The Three-dimensional Structures of Two (cid:1) -Agarases*

Agars are important gelifying agents for biochemical use and the food industry. To cleave the (cid:1) -1,4-linkages between (cid:1) - D -galactose and (cid:2) - L -3,6-anhydro-galactose residues in the red algal galactans known as agars, marine bacteria produce polysaccharide hydrolases called (cid:1) -agarases. (cid:1) -Agarases A and B from Zobellia galactanivorans Dsij have recently been biochemically characterized. Here we report the first crystal structure of these two (cid:1) -agarases. The two proteins were overpro-duced in Escherichia coli and crystallized, and the crystal structures were determined at 1.48 and 2.3 Å for (cid:1) -agarases A and B, respectively. The structure of (cid:1) -aga-rase A was solved by the multiple anomalous diffraction method, whereas (cid:1)

1a). On the basis of x-ray fiber diffraction, optical rotation calculations, and solution gel transition, both a parallel double helix as well as a single helix structure for agarose have been proposed, where the individual polysaccharide chains have a left-handed 3-fold helix symmetry and a pitch of 1.90 nm (3,4). Agarose forms thermo-reversible gels structured by aggregates of agarose chains. These gels exhibit unique rheological properties and are widely used as texturing agents for various applications in the food industry or as a common laboratory medium for chromatographic separation or bacterial colony growth (5).
Although the sequence-based GH families contain enzymes of similar fold and identical molecular mechanism (16 -18), they are known to often group together enzymes of varying substrate specificity (7)(8)(9). Interestingly, family GH-16 is such * This work was supported by the region "Provence-Alpes-Côte d'Azur" (France), Goëmar, St. Malo (France), and the CNRS (France). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The atomic coordinates and structure factors ( a "polyspecific" family and contains enzymes hydrolyzing widely different substrates such as keratan-sulfate, ␤-1,3-glucan, mixed linkage ␤-1,3 (4)-glucans, xyloglucan, -carrageenan, and agarose. These enzymes share a common ancestor and have diverged significantly in their primary sequence (17,19). Pairwise comparisons of the sequences of family GH-16 reveals the existence of several subfamilies that coincide with the experimentally determined substrate specificity of the enzymes. Indeed five such subfamilies can be identified: (i) ␤-agarases, (ii) endo-1,3-␤-glucanases (laminarinases) (iii), endo-␤-1,3-1,4-glucanases (lichenases), (iv) -carrageenases, and (v) xyloglucan endotransferases. Several additional subfamilies can be detected, but the substrate specificity of their members is not known at present. The few endo-␤-galactosidases that have been characterized have too few homologues to allow the building of any subfamily. Although the overall sequence identity between members of a subfamily is typically better than 30 -35%, inter-subfamily sequence identity is low (10 -25%), and the similarity is apparently restricted to a few invariant residues that include the catalytic machinery. As a consequence of this conservation, all the glycoside hydrolases of family GH-16, including the two Z. galactanivorans Dsij ␤-agarases (␤-agarases A and B), proceed with a molecular mechanism leading to overall retention of the anomeric configuration (20) and are hence capable of transglycosylating activity (18). This activity is actually the dominant feature of the family GH-16 enzymes that act on xyloglucan (xyloglucan endotransferases). The low degree of sequence similarity observed between the subfamilies within family GH-16 prevents successful homology modeling of an enzyme based on the structure of an enzyme with a different substrate specificity. Instead, and despite a predictable overall fold similarity, structural determinations are required to understand the details of substrate recognition and specificity in each subfamily. Three-dimensional structures have already been solved for the ␤-1,3-1,4glucanase (20 -23) and for the -carrageenase (24) subfamilies. Structural data representative of xyloglucan endotransferases will soon be available, because the crystallization of this type of enzyme from Populos tremula ϫ tremuloides has been reported recently (25). Because there is currently no three-dimensional structure available for any agarase, we have started a program aiming at the structural resolution of the two Z. galactanivorans Dsij ␤-agarases. The structural comparison between both enzymes as well as with the other GH-16 enzymes with known structures is expected to help define the factors governing substrate specificity within this class of enzymes.
Sequence analysis shows that the Z. galactanivorans Dsij ␤-agarase A gene (agaA; GenBank TM accession number AX008608) encodes for a modular protein comprising a signal peptide of 19 amino acids, the GH-16 catalytic module (residues 20 -295), and a C-terminal domain (residues 296 -539) of unknown function. In contrast to ␤-agarase A, the ␤-agarase B gene (agaB; GenBank TM accession number AX008610) encodes a single module protein consisting of 334 amino acids (residues 19 -353). The first 18 amino acids are supposed to be a signal peptide of type II where residue Cys-18 might be linked to a lipid moiety (Fig. 1b). To avoid the known problems associated with the crystallization of multimodular proteins, we expressed only the catalytic modules, referred to below as ␤-AgaA_CM (275 residues) and ␤-AgaB (334 residues), respectively.
We have determined the structures of ␤-AgaA_CM and ␤-AgaB by the MAD method at 1.48 Å resolution and by molecular replacement at 2.3 Å resolution, respectively. These three-dimensional structures help to define the amino acid residues involved in the recognition and cleavage of agarose and to understand the determinants of substrate specificity by comparison with the known structures of Pseudoalteromonas carrageenovora -carrageenase and the Bacillus licheniformis and Paenibacillus macerans ␤-1,3-1,4-glucanases. In addition, the number of sugar-binding subsites in both ␤-agarases was determined biochemically by high performance anion exchange chromatography analysis of the reaction products using different starting oligosaccharides.
Full details of the expression and purification of native ␤-AgaA_CM and ␤-AgaB have also been reported by Flament et al. 2 Briefly, ␤-AgaA_CM and ␤-AgaB were expressed using the pET20b vector (Novagen) as a His-tagged fusion protein in the periplasm of E. coli ORIGAMI (DE3) pLysS strain and purified first by metal affinity chromatography on a column of Chelating Fast Flow-Sepharose loaded with NiSO 4 by using a concentration gradient from 30 to 400 mM imidazole. A second purification step by size exclusion chromatography on S100-Sephacryl (Amersham Biosciences) by using 20 mM Tris, pH 7.5, 200 mM NaCl yielded an electrophoretically pure protein (data not shown). The yield was 2 mg/liter culture medium, and the purified protein was concentrated using a dialyzing concentrator equipped with a polyethersulfone membrane (10 kDa cut-off) (Amicon).
Crystallization-␤-AgaA_CM and ␤-AgaB were concentrated to 2.5 mg/ml and stored in 50 mM Tris buffer, pH 7.5, with 25 mM NaCl. Crystallization conditions were first investigated using two sparsematrix sampling kits (Molecular Dimensions and Stura Footprint).
The crystallization solutions contained 28% PEG 8000, 200 mM ammonium sulfate, and 100 mM sodium acetate, pH 4.6, for ␤-AgaA_CM and 58% methyl pentanediol, 20 mM calcium chloride, and 100 mM Hepes, pH 7.5, for ␤-AgaB. For both enzymes, crystals were grown by mixing 2 volumes of protein with 1 volume of precipitant solution in a hanging-drop vapor diffusion set up at 20.5°C. Crystals grew within 1 week. Thereafter, a single crystal was soaked for 30 s in successive a Values in parentheses correspond to the highest resolution shell. b R sym ϭ ⌺ ͉IϪI av ͉/⌺I, where the summation is over all symmetry equivalent reflections. c R ano ϭ ⌺ ͉I(ϩ) Ϫ I (Ϫ)͉/⌺ ͉I av ͉ for anomalous differences, where I av is the average of Friedel amplitudes at a single wavelength. d R iso ϭ ⌺ ͉I j Ϫ I j ͉/⌺ ͉I av ͉ for dispersive differences, where I av is the average amplitude at two wavelengths. e FOM, figure of merit. f R calculated on 5% of data excluded from refinement. g r.m.s.d., root mean square deviation.
solutions in which the glycerol concentration was increased by steps of 5% to a final concentration of 10% for ␤-AgaA_CM. For ␤-AgaB, the crystal was soaked 30 s in a 5% glycerol solution. All crystals were mounted on a loop, transferred to the goniometer head, and kept at 100 K in a nitrogen stream. Data Collection and Phasing-Diffraction data at resolutions of 1.48 and 2.3 Å were collected on beamlines ID14-EH1 and EH2 (ESRF, Grenoble, France) for ␤-AgaA_CM and ␤-AgaB, respectively.
The structure was solved by the multiple anomalous diffraction (MAD) method (28) using ytterbium as anomalous diffuser. Ytterbium ions were incorporated by soaking with a mother solution containing 0.33 mM ytterbium chloride (exposure time 1 h and 15 min). The crystal was then back-soaked in the above-mentioned cryo-solution. The data were collected at three wavelengths around the LIII absorption edge of ytterbium (beamline ID29, ESRF, Grenoble, France). An ADSC image plate detector was used. All intensity data were integrated and reduced using MOSFLM/SCALA (29). The position of one ytterbium atom and the phase problem were solved using SOLVE with an overall Z score of 4.2 (30). The experimental phases from SOLVE resulted in a figure of merit of 0.67 at 2.5 Å resolution.
The electron density was improved by solvent flattening with the program DM (31). By using the native data, the phases were then extended to 1.48 Å resolution yielding a figure of merit of 0.91.
For ␤-AgaB, the space group was identified by autoindexing as P2 1 with cell constants a ϭ 72.4 Å, b ϭ 105.3 Å, c ϭ 97.4 Å, and ␤ ϭ 94.1°, suggesting a V m value of 2.3 Å 3 /Da with four molecules in the asymmetric unit (26). The data were processed with DENZO (27). The structure was solved by molecular replacement using AMoRe (31, 32) and ␤-AgaA_CM as models. Molecular replacement gave four solutions corresponding to four molecules in the asymmetric unit. Therefore, the program DM was used for solvent flattening and non-crystallographic symmetry averaging leading to a final figure of merit of 0.83 at 2.3 Å resolution.
Both ␤-AgaA_CM and ␤-AgaB were built manually using TURBO (33). The refinement was performed with REFMAC 5 (31). Water molecules were added using ARPwARP (34). The stereochemistry of the final structures was evaluated using PROCHECK (35). The structural superimposition between the different structural members of family GH-16 was performed with TURBO.
Digestion of Calibrated Oligo-agarose-Agarose (1% w/v, Invitrogen) was melted in 100 ml of boiling deionized water. The solution was cooled down to 40°C and maintained at this temperature to prevent gelation of the polysaccharide. Agarose was incubated with 500 l of overexpressed ␤-AgaB (1 nmol/ml) at 40°C for 3 h. Then the reaction was stopped by boiling the incubation medium for 15 min. After cooling, the hydrolysate was freeze-dried and stored at 4°C.
The method applied to the purification of calibrated oligo-agaroses was adapted from Rochas et al. (4). 1 ml of 1.5% oligo-agarose mixture in deionized water was injected on a column (100 ϫ 5 cm) of Bio-Gel P2 (Bio-Rad) and was eluted with deionized water at the flow rate of 35 ml/min. Fractionation was monitored with a computer running AZUR 3.0 software (Datalys) and connected to a SpectraSystem RI 150 refractometer. The most homogeneous fractions collected were directly subjected to enzymatic assays without further treatment. 200 l of purified oligo-agarose was incubated at 40°C with 50 l of purified ␤-AgaA_CM (70 pmol/ml) or ␤-AgaB (1.4 pmol/ml). The degradation products were analyzed by high performance anion exchange chromatography (HPAEC, Dionex) using pulsed amperometric detection. Samples (50 l) were injected on a CarboPac PA 100 column (4 ϫ 250 mm, Dionex) and eluted with a gradient starting from 150 mM NaOH which reached a final concentration of 150 mM NaOH, 500 mM NaOAc according to the method described by Weinberger et al. (36). DP refers to the degree of polymerization of the disaccharides repeat unit of agarose. For example, DP1, DP2, and DP3 are disaccharide, tetrasaccharides, and hexasaccharides, respectively. The DP of oligosaccharides was determined using calibrated neo-oligoagarose provided by Dextra Laboratories (Reading, UK).

RESULTS AND DISCUSSION
Overall Structure of ␤-AgaA_CM and ␤-AgaB-The threedimensional crystal structure of ␤-AgaA_CM was determined at 1.8 Å by the MAD method, using a native crystal soaked in an ytterbium solution. The resulting experimental electron density map was of high quality (Fig. 2). The model was then refined at 1.48 Å resolution to a crystallographic R factor of 14.9% and R free factor of 17.6%. The three-dimensional crystal structure of ␤-AgaB was determined at 2.3 Å by molecular replacement with AMoRe using ␤-AgaA_CM as the search model and was then refined to a crystallographic R factor of 19.9% and R free factor of 22.4%. Crystallographic statistics for both ␤-agarases are reported in Table I. In ␤-AgaA_CM, the asymmetric unit encompasses amino acids 20 -289. Residue Ser-290 is not visible in the (2F o Ϫ F c ) electron density map and is thus presumed to be in a disordered conformation. The high resolution electron density allowed the detection of a sequence discrepancy (Ala-84 is in fact a proline). The triplet CCG coding for a proline differs by a single nucleotide from the triplet GCG leading to an alanine, which easily accounts for an error that might have occurred during the sequence determination. In ␤-AgaB, the asymmetric unit encompasses amino acids 59 -353. Residues 19 -58 are not visible in the (2F o Ϫ F c ) electron density map and are probably either truncated or in a disordered conformation.
As expected, the overall fold of both agarases is similar to the other members of family GH-16 with known structures (20 -22, 24). Both ␤-agarases fold into globular ellipsoids, 58 ϫ 47 ϫ 42 Å and 56 ϫ 47 ϫ 40 Å in dimensions for ␤-AgaA_CM and ␤-AgaB, respectively, and consist almost exclusively of ␤-sheets and surface loops (Fig. 3a). Each sheet is formed by seven ␤-strands. ␤-AgaB has three supplementary loops compared with ␤-AgaA_CM; two of these loops are located on the same side, connecting strands at the C-terminal extremity of the internal ␤-sheet, and are close in space; the third is on the opposite side connecting strands at the N-terminal end of the external ␤-sheet. The molecules have a slightly elongated shape with a prominent cleft crossing one side. This cleft originates from the twist and bending of the two ␤-sheets, arranged on top of each other in a sandwich-like manner, and from the loop regions as already described for lichenase (37). The cleft is at least 35 Å long for both ␤-AgaA_CM and ␤-AgaB and is equally lined with aromatic residues, listed in Table II.
Structural Comparison with Family GH-16 Enzymes- Fig. 4 shows the amino acid sequence alignment of ␤-AgaA_CM and ␤-AgaB with the -carrageenase of P. carrageenovora (PDB code 1DYP) and the ␤-1,3-1,4-glucanases of B. licheniformis (PDB code 1GBG) and P. macerans (PDB code 1MAC). Despite the low sequence similarity of agarases with other members of family GH-16, most of the amino acids comprising the basic secondary structures of the ␤-jelly roll motif are well conserved. Consequently, the core region of the ␤-sandwich is perfectly superimposable, reflected by the lower root mean square deviations for the C␣ atoms of the ␤-sandwich. The overall root mean square deviations of all structurally known members of family GH-16 enzymes are summarized in Table III.
In the two ␤-agarases and in the -carrageenase, strands ␤11 and ␤12 are separated by a short loop insertion. In the ␤-1,3-  (labeled as k-car[Pc]) and lichenases of B. licheniformis and P. macerans  (labeled by lich1[bl] and lich2[Pm], respectively). The alignment of known structures of family GH-16 was prepared based on the superimposed models. ␤-Strands are shown as black arrows, and loops are shown as red arrows. Helices are shown as black rectangles when they are found in all structures and as red rectangles when they are only found in ␤-agaA_CM and ␤-agaB. Red circles highlight the strictly conserved catalytic residues and black circles the residues involved in binding of calcium. The figure was drawn using the program ESPript 2.0 (46). anhydro-galactose units. This should have a global consequence on the form of the binding cleft, because the 3,6-anhydro-galactose units whether in the L-or D-galacto configuration are much more bulky than regular pyranose units, and consequently stacking interactions are only possible on one face of the sugar unit. The largest differences with ␤-1,3-1,4-glucanases are therefore expected in the Ϫ2 and Ϫ4 subsites on the glycone-binding side and at the ϩ1 and ϩ3 aglycone-binding subsites, following the subsite naming convention of Davies et al. (38). Indeed, the inserted loops mentioned above in both agarases and the -carrageenase give different forms to the cleft at the level of the extremities of the binding subsites, which appear to be less linear in the cases of the ␤-agarases (Fig. 3b). Differences between agarases and the other members of family GH-16 at the Ϫ1 subsite should only concern the O-4 position and are discussed below under "Substrate-binding Region." Two short helices (␣3 and ␣4), only present in the agarases, are located in loops on the side of the external ␤-sheet and are therefore not responsible for differences in substrate specificity.
In ␤-AgaA_CM, a calcium ion is bound with nearly perfect octahedral geometry on the convex face of the protein, coordinating to the backbone carbonyl oxygen atoms of Ser-47, Ser-91, and Asp-279, a carboxylate oxygen of Asp-279, one of Asn-49, and one of Asp-22, and one water molecule. In ␤-AgaB, a calcium ion is located at the same position but coordinated in a pentahedral manner to the backbone carbonyl oxygen atom of Asn-83, Gly-127, and Asp-343, a carboxylate oxygen of Asp-343, and one water molecule. It is interesting to note that this structural calcium ion is also found in the other GH-16 mem-bers with known three-dimensional structures (lichenases and -carrageenase) (24,37) and that the aspartic acid responsible for the only side-chain interaction with calcium is conserved (Fig. 4).
Catalytic Machinery-Glycosidic bond cleavage by retaining glycosidases such as those in family GH-16 takes place via a two-step mechanism involving two enzymatic carboxyl groups (39 -41). In the first step (glycosylation), one of the carboxyl group acts first as a general acid, which protonates the glycosidic oxygen and thereby facilitates aglycone departure. Simultaneously, the other catalytic residue acts as a nucleophile, attacking the anomeric carbon and forming a covalent glycosylenzyme intermediate. In the second step (deglycosylation), the general acid now acts as a general base and deprotonates a water molecule, in order to hydrolyze the glycosyl-enzyme intermediate and to release the product. Both steps occur via transition states with substantial oxocarbenium ion character (40,41). In family GH-16 glycoside hydrolases, the catalytic nucleophile and acid/base residues have been unambiguously identified in the P. macerans and B. licheniformis 1,3-1,4-␤glucanases (21,22) and have been confirmed in the P. carrageenovora -carrageenase (24). Sequence and structure similarities between members of family GH-16 allowed the rapid identification of the catalytic machinery of the two ␤-agarases on strand ␤9 with Glu-147 and Glu-184 as the nucleophile and Glu-152 and Glu-189 as the acid/base in ␤-AgaA_CM and ␤-AgaB, respectively. These two residues are found in the deepest region of the cleft of both enzymes. A third conserved acidic residue (Asp-149 and Asp-186, respectively, in ␤-AgaA_CM and ␤-AgaB) points toward the cleft and is probably important to maintain the charges in the environment of the catalytic amino-acids.
In -carrageenase, oxygen OD2 of Asp-165 makes a 2.74-Ålong hydrogen bond to His-183. Michel et al. (24), have suggested that Asp-165 and His-183 might be implicated in proton traffic, governing the regeneration of the protonated states of the catalytic residues during catalysis and cooperating in proton trafficking during the deglycosylation step. This histidine is conserved in both agarases; however, the positioning is not the same as in the -carrageenase. In ␤-AgaB, ND1 of His-215 is at 3.5 Å from Asp-186, whereas the high resolution (1.48 Å) of ␤-AgaA_CM shows clearly that the side-chain of His-172 has an opposite orientation, and its ND1 atom is hydrogen-bonded to OD1 of Asp-181 (2.68 Å) and not to Asp-149. In the lichenases this histidine residue is replaced by a tyrosine, which is not involved in hydrogen-bonding with the catalytic residues.
Substrate-binding Region-In the absence of a complex with a substrate molecule, it is difficult to analyze the elements governing substrate binding. No complex is available either with the other GH-16 enzymes with known three-dimensional structures. However, it has been shown that family GH-16 and family GH-7 are structurally and mechanistically related despite the absence of detectable sequence similarity outside of a common sequence motif, E(ILV)D(IVAF)(VILMF) (0,1) E. The  structural similarity between the two families, accompanied by a perfect correspondence of the catalytic machinery, has allowed their grouping in a superfamily called "clan GH-B" (22,24). We have taken advantage of the availability of the threedimensional structure of Fusarium oxysporum endoglucanase I of family GH-7 complexed with a nonhydrolysable substrate analogue (PDB code 1OVW) (42) in order to infer by similarity the approximate location and direction of the agarose substrate in the active site of ␤-agarases. After superimposition of the catalytic residues and of the secondary structure elements, the cleft on the concave side of both ␤-agarases A and B has a position similar to that of F. oxysporum endoglucanase (42). From this superimposition, some differences between the family GH-16 members may be highlighted at the level of subsites Ϫ1 and ϩ1. As mentioned above, the only difference in the ␤-linked pyranose unit that binds the Ϫ1 subsite of the lichenases, -carrageenase, and ␤-agarases is the O-4 position. In the substrate of lichenases, this unit is a glucose residue where the OH-4 group is equatorial. In contrast this group is axial in the galactose unit of agarose and axial and sulfated in -carrageenan. Indeed, corresponding differences are seen with respect to the residues of the different enzymes that are located in the area involved in potential hydrogen bonds to O-4. In both ␤-agarases a glutamic acid (308 in ␤-AgaB and 254 in ␤-AgaA_CM) and a glutamine (310 in ␤-AgaB and 256 in ␤-AgaA_CM) are candidates for hydrogen bonding with the axial O-4 group of the sugar unit in subsite Ϫ1. In the -carrageenase these residues are replaced by Gly-258 and Arg-260, the latter clearly being in favor of binding to a sulfate group, whereas the former is probably necessary to create room to accommodate the sulfate group. In the lichenases, the glutamic acid (Glu-254 in ␤-AgaA_CM) is replaced by a tryptophan residue, which could form a hydrogen bond to the equatorial O-4 of the glucose unit that comes to lie in this subsite. In the place of Gln-256 of ␤-AgaA_CM (Gln-310 in ␤-AgaB) the lichenases possess a phenylalanine (Phe-30 in B. licheniformis lichenase (1GBG)), which comes from a completely different structural element than in both the agarases or the -carrageenase. Interestingly, the alignment of lichenases and xyloglucan endotransferases, which are all members of family GH-16 that degrade polysaccharides having the OH-4 group in equatorial position, reveal that this tryptophan is strictly conserved within these members. The position of Phe-30 in B. licheniformis lichenase is less conserved but is most often a tyrosine or a phenylalanine and in all cases a hydrophobic residue in most lichenases and xyloglucan endotransferases. As expected, the Differences in the ϩ1 subsite are more difficult to infer, because this subsite accommodates the sugar unit that undergoes the largest positional shift, caused by the distortion of the sugar unit in the Ϫ1 subsite on the catalytic pathway (42,43). It is worth mentioning that His-172 in ␤-AgaA_CM (His-215 in ␤-AgaB and His-183 in -carrageenase) might play a role in binding of the axial OH-2 group of a 3,6-anhydro-galactose unit in subsite ϩ1. A second histidine (His-168 and His-211 in ␤-agarases A and B respectively), unique to the ␤-agarases, is situated on the opposite side of the cleft at the level of the putative ϩ1 subsite and is therefore a candidate for specific interaction with the ␣-L-3,6-anhydro-galactose unit of agarose (Fig. 6).
The -carrageenase active site cleft contains several positively charged residues (Arg-151, Lys-172, Lys-191, and Arg-196, numbering of -carrageenase) important for the recognition of its sulfated substrate (24). These residues are frequently replaced by negatively charged or aromatic residues in the cleft of the ␤-agarases, in line with the requirement to interact with agarose. The assignment and inspection of the aromatic residues that line the cleft of the two ␤-agarases (Table II) suggest that it may contain eight subsites. In order to determine the number of subsites in the two agarases, we have carried out an analysis of the reaction products of ␤-agarases A and B by using oligosaccharides of defined degree of polymerization as substrates and high performance anion exchange chromatography detection. The degradation products observed after incubation of calibrated oligo-agaroses either with ␤-agarase A or B were similar. The apparent rate of hydrolysis was found to be similar when the concentration of ␤-AgaA_CM was 50 times higher than that of ␤-AgaB (Fig. 7a). This lower specific activity of ␤-AgaA_CM toward oligosaccharides was more pronounced in the case of the hexasaccharide neoagarohexaose. When this oligosaccharide was subjected to enzymatic assays, the amount of neoagarotetraose detected was about three times lower for ␤-AgaA_CM than for ␤-AgaB (Fig. 7b). Neoagarooctaose was fragmented only in tetrasaccharides, whereas neoagarodecaose was disrupted in tetra-and hexasaccharides (Fig.  7c). This degradation pattern indicates that the active site can accommodate a maximum of eight sugar units, in agreement with the structural analysis above. Both enzymes degrade agarose oligosaccharides comprising at least six sugars to the same final length. However, ␤-AgaA_CM preferably degrades longer oligosaccharides and is overall less efficient than ␤-AgaB. The difference between both enzymes is kinetic and not in the release of different reaction products. We have found only a single amino acid difference between the two enzymes in the close vicinity of the catalytic residues. At subsite Ϫ1, the aspartic acid 173 in ␤-AgaB is replaced by Ala-136 in ␤-agaA_CM. It is conceivable that this change could alter slightly the pK a of the general acid/base, thereby explaining the subtle difference in catalytic activity between the two enzymes.
Our structural and biochemical data clearly show that ␤-agarases A and B from Z. galactanivorans Dsij are endo-agarases, having a long and open active site cleft that accommodates at least eight sugar units. The comparison with the other members in family GH-16 with known structures allowed us to identify the residues that are most probably involved in substrate specificity. More precise details of agarose recognition obviously require the production of a catalytically inactive mutant and the co-crystallization with a substrate molecule. This work is currently underway in our laboratories.