style (6). However, because many putative protein cores exist and the structures of the carbohydrate moieties are complex, it has been difficult to differentiate one AGP species from another in plant tissues. This, in turn, has made it difficult to assign specific roles to individual AGPs. Despite significant physiological interest in AGPs, there are few studies on glycoside hydrolases that cleave the sugar moieties of these proteins. It is important to study such enzymes because hydrolytic enzymes specific to particular sugar residues or to a type of glycosidic linkage would be useful tools in the structural analysis of AGPs.
Because S. avermitilis has two different kinds of galactanases, we focused on finding novel AGP-degrading enzymes. We have cultivated the actinomycete using gum arabic as a carbon source, and isolated a novel ␤-L-arabinopyranosidase.
To the best of our knowledge, the only report on ␤-L-arabinosidase (EC 3.2.1.88) has been on its purification from Cajanus indicus (12). The amino acid composition of the enzyme was investigated (13), but its sequence remains unknown. In this article, we cloned ␤-L-arabinopyranosidase from S. avermitilis (SaArap27A), analyzed its catalytic properties, and analyzed the crystal structure of the recombinant enzyme. The results clearly showed that this enzyme is ␤-Larabinopyranosidase and is a novel member of the glycoside hydrolase family 27 (GH27). This is the first detailed report on ␤-L-arabinopyranosidase.
The sugar composition of the polysaccharides was analyzed as described previously (14,15). Briefly, the substrate was hydrolyzed with 2 M trifluoroacetic acid by incubating at 121°C for 1 h, and the acid was evaporated under a stream of N 2 gas. The sugar composition of the sample was analyzed by high performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) using a CarboPac TM PA20 column (Dionex Corp., Sunnyvale, CA) at a flow rate of 0.5 ml/min, as reported earlier (15).
Partial Purification of Native SaArap27A-S. avermitilis NBRC14893 was obtained from the National Institute of Technology and Evaluation (Kazusa, Japan). The strain was grown on liquid medium containing 1% gum arabic, 0.1% yeast extract, 0.1% peptone, 0.5% potassium dihydrogen phosphate, and 0.05% magnesium sulfate (11) in a baffle flask at 28°C for 6 days. The culture was filtered, and the supernatant was used as the crude enzyme.
The crude enzyme was concentrated 10-fold using a 100-kDa polyethersulfone ultrafiltration membrane (Biomax, Millipore Corp.). Ammonium sulfate was added with stirring to the concentrated fraction to 70% saturation. The precipitated protein was redissolved in 50 mM acetate buffer (pH 4.0) containing 2 mM CaCl 2 and dialyzed against the same buffer. After centrifugation, the supernatant was applied to a SP-Sepharose Fast Flow column (GE Healthcare UK Ltd.; HR16/10 column, 0.8 ϫ 10 cm) equilibrated with 50 mM acetate buffer (pH 4.0) containing 2 mM CaCl 2 . ␤-L-Arabinopyranosidase activity was eluted with a linear gradient of 0 -1 M sodium chloride. The active fractions were combined, dialyzed against 50 mM acetate buffer (pH 4.0) containing 2 mM CaCl 2 , and loaded onto a Mono S HR 5/5 cation-exchange column (GE Healthcare) equilibrated with the same buffer. ␤-L-Arabinopyranosidase activity was eluted by a linear gradient of 0 -0.5 M sodium chloride. The pooled active fractions were assayed for total protein and ␤-L-arabinopyranosidase activity, and protein purity was determined by SDS-PAGE, according to the method of Laemmli (16).
Molecular Cloning, Mutagenesis, and Expression of Recombinant SaArap27A-The pooled fractions from Mono S cationexchange chromatography were separated by SDS-PAGE and electroblotted onto a polyvinylidene difluoride membrane (Millipore Corp.). The N-terminal amino acid sequence was analyzed on an HP G105A protein sequencer (Hewlett Packard).
The gene encoding a putative ␤-L-arabinopyranosidase (SAV_2186; GenBank TM accession number BAC69897) was cloned and expressed as a mature protein using the Streptomyces expression system as described previously (17). Briefly, the full-length gene amplified from S. avermitilis genomic DNA by PCR was cloned into a Streptomyces expression vector (18), and the plasmid was transformed into Streptomyces lividans 1326, which served as the host. The recombinant protein SaArap27A was secreted at high levels in the culture broth and was purified on lactosyl-Sepharose as described previously (19). The protein was detected by SDS-PAGE and dialyzed against 50 mM sodium phosphate buffer (pH 7.0). The final preparation thus obtained was used as the purified enzyme. The E99D mutant of SaArap27A was generated by PCR using appropriate primers (supplemental Table S1). Mutation was confirmed by DNA sequencing. Expression and purification of the mutant were carried out in the same way as for the wild-type enzyme.
Enzymatic Properties-␤-L-Arabinopyranosidase activity was determined using a mixture containing 25 l of 2 mM PNP-␤-L-Arap, 20 l of McIlvaine buffer (0.2 M Na 2 HPO 4 and 0.1 M citric acid, pH 4.0), and 5 l of enzyme solution. The reactions were carried out at 40°C for 10 min and terminated by the addition of 50 l of 0.2 M Na 2 CO 3 . The amount of PNP released was detected at 400 nm (extinction coefficient ϭ 19,200 M Ϫ1 cm Ϫ1 ). One unit of enzyme activity is defined as the amount of enzyme that released 1 mol of PNP per minute. The protein amounts were determined by the BCA protein assay reagent kit (Pierce) using bovine serum albumin as the standard.
The effects of pH and temperature on enzyme activity were investigated as described previously (7,20). The following buffers were used to study the effect of pH on ␤-L-arabinopyranosidase activity: McIlvaine buffer (pH 2.6 -7.6), Atkins-Pantin buffer (pH 7.6 -11.0), and glycine-HCl buffer (pH 1-2.6). The activity of ␤-L-arabinopyranosidase was assayed under the conditions described above. The effects of metal ions and chemicals on enzyme activity were also examined by treatment with 2 mM solutions of metal ions, including Ca 2ϩ , Cd 2ϩ , Co 2ϩ , Cu 2ϩ , Fe 2ϩ , Mg 2ϩ , Mn 2ϩ , Zn 2ϩ , Hg 2ϩ , and Ag 2ϩ , and with 2 mM solutions of chemicals such as EDTA, p-chloromercuribenzoic acid, and SDS.
Substrate Specificity-The substrate specificity of SaArap27A toward various PNP glycosides was determined. The assay method was identical to that described for PNP-␤-L-Arap. The kinetic parameters of wild-type and mutants of SaArap27A ␤-L-Arabinopyranosidase from S. avermitilis were determined as follows. The reactions were performed in McIlvaine buffer (pH 4.0) containing 0.5-5 mM substrates, 0.1% (w/v) bovine serum albumin, and 2.6 -260 nM enzyme at 37°C for up to 15 min. The amount of PNP released was determined from the absorbance at 400 nm. The assay was performed in triplicate.
The substrate specificity of SaArap27A toward polysaccharides was also determined at 40°C in McIlvaine buffer (pH 4.0) with 0.5% (w/v) polysaccharide as the substrate and 2.6 M enzyme. After incubation for the appropriate reaction time, the initial hydrolysis rate was determined by the Somogyi-Nelson method (21). After incubation for 24 h, the amount of released arabinose was quantified by the HPAEC-PAD system using a CarboPac PA1 column (Dionex Corp.) at a flow rate of 1 ml/min, as described previously (20). The substrate hydrolysis rate was estimated by assuming that 100% L-arabinose was released from the substrate.
Crystallization, Data Collection, and Structure Determination-Crystallization procedures and native data collection have been reported elsewhere (17,22). The data collection statistics are summarized in Table 3. The Matthews coefficient was calculated to be 2.38 Å 3 Da Ϫ1 (23), and this indicated that the asymmetric unit of the crystal contained two SaArap27A molecules corresponding to a solvent content of 48.3% in the crystal.
Crystals complexed with L-arabinose or galactose were prepared by adding the same volume of the reservoir solution containing 10% sugar into the crystal drop one night prior to x-ray exposure. The data for the sugar complexes were collected at beamline BL-6A at the Photon Factory, High Energy Accelerator Research Organization, Tsukuba, Japan. The crystals were mounted in a 0.3-mm quartz glass capillary and flash cooled under a nitrogen stream at 95 K. Diffraction data were collected with 20-s exposures for 1°oscillations for a total of 360°at a wavelength of 0.978 Å with a Quantum R4 CCD detector (ADSC, CA). Data were integrated and scaled using the DENZO and Scalepack programs in the HKL2000 program suite (24) up to a resolution of 1.9 Å.
The crystal structure was first analyzed by the molecular replacement method using the 1.5-Å resolution ligand-free data. Homology models of the first two domains and the C-terminal carbohydrate-binding module family 13 (CBM13) of SaArap27A were built with the MODELER program (Accelrys Software Inc., San Diego, CA) using the known crystal structures of rice ␣-galactosidase and CBM13 of Streptomyces olivaceoviridis ␤-xylanase (SoCBM13), respectively, as reference models (25,26). The first run of the MOLREP program (27) in the CCP4 program suite (28) using the ␣-galactosidase-derived model as the reference against the native data resulted in two prominent solutions yielding an R factor of 0.553. This was followed by restrained refinement of the resultant models with the REFMAC5 program (29). With the refined models fixed, the second run of MOLREP using the CBM13-derived model as the reference yielded two solutions resulting in an R factor of 0.501. The second run of restrained refinement yielded an R factor of 0.433 and figure of merit of 0.521, and the resultant electron density map was sufficient for tracing the overall structure of two ␤-L-arabinopyranosidase molecules. The model was initially built using the automodeling program ARP/wARP (30), and several cycles of manual rebuilding and refinement followed using COOT (31) and REFMAC5 programs. In addition to the protein, models of water, glycerol, polyethylene glycol, sulfate ion, and HEPES were included. Structural analyses of the L-arabinose and galactose complexes were performed by isomorphous replacement, and the bound sugar molecules were identified by the F obs Ϫ F calc electron density map. The structure refinement statistics are shown in Table 3. The stereochemistry of the models was analyzed by the PROCHECK program (32). Figures were prepared using MOLSCRIPT (33) and RASTER3D (34).

RESULTS
Purification of Native SaArap27A and Analysis of the Primary Structure-S. avermitilis was cultivated in liquid medium using gum arabic as the sole carbon source. The enzyme activities of the culture supernatant were tested using various PNP glycosides (see supplemental Fig. S1A). Interestingly, the culture supernatant showed the highest activity toward PNP-␤-L-Arap among all tested substrates. To analyze the enzymatic activity, we partially purified ␤-L-arabinopyranosidase from the culture supernatant of S. avermitilis (Table 1). After cationexchange chromatography, the enzyme appeared as a major band of a 64-kDa protein on SDS-PAGE (supplemental Fig. S2, lane 2, indicated by the arrow). The N-terminal amino acid sequence of the protein was determined to be AVTTR-QITVPSA. A BLASTP search against the S. avermitilis NBRC14893 sequence data base revealed that the determined amino acid sequence corresponded to the open reading frame SAV_2186. The open reading frame SAV_2186 of S. avermitilis (1977 bp) encoded a putative ␤-L-arabinopyranosidase (658 amino acids) (Fig. 1, SaArap27A). The N-terminal 12-amino acid sequence of the mature enzyme corresponds to amino acid residues 45-56 of the deduced amino acid sequence (Fig. 1,  underlined). The N terminus (amino acids 1-44) was consistent with the signal sequence predicted by SignalP.
Similar investigations of the deduced amino acids indicated that this enzyme has a modular architecture. Residues 45-430 of the protein showed high sequence similarity to GH27 enzymes, whereas residues 532-658 showed similarity to CBM13 (Fig. 1). Among the proteins whose three-dimensional Based on the crystal structures, almost all of the amino acids involved in sugar binding were conserved, but only one aspartic acid, which coordinates to the O 6 atom of the bound sugar, was replaced by glutamic acid (Glu 99 ) in SaArap27A.
Expression and Characterization of SaArap27A-The DNA fragment encoding the full-length protein was cloned. Recombinant protein SaArap27A was successfully expressed in S. lividans in the secreted form and purified as a single band with an apparent molecular size of 64 kDa on SDS-PAGE (supplemental Fig. S2, lane 3); this was in agreement with the size of the native enzyme (supplemental Fig. S2, lane 2, indicated by the arrow).
Using PNP-␤-L-Arap as the substrate, maximal enzyme activity was detected at pH 4.0 and 40°C. The enzyme was stable between pH 4.0 and 8.0 at 30°C for 1 h and was also stable below 45°C for 1 h (supplemental Fig. S3). Some metal ions and chemicals did not affect the enzyme activity, but Hg 2ϩ , Ag 2ϩ , p-chloromercuribenzoic acid, and SDS completely inhibited the enzyme activity (data not shown). The specific activity of SaArap27A with PNP-␤-L-Arap as the substrate was 18 units/ mg, which is almost the same as that of the native enzyme (21 units/mg, see Table 1).
The activities of SaArap27A toward various PNP glycosides were tested (supplemental Table S2). The enzyme was active on PNP-␤-L-Arap. It also showed weak activity toward PNP-␣-Galp and 1.5% activity toward PNP-␤-L-Arap but did not hydrolyze any other PNP glycosides, including PNP-␣-Nacetylgalactosaminide. Because the structure of ␤-L-arabinose and ␣-galactose are similar, kinetic study was performed using PNP-␤-L-Arap and PNP-␣-D-Galp. ␤-L-Arabinopyranose and ␣-galactopyranose have similar orientations of hydrogen and hydroxyl groups at C-2, C-3, and C-4. The only difference is the presence of the hydroxymethyl group at C-5 in ␣-galactopyranose (supplemental Fig. S1B, see also Fig. 3). The K m and k cat values of SaArap27A for PNP-␤-L-Arap were 3.6 Ϯ 0.4 mM and 317 Ϯ 10 min Ϫ1 , respectively (see Table 4). In contrast, a significant difference in the k cat value was observed when PNP-␣-D-Galp was used as the substrate. The K m and k cat values of SaArap27A for PNP-␣-D-Galp were 5.1 Ϯ 0.3 mM and 2.3 Ϯ 0.1 min Ϫ1 , respectively (see Table 4).
The substrate specificities of the enzyme toward polysaccharides were investigated. The enzyme hydrolyzed gum arabic and larch arabinogalactan but not guar gum, locust bean gum, arabinan, debranched arabinan, wheat arabinoxylan, or corn hull arabinoxylan. SaArap27A removed 0.1% L-arabinose from gum arabic or 45% L-arabinose from larch arabinogalactan ( Table 2). The reaction products of gum arabic by SaArap27A were analyzed by HPAEC-PAD, and the only L-arabinose was detected (supplemental Fig. S4). These data suggest that SaArap27A is a ␤-L-arabinopyranosidase.
Overall Structure of SaArap27A-The crystal structure of SaArap27A in the ligand-free state and in complexes with Larabinose and galactose were determined by the molecular replacement method at 1.5-, 1.9-, and 1.9-Å resolution, respectively, and the structures were refined to R/R free factors of 0.157/0.179, 0.162/0.207, and 0.142/0.177, respectively ( Table  3). The final models included two noncrystallographic symmetry-related SaArap27A molecules (molecules A and B) as well as the surrounding water, glycerol, polyethylene glycol, HEPES molecules, and sulfate ions. The SaArap27A molecule was composed of a single polypeptide chain of 614 amino acids (45-658), and molecule A was modeled through the chain, but the two N-terminal residues Ala 45 and Val 46 of molecule B could not be identified due to the lack of electron density. The root mean square difference of the C␣ atoms of these two molecules was calculated to be 0.31 Å, and two noncrystallographic symmetry molecules had almost the same overall structures.
The SaArap27A monomer consists of four domains ( Fig. 2A). The N-terminal catalytic domain (domain I, residues 45-339) has a (␤/␣) 8 barrel, which is observed in many glycoside hydrolases. The second domain (domain II, residues 340 -430) is an eight-stranded anti-parallel ␤-domain containing tandemly repeated Greek key motives, but in imperfect shapes. The relative arrangement of these two domains is the same as that of other GH27 enzymes. Domain II is located at the 7th and 8th ␣-helices of the catalytic domain. The third domain (domain III, residues 431-531) also contains eight antiparallel ␤-strands but comprises a ␤-jellyroll domain. This domain is located adjacent to domain II and also contacts the catalytic domain over the 5th ␣-helix and the loop after the 6th ␣-helix. The last domain, i.e. the C-terminal domain (domain IV, residues 532-658), is a ricin-type lectin domain consisting of the ␤-trefoilfold and is of the CBM13 type. This domain is in front of the catalytic domain covering the 6th ␣-helix and the loop before the 5th ␣-helix so that it contacts all three other domains forming a compact entity. No remarkable linker peptide could be observed between domains. The SaArap27A molecule contains four disulfide bonds: Cys 148 -Cys 188 in the catalytic domain and Cys 543 -Cys 562 , Cys 585 -Cys 604 , and Cys 628 -Cys 647 in CBM13.
Sugar complex crystals were prepared by soaking the SaArap27A crystals with L-arabinose or galactose solution, and the structures were determined. The relative positions of the domains did not change in comparison to the ligand-free state. The bound sugars were identified in the F obs Ϫ F calc electron density maps (Fig. 3). In the L-arabinose complex, one L-arabinose molecule was bound in each catalytic domain and four (in molecule A) or three (in molecule B) L-arabinose molecules were found in the CBM domain ( Fig. 2A). On the other hand, in  The L-arabinose content in the residue after the enzymatic reaction was determined, and the hydrolysis rate was estimated by assuming that 100% arabinose was released from the substrate.
␤-L-Arabinopyranosidase from S. avermitilis the galactose complex, one galactose molecule was bound in each catalytic domain but only one galactose molecule was found in the CBM domain of molecule B. No bound sugars were observed in domains II and III. Substrate Binding Structure in the Catalytic Module-The structure of the L-arabinose complex revealed that one L-arabinose molecule was bound in the active site of the catalytic domain, and L-arabinose was mainly in the ␣-anomeric form (Fig. 4A). Two aspartic acid residues, Asp 186 and Asp 247 , are regarded as the catalytic residues of this enzyme. Asp 186 forms a hydrogen bond from its O ␦2 atom to the L-arabinose (Ara)-O 1 atom, whereas Asp 247 forms a hydrogen bond from its O ␦1 atom to the Ara-O 2 atom. But if the bound L-arabinose adopts the ␤-anomeric form as in the case of the natural substrate, the Asp 247 -O ␦2 atom is located at hydrogen bonding distance to the Ara-O 1 atom, and these residues maintain the stereochemical arrangement that is often observed in the retaining glycoside hydrolases (Fig. 4A). Apart from the catalytic residues, many other amino acids participate in sugar binding. The is bound in a manner quite similar to that in which galactose is bound by rice ␣-galactosidase (Fig. 4C). These residues are completely conserved in their positions and functions.
At the entrance of the catalytic pocket, one glycerol molecule was observed (Fig. 4A). It was surrounded by Tyr 152 , Trp 226 , and Tyr 250 . The position seems to be the aglycon subsite of the enzyme. However, Tyr 152 and Typ 250 are not conserved in the other GH27 enzymes because they are located in the inserted peptides unique to SaArap27A.
The structure of the galactose complex revealed that one galactose molecule was bound in the active site, and galactose was observed to mainly adopt the ␤-anomeric form (Fig. 4B). The B factors of the bound galactose atoms are comparable with those of the surrounding amino acids, and sugar occupancy is close to 100%. The position of the bound galactose is almost the same as that of bound L-arabinose in the L-arabinose complex (Fig. 4A), and the sugar-binding mechanism is also well conserved. However, a difference was observed in the case of the galactose (Gal)-O 6 atom, which is not present in the L-arabinose molecule. The Gal-O 6 atom has a unique hydrogen bond to the Glu 99 -O ⑀1 atom. In comparison to the structure of rice ␣-galactosidase, there were no obvious differences in the galactose-binding manner, but Glu 99 , the hydrogen-bonding partner of the Gal-O 6 atom, was replaced with Asp 52 (Fig. 4C).
Role of Glu 99 in SaArap27A to Modulate ␤-L-Arabinopyranosidase/␣-D-Galactopyranosidase Activity-The K m value of SaArap27A for PNP-␣-Galp was almost the same as for PNP-␤-L-Arap (Table 4). However, the k cat values of the enzyme for PNP-␣-Galp was ϳ140 times lower than for PNP-␤-L-Arap. To investigate the role of Glu 99 for enzyme activity, a mutant enzyme (SaArap27A/E99D) was constructed in which Glu 99 was replaced by Asp. When the enzyme activities for PNP glycosides were tested (supplemental Table S2), the mutant gained higher activity for PNP-␣-Galp than for PNP-␤-L-Arap. The specific activity of SaArap27A/E99D for PNP-␣-Galp was 9 units/mg at pH 4.0 at 40°C, and was 9 times higher than that for PNP-␤-L-Arap. The K m and k cat values of the mutant for PNP-␣-Galp and PNP-␤-L-Arap were 4.3 Ϯ 0.1 mM and 29 Ϯ 0.6

TABLE 3 Data collection and structure refinement statistics of SaArap27A
The values in parentheses represent the highest resolution shell.

Data
Ligand-free L-Arabinose complex D-Galactose complex Data collection P2 1  ␤-L-Arabinopyranosidase from S. avermitilis min Ϫ1 and 11.1 Ϯ 0.9 mM and 17 Ϯ 1 min Ϫ1 , respectively ( Table  4). The enzyme activity of the mutant converted ␤-L-arabinopyranosidase to ␣-galactosidase, suggesting that Glu 99 is critical for modulating the enzyme activity of SaArap27A. Sugar-binding Structure of CBM13-The structure of the L-arabinose complex revealed that three L-arabinose molecules were bound in three subdomains of SaArap27A domain IV (Fig. 4, D-F). In subdomain ␣, the L-arabinose was docked with its hydrophobic face (C 3 -C 4 -C 5 ) into the aromatic indole ring of Trp 560 to form a stacking interaction (Fig. 4D). There are six hydrogen bonds between subdomain ␣ and the bound L-arabinose molecule. These are from Asn 548 -O ␦1 to the Ara-O1 atom; from Asn 563 -N ␦1 , Asn 567 -N ␦1 , and Asp 545 -O ␦2 to the Ara-O 3 atom; and from Asn 548 -N and Asp 545 -O ␦1 to the Ara-O 4 atom. Another hydrogen bond was observed from the side chain of Thr 481 of domain III to the Ara-O 1 atom. On the side of the binding pocket, the aromatic plane of the side chain of Tyr 547 faces the C 5 -O 5 bond of the bound L-arabinose. Behind Tyr 547 of molecule A, another L-arabinose molecule was observed that forms a stacking interaction between the aromatic ring of Tyr 547 and the hydrophobic surface of L-arabinose. Two carboxylate oxygen atoms of Glu 558 form two hydrogen bonds with the O 3 and O 4 atoms of the L-arabinose. However, no bound L-arabinose was observed at this site in molecule B. The average B factor of this L-arabinose molecule was high in comparison to that of the bound L-arabinose in the binding pocket, and this binding did not seem to be specific. In subdomains ␤ and ␥, the bound L-arabinose was docked in a manner similar to that in subdomain ␣ (Fig.  4, E and F). The detailed binding manners of each subdomain are described under supplemental materials.
When the crystal was soaked with galactose, electron density in the sugar-binding sites of CBM13 showed that the bound molecules were mostly glycerol, which were derived from the crystallization condition. Additionally, the structure of the complex revealed that the CBM13 of SaArap27A bound to Larabinose in three subdomains when the crystals were soaked in a mixture of L-arabinose and galactose.

DISCUSSION
␤-L-Arabinopyranose was found in AGP as 3-O-␤-L-arabinopyranosyl-L-arabinose determined by structural analysis of carbohydrate moieties (3,35), implying the existence of a ␤-Larabinopyranosidase. However, to the best of our knowledge, the only report on ␤-L-arabinopyranosidase has been on its purification from C. indicus, and its sequence information has not yet been elucidated (12,13).

␤-L-Arabinopyranosidase from S. avermitilis
In this study, for the first time, we cloned the ␤-L-arabinopyranosidase gene from S. avermitilis and determined the threedimensional structure of the enzyme. Sequence analysis showed that SaArap27A has a unique modular structure composed of four structural modules. The resultant three-dimensional structure showed that this enzyme has a compactly packed modular architecture in which all four domains are in contact with each other, and the catalytic pocket and three sugar-binding sites are open to the solvent. Interestingly, they face the same solvent region, i.e. in Fig. 2B, the catalytic site faces the left, and three L-arabinose-binding sites of CBM13 face upwards. We noticed that there are many tyrosine residues exposed to this solvent region. Tyr 150 , Tyr 151 , Tyr 152 , Tyr 228 , Tyr 250 , and Tyr 251 are in the catalytic domain, and five of these are in the inserted peptides around the catalytic cleft (supplemental Fig. S5B). Tyr 547 , Tyr 589 , and Tyr 632 are exposed from the CBM13 domain, and these residues were also used in the arabinose binding pocket mentioned under "Results" (Fig. 4, D-F The substrate type II arabinogalactan contains both L-arabinopyanosyl and L-arabinofuranosyl residues in addition to galactose. CBM13 is often represented by the plant lectin ricin; therefore, its main function is considered to be galactose binding. However, the structure of the sugar complex revealed that CBM13 of SaArap27A preferred to bind to L-arabinose when the crystals were soaked in a mixture of L-arabinose and galactose. The sugar binding pocket of the ␣, ␤, and ␥ subdomains of SaArap27A CBM13 bound L-arabinose in the same manner (Fig. 4, D-F). The affinity of CBM13 for L-arabinose was reported for SlCBM13 (36). However, the association constants of L-arabinose and galactose for SlCBM13 were almost the same, i.e. 1.5 ϫ 10 2 and 6.1 ϫ 10 2 M Ϫ1 , respectively (36). Although the structures of many CBM13 complexes are cur-  ␤-L-Arabinopyranosidase from S. avermitilis rently available (on the CAZy website), these are with ligands such as galactose, GalNAc, xylose, glucose, and lactose, and there are no structures of complexes with L-arabinose. CBM13 of SaArap27A is the first one of its kind that preferentially binds to L-arabinose. Amino acid sequence comparison revealed that SaArap27A had a high similarity to ␣-galactosidases, particularly rice ␣-galactosidase (25). Comparison of the primary structure of SaArap27A and rice ␣-galactosidase indicated that there was a single amino acid substitution in the substrate-binding residues (Fig. 1, domain I). The aspartic acid to glutamic acid substitution occurred at Glu 99 of SaArap27A corresponding to Asp 52 of rice ␣-galactosidase (Figs. 1 and 4C). The results of the soaking experiment suggested that there were no large differences in the sugar binding manner; however, the Gal-O 6 atom has a tight hydrogen bond to the Glu 99 -O ⑀1 atom (Fig. 4B). It is also apparent from the kinetic parameters toward PNP-␤-L-Arap and PNP-␣-Galp. The K m value of SaArap27A for PNP-␣-Galp was almost the same as that of PNP-␤-L-Arap although the k cat values of the enzyme were different ( Table 4). The k cat value of the enzyme for PNP-␣-Galp was extremely lower than that for PNP-␤-L-Arap, indicating that the galactose tightly bound to the catalytic pocket resulted in the reduction of turnover of the catalysis. Therefore, this substitution would be one factor why SaArap27A selects ␤-L-arabinopyranose, not ␣-galactopyranose, as its substrate. The results of the mutagenesis study clearly demonstrated the critical role of Glu 99 for modulating the enzyme activity to possess ␤-L-arabinopyranosidase activity. The E99D mutant showed higher activity for PNP-␣-Galp than for PNP-␤-L-Arap (see Table 4 and supplemental Table S2).
The single aspartic acid to glutamic acid substitution is found in only SaArap27A, and Asp residues are conserved in all other GH27s whose three-dimensional structures are known, including ␣-galactosidases and ␣-N-acetylgalactosaminidases (Fig. 1). BLAST searches of the NCBI nonredundant protein sequence data base using the deduced amino acid sequence of SaArap27A revealed that the same substitution was also present in a putative ␣-galactosidase from S. sviceus ATCC 29083 (accession number EDY60900). The amino acid sequence shows 82% identity and 89% similarity. Additionally, the putative protein has a C-terminal module classified as a CBM13 domain. The putative protein also appears to be a ␤-L-arabinopyranosidase. The results of BLAST searches showed a different kind of amino acid substitution (aspartic acid-aspartic acid (Asp 52 in rice ␣-galactosidase)-cysteine-tryptophan to aspartic acid-cysteine-glycine-tryptophan) in a putative protein annotated as an ␣-galactosidase. The putative ␣-galactosidase from Aspergillus nidulans (accession number ABF50881) was expressed in Pichia pastoris X-33 as a plant cell wall polysaccharide-degrading enzyme (37). The crude enzyme showed weak activity toward PNP-␣-Galp (see Table 1 in Ref. 37). However, the enzyme did not show any activity toward raffinose, locust bean gum, and guar gum. A possible interpretation of these properties of the enzyme is that the real activity differs from that of ␣-galactosidase because all ␣-galactosidases with confirmed activity have aspartic acid, not glutamic acid, similar to SaArap27A, at the Gal-O 6 position in the substrate binding pocket. Aspartic acid coordinating the Gal-O 6 atom seems important for ␣-galactosidases or ␣-N-acetylgalactosaminidase activity and substitution of this aspartic acid would result in different substrate specificity. Similarly, we reported that some amino acid substitutions around the Gal-O 2 atom differentiate the enzyme activity between ␣-galactosidase or ␣-N-acetylgalactosaminidases. These sugar complex structures of GH27s showed the possibility of alternation of substrate specificity by some amino acid mutations and gave a structural base for further molecular design of GH27 enzymes.
In conclusion, we purified an enzyme with ␤-L-arabinopyranosidase activity from S. avermitilis. The gene encoding the enzyme was cloned, and the amino acid sequence was determined. In addition, we successfully determined the intact three-dimensional structure. The protein is classified as GH27 and consists of four modules. The enzymatic activity differs from that of other ␣-galactosidases due to a single amino acid substitution. The structure of the L-arabinose complex clearly indicates that the enzyme is ␤-L-arabinopyranosidase. Furthermore, CBM13 of this enzyme has a novel L-arabinose binding property. This is the first report of ␤-L-arabinopyranosidase as a new member of the GH27 family and CBM13 as an L-arabinose-binding module.