Crystal Structure of an Exo-1,5-α-l-arabinofuranosidase from Streptomyces avermitilis Provides Insights into the Mechanism of Substrate Discrimination between Exo- and Endo-type Enzymes in Glycoside Hydrolase Family 43*

Exo-1,5-α-l-arabinofuranosidases belonging to glycoside hydrolase family 43 have strict substrate specificity. These enzymes hydrolyze only the α-1,5-linkages of linear arabinan and arabino-oligosaccharides in an exo-acting manner. The enzyme from Streptomyces avermitilis contains a core catalytic domain belonging to glycoside hydrolase family 43 and a C-terminal arabinan binding module belonging to carbohydrate binding module family 42. We determined the crystal structure of intact exo-1,5-α-l-arabinofuranosidase. The catalytic module is composed of a 5-bladed β-propeller topologically identical to the other family 43 enzymes. The arabinan binding module had three similar subdomains assembled against one another around a pseudo-3-fold axis, forming a β-trefoil-fold. A sugar complex structure with α-1,5-l-arabinofuranotriose revealed three subsites in the catalytic domain, and a sugar complex structure with α-l-arabinofuranosyl azide revealed three arabinose-binding sites in the carbohydrate binding module. A mutagenesis study revealed that substrate specificity was regulated by residues Asn-159, Tyr-192, and Leu-289 located at the aglycon side of the substrate-binding pocket. The exo-acting manner of the enzyme was attributed to the strict pocket structure of subsite −1, formed by the flexible loop region Tyr-281–Arg-294 and the side chain of Tyr-40, which occupied the positions corresponding to the catalytic glycon cleft of GH43 endo-acting enzymes.

L-Arabinose residues are widely distributed in plant cell walls, where they are present in polymers such as arabinans, arabinoxylans, arabinogalactans, and arabinogalactan proteins (1). Research on plant cell walls is becoming a necessity because worldwide attention has now focused on bioethanol production to combat global warming and to improve global energy security. Because of competition between food and fuel, lignocellulose is expected to be used as a material for fuel ethanol production in the future. Generally, lignocellulose contains cellulose, which makes up ϳ40% of the total amount of cell wall components, together with ϳ20% hemicellulose, which is mainly composed of pentoses such as xylose and arabinose (2). Hemicelluloses often become bad factors in bioethanol production because the efficiency of ethanol conversion from pentoses is significantly lower than that from hexoses (3,4).
In contrast, L-arabinose is used as a functional sugar in the food industry. This sugar has a sweet taste and selectively inhibits intestinal sucrase activity in a noncompetitive manner and consequently suppresses plasma glucose increase due to sucrose ingestion (5)(6)(7). Therefore, L-arabinose may also be useful in preventing excess sucrose utilization.
In this article we analyzed the crystal structure of SaAraf43A. A mutagenesis study of SaAraf43A provides insights into the mechanism of substrate discrimination between exo-and endo-type enzymes in glycoside hydrolase family 43.
␣-L-Arabinofuranosyl azide (Araf␣Az) was synthesized as follows. In 50 ml of N,N-dimethylformamide, 5 g of 2,3,6tribenzoyl-␣-L-arabinofuranosyl bromide (23) (9.5 mmol) and 5 g of sodium azide (77 mmol) were dissolved, and the reaction was allowed to proceed at 95°C for 3.5 h. Then the solution was rotoevaporated to remove N,N-dimethylformamide. The residue was dissolved in 100 ml of CH 2 Cl 2 , filtrated, washed with aqueous NaHCO 3 , dried over Na 2 SO 4 , and evaporated into syrup. The syrup was then applied to a silica gel dry chromatography (3-cm diameter ϫ 20 cm) with hexane-ethyl acetate (3:2 by volume) as solvent. Fractions containing the main product were collected, and 2,3,6-tribenzoyl-␣-L-arabinofuranosyl azide was crystallized from methanol. Finally, 2.2 g of 2,3,6tribenzoyl-␣-L-arabinofuranosyl azide ( 4.71 (1H, dd, J ϭ 5.0, 11.0 Hz H5). Then, 300 mg of 2,3,6-tribenzoyl-␣-L-arabinofuranosyl azide (0.62 mmol) was dissolved in 3 ml of MeOH containing a catalytic amount of sodium methoxide and left to stand for 2 h at 25°C. The solution was then deionized with Dowex 50ϫ, filtered, and evaporated into syrup. The syrup was dissolved in 3 ml of H 2 O, and the remaining methyl benzoate was extracted with CHCl 3 3 times. The aqueous layer was collected and lyophilized to obtain 108 mg of Araf␣Az (0.62 mmol, yield 100%) as white powder. Protein Expression and Mutant Generation-Recombinant SaAraf43A was expressed in Escherichia coli BL21gold (DE3) (Stratagene, La Jolla, CA) using an expression vector pET30 (Novagen, Madison, WI) purified as described previously (18). Amino acid substitutions of SaAraf43A were generated by inverse PCR using pET30/SaAraf43A as template DNA and the appropriate primers (supplemental Table S1). Mutations were confirmed by DNA sequencing. The plasmids were transformed into E. coli BL21gold (DE3). Expression and purification of wild type and mutants were carried out in the same way as for wild type SaAraf43A (18).
Crystallization, Data Collection, and Structure Determination-Crystallization procedures have been reported previously (21) Approximately equal volumes of the reservoir solution containing 10% glycerol was added into the crystal drop, scooped in a nylon loop, and flashfrozen in a nitrogen stream at 95 K. Diffraction data were collected with ADSC CCD detectors. Data were integrated and scaled using the program DENZO and Scalepack in the HKL2000 program suite (24). SaAraf43A crystals diffracted to ϳ2.0 Å resolution (space group P2 1 2 1 2 1 ).
Structural analysis of SaAraf43A was conducted through the multiwavelength anomalous dispersion method using Se-Metlabeled SaAraf43A crystals. Four selenium atom positions were determined, and initial phases were calculated using the program SOLVE/RESOLVE (25). Manual model rebuilding, introduction of water molecules, and molecular refinement were conducted using Coot and Refmac5 (26,27). One sodium ion and 1 chloride ion were added into the model. For the analyses of ligand binding structures of SaAraf43A, ␣-1,5-L-arabinofuranobiose, ␣-1,5-L-arabinofuranotriose, or Araf␣Az was dissolved in the crystallization precipitant to the concentration of 5% (w/v); 1 l of the ligand solution was added to the crystal drops and incubated for 20 -30 min, and the crystals were subjected to diffraction experiments. Structural determination was conducted through the molecular replacement method using the ligand-free structure as the starting model. Data collection and structure refinement statistics are given in Table 1. Structural drawings were prepared by the program PyMol (DeLano Scientific LLC, Palo Alto, CA).
The catalytic activity of SaAraf43A and its mutants was assayed using 100 nmol to 1 mol of enzyme incubated with 0.25-5 mM PNP-␣-L-Araf in McIlvaine buffer, pH 6.0, in a total volume of 0.5 ml at 30°C for up to 30 min. The amount of PNP released was determined at 400 nm with an extinction coefficient of 19,608 M Ϫ1 cm Ϫ1 . The assay was performed in triplicate.

RESULTS AND DISCUSSION
Overall Structure of SaAraf43A-The crystal structure of SaAraf43A was determined by the multiwavelength anomalous dispersion method using Se-Met derivative data. Successively, native and three ligand complex structures, SaAraf43A/␣-1,5-L-arabinofuranobiose, SaAraf43A/ ␣-1,5-L-arabinofuranotriose, and SaAraf43A/Araf␣Az, were determined. Structure refinement statistics are summarized in Table 1. The quality and accuracy of the final structures were further demonstrated in that more than 98% of their residues fall within the common regions of the Ramachandran stereochemistry plot. Recombinant SaAraf43A molecule is composed of a single polypeptide chain of 468 amino acids (0 -467), where N-terminal Met-0 and C-terminal 460 LEHHHHHH 467 were derived from the expression vector and purification tag. The N-terminal 9 residues Met-0 -Val-8 and the C-terminal 12 residues Leu-456 -His-467 were not identified because of lack of electron density. The final model consisted of 1 SaAraf43A molecule accompanied with 1 sodium ion, 1 chloride ion, and several glycerol molecules.
SaAraf43A is composed of 2 distinct domains (Fig. 1). The N-terminal catalytic domain is composed of a 5-bladed ␤-propeller that is built of 5 radially oriented "blades" (marked here as I-V, Fig. 1A), distributed almost equally around a full circle. Such a fold was first reported for tachylectin (29) and was found in three glycoside hydrolase families, GH32, GH43, and GH68, represented by invertase, ␣-L-arabinanase, and levansucrase, respectively. The C-terminal domain is made of three repeated-peptide segments referred to as subdomains ␣, ␤, and ␥ (Figs. 1B and 2). A combination of the three subdomains results in a fold similar to the "␤-trefoil fold" proposed by Murzin et al. (30). The ␤-trefoil fold is observed in two families, CBM13 and CBM42.
Sugar complex crystals were prepared by adding the ␣-1,5-Larabinofuranobiose, ␣-1,5-L-arabinofuranotriose, and Araf␣Az solution into SaAraf43A crystal drops. Then the structure of the sugar complex was determined. The relative positions of the domains did not change in comparison with the ligand-free structure. In the ␣-1,5-L-arabinofuranobiose complex, one L-arabinofuranosyl moiety was bound in the catalytic domain and in subdomain ␥ of the CBM42 domain. On the other hand, two L-arabinose moieties were found in subdomain ␣ of the CBM42 domain. In the ␣-1,5-L-arabinofuranotriose complex, three L-arabinofuranosyl moieties were observed in the catalytic domain, but two moieties were observed in subdomains ␣  OCTOBER 29, 2010 • VOLUME 285 • NUMBER 44 and ␥ (Fig. 1A). In contrast, in the Araf␣Az complex, one molecule was bound in the catalytic domain, and three molecules were found in the CBM42 domain, in all subdomains (Fig. 1B).

Exo-1,5-␣-L-arabinofuranosidase from S. avermitilis
Structure and Sugar Binding Manner of the Arabinan Binding Module-The C-terminal CBM42 module of SaAraf43A (SaCBM42) consisted of 3 repeated subdomains of ϳ50 residues, and 3 subdomains generated a ␤-trefoil fold (Figs. 1B and 2). The crystal structure of CBM42 had been reported only for the A. kawachii ␣-L-arabinofuranosidase, which is a modular enzyme consisting of GH54 and CBM42 (AkCBM42) (19). In AkCBM42, two sugar binding pockets (subdomains ␤ and ␥) were active. Comparison of the sugar complex structures showed that the L-arabinose binding structures were conserved in two proteins, and histidine and aspartate residues involved in arabinose binding are strictly conserved among all three subdomains of SaCBM42. However, the bound sugars were observed in only two subdomains ␣ and ␥ in the ␣-1,5-L-arabinofuranobiose or ␣-1,5-L-arabinofuranotriose complex structures (Fig. 1A).
The relative positions of the L-arabinofuranosyl moiety at the non-reducing end of the bound sugars and their binding manners were quite conserved between two subdomains as well with those of AkCBM42, whereas the positions of the reducingend L-arabinofuranosyl moiety differed between the two subdomains. This difference was caused by lesser contacts with the binding site and the close contact with symmetry-related molecules. The electron density for the reducing-end L-arabinofuranose was rather vague, and they were observed to be bound nonspecifically by the crystallographic circumstances. Therefore, the sugar-binding site of SaCBM42 basically recognized only one moiety of ␣-1,5-L-arabinofuranooligosaccharides.
In subdomain ␤ of CBM42, His-381 and Asp-400 hydrogenbonded to one glycerol molecule (Fig. 3B). This glycerol might have originated from the cryoprotectant. In contrast, in the complex structure with Araf␣Az, the bound Araf␣Az molecule was observed in subdomain ␤ as well as in subdomains ␣ and ␥ of SaCBM42. Therefore, subdomain ␤ would have the potential to bind L-arabinooligosaccharides. The side chains of Tyr-431 and Glu-432 in the symmetry-related molecule were considered to prevent the sugars from binding to the crystal through the sugar-soaking method. The binding manner of Araf␣Az was similar to those of the other subdomains, and four hydrogen bonds were present between the protein and the sugar (Fig. 3D).
The O5 atoms of the bound L-arabinofuranoses were located at the bottom of the binding site, recognized by two hydrogen bonds. That the O5 atom was buried indicated the sugar-binding sites of SaCBM42 specifically recognizes the terminal L-arabinofuranosyl moiety of the L-arabinofuranooligosaccharides and cannot bind the linear 1,5-L-arabinooligosaccharides across the two binding sites. This specificity agreed with the exo-type catalytic mechanism of the enzyme and the binding specificity of SaCBM42 that bound terminal and branched L-arabinofuranosyl residues, such as arabinan and arabinoxylan, described in the previous paper (18). SaCBM42 would play a role of finding the L-arabinofuranosyl residues among the insoluble substrate and aid catalysis by the catalytic module. CBM13 also has a ␤-trefoil fold and is separated into three homologous subdomains, ␣, ␤, and ␥. We have shown that CBM13 of S. avermitilis ␤-L-arabinopyranosidase or Streptomyces olivaceoviridis endo-␤-1,4-xylanase also has three sugar-binding sites (31,32). Each subdomain has one sugar-binding site. In these enzymes, three sugar-binding sites were believed to enhance the possibility of substrate binding. Similarly, the meaning of three binding sites of SaCBM42 should also be increasing the probability of insoluble substrate binding possessing L-arabinosyl terminus or side chains.
Structure and Sugar Binding Manner of the Catalytic Module-The 5-bladed ␤-propeller fold is a common structure for GH43, which includes inverting enzymes and is classified into clan GH-F together with GH62 (33). Families GH32 and GH68 also share a similar five-bladed ␤-propeller fold, but they are composed of retaining enzymes and are classified into clan GH-J. GH43 contains both endo-and exo-acting enzymes and has broad substrate specificities. A structural comparison of the catalytic cleft elucidated the difference in a substrate binding manner (Fig. 5). The active site of SaAraf43A is located on the central cavity of the ␤-propeller fold, forming a substrate binding pocket (Fig. 5A). This is very typical for the exo-mode action, by which SaAraf43A releases a single L-arabinose unit from the non-reducing end of linear arabinan or arabinooligosaccharides by cleaving the ␣-1,5-linkage. Geobacillus stearothermophilus ␤-xylosidase (XynB3) is also an exo-acting enzyme and has a catalytic pocket at the center of the ␤-propeller fold like SaAraf43A; however, the structures surrounding its catalytic pocket were not conserved (Fig. 5D) (34). In contrast, the endo-type enzymes Bacillus subtilis arabinan endo-␣-1,5-L-arabinosidase (BsArb43A) and Cellvibrio japonicus endo/ exo-␣-1,5-L-arabinanase (CjArb43A) possess a catalytic cleft across the surface of the ␤-propeller fold. The clefts of the two enzymes lie in almost the same orientation (Fig. 5, B and C) (33,35) and are deep enough so that the linear substrate could dock through the catalytic center. The length of the cleft of BsArb43A was as long as the L-arabinopentaose, and multiple subsites specific for arabinooligosaccharides might enable the endo-mode action of the enzyme. In the case of SaAraf43A, the longer loop region Tyr-281-Arg-294 (Asp-283-Asn-290 is shown in magenta in Figs. 1 and 5A) as well as the side chain of Tyr-40 occupied the positions corresponding to the glycon side of the cleft of BsArb43A. This hindrance probably conferred an exo-mode manner on SaAraf43A. B. subtilis arabinoxylan arabinofuranohydrolase (BsAXH-m2,3) also liberated L-arabinose from the substrate, although it is a side-chain-releasing enzyme (36). The xylotetraose-bound structure of BsAXH-m2,3 showed a different orientation of the sugar backbone in comparison with the substrate-bound structures of SaAraf43A or CjArb43A (Figs. 5, A, C, and E). BsAXH43A had a short loop region connecting ␤-strands 2 and 3 in blade I. In contrast, the corresponding loop region of CjArb43A and SaAraf43A is much longer, and therefore, the linear arabinan backbone could not bind the active site in the same orientation as BsAXH-m2,3. Thus, GH43 enzymes could change their substrate specificities by changing the structure of the loop region of the catalytic domain, yielding a variety of enzymatic activities.
Like other GH43 members, the three important acidic residues, Asp-20, Asp-135, and Glu-196, were conserved in SaAraf43A and were located at equivalent positions in the depth of the catalytic pocket (Figs. 5 and 6). The ␣-1,5-L-arabinofuranotriose complex structure of SaAraf43A revealed the recognition mechanism of the L-arabinofuranosyl substrate, where three L-arabinofuranosyl moieties were observed (Figs. 4B and 6). The three moieties seemed to be divided into

Exo-1,5-␣-L-arabinofuranosidase from S. avermitilis
L-arabinose and ␣-1,5-L-arabinofuranobiose. The non-reducing-end L-arabinofuranose was located at the bottom of the catalytic pocket, whereas the two reducing-end L-arabinofuranose residues were positioned at the outside of the pocket; their positions were called subsite Ϫ1 and subsites ϩ1 and ϩ2, respectively, where the boundary was proximal to the side chain of Glu-196. The electron density of L-arabinose at subsite Ϫ1 or subsite ϩ2 was rather obscure, and the average B-factor values were 45.9, 29.9, and 43.9 A 2 for the 3 moieties at subsites Ϫ1, ϩ1, and ϩ2. The bound ligand, therefore, could not be identified as a product or a substrate, but the structure of L-arabinose at subsite Ϫ1 was identified as ␣-furanoside for the present model. The side chain of Glu-196, which is 1 of the 3 conserved acidic residues, was proximal to the O1 atom of the L-arabinose moiety at subsite Ϫ1, or the O5 atom at subsite ϩ1. Therefore, Glu-196 was considered to be the catalytic acid of the enzyme. Subsite Ϫ1 consisted of Asp-20, Thr-36, Tyr-40, Trp-76, Leu-134, Asp-135, His-260, and Arg-294. Five hydrogen bonds were observed: between the O2 and Asp-135-O␦2 atoms, between the O3 and Asp-135-O␦1 atoms, and between the O5 atom and the 3 protein atoms Asp-20-O␦2, Arg-294-N1, and Arg-294-N2. The indole group of Trp-76 provided a platform contacting with the C3-C4-C5 plane. Trp-76 was conserved in ␣-Larabinanases of GH43, such as CjArb43A and BsArb43A, and subsite Ϫ1 could be mentioned to have similar structure among GH43s, although Tyr-40 was not conserved in CjArb43A and BsArb43A. SaAraf43A hydrolyzes ␣-1,5-linked arabinan in the exo-acting manner and liberates L-arabinose. The exo-acting mechanism was expressed probably because subsite Ϫ1 was completely surrounded by amino acid residues including the loop region Tyr-281-Arg-294, as mentioned above, and only one L-arabinosyl moiety could be docked in the glycon subsite.
The electron density of the L-arabinofuranosyl moiety at subsite ϩ1 was observed clearly, and it has the envelope 1 E conformation (Fig. 6).

Exo-1,5-␣-L-arabinofuranosidase from S. avermitilis
the wall on the other side. Hydrogen bonds were observed between the O1 and O3 atoms and the Asn-159-N␦2 atom and the O5 and Glu-196-O⑀1 atoms. The position of the O5 atom corresponded to the scissile ␣-1,5-glycoside bond of the catalysis. Another ␣-1,5-glycoside bond between the L-arabinofuranosyl moiety subsites ϩ1 and ϩ2 was held by the Tyr-192, and its O5 atom was hydrogen-bonded by the Asn-159-N␦2 atom. The structure of subsite ϩ1 would be important for the substrate specificity against the ␣-1,5-glycosidic bond. Three bound water molecules were clearly observed, which also mediated hydrogen network to the protein.
The L-arabinofuranosyl moiety at subsite ϩ2 was embedded on the indole ring of Trp-100 and was partially supported by the aromatic rings of Phe-132 and Tyr-192. Including Asn-159, 4 residues in total constructed subsite ϩ2. When the catalytic efficiency of SaAraf43A against ␣-1,5-linked L-arabinofuranooligosaccharides was investigated, the activity of the enzyme increased according to the degree of polymerization of the substrate, namely k cat /K m values of 426 min Ϫ1 ⅐mM Ϫ1 for arabinobiose increased to 5,556 min Ϫ1 ⅐mM Ϫ1 for arabinotriose. The hydrolysis rates between arabinotriose and arabinopentaose were almost the same, implying that the enzyme contains three significant L-arabinose-binding subsites (Table 2). Indeed, the sugar binding structure showed that the active site is composed of three subsites (Figs. 4B and 6). Subsite ϩ2 was located on the surface of the catalytic pocket, and therefore, the activity of SaAraf43A against ␣-1,5-linked arabinotriose and arabinopentaose would be almost at the same level.
The O2 and O3 atoms of the arabinofuranose at subsite ϩ2 in the ␣-1,5-L-arabinofuranotriose complex structure of SaAraf43A were exposed to solvent, whereas those atoms at subsites ϩ1 and Ϫ1 were not exposed. The structure indicated that SaAraf43A could not degrade decorated arabinan when its terminal or second L-arabinose was substituted by the sugar side chain, and this is why SaAraf43A did not show hydrolyzing activity against decorated arabinan, reported previously (18).
Mutagenesis Study of SaAraf43A-To verify the catalytic residues of SaAraf43A, some mutants were generated by PCR at the putative catalytic centers, Asp-20, Asp-135, and Glu-196. These residues hydrogen bonded to the L-arabinofuranosyl moiety at subsite Ϫ1 in the ␣-1,5-L-arabinotriose complex structure and were conserved among GH43 enzymes.  was considered to be the catalytic acid of the enzyme as the side chain of Glu-196 was located close to the O1 atom of the Larabinose moiety at subsite Ϫ1. Enzymatic analyses showed that the catalytic activities of the single amino acid variants, D20A, D135A, D135N, and E196A, vanished, and these residues were proven to be indispensable for catalysis (Table 3). However, only D20N maintained an extremely weak activity; mutation of this amino acid reduced the values of k cat and K m , suggesting that the mutant has an increased affinity for a substrate but hardly released a product. The Asp-20-O␦2 atom hydrogen-bonded to the O5 atom of L-arabinofuranosyl moiety at subsite Ϫ1, and the Asp-20-O␦2 atom was 4.4 Å away from the anomeric C1 atom. However, there was one bound water molecule hydrogen bonded to the Asp-20-O␦2 atom and located close to the C1 atom (2.7 Å distance; Fig. 4B). Catalysis by family 43 enzymes occurs with inversion of the anomeric configuration (37), and this water could be considered to play as a catalytic nucleophile, representative for the inversion mechanism. When Asp-20 was substituted by asparagine, nucleophilicity of the water would decrease, but it was considered to maintain the attacking ability, resulting to the weak activity of the mutant. Mutagenesis studies of CjArb43A suggested that Glu-221 functions as a catalytic acid, Asp-38 functions as a catalytic base, and Asp-158 functions in pK a modulation and in maintaining the correct orientation of the general acid residue (33). Our analyses on SaAraf43A provided equivalent results.
The substrate recognition mechanism of exo-1,5-␣-Larabinofuranosidase was also assessed by mutagenesis. According to the above structural analysis, three amino acids, Asn-159, Tyr-192, and Leu-289, were selected for the mutagenesis study. First, kinetic analysis with PNP-␣-L-Araf was performed ( Table  3). The K m values of N159A and N159L mutants were 1.7 and 1.1 mM, respectively, which were almost similar to the K m value of the wild type enzyme. However, the k cat values of the mutants were 5-40 times larger than that of the wild type. The hydrolysis activities of the N159A and N159L mutants were higher than those of the wild type enzyme, suggesting that Asn-159 has no considerable effect on substrate binding and probably discriminates L-arabinofuranose from the other sugars. The muta-  tion of Asn-159 vanished two hydrogen bonds to the substrate but created hydrophobic environment at subsite ϩ1, which would be suitable for the binding of hydrophobic substances such as PNP. Hydrophobicity produced by leucine or alanine might enhance the hydrolytic activity against the PNP-␣-L-Araf. Unlike Asn-159 mutants, the k cat values of Y192A and L289A were increased; however, the K m values of the mutants were 20 -40 times larger than that of the wild type. The activities of Y192A and L289A were lower than that of the wild type enzyme, suggesting that Tyr-192 and Leu-289 were involved in substrate binding. As shown by the SaAraf43A/␣-1,5-L-arabinotriose complex structure, Tyr-192 contacts with the plane of the ␣-1,5-glycoside bond between the L-arabinofuranosyl moieties of subsites ϩ1 and ϩ2, but the side-chain Leu-289 was located away from the bound sugars with distances of more than 6 Å. Thereby, the position of Leu-289 was examined in the other structural models, ligand-free and sugar complex structures with ␣-1,5-L-arabinofuranobiose and Araf␣Az (Fig. 7). The region of Asp-283-Asn-290 formed an extended loop, and Pro-288 -Asn-290 formed a 3 10 -helix on the tip of the loop (Fig. 1). The position of the loop point was different between the structures, and the largest shift of 2.6 Å was observed for the main chain of Asp-287. The B-factor values for this region were higher than the average values, and this loop was considered to be flexible. The side chain of Leu-289 was considered to possibly move toward subsites Ϫ1 and ϩ1, and the estimated distance to the bound sugars was minimized to 4.2 Å as judged from the superposed model, which could be considered as a hydrophobic contact. Therefore, this loop region, containing Leu-289, was shown to bind the substrate by capping the subsites and also to play a role in uptake the substrate by taking an open form. Second, the detailed substrate specificity of the enzymes was characterized by using various L-arabinofuranobiose with different types of glycosidic bond. The hydrolysis rate of three kinds of regioisomer of biosides by wild type and mutants are shown in Table 4. Interestingly, unlike against PNP-␣-L-Araf, the activity of all the mutants against ␣-1,5-linked L-arabinofuranobiose decreased compared with the wild type. Although Asn-159 mutants exhibited higher activities toward PNP-␣-L-Araf than the wild type enzyme, their activity toward ␣-1,5-linked L-arabinofuranobiose was lower than the wild type enzyme. Asn-159 is considered in the form making a suitable space for the fitting of the natural substrate such as ␣-1,5linked L-arabinofuranobiose by forming hydrogen bonds at subsite ϩ1. For Y192A and L289A mutants, a low activity toward both PNP-␣-L-Araf and ␣-1,5-linked L-arabinofuranobiose was observed; however, improved substrate specificities are shown in Table 4 and Fig. 8. The mutants showed activity against ␣-1,2-linked L-arabinofuranobiose

Exo-1,5-␣-L-arabinofuranosidase from S. avermitilis
in addition to their activity against ␣-1,5-linked L-arabinofuranobiose. As described above, Tyr-192 and Leu-289 regulate the substrate specificity of SaAraf43A. Tyr-192 is considered to fix the substrates at subsites ϩ1 and ϩ2 by hydrophobic interaction with its aromatic ring. In the Y192A mutant, strictness of the subsite ϩ1 decreased, and ␣-1,2-linked L-arabinofuranobiose could be bound at subsite Ϫ1 and ϩ1. In contrast, Leu-289 did not show the direct interaction with substrates, but it would have an important role to fix the substrates at both subsites Ϫ1 and ϩ1, as mentioned previously. L289A mutation would decrease rigidity of the substrates around these subsites. Therefore, the k cat /K m value for ␣-1,5-linked L-arabinofuranobiose was significantly reduced comparing with wild type enzyme or Y192A mutant. L289A mutant also showed hydrolyzing activity against ␣-1,2-linked L-arabinofuranobiose at the same level with Y192 mutant but broader glycosidic-linkage specificity than Y192A mutant. XynB3, an exo-type enzyme belonging to GH43 like SaAraf43A, hydrolyzes substrates with various aglycon leaving groups (38). In the XynB3-xylobiose complex, the position of the xylose at subsite Ϫ1 is maintained by a large number of hydrogen bonds, whereas the aglycon xylose unit at subsite ϩ1 was bound much less tightly (34). In the case of SaAraf43A, the residues surrounding subsite ϩ1, such as Tyr-192 and Leu-289, would cause tight substrate specificity.
This study provides a good model for investigating the hydrolytic mechanism of the modular enzyme. As the function of SaCBM42 has been demonstrated biochemically (18), our structural analysis showed that SaCBM42 assists in the catalysis by binding the L-arabinose moiety of an insoluble substrate through three arabinan-binding sites. This study also provides a good model for investigating the mode of action of exo-and endo-type enzymes because the GH43 family includes different types of arabinan-degrading enzymes. For example, amino acids involved in aglycon subsites ϩ1 and ϩ2 and their SaAraf43A structures were not conserved in CjArb43A or BsArb43A, which showed endo-type catalytic properties. Therefore, the difference between the endo-and exo-modes of action against arabinan or arabinooligosaccharides in GH43 was attributed not only to the structure of the glycon site but also to the fact that the overall structures of the catalytic clefts were basically different in different enzymes.