Structural Enzymology of Cellvibrio japonicus Agd31B Protein Reveals α-Transglucosylase Activity in Glycoside Hydrolase Family 31*

Background: Transglycosylases are important enzymes in bacterial glycogen metabolism. Results: The tertiary structure and function of a novel α-transglucosylase have been defined. Conclusion: In addition to previously known activities, glycoside hydrolase family 31 (GH31) contains a group of enzymes with 1,4-α-glucan 4-α-glucosyltransferase activity. Significance: This gives new insight into bacterial glycogen utilization and will inform future bioinformatics analyses of (meta)genomes. The metabolism of the storage polysaccharides glycogen and starch is of vital importance to organisms from all domains of life. In bacteria, utilization of these α-glucans requires the concerted action of a variety of enzymes, including glycoside hydrolases, glycoside phosphorylases, and transglycosylases. In particular, transglycosylases from glycoside hydrolase family 13 (GH13) and GH77 play well established roles in α-glucan side chain (de)branching, regulation of oligo- and polysaccharide chain length, and formation of cyclic dextrans. Here, we present the biochemical and tertiary structural characterization of a new type of bacterial 1,4-α-glucan 4-α-glucosyltransferase from GH31. Distinct from 1,4-α-glucan 6-α-glucosyltransferases (EC 2.4.1.24) and 4-α-glucanotransferases (EC 2.4.1.25), this enzyme strictly transferred one glucosyl residue from α(1→4)-glucans in disproportionation reactions. Substrate hydrolysis was undetectable for a series of malto-oligosaccharides except maltose for which transglycosylation nonetheless dominated across a range of substrate concentrations. Crystallographic analysis of the enzyme in free, acarbose-complexed, and trapped 5-fluoro-β-glucosyl-enzyme intermediate forms revealed extended substrate interactions across one negative and up to three positive subsites, thus providing structural rationalization for the unique, single monosaccharide transferase activity of the enzyme.

Glycogen is a highly branched, mixed linkage ␣(134)/ ␣(136)-glucan polymer that serves as a readily accessible, osmotically neutral, cellular energy reserve in all domains of life (1)(2)(3). Glycogen is structurally related to amylopectin, which together with the linear polysaccharide amylose (␣(134)-glucan) comprises the plant storage reserve starch (4). Prokaryotic and eukaryotic glycogen biosynthesis and degradation are a complex, highly conserved, and tightly controlled process involving a myriad of enzymes and regulatory factors (2,3). In bacteria such as Escherichia coli, glycogen is synthesized from ADP-glucose by the combined action of glycogen synthase, which builds linear ␣(134)-glucan chains, and glycogen branching enzyme, which catalyzes chain rearrangement via ␣(134)-to-␣(136) transglycosylation, thereby yielding polydisperse molecules with molar masses of up to 10 7 -10 8 Da (2,3). In turn, catabolism under carbon-limited conditions occurs via the sequential action of glycogen phosphorylase and debranching enzyme to yield glucose 1-phosphate.
The Gram-negative soil saprophyte Cellvibrio japonicus is best known for its ability to efficiently utilize a plethora of plant cell wall polysaccharides as energy sources (24). Additionally, the genome sequence of this organism has revealed a large number of predicted ␣-glucan-active enzymes. In total, the C. japonicus genome encodes 22 enzymes from GH13, GH15, GH31, GH57, and GH77 (25) that may be predicted to act on starch and/or glycogen. However, none of these have been biochemically or structurally characterized (6,25). GH31 in particular is one of the major ␣-glucosidase-containing glycoside hydrolase families. This family is functionally diverse; it also contains ␣-xylosidases and ␣-glucan lyases in addition to the aforementioned CtsY and CtsZ ␣-transglycosylases. A phylogenetic analysis has recently been presented that partially delineates these activities in clades, although sequence-based functional prediction is not absolute (26). The generally exo-acting GH31 enzymes, which are members of Clan GH-D together with GH27 and GH36, have been suggested to share a common ancestor with members of clan GH-H, which comprises generally endo-acting ␣-glucan-active enzymes of GH13, GH70, and GH77 (27).
Building upon our interest in the postgenomic characterization of GH31 enzymes from C. japonicus (26), we present here a detailed structural enzymology study of CjAgd31B, whose coding sequence resides within a gene cluster encoding predicted ␣-glucan-active enzymes and sugar transporters. Biochemical analysis revealed that CjAgd31B is a predominant transglucosylase with strict ␣(134) linkage specificity, which represents a previously undiscovered activity in GH31. Crystal-lography of the enzyme in free, acarbose-complexed, and trapped 5-fluoro-␤-D-glucopyranosyl-enzyme intermediate forms has highlighted the structural basis for the strict transfer of a single glucosyl residue and preference for maltotriose and longer substrates. Taken together, the data suggest a biological role for CjAgd31B in glycogen or maltodextrin metabolism that may be complementary to that predicted for the GH77 homologue CjMal77Q.

EXPERIMENTAL PROCEDURES
Curve fitting and processing of kinetics data were performed using Origin 8 software (OriginLab). p-Nitrophenyl (pNP) ␣-glycosides, sucrose, D-maltose, and starch from corn were purchased from Sigma. Malto-oligosaccharides (maltotriose to maltohexaose), isomaltose, melibiose, and acarbose were purchased from Carbosynth. ␣-Glucosyl fluoride and 5-fluoro-␣-D-glucopyranosyl fluoride were kind gifts from Professor Stephen Withers (Department of Chemistry, University of British Columbia, Canada). Ultrapure water was used in all experiments and refers to water purified on a Milli-Q system (Millipore) with a resistivity () Ͼ18.2 megaohms⅐cm.

Cloning of CjAgd31B
The open reading frame encoding CjAgd31B (GenBank accession number ACE84782.1) was amplified by PCR from genomic DNA of C. japonicus Ueda107 using Phusion polymerase (Finnzymes) and the following primers (Thermo Fischer Scientific): 5Ј-CACCATGAATCCGGTCAAACG-3Ј and 5Ј-ATGCAACCTG-AGGTTAAGCGCTTC-3Ј with the forward primer incorporating the CACC overhang (underlined) needed for TOPO cloning and excluding the predicted signal peptide (cleavage site between amino acid residues 24 and 25). The PCR product was cloned into the pENTR/SD/D-TOPO entry vector (Invitrogen) and recombined into the pET-DEST42 destination vector (Invitrogen) as described previously (26).

Gene Expression and Protein Purification
Plasmids harboring the CjAgd31B gene were transformed into E. coli BL21(DE3) by electroporation, the gene was expressed and the resulting protein was purified by immobilized metal affinity chromatography following an established protocol (26). Analysis by SDS-PAGE showed the protein to be electrophoretically pure. LC electrospray ionization MS was used for protein molar mass determination as described previously (28). For crystallization studies, the protein was further purified by size exclusion chromatography and ion exchange chromatography. The eluted protein solution was concentrated to 5 ml by a Vivaspin 20 concentrator (Sartorius Stedim Biotech) and loaded onto a HiLoad 16/60 Superdex 200 prep grade column (GE Healthcare) equilibrated with 20 mM Tris (pH 8.0), 300 mM sodium chloride. The eluted protein solution was dialyzed into 20 mM Tris (pH 8.0) at 4°C for 16 h. The dialyzed protein solution was loaded onto a Resource Q column (GE Healthcare) equilibrated with 20 mM Tris (pH 8.0) and eluted with a linear gradient of 20 mM Tris (pH 8.0), 400 mM sodium chloride. Two major peaks of CjAgd31B were obtained, and the peak eluted in lower salt concentration was collected and used for protein crystallization. Protein concentrations were deter-mined from A 280 values of suitably diluted samples using an extinction coefficient of 139,245 M Ϫ1 ⅐cm Ϫ1 as calculated by the ProtParam tool on the ExPASy server (29).

Thin Layer Chromatography (TLC)
TLC was performed using normal phase silica on aluminum plates eluted with acetonitrile-water (2:1). Analytes were visualized by immersion in 8% H 2 SO 4 in ethanol followed by charring.

High Performance Anion Exchange Chromatography with Pulsed Amperometric Detection (HPAEC-PAD)
Oligo-and monosaccharides were analyzed on a Dionex ICS-3000 HPLC system operated by Chromelion software version 6.80 (Dionex) using a Dionex Carbopac PA200 column. Solvent A was water, solvent B was 1 M sodium hydroxide, and solvent C was 1 M sodium acetate.

Enzyme Assays
Activity on pNP-glycosides was analyzed by a stopped assay as described previously (26) using an enzyme concentration of 6.5 M. The transglycosylation activity of CjAgd31B on various oligosaccharides was performed in 100-l reactions at 25°C in 50 mM citrate buffer (pH 6). For initial rate saturation kinetics experiments, CjAgd31B was added to a final concentration of 1.6 M for maltose and 270 pM for maltotriose, maltotetraose, and maltopentaose. Reactions typically proceeded for 10 min and were stopped by addition of 4 l of 5 M sodium hydroxide.
HPAEC-PAD was used for product analysis using Gradients A, B, D, and E for reactions on maltose, maltotriose, maltotetraose, and maltopentaose, respectively. Commercial malto-oligosaccharides (maltose to maltohexaose) were used as standards.
To test different acceptors, starch was used as a glucosyl donor. Starch from corn was dissolved to 1% (w/v) in water followed by dialysis in deionized water using a 5-kDa-cutoff membrane to remove monosaccharides and small oligosaccharides. 50-l reactions containing 0.4% starch as donor, 1 mM acceptor (glucose or isomaltose), and 2 M enzyme were incubated at 25°C for 10 min and terminated by addition of 2 l of 5 M sodium hydroxide. Products were analyzed by HPAEC-PAD using Gradient C.

IC 50 Measurements
The inhibition of CjAgd31B by acarbose (0 -1050 M) was determined by using maltotriose (100 M) as a substrate in reactions as described above. Product formation was analyzed by HPAEC-PAD (Gradient B).

Crystallization and Data Collection
CjAgd31B was stored in 5 mM Bistris propane (pH 8.5) and concentrated to 7 mg/ml by using a Vivaspin 20 concentrator. In initial crystal screens using Crystal Screen HT, Index HT, SaltRx HT (Hampton Research), and modified Newcastle Screen prepared at the York Structural Biology Laboratory, small single crystals were obtained in several conditions. Well diffracting crystals were obtained after 3-4 days in 1.8 M ammonium sulfate, 0.1 M HEPES (pH 7.0), 2% PEG 400 at 20°C by the sitting drop vapor diffusion method. The structure was solved using experimental phasing with an iodine derivative. This was prepared by placing ϳ1 l of a 0.25 g/ml potassium iodide solution into a 2-l crystallization droplet to allow slow diffusion of iodine into the crystal. Crystallization droplets with iodine solution were left at 20°C for 16 h prior to freezing and data collection. For the complex structures, crystals were soaked in 1.8 M ammonium sulfate, 0.1 M HEPES (pH 7.0), 2% PEG 400 with either 5 mM 5-fluoro-␣-D-glucopyranosyl fluoride or 5 mM acarbose for 1 h at 20°C. All crystals were cryoprotected by 2.0 M lithium sulfate, 0.1 M HEPES (7.0), 2% PEG 400. The x-ray data for the free enzyme and iodine-soaked crystals were collected at 100 K on an ADSC Q315 charge-coupled device detector on beamline I02 of the Diamond Light Source. X-ray data for the complex with acarbose and the 5F␤Glc-enzyme were collected at 100 K on an ADSC Q315 charge-coupled device detector at BL-ID14-4 and BL-ID29 at the European Synchrotron Radiation Facility, respectively. The details of the data collections are listed in Table 1. All data were processed using iMOSFLM (30) and programs from the CCP4 suite (31) unless otherwise stated. The statistics of the data processing and structure refinement are listed in Table 1.
Experimental phasing was performed by single wavelength anomalous diffraction methods at a wavelength of 1.8 Å. Heavy atom substructure solution and initial phasing was performed on the 2.9-Å resolution iodine-derivatized crystal data with autoSHARP (32) followed by phase extension with the 1.9-Å free enzyme data set using DM (33). The 1.9-Å data were used as a starting point for automatic model building using ARP/ wARP (34) ( Table 2). Structure refinement, including TLS refinement of molecular motions, was performed using PHENIX (35) interspersed with manual rebuilding using COOT (36). Complex structures were solved by molecular replacement using MOLREP (37) with the free enzyme structure as a search model and refined as above.

RESULTS
Bioinformatics Analyses-CjAgd31B is found among putative receptors, transporters, and ␣-glucan-active enzymes in the genome of C. japonicus (supplemental Fig. S2). Notably, these genomic neighbors include a predicted ␣-amylase, cyclomaltodextrin glucanotransferase, 6-phospho-␤-glucosidase, and a glucokinase. CjAgd31B has a predicted secretion signal peptide and is thus likely to be localized in the periplasm or to be secreted extracellularly. In addition to CjAgd31B, the genome of C. japonicus encodes two other GH31 members, the putative ␣-glucosidase CjAgd31A and the biochemically and structurally characterized ␣-xylosidase CjXyl31A (26). Although all are members of GH31, CjAgd31B has a low sequence similarity to both CjAgd31A and CjXyl31A with amino acid identities of 28 and 27% and similarities of 43 and 45%, respectively. From our recent phylogenetic analysis (26), the biochemically characterized member of GH31 most similar to CjAgd31B is YihQ from E. coli (sequence identity, 28%; similarity, 44%). E. coli YihQ has been annotated as an ␣-glucosidase based on a weak ability to hydrolyze the artificial substrate ␣-glucosyl fluoride, although the enzyme was impotent toward a range of other ␣-glucosides (38).
Gene Expression-A gene construct encoding CjAgd31B with a C-terminal hexahistidine tag and lacking the predicted native signal peptide was expressed in E. coli BL21(DE3) cells. The protein product was purified by immobilized metal affinity chromatography for kinetics analyses and additionally by size exclusion chromatography and ion exchange for crystallization studies; purity in both cases was confirmed by SDS-PAGE (data not shown). The molar mass of CjAgd31B, corresponding to the C-terminal His-tagged enzyme starting at Asn-25 (the natural site of signal peptide cleavage), was verified by LC electrospray ionization MS (expected, 94,478.8 Da; observed, 94,478.1 Da; supplemental Fig. S3). The overall yield was typically around 100 mg/liter of culture broth.
Transglycosylation Activity on Malto-oligosaccharides-Based on membership in GH31, the substrate specificity of CjAgd31B was initially tested using pNP-␣-glucoside and pNP-␣-xyloside; the enzyme showed no apparent liberation of the aglycone from either of these substrates after extended incubation (1 mM substrate and 6.5 M enzyme and up to 4-h incubation). The enzyme also displayed no detectable activity on sucrose, meli-   (38)). On maltose (Glcp␣(134)Glc), however, formation of both glucose and longer oligosaccharides could be observed using TLC (data not shown). The transglycosylation potential of CjAgd31B was confirmed by HPAEC-PAD following incubation of 6.5 M enzyme with 10 mM maltose for 30 min, which led to a buildup of malto-oligosaccharides. Products with a degree of polymerization of up to 14 glucose residues could be detected (Fig. 1A).
To further analyze the catalytic properties of CjAgd31B, HPAEC-PAD was used to measure product formation from reactions on malto-oligosaccharides. Under conditions of low substrate conversion (Ͻ10% of substrate consumed), the enzyme was shown to transfer a single glucose moiety from a donor to an acceptor molecule. Incubation of the enzyme with linear malto-oligosaccharides (maltotriose to maltopentaose) exclusively yielded Glc n Ϫ 1 and Glc n ϩ 1 products via transglucosylation (Fig. 1B). On maltotriose, maltotetraose, and maltopentaose, the production of glucose, which would indicate competing substrate hydrolysis, was not observed under initial rate conditions. Thus, apparent Michaelis-Menten kinetics parameters for these substrates could be directly determined from plots of v 0 /[E] t versus [Glc n ] ( Fig. 2A) where the rate of transglycosylation product formation is given by Equation 1 (in this case, the rate of Glc n Ϫ 1 formation can also be used). The best substrate for the enzyme was maltotriose with a (k cat / K m ) app value of 196 s Ϫ1 ⅐mM Ϫ1 , whereas maltotetraose and maltopentaose displayed comparable (k cat /K m ) app values of 72 and 58 s Ϫ1 ⅐mM Ϫ1 , respectively (Table 3). Hydrolysis could, however, be detected when maltose was used as a substrate. Here, the production of Glc was measurably higher than the 1:1 stoichiometric ratio of Glc to Glc 3 expected for disproportionation (Glc 2 3 Glc 3 ϩ Glc). In this case, the velocity of the transglycosylation reaction was directly measured according to Equation 1 by quantifying the maltotriose produced. Determination of the hydrolytic rate required subtraction of the amount of glucose co-produced by dispropor-tionation from that arising from hydrolysis, including accounting for the stoichiometry of glucose release by hydrolysis (2 mol of Glc/mol of maltose) (Equation 2). From the specific transglycosylation activity on 0.5 mM maltose, the pH profile was found to be broad with an optimum of pH 6.5 (supplemental Fig. S1).  (Table 3).
Although the (k cat ) app value for maltose transglycosylation is actually higher than that of maltotriose, the lower (k cat /K m ) app value on the disaccharide is due to a high (K m ) app value. This high value may reflect the observation of up to four enzyme subsites (Ϫ1 to ϩ3; nomenclature according to Ref. 39) in the crystal structure (see below) of which only two would be occupied by maltose acting as a donor substrate. Across the range of maltose concentrations examined (50 M to 50 mM), the rate of hydrolysis was consistently low (Fig. 2B). At 50 M maltose, the rate of hydrolysis was equal to the rate of transglycosylation (v 0 /[E] t ϭ 2 s Ϫ1 ) after which the hydrolytic rate increased to a maximum of 10 s Ϫ1 at 2 mM substrate. Nonetheless, the hydrolytic rate was only 13% of the transglycosylation rate at this maximum. The rate of hydrolysis steadily decreased to ϳ8 s Ϫ1 at 50 mM maltose as the transglycosylation rate continued to increase, resulting in a relative hydrolysis rate of Ͻ2% at this concentration.
Starch and Isomaltose as an Alternate Donor/Acceptor Pair-Starch was also tested as a donor substrate with high degree of polymerization with glucose and isomaltose as alternate acceptor substrates. Despite dialysis to reduce the content of short malto-oligosaccharides, the commercial starch sample contained a minor amount of ␣(134)-glucans with a degree of polymerization of 2 and higher. When CjAgd31B was incubated with starch alone, the amounts of shorter malto-oligosac- charides (maltose to maltotetraose) still present in the substrate mixture decreased significantly, whereas longer malto-oligosaccharides (maltopentaose to maltododecaose) increased in keeping with previous observations on purified malto-oligosaccharides during kinetics analyses (Fig. 3A). In contrast, when the reaction was supplemented with 1 mM glucose, a large increase of malto-oligosaccharides (maltose to maltododecaose) could be observed (Fig. 3B). Here, glucose (which as a monosaccharide cannot act as a glycosyl donor) acted as viable acceptor substrate in the breakdown of the glycosyl-enzyme intermediate formed via initial attack of starch chain ends. This reaction is the reverse of glycosyl-enzyme formation when maltose acts as a donor substrate, consistent with the principle of microscopic reversibility. The large increase in medium length malto-oligosaccharides compared with the glucose-free reaction both demonstrates how the non-reducing ends present in starch are utilized as glucose donors and suggests that glucose moieties are mainly "shuffled" between the non-reducing ends of the starch molecules when no additional acceptor is included in the reactions.
The disaccharide isomaltose (Glcp␣(136)-Glc) on its own was not a substrate for CjAgd31B, indicating that the enzyme was unable to utilize ␣(136)-linked glucosides as donor substrates. Glycogen and starch both contain ␣(136)-linked branch points, which prompted us to test whether such branches might be extended by CjAgd31B using isomaltose as a model. Indeed, the enzyme transferred glucosyl moieties from starch to isomaltose (1 mM) as indicated by the appearance of an alternate series of peaks on the chromatogram, each slightly preceding the corresponding all-␣-linked congener, which suggests the formation of Glc n -␣(134)-Glc-␣(136)-Glc saccharides (Fig. 3B). As such, these data indicated that the ϩ2 subsite of the active site is not strictly specific for ␣(134)-linked sugars but will accommodate ␣(136)-linked isomers.
Inhibition by Acarbose-The inhibitory effect of the pseudotetrasaccharide acarbose, a common ␣-glucanase and ␣-glucosidase inhibitor, was assayed using maltotriose as the substrate at a fixed concentration of 100 M. The IC 50 value was determined to be 75.1 Ϯ 3.4 M by plotting the relative activity versus the concentration of acarbose and fitting Equation 3 by nonlinear regression (Fig. 4). With reactions performed at a substrate concentration of 100 M, which is much lower than the apparent K m value (1.2 mM), the IC 50   Tertiary Structures of CjAgd31B-Three tertiary structures of CjAgd31B were obtained at 1.9-, 2.0-, and 1.85-Å resolution, respectively: the free enzyme, a non-covalent complex with the inhibitor acarbose, and a trapped 5-fluoro-␤-D-glucopyranosyl-enzyme intermediate. All crystals contained one molecule in the asymmetric unit. The CjAgd31B structures consisted of a typical GH31 fold comprising four domains with two insertions (Fig. 5A): the N-terminal domain (N-terminal; residues 35-240), the catalytic (␤/␣) 8 domain (residues 241-586) with insertion domain 1 (Insert 1; residues 345-384) and insertion domain 2 (Insert 2; residues 415-435), the C-terminal proximal domain (C-proximal; residues 587-667), and the C-terminal distal domain (C-distal; residues 668 -817). In all structures, the electron density map of the 10 N-terminal residues from 25 to 34 and C-terminal residues from 818 to the end (859), including the V5 epitope and His tag provided by the expression vector, were disordered. The free and acarbose structures had a disordered region from 137 to 140, and the 5-fluoro-␤-glucosyl-enzyme structure had a disordered region from 139 to 140. The visible secondary structures of the four domains, including the insertion domains, in CjAgd31B were well conserved with the human sucrase-isomaltase (Protein Data Bank code 3lpp) of GH31 with a root mean square deviation of 1.9 -2.4 Å.
The free enzyme structure reveals a water-lined pocket where the conserved catalytic aspartic acid residues (Asp-412 and Asp-480) are located (45,46). The pocket was 318 Å 3 as calculated by the Pocket-Finder server (47) Fig. S4). To define enzyme-substrate interactions, two complexes, with the inhibitory tetrasaccharide acarbose (see above) and the covalent 5-fluoro-␤-Dglucopyranosyl-enzyme intermediate, were obtained.
Structure of the Acarbose Complex-The CjAgd31B complex with acarbose revealed clear, unambiguous density for the inhibitor in the Ϫ1 to ϩ3 subsites ( Fig. 5B; subsite nomenclature according to Ref. 39). In the Ϫ1 subsite, the enzyme-derived nucleophile, Asp-412, is indeed poised for nucleophilic attack, lying 3.2 Å "above" the pseudo-anomeric carbon of acarbose and with a nucleophile-C1-NH angle of 164.1°. The catalytic acid, Asp-480, lies 2.5 Å from the "interglycosidic" nitrogen of acarbose as expected. The hydrophobic residues Leu-300, Ile-341, Trp-410, Trp-477, Phe-271, and Phe-513 all lie within a 4-Å distance from the Ϫ1 subsite pseudosugar of acarbose (supplemental Fig. S4).  The ϩ1 subsite contains the 6-deoxyglucosyl moiety of acarbose. Hydrogen bonds are made to Arg-463, Glu-417, and a water molecule, and enzyme-substrate distances suggest van der Waals contacts to Phe-377. In most other solved structures of GH31, the hydrogen bonds provided here by Glu-417 (Insert 2) are made instead by an aspartate residue from a loop in the N-terminal domain (42, 48 -50); the other exception to this is found in CjXyl31A in which a PA14 domain insert in the N-terminal domain extends the active site (26, 42, 48 -50).
The substrate-interacting residues in the Ϫ1 and ϩ1 subsites of the CjAgd31B structure are essentially homologous to those in the human maltase-glucoamylase and the Ro-␣G1 ␣-glucosidase from Ruminococcus obeum with the exception of Phe-271, Leu-300, Phe-377, and Leu-413, which instead are Trp, Ile, Trp, and Met, respectively, in the maltase-glucoamylase and Ro-␣G1 structures (supplemental Fig. S5 and Refs. 48 and 49). These side chains all make van der Waals contacts to acarbose as well as the 5-fluoro-␤-glucosyl residue (see below). It is possible that Phe-271 of CjAgd31B, corresponding to Trp-169 and Tyr-299 in Ro-␣G1and maltase-glucoamylase, respectively, contributes to substrate specificity as wild type Ro-␣G1 prefers isomaltose to maltose as a substrate, whereas the W169Y mutant inverts this preference (49).
Notable features of the ϩ2 and ϩ3 subsites are the tyrosine clamp of Tyr-179 (part of a long loop extending from the N-terminal domain) and Tyr-376 of Insert 1, which together form van der Waals contacts to the internal glucose moiety (third ring) of acarbose (cf. Fig. 5B and supplemental Fig. S4). A similar hydrophobic clamp has not been found in other GH31 structures apart from the PA14-mediated protein-sugar interaction in CjXyl31A (26).   (Fig. 5B). Of particular note is that neither (nearby) water of the intermediate complex is in an appropriate position for nucleophilic attack of the intermediate by hydrolysis, implying that hydrogen bonding is optimized to prevent hydrolysis and facilitate transglycosylation. This figure was drawn with CCP4MG (66).

JOURNAL OF BIOLOGICAL CHEMISTRY 43295
Structure of the Trapped Covalent 5-Fluoro-␤-glucosyl-enzyme Intermediate-To assess factors that may lead to the strict transglycosylation activity of CjAgd31B, a near mimic of the covalent glycosyl-enzyme intermediate was accessed using a classic "Withers" reagent, 5-fluoro-␣-D-glucopyranosyl fluoride. The electron density map clearly reveals the trapped 5-fluoro-␤-D-glucopyranosyl-enzyme observed in 1 S 3 skew boat conformation (conformational aspects of catalysis are reviewed in Ref. 51) formed via covalent linkage to the O␦2 atom of Asp-412 with 1.34-Å distance (Fig. 6A). The majority of interactions of this sugar are the same as previously observed for the Ϫ1 subsite sugar of the acarbose pseudotetrasaccharide (supplemental Fig. S4), but in addition, two water molecules (Fig. 6A) bind to O␦2 atom of Asp-480, the acid-base residue. These water molecules form a hydrogen bond network with Asp-480, Arg-463, Glu-417, and a third water molecule binding to Gln-542. Although transglycosylation is always kinetically favored over hydrolysis (52), what is unusual about CjAgd31B and indeed other transglycosylases is how they overcome the thermodynamically favored hydrolysis reaction in 55 M water as discussed below.

DISCUSSION
Through a combination of enzymological and structural analysis, we have revealed that CjAgd31B from the soil saprophyte C. japonicus possesses the ability to exclusively transfer single glucosyl units from ␣(134)-glucans to the non-reducing terminal 4-OH of glucose and ␣(134)and ␣(136)-linked glucosyl residues; weak hydrolysis activity is only observed on the disaccharide maltose. As outlined in the Introduction and discussed below, this type of transglycosylase has not previously been described in GH31 nor any other CAZyme family to our knowledge.
GH31 enzymes utilize a double displacement mechanism involving a covalent glycosyl-enzyme intermediate, which, as was the case here for CjAgd31B, can be trapped and directly observed using kinetic probes derived from fluorosugars (46). In the natural reactions catalyzed by GH31 members, the glycosyl-enzyme is most commonly decomposed by water, yielding substrate hydrolysis. However, this intermediate can also be intercepted by saccharide acceptor substrates to generate transglycosylation products with varying efficiencies in a substrate-and enzyme-dependent manner (53,54). Indeed, several members of GH31 have been shown to possess transglycosylation ability, although yields are typically low due to dominating hydrolytic reactions (21-23, 55, 56). In this context, the strict transglycosylating activity of CjAgd31B on malto-oligosaccharide substrates with degree of polymerization Ն3 is particularly noteworthy.
GH77 enzymes are structurally related to GH13 enzymes in clan GH-H and thus generally possess open cleft-shaped active sites (60 -62) that confer specificity for longer glucan donor substrates (9). For example, the eukaryotic Solanum tuberosum (potato) starch disproportionating ("D") enzyme has such an extended active site (Protein Data Bank code 1x1n), transfers long ␣-glucan chains, and does not use maltose as a donor substrate (14,63,64). The disaccharide is also not a substrate for the Thermus aquaticus amylomaltase of GH77 that is distinguished by its propensity to form large cyclic ␣-glucans from long ␣(134)-glucan donors (14,60). In the context of bacterial malto-oligosaccharide metabolism, the E. coli GH77 amylomaltase MalQ appears to favor the transfer of longer ␣-glucan chains, although there appears to be some debate whether this enzyme can utilize maltose as a donor, thereby transferring a single glucosyl residue to longer congeners.
CjAgd31B thus occupies a unique catalytic place as a 4-␣glucosyltransferase among the broader spectrum of 4-␣-glucanotransferases. As a member of GH31, CjAgd31B belongs to clan GH-D, which is also composed of GH27 and GH36. Clan GH-D and GH-H members are built on a common triose isomerase (␤/␣) 8 barrel scaffold and may share a distant evolutionary relationship (27). However, in contrast to the clefted clan GH-H members, clan GH-D members are typified by shallow, pocket-shaped active sites comprising only one negative subsite accommodating a monosaccharide residue, the glyco-sidic bond of which undergoes catalysis. CjAgd31B likewise presents an active site pocket as revealed by acarbose and 5-fluoro-glycosyl-enzyme complex structures, allowing speculation regarding its strict glucosyl transfer capacity.
The inherent challenge for transglycosylases is overcoming the thermodynamic preference for water as a nucleophile versus saccharide acceptor substrates. It has long been established by Withers et al. (52) through analysis of the reactivation of trapped intermediates that transglycosylation is kinetically favored over hydrolysis. Crystallographic analysis of the trapped covalent glycosyl-enzyme intermediate of CjAgd31B suggests that the hydrogen-bonding scheme does not place a water molecule with appropriate geometry or interaction with the catalytic acid-base residue Asp-480 to facilitate hydrolysis of the intermediate. Instead, two water molecules (one of them disordered) appear to lie on either side of the position expected of a catalytically competent nucleophile. However, the interglycosidic nitrogen of acarbose does interact with the acid-base (and with its O3 and C5 groups, binding in positions corresponding to the observed waters of the trapped intermediate).
This solvent hydrogen-bonding arrangement suggests that the CjAgd31B active site has evolved to avoid deprotonation and activation of water, whereas optimization of hydrogen bonds to O6 and O3 of acceptor glucosides in the ϩ1 subsite allows a favorable placement of the O4 atom for deprotonation and concerted electrophilic migration of C1 of the ␤-glucosylenzyme. In this process, Glu-417 and Arg-463 play particularly important roles in binding the O3 of the ϩ1 sugar and legislating against a water molecule positioned to enable hydrolysis. A caveat is that the trapped intermediate observed is that of a 5-fluoroglycoside, so it is possible that the observed solvent network is perturbed by the unnatural 5-fluoro substituent. However, the solvent network of the free enzyme structure is similar to that of the trapped intermediate complex especially in context of the O6-and O3-mimicking water molecules, where there is no suitably poised "nucleophilic" water molecule bound to Asp-480.
It is currently unclear what role(s) CjAgd31B might play in the biology of C. japonicus. Interestingly, the gene encoding CjAgd31B is located among a cluster of genes predicted to encode ␣-glucan-active enzymes (␣-amylase, cyclomaltodextrin glucanotransferase, 6-phospho-␤-glucosidase, and glucokinase) and transporter proteins (TonB-dependent receptors and ATP-binding cassette transporters; supplemental Fig. S2). Whether these genes are co-regulated or comprise an operon is currently not known. This genomic association together with the observation that CjAgd31B is encoded with a native secretion signal peptide hints toward a role in glycogen or starch metabolism in the periplasm. Indeed, CjAgd31B could possibly have a function similar to the GH77 amylomaltase MalQ from E. coli, which creates longer ␣-glucan chains from shorter malto-oligosaccharides as substrates for maltodextrin phosphorylases; these phosphorylases require maltopentaose as a minimal substrate to generate glucose 1-phosphate for further metabolism (5,65,66). C. japonicus, however, does possess a predicted GH77 homologue (CjMal77Q, CJA_1882; Ref. 25), which is located elsewhere in the genome and in proximity to other predicted glycogen/starch-active enzymes. This would suggest that CjAgd31B and CjMal77Q most likely have independent or perhaps complementary functions.
Another possible clue to the physiological function of CjAgd31B can be gleaned from analysis of potential GH31 orthologs. The biochemically characterized GH31 member closest to CjAgd31B is the ␣-glucosidase YihQ from E. coli K12 MG1655 (38). Biological data on E. coli YihQ are currently lacking; however, a reverse genetics analysis of a YihQ orthologue in Salmonella enterica serovar Enteriditis indicates that ⌬yihQ mutants are deficient in capsular polysaccharide formation (67). Notably, the LPS of this organism consists of a repeating core glycan comprising tyvelose, L-rhamnose, galactose, and mannose that is appended with extended ␣(134)-glucan chains. It is therefore tempting to speculate that YihQ and by extension CjAgd31B may act as a transglucosylase to extend or restructure these chains. In this context, it is interesting to note that E. coli YihQ has previously been designated as an ␣-glucosidase based on a weak activity on ␣-glucosyl fluoride but no other ␣-glucosides (38). A reassessment of YihQ activity both in vitro and in vivo in light of the transglycosylation capacity of GH31 enzymes demonstrated in the present study may well be warranted. In conclusion, the detailed enzyme structure-function analysis of Agd31B from the model soil bacterium C. japonicus presented here that has defined a previously unknown ␣-transglucosylase activity in GH31 will inform future functional genomics studies in bacteria and other microorganisms.