Structural and Biochemical Analyses of Glycoside Hydrolase Families 5 and 26 β-(1,4)-Mannanases from Podospora anserina Reveal Differences upon Manno-oligosaccharide Catalysis*

Background: Fungal mannanases contribute to enzymatic degradation of lignocellulose. Results: New fungal mannanases reveal striking differences in substrate specificities. A rigid linker tightly connects the family 26 glycoside hydrolase to its binding module. Conclusion: Podospora anserina mannanases display differences in substrate binding modes, transglycosylation activity, and modular organization. Significance: Information on the structure-function relationships of fungal mannanases is essential to improve the comprehension of biomass deconstruction. The microbial deconstruction of the plant cell wall is a key biological process that is of increasing importance with the development of a sustainable biofuel industry. The glycoside hydrolase families GH5 (PaMan5A) and GH26 (PaMan26A) endo-β-1,4-mannanases from the coprophilic ascomycete Podospora anserina contribute to the enzymatic degradation of lignocellulosic biomass. In this study, P. anserina mannanases were further subjected to detailed comparative analysis of their substrate specificities, active site organization, and transglycosylation capacity. Although PaMan5A displays a classical mode of action, PaMan26A revealed an atypical hydrolysis pattern with the release of mannotetraose and mannose from mannopentaose resulting from a predominant binding mode involving the −4 subsite. The crystal structures of PaMan5A and PaMan26A were solved at 1.4 and 2.85 Å resolution, respectively. Analysis of the PaMan26A structure supported strong interaction with substrate at the −4 subsite mediated by two aromatic residues Trp-244 and Trp-245. The PaMan26A structure appended to its family 35 carbohydrate binding module revealed a short and proline-rich rigid linker that anchored together the catalytic and the binding modules.

Several types of glycoside hydrolases (GH) 2 are required for complete degradation of mannans, and endo-␤-1,4-mannanases are the key enzymes. In the CAZy database (2), ␤-1,4mannanase activities are found in families GH5, GH26, and GH113. The three families belong to clan GH-A; they share the same (␤/␣) 8 -barrel protein fold, catalytic machinery, and retaining double displacement mechanism (3)(4)(5). Because of this retaining double displacement mechanism, some of these enzymes are able to perform transglycosylation in which a carbohydrate hydroxyl group can act as an acceptor molecule rather than water as is the case in hydrolysis. Transglycosylation thus leads to the synthesis of new glycosides or oligosaccharides longer than the original substrate. GH5 and GH113 mannanases have been described as able to catalyze transglycosylation reactions (6 -9), whereas to date no evidence of transglycosylation has been reported for GH26 mannanases (10). ␤-Mannanases are frequently encountered as modular enzymes. Indeed, some harbor carbohydrate binding modules (CBMs) from families CBM1, CBM6, CBM10, CBM31, and CBM35 (11,12). It is generally observed that the linker regions between catalytic module and CBM display a great deal of structural flexibility to maximize substrate accessibility, as has been confirmed by the few crystal structures of bacterial modular enzymes (13,14).
The characterization of endo-␤-1,4-mannanases biochemical properties and substrate specificities revealed that many release essentially mannobiose and mannotriose as end products (9,17,24,25) and that their active site displays generally 5-6 subsites able to accommodate the substrate (10,15). Although GH5 and GH26 mannanases share some characteristics, several studies revealed different modes of action. In particular, biochemical studies pointed to divergence in specificity between GH5 and GH26 bacterial mannanases, a suggesting different biological role (20,26).
The coprophilic fungus Podospora anserina has one of the largest fungal sets of candidate enzymes for cellulose and hemicellulose degradation described to date and one of the highest numbers of CBMs of all the fungal genomes available (27). In a previous study comparative genomics were used that identified two mannanases from families GH5 (PaMan5A) and GH26 (PaMan26A) in the P. anserina genome (28). Investigation of the contribution made by each P. anserina mannanase to the saccharification of spruce demonstrated that they individually supplemented the secretome of the industrial T. reesei CL847 strain. The most striking effect was obtained with PaMan5A that improved the release of total sugars by 28% and of glucose by 18% (28). In the present study P. anserina GH5 and GH26 mannanases were subjected to detailed comparative analysis of their substrate specificities, active site organization, and transglycosylation capacity. The three-dimensional structures of PaMan5A and PaMan26A linked to a CBM35 module were solved in their native form at 1.4 and 2.85 Å resolution, respectively.

EXPERIMENTAL PROCEDURES
Production and Purification of PaMan5A and PaMan26A-PaMan5A and PaMan26A were produced in P. pastoris 2-liter cultures and purified as described previously in (28). Enzyme purification was completed by an additional size exclusion chromatographic step. After the nickel chelate purification step, the eluate containing PaMan5A or PaMan26A was con-centrated using a Vivaspin with 10-kDa cut-off polyethersulfone membrane (Sartorius, Palaiseau, France) and dialyzed against the buffer used for the size exclusion chromatography (20 mM Hepes, pH 7.5, 150 mM NaCl). The concentrated fraction was subsequently loaded onto a Superdex S200 HiLoad 16/60 column (Amersham Biosciences). The fractions containing PaMan5A or PaMan26A were pooled and concentrated as described above.
Construction of Site-specific Variants-Site-directed mutagenesis was performed using the QuikChange kit (Stratagene), with primers listed in Table 1, according to the instructions of manufacturer. Using the wild-type PaMan5A and PaMan26A plasmids described in Couturier et al. 28), active-site variants were designed for each enzyme. Two single-site mutants were constructed for each enzyme: E177A and E283A for PaMan5A and E300A and E390A for PaMan26A. Transformation was performed in P. pastoris, and production and purification of enzyme variants were carried out as described above.
Deglycosylation Assay-N-Glycosylation sites were predicted using the NetNGlyc 1.0 Server. To remove N-linked glycans, purified enzymes were treated with peptide N-glycosidase F New England Biolabs, Ipswich, MA) under denaturing conditions according to the manufacturer's instructions. Briefly, 10 g of protein were incubated in 0.5% SDS and 40 mM DTT and boiled for 10 min for complete denaturation. Denaturated samples were subsequently incubated with 1500 units of peptide N-glycosidase F in appropriate buffer for 1 h at 37°C. Deglycosylated and control samples were analyzed by SDS-PAGE (Bio-Rad).
Analysis of End Products Release from Polysaccharides-The activity of PaMan5A and PaMan26A was assayed toward glucomannan, galactomannan, and linear mannan. Briefly, a 1% w/v solution was prepared in 50 mM sodium acetate buffer, pH 5.2. The assay was performed by incubating 75 g of enzyme with 90 l of 1% w/v substrate solution or suspension at 40°C for 30 min. After hydrolysis, mono-and oligo-saccharides were analyzed using high performance anion exchange chromatography (HPAEC) coupled with pulsed amperometric detection (PAD) (ICS 3000; Dionex, Sunnyvale, CA) equipped with a carbo-PacPA-1 analytical column (250 ϫ 4 mm). 10-l samples of enzymatic reactions were stopped by the addition of 90 l of 100 mM NaOH before injection (5 l) into the HPAEC system. Elution was carried out in 130 mM NaOH using a 25-min linear gradient program from 100% A (130 mM NaOH) to 60% A and 40% B (NaOAc, 500 mM; NaOH, 130 mM). All the assays were carried out in triplicate.

Primer
Nucleotide sequence 5 to 3 Hydrolysis Product Formation from Oligosaccharides and Determination of Kinetic Parameters-Products generated after hydrolysis of manno-oligosaccharides were analyzed using HPAEC-PAD as described above. 20 l of suitably diluted enzyme were incubated at 40°C for various time lengths with 180 l of 100 M substrate in 50 mM acetate buffer, pH 5.2. Calibration curves were plotted using ␤-1,4-manno-oligosaccharides as standards from which response factors were calculated (Chromeleon program, Dionex) and used to determine the amount of products released at different time points. All the assays were carried out in duplicate. The data were fitted to the equation of Matsui (29,30), k ϭ ln[S 0 ]/[S t ], where k ϭ (k cat / K m )[enzyme] ϫ time, and [S 0 ] and [S t ] represent substrate concentration before the start of the reaction and at a specified time during the reaction, respectively.
Hydrolysis of M 5 and M 6 in H 2 18 O-To determine and compare the hydrolytic cleavage patterns of M 5 and M 6 by PaMan5A and PaMan26A, HPAEC-PAD data on the hydrolysis products (as described above) was combined with the analysis of hydrolysis performed in H 2 18 O as described previously (31,32). Each productive binding of M 5 or M 6 gives rise to two products (e.g. M 6 cleaved to either two molecules of M 3 or to M 5 and M 1 or to M 4 and M 2 ). Quantitative HPAEC-PAD analysis of one product per cleavage (M 3 , M 4 , and M 5 , respectively) was used to calculate the relative frequency of the productive binding modes of M 5 and M 6 that give rise to these products. Each of these products can further be produced by either of two binding modes, and to distinguish between these two modes, the ratio of non-labeled ( 16  O and 4% from the enzyme and substrate stock solutions) containing 1 mM sodium acetate buffer, pH 5, 0.8 mM substrate, and 0.1 M enzyme. Samples (0.5 l) were withdrawn at different time points (0 -60 min) and spotted directly on a stainless steel plate for matrix-assisted laser desorption ionization-time-of-flight mass spectrometry (MALDI-TOF MS) analysis. Matrix (10 mg⅐ml Ϫ1 2,5-dihydroxybenzoic acid in H 2 O) was applied immediately to the sample, and it was dried under warm air. Samples from 40 and 60 min had sufficient product build-up, and the determined ratios were in good agreement (2-8% variation between the 40-and the 60-min samples from each incubation). The data from 40 min samples were used.
MALDI-TOF MS Data Acquisition and Analysis-MALDI-TOF MS spectra were recorded in positive reflector mode using a 4700 Proteomics Analyzer (Applied Biosystems, Framingham, MA). The laser intensity was set at 5500, and 50 subspectra with 20 shots on each were accumulated from each sample spot. The program Data Explorer version 4.5 was used for analysis of the data. The relative frequencies of the different productive binding modes resulting in the same products were calculated from the relative areas of the monoisotopic peaks of 16 O-and 18 O-labeled products as previously described (31). Two corrections were made; one for the [M ϩ 2] natural isotope peak (5.3% of the monoisotopic peak for M 3 , 8% of the monoisotopic peak for M 4 , and 11% of the monoisotopic peak for M 5 ) of the light ( 16 O) species that overlaps with the heavy ( 18 O) peak followed by another correction for the 7% H 2 16 O contamination in the hydrolysis assays. M 1 was not detectable because of matrix suppression of low masses.
Transglycosylation Analysis-Reactions were set up to aim for detection of transglycosylation products with MALDI-TOF-MS, similarly to what was done in Rosengren et al. (32). Five mM M 5 was incubated with 0.5 M PaMan5A at 40°C in 10 mM sodium acetate buffer pH 5 for 0 -15 min. Samples (0.5 l) were withdrawn at different time points and spotted directly onto a stainless steel MALDI plate. Matrix solution (10 mg⅐ml Ϫ 1 2,5-dihydroxybenzoic acid in water) was applied (0.5 l), and the samples were dried under warm air. Data acquisition and analysis was performed as described above.
Protein Crystallization, Data Collection, and Processing-All crystallization trials were carried out by the vapor diffusion method at 20°C. PaMan5A was concentrated to 8 mg⅐ml Ϫ1 in 20 mM Hepes, pH 7.5, 150 mM NaCl buffer. Initial crystallization trials were performed using Wizard and MDL screens (Qiagen) on a cartesian robot. For each condition, three drops (100 nl of screen buffer ϩ 100, 200, and 300 nl of protein) were formed. Optimization was then carried out by varying the pH and the concentration of precipitant. The final crystallization conditions were Tris 0.1 M pH 8.5, 0.2 M sodium acetate, 30% PEG 4000. Glycerol was used at a concentration of 25-30% as the cryoprotectant in the subsequent data collection stage. PaMan5A crystals belonged to the P2 1 2 1 2 1 space group with the cell dimensions a ϭ 56.9 Å, b ϭ 58.0 Å, and c ϭ 98.2 Å and diffracted to 1.4 Å resolution. X-ray diffraction data of a PaMan5A crystal were collected at 100K at the European Synchrotron Research Facilities (ESRF, Grenoble, France) beam line ID29.
PaMan26A was concentrated to ϳ26 mg⅐ml Ϫ1 in 20 mM Hepes, pH 7.5, 150 mM NaCl buffer. Small PaMan26A crystals were obtained in the conditions (i) 0.1 M Tris pH 7, 0.2 M Li 2 SO 4 , 1 M potassium sodium tartrate and (ii) 0.1 M imidazole, pH 8, 0.1 M potassium sodium tartrate, 0.2 M NaCl, both conditions of the Wizard screen. The best crystals were obtained after optimization in a solution containing 0.1 M Tris, pH 7, 0.2 M NaCl, 0.8 M potassium sodium tartrate, 1 mM HgCl 2 . For cryoprotection, crystals were transferred in a solution containing 25% (v/v) glycerol, 1.5 M Li 2 SO 4 , 100 mM Bistris propane, pH 7.4. The crystals belonged to the P6 5 22 space group with the following cell dimensions: a ϭ b ϭ 97.5 Å, c ϭ 268.7 Å. Several x-ray diffraction data sets were collected on beam line Proxima1 at the French synchrotron SOLEIL (Saint-Aubin, France) and on beam lines ID14-4 and ID29 at the European Synchrotron Research Facilities. The best x-ray diffraction data were collected to 2.85 Å resolution at the European Synchrotron Research Facilities beam line ID14-4.
All the data sets were processed with the programs XDS (33) and SCALA (34). The data collection statistics are summarized in Table 2.
Structure Determination and Refinement-The structure of PaMan5A was determined with the molecular replacement method using the AMoRe program (35) and the T. reesei GH5 mannanase coordinates (PDB code 1QNO). The rotation function yielded one solution, and the translation function yielded a unique solution, with a correlation coefficient and an R factor of 38.1 and 44.5%, respectively, for data between 10 and 4 Å. After rigid body refinement, the correlation coefficient was 59.2% for an R factor of 35.9%. After refinement using the programs Refmac (36) and Buster (37), the final crystallographic R factor and R free were 15.0 and 17.2%.
The structure of PaMan26A was also determined with the molecular replacement method using the AMoRe program (35). The superposition of four structures of GH26 CDs (PDB codes 2QHA, 2BVT, 2VX4, and 2WHK) plus the homology model given by the Phyre server (38) has been used as an ensemble search model for molecular replacement. The rotation function yielded one solution, and the translation function yielded a unique solution, with a correlation coefficient and an R factor of 32.4 and 49.2%, respectively, for data between 10 and 4 Å. After rigid body refinement, the correlation coefficient was 42.7% for an R factor of 43.7%. A modified structure of the CBM35 from Clostridium thermocellum (PDB code 2W47) with most of the loops deleted was located manually in the difference Fourier electron density map and was used as a starting point to build the CBM domain of PaMan26A. After performing several cycles of refinement using Refmac (36) and Buster (37) programs and manual replacement and building on the graphic display with the Turbo-Frodo program (39), the R factor has decreased to 20.7% (R free 25.8%). All representations of the structure in the figures were prepared with the program PyMOL. Coordinates for the structure PaMan5A and PaMan26A have been deposited in the Protein Data Bank under the accession number 3ZIZ and 3ZM8, respectively.

RESULTS AND DISCUSSION
Hydrolytic Activity of PaMan5A and PaMan26A toward Polysaccharides-In a recent study we showed that PaMan5A and PaMan26A displayed similar kinetic parameters toward a range of mannan substrates. To further compare the P. anserina mannanases, we measured the release of mannooligosaccharides after hydrolysis of ivory nut mannan and carob galactomannan. Toward the end of the reaction, PaMan5A yielded mainly M 2 and M 3 and smaller amount of M 1 (data not shown), consistent with other GH5 mannanases such as T. reesei (40). PaMan26A produced mainly M 4 and smaller amounts of M 1 , M 2 , and M 3 (data not shown). In other GH26 mannanases, different profiles have been observed; C. fimi CfMan26A and C. japonicus CjMan26B released M 2 and M 1 (17,26); and B. subtilis BCMan released M 2 and M 4 (21). The nature of oligosaccharide products released upon mannan hydrolysis confirms (i) the endo-mode of action of the two enzymes and (ii) differences between the two enzymes in substrate binding.
Hydrolytic Activity of PaMan5A and PaMan26A toward Oligosaccharides-The capacity of PaMan5A and PaMan26A to hydrolyze a range of manno-oligosaccharides was evaluated by ionic chromatography to get further insights into their active site architecture (Fig. 1). PaMan5A had very low activity on M 3 , higher activity on M 4 , and cleaved M 5 and M 6 rapidly (Table 3). A decrease of k cat /K m was observed with decreasing degree of polymerization. The relative k cat /K m values of PaMan5A on M 3 , M 4 , M 5 , and M 6 were 1:358:1127:1782. The increase of the degree of polymerization from 4 to 5 (M 4 and M 5 ) resulted in a 3.1-fold increase in k cat /K m , suggesting that at least four subsites are required to achieve efficient hydrolysis. In contrast, PaMan26A had no detectable activity on M 3 , very low activity on M 4 , and cleaved M 5 and M 6 rapidly. For PaMan26A the relative k cat /K m values on M 4 , M 5 , and M 6 were 1:195:365 with an increase of k cat /K m of 1.9-fold between M 5 and M 6 hydrolysis. These data suggest that PaMan26A requires at least five subsites to achieve maximum manno-oligosaccharide hydrolysis efficiency.
The nature of the hydrolysis products yielded from mannooligosaccharides (summarized in Table 4) also revealed striking differences between the P. anserina mannanases.   MAY 17, 2013 • VOLUME 288 • NUMBER 20

JOURNAL OF BIOLOGICAL CHEMISTRY 14627
whereas CfMan26A is less efficient toward M 4 and requires substrate binding at five subsites to achieve efficient hydrolysis (17). For PaMan26A, the occupation by substrate of at least five subsites to achieve efficient hydrolytic activity is even more pronounced, with a dramatic increase in k cat /K m between M 4 and M 5 .
Productive Binding Mode of M 5 and M 6 by PaMan5A and PaMan26A-␤-Mannanases usually bind oligomeric substrates in multiple productive binding modes that can generate identical products. The simplest example of this is M 3 hydrolysis to M 2 and M 1 , where mannose would be released from either the reducing end or the non-reducing end. In the former case M 3 binds productively from the Ϫ1 subsite to the ϩ2 sub-site and in the latter case from the Ϫ2 to the ϩ1 subsite, following the established subsite nomenclature (41). As another example, from M 5 , each of the products M 3 and M 4 , respectively, can be produced by either of two binding modes (see the scheme in Fig. 2A). Binding of M 5 from subsite Ϫ2 to ϩ3 or from subsite Ϫ3 to ϩ2, both, generates M 3 , and binding from subsite Ϫ4 to ϩ1 and from subsite Ϫ1 to ϩ4, both, generates M 4 . Thus, product analysis using HPAEC-PAD data alone cannot distinguish between binding modes giving the same products. However, this can be achieved when the HPAEC-PAD product analysis (as in previous paragraph) is combined with in situ product isotope labeling using 18 O-labeled water followed by mass spectrometric analysis as shown previously (31,32).
Relative quantities of the produced M 3 , M 4 , and M 5 from the HPAEC-PAD data of M 5 and M 6 hydrolysis (Table 4) were used to calculate the relative molar distribution of these products (Table 4, values in parentheses), and thus the frequencies of productive binding modes that give rise to these products could be estimated ( Fig. 2A, far left and far right column). MALDI-TOF-MS analysis was conducted to determine the ratio of nonlabeled ( 16 O) and labeled ( 18 O) species of each product (light versus heavy M 3 , M 4 , or M 5 ), which was then used to estimate the relative frequency of the productive binding modes of M 5  and M 6 that give rise to these same products. The combined results of the HPAEC-PAD and MALDI-TOF-MS data are summarized in Fig. 2A, showing the relative frequencies (%) of productive binding modes of M 5 and M 6 . The calculation procedure is explained in supplemental Table S1. To exemplify, determined from HPAEC-PAD data ( Table 4), 80% of the productive binding during the hydrolysis of M 5 by PaMan5A gen-erated M 3 . MALDI-TOF analysis then determined the ratio between the two binding modes that give M 3 . The analysis gave a ratio of M 3 /M 3 O18 of 1:2.9 (Fig. 2B), which shows that the enzyme binds M 5 preferably from subsite Ϫ3 to ϩ2 to produce M 3, giving a 59% frequency of this binding mode ( Fig. 2A). Small amounts of M 1 and M 4 were also produced, and the ratio of M 4 /M 4 O18 was 1:0.7. Hydrolysis of M 6 by PaMan5A produced a The values in parentheses represent the relative molar distribution (%) between the products M 3 , M 4 , and M 5 from each of the M 5 and M 6 incubations, which were used to estimate the relative frequency of productive binding modes yielding these products (presented in Fig. 2A). One product (M 3 , M 4 , or M 5 ) per productive binding was used for calculations; thus, only half of the produced M 3 (bold) from M 6 incubations was accounted for (two molecules of M 3 are produced from each molecule of M 6 ). mainly M 3 (55% binding frequency) but also smaller amounts of M 2 and M 4 . The ratio of M 4 /M 4 O18 was 1:3.2, which shows that PaMan5A binds M 6 preferably from subsite Ϫ4 to ϩ2 to produce M 4 (34% binding frequency). Hydrolysis of M 5 by PaMan26A yielded M 1 and M 4 with a M 4 /M 4 O18 ratio of 1:5.0 (Fig. 2B), which shows that the enzyme binds M 5 preferentially from subsite Ϫ4 to ϩ1 (83% binding frequency, see Fig. 2A). For hydrolysis of M 6 , major product ratio analysis showed that the ratio of M 4 /M 4 O18 was 1:10.8, which shows that PaMan26A prefers to bind M 6 from subsite Ϫ4 to ϩ2 (69% binding frequency). The ratio of the minor product M 5 /M 5 O18 was 1:4.2, showing that M 1 and M 5 are mainly produced without binding at the ϩ2 subsite. Thus, these data reveal clear differences in the binding mode of the two P. anserina mannanases; the predominant modes of binding of M 5 and M 6 were significantly different ( Fig. 2A). PaMan5A showed a classical pattern of hydrolysis products (M 3 and M 2 mainly were released from M 5 ) as described in several studies (B. subtilis, C. japonicus, M. edulis), whereas PaMan26A showed release of M 4 and M 1 from M 5 , which is unusual when compared with other GH26 endo-mannanases (B. subtilis BCMan, C. fimi CfMan26A, C. japonicus CjMan26A), suggesting an unusual arrangement of subsites in the catalytic center.
Transglycosylation Ability-To detect potential transglycosylation ability of the two enzymes, they were incubated with M 5 as substrate. The resulting short time course study of the product formation clearly showed that PaMan5A, in addition to hydrolysis products, also produces transglycosylation products with higher degree of polymerization than the original substrate (Fig. 3). PaMan5A was able to transglycosylate yielding to oligosaccharide structures of up to a degree of polymerization of 8 (n ϩ 1 to n ϩ 3), in good agreement with GH5 mannanases described before, T. reesei (n ϩ 1 to n ϩ 3) (31), and Aspergillus nidulans ManA (n ϩ 1 to n ϩ 3) and ManC (n ϩ 1 and n ϩ 2) (6). No transglycosylation products could be detected with PaMan26A incubated with M 5 in the same experimental conditions, which is consistent with some other family GH26 mannanases that have been described as non transglycosylating enzymes (10).
Structure of PaMan5A-The crystal structure of PaMan5A was solved in its free form. The crystal contained one monomer in the asymmetric unit, and light-scattering experiments indicated that the protein is a monomer in solution (data not shown). The overall structure of PaMan5A (Fig. 4A) revealed a (␤/␣) 8 -barrel fold as expected for enzymes belonging to clan GH-A. When superimposed with TrMan5A (PDB code 1QNR) and Thermomonospora fusca mannanase (PDB code 3MAN) structures (supplemental Fig. S1), the overall fold of PaMan5A is very similar to that of TrMan5A, with structural differences being confined mainly in the loop regions (Fig. 4A)  peptide N-glycosidase F, no shift in the apparent molecular mass (46 kDa) was observed on SDS-PAGE compared with untreated sample (data not shown). This observation was in good agreement with NetGlyc predictions from the PaMan5A primary sequence (no predicted N-glycosylation site) and analysis of the PaMan5A crystallographic data that confirmed absence of glycosylation units.
The active site of PaMan5A was clearly identified in the groove, with the two conserved catalytic glutamate residues (acid-base and nucleophile) positioned near the C-terminal ends of ␤-strands four and seven of the (␤/␣) 8 barrel (41), Glu-177 and Glu-283, respectively. Mutant E283A showed no catalytic activity for glucomannan, thus indicating that Glu-283 should be the nucleophile. E177A had a specific activity of 0.47 units⅐mg Ϫ1 toward glucomannan, which is roughly 100-fold lower than the wild-type enzyme (45 units⅐mg Ϫ1 ), thus indicating that Glu-177 should be the acid-base catalytic residue. These results are in agreement with other homologous GH5 enzymes where catalytic residues have been determined (42,43). Despite several attempts, no structure of PaMan5A inactive mutants alone or in complex with its substrate has been obtained. Consequently, we performed comparative structural analysis of PaMan5A with other GH5 mannanases complexes (T. reesei PDB code 1QNO, Thermotoga petrophila PDB code 3PZ9, and S. lycopersicum PDB code 1RH9) to map the substrate binding subsites (Fig. 4B). In the Ϫ1 and ϩ1 subsites where the catalytic cleavage occurs, 7 of 8 residues highly conserved in GH5 mannanases (44) are found in PaMan5A, among which are the catalytic residues Glu-177 and Glu-283 and Arg-62, Asn-176, His-248, Tyr-250, Trp-315 (Fig. 4B). PaMan5A also has an arginine equivalent to Arg-171 in the ϩ2 subsite of TrMan5A (15), which is semi-conserved among GH5 mannanases and which was shown to play a significant role in the transglycosylation ability of TrMan5A (32).
Structure of PaMan26A Catalytic Module-The structure of PaMan26A was successfully solved using molecular replacement. The search model was composed of the superimposition of four structures of bacterial mannanases (PDB codes 2QHA, 2BVT, 2VX4, and 2WHK). The final structure comprising 443 residues was refined at 2.85 Å resolution. The overall structure of PaMan26A CD revealed a (␤/␣) 8 -barrel fold (Fig. 5A) as expected for enzymes belonging to clan-GHA. The active site was clearly identified in the groove, with the two conserved catalytic glutamate residues (Glu-300 and Glu-390) positioned at the end of the (␤/␣) 8 barrel and several aromatic residues forming the subsites of catalytic cleft. Mutant E390A showed no catalytic activity for glucomannan, indicating that Glu-390 should be the nucleophile. E300A had a specific activity of 0.33 units⅐mg Ϫ1 , which is roughly 200-fold lower than the wild-type enzyme (65 units⅐mg Ϫ1 ), thus indicating that Glu-300 should be the acid-base catalytic residue. These results are in agreement with other homologous GH26 enzymes where catalytic residues have been determined (45).
Electron density was observed for two carbohydrate sugar residues at one glycosylation site, Asn-268, which is located in  the CD on the external side of the barrel. As modeled from electron density, 2 ␤-1,4-linked N-acetylglucosamine (GlcNAc) units are attached to this N-glycosylation site. N-Deglycosylation of PaMan26A using peptide N-glycosidase F was associated with a 2-3-kDa shift in the apparent molecular mass on SDS-PAGE compared with untreated sample (data not shown). These results confirm that PaMan26A is N-glycosylated and are in agreement with the NetNGlyc prediction (one predicted N-glycosylation site at position Asn-268).
Several  (Fig. 5, B and C). As described for CfMan26A and CjMan26A, PaMan26A Tyr-362 is probably involved in a hydrogen bond with the catalytic nucleophile Glu-390, whereas PaMan26A Trp-305 and Trp-413 could play a role as aromatic platforms to stabilize mannopyrannose rings at the ϩ1 and Ϫ1 subsites, respectively (Fig. 5C). In the Ϫ2 subsite of BCMan, binding is not favorable because of steric hindrance due to the position of Tyr-40 (21). In the case of CfMan26A, the two aromatic Phe-123 and Tyr-124 residues that are superimposed with PaMan26A F248 and Tyr-249 stabilize the interaction with a mannose unit at the Ϫ2 subsite (Fig. 5C).
Our experimental data indicate that PaMan26A displays strong interactions at the Ϫ4 subsite. Indeed, PaMan26A was poorly active toward M 4 probably due to the formation of an unproductive complex between Ϫ4 and Ϫ1 subsites. We further analyzed the Ϫ4 subsite in the PaMan26A structure and identified two aromatic residues, Tro-244 and Trp-245, located in loop 2 that could stabilize mannopyrannose rings in the Ϫ4 subsite (Fig. 5, B and C). As PaMan26A, CfMan26A active site also contains four glycone binding subsites, but experimental results provided evidence for the existence of a strong Ϫ3 subsite, and residues involved in the Ϫ4 subsite were described as making a minor contribution to binding (17). In PaMan26A, there is no equivalent to the Phe-42, Phe-325, and Gln-329 CfMan26A Ϫ3 subsite residues. The lack of a strong Ϫ3 subsite and the presence of a strong Ϫ4 subsite in the structure is in agreement with our experimental results that suggest a predominant substrate binding mode involving the Ϫ4 subsite. Lacking a strong Ϫ4 subsite, CfMan26A produces M 2 and M 3 as major products from M 5 with only minor amounts of M 1 and M 4 (31).
PaMan26A Modular Organization-PaMan26A harbors a family 35 CBM at its N-terminal end, and the closest characterized enzyme is Humicola insolens ␤-mannanase (GenBank TM AAQ31840 (46)) with 78% amino acid identity. After a BlastP search using the PaMan26A amino acid sequence, it is interesting to note that all related bacterial and fungal sequences harbor a CBM35 module at their N terminus. In fungi, in addition to PaMan26A CBM35, only one CBM35 module binding to galactan has been characterized to date in a Phanerochaete chrysosporium exo-␤-1,3-galactanase (47). We previously suggested that the N-terminal CBM35 module of PaMan26A displayed dual binding specificity toward xylan and mannan (28), and the phylogenetic analysis was performed by Correia et al. (48) clustered PaCBM35 in the subfamily II that is proposed to target ␤-1,4 mannan.
Although the structures of fungal GH bearing a CBM are generally determined separately, this is the first intact structure that allows visualization of the juxtaposition of the CBM35 module relative to the GH26 CD. The linker region of PaMan26A is short without any glycosylation sites, whereas modular fungal GHs usually display long and highly glycosylated linkers. The PaMan26A linker sequence was rich in proline residues, i.e. it contains 4 prolines (Pro-132, -134, -135, and -140) of 12 residues that may confer rigidity to the modular enzyme (Fig. 6A). The linker starts on residue Ser-130 at the end of the last ␤-strand of the N-terminal CBM domain. Only two residues (Ala-131 and Pro-132) have no interaction with the rest of the molecule. The region from residue R133 to residue N141, which may be considered as the end of the linker, is tightly bound to the CD. Arg-133 and His-136 side chains make hydrogen bond with Asp-382, Asn-374, and Gln-404, whereas the side chain of Ile-138 fits into a hydrophobic cavity made of Arg-159 (aliphatic part of the side chain), Tyr-162, and Mer-385. The CBM and the catalytic module are thus in close association thanks to the embedded linker (Fig. 6B), and it may explain why attempts to express the catalytic module alone were unsuccessful (data not shown). Alignment of PaMan26A-CBM35 with 60 microbial GH26 mannanases sequences bearing a CBM35 module revealed that they all display a short linker region (12-14 residues) rich in proline residues (data not shown).
The CBM35 domain comes into contact with the CD through hydrophobic interactions. Indeed a hydrophobic patch comprising Leu-58 and Leu-103 on the surface of the CBM35 domain stands in front of a cluster of hydrophobic residues (Ala-402, Tyr-403, and Leu-399) of the CD. The rationale of the tight modular association of bacterial and fungal GH26-CBM35 mannanases will need further work to gain insights into their function.
PaCBM35 Domain-The CBM35 domain overall structure consists of 2 antiparallel sheets consisting of 4 and 5 antiparallel ␤-strands, respectively. The two sheets are packed in a ␤-sandwich conformation enclosing a highly hydrophobic core. The closest structural homologue found using the DALI server (49) is a CBM35 from C. thermocellum (PDB code 2W1W (50)), with a Z-score of 18.6. Its superimposition with PaCBM35 shows that 57 C␣ of 125 C␣ (45%) have equivalent positions in both molecules, with the distance between the superimposed C␣ atoms Ͻ1 Å. The main differences occur in the loops connecting the ␤-strands as shown in Fig. 6C. A metal ion is present that has been modeled as calcium based on its coordination geometry exclusively with oxygen atoms. A calcium ion is also present at a similar location in the structure of the CBM35 domain from C. thermocellum (50). However, the second calcium evidenced in all of the other CBM35s and involved in carbohydrate recognition (50) is not conserved in PaCBM35. A platform of three aromatic residues (Phe-87, Trp-117, and Trp-119) was observed at the surface of the PaCBM35 (Fig.  6B). These residues are aligned with the PaMan26A catalytic cleft, suggesting that they could play a role in substrate binding. Conclusions-The P. anserina CAZome (the genome-wide inventory of CAZymes) includes three genes encoding ␤-(1,4)mannanases: two GH5 mannanases without CBM (including PaMan5A) that both belong to the GH5 subfamily 7 (51) and one GH26 mannanase bearing a CBM35 (i.e. PaMan26A) with affinity for hemicellulosic polysaccharides (28). Based on our kinetic analysis, we can conclude that PaMan5A and PaMan26A are complementary in terms of hydrolysis profile and could act in synergy to deconstruct mannan polysaccharides. Indeed, PaMan26A produces larger manno-oligosaccharides that could be processed by PaMan5A. In C. japonicus, a bacterium also producing both GH5 and GH26 ␤-mannanases, the catalytic modules of GH5 mannanases were linked to various CBMs, whereas GH26 mannanases were found as single CD (25). Therefore, it has been suggested that GH26 mannanases were involved in degradation of storage tissues, whereas GH5 mannanases harboring cellulose-specific CBMs were involved in degradation of plant cell wall (20,26). It is interesting to note that the P. anserina mannanase system does not seem to fit with this model, suggesting a difference in the strategies to degrade mannan between these two microbes.
Together with our previous studies on P. anserina CAZymes (28,52,53), the present findings give more insights into the P. anserina enzymatic machinery for the deconstruction of plant cell wall polysaccharides. This knowledge is essential to design tailor-made biocatalysts, which can then be used in the biofuel and bioprocessing industries.