Structural features of a bacterial cyclic α-maltosyl-(1→6)-maltose (CMM) hydrolase critical for CMM recognition and hydrolysis

Cyclic α-maltosyl-(1→6)-maltose (CMM, cyclo-{→6)-α-d-Glcp-(1→4)-α-d-Glcp-(1→6)-α-d-Glcp-(1→4)-α-d-Glcp-(1→})is a cyclic glucotetrasaccharide with alternating α-1,4 and α-1,6 linkages. CMM is composed of two maltose units and is one of the smallest cyclic glucooligosaccharides. Although CMM is resistant to usual amylases, it is efficiently hydrolyzed by CMM hydrolase (CMMase), belonging to subfamily 20 of glycoside hydrolase family 13 (GH13_20). Here, we determined the ligand-free crystal structure of CMMase from the soil-associated bacterium Arthrobacter globiformis and its structures in complex with maltose, panose, and CMM to elucidate the structural basis of substrate recognition by CMMase. The structures disclosed that although the monomer structure consists of three domains commonly adopted by GH13 and other α-amylase–related enzymes, CMMase forms a unique wing-like dimer structure. The complex structure with CMM revealed four specific subsites, namely −3′, −2, −1, and +1′. We also observed that the bound CMM molecule adopts a low-energy conformer compared with the X-ray structure of a single CMM crystal, also determined here. Comparison of the CMMase active site with those in other enzymes of the GH13_20 family revealed that three regions forming the wall of the cleft, denoted PYF (Pro-203/Tyr-204/Phe-205), CS (Cys-163/Ser-164), and Y (Tyr-168), are present only in CMMase and are involved in CMM recognition. Combinations of multiple substitutions in these regions markedly decreased the activity toward CMM, indicating that the specificity for this cyclic tetrasaccharide is supported by the entire shape of the pocket. In summary, our work uncovers the mechanistic basis for the highly specific interactions of CMMase with its substrate CMM.

We have previously identified a novel enzymatically produced cyclic tetrasaccharide, cyclic ␣-maltosyl-(136)-maltose  (29). CMM consists of two maltose molecules cyclized with two ␣-1,6 linkages and thus has alternating ␣-1,4 and ␣-1,6 linkages ( Fig. 1). CMM was initially obtained by reacting the culture supernatant of a soil-isolated bacterium, Arthrobacter globiformis M6, on starch. Further studies showed that the action of a single enzyme contained in the culture supernatant, 6-␣-maltosyltransferase (6MT, EC 3.2.1.-), was responsible for the synthesis of CMM (30). It was also shown that 6MT acts on maltooligosaccharides of a degree of polymerization ϭ 3 or higher and produces CMM by a twostage catalysis of inter-and intramolecular ␣-1,6-transglucosylation. A. globiformis M6 can grow on CMM as the sole carbon source, but no CMM decomposition product was detected in the culture supernatant. A further study revealed that an intracellular enzyme, CMM hydrolase (CMMase, EC 3.2.1.-), plays a key role in the CMM catabolism (31)(32)(33). CMMase degrades CMM into two maltose molecules by a two-stage hydrolysis of the ␣-1,6 linkages via maltosyl-maltose (MM, Fig. 1). CMMase belongs to GH13_20, whereas 6MT is unclassified in the currently established 40 subfamilies of GH13. Interestingly, CMMase is highly specific to CMM and MM, although other GH13_20 enzymes generally do not hydrolyze CMM and exhibit wide substrate specificities toward CD, ␣-1,4, and ␣-1,6 linkages.
Recently, crystal structures of enzymes and proteins related to the metabolism of CNN, which is the other type of cyclic glucotetrasaccharide linked by alternating ␣-1,3 and 1,6 linkages, by Listeria monocytogenes have been reported (34,35). However, the structural basis for the CMM metabolism has not yet been elucidated.
In this study, we determined the crystal structures of CMMase in a ligand-free form and in complex forms with three glucooligosaccharide ligands, including CMM. A comparison of the active site of CMMase with the GH13_20 enzymes revealed significant differences responsible for the substrate specificity. The X-ray structure of a single CMM crystal was also determined, and its conformation was compared with the CMMase-bound CMM molecule. Furthermore, a site-directed mutational analysis of the key residues in the substrate-binding site confirmed their importance in the recognition of CMM.

Overall structure of CMMase
The crystal structure of CMMase was solved by molecular replacement using the GsNPL structure (33.8% sequence identity by EMBOSS Needle Pairwise Alignment, PDB code 1J0H) as a search model. Structures in a ligand-free form and the complex forms with maltose (␣-D-Glcp-(134)-␣-D-Glcp), panose (␣-D-Glcp-(136)-␣-D-Glcp-(134)-␣-D-Glcp), and CMM were determined at 1.6 -2.4 Å resolutions ( Table 1). The monomer enzyme is composed of three domains: a catalytic (␤/␣) 8barrel domain A (residues 1-119 and 175-377); domain B (residues 120 -174), which protrudes from domain A; and domain C (residues 378 -450) (Fig. 2B). The three-domain architecture is generally present in GH13 and ␣-amylase-related clan GH-H enzymes (e.g. GH70 and GH77) (36). In the GH13_20 subfamily, crystal structures of six bacterial and three archaeal enzymes, which have activities of ␣-amylase (EC 3. The molecular masses of CMMase, as deduced from the amino acid sequence, estimated by SDS-PAGE and calibrated by size-exclusion chromatography (in 150 mM NaCl and 20 mM Tris-HCl (pH 8.0)), were 51.6, 53.4, and 105.2 kDa, respectively, suggesting that it is dimeric in solution. In the ligand-free structure and those of two complex forms (maltose and panose), a wing-like dimer structure was observed in the crystal packing in space group P2 1 (two molecules in the asymmetric unit) or C222 1 (one molecule in the asymmetric unit) ( Fig. 2A). The dimer interface comprises two ␤-sheets in domain C. A molecular interface analysis using PISA (44) indicates that the interface area is 1,060 Å 2 , with 19 hydrogen bonds, 11 salt bridges, and an estimated ⌬ i G value of Ϫ3.2 kcal/mol, and it is implied to be the most likely dimer interface among crystal packing interfaces (complexation significance score ϭ 1). However, this dimer structure is not present in the CMM complex structure of the P4 1 2 1 2 space group. This is probably because the CMM complex crystal was grown under alkaline conditions with a high-salt concentration (0.1 M glycine NaOH (pH 9.3), 0.2 M lithium sulfate, and 0.8 M sodium/potassium tartrate), and the salt bridges and the hydrogen bonds in the interface may have been broken during the crystal growth. The two active sites are well-separated in the dimer structure, and the dimerization apparently does not affect the enzymatic function.
To the best of our knowledge, the C domain-mediated winglike dimer structure of CMMase (Fig. 3A) is unique among

Structure and substrate recognition of CMMase
GH13 enzymes, and no similar type of dimer structures have been reported. In GH13_20 enzymes, GsNPL forms a dimer with the characteristic domain N (Fig. 3B) (42), and the dimer interface is substantially involved in the formation of the active site. Similar dimer formation has also been observed in other GH13_20 enzymes containing domain N, such as TVAII and ThMAA (22,41). As shown in Fig. 3C, NpDBE exhibits a unique boat-shaped dimer structure mediated by domains A and B (38). A neighboring molecule is only slightly involved in the formation of the active site, and it has been shown that dissociation into monomers does not affect the activity (45).
A calcium-binding site is observed in the CMM complex structure (Fig. 2B). The same calcium ion was observed in chain A of the ligand-free and panose complex structures but was not found in chain B of these structures or in either chain of the maltose complex. The calcium ion is coordinated by the sidechain oxygens of Asn-23, Asp-29, and Asp-49, the main chain carbonyl of Asp-25 and Gly-47, and one water molecule. The calcium-binding site in CMMase is different from the conserved site in other GH13 subfamilies, e.g. that of GH13_5 ␣-amylase from Bacillus licheniformis (46) but is similar to that of GsNPL (42). The calcium ion of CMMase is not required for activity; we previously reported that EDTA treatment or the addition of calcium ion did not change the activity (31).

Active site of CMMase complexed with substrate and products
In addition to the natural substrate (CMM), a partial structure of the intermediate product/substrate MM (panose) and the final degradation product (maltose) of CMMase ( Fig. 1) were located in the active site of the complex structures. The electron densities of these ligands were clearly observed ( Fig. 2, C-E) near the catalytic residues (Asp-201 as the nucleophile and Glu-230 as the acid/base catalyst), which are conserved in GH13. To avoid the hydrolysis of CMM, a nucleophile residue variant (D201N) was made and used for the co-crystallization with CMM.
In the panose complex structure, two panose molecules are bound at subsites Ϫ3Ј to Ϫ1 (panose 1) and at subsites ϩ1 to ϩ3 (panose 2) (Fig. 4). Notably, the nonreducing end Glc moiety of panose 1 was significantly deviated from the normal subsite Ϫ3 position of GH13_20 enzymes (discussed below); thus, we designated this subsite as Ϫ3Ј. The ␣-1,6 linkages of panose 1 and 2 are located between the subsites Ϫ3Ј and Ϫ2 and subsites ϩ1 and ϩ2, respectively. Therefore, panose 1 mimics the binding of MM, whereas panose 2 appears to represent binding of linear ␣-glucans such as pullulan. The maltose moiety of panose 1 is bound very similarly to the maltose complex structure, but subsite Ϫ3Ј of CMMase is located in a vertical position (Fig. 5B). The subsite Ϫ3Ј is not extensively recognized by the protein, but the side chains of Ser-164 and Cys-163 make a hydrogen bond to the O6 hydroxyl and a hydrophobic interaction, respectively. Panose 2 is loosely recognized compared with panose 1. The side chains of Tyr-204 and Glu-231 form

Structure and substrate recognition of CMMase
direct hydrogen bonds and a stacking interaction at subsites ϩ2 and ϩ3, and several water-mediated hydrogen bonds additionally support the recognition. The reducing end Glc of panose 2 was observed as the ␤-anomer (Fig. 2E), probably due to the loose recognition at this subsite (ϩ3). The observed oxygenoxygen distances between the O1 atom of Glc at subsite Ϫ1 and the O6 atom of Glc at subsite ϩ1 are too close (1.2 Å, dotted line in Fig. 4); thus, we set the occupancies of the two panose molecules to 0.5. This implies that the two Glc moieties mimic an ␣-1,6 linkage of isopanose and pullulan, which can be cleaved by CMMase ( Fig. 1) (31). The O1 and C1 atoms of Glc at subsite Ϫ1 are located near the oxygen atoms of the side chains of the acid/base catalyst (Glu-230, distance ϭ 2.6 Å) and the nucleophile (Asp-201, distance ϭ 2.9 Å), suggesting that the Glc moiety canonically occupies the catalytic subsite.
In the complex structure with CMM, the ␣-1,6 linkage is positioned near the catalytic residues. The distances between the C1 of Glc at subsite Ϫ1 and the side-chain oxygen of

Structure and substrate recognition of CMMase
Asp(Asn)-201 and the scissile glycosidic bond oxygen and that of Glu-230 are 3.1 and 3.2 Å, respectively (Fig. 5C). Because the Glc moiety connecting the sugars in subsites Ϫ3Ј and Ϫ1 is again significantly deviated from the normal subsite ϩ1 position for linear glucan substrates, we designated this subsite as ϩ1Ј (Fig. 4). The Glc at subsite ϩ1Ј forms hydrogen bonds with Asp-297 and several water molecules. The binding mode of CMM to subsites Ϫ3Ј to Ϫ1 is similar to that of panose 1. At subsite Ϫ3Ј, there are also interactions with Cys-163 and Ser-164. The maltose moiety at subsites Ϫ2 and Ϫ1 are extensively recognized with similar interactions as those of the maltose complex, as described above.
In summary, for the structural features of the active site, CMMase binds a Glc moiety at subsite ϩ1 by placing the O6 hydroxyl near the acid/base catalyst (cleavage preference for an ␣-1,6 bond). In addition, CMMase has strong subsites for the maltose moiety at subsites Ϫ2 and Ϫ1, whereas the flanking subsites (Ϫ3Ј and ϩ1Ј/ϩ1) appear to be relatively weak. The extensive interactions at subsites Ϫ2 and Ϫ1 suggest that only ␣-1,4-linked Glc moieties are allowed to bind here. These structural features are consistent with the substrate preference for CMMase (31), with hydrolysis ratios of 99.4% for CMM, 76.2% for MM, 34.6% for pullulan, 28.7% for isopanose, 4.2% for maltotetraose (G4), and 1.8% for maltopentaose. Furthermore, we have measured the kinetic parameters of CMMase to CMM and MM. CMMase exhibited typical Michaelis-Menten-type kinetics (hyperbolic saturation curve) to the both substrates (data not shown). The K m , k cat , and k cat /K m values for CMM were 3.6 Ϯ 0.2 mM, 59.7 Ϯ 2.9 s Ϫ1 , and 17.8 Ϯ 1.4 s Ϫ1 mM Ϫ1 , respectively ( Table 2). The K m , k cat , and k cat /K m values for MM were 51 Ϯ 14 mM, 250 Ϯ 39 s Ϫ1 , and 5.0 Ϯ 2.8 s Ϫ1 mM Ϫ1 , respectively. Therefore, CMMase preferentially binds CMM for hydrolysis compared with MM. From a structural comparison of the ligand-free and complex structures, it was found that CMMase did not significantly change its structure upon substrate (ligand) binding, as the RMSD values for the C␣ atoms

Structure and substrate recognition of CMMase
between all pairs are within 0.19 -0.47 Å. A significant movement was only observed for Arg-233 in the panose complex (Fig. 5B), whose side chain moves into the subsite ϩ3 when it is occupied.
We also examined inhibition behavior and potency of maltose and panose, whose binding interactions with CMMase were revealed by the crystallography. Both of the compounds exhibited typical competitive inhibition with CMM substrate (data not shown), and the K i values for maltose and panose were 4.8 Ϯ 0.3 and 2.8 Ϯ 0.4 mM, respectively. The K i values were comparable with the K m value for CMM (3.6 mM). The relatively small K i value of panose is consistent with the more extensive interactions of the panose 1 molecule in subsites Ϫ3Ј to Ϫ1 compared with maltose in subsite Ϫ1 and Ϫ2 (Fig. 5, A  and B). Fig. 6 shows a comparison of the active sites of the two representative GH13_20 enzymes, CMMase and GsNPL. The

Structure and substrate recognition of CMMase
complex structures of CMMase with CMM, panose, and maltose and those of GsNPL with isopanose and maltotetraose (G4) are shown by superimposing the ligands. As expected by the moderate sequence homology, these enzymes have similar overall structures, as described above. Accordingly, the activesite residues are basically conserved, especially for those depicted on the left side and the centerline in Fig. 6, B and D. In particular, residues involved in the strong interactions at subsites Ϫ2 and Ϫ1 are highly conserved. However, residues depicted on the right side in Fig. 6, B and D (circled by dotted lines), are not conserved, and thus significant differences at other subsites (Ϫ3/Ϫ3Ј, ϩ1, ϩ2, and ϩ3) are found. As shown in Fig. 6, A and C, by surface models with hydrophobicity from white to red, the active-site cleft of GsNPL is elongated and boat-shaped, with a hydrophobic protrusion above subsite Ϫ1, whereas that of CMMase is bowl-shaped with a relatively hydrophobic platform at subsites ϩ2 and ϩ3.
The residues involved in formation of the characteristic walls of the cleft of each enzyme can be grouped as follows: PYF (Pro-203, Tyr-204, and Phe-205), CS (Cys-163 and Ser-164), and Y (Tyr-168) in CMMase, and ANE (Ala-330, Asn-331, and Glu-332), FA (Phe-289 and Ala-290), and Q (Gln-294) in GsNPL (Fig. 6, B and D). The cleft of GsNPL is relatively deeper at subsites ϩ1 and ϩ2, whereas that of CMMase is shallower at subsites ϩ2 and ϩ3. Moreover, the cleft of GsNPL is narrower at subsites ϩ1 and Ϫ1 compared with that of CMMase. The changes at the PYF/ANE and CS/FA walls are responsible for these differences. The recognition of Glc at subsite ϩ1 in GsNPL is supported by hydrophobic interactions with Phe-289, Val-329, Trp-359, and Tyr-45* (Fig. 6D, asterisk indicates that this residue is from the neighboring monomer) (42), and this subsite appears to be relatively stronger than that of CMMase. In particular, the unique subsite Ϫ3Ј of CMMase is formed by the CS wall. Subsite Ϫ3 (and also possible subsite Ϫ4) of GsNPL is open to the solvent, suggesting that it prefers linear glucan substrates. In contrast, the cleft of CMMase is blocked at the corresponding position by the side chain of Tyr-168 (Fig. 6B), and the subsite Ϫ3Ј is located at an elevated position. Therefore, the difference at subsite Ϫ3/Ϫ3Ј appears to be caused by the amino acid change of Q/Y.
Amino acid sequence alignment revealed that, among the GH13_20 enzymes, the key residues of the three walls are unique to CMMase (Fig. 7). The ANE and FA motifs in GsNPL are conserved in other GH13_20 enzymes that exhibit neopul-lulanase-like activity (TVA II and ThMAA), but the corresponding residues of Gln-294 vary (Ala or His).

Identification of important residues for activity by site-directed mutational analysis
To investigate the importance of these residues for the activity of CMMase, we constructed 10 variant enzymes containing single to quadratic site-directed mutations at the three wall positions by mimicking the active site of GsNPL with substitution of PYF, CS, and Y for ANE, FA, and Q, or by substituting them with Ala. Their kinetic parameters for CMM (Table 2) were compared with the WT enzyme. A single Ala substitution in the PYF wall (Y204A) caused a concomitant increase of the k cat and K m values for CMM, resulting in slight increase of the k cat /K m value. A single Ala substitution at the Y wall (Y168A) also caused increases in both the k cat and K m values, but the significant increase of the K m value had a large impact on the k cat /K m value (7.4-fold decrease compared with the WT). The triple substitutions from PYF to ANE (P203A/Y204N/ F205E) and the combinational double mutation of Tyr-204 and Tyr-168 (Y168Q/Y204N) also significantly decreased the k cat /K m value due to increases of the K m value. The quadratic variant from PYF-Y to ANE-Q (Y168Q/P203A/Y204N/F205E) exhibited the highest K m and the lowest k cat /K m values among all variants tested. These results indicate that the PYF and Y walls at the Ϫ2 and plus subsites concertedly support formation of the bowl-shaped cleft of CMMase that is suitable for the small cyclic substrate. Substitutions at the CS wall also significantly increased the K m value for CMM. The effect of the substitution at Ser-164 (S164A) was milder than those at Cys-163 (C163A, C163V, C163L, and C163S). Among the four Cys-163 variants that we tested, C163V and C163S showed the highest and lowest k cat /K m values, respectively. Therefore, we conclude that the small side chain of Cys-163 mainly contributes to the substrate affinity at the unique Ϫ3Ј subsite of CMMase with its hydrophobicity.
We also measured activities (hydrolysis ratio) of the variant enzymes toward various substrates and compared them with those of the WT enzyme (Table 3). Notably, the Y204A and P203A/Y204N/F205E variants exhibited a significantly increased activity toward G4, suggesting that elimination of the large side chain of Tyr-204 at subsite ϩ2 shifted the specificity of CMMase from ␣-1,6 linkage toward the linear ␣-1,4-linked substrate. This result is consistent with the conservation pattern of the residue corresponding to Tyr-204 of the PYF motif (Fig. 7). Enzymes with neopullulanase-like activity (GsNPL, TVAII, and ThMAA), which have intermediate specificities to ␣-1,4 and ␣-1,6 linkages, have Asn at this position, although the ␣-1,6-specific debranching enzyme (NpDBE) has an aromatic residue (Phe-213). In the study of TVAII, substitution of Val-326 (the residue corresponding to Val-202) could modulate the preference for ␣-1,6 and ␣-1,4 linkages (47). The V326A variant of TVAII favored the ␣-1,4 linkage, although V326I favored the ␣-1,6 linkage, suggesting that size and hydrophobicity of the residue at this position modulate the linkage preference. As shown in Fig. 5, Val-202 of CMMase is located close to the PYF wall.

Table 2 Kinetic parameters of the wild-type CMMase and variants for CMM
The enzymatic reaction was performed in 50 mM sodium acetate buffer (pH 6.0) at 25°C, and reducing power was measured.

Structure and substrate recognition of CMMase
The substitution at Tyr-168 (Y168A, Y168Q/Y204N, and Y168Q/P203A/Y204N/F205E) significantly decreased the activity toward pullulan and isopanose. Moreover, the substitutions at the CS wall generally decreased the activities toward substrates other than CMM. In particular, the Cys-163 substitutions showed significantly decreased activities toward MM. These results indicate that changes at the CS and Y walls, which constitute the essential part of the pocket of CMMase, are also destructive for binding of the linear substrates.

Crystallography of CMM
Previously, we reported the preparation process and characteristics of pentahydrate crystals of CMM (29). Here, we deter-mined the X-ray crystal structure of a single CMM crystal (Table 4 and supporting information). Fig. 8 shows superimposition of the structure of CMM in the single crystal (Fig. 8, cyan) with the CMM molecule bound to CMMase (white). The two CMM molecules unexpectedly exhibit large conformational differences. We also modeled an energy-minimized CMM molecule in the active site (Fig. 8, thin orange sticks), and we found that it adopts a similar conformation as the CMMase-bound CMM molecule. The CMM in the single crystal shows the largest deviations from the other two at the Glc unit at subsite Ϫ3Ј, swinging to the left side in Fig. 8. The conformational differences arise from the two ␣-1,6 bonds (b1 between Ϫ3Ј and Ϫ2 and b3 between Ϫ1 and ϩ1Ј) and the two ␣-1,4 bonds (b2

Structure and substrate recognition of CMMase
between Ϫ2 and Ϫ1 and b4 between ϩ1Ј and Ϫ3; see Fig. 4 for the designation). Therefore, we measured the torsion angles of the four glycosidic bonds of the three CMM structures (Table  5). Dowd et al. (48) calculated isoenergy surfaces of the ␣-1,6 bond of isomaltose based on the MM3 force field. The conformations of the b1 and b3 bonds in the CMM single crystal are placed near the second lowest local minimum of the energy map of ␣-isomaltose ( ϭ Ϫ44.3°, ϭ Ϫ174.6°, and ϭ Ϫ50.5°), and those in CMM complexed with CMMase are close to the third minimum ( ϭ Ϫ43.5°, ϭ 179.7°, and ϭ 60.6°). Therefore, the two ␣-1,6 bonds are not high-energy conformers even though they adopt distinct conformations ( ϳ Ϫ58°in a single crystal and ϳ ϩ40°in complex in CMMase). For the ␣-1,4 bond of the ␣-maltose unit, the GlycoMapsDB web tool was used to map the torsion angles with 2,512 PDB entries (black for 1,988 disaccharide fragments and red for 524 exact structures) onto a calculated energy map (Fig. 9) (49). The b2 and b4 bonds in CMMase were mapped in the most frequently observed low-energy conformation area, although those in the single CMM crystal were mapped in a rarely observed highenergy conformer area, whose calculated energy is more than 5 kcal/mol higher than that of the global minimum. Therefore,  Table 3 Substrate specificity of the wild-type CMMase and variants toward various substrates (10 mM oligosaccharides or 1% pullulan) The enzymatic reaction was performed in 50 mM sodium acetate buffer (pH 6.0) with 0.1 mg/ml enzyme at 25°C for 18 h. The hydrolysis was monitored by TLC and classified by the following marks: Ϫ, not hydrolyzed; ϩ, Ϲ10% hydrolyzed; ϩϩ, 10 -50% hydrolyzed; and ϩϩϩ м50% hydrolyzed.

Structure and substrate recognition of CMMase
the single-crystal structure of CMM may be a relatively highenergy conformer trapped during the crystal formation, and the CMM molecule bound in CMMase represents one of the lowest energy conformations in solution. Fig. 10 illustrates our proposed reaction steps of the twostage hydrolysis of CMM to maltose by CMMase, according to the results of this study: step 1, CMM binds to the active site; step 2, CMM is cleaved at one of the two ␣-1,6 linkages, and the reaction product, MM, is tentatively released from the active site. For step 3, MM rebinds so that the remaining ␣-1,6 linkage is placed near the catalytic residue, and for step 4, the second ␣-1,6 linkage is hydrolyzed, and the products (two molecules of maltose) are released from the active site. The Ϫ2 and Ϫ1 subsites strongly recognize the maltose molecule and play a critical role in the substrate binding. In addition, interaction at Ϫ3Ј subsite, which is mainly supported by the hydrophobic interaction by Cys-163, fixes the cyclic molecule. The PYF and Y walls also support formation of the bowl-shaped cleft, which is complementary for CMM. Comparison between the apo and complex structures suggested that the active site of CMMase does not have an induced-fit-type feature. Our kinetic analysis indi-  and ␣-maltose-containing glycans (black) in the PDB are plotted on a calculated conformational free energy map (coded green to red for low-to highenergy conformations). The torsion angles of CMM molecules in the single crystal (blue) and in the CMMase complex structure (magenta) are also plotted. The b2 and b4 bonds of CMM are designated in Fig. 4.   Fig. 4. Definitions of the torsion angles are as follows: H1Ј-C1Ј-O4 -C4 for , C1Ј-O4 -C4 -H4 for , and O1-C6Ј-C5Ј-H5Ј for (␣-1,6 bond).

Structure and substrate recognition of CMMase
cated that the enzyme preferentially binds CMM for hydrolysis compared with MM. This is probably due to the rigid and cyclic structure of the CMM substrate, whose low-energy conformer fits into the static active site. Among the several known cyclic oligosaccharides, the metabolic pathways of CDs have been extensively studied for various bacteria and archaea, such as Klebsiella oxytoca, Thermococcus sp., Bacillus subtilis, and T. vulgaris (50 -53). The microbial CD metabolism generally consists of three stages: synthesis of the cyclic oligosaccharides (CDs) outside the cells, uptake into the cells via ATP-binding cassette transporters, and degradation by intracellular enzyme(s). It has been suggested that this type of metabolic pathway is advantageous in the competition of carbon source acquisition by transiently changing the molecular form of a digestible glucan (starch) into a special cyclic form (CDs), which is rarely assimilated by other microorganisms. A similar three-stage metabolic pathway for CNN has been studied for L. monocytogenes (35). The CNN metabolic pathway is coded in two operons with 10 genes in total (lmo2446--lmo2444 and lmo0178--lmo0184). The functional and structural study revealed that the CNN metabolic pathway involves at least two extracellular enzymes for CNN formation, a specific ABC transporter, several intracellular enzymes, and a translational regulation system of a ROK family protein. In the CNN metabolic pathway, the crystal structure of GH31 Trueperella pyogenes cycloalternan-degrading enzyme (TpCADE), which is a homolog of the intracellular CNN-hydrolyzing enzyme of L. monocytogenes (Lmo0182), has been reported (34,35). Fig. 11 shows a comparison of the active sites of CMMase and TpCADE by aligning the structures with the catalytic components for ␣-glycosidic bond hydrolysis. However, the two enzymes having the hydrolytic activity toward cyclic glucotetrasaccharides adopt completely different substrate-recognition architectures. TpCADE has the cleavage specify toward the ␣-1,3 bond of CNN and recognizes the substrate by a stacking interaction ranging from subsites Ϫ1 to ϩ1, and two direct hydrogen bonds to sugars in subsites Ϫ1 and Ϫ2.
The metabolic pathway of A. globiformis M6 for CMM also consists of three-stage components (29 -31, 33), but it is relatively simpler than that for CNN. The CMM metabolic pathway consists of a single cluster containing seven genes (cmmA-G). Among them, an extracellular CMM-forming enzyme, 6MT (CmmA), and an intracellular CMM-degrading enzyme, CMMase (CmmF), have been characterized. The cmmB gene encodes an intracellular GH13_30 ␣-glucosidase (33). The ␣-glucosidase exhibits high activity toward MM, panose, and maltose, and it has been suggested that CMM is degraded to glucose by the synergistic action of CMMase and the ␣-glucosidase CmmB (31). The cmmC, -D, and -E genes show 27, 44, and 49% sequence identity (by BLAST) to a solute-binding protein from Streptomyces avermitilis, a putative permease from Deinococcus geothermalis, and a putative permease from Bacillus clausii of the ABC sugar transport system, respectively. Therefore, CmmC, CmmD, and CmmE likely form an ABC importer system specific for CMM. The cmmG gene encodes a putative LacI/PurR family transcriptional regulator.
Our study revealed that the shape and interactions in the active site of CMMase are highly specific for the CMM substrate, and we identified several critical residues for the recognition. CMMase might have emerged through molecular evolution from a GH13_20 family enzyme, which has a wide substrate specificity similar to that of neopullulanase, under pressure from the "selfish" metabolic pathway for the cyclic oligosaccharide, in combination with the molecular evolution of the CMM-forming enzyme, 6MT. To date, the CMMase activity has not been confirmed for enzymes other than CMMase from A. globiformis M6, even though many GH13 enzymes show a certain sequence identity (Ͼ40% by BLAST). For example, neopullulanase-like enzymes, cyclomaltodextrinases, general type ␣-amylases, dextranases (EC 3.2.1.11), isoamylases (EC 3.2.1.63), pullulanases, and debranching enzymes cannot hydrolyze CMM. To investigate putative CMMases from the gene database, we performed a protein BLAST search.  Table 6 shows the amino acid sequence alignment of the top 10 homologous proteins from different organisms. Most of the bacterial species belong to the Actinobacteria class, except for Chlamydia. These putative proteins exhibit relatively higher sequence identities (Ͼ55%) to CMMase than to the neopullulanase-like enzymes, and the critical residues forming the PYF, CS, and Y walls are basically conserved. Therefore, we assume that these putative homologs also have CMMase activity and that their source organisms likely have a threestage metabolic pathway for CMM, similar to that of A. globiformis M6. Further studies on 6MT and the putative ABC transporter (CmmC solute-binding protein) will identify key residues for their functions and clarify the whole picture of the three-stage metabolic pathway of CMM, which may be prevalent in Actinobacteria.

Protein preparation
The CMMase-encoding gene was amplified from pRSET-CMMase (33) by PCR to express as an N-terminally His-tagged (His 6 ) protein, and inserted between the NdeI and BamHI sites of pET-28b(ϩ) vector (Novagen, Madison, WI). The following primers were used: 5Ј-GGAATTCCATATGACCGCTCCC-GACTGG-3Ј; 5Ј-CGCGGATCCTTACGCAGAGCTCCC-GGG-3Ј (restriction enzyme sites are underlined). This plasmid is designated pET28b_CMMase. Escherichia coli Rosetta2 (DE3) (Novagen) transformed by this plasmid was cultured in Luria-Bertani medium containing antibiotics (50 mg/liter kanamycin and 34 mg/liter chloramphenicol) at 37°C until A 600 nm ϭ 0.6. To induce protein expression of the transformant, 0.1 mM (final concentration) of isopropyl 1-thio-␤-D-galactopyranoside (FUJIFILM Wako Pure Chemical Co., Osaka, Japan) was added to the medium. The medium was then continuously cultured at 15°C for 24 h. The cultured cells were harvested by centrifugation at 8,000 ϫ g for 15 min and suspended in 50 mM Tris-HCl (pH 8.0) and 500 mM NaCl. To obtain cell-free extracts, the suspended solution was sonicated and centrifuged at 18,000 ϫ g for 45 min. First, CMMase was purified by nickel affinity column chromatography using cOmplete His-Tag Purification Resin (Sigma). CMMase was then further purified by column chromatogra-phy with a Mono Q 10/100 GL and a Superdex 200 pg 16/60 using the ÄKTA system (GE Healthcare, Buckinghamshire, UK). The concentration of CMMase was determined by a NanoDrop ND-1000 spectrophotometer (ThermoFisher Scientific, Waltham, MA) using the extinction coefficient ⑀ 280 nm ϭ 90,090 M Ϫ1 cm Ϫ1 , which was estimated from the amino acid sequence.

Structure and substrate recognition of CMMase Protein crystallography of CMMase
Protein crystallization was performed by the sitting drop vapor diffusion method at 20°C. WT crystals were obtained by mixing 0.5 l of protein solution containing 20 mg/ml CMMase in 20 mM Tris-HCl (pH 8.0), and an equal volume of reservoir solution containing 0.1 M sodium citrate (pH 5.6), 0.22 M ammonium sulfate, and 30% (w/v) PEG4000. The crystals were cryoprotected in the reservoir solutions supplemented with 30% (w/v) PEG400 or 30% maltose (for maltose complex). To obtain the complex structure with panose, 50 mM panose was added to the cryoprotectant solution. D201N-CMM complex crystals were obtained by co-crystallization by mixing 0.5 l of protein solution containing 13 mg/ml CMMase in 20 mM Tris-HCl (pH 8.0) and 5 mM CMM with an equal volume of reservoir solution containing 0.1 M glycine-NaOH (pH 9.3), 0.2 M lithium sulfate, and 0.8 M sodium/potassium tartrate. The D201N-CMM complex crystals were cryoprotected in the reservoir solution supplemented with 30% (w/v) PEG400 and 10 mM CMM. X-ray diffraction data were collected at BL26B1 at SPring-8 (Hyogo, Japan) and at BL1A, BL17A, and NW12A at the Photon Factory of the High Energy Accelerator Research Organization (KEK-PF, Tsukuba, Japan). The diffraction images were processed with HKL2000 (54). Molecular replacement was performed with MOLREP (55). The model was further built manually with COOT (56) and refined with REFMAC5 (57). Molecular graphic images were prepared using PyMOL (Schrödinger LLC, New York).

Crystallography of CMM
Crystals of CMM were prepared as described previously (29). Details of the X-ray data collection, data reduction, structure solution, and crystallographic refinement are described in Table 4 and the supporting information.

Enzyme assay and kinetic analysis
For the kinetic analysis using CMM as substrate, the enzymatic reaction was initiated by mixing enzyme solution (40 l), containing 2-10 g/ml CMMase (or its variant), in 50 mM sodium acetate (pH 6.0) and substrate solution (160 l), containing 0.625-62.5 mM CMM in 50 mM sodium acetate (pH 6.0). The mixture was incubated at 25°C for 6 min. After stopping the reaction by heat treatment at 100°C for 10 min, the reducing power of each mixture was measured by the bicinchoninic acid assay according to the method of Utsumi et al. (58). For the kinetic analysis for MM, substrate solutions containing 1.25-75 mM MM were used. After stopping the reaction by heat treatment at 100°C for 10 min, a decreased amount of MM was measured by the HPLC system (Shimadzu Corp., Kyoto, Japan) by using two YMC-Triart C18 columns (YMC Co., Ltd., Kyoto, Japan) in tandem. The kinetic parameters and their fitting errors were calculated by nonlinear fitting of the experimental data to the Michaelis-Menten equation using KaleidaGraph 3.6 (Synergy Software, Reading, PA).
For the inhibition kinetics assay for maltose and panose, reaction mixtures containing 1, 4, or 12 mM CMM and 0, 5, 10, or 30 mM inhibitor (maltose or panose) at final concentrations were used, and an increased amount of MM after 6 min of incubation at 25°C was measured by HPLC. The kinetic parameters were calculated by nonlinear global curve-fitting the experimental data to the theoretical equation for competitive inhibition using R 3.51 (R Foundation for Statistical Computing, Vienna, Austria).
For the substrate specificity assay, the enzymatic reaction was initiated by mixing enzyme solution (1 l), containing 1 mg/ml CMMase (or its variant), in 20 mM Tris-HCl (pH 8.0), and substrate solution (9 l), containing 10.0 mM CMM, 10.0 mM MM, 10.0 mM maltotetraose, 10.0 mM isopanose, or 1.00% pullulan, in 50 mM sodium acetate (pH 6.0). The mixture was incubated at 25°C for 18 h. After stopping the reaction by heat treatment at 100°C for 10 min, 2 l of the mixture was spotted onto a TLC Silica Gel 60 F 254 plate (Merck, Darmstadt, Germany). Subsequently, the TLC plate was developed by a solution consisting of 55% (v/v) 1-butanol, 36% (v/v) pyridine, and 9% (v/v) water. Sugar spots were detected by spraying 20% (v/v) sulfuric acid/methanol and heating in an oven for several minutes. The contrast of the spot was then visually judged.