Crystal Structures of 4- (cid:1) -Glucanotransferase from Thermococcus litoralis and Its Complex with an Inhibitor*

Thermococcus litoralis 4- (cid:1) -glucanotransferase (TLGT) belongs to glucoside hydrolase family 57 and catalyzes the disproportionation of amylose and the formation of large cyclic (cid:1) -1,4-glucan (cycloamylose) from linear amylose. We determined the crystal structure of TLGT with and without an inhibitor, acarbose. TLGT is composed of two domains: an N-terminal domain (domain I), which contains a ( (cid:2) / (cid:1) ) 7 barrel fold, and a C-terminal domain (domain II), which has a twisted (cid:2) -sandwich fold. In the structure of TLGT complexed with acarbose, the inhibitor was bound at the cleft within domain I, indicating that domain I is a catalytic domain of TLGT. The acar-bose-bound structure also clarified that Glu 123 and Asp 214 were the

(TLGT) 1 (EC 2.4.1.25) plays a key role, producing glucose and a series of maltodextrins through intermolecular transglycosylation of maltose that has been transported into the cells (1). In addition to intermolecular transglycosylation, TLGT also catalyzes intramolecular transglycosylation in vitro, where it cyclizes amylose to produce cyclic ␣-1,4-glucans (cycloamyloses, CAs) (2) with 16 to several hundred glucose units. 2 The degree of polymerization of known CAs varies, from six to several hundreds of glucose units. ␣-, ␤-, and ␥-cyclodextrins, which are the smallest CA species, consisting of 6, 7, and 8 glucose units, respectively, are well known doughnut-shaped rigid molecules and are able to accommodate guest molecules in their central cavity, yielding inclusion complexes. Although larger species of CA, which were recently found to be products of potato 4-␣glucanotransferase (3), also form inclusion complexes, they have flexible single-helical conformations in an aqueous solution unlike cyclodextrins (4), and they lose their flexibility and fold into a compact structure during complex formation (5). Because of this structural difference from cyclodextrins, large CAs show the advantageous features of the formation of inclusion complexes and higher solubility in water and thus are expected to be valuable for future industrial use (6). Interestingly, an artificial chaperone activity of large CAs has also been reported (7).
According to Henrissat's classification (8,9), TLGT belongs to family 57 of the glycoside hydrolases. Most family 57 enzymes catalyze reactions similar to those of some ␣-amylase family members (families 13, 70, and 77). However, no sequence similarity has been detected between family 57 and ␣-amylase family enzymes. The three-dimensional structures of many ␣-amylase family enzymes, including that of Thermus aquaticus amylomaltase (10), which produces large CAs, have been determined (11)(12)(13), and the amino acid residues involved in the catalysis have also been studied extensively (for a review, see Ref. 14). In contrast to ␣-amylase family enzymes, family 57 enzymes have received less investigation. Although the catalytic nucleophile of TLGT was recently determined (15), its three-dimensional structure, acid/ base catalyst, and mechanism for large CA production remain unknown.
Previously, we made a preliminary report of the crystal structure of TLGT (16). In this paper, we describe the detailed structures of TLGT with and without a tetrasaccharide inhibitor, acarbose. The structures revealed the residues involved in catalysis and substrate binding of the enzyme and provided insight to investigate the mechanism for the production of large CAs.

EXPERIMENTAL PROCEDURES
Purification and Crystallization-The nonlabeled enzyme was expressed in Escherichia coli strain BL21(DE3) and purified as described previously (15) with some modifications. In the last purification step, the buffer was changed to 5 mM Tris-HCl, pH 7.5, and the enzyme was concentrated to 10 -20 mg/ml by centrifugal filtration using a Centriplus 30 (Millipore). The seleno-methionine-labeled enzyme was coexpressed in the methionine auxotrophic E. coli strain B834(DE3) (Novagen) with GroELS and tRNA cognates for AGA and AGG in a medium containing seleno-methionine. The purification procedures for the labeled enzyme were the same as those for the nonlabeled enzyme, except that the buffer used in the last step was replaced by 5 mM Tris-HCl, pH 7.5, containing 10 mM dithiothreitol.
Form I crystals were obtained at 25°C by the hanging drop vapor diffusion method by mixing 5 l of protein solution with 5 l of a reservoir solution comprising 1.9 M ammonium sulfate, 2% (w/v) polyethylene glycol 400, and 0.1 M HEPES-NaOH, pH 7.5. The crystals were harvested in a solution comprising 2.2 M ammonium sulfate, 2% (w/v) polyethylene glycol 400 and 0.1 M HEPES-NaOH, pH 7.5. The concentration of trehalose (a gift from Hayashibara (Okayama, Japan), recrystallized before use) in the harvesting solution was increased stepwise to 25% (w/v), and then the crystals were subjected to flash freezing before data collection. Form II crystals were grown at 25°C by the sitting drop vapor diffusion method. The drops were made by mixing 5 l of protein solution with 5 l of a reservoir comprising 35% (w/v) 2,4-dimethylpentanediol (Hampton Research), 20 mM calcium chloride, and 0.1 M Tris-HCl, pH 8.0. The crystals were transferred to a solution comprising 40% (w/v) 2,4-dimethylpentanediol, 20 mM calcium chloride, and 0.1 M Tris-HCl, pH 8.0, followed by flash freezing. Acarbose-bound crystals were obtained by soaking FIG. 1. Schematic representation of TLGT. a, two TLGT monomers in an asymmetric unit of form II-free. Calcium ions are presented as yellow space-filled models. Two catalytic residues and Tris are presented as stick models. b, stereoview of a C ␣ trace of a TLGT monomer. Chain A of form II-free is shown. Every 40 th C ␣ is represented as a sphere. The molecule is viewed from the same perspective as in a. Figs. 1, 3, 4c, 6, and 8 were prepared with Molscript (44) and Raster3d (45). crystals overnight in a solution comprising 40% (w/v) 2,4-dimethylpentanediol, 20 mM calcium chloride, 10 mM acarbose (a gift from Bayer), and 0.1 M Tris-HCl, pH 8.0, followed by flash freezing. Data Collection-A multiple wavelength anomalous dispersion data set for a seleno-methionine-labeled form I crystal was collected at wavelengths of 1.0400 Å (remote), 0.9793 Å (peak), and 0.9795 Å (edge) at 100 K on beamline BL45PX of SPring-8 (Hyogo, Japan). A native data set for a form II crystal was collected at a wavelength of 1.0200 Å at 100 K on beamline BL45PX of SPring-8. An acarbose-bound data set for a form II crystal was collected at a wavelength of 1.0000 Å at 100 K on beamline BL18B of the Photon Factory (Tsukuba, Japan). The first two data sets were processed with DENZO and Scalepack (17), and the last one was processed with Mosflm (18).
Phase Calculation and Refinement-For phase calculation of the multiple wavelength anomalous dispersion data set, the program SOLVE (19) was used. The multiple wavelength anomalous dispersion map was improved by density modification using the program DM (20). The model for "form I-Se" was built into the resultant electron density map using the program O (21) and refined to 2.8 Å resolution, including simulated annealing, bulk solvent correction, and grouped B-factor refinement with the program CNS (22). Water molecules were picked automatically from F o Ϫ F c electron density maps using the program CNS and checked manually at a graphic station. The phase for form II-free crystal was calculated by molecular replacement using form I-Se as the initial model using the program CNS. Noncrystallographic sym-metry was found on the self-rotation function, but it was not used for cross-rotation function nor translation function. Density modification was not applied to the molecular replacement solution. The calculated model was refined to 2.4 Å resolution, including rigid body minimization, simulated annealing, and grouped B-factor refinement with the program CNS. The two TLGT monomers in the asymmetric unit were not restrained during refinement, because there are some variations in the conformation of the two molecules as described later. The final model was found to exhibit good geometry, as determined using the program Procheck (23); 88.7% of the residues have / angles in the "most favored region" of a Ramachandran plot. The model for form II complex was also refined using the program CNS. During refinement of form II complex, the ligands were added based on 2F o Ϫ F c and F o Ϫ F c electron density maps. The refinement statistics are presented in Table I. A structural data base search was performed using the DALI server (24). Least mean square fitting of the structures was carried out with the program LSQMAN (25). The oligosaccharide model shown in Fig. 7 was constructed using the program XtalView (26).
Site-directed Mutagenesis and Enzyme Assay-The D214N mutant was generated with a QuikChange site-directed mutagenesis kit (Stratagene). An oligonucleotide with the sequence 5Ј-GTGTTCCATGA-CAATGGTGAAAAGTTCGG-3Ј and a complementary oligonucleotide, which replaced the codon for Asp 214 (GAC) with AAT and introduced a HphI site for rapid screening of the mutation, were used. The mutation was reconfirmed by sequencing with an ABI PRISM 310 DNA se-  (46). The secondary structure was assigned with DSSP (47). Identical residues are shown in white on red, and homologous residues are in red letters. Arrowheads indicate the two catalytic residues. The closed blue circles represent residues whose side chains are involved in substrate binding. The residues indicated by closed yellow circles are involved in calcium binding. Open purple and green circles indicate the residues that contribute to the intermonomer interactions via hydrophobic interactions and hydrogen bond networks, respectively. Functions of the residues were determined for TLGT as described in this study. This figure was generated using ESPript (48). quencer (Applied Biosystems). Activity toward maltotriose was measured as described previously (27). One unit of activity was defined as the amount of enzyme that liberated 1 mol of glucose from maltotriose per min at 80°C.

RESULTS
Structure Determination-TLGT was crystallized in two forms (forms I and II). Form I crystals belonged to hexagonal space group P6 4 22 and form II crystals to orthorhombic space group P2 1 2 1 2. First, we determined the structure at 2.8 Å resolution by multiple wavelength anomalous dispersion method using a form I crystal labeled with seleno-methionine. The model was refined to an R-factor of 22.1% (Table I). We called this structural model form I-Se. The form I crystal contained one TLGT monomer/asymmetric unit. Next, we determined the structures of a form II crystal through molecular replacement in the free form, designated as form II-free, and in a complex form with acarbose, designated as the form II-complex. Acarbose, an inhibitor of ␣-amylase family enzymes, ␤-amylases, and glucoamylases, is a maltotetraose analog composed of a nonreducing end inhibitory group, acarviosine, and a reducing end maltose. Acarbose also inhibits the activity of TLGT. 3 The form II crystal contained two TLGT monomers (chains A and B)/asymmetric unit (Fig. 1a). The form II-free structure was refined to an R-factor of 19.5% at 2.4 Å resolution (Table I).
Domain I-The core of domain I is a (␤/␣) 7 barrel (residues 1-258), in which seven central parallel ␤-strands (␤1, ␤2, ␤3, ␤4, ␤5, ␤7, and ␤8) are flanked by helices (Fig. 3a). The (␤/␣) 7 barrel is followed by a helical region that consists of 10 helices including the three-helix bundle made of ␣16, ␣17, and ␣18 (Figs. 1a and 3b) and covers the C termini of the (␤/␣) 7 barrel, forming a cleft between them. The observation that acarbose was bound in the cleft in the form II complex (see below) and that the catalytic nucleophile Glu 123 (15) was located in the middle of the cleft indicates that the cleft is an active site and domain I is a catalytic domain.
Calcium-binding Site-A calcium ion was bound at the loop between ␤10 and ␤11 in domain II (Figs. 1a and 2). The calcium ion was coordinated with O␦1 of Asp 392 , O␦2 of Asp 394 , O␦1 of Asp 396 , the main chain carbonyl oxygen of Arg 398 , and O⑀1 and O⑀2 of Glu 400 .
In form II-free, one additional calcium ion was identified between Glu 60 of chain B and Asn 248 of chain A of an adjacent asymmetric unit (data not shown). Because no form II crystals were observed when calcium chloride was omitted from the crystallization buffer, the latter calcium ion may promote crystallization by tightening the interaction between the two TLGT molecules.
Catalytic Residues-An acarbose-TLGT complex was obtained by soaking a form II crystal in a buffer containing acarbose. The structure of the complex was determined at 2.4 Å resolution and refined to an R-factor of 19.8% (Table I). Electron density corresponding to an intact acarbose molecule was clearly observed for the active site of chain A (Fig. 4a). As expected, the acarbose molecule bound to subsites Ϫ1 to ϩ3, where the acarviosine moiety, the inhibitory disaccharide group of acarbose, occupied subsites Ϫ1 and ϩ1 and the maltose moiety occupied subsites ϩ2 and ϩ3 (Fig. 5a). The nomenclature for the subsites is according to Davies et al. (29). This is the inhibitory binding mode observed for most of the structures of ␣-amylase family enzymes in complex with acarbose (30). 3 H. Imamura and H. Matsuzawa, unpublished results. Enzymatic hydrolysis of glycosidic linkages can be classified into two major types according to the anomeric configuration of the product, retaining and inverting, and in both cases the catalytic residues are typically two carboxylates (31). In retaining glycoside hydrolases, such as TLGT (2, 15), one residue acts as a nucleophile and the other as an acid/base catalyst. Glu 123

FIG. 4. Bound ligands in the form II complex of TLGT.
Stereoviews of the refined models of acarbose in chain A (a) and maltose in chain B (b) are presented, together with F o Ϫ F c electron density maps contoured at 4 . Each ligand was omitted from the phase calculation. c, acarbose and the two catalytic residues. a and b were generated using XtalView (26) and Raster3d (45). is close (3.15 Å) to the C1 atom of the valienamine moiety at subsite Ϫ1, allowing nucleophilic attack (Fig. 4c), which is consistent with the results of a cross-linking study that demonstrated Glu 123 to be a catalytic nucleophile (15). It is considered most likely that Asp 214 is the acid/base catalyst of TLGT, because O␦2 of Asp 214 is only 2.96 Å away from the amide group of the valienamine moiety (this amide group is replaced by a glucosidic oxygen in a native substrate) (Fig. 4c), and there is no other acidic residue nearby. The average distances between all four pairs of O atoms of Glu 123 and Asp 214 (6.72 and 6.97 Å in the acarbose-free and acarbose complex structures, respectively) are in appropriate range for retaining enzymes (32). We generated the D214N mutant, in which Asp 214 was replaced by Asn. The specific activity of the D214N mutant (0.0016 units/mg) was decreased about 10,000-fold as compared with that of the wild-type enzyme (17.7 units/mg). These results indicate that Asp 214 is the acid/base catalyst.
Subsite Structure- Fig. 5a shows interactions between acarbose and the protein. The valienamine moiety at subsite Ϫ1 is within hydrogen bonding distance of His 11 and Asp 354 and interacts with His 13 , Glu 216 , and Trp 357 via water molecules. The 6-deoxyglucose moiety is fixed at subsite ϩ1 with Arg 124 , Asp 213 , and Asp 214 through hydrogen bonds, with Tyr 272 through aromatic stacking, and with Trp 221 through a hydrophobic interaction. The glucose moiety at subsite ϩ2 is bound to Arg 182 and Asp 213 though hydrogen bonds, to Tyr 183 through an aromatic stacking interaction, and to Trp 221 through a hydrophobic interaction. The glucose moiety at subsite ϩ3 seems to be more flexible, because it only exhibits a stacking interaction with Phe 187 and a hydrophobic interaction with Trp 221 , i.e. no hydrogen bonding interaction with the protein.
In contrast to chain A, acarbose did not bind at the active site of chain B of the form II complex, and only a Tris molecule was identified there (data not shown). Tris was also found at the active site of chain B of the form II-free (Fig. 1a). Although acarbose did not bind to the active site, we found electron density corresponding to disaccharide, which seems to be maltose, at the edge of the active site cleft of chain B (Fig. 4b). The reducing end of maltose is ϳ14.5 Å apart from subsite Ϫ1, suggesting that this binding site corresponds to subsites Ϫ5 and Ϫ6. The glucose moiety at subsite Ϫ5 is bound to His 368 through a hydrogen bond and to Phe 19 through a hydrophobic interaction (Fig. 5b). The glucose moiety at subsite Ϫ6 is within hydrogen bonding distance of Arg 371 and exhibits hydrophobic interactions with Phe 19 and Tyr 601 (Fig. 5b). In chain A, maltose did not bind to this site, because this region was involved in the crystal contact. From the complex structure with acarbose, TLGT was revealed to possess at least nine subsites, Ϫ6 to ϩ3, although the residues forming subsites Ϫ4 to Ϫ2 could not be determined in this study.
Unexpectedly, a glucose molecule was found at subsite ϩ1 of form I-Se (data not shown). Glucose interacted with Arg 124 , Asp 214 , Asp 213 , and Tyr 272 . Interactions between glucose and these residues in form I-Se are the same as those observed between the ϩ1 glucose moiety of acarbose and the corresponding residues in the form II complex. Glucose was probably derived from trehalose reagent, which was used as a cryoprotectant and contained a trace amount of glucose.
The active site cleft of TLGT is tunnel-like in shape, as evidenced by the three lids that cover the cleft (Figs. 6 and 7). The first lid (lid 1, residues 220 -224) protrudes from the (␤/␣) 7 barrel (Fig. 2). The second (lid 2, residues 358 -363) and third (lid 3, 627-630) lids protrude from the three-helix bundle and domain II, respectively (Fig. 2). Upon binding of acarbose, the conformations of lids 2 and 3 change significantly (Fig. 6). In the absence of acarbose, the side chains of Val 360 and Phe 361 are directed toward the active site cleft. The bound acarbose collides with the side chains of Val 360 and Phe 361 , leading to movement of lid 2. This movement induces a large movement of lid 3 (Fig. 6). For example, the C ␣ of Ser 627 and Glu 628 move 4.2 and 6.2 Å, respectively. In addition to the movements of these two regions, the 1 axis of Tyr 183 and the 2 axis of Phe 187 rotated to interact with the pyranose rings of the maltose moiety via hydrophobic stacking at subsites ϩ2 and ϩ3, respectively (Fig. 6).
Subunit Interface-Although the TLGT gene encodes a 78-kDa polypeptide, purified TLGT is eluted as a 168-kDa protein in gel filtration chromatography using Superdex 200 column (Amersham Biosciences) (data not shown). This indicates that TLGT is a homodimer, as previously suggested by Xavier et al. (1). In a form II crystal, two TLGT monomers, which are in an asymmetric unit, interact via the same surface of domain I, because there is a pseudo-2-fold axis between them (Fig. 1a). In a form I crystal, two TLGT monomers, which are in adjacent asymmetric units associated through a 2-fold axis, interact in FIG. 5. Schematic drawing of subsite structures of TLGT. a, interactions between acarbose and residues at subsites Ϫ1 to ϩ3. b, interactions between maltose and residues at subsites Ϫ6 and Ϫ5. the same manner (data not shown). This observation indicates that the two TLGT monomers interact in the same manner as observed in these crystals. Two hydrogen bond networks and a hydrophobic patch form the primary contribution to the interactions between the two subunits (Fig. 8a). One hydrogen bond network is constructed from two water molecules and six residues: Glu 166 and Tyr 266 from one monomer, and Ala 91 , Lys 314 , Asn 324 , and Lys 328 from the other (Fig. 8a). The hydrophobic patch is formed by Leu 287 , Phe 288 , Phe 291 , Leu 295 , Tyr 304 , Phe 307 , and Val 308 (Fig. 8b). In particular, Leu 295 and Val 308 are conserved (Fig. 2) and in contact with their counterparts in the other monomer (Fig. 8b), which suggests that they play a central role in the hydrophobic contact in this region. The proportion of the buried surface area (ϳ1500 Å 2 ) is large enough for dimerization, compared with the total molecular surface of a monomer (ϳ22300 Å 2 ). Oligomerization is known to be one of the strategies by which proteins acquire thermostability (33). Because the dimer interface is located on the opposite side of the active site, dimerization seems to contribute to the thermostability rather than the activity including amylose cyclization.

DISCUSSION
One of the interesting features of TLGT is the (␤/␣) 7 barrel fold, which forms the core of domain I (Fig. 3a). Proteins having the (␤/␣) 7 barrel fold are very rare (34), although there is a large number of ␤-barrel proteins. Cellobiohydrolase (35) and endoglucanase (36), which both belong to glycoside hydrolase family 6, are two examples of (␤/␣) 7 barrel proteins. The central ␤-sheet of the barrel of family 6 proteins is not completely closed, because there is only one hydrogen bond between the main chains of the first (␤1) and last (␤7) strands. The barrel of TLGT has a more open conformation, because there are two such "nonclosures" in the barrel of TLGT: between ␤3 and ␤4 and between ␤7 and ␤8.
Previously, we found that the nucleophilic residues of TLGT and class II ␣-mannosidases (glycoside hydrolase family 38) are located at the same position in the amino acid sequences despite their low sequence similarity (15). When the two families were compared at the three-dimensional level, striking structural similarities were observed in the catalytic domains of TLGT (Fig. 3b) and Drosophila melanogaster Golgi ␣-mannosidase II (37) (Fig. 3c): a ␤-sheet core surrounded by ␣-helices, a three-helix bundle, and a catalytic nucleophile at the fourth ␤-␣ loop. These findings strongly suggest that the two families have evolved from a common ancestor. Despite the similarities, however, there are also significant structural differences between the two enzymes. The most important one is that the catalytic domain of ␣-mannosidase II does not have a barrel structure, because there is no hydrogen bond that connects the two ␤-strands (7 and 8 in Fig. 3c) corresponding to ␤7 and ␤8 of TLGT. The ␣␤ fold in ␣-mannosidase II may be a result of protein evolution from the (␤/␣) 7 barrel fold found in TLGT and vice versa. This difference is interesting from the standpoint of the evolution of protein folding.
In the crystal structures of amylase-acarbose complexes, acarbose was often found in a modified form, because of the transglycosylation activity of amylase (see Ref. 30). However, it is clear that acarbose bound to the active site of TLGT (chain A in the form II-complex) is an intact one, because the electron density indicates that the sugar ring at the subsite Ϫ1 is somewhat flattened and that O-6 is not present at the subsite ϩ1. Why maltose bound to the subsite Ϫ5 and Ϫ6 of chain B in the form II-complex is uncertain. Because maltose was not detected in the acarbose reagent when analyzed on thin layer chromatography (data not shown), acarviosine moiety of acarbose may have flexible conformation, and only maltose moiety may be observed as a clear electron density. Additional electron density at the nonreducing end seen in Fig. 4b can be explained by this idea. Such a case has been also reported in the crystal structure of Bacillus circulans xylanase complexed with xylotetraose, in which electron density for only two xylose residues was observed (38). It is also unclear why acarbose did not bind to the active site of chain B in the form II crystal. Although TLGT has a larger K i value (ϳ0.6 ϫ 10 Ϫ3 M) for acarbose than ␣-amylase family enzymes (0.6 ϫ 10 Ϫ7 to 0.8 ϫ 10 Ϫ4 M) (39), the acarbose concentration (10 mM) in the soaking solution is sufficient. One possibility is that conformational changes in chain B prevent the binding of acarbose. When compared with form I and chain A of form II, the slight movements of ␣8, ␣9, ␣10, ␣11, ␣13, ␣14, 3 10 5, ␤22, and ␤23, and the destruction of some ion pairs in chain B were observed (data not shown), probably because of crystal packing. These movements seem to change chain B into an inactive form that cannot bind the substrate.
There are several kinds of 4-␣-glucanotransferases, including cyclodextrin glucanotransferase (CGTase), many of which produce CAs (6). The minimum ring size of CAs varies with the enzyme. The smallest products of CGTase are CAs with 6 -8 glucose units. Potato 4-␣-glucanotransferase (3) and T. aquaticus amylomaltase (40) produce CAs with 17 or more and 22 or more glucose units, respectively. Sixteen is the minimum ring size of CAs produced by TLGT 2 and Thermococcus kodakaraensis 4-␣-glucanotransferase (also a family 57 enzyme) (41). What determines the minimum ring size of CAs produced by these enzymes? It has been proposed that in the case of large CA formation by T. aquaticus amylomaltase, the product wraps around the bulky loop, which partially covers the active site (30). In addition to the loop structure, it was thought that the existence of a second ligand-binding site apart from the catalytic site also contributed to the large CA formation (30). In contrast, there is no such large steric hindrance in CGTase. Although T. aquaticus amylomaltase and TLGT have different structures, they seem to have adapted a similar strategy to produce large CAs. In the form II-complex of TLGT, acarbose partially wraps around lid 1 (Fig. 7). Lid 1 seems to prevent the formation of small CAs because of its considerable steric hindrance. Because amylose forms a helical structure in which glucose units are connected through O-3-O-2Ј hydrogen bonds, ␣-1,4-linked glucose polymers tend to become circularized in an aqueous solution. In fact, the reorganization of O-3-O-2Ј hy-drogen bonds is proposed to be one of the forces in the circularization step responsible for the formation of cyclodextrins by CGTase (42). If there is no steric hindrance by the loop observed in TLGT and amylomaltase, smaller CAs might be produced. When we included a polysaccharide chain in the structure of chain A of the form II complex, a chain of 14 glucose residues was built into the structure, in addition to the tetrasaccharide inhibitor at the catalytic site (Fig. 7). This polysaccharide model and acarbose correspond to a CA with 18 glucose units. Because TLGT predominantly produces a CA with 18 -20 glucose units, 2 this model seems to represent the most common mode of CA binding in TLGT. When the minimum ring size of CA (16 glucose residues) is produced, some assumption need to be made. From this model, in addition to lid 1, the large distance (ϳ27.5 Å) between subsites Ϫ6 and ϩ3, which is caused by the relatively extended structure of the cleft, coupled with the steric hindrance of the side chains of Phe 19 and Trp 21 also seem to prevent the formation of small CAs in TLGT (Fig. 7). When amylose circularizes, the nonreducing end of a polysaccharide chain must travel a considerable distance to the acceptor site in the active site. It has been proposed that the movement of some hydrophobic residues assists in the circularization of a substrate in CGTase (43). The structural movements observed in TLGT (Fig. 6) may also facilitate the circularization of amylose.