Substrate Recognition and Hydrolysis by a Family 50 exo-β-Agarase, Aga50D, from the Marine Bacterium Saccharophagus degradans*

Background: The catalytic mechanism and substrate recognition required for exo-β-agarase activity are unclear. Results: Structural analysis of a family 50 glycoside hydrolase details substrate recognition and supports a retaining mechanism. Conclusion: The active site architecture dictates exo-activity, whereas substrate distortion aids glycosidic bond hydrolysis. Significance: This structural work offers the first snapshots of Michaelis complexes formed during hydrolysis of the β-linkages in agarose. The bacteria that metabolize agarose use multiple enzymes of complementary specificities to hydrolyze the glycosidic linkages in agarose, a linear polymer comprising the repeating disaccharide subunit of neoagarobiose (3,6-anhydro-l-galactose-α-(1,3)-d-galactose) that are β-(1,4)-linked. Here we present the crystal structure of a glycoside hydrolase family 50 exo-β-agarase, Aga50D, from the marine microbe Saccharophagus degradans. This enzyme catalyzes a critical step in the metabolism of agarose by S. degradans through cleaving agarose oligomers into neoagarobiose products that can be further processed into monomers. The crystal structure of Aga50D to 1.9 Å resolution reveals a (β/α)8-barrel fold that is elaborated with a β-sandwich domain and extensive loops. The structures of catalytically inactivated Aga50D in complex with non-hydrolyzed neoagarotetraose (2.05 Å resolution) and neoagarooctaose (2.30 Å resolution) provide views of Michaelis complexes for a β-agarase. In these structures, the d-galactose residue in the −1 subsite is distorted into a 1S3 skew boat conformation. The relative positioning of the putative catalytic residues are most consistent with a retaining catalytic mechanism. Additionally, the neoagarooctaose complex showed that this extended substrate made substantial interactions with the β-sandwich domain, which resembles a carbohydrate-binding module, thus creating additional plus (+) subsites and funneling the polymeric substrate through the tunnel-shaped active site. A synthesis of these results in combination with an additional neoagarobiose product complex suggests a potential exo-processive mode of action of Aga50D on the agarose double helix.

The bacteria that metabolize agarose use multiple enzymes of complementary specificities to hydrolyze the glycosidic linkages in agarose, a linear polymer comprising the repeating disaccharide subunit of neoagarobiose (3,6-anhydro-L-galactose-␣-(1,3)-D-galactose) that are ␤-(1,4)-linked. Here we present the crystal structure of a glycoside hydrolase family 50 exo-␤-agarase, Aga50D, from the marine microbe Saccharophagus degradans. This enzyme catalyzes a critical step in the metabolism of agarose by S. degradans through cleaving agarose oligomers into neoagarobiose products that can be further processed into monomers. The crystal structure of Aga50D to 1.9 Å resolution reveals a (␤/␣) 8 -barrel fold that is elaborated with a ␤-sandwich domain and extensive loops. The structures of catalytically inactivated Aga50D in complex with non-hydrolyzed neoagarotetraose (2.05 Å resolution) and neoagarooctaose (2.30 Å resolution) provide views of Michaelis complexes for a ␤-agarase. In these structures, the D-galactose residue in the ؊1 subsite is distorted into a 1 S 3 skew boat conformation. The relative positioning of the putative catalytic residues are most consistent with a retaining catalytic mechanism. Additionally, the neoagarooctaose complex showed that this extended substrate made substantial interactions with the ␤-sandwich domain, which resembles a carbohydrate-binding module, thus creating additional plus (؉) subsites and funneling the polymeric substrate through the tunnel-shaped active site. A synthesis of these results in combination with an additional neoagarobiose product complex suggests a potential exo-processive mode of action of Aga50D on the agarose double helix.
Heterotrophic marine bacteria drive the ocean carbon cycle by recycling carbohydrates from algae, the base of marine food webs (1). Many algal polysaccharides are highly heterogeneous, with different sugar monomers, glycosidic linkage types, and chemical modifications within the same macromolecule. These complex carbohydrates are metabolized with sophisticated systems of enzymes (2), many of which work in a concerted manner to complete polymer deconstruction. For instance, endoacting enzymes cleave within the polysaccharide chain and produce longer oligosaccharides as final products. exo-Acting enzymes continue degradation and cleave these longer oligomers, from their ends, into shorter oligomer or monomer products that are ultimately amenable to complete catabolism. Because carbohydrates represent one of the most abundant carbon resources in the ocean, it is fundamental to understand carbohydrate cycling by marine bacteria and its impact on the ecology of heterotrophic microbes in the wild (3,4).
Agarose belongs to the agar class (5) of linear and frequently sulfated polysaccharides that represent up to 70% of the cell wall polymers in red seaweeds (Agarophytes). Because of its abundance, agarose is a good model substrate to study the mechanism of degradation of gel-forming polysaccharides by marine microbes. Agarose is a neutral agar consisting of alternating ␤-1,3-linked D-galactose and ␣-1,4-linked 3,6-anhydro-L-galactose residues forming a parallel double helix (6 -8). Many marine microbes have evolved degradation enzymes to consume agarose as an energy and carbon source (for a review, see Refs. 9 -15). Agarose degradation has been intensively characterized in the heterotrophic microbe Saccharophagus degradans (9). This bacterium contains multiple enzymes, displaying different endo-and exo-modes of action to hydrolyze the ␣-1,3 and ␤-1,4 glycosidic bonds in agarose. S. degradans contains five ␤-agarases, which cleave the ␤-1,4 glycosidic bond, belonging to glycoside hydrolase families 16 (AgaB), 50 (AgaA and -D; here AgaD is referred to as Aga50D), and 86 (AgaC and -E) (16). This bacterium also has a single family GH117 ␣-agarase that hydrolyzes the ␣-1,3 bond (17). Formally, the process of agarose degradation can be divided into three steps, as inferred from biochemical, genetic, and in vivo experiments. In the first step, the endo-␤-agarases GH16 (AgaB) and GH86 cleave agarose into neoagarooligosaccharide products, with a degree of polymerization (DP) 4 of 4 -6 (GH16) and DP6 -DP8 (GH86) (9, 16, 18 -23). In an intermediate step, the exo-␤-agarases GH50A, GH50D, and possibly GH86 act as neoagarobiohydrolases to cleave these larger neoagarooligosaccharides into neoagarobiose (DP2) (24). Finally, the GH117 neoagarobiose hydrolase completes hydrolysis by cleaving the disaccharide into monomeric sugars. Deletion of the genes encoding the individual enzymes in this pathway has impaired or, as for AgaA, even abolished the ability of the bacterium to metabolize agarose (9,19,20), highlighting that these diverse agarases form finely tuned systems to completely break down agarose into its constituent monosaccharides.
Detailed molecular insight into the enzymatic mechanism of such systems has been advanced through the structural characterization of agarases from different families and microbes (15). In accordance with their role of generating longer oligosaccharides, the crystal structures of GH16 endo-␤-agarases (AgaA, AgaB, and AgaD) from Zobellia galactanivorans and the GH86 endo-␤-porphyranase from Bacteroides plebeius, revealed open and extended substrate binding clefts (11,21,25,26). In contrast, the structures of GH117 exo-␣-neoagarobiose hydrolases, from S. degradans (27), Z. galactanivorans (28), and B. plebeius (17), displayed small substrate binding pockets sufficient to bind and cleave a disaccharide. Structural analysis of a neoagarobiohydrolase, such as those from GH50, which have been found critical for agarose metabolism in S. degradans (9), has not been reported, limiting our understanding of the agarolytic system of this and other marine microbes.
Aga50D from S. degradans is a well characterized enzyme whose exo-lytic mode of neoagarobiose production from agarose, kinetic properties, and likely processivity have been described in detail and provide the motivation for exploring the molecular basis of its activity (24). Here we report the structural analysis of the ␤-agarase Aga50D from S. degradans 2-40 (Aga50D) in complex with a series of agarose oligomers. The structure of Aga50D revealed a fold comprising an elaborated (␣/␤) 8 barrel fused to a ␤-sandwich domain that bears strong resemblance to carbohydrate-binding modules (CBMs; for a review, see Boraston et al. (29)). This architecture creates a deep, partially covered active site whose properties were explored by trapping Michaelis complexes for non-hydrolyzed neoagarotetraose and neoagarooctaose using a catalytically inactive mutant of Aga50D. These complexes revealed a puta-tive catalytic machinery that is most consistent with a retaining catalytic mechanism (30) and a tunnel-shaped active site, suggesting an exo-processive mechanism (25). Furthermore, the structural data allowed us to propose a structural model for the degradation of agarose helices by Aga50D. Combined, these results advance our understanding of the role of GH50 exo-␤agarases within the agarolytic enzyme systems of marine microbes.

EXPERIMENTAL PROCEDURES
Materials-All reagents and chemicals were purchased from Sigma unless otherwise specified.
Neoagarooctaose, neoagarohexaose, and neoagarotetraose were produced from agarose hydrolysis using Aga50D and purified by chromatography on Bio-gel P2 resin (Bio-Rad). Briefly, a solution of agarose (2%) in water was prepared by heating until the agarose was dissolved. The solution was cooled down to and kept at 40°C, and Aga50D at a final concentration of ϳ10 g/ml was added. The digestion was carried out overnight with shaking at 200 rpm. After digestion, the enzyme was heat-inactivated for 20 min at 90°C. The solution was cooled down to 20°C, and denatured protein and aggregates of high molecular weight agarose were removed by centrifugation. The clarified solution was lyophilized. 500 mg of the sample was redissolved in 1 ml of water prior to filtration through a 0.2-m membrane (Millipore) and size exclusion chromatography using Bio-gel P2 resin equilibrated with 50 mM ammonium carbonate buffer (pH 7.5). Samples from the 2-ml fractions were analyzed by TLC as described previously, and oligomers were identified by comparison with standards of the neoagarobiose series (17). Fractions with purified oligosaccharides were pooled and lyophilized.
Cloning-The gene fragment encoding Aga50D lacking a predicted secretion signal peptide (amino acids 47-793) was amplified by PCR from S. degradans 2-40 genomic DNA using primers to introduce 5Ј NdeI and 3Ј XhoI restriction sites: 5Ј-TTT GGA CAT ATG ATG TTA TTC GAT TTT GAA AAC G-3Ј (forward) and 5Ј-CTT TGT CTC GAG TTA TTT GCT GCC TAG CCT TTC GG-3Ј (reverse). Site-directed mutagenesis was used to create catalytically inactive protein (referred to as Aga50D-EQ). The E534Q mutation was introduced with the forward primer 5Ј-GTA TTT ATC GAT AAC CAA AAA AGC TTC GGT CGC-3Ј and the reverse primer 5Ј-GCG ACC GAA GCT TTT TTG GTT ATC GAT AAA TAC-3Ј (boldface letters indicate the mutated codon) using standard PCR mutagenesis. The amplified products were cloned into pET28a via the engineered restriction sites using standard molecular biology procedures to generate pETAga50D and pETAga50D-EQ for the wild-type and mutant, respectively. The resulting recombinant gene encoded a N-terminal His 6 tag fused to the protein by an intervening thrombin protease cleavage site. Bidirectional DNA sequencing was used to verify the fidelity of each construct.
Protein Expression and Purification-pETAga50D and pETAga50D-EQ were transformed into Escherichia coli BL21 Star (DE3) cells (Invitrogen), and proteins were produced using LB medium supplemented with kanamycin (50 mg⅐ml Ϫ1 ). Briefly, bacterial cells transformed with the appropriate expres-sion plasmid were grown at 37°C until the culture reached an optical density of 0.8 -0.9 at 600 nm. Protein production was then induced by the addition of isopropyl ␤-D-1-thiogalactopyranoside to a final concentration of 0.5 mM, and incubation was continued overnight at 16°C with shaking at 200 rpm. Cells were harvested by centrifugation and disrupted by chemical lysis (31). Proteins were purified from the cleared cell lysate by Ni 2ϩ -immobilized metal affinity chromatography followed by size exclusion chromatography using a Sephacryl S-200 column (GE Healthcare). Purified protein was concentrated using a stirred cell ultrafiltration device with a 10,000 Da molecular weight cut-off membrane (Millipore). Protein concentration was determined by measuring the absorbance at 280 nm and using a calculated molar extinction coefficient of 165,590 cm Ϫ1 ⅐M Ϫ1 for Aga50D and Aga50D-EQ (32).
Activity Assay-High performance anion exchange chromatography with pulsed amperometric detection analysis was performed using a Dionex ICS3000 instrument equipped with a CarboPac PA-200 column and pulsed amperometric detection. Samples were loaded onto the column as 20-l volumes followed by elution with a linear gradient of sodium acetate (0 -300 mM) in 100 mM sodium hydroxide over 30 min. Agarose polymers (80 M) were incubated at 37°C in the presence of 20 nM Aga50D, and then aliquots of the reaction were stopped at various times by the addition of 1 reaction volume of sodium hydroxide followed by centrifugation at 10,000 ϫ g prior to high performance anion exchange chromatography analysis. The catalytically inactive mutant Aga50D-EQ was used as a negative control (not shown). Neoagarobiose (DP2), neoagarotetraose (DP4), neoagarohexaose (DP6), and neoagarooctaose (DP8) were run individually at a concentration of 40 M as standards. All distinctive peak positions in the resulting chromatograms were then compared with those of the standards for identification of the carbohydrates. Experiments were performed at least two times and were highly reproducible.
General Crystallography-Crystallization experiments were performed using the vapor diffusion method at 18°C with sitting drops for screening and hanging drops for optimization. For data collection, single crystals were flash-cooled with liquid nitrogen in crystallization solution supplemented with a cryoprotectant optimized for each crystal form as given below. Diffraction data were collected either on Beamline 9-2 of the Stanford Linear Accelerator Center (Stanford Synchrotron Radiation Lightsource (SSRL)) or 08ID-1 or 08B1-1 at the Canadian Light Source (CLS, Saskatoon, Canada), as indicated in Table 1. All diffraction data were processed using MOSFLM and SCALA (33,34). Data collection and processing statistics are shown in Table 1. For all structures, manual model building was performed with COOT (35), and refinement of atomic coordinates was performed with REFMAC (36). The addition of water molecules was performed in COOT with FINDWATERS and manually checked after refinement. In all data sets, refinement procedures were monitored by flagging 5% of all observation as "free" (37). Model validation was performed with SFCHECK (38), PROCHECK (39), and MOLPROBITY (40, 41). All model statistics are shown in Table 1.   Crystallization and Structure Determination-Crystals of a selenomethionine-labeled Aga50D (25 mg⅐ml Ϫ1 ) were obtained in 17% (w/v) PEG 3350, 0.24 M lithium sulfate. Crystals were cryoprotected in crystallization solution containing 25% glycerol. A single-wavelength anomalous dispersion data set optimized for selenium (energy ϭ 12,661.31 eV, fЈ ϭ Ϫ8.11e, fЉ ϭ 5.61e) was collected. Heavy atom substructure determination, phasing, and density modification was performed with AutoSHARP (42). Thirty-three selenium positions corresponding to approximately 16 selenomethionine residues present in each of the two Aga50D monomers in the asymmetric unit were used for phasing with the full 2.65 Å resolution data set (acentric/centric figures of merit 0.40/0.20; phasing power, 1.35). The phases resulting from density improvement were of sufficient quality for BUCCANEER (43) to build a virtually complete model. This model was used as a reference for molecular replacement.
Crystals of native Aga50D (25 mg⅐ml Ϫ1 ) were grown in 19% (w/v) PEG 3350, 0.22 M lithium sulfate, 0.1 M Tris-HCl, pH 8.6. A data set was collected as described above on a single crystal that was cryoprotected in crystallization solution supplemented with 25% (v/v) glycerol. The initial model built from the selenomethionine derivative data were used to solve the structure of native Aga50D by molecular replacement using PHASER (44). A combination of automatic model building with BUCCANEER, manual building with COOT, and refinement with REFMAC was used to complete the model.
Crystals of Aga50D were soaked in an excess of neoagarobiose for 30 min in the crystallization conditions prior to freezing using 25% (v/v) glycerol in the crystallization solution. The completed model of Aga50D was used as a starting point to solve the structure of this complex.
Crystals of Aga50D-EQ (20 mg⅐ml Ϫ1 ) were obtained in the presence of 19 -21% (w/v) PEG 3350, 0.22-0.24 M lithium sulfate, 0.1 M Tris-HCl, pH 8.5. These crystals were soaked for 30 min in the crystallization solution containing an excess of neoagarotetraose or neoagarooctaose prior to freezing using 25% (v/v) glycerol in crystallization solution as a cryoprotectant. These structures were solved by molecular replacement as described above.

RESULTS
The Molecular Architecture of Aga50D-Aga50D was previously determined to be an enzyme that releases only neoagarobiose from agarose in an exo-lytic mode of action (24). Using high performance anion exchange chromatography with pulsed amperometric detection with defined neoagarooligosaccharides of DP4, DP6, and DP8, we demonstrated that our recombinant Aga50D is active and breaks these substrates down to neoagarobiose (supplemental Fig. 1), consistent with a previous report (24). The oligosaccharide substrates used for these experiments were produced from soluble agarose using Aga50D, which is at odds with the apparent exo-acting activity of the enzyme. Unlike the product profile analysis, the soluble agarose degradation was performed with high substrate concentrations over a long period of time, ultimately producing concentrated product. We propose that the longer oligosaccharides generated by extensive digestion of concentrated agarose with Aga50D result from transglycosylation reactions.
In order to provide insight into the molecular basis of agarose degradation by GH50 enzymes, we used x-ray crystallography to determine the structure of Aga50D. Selenomethionine-labeled Aga50D crystallized in the space group P2 1 2 1 2 1 and allowed the structure to be determined to 2.65 Å resolution by the single-wavelength anomalous dispersion method. An initial model comprising the two monomers in the asymmetric unit was built by autobuilding with minimal manual intervention. The most complete monomer from this initial model was used as a search model to solve the structure of the native protein to 1.9 Å resolution in the space group P4 1 . The final refined model based on the x-ray diffraction data from this crystal form contained the four molecules of Aga50D present in the asymmetric unit. In all of the monomers, residues 45-790 could be traced, with three gaps present in only one of the monomers.
Aga50D adopts a hybrid fold comprising a ␤-sandwich domain at the N terminus of the enzyme fused to an elaborated (␣/␤) 8 barrel at the C terminus. These two domains are joined by an ϳ30-amino acid linker with a central ␣-helix that packs against the (␣/␤) 8 barrel (Fig. 1A).
The ␤-sandwich domain comprises two 5-stranded ␤-sheets that pack against one another (Fig. 1B). The domain contains a metal binding site, which on the basis of B-factor analysis and coordination geometry was modeled as occupied by a calcium ion. This fold is immediately reminiscent of the ␤-sandwich CBMs (29), including three solvent-exposed aromatic amino acids. Structural similarity searches (PDBeFold (45)) with only the ␤-sandwich domain of Aga50D revealed significant structural identity with the family 11 CBM of Clostridium thermocellum Lic26A-Cel5E (PDB code 1V0A; Z-score 9.6, RMSD 1.98 over 129 aligned residues (46)) ( Fig. 1C) and the family 29 CBM from Pyromyces equii (PDB code 1GWK; Z-score 7.5, RMSD 2.28 over 123 aligned residues (47)). The calcium binding site is also conserved with one of the two sites in CtCBM11 (Fig. 1C). Despite the structural similarities of these domains, only one amino acid residue involved in carbohydrate binding by CtCBM11 is structurally conserved with Tyr-94 in the Aga50D CBM-like domain (Fig. 1C). CBMs are often joined to their cognate catalytic domains by flexible linkers, allowing conformational freedom of the domains (48). Aga50D possesses a fairly long ϳ30-amino acid linker; however, this linker possesses defined secondary structure and is well ordered as it packs against the (␣/␤) 8 barrel, suggesting a complete lack of flexibility. Furthermore, the molecular interface between the CBM-like domain and the (␣/␤) 8 barrel is extensive, burying 1460 Å 2 of surface area with numerous hydrogen bonds and salt bridges. This interaction firmly fuses the CBM-like domain to the (␣/␤) 8 barrel in a conformation that is identical not only in the four monomers present in the asymmetric unit but also in the two monomers of the P2 1 2 1 2 1 crystal form used to generate substrate complexes (see below). Thus, the fused domain architecture creates an overall well ordered asymmetric globular fold (Fig. 1D).
An examination of the solvent-accessible surface of Aga50D reveals the presence of a channel that is blocked at one end and partially covered by a bridge, making a short tunnel (Fig. 1F). The putative catalytic residues reside in the tunnel, making this a candidate active site. Notably, the CBM-like domain in Aga50D is fused in such a way that it appears to extend the surface leading into the proposed catalytic machinery. This ori- entation suggests that the CBM-like domain may be involved in substrate recognition.
Substrate Recognition by Aga50D-Based on the postulated role of Glu-534 and Glu-695 as catalytic residues, we sought to mutate these in order to trap complexes of Aga50D with nonhydrolyzed substrate. The Aga50D E534Q mutant (Aga50D-EQ) was successfully crystallized and proved to be sufficiently inactive to obtain complexes of Aga50D-EQ with neoagarotetraose and neoagarooctaose. The electron density for both sugars was unambiguous and revealed them to be non-hydrolyzed (supplemental Fig. 2, A and B). The four sugar residues at the nonreducing ends of the two oligosaccharides spanned the proposed catalytic machinery, confirming the location of the active site as the blocked tunnel (Fig. 2, A and B).
The four sugar residues at the non-reducing terminus of the substrates occupied four clear active site subsites; Ϫ2 and ϩ1 subsites accommodate 3,6-anhydro-L-galactose residues, whereas Ϫ1 and ϩ2 subsites recognize D-galactose residues. The sugar residues occupying these subsites contribute a series of potential hydrogen bonds with polar residues in the active site (Fig. 2, A and B). Together with a platform created by a series of aromatic amino acid side chains, the active site provides the ability to specifically recognize the unique twisted conformation of neoagaroligosaccharides, its particular pres- entation of hydrogen bonding groups, and the orientation of the sugar planes.
A notable set of interactions is made in the Ϫ1 subsite. A D-galactose residue and a 3,6-anhydro-L-galactose occupy the Ϫ1 subsite and ϩ1 subsite, respectively, with the ␤-glycosidic bond spanning the catalytic machinery. In both the neoagarotetraose and neoagarooctaose complexes, the D-galactose residue in the Ϫ1 subsite is distorted into a 1 S 3 skew boat conformation ( Fig. 2A), revealing for the first time substrate distortion by a ␤-agarase. This conformation is stabilized by extensive hydrogen bonding interactions with Glu-757, Glu-695, Arg-752, and Asn-533 and an interaction between the C3-C4-C5 plane of the B-face of the galactose ring and the side chain of Phe-742 ( Fig. 2A), providing specificity for a galactose residue in this subsite.
An additional interaction of note occurs in the ϩ2 subsite, where the B-face of the D-galactose aligns with the side chain of Trp-199, forming a classic carbohydrate-aromatic ring interaction. This aromatic side chain is located in the N-terminal CBM-like module. The following 3,6-anhydro-L-galactose residue occupies what may be a ϩ3 subsite, where it makes direct and water-mediated hydrogen bonds with Trp-199, Arg-539, and Thr-142; the latter residue is present in the CBM-like domain. Beyond the potential ϩ3 subsite, few interactions are made, suggesting that these additional residues on the substrate do not make a substantial contribution to substrate recognition.
Overall, the active site is a covered channel of ϳ25-30 Å in length, although the channel is most pronounced over the ϳ15 Å that accommodates the four sugar residues at the non-reducing end of the substrate. The end of the channel that accommodates the non-reducing end 3,6-anhydro-L-galactose is closed by a loop comprising residues 351-379, thus creating a pocketshaped Ϫ2 subsite for the non-reducing end residue (Fig. 2C). This loop also contributes Asp-362 to the active site, which hydrogen-bonds with O4 of the terminal 3,6-anhydro-L-galactose (Fig. 2, A and C). Three loops contributing His-409, Trp-494, Phe-493, and Asn-173, the latter of which is present in the CBM-like domain, form a roof over the ϩ1 and ϩ2 subsites to create a tunnel and completely cover the two sugar residues bound at these positions, respectively (Fig. 2D). Notably, the gap between the blocked end of the active site channel and the roof over the channel leaves a solvent-exposed opening of ϳ7 Å in diameter over the Ϫ2 and Ϫ1 subsites (Fig. 2C).
Product Recognition by Aga50D-X-ray diffraction data from crystals of Aga50D soaked in excess neoagarobiose gave clear electron density for a neoagarobiose molecule bound in the ϩ1 and ϩ2 subsites of the active site (supplemental Fig. 2C). The set of interactions was identical to that seen for occupation of these subsites by neoagarotetraose and neoagarooctaose (Fig.  3). This binding pattern suggests a preference for the product binding the ϩ1 and ϩ2 subsites, which is in contrast to the GH16 ␤-agarases, where products appear to bind with preference to the minus (Ϫ) subsites (7,11).

DISCUSSION
Hydrolysis of the Agarose ␤-Glycosidic Bond by Aga50D-Hydrolysis of the glycosidic bond by glycoside hydrolases proceeds primarily via two mechanisms. One uses a single displacement mechanism that inverts the stereochemistry at the anomeric carbon of the glycon, and the other uses a double displacement mechanism that retains the stereochemistry (30). A large body of structural evidence shows excellent correlation between the arrangement of the catalytic machinery and the type of catalytic mechanism used (25,54). In the case of Aga50D, the distortion of the D-galactose in the Ϫ1 subsite into a 1 S 3 skew boat conformation places the glycosidic bond to be cleaved in a pseudoaxial orientation (Fig. 4A). This in turn places an oxygen of the Glu-695 carboxylate group ϳ3.3 Å beneath C1 and nearly inline with the scissile bond. Gln-534, which in the wild-type enzyme would be a Glu residue, is then positioned ϳ2.9 Å from the oxygen in the glycosidic bond. These residues are conserved in the GH50 family (Fig. 4B). This arrangement and geometry is consistent with a retaining catalytic mechanism whereby Glu-695 acts as the nucleophile that attacks at C1 while Glu-534 is positioned for anti-protonation of the glycosidic oxygen to aid departure of the leaving agarose chain. Indeed, this arrangement of catalytic residues and the conformation of the distorted sugar in the Ϫ1 subsite are very similar to those observed for the retaining GH5 endoglucanase from Bacillus agaradherens (55) (Fig. 4C), providing additional support for the retaining mechanism proposed for Aga50D. This conservation of catalytic machinery and fold between Aga50D and GH5 enzymes is in agreement with the assignment of GH50 enzymes into clan GH-A of the glycoside hydrolase classification (56).
Insights into the exo-Acting Mode of a ␤-Agarase-The outcome of agarose and neoagarooligosaccharides hydrolysis by Aga50D is the production of only neoagarobiose. The structures of Aga50D in complex with substrates show an active site architecture that comprises a channel with one end blocked and partially roofed to create a short tunnel. The two nonreducing end residues of the substrate, which constitute the neoagarobiose product, fit up against the blocked end of the channel to occupy Ϫ1 and Ϫ2 subsites. The lack of additional minus (Ϫ) subsites limits the length of the hydrolysis product to the disaccharide neoagarobiose, thus explaining the exo-acting mode of catalysis. The GH50 family, however, has members that are able to release neoagarotetraose from agarose, suggesting varying active site architectures. The loop blocking the active site channel at the non-reducing end of the substrate comprises residues 351-379 of Aga50D. An alignment with characterized members of GH50 reveals this loop to be the most diverse part of the protein sequence, where in many cases the loop is entirely absent (Fig. 2E). There is no clear correlation between the presence or absence of the loop and whether the enzyme is known to produce neoagarobiose or neoagarotetraose as a product of hydrolysis. Given the structural role of this loop in Aga50D, however, it seems highly likely that the structure in this region of the protein plays a role in determining the length of the product released by the GH50 enzymes. It is possible that other structural elements elsewhere in the protein that cannot be identified on the basis of our single structural representative compensate for the presence or absence of this loop and influence the architecture of the minus (Ϫ) subsites.
Three loops contributed by different regions of the protein (Fig. 2, D and E) form the roof over the ϩ1 and ϩ2 subsites, creating a tunnel of ϳ8 -10 Å in length. These loops were found in the same conformation in the monomers of all of the determined structures and were well ordered, with B-factors approximately the same as those found in residues comprising neighboring secondary structure elements, indicating that the tunnel is a stable structure. Furthermore, the loops comprising the active site roof appear to be relatively well conserved among the characterized GH50 enzymes (Fig. 2E). Davies and Henrissat (25) forwarded the observation that enzymes possessing a tunnel-shaped catalytic groove typically act as processive enzymes. Processivity has been described for several types of glycoside hydrolases, including amylases, ␤-glucanases, cellulases, chitinases, carrragenases, and polygalacturonases (57)(58)(59)(60)(61)(62). The particular topology of such a catalytic site has been shown to be a key factor in the efficiency of microcrystalline polysaccharide degradation (63)(64)(65). In their study of Aga50D, Kim et al. (24) described the enzyme as probably being processive on the basis of residual substrate analysis.
To achieve this, the agarose chain would thread into the tunnelshaped active site, followed by hydrolysis to generate the neoagarobiose product in subsites Ϫ1 and Ϫ2. There are two possibilities for the release of this product. The first is that the shortened agarose chain dissociates, allowing the neoagarobiose product to leave the active site through the substrate's entry point. A second possibility, however, is that the neoagarobiose product leaves the active site via the opening over the Ϫ1 and Ϫ2 subsites, allowing the shortened agarose chain to remain bound to the enzyme. Indeed, the complex of Aga50D with neoagarobiose, which occupied only the ϩ1 and ϩ2 subsites, suggests significant binding affinity at these subsites, consistent with the possibility that an agarose chain may remain bound to the enzyme after hydrolysis, promoting a processive mode of agarose degradation.
The Role of Domain Architecture-The CBM-like domain is structurally very similar to family 11 and 29 CBMs. Although these families of CBMs are known to have independent binding functions to glucans and mannans, this is not the case for the Aga50D CBM-like domain. Indeed, the CBM-like domain of Aga50D appears to have evolved, perhaps from an ancestral CBM, to be fused with and extend the catalytic site of Aga50D to contribute to forming the ϩ2 subsite and a possible ϩ3 subsite (Fig. 2B). Additionally, a loop in the CBM-like domain is involved in forming the roof of the active site channel. The contribution of the CBM-like domain to formation of the active site of the enzyme supports a role in substrate recognition. This is not a unique observation for a glycoside hydrolase, but it is highly rare. The first example of a CBM contributing to the formation of a glycoside hydrolase active site was observed in GH9 endoglucanases, where a family 3c CBM, which is rigidly fused to the catalytic module, extends the active site along the binding face of the CBM (66,67). A more recent example is seen in the glycogen-degrading enzyme SpuA from Streptococcus pneumoniae, where the binding site of the CBM contributes to the formation of two additional active site subsites, allowing the recognition of glycogen branch points (68).
Quaternary Structure and Agarose Recognition-In this study, Aga50D was crystallized in two different crystal forms, FIGURE 4. The catalytic mechanism of Aga50D. A, neoagarotetraose spanning the catalytic machinery of Aga50D. The gray mesh shows the electron density map as a maximum likelihood/ a -weighted F o Ϫ F c map contoured at 3 (0.17e/Å 3 ). The putative nucleophile (Glu-695 (E695)) and the putative acid/base (Glu-534 (E534), which has been mutated to a glutamine) are shown as green sticks. Relevant distances are shown as dashed lines and labeled. B, conservation of the putative catalytic residues in known GH50 ␤-agarases. The nucleophile is indicated above the alignment with a blue circle, and the acid/base is indicated with a red circle. C, structural overlap of the Aga50D neoagarotetraose complex with the Michaelis complex of Cel5A from B. agaradherens with 2,4-dinitrophenyl 2-deoxy-2-fluoro-␤-cellobioside (PDB code 4A3H (77)). The substrate and catalytic residues of Cel5A are shown as blue sticks. Cel5A catalytic residues are labeled in blue. P4 1 and P2 1 2 1 2 1 , with four and two molecules in the asymmetric units, respectively. The two monomers in the P2 1 2 1 2 1 crystal form are arranged as a dimer with a 2-fold non-crystallographic symmetry (Fig. 5A). The four monomers in the P4 1 crystal form make identical dimers, although these are generated by crystallographic symmetry (not shown). This packing suggests a propensity, at least in crystals, for this enzyme to adopt a consistent dimer structure. This becomes potentially relevant when examining the dimer present in the neoagarooctaose complex. Agarose chains naturally adopt a right-handed double helix. In the Aga50D-neoagarooctaose complex, the reducing end of the neoagarooctaose, which does not specifically interact with the protein, starts to adopt a conformation consistent with a helical structure (Fig. 5A). In the context of the Aga50D dimer, the reducing ends of the neoagarooctaose come in proximity (17 Å separation) in an overall arrangement that is remarkably reminiscent of the initiation of a right-handed double helix. Indeed, these termini require only ϳ3 additional residues to wind into the termini of an idealized agarose double helix (Fig.  4B) (PDB code 1AGA (6)). Although the Aga50D dimer is clearly driven to form in crystals, the interaction interface between the monomers in the Aga50D dimer is not extensive at only 550 Å 2 , suggesting a weak interaction. Consistent with this weak interaction, we were unable to observe significant dimer formation by dynamic light scattering at protein concentrations up to 10 mg⅐ml Ϫ1 . The interaction of free Aga50D monomers with the non-reducing ends of an agarose helix, however, would potentially bring the monomers in proximity, thus creating a pseudo-unimolecular interaction and therefore driving dimer formation by this recruitment. Overall, this evokes a model whereby recruited Aga50D dimers unwind the helices of agarose or agarose fragments as the non-reducing ends are shortened by neoagarobiose units, a model that is made more attractive by the potential processivity of the enzyme.
A potential difficulty with the model is that the neoagarobiose units are rotated by ϳ180º relative to one another around the long axis of the agarose chain. Thus, threading into the active site would require a 180º rotation of the chain end after each round of hydrolysis. Either the chain has to flip in the active site or the enzyme must flip, the latter being unlikely in the proposed model. It is plausible, however, that as the agarose helix is unwound, this enables the chain to flip ϳ180 o and thread into the active site. The proposed model is suggestive of that put forward for the GH16 Agarase A from Z. galactanivorans. This enzyme possesses a non-catalytic binding site on the side of the protein, which is proposed to work in concert with the catalytic site on the same monomer by wedging the strands of agarose helices apart to unwind them and perhaps promote catalysis (7). A potentially important difference in our proposed model is that GH50 enzymes are proposed to act on the long fragments of agarose generated by endo-acting agarases rather than the agarose chains in gels (69) (supplemental Fig. 3). Nonetheless, our tentatively proposed model of Aga50D action on agarose helices remains to be examined in more detail.
Conclusion-The exo-␤-agarase, or neoagarobiohydrolase, Aga50D adopts a relatively complex fold comprising an elaborated (␣/␤) 8 barrel and a CBM-like ␤-sandwich domain. Loop structures and the involvement of the CBM-like domain create a partially covered active site channel that is blocked at one end, explaining the exo-mode of ␤-agarase action. The catalytic machinery, which is present in the active site tunnel, has an arrangement that is most consistent with retaining a catalytic mechanism, whereas the substrate adopts a distorted conformation that is in keeping with other GH-A clan members, such as the family 5 endoglucanase from B. agaradherens. The architecture of the Aga50D active site combined with possible quaternary structure is suggestive of a potentially complex processive mode of action. Within the agarolytic system, Aga50D occupies a critical function. Whereas endo-acting ␤-agarases, such as GH16 and GH86, cleave randomly along the agarose polymer chains (9,21), Aga50D appears as a key component to degrade those neoagarooligosaccharides into neoagarobiose units. The disaccharide units produced are then further degraded by ␣-agarases, such as GH117 (17), into monosaccharides useable as carbon and energy sources by the bacterium.
The data presented here provide new molecular insight into this important step of agarose degradation and therefore an increased understanding of how bacteria degrade seaweed polysaccharides, insight that is much needed for a refined understanding of the ocean carbon cycle.