Structural and Functional Basis for Substrate Specificity and Catalysis of Levan Fructotransferase*

Background: Levan fructotransferase converts polymeric β-2,6-linked levan into cyclic difructose dianhydrides. Results: The active site accommodates difructosaccharide in a catalytic β-propeller fold. Conclusion: The shape of the active site pocket dictates substrate specificity and formation of cyclic difructose dianhydrides. Significance: Sequence conservation in the loop region among levan fructotransferases is the molecular basis for the exo-type cleavage of difructosaccharide and product formation. Levan is β-2,6-linked polymeric fructose and serves as reserve carbohydrate in some plants and microorganisms. Mobilization of fructose is usually mediated by enzymes such as glycoside hydrolase (GH), typically releasing a monosaccharide as a product. The enzyme levan fructotransferase (LFTase) of the GH32 family catalyzes an intramolecular fructosyl transfer reaction and results in production of cyclic difructose dianhydride, thus exhibiting a novel substrate specificity. The mechanism by which LFTase carries out these functions via the structural fold conserved in the GH32 family is unknown. Here, we report the crystal structure of LFTase from Arthrobacter ureafaciens in apo form, as well as in complexes with sucrose and levanbiose, a difructosacchride with a β-2,6-glycosidic linkage. Despite the similarity of its two-domain structure to members of the GH32 family, LFTase contains an active site that accommodates a difructosaccharide using the −1 and −2 subsites. This feature is unique among GH32 proteins and is facilitated by small side chain residues in the loop region of a catalytic β-propeller N-domain, which is conserved in the LFTase family. An additional oligosaccharide-binding site was also characterized in the β-sandwich C-domain, supporting its role in carbohydrate recognition. Together with functional analysis, our data provide a molecular basis for the catalytic mechanism of LFTase and suggest functional variations from other GH32 family proteins, notwithstanding the conserved structural elements.

Fructans are polymeric carbohydrates of fructose linked via O-glycosidic bonds, with a terminal sucrose. The ␤-2,1-linked inulin and ␤-2,6-linked levan are the two major types of fructans identified to date, and they represent one of the primary carbohydrate storage molecules in some plants, bacteria, and fungi (1). In addition to their role as an energy reserve, plant fructans also function in cold and drought tolerance (1)(2)(3). Their biosynthetic pathways in bacteria are relatively well established; in these pathways, a single multifunctional enzyme of levansucrase (4,5) and inulosucrase (6) is responsible for the synthesis of levan and inulin, respectively, with sucrose as a substrate. However, those pathways have not yet been conclusive in plants (7). Once the fructosyl moiety is reserved in fructans, its mobilization is achieved by various hydrolytic enzymes (7). For example, plant fructan exohydrolases successively remove one terminal fructose unit from inulin and levan through an exo-type reaction, although in bacteria levanase and inulinase perform both endo-and exo-type reactions toward these fructans.
Although the fructan-utilizing enzymes exhibit diverse substrate specificities and catalytic properties, these enzymes-including LFTase-are either O-glycoside hydrolase (GH) 32 or GH68 family enzymes, based on sequence similarity classifications (CAZy) (15). More specifically, most of these enzymes are members of the GH32 family; however, bacterial levansucrase and inulosucrase are in the GH68 family. Despite their limited sequence homology, these two families exhibit high similarities in their three-dimensional structures and belong to the GH-J clan by a structure-based classification (16). In particular, a fivebladed ␤-propeller fold that acts as a catalytic domain is common to the GH-J clan, whereas an additional ␤-sandwich domain has been characterized in the GH32 family C-terminal region. The only exception to these general features is with IFTase, which is a member of the GH91 family and adopts a homotrimeric ␤-helical fold with its active site located in the intersubunit interface (11). Notably, the proposed catalytic residues and their mechanistic features account for characteristics of IFTase that differ from those of the GH32 and GH68 families.
Unlike IFTase, the DFA-IV-producing LFTase from Arthrobacter ureafaciens K-2032 (AuLFTase) is a member of the GH32 family (12) (supplemental Fig. S1). Therefore, how the unusual functional features of AuLFTase can be accommodated within the structural fold conserved in the GH-J clan remains to be resolved. We determined the crystal structure of AuLFTase and conducted a functional analysis to understand its structural and functional features. The structure of AuLFTase was determined in apo form at a resolution of 2.57 Å, as well as in two different binary complexes, one with sucrose at a resolution of 2.2 Å and the other at a resolution of 2.3 Å with levanbiose (O-␤-D-fructofuranosyl-(2,6)␤-D-fructofuranoside), a difructosaccharide with a ␤-2,6-glycosidic bond (supplemental Fig. S2). Our analyses provide the molecular basis for substrate specificity and the exolytic cleavage of levan, as well as structural similarities to and differences from members of the GH32 family. These results serve as a framework for understanding DFA-IV production by other LFTases.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification-The gene for AuLFTase (12) (GenBank TM accession number AF181254) was amplified by PCR with sequence-specific primers (supplemental Table S1). The resulting PCR product contained residues Ala-41 to Ile-521 by excluding the N-terminal 40 residues as a signal peptide, Met-1 to Ala-33 (supplemental Fig. S1), and was subcloned into the pET28b expression vector (Merck) with the C-terminal noncleavable His tag.
Seleno-L-methionine-substituted, C-terminal His-tagged AuLFTase (residues Ala-41 to Ile-521) was expressed in Escherichia coli B834 (DE3) methionine auxotroph cells (Merck) for structural studies. In particular, we found that methylation of AuLFTase is essential for crystallization. Therefore, a methyla-tion step was included in the purification, according to a protocol published previously (17,18). Cells were grown at 37°C in minimal medium supplemented with seleno-L-methionine to an A 600 of 0.7 and then induced for 16 h at 20°C by the addition of 0.5 mM isopropyl-␤-D-thiogalactopyranoside. The cells were harvested and sonicated in buffer A (50 mM HEPES, pH 7.6, 150 mM NaCl). The enzyme was purified using immobilized metal affinity chromatography with buffer A plus 500 mM imidazole. After the purified AuLFTase was dialyzed against buffer A, the methylation reaction using the dimethylamine-borane complex, and formaldehyde was performed overnight at 4°C with 1-2 mg/ml of AuLFTase. After the soluble fraction from the reaction mixture had been obtained by centrifugation, size exclusion chromatography using a Superdex 200 column (GE Healthcare) was carried out with buffer A to remove any possible aggregated form of the protein. Elution fractions for AuLFTase MET (hereafter, a superscript of MET denotes the methylated enzyme) were pooled, resulting in a yield of 15 mg/ml, based on its molar extinction coefficient of 151,260 M Ϫ1 cm Ϫ1 at 280 nm. AuLFTase MET was catalytically competent (supplemental Fig. S3A), and four of the five lysine residues were estimated to be methylated at the ⑀-amino group of lysine, based on mass spectroscopy-time of flight analysis.
For structural analysis of AuLFTase MET in complex with either sucrose or levanbiose, we used a mutant AuLFTase-(D54N) MET enzyme, in which the catalytic nucleophile Asp-54, common to members of the GH32 family (supplemental Fig.  S1), was mutated into asparagine to exclude any possible involvement of Asp-54 in catalysis of the ligand used in this experiment. Site-directed mutagenesis was performed using a QuikChange (Agilent Technologies), with mutagenic primers (supplemental Table S1), and the resulting plasmid that encodes the AuLFTase(D54N) enzyme was then transformed into the BL21(DE3) codon plus-RIL strain (Agilent Technologies) and expressed in Luria Bertani medium. The purification steps for AuLFTase(D54N) MET were identical to the procedure described above, with the exception of addition of 1 mM dithiothreitol. We also confirmed that mutant AuLFTase(D54N) was catalytically inactive (supplemental Fig. S3A). During protein purification, we noticed that both the native and methylated enzyme were monomeric, based on their size exclusion chromatography elution profiles.
Crystallization-The apo form of seleno-L-methionine-substituted AuLFTase MET was crystallized at 22°C using the hanging drop vapor diffusion method in a crystallization buffer of 0.1 M Bis-Tris (pH 5.5), 1% (w/v) PEG 3350, 1 M ammonium sulfate, and 3% 2-propanol. Unlike the WT enzyme, crystals for the apo form of the AuLFTase(D54N) MET mutant enzyme were produced at 22°C in different crystallization buffers using the hanging drop vapor diffusion method; the crystallization buffer consisted of 0.1 M MES (pH 6.5) and 1 M lithium sulfate. Formation of the AuLFTase(D54N) MET -ligand complex, either with sucrose or levanbiose, was achieved by a soaking procedure. Sucrose is not a substrate of AuLFTase (supplemental Fig.  S3A), but the enzyme structure in a complex with sucrose reveals crucial structural features of AuLFTase substrate binding.
For the AuLFTase(D54N) MET -sucrose complex, a pregrown crystal of AuLFTase(D54N) MET was soaked for 60 min in a solution of 1 M sucrose in 0.1 M MES (pH 6.5), 1 M lithium sulfate, and 5% (v/v) glycerol. The AuLFTase(D54N) MET -levanbiose complex was obtained by soaking a crystal for 60 min in 30% of the levanbiose reaction product (supplemental "Methods") in 0.1 M MES (pH 6.5), 1 M lithium sulfate, and 10% (v/v) glycerol. In both experiments, glycerol was included as a cryoprotectant. We also determined the crystal structure of AuLFTase MET in complex with the DFA-IV product. Specifically, DFA-IV was produced by reacting 3% (w/v) levan (average molecular weight of 50,000 from Realbiotech, Chungnam, Korea) in buffer A with 0.28 mM AuLFTase MET . The reaction mixture was incubated at 30°C for 3 h, and the entire reaction mixture, including AuLFTase MET , was crystallized under conditions identical to those for AuLFTase(D54N) MET . Subsequent structural analyses, as described below, indicated that DFA-IV is bound to the C-terminal domain of AuLFTase MET , not to the catalytic domain (supplemental Fig. S3B).
Data Collection and Structure Determination-Multiwavelength anomalous dispersion data using a crystal of the apo form of seleno-L-methionine-substituted AuLFTase MET were collected at a resolution of 2.57 Å on Beamline 4A and 6C at the Pohang Accelerator Laboratory (Pohang, Korea). Subsequently, single-wavelength diffraction data were collected on Beamline 17A at the Photon Factory (Tsukuba, Japan), for the AuLFTase(D54N) MET -sucrose complex at a resolution of 2.2 Å and the AuLFTase(D54N) MET -levanbiose complex at a resolution of 2.3 Å. We collected all data at 100 K and processed them using the program HKL2000 (19). All of the crystals belonged to the space group of P2 1 2 1 2 1 , with four monomers per asymmetric unit (supplemental Table S2).
The structure of AuLFTase MET in apo form was determined using the programs SOLVE and RESOLVE (20,21). Manual model building and refinement were performed using the programs COOT (22) and CNS (23). Of 24 possible selenium atoms, 20 sites were identified and used to calculate the initial phases. The figure of merit was 0.71 by density modification with RESOLVE. After several iterative cycles of manual inspection and refinement, the model was built including Ala-41 to Gln-519 for all four monomers. The refined apo form model was then used as an initial structure to solve the structures of the binary complexes. The CNS program was used to perform a rigid body refinement followed by simulated annealing for the AuLFTase(D54N) MET -sucrose complex. The resulting electron density map showed one molecule of sucrose bound in the pocket of a ␤-propeller fold of monomers A and B, whereas the corresponding density was highly disordered in the other two monomers. Unexpectedly, we also found one molecule of sucrose bound in the C-terminal ␤-sandwich domain of all four monomers (supplemental Fig. S4A). An almost identical refinement strategy was used for the structural determination of the AuLFTase(D54N) MET -levanbiose complex using the PHENIX program (24). Analogous to the AuLFTase(D54N) MET -sucrose complex, one molecule of levanbiose was clearly identified in the pocket of the ␤-propeller fold in monomer C, as well as in the C-terminal ␤-sandwich domain of monomers A, C, and D (supplemental Fig. S4B). Furthermore, continuous electron density for levan was observed at the C-terminal ␤-sandwich domain of monomer B, but the ordered density was for levantriose (supplemental Fig. S4C). Interestingly, we also characterized the binding of levanbiose on the surface of the ␤-propeller fold in the AuLFTase(D54N) MET -levanbiose complex for all four monomers. However, this binding seems to be nonspecific and unrelated with enzyme function, and details of those interactions are not described in this study.
At the final refinement stage, water molecules, whose refined temperature factors are less than 50 Å 2 , were assigned based on the possible hydrogen bonding to the enzyme or a nearby water molecule. The final values for R work and R free were 21.2 and 25.6% for the apo form of AuLFTase MET , 19.1 and 21.8% for the AuLFTase(D54N) MET -sucrose complex, and 21.5 and 25.1% for the AuLFTase(D54N) MET -levanbiose complex. Details of the refinement are described in supplemental Table S2. Structural comparisons and analyses were carried out using the CCP4 suite program (25), and the figures were prepared using PyMOL (Molecular Graphics System, Palo Alto, CA).
Functional Analysis-Various mutant enzymes were prepared by site-directed mutagenesis using a QuikChange kit (Agilent Technologies), with mutagenic primers (supplemental Table S1). All enzymes used for the functional analysis were purified without methylation; their expressions, as well as the WT enzyme, were carried out in Luria Bertani medium, and the C-terminal His-tagged proteins were purified using immobilized metal affinity chromatography. The purified enzymes were dialyzed against 50 mM HEPES (pH 8.0) and diluted to 0.2 mg/ml for the functional analysis.
Enzyme assays were performed according to a method published previously (26,27). Briefly, 300 l of enzyme was incubated for 24 h at 40°C with 600 l of 1% (w/v) levan (Sigma), and the reaction was stopped by boiling for 5 min. After centrifugation, the reaction product was analyzed by TLC and HPLC (supplemental "Methods").

RESULTS
Overall Structure of AuLFTase MET in Apo Form-AuLFTase is a monomeric protein in solution as characterized using size exclusion chromatography. Consistent with this observation, we could not identify any possible oligomeric conformations of AuLFTase MET , either by noncrystallographic or crystallographic symmetry, even with four monomers in an asymmetric unit of the P2 1 2 1 2 1 space group. Those four monomers in the asymmetric unit are structurally identical within a root mean square deviation of 0.19 -0.48 Å for 479 C␣ atoms between Ala-41 and Gln-519. We used Monomer B to describe the overall structure of AuLFTase.
The AuLFTase MET structure consists of the N-domain (Ala-41 to Val-364) and C-domain (Val-371 to Gln-519) ( Fig.  1B and supplemental Fig. S1). Specifically, the N-domain has 18 ␤-strands (␤1 to ␤18) and three short ␣-helices arranged in a five-bladed ␤-propeller fold (Blades I-V) (Fig. 1B), whereas a ␤-sandwich fold of 14 ␤-strands (␤19 to ␤32) is characterized in the C-domain, with the 3 10 -helix as a linker between the two domains. This overall architecture is common to members of the GH32 family, in which the ␤-propeller N-domain functions as the catalytic domain (7). The proposed catalytic residues of AuLFTase conserved in the GH32 and GH68 family also form a cluster for a "catalytic triad" at the center of the N-domain ( There are two layers of antiparallel ␤-sheets in the C-domain, in which the top layer of the ␤-sheet is composed of ␤21, ␤30, ␤23, ␤24, ␤25, and ␤26 in a spatial order and the other ␤-sheet consists of ␤19, ␤31, ␤32, ␤22, ␤27, ␤28, and ␤29. These two ␤-sheets are packed into a parallel orientation to form the ␤-sandwich fold of the C-domain, generating a concave surface on the top ␤-sheet. Notably, the concave surface of the C-domain is on the same face as the entrance to the center of the ␤-propeller N-domain (Fig. 1B).
Structure of AuLFTase(D54N) MET in Complex with Sucrose-An AuLFTase(D54N) MET mutant enzyme, which is catalytically inactive (supplemental Fig. S3A), was used for the structural analysis of the complex with sucrose. Of the four monomers in the asymmetric unit, two exhibited an ordered electron density for sucrose bound in the ␤-propeller N-domain pocket ( Fig. 2A), as well as on the surface of the ␤-sandwich C-domain (supplemental Fig. S4A).
The pocket identified in the N-domain begins at the center of the ␤-propeller fold and extends to the edge of the enzyme enclosed by Blades I and II, resulting in an elongated shape (Fig.  2, B and C). Specifically, the innermost floor of the pocket is laid with residues in ␤1 (Trp-51, Cys-53, and Asn-54) and ␤3 (Leu-70 and Ser-72) strands in Blade I, whereas the side walls along the long axis of the pocket consist of two loops; that on the wall includes a loop containing Pro-106 and Trp-108 between Blades I and II, and the other involves a loop with Trp-322 between ␤16 and ␤17 in Blade V. One end of the pocket is occluded at the center of the domain by residues from Blades III, IV, and V, including residues Glu-236 and Asp-186, as well as Arg-185 and Tyr-298. The other end of the pocket near the Blade I region is also sealed by a loop between ␤3 and ␤4 in Blade I; specifically, residues such as Asn-76, Gly-77, Pro-78, and Gly-79 constitute the end wall of the pocket. However, the arrangement of these residues left the top of the pocket wide open to the surface (Fig. 2B).
In the pocket, sucrose is oriented such that the fructosyl moiety faces the bottom of the pocket as a stacked orientation, and the pyranosyl ring of glucose points upward to the surface of the N-domain, with its conformation almost perpendicular to the fructose furanosyl ring (Fig. 2, B and C). There are extensive hydrogen bonding and hydrophobic interactions with sucrose. In particular, the catalytic triad clustered at the center of the N-domain remains in the vicinity of the sucrose glycosidic bond; the side chain of Asn-54 in the AuLFTase(D54N) MET mutant enzyme, which replaced the WT nucleophile Asp-54, is located ϳ3.6 Å below the anomeric carbon (C2Ј) of fructose, and the carboxyl group of an acid/base catalyst Glu-236 is at 3.1 Å from an oxygen atom of the glycosidic bond. The Asp-186 intermediate-stabilizing residue is located within 2.8 Å from the hydroxyl group of fructose (Fig. 2C). The geometric locations of these three residues fulfill the well established stereochemical requirements of catalysis in the GH32 family (7), in which the glycosidic bond near these residues is susceptible to cleavage. Therefore, we tentatively named the binding site for the fructose and glucose units as the Ϫ1 and ϩ1 subsites, respectively, according to Davies et al. (28). In addition to these interactions, fructose at the Ϫ1 subsite mediates possible hydrogen bonding with Arg-185, Asp-186, and Ser-109, as well as extensive van der Waals interactions within 4.5 Å of the residues in the bottom and wall of the pocket, including Leu-70, Trp-108, Tyr-298, and Trp-322 (Fig. 2C). Extensive interactions were also identified around glucose at the ϩ1 subsite through the residues identified at Ϫ1 subsite; residues such as Trp-108, Arg-185, and Trp-322 from the pocket side wall also maintain the interactions. Specifically, the Arg-185 side chain is in a stacking mode within 4.0 Å to the glucose pyranosyl ring, and the side chains of Trp-108 and Trp-322 are almost perpendicular to the pyranosyl ring. Arg-139 and Glu-236 acid/base catalyst mediate several hydrogen bonds at a distance of 2.6 -3.5 Å to the hydroxyl groups of glucose (Fig. 2C). In particular, Arg-139 mediates various interactions exclusively with glucose at the ϩ1 subsite. It is noticeable that the binding of sucrose occupied only half of the pocket near the catalytic triad at the center of the N-domain, leaving space for the other half near a region of Blade I (Fig. 2B).
Unlike the catalytic N-domain, the function of the ␤-sandwich C-domain has been elusive but is proposed to be involved in carbohydrate recognition based on its structural similarity to a lectin domain (7). In this complex, we identified the binding site of one sucrose molecule on the concave surface of the C-do- main (supplemental Fig. S4A). The fructosyl and glucosyl moieties in this sucrose molecule also adopted a conformation essentially identical to that of sucrose in the ␤-propeller N-domain and was bound to the concave region that is comprised of ␤30, ␤23, ␤24, and ␤25. Specifically, the side chains of two tyrosine residues on the edge of this region, Tyr-433 in ␤25 and Tyr-501 in ␤30, are in a head to head orientation with a possible hydrogen-bonding distance of 2.8 Å between the two hydroxyl groups. This unique structural feature forms a hydrophobic platform on the surface of the C-domain, on which a fructose unit of sucrose sits in a stacking orientation. In addition to these extensive hydrophobic interactions, the residues around fructose further stabilize its conformation through hydrogen bonding, including Asn-408, Asn-424, Asp-435, and Arg-451. In contrast, a pyranosyl ring of glucose, almost perpendicular to the fructose furanosyl ring, has a few interactions; the side chain of Tyr-428 in ␤24 is in an upright position along sucrose and mediates extensive hydrophobic interactions, mainly with a glucose unit. Further structural analysis indicated that this sucrose-binding site is indeed capable of binding other types of carbohydrates, including DFA-IV, levanbiose, and levantriose (supplemental Figs. S3B and S4, B and C), suggesting a possible role for the C-domain in carbohydrate recognition.
Structure of AuLFTase(D54N) MET in Complex with Levanbiose-The crystal structure of the AuLFTase(D54N) MET -levanbiose complex revealed a levanbiose-binding site in an N-domain pocket and in the concave surface of the C-domain. Levanbiose, which is derived from a ␤-2,6-linked levan (supplemental Fig.  S2), consists of two fructose units, one each of reducing and nonreducing fructoses. The electron density for levanbiose clearly exhibits a typical ␤-2,6-O-glycosidic linkage between the fructose units and an extended conformation with a torsion angle of for the C6 -O bond in the ␤-2,6-O-glycosidic linkage (Fig. 3A) (29). Because of this stereochemical restraint, the furanosyl ring of the two fructose units is perpendicular, similar to those of glucose and fructose in the AuLFTase(D54N) MET -sucrose complex.
However, the binding site of levanbiose differs from that of sucrose in the AuLFTase(D54N) MET -sucrose complex. In general, levanbiose remains in the pocket in which sucrose was identified, but its binding site includes the Blade I region (Fig. 3,  B and C), which was empty in the sucrose complex. Levanbiose is bound in this elongated pocket with an extended conformation and sits on the floor of the pocket parallel to the ␤1 and ␤3 strands. In particular, a reducing fructose with a free hydroxyl group (O 2 Ј) at an anomeric carbon is located at the immediate vicinity of the Ϫ1 subsite identified in the AuLFTase (D54N) MET -sucrose complex, whereas a nonreducing fructose with a free hydroxyl group (O6Љ) is bound in the region near Blade I. Therefore, we named this newly characterized site as the Ϫ2 subsite for a nonreducing fructose (Fig. 3, B and C; hereafter, the prime and double prime symbol denote atoms belonging to the Ϫ1 and Ϫ2 subsites, respectively). Many interactions occur along levanbiose; however, interactions around the Ϫ1 subsite are similar to those in the sucrose complex. A furanosyl ring with an orientation perpendicular to the bottom of the pocket is surrounded by Trp-108 and Trp-322 at the Ϫ1 subsite, and its anomeric carbon (C2Ј) is 4.8 Å from the Asn-54 side chain. Except for hydrogen bonding with Cys-53 from the floor of the pocket, fructose at the Ϫ1 subsite maintains interactions similar to those at the Ϫ1 subsite of the AuLFTase(D54N) MET -sucrose complex. In contrast, additional structural features are involved in stabilizing the fructose conformation at the Ϫ2 subsite. A furanosyl ring at the Ϫ2 subsite is perpendicular to the bottom of the pocket and faces in a distance of 5.5-7.0 Å against the Pro-78 loop residue between ␤3 and ␤4 in a stacked orientation. Residues in Blade I make extensive hydrophobic and hydrogen bonding interactions with fructose at the Ϫ2 subsite, which are mediated mainly by a loop region between ␤3 and ␤4. Specifically, this loop region, including Asn-76, Gly-77, Pro-78, and Gly-79, is responsible for extensive hydrophobic contacts to fructose at the Ϫ2 subsite. The elongated pocket in the N-domain is now fully occupied by levanbiose, suggesting that difructosaccharide is a basic unit recognized by this enzyme.
Levanbiose is also bound to the C-domain region identified in the AuLFTase(D54N) MET -sucrose complex (supplemental Fig. S4B). However, its conformation is different from that in the N-domain. Specifically, the relative orientation of the two fructosyl rings is completely inverted, with a torsion angle of 87°. Our modeling study indicates that this conformation allows the polymeric fructose to take a linear form, rather than a circular form dictated by the torsion angles characterized in the N-domain (Fig. 3A). However, its binding environment is A, a modeled sucrose is overlaid with an F o Ϫ F c electron density map contoured at 2.4 and also with labels for a possible subsite for a fructose (Ϫ1) and a glucose (ϩ1) unit. One of the hydroxyl groups in a glucose unit has a disordered density. B, shown is the surface representation of the active site with sucrose in a stick model. Note that a glucose moiety is perpendicular to a fructose unit. C, the active site pocket in the AuLFTase(D54N) MET -sucrose complex is displayed, with the residues that form the pocket. Labels in magenta indicate the Blade numbers. The orientations in B and C are identical. Hydrogen bonding is not indicated but is described in the text.
largely reminiscent of that in the AuLFTase(D54N) MET -sucrose complex, and more interactions are identified because of the extended conformation of levanbiose.
Structure Comparison of AuLFTase and Other Members of the GH32 Family-No significant changes in the conformation of AuLFTase occurred following binding of a ligand; superposition of the three structures described in this study showed a root mean square deviation of less than 0.48 Å for 477 C␣ atoms. These structural features extended into the active site residues, including a catalytic triad, the side chains of which differ only by 0.1-0.5 Å. These observations suggest that no noticeable structural rearrangement was induced by mutation of Asp-54 to Asn-54 in AuLFTase (D54N). Therefore, the active site environment of AuLFTase(D54N) is almost identical to the WT enzyme, except for the immediate vicinity of Asn-54. However, some differences were observed in the orientation of the ligand bound to the active site. Specifically, the relative orientation of fructose at the Ϫ1 subsite was different in the two complexes (supplemental Fig. S5A); the anomeric carbon (C2Ј) at the Ϫ1 subsite of levanbiose was located ϳ1.7 Å farther from Asn-54, and its furanosyl ring adopted a different orientation. Our analysis indicated that fructose at the Ϫ1 subsite in the complex with sucrose has geometrical features more appropriate for an enzymatic reaction (see "Discussion").
Consistent with the general features of GH32 family proteins, the AuLFTase structure in this study showed high similarity with the known structures of the GH32 family (see "Discussion"), but noticeable differences in the ␤-propeller N-domain pocket dimensions were observed. All GH32 family structures determined to date liberate a monomeric unit of carbohydrate, either glucose or fructose, by hydrolysis of the glycosidic bond of various substrates. Accordingly, the substratebinding pocket, which is located at the center of the N-domain near the catalytic triad, accommodates a monosaccharide (Fig.  4A) (30). Structural comparison of AuLFTase with an exo-inulinase of the GH32 family clearly indicates that the Ϫ2 subsite identified in this study is absent in other GH32 family structures (Fig. 4B); there is no space available in other GH32 family proteins for the Ϫ2 subsite, because of sequence changes in the loop between ␤3 and ␤4 of Blade I. In particular, residues such as Asn-76 and Gly-79 are conserved in the corresponding loop among members of LFTase but are changed in other GH32 proteins with residues for a larger side chain such as tryptophan and isoleucine, respectively (supplemental Fig. S1). These features are likely an essential for the substrate specificity of LFTase, i.e., the production of DFAs that consist of difructosaccharide.
Functional Analysis-We carried out a functional analysis of various AuLFTase mutant enzymes by reacting them with levan, the products of which were characterized by TLC and HPLC. Sixteen residues were selected for these analyses, including the proposed catalytic residues for the GH32 family, as well as residues near the pocket.
Consistent with the general enzymatic mechanism of the GH32 family (7), mutant enzymes for the catalytic triad, D54N, D186A, and E236A, are catalytically incompetent  Fig. S2) are 60°for , 176°for , and 171°for . B, shown is surface representation of the active site, with levanbiose in a stick model. C, the active site is shown, with neighboring residues forming the active site pocket. This orientation is identical to that in B. Water molecules are not included for clarity. D, represented is a scheme for the interactions around levanbiose. Residues involved in hydrophobic interactions within 4.5 Å and hydrogen bonding are indicated in magenta and black, respectively, together with the interatomic distance values. Water molecules are represented by red circles.
(supplemental Figs. S3A and S6A), indicating their involvement in DFA-IV production. In addition to these catalytic residues, many others surrounding the ϩ1 and Ϫ1 subsites also play a crucial role in enzyme activity (Figs. 2C and 3C). The catalytically inactive or greatly impaired mutants (supplemental Fig.  S6A) are L70A, W108A, S109A, R185A, and W322A, which are involved in hydrogen bonding or van der Waals interactions to ligands, suggesting that these residues are involved in substrate binding. The only exception is Arg-139, which exclusively interacts with the ϩ1 glucose moiety, and the R139A mutant retains enzymatic activity (see "Discussion"). In contrast, many residues surrounding the Ϫ2 subsite were to some extent tolerant of mutation (Fig. 5), suggesting their roles in pocket formation, not in catalysis. Only the W51A mutant lost its activity, whereas others, such as C53A and S72A, retained their DFA-IV production activity to a level identical to that of the WT enzyme. In particular, a larger side chain was introduced by mutation at positions 76 and 79, which mimics the structural characteristics of the pocket-forming residues for other GH32 proteins (Fig. 4B). Mutants such as N76A, N76W, G77A, and P78G maintained activities as high as the WT enzyme, but G79I and a double mutant of N76W/G79I produced fructose as a major product, rather than DFA-IV, consistent with the structural features of other monosaccharide-producing GH32 proteins. The HPLC analysis further supports our conclusion (supplemental Fig. S6B), because the major product of both mutant enzymes is indeed fructose. These observations suggest that the presence of a larger side chain at position 79 eliminates the fructosyl moiety binding site at the Ϫ2 subsite, and therefore those mutants catalyze hydrolysis of a monomeric fructose, possibly the terminal nonreducing fructose of levan, instead of producing DFA-IV.

DISCUSSION
We presented a crystal structure of AuLFTase, a member of the GH32 family, in apo form and in complexes with sucrose or levanbiose. The overall structural features of AuLFTase are essentially identical to those of other GH32 family members, in that the enzyme consists of two domains: a five-bladed ␤-propeller N-domain and a ␤-sandwich C-domain (Fig. 1B). A structural similarity search using the DALI program (31) indicated that there are more than a dozen GH32 family proteins homologous with AuLFTase, with Z-scores of 20.0 -42.3 and a root mean square deviation of 2.3-4.2 Å. From a functional perspective, the ␤-propeller N-domain is a catalytic domain as revealed by functional analysis, and our structural data further suggest the possible role of the ␤-sandwich C-domain in carbo-hydrate recognition. Given that the carbohydrate-binding site on the concave surface of the C-domain is on the same face as an entrance to the center of the catalytic ␤-propeller N-domain, it could be considered that the polymeric substrate levan binds the C-domain along the concave surface and that its nonreducing end reaches into the active site for catalysis. Recent structural analysis of ␤-fructofuranosidase in the GH32 family also indicated the binding of substrate in the ␤-sandwich C-domain (32).
Despite the marked overall structural similarities, LFTase exhibits novel features in terms of its substrate specificity and the chemical nature of its DFA-IV product compared with other members of the GH32 family that hydrolyze one monosaccharide from the substrate. These functional characteristics of LFTase are imprinted in the active site of the N-domain; it accommodates levanbiose using the Ϫ1 and Ϫ2 subsites (Fig.  3B). These unique features are due to sequence variations in the Blade I loop region between LFTase and other GH32 proteins ( Fig. 4B and supplemental Fig. S1), within which smaller side chains are present at the corresponding region in LFTase, unlike the presence of a bulky side chain in other GH32 proteins, indicating that they are key structural elements for the presence of the Ϫ2 subsite in LFTase. Except for those loop residues, most of the other pocket-forming residues are well conserved in the GH32 family (supplemental Fig. S1). In addition to LFTase, other carbohydrate-degrading enzymes, such as cellobiohydrolase (33) and IFTase (11), release a disaccharide or DFAs as products, but these enzymes belong to a different structural fold with unrelated sequences. However, common to these enzymes is the shape and dimension of the active site. It is  (30), with fructose in a stick model. B, the active site of AuLFTase (gray) is compared with an exo-inulinase (cyan). AuLFTase residues are labeled in black. FIGURE 5. TLC analysis of the reaction products. TLC analysis of the reaction products is shown, using various mutant enzymes associated with the Ϫ2 subsite. Marker is 1% (w/v) glucose, maltose, and maltotriose for G1, G2, and G3, respectively. Levanbiose standard was produced by reacting 1% levan with levanbiohydrolase (27) and further purified. Enzyme reactions by WT and the mutants were carried out as described under "Experimental Procedures." Details for TLC analysis are described in supplemental "Methods".
pocket-shaped, with dimensions suitable for a disaccharide, providing a structural basis for the exo-type cleavage of disaccharide from the substrate.
Structural and functional analyses in this study led us to propose a possible mechanism for LFTase. In particular, our structural analysis suggested that fructose at the Ϫ1 subsite in the AuLFTase(D54N) MET -sucrose complex, rather than in the levanbiose complex, has an orientation favorable for nucleophilic attack by Asp-54 (supplemental Fig. S5A). Consistent with this suggestion, structures of other GH32 enzymes have a fructose conformation at the Ϫ1 subsite identical to that of the AuLFTase(D54N) MET -sucrose complex (supplemental Fig.  S5B). Therefore, our proposed mechanism is based on a levanbiose conformation, in which the Ϫ1 subsite is superposed on that of the AuLFTase(D54N) MET -sucrose complex (Fig. 6A).
In the proposed mechanism (Fig. 6B), the levan substrate binds to the active site of LFTase, with such an orientation that a nonreducing terminal fructose sits at the Ϫ2 subsite, and the preceding fructosyl moiety is located at the Ϫ1 subsite near the Asp-54 nucleophile. The ␤-2Ј,6-glycosidic bond between the Ϫ1 and ϩ1 subsite is cleaved following a nucleophilic attack by Asp-54 on the anomeric carbon (C2Ј) of the Ϫ1 subsite and subsequent reaction by the Glu-236 acid/base catalyst. Consequently, an intermediate between a levanbiosyl moiety and LFTase is formed, in which the anomeric carbon of the Ϫ1 subsite in levanbiosyl moiety is covalently attached to the Asp-54 of LFTase. A further reaction is carried out via an intramolecular fructosyl transfer reaction within a levanbiosyl unit; a nonreducing fructose at the Ϫ2 subsite is swung out about the ␤-2Љ,6Ј-O-glycosidic bond by a rotation around the torsion angle . Because of this rotation the hydroxyl group (O6Љ) of the nonreducing fructose is located near the unprotonated side chain of Glu-236 and serves as an acceptor. Deprotonated by Glu-236, the hydroxyl group (O6Љ) on the nonreducing fructose mediates an S N 2-like nucleophilic attack on the anomeric carbon of the Ϫ1 subsite fructosyl unit, regenerating the carboxyl group of Asp-54 and producing DFA-IV by forming an anhydride bond between the anomeric carbon (C2Ј) of the Ϫ1 subsite and the hydroxyl group (O6Љ) of the nonreducing fructose.
This proposed mechanism is consistent with the general retaining mechanism of the GH32 family (7) and DFA-IV stereochemistry (supplemental Fig. S3B), as well as functional analyses in this study. Our analysis further supports this proposal. First, from a stereochemistry perspective ( Fig. 3A and supplemental Fig. S3B), rotation around in the levanbiosyl moiety is a most feasible and effective way to form DFA-IV. By changing the torsion angle , the hydroxyl group (O6Љ) on the nonreducing fructose approaches the anomeric carbon at Ϫ1 subsite from above the furanosyl plane, fulfilling the stereochemical requirements for an S N 2-like nucleophilic attack and avoiding any possible unfavorable interactions within the levanbiosyl moiety. Second, this rotation does not entail any possible steric hindrance against neighboring enzyme residues. Specifically, the proposed rotation occurs toward the wide open surface of the pocket, eliminating any possible unfavorable hindrance. Consistent with this proposal, the dynamic features of the Ϫ2 subsite fructosyl unit were characterized in the structure of WT AuLFTase MET in complex with levanbiose (supplemental Fig. S5C); a relatively well ordered electron density of a fructose moiety was found at the Ϫ1 subsite, but a highly disordered density was associated with the Ϫ2 subsite. Our structural analysis indicated the presence of Ϫ1 and Ϫ2 subsites in the ␤-propeller N-domain pocket, but the ϩ1 subsite, tentatively assigned based on a glucosyl unit, was unlikely to be a cognate site for the levan substrate. In fact, in the structural analysis of WT AuLFTase MET in complex with levanbiose, we observed a continuous electron density from the Ϫ1 subsite to the rim of the ␤-propeller N-domain in one of the monomers, which was in a different direction from the currently assigned ϩ1 subsite, although its density was not well ordered (supplemental Fig. S5D). A putative ϩ1 subsite is likely located between Trp-322 and Arg-185, suggesting the means by which the R139A mutant retains its activity (supplemental Fig. S6, A  and B), even though Arg-139 interacts exclusively with the glucosyl unit at the currently assigned ϩ1 subsite. Notably, mutation of Trp-322 or Arg-185 abolished enzyme activity (supplemental Fig. S6A).
We characterized the structural features of AuLFTase in apo form and in complex with sucrose and levanbiose. The structural analysis indicated the presence of Ϫ1 and Ϫ2 subsites in the active site and revealed the pocket that accommodates a difructosaccharide. This unique pocket dimension feature is conserved among LFTase members through sequence conservation in the loop region between ␤3 and ␤4 in Blade I but is not FIGURE 6. Proposed binding mode of levanbiose and a reaction mechanism. A, a proposed binding mode of levanbiose (yellow) is shown after the Ϫ1 subsite fructose was superimposed on that of fructose in sucrose (green) characterized in the AuLFTase(D54N) MET -sucrose complex. B, a catalytic mechanism is suggested with possible involvement of the catalytic residues. applicable to other GH32 enzyme family members. Our analyses suggest a possible reaction mechanism for LFTase in which the catalytic triad conserved in the GH32 family plays an essential role, and the terminal nonreducing fructosyl unit serves as acceptor. Our study thus provides a molecular basis for DFA-IV production by LFTase, which exhibits functional variations compared with GH32 family proteins, even within the conserved structural elements.