Mechanistic Strategies for Catalysis Adopted by Evolutionary Distinct Family 43 Arabinanases*

Background: Arabinanases are key enzymes involved in hemicellulose degradation. Results: Crystallographic, mutational, and biochemical assays of three arabinanases reveal the molecular mechanisms governing their catalysis and activation. Conclusion: Accessory domain and metal ion are essential for catalysis. Structural adaptations in the catalytic interface confer unique action modes to ruminal arabinanases. Significance: This work provides new molecular strategies for arabinan hydrolysis. Arabinanases (ABNs, EC 3.2.1.99) are promising catalysts for environmentally friendly biomass conversion into energy and chemicals. These enzymes catalyze the hydrolysis of the α-1,5-linked l-arabinofuranoside backbone of plant cell wall arabinans releasing arabino-oligosaccharides and arabinose, the second most abundant pentose in nature. In this work, new findings about the molecular mechanisms governing activation, functional differentiation, and catalysis of GH43 ABNs are presented. Biophysical, mutational, and biochemical studies with the hyperthermostable two-domain endo-acting ABN from Thermotoga petrophila (TpABN) revealed how some GH43 ABNs are activated by calcium ions via hyperpolarization of the catalytically relevant histidine and the importance of the ancillary domain for catalysis and conformational stability. On the other hand, the two GH43 ABNs from rumen metagenome, ARN2 and ARN3, presented a calcium-independent mechanism in which sodium is the most likely substituent for calcium ions. The crystal structure of the two-domain endo-acting ARN2 showed that its ability to efficiently degrade branched substrates is due to a larger catalytic interface with higher accessibility than that observed in other ABNs with preference for linear arabinan. Moreover, crystallographic characterization of the single-domain exo-acting ARN3 indicated that its cleavage pattern producing arabinose is associated with the chemical recognition of the reducing end of the substrate imposed by steric impediments at the aglycone-binding site. By structure-guided rational design, ARN3 was converted into a classical endo enzyme, confirming the role of the extended Arg203–Ala230 loop in determining its action mode. These results reveal novel molecular aspects concerning the functioning of GH43 ABNs and provide new strategies for arabinan degradation.

Endo-1,5-␣-L-arabinanases (ABNs) 3 (E.C. 3.2.1.99) are key enzymes in the breakdown of plant cell wall arabinans and their products such as arabinose and arabino-oligosaccharides have potential applications as low-calorie sweetener and prebiotics, respectively (1,2). These enzymes are also considered promising tools in the environmentally friendly biomass conversion into energy and chemicals (3) because pentoses can also be used as carbon sources and hemicellulosic oligosaccharides can competitively inhibit the cellulolytic system, lessening the hydrolysis efficiency. Indeed, recent studies showed that the supplementation of fungal enzyme mixtures with ABNs and arabinofuranosidases (ABFs, E.C. 3.2.1.55) improved the hydrolysis rate of plant biomass (4 -6). Moreover, genomics and proteomics approaches have revealed an impressive expansion of the GH43 family, which includes ABNs and ABFs, in a number of organisms specialized in plant cell wall degradation. This observation highlights the importance of this CAZy family for depolymerization of lignocellulosic material and, probably, reflects a broader range of specificities than described so far.
GH43 ABNs consist of a five-bladed ␤-propeller catalytic core that can be found as a discrete protein (7)(8)(9)(10) or fused to an ancestral ancillary domain with a ␤-barrel-like architecture (11). Although the accessory ␤-sandwich domain, found in GH43 ␤-xylosidases and ABFs, contributes to the active site formation of its parental protein (12)(13)(14), in ABNs the ␤-barrellike domain does not participate in the groove formation. To date, there are no structural and functional studies with deletion mutants of two-domain ABNs to determine the correlation of this ancestral domain with catalysis and stability of these enzymes. Moreover, all structurally characterized ABNs so far (7)(8)(9)(10)(11) contain a divalent ion bound to a solvated channel located at the inner region of the propeller-fold, which recently has been demonstrated to be essential for catalysis (11). Nevertheless, the structural basis for the ion role remains unresolved and there has been no evidence for the existence of divalent ion-independent GH43 ABNs.
The synergism between endo-and exo-acting hydrolases is a sine qua non for complete saccharification of complex and recalcitrant polysaccharides (15,16) such as arabinan that consists of a backbone of ␣-1,5-linked L-arabinofuranosyl residues decorated with ␣-1,3and ␣-1,2-linked arabinofuranosides and can be found freely or attached as a side chain of xylogalacturonan and rhamnogalacturonan (17). However, only two GH43 exo-ABNs have been characterized so far. BsArb43A releases a trisaccharide from the non-reducing end of arabinan and the structural basis for its action mode is the partial blocking at the Ϫ3 subsite (8). ARN3 is an exo-ABN that releases arabinose from both arabinan and arabinooligosaccharides for which there is no structural information (18). Most GH43 ABNs (8, 10, 19 -21) show preference for linear substrates, despite their natural substrate, arabinan, is commonly a branched chain. ARN2, a functionally characterized two-domain ABN presented similar affinity for both linear and branched arabinan (22), which is interesting and also relevant to biomass degradation, but the structural basis for this feature is still unknown.
Thus, to shed light on the structural determinants for the distinct molecular mechanisms governing the catalysis of this important family of glycosidases, we have investigated three GH43 ABNs: two endo-acting ABNs, TpABN and ARN2, which are two-domain enzymes, from Thermotoga petrophila RKU-1 (23)(24) and from a bovine rumen metagenomic library (22), respectively; and, a single-domain exo-ABN (ARN3) recovered from the same metagenomic DNA library (18). Biochemical and biophysical studies with TpABN contributed to elucidate the structural basis for calcium dependence of some GH43 ABNs and the role of the ancestral ancillary module for function and stability of two-domain ABNs. The structure, mutagenesis, and kinetics of ARN3 revealed that its exo-acting mode (18) is related to the chemical recognition of the reducing ends of arabinan imposed by a steric barrier at the aglyconebinding region. Moreover, structural data of ARN2 showed that its unique ability to efficiently degrade branched substrates as observed by Wong and colleagues (22) is due to a larger catalytic interface with higher solvent accessibility than that observed in other ABNs with preference for linear arabinan. These molecular findings provide a further understanding in the functional aspects, stability, and regulation of GH43 ABNs and represent an advance toward new strategies for hemicellulose degradation.
Expression and Purification-All of the proteins were expressed in BL21(DE3) cells supplemented with pRARE2 plasmid for 4 h at 30°C with 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside at A 600 nm of 0.6 -0.8. The cells were collected, resuspended in lysis buffer (20 mM sodium phosphate, pH 7.5, 500 mM NaCl, 5 mM imidazole, 1 mM PMSF, and 5 mM benzamidine), and disrupted by lysozyme treatment (80 g/ml, 30 min, on ice), followed by sonication (ϳ6 pulses of 15 s on ice using a Vibracell VCX 500, Sonics and Materials). The extracts were centrifuged at 10,000 ϫ g for 30 min and the supernatants were loaded onto nickel-charged 5-ml HiTrap chelating columns (GE Healthcare) with a flow rate of 1 ml/min. The proteins were eluted using a nonlinear gradient of imidazole ranging from 5 to 500 mM. The fractions containing the target protein were concentrated using Amicon Ultra centrifugal filters (Millipore) and submitted to size exclusion chromatography using a Superdex 75 16/60 column (GE Healthcare) and a flow rate of 0.5 ml/min. The TRX tag was cut out of the ARN3 construct before the size exclusion chromatography experiment by limited proteolysis using 0.01 mg/ml of trypsin for 30 min at room temperature. All chromatographic steps were performed in an Ä KTA FPLC system (GE Healthcare). Sample homogeneity was confirmed by SDS-PAGE and protein concentration was estimated by absorbance at 280 nm using molar extinction coefficients derived from the sequence of the constructs.
Crystallization, Data Collection, Structure Determination, and Refinement-Crystals of TpABN were promptly obtained without His 6 tag removal. Crystals of ARN3, a single-domain ABN, were obtained only after sample treatment with trypsin that efficiently cleaved out the macromolecular tag. Trials to remove the TRX tag from ARN2 yielded a heterogeneous sample with low stability, suggesting an interaction between the enzyme and macromolecular tag. Based on this observation, we preceded the crystallization with the TRX-ARN2 chimeric protein. The purified TpABN, TRX-ARN2, and ARN3 proteins were concentrated to 29, 5.8, and 12.9 mg/ml, respectively, using Amicon Ultra centrifugal filter units (Millipore) and dialyzed against 20 mM Tris-HCl, pH 7.5 (TpABN), or 20 mM HEPES, pH 7.5 (TRX-ARN2 and ARN3). Crystallization experiments were performed by the vapor diffusion method using a HoneyBee 963 robot (Genomic Solutions). Sitting drops were prepared by mixing 0.5 l of the protein solution with an equal volume of mother liquor and equilibrated against 80 l of the reservoir at 18°C. Formulations based on the following commercial crystallization kits were tested: SaltRX, Crystal Screen, Crystal Screen 2 (Hampton Research), Precipitant Synergy, Wizard I and II (Emerald BioSystems), PACT and JCSG (Qiagen/Nextal), totalizing 544 solutions. Crystals of TpABN with a ligand at the active site were obtained within 1 week in 100 mM sodium citrate, pH 5.5, 40% (v/v) PEG400, and 5% (w/v) PEG3350. The reduction of the crystallization condition pH to 5.0 yielded crystals with better diffraction quality.
Crystals were directly flash-cooled in a 100 K nitrogen gas stream because PEG400 is a cryoprotectant compound. TpABN crystals in the native state were obtained by changing the dialysis buffer, prior to protein concentration, from 20 mM Tris-HCl, pH 7.5, to 20 mM HEPES, pH 7.5. Small crystals of TRX-ARN2 were obtained within 1 day in 2.1 M sodium malonate and bigger crystals were formed in hanging drops by reducing the precipitant concentration to 1.8 M. Crystals of ARN3 were obtained within 45 days in 100 mM Tris-HCl, pH 8.5, 2.4 M ammonium phosphate and better crystals were obtained reducing the precipitant concentration to 2.1 M. Crystals of ARN2 and ARN3 were cryoprotected using 30% (v/v) glycerol before flash cooling. X-ray diffraction data were collected at the W01B-MX2 beamline (Brazilian Synchrotron Light Laboratory, Campinas, Brazil). Data processing was carried out with HKL2000 or XDS (27,28). All structures were solved by the molecular replacement method, using the program MolRep (29). The native and ligand-complex structures of TpABN were initially solved by molecular replacement in P2 1 2 1 2. However, the high values of R factor /R free (26.1/30.1 for TpABN and 26.2/29.7 for TpABN-ligand complex) after model refinement were not compatible with high data quality and resolution indicating an incorrect space group and/or a packing disorder (30,31). Besides that, the high solvent content (63%) was very suspicious for a well diffracting crystal, but all molecular replacement attempts to find another molecule in the asymmetric unit were unsuccessful. Therefore, data sets were processed in a lower symmetry space group (P2 1 ) and three molecules were found in asymmetric units with interpretable maps and lower R factor /R free values (16.0/19.7 for TpABN and 15.3/19.7 TpABNligand complex). The third molecule (chain C) presented a poor map in comparison with chains A and B, probably due to several orientations of this molecule along the crystal lattice. The absence of significant off-origin peaks in the native Patterson map and a few diffused reflections in some frames suggest the presence of rotational order-disorder structure (30). Besides the order-disorder problem, chain C lies on a 2-fold symmetry axis and, because the molecule does not have an intrinsic symmetry, it is not possible to expand symmetry to P2 1 2 1 2. The YxiA protein from Bacillus licheniformis (PDB code 3LV4), sharing 54% identity with TpABN was used as a search model to solve the TpABN structure. The ABN from Geobacillus stearothermophilus (AbnB, PDB code 3CU9) was used as search model for ARN3 (43% of identity). The TRX-ARN2 structure was solved using TpABN (identity of 34%) and E. coli TRX structures as search models. The structures were refined using the program REFMAC5 (32). After each cycle of refinement, the model was inspected and manually adjusted into the (2F o Ϫ F c ) and (F o Ϫ F c ) electron density maps using the program COOT (33). Solvent molecules were manually added at positive peaks above 3.0 in the difference Fourier maps. The local and global quality of the structures was evaluated using the Molprobity server (34). Processing and refinement statistics are reported in Table 1. Interestingly, analysis of the ARN2 crystal lattice showed that the macromolecular tag assisted the crystallization process by providing new molecular surfaces entropically favorable for crystal packing. In fact, the main crystalline interface corresponds to a buried area of 1330 Å 2 and involves inter-chain interactions between TRX and ARN2 from the 2-fold symmetrically related molecules found in the asymmetric unit (Fig. 1A). Moreover, when the two molecules of ARN2 from the asymmetric unit were superimposed, the TRX did not superimpose, indicating that the two chimeras adopt different orientations, possibly to maximize the intermolecular contacts (Fig. 1B). These observations corroborate that TRX crystalline contacts played a key role in the crystal formation. This strategy was also successfully described for other proteins (35)(36)(37)(38)(39)(40)(41)(42)(43) and the use of a short linker connecting the protein target with the macromolecular is essential to produce a rigid and crystallizable chimera (41,42).
Enzyme Assays-Standard arabinanase activity assays were performed in 50 mM citrate/phosphate/glycine buffer, contain-ing 0.25% (w/v) debranched arabinan (Megazyme International, Ireland) and 8 g/ml of enzyme. The amount of reduction sugar was determined by the DNS method (44). The enzymes and solutions were previously incubated with Chelex 100 (Bio-Rad) to remove any trace of metal ion. Chelators (EDTA and EGTA) and metal ions (chloride salts of Ca 2ϩ , Mg 2ϩ , Mn 2ϩ , Ni 2ϩ , and Co 2ϩ ) were used in concentrations of 5 and 10 mM, respectively. The reactions were carried out at pH 6.0 and 70°C for TpABN/mutants and pH 6.0 and 45°C for ARN2/ARN3. Temperature profiles were obtained from reactions incubated at pH 6.0 for 10 min at temperatures ranging from 20 to 85°C. pH curves were obtained from reactions incubated at 70°C for TpABN/mutants and 45°C for ARN2/ARN3/ mutants for 10 min at pH values from 2 to 10. The apparent kinetic parameters K m , V max , k cat , and k cat /K m were calculated by nonlinear regression analysis of the Michaelis-Menten plot. Debranched arabinan was used as substrate in a concentration range from 0.8 to 25.0 mg/ml. Assays were conducted in 50 mM  citrate/phosphate/glycine buffer, pH 7.0, at 45°C for 10 min. Mathematical calculations were made using GraphPad Prism 5.0 (GraphPad Software). These assays were performed in sextuplicate.
Capillary Zone Electrophoresis-WT and mutant versions of ARN3 (SRGEEP-ARN3, ARN3/F237A, ARN3/D37W, ARN3/ S329P, ARN3/D37W/S329P) were incubated with 0.25% (w/v) linear arabinan (Megazyme) for 15 min at 20°C. Alternatively, 7 mM arabinoheptaose (Megazyme) was used as substrate and the reaction was stopped after 5 and 10 min of incubation at 20°C. The reaction products were derivatized with 8-aminopyrene-1,3,6-trisulfonic acid by reductive amination as described previously (45). Capillary electrophoresis experiments were performed in a P/ACE MDQ system (Beckman Coulter) with laser-induced fluorescence detection using a fused-silica capillary (TSP050375, Polymicro Technologies) of 50-mm internal diameter and 31-cm length. Samples were injected by application of 0.5 p.s.i. for 0.5 s. Electrophoresis conditions were 15 kV/70 -100 mA with the cathode at the inlet, 0.1 M sodium phosphate, pH 2.5, as running buffer and a controlled temperature of 20°C. The capillary was rinsed with 1 M NaOH followed by running buffer with a dipcycle to prevent carry over after injection. Oligosaccharides labeled with 8-aminopyrene-1,3,6trisulfonic acid were excited at 488 nm and emission was collected through a 520-nm band pass filter. Retention times varied slightly in different runs due to small volumes of the capillary and variations in the buffer strength and temperature. The resultant peaks were assigned by comparison with electrophoretic behavior of standards: arabinose and 1,5-␣-L-arabinooligosaccharides (Megazyme).
Spectroscopy Characterization-Far UV circular dichroism spectra were recorded on a Jasco J-810 spectropolarimeter (Jasco International Co., Ltd., Tokyo, Japan) coupled to a Peltier temperature controller and a 1-mm quartz cuvette. The proteins were previously dialyzed against 20 mM sodium phosphate, pH 7.5, and used at a concentration of 10 M. The data collection parameters were set to scan rate of 50 nm/min, response time of 4 s, sensitivity of 100 mdeg, accumulation of 10, and delay time for spectrum collection of 60 s. Unfolding experiments were monitored at 230 nm. For thermal denaturation experiments the cuvette was heated at a rate of 1°C/min.
Computational Analysis-The ARN3 structure was prepared for molecular dynamics simulations using the program YASARA (46). All hydrogen atoms and other missing atoms were created using force field parameters from YAMBER3. A simulation box was defined at 15 Å around all atoms of each structure. Protonation was performed based on pH 7. Cell neutralization was reached filling the box with water molecules and Na/Cl counterions. A short molecular dynamics simulation was performed for solvent relaxation, deleting water molecules until the water density of 0.997 g/ml was reached. A short steepest descent energy minimization was carried until the maximum atom speed dropped below 2200 m/s. Then 500 steps of simulated annealing were performed at 0 K. Finally, a 10-ns simulation at 298 K and with non-bonded cutoff of 7.86 Å was performed (47). A snapshot was saved every 25 ps. Each of the structures was simulated with and without a calcium ion at the coordination site. To analyze the backbone flexibility and sta-bility, the root mean square deviation of each C␣ atom was calculated for each structure. Using the algorithm KVFinder (developed by R. V. Honorato, P. S. L. Oliveira, and others) we developed a custom routine that was able to calculate the volume of the coordination site for each simulation snapshot. The matrix-filling procedure for cavity prospection of KVFinder is able to detect and account for slight fluctuations in the coordination site.

RESULTS
Molecular Basis of the Accessory Domain Influence on Stability and Catalytic Activity-The crystal structure of the hyperthermostable GH43 ABN from T. petrophila (TpABN), was determined at 1.7-Å resolution, revealed a canonical 5-bladed ␤-propeller catalytic core with a C-terminal ␤-barrel-like domain, which lacks sequence similarity with characterized carbohydrate-binding modules. The only structural study addressing a two-domain ABN attempted to produce truncated versions of the protein lacking this C-terminal domain, but no expression was observed in the soluble fraction (11). Herein, to characterize the influence and importance of this accessory domain for stability and catalysis, we designed a truncated construct of TpABN devoid of the C-terminal accessory domain (TpABN⌬C, Ala 28 -Leu 371 ) that was successfully produced in E. coli. The truncated protein TpABN⌬C was correctly folded, as suggested by circular dichroism analysis ( Fig. 2A), but presented decreased thermal stability, with a melting temperature (T m ) of 66.4°C (Fig. 2B), whereas the full-length protein TpABN showed a T m of 88.7°C (Fig. 2B). The unusual circular dichroism spectra of TpABN constructs, with a positive band at 230 nm, are the result of multiple tryptophan side chain absorption in the ␤-propeller core (48). Structural analysis using the Protein Interactions Calculator (49) showed a large inter-domain area (2,297 Å 2 ) stabilized by extensive polar and hydrophobic contacts ( Table 2) that are affected by ionic strength, because the WT protein presented an increase of 7°C in the T m under moderate ionic strength (Fig. 2B). In single-domain ABNs, the corresponding region is populated by charged and polar residues. These results suggest a direct structural role of the accessory domain in two-domain ABNs and explain the poor solubility and stability of the deletion mutant. Interestingly, despite the native-like properties of the catalytic core, the enzymatic activity of the enzyme was abolished after deletion of the accessory domain, indicating an essential role in catalysis as well. Structural data revealed that the accessory domain participates in the conformational stabilization of the Gly 243 -Gly 250 loop (GGLDYRGG, Fig. 2C). This loop is adjacent to the active site and forms the substrate-binding groove. The accessory domain maintains the Gly 243 -Gly 250 loop in a conformation favorable for substrate recognition and binding, and in the absence of the accessory domain, this loop may adopt a different or non-ordered conformation, disrupting the substrate channel and displacing the substrate. The leucine residue, located at the tip of this loop (Leu 245 in TpABN, Fig. 2C), is conserved in all two-domain ABNs (Fig. 3) and possibly plays a role in the substrate interaction because it establishes aliphatic contacts with carbon atoms from a sugar mimetic (results not shown). In single-domain ABNs, a conserved disulfide bond (Cys 254 -Cys 255 in the ruminal enzyme, ARN3) is observed at the equivalent position of the leucine residue (Fig. 2D). This disulfide bridge also participates in the formation of the substrate-binding channel and is important to stabilize the Gly 243 -Gly 250 loop (numbering based on the sequence of TpABN) in the proper geometry for substrate binding, enabling single-domain ABNs to be functional without the presence of an accessory domain. The conservation of this disulfide bridge in Cellvibrio japonicus exo-arabinanase CjArb43A (Fig. 3) indicates that exoacting single-domain ABNs share a common mechanism of active site formation with endo-acting single-domain ABNs.
Calcium Ion Caged into ␤-Propeller Core Is Essential for Catalysis-In all single-and two-domain ABNs with crystallographically determined structures (7-11), a calcium ion was modeled in the highly hydrated cavity located at the inner core of the structure, which connects the catalytic interface with the opposite face of the propeller (Fig. 4A). Functional analysis showed that TpABN, as well as BsArb43B (11), was inhibited by chelating agents and only recovered its maximum enzymatic activity in the presence of calcium (Fig. 4B). In these calciumdependent ABNs, the ion interacts with the catalytically relevant His 314 (TpABN sequence numbering) and six water molecules forming a distorted pentagonal bipyramidal molecular geometry (Fig. 4C). These water molecules interact with main chain carbonyl oxygen atoms of Pro 33 , Ala 96 , Asp 170 , Pro 171 , Gly 224 , Pro 225 , and Asn 315 , which permits the calcium ion to form water-mediated interactions with all five blades of the propeller. In addition, the side chains of Ser 34 and Gln 98 interact with W1, W2, and W6 water molecules (Fig. 4C), but these interactions seem not to be essential for calcium binding because in the BsArb43B structure, the corresponding residues, Ser 40 and Asp 103 , adopt different orientations and are distant from the calcium-coordinating water molecules. A possible structural role of the calcium ion in GH43 enzymes is not evident because thermal stability assays in the presence or absence of the ion did not result in significant changes in the T m (⌬t Ͻ 2°C).

Ruminal Enzymes Display a Calcium-independent Catalytic
Mechanism-Interestingly, the catalytic activity of the ruminal enzymes, ARN2 and ARN3, did not undergo any change by the presence of calcium ions or chelating agents (Fig. 4B), indicating a calcium-independent mechanism. In fact, the ARN2 crystalline structure did not present the calcium ion bound to the hydrated cavity. Based on electron density, coordination sphere, and atom distances, we deduced the presence of a sodium ion with distorted pentagonal bipyramidal coordination at the corresponding position (Fig. 4D). The geometry is slightly different from calcium coordination, but all direct and water-mediated interactions are fully conserved. The presence of a potassium ion was discarded due to larger ionic radius and typical bond lengths with oxygen atoms, which is not compatible to the hydrated pocket. Although the water molecule has a similar electron density to sodium ion, the number of interactions and geometry bring it less plausible. Thus, the sodium is the most likely substituent for calcium ions in ARN2 and ARN3 enzymes. To elucidate the structural determinants for calcium dependence, mutants G172C, G224A, and G172C/G224A of TpABN were produced according to comparative structural analysis of the ion-binding pocket. Nonetheless, all mutants showed similar residual activity to the WT enzyme under calcium deprivation (Fig. 4B), indicating that these elements are not responsible for the calcium-independent behavior observed in the ruminal enzymes, ARN2 and ARN3. Other molecular aspects were also analyzed including electrostatic surface charge distribution, the presence or absence of the accessory domain, and volume of the cavity. However, these analyses did not indicate a clear correlation of these parameters with calcium dependence. Thus, collectively these structural and functional evidences suggest a global effect, resulting from multiparameter attributes, which select some GH43 ABNs to be activated by calcium or sodium ions.
Aromatic Gateway Controls Solvent Dynamics at the Hydrated Pocket-The molecular complexity regarding the catalytic machinery of GH43 ABNs is not limited to the direct role of His 314 and the presence of metallic cofactors. The solvent dynamics at the hydrated pocket is also an important issue for the proper functioning of these enzymes. The ion coordination is almost exclusively formed by water molecules, thus it might be labile to solvent perturbations in the hydrated pocket. To investigate this feature of the ion-binding pocket, the fully conserved aromatic residue located at the entrance of the hydrated pocket (Tyr 226 in TpABN (Fig. 4E) and Phe 237 in ARN3) was mutated to alanine causing a significant reduction of the catalytic activity in both enzymes (Fig. 4B). In ARN3, the mutation completely abolished the catalytic activity, whereas in TpABN the activity was reduced up to ϳ60% in the presence of calcium ions (Fig. 4B). This residual activity observed in TpABN can be interpreted as a result of the stronger coordination of calcium ions in comparison to sodium ions, being less labile to solvent perturbations caused by deletion of the aromatic residue. Therefore, these results suggest that the conserved aromatic residue acts as a gateway controlling the solvent exchange rate at this pocket and, independent of calcium or sodium ions coordinating the catalytically relevant histidine, the solvent dynamics at this pocket were shown to be relevant for catalysis.
An Unusual Catalytic Interface Accommodates Branched Substrates-The crystallographic structure of ARN2, a ruminal endo-acting two-domain enzyme, showed a classical overall fold of the GH43 family (Fig. 1B) with all the catalytically relevant residues (10) conserved including the three acidic residues involved in the catalytic reaction (base, Asp 126 , acid, Glu 376 , and pK a modulator, Asp 308 ) and the ion-polarizable histidine (His 125 ). ARN2 contains four main amino acid segment insertions in its sequence compared with other two-domain ABNs such as TpABN, YxiA from B. licheniformis, and endo-arabinanase 2 from Bacillus subtilis (BsArb43B). Interestingly, three are located at the opposite side of the accessory domain (Glu 178 -Ser 198 , Asn 270 -Arg 302 , and Val 346 -Tyr 362 ) forming an unexpected motif (Fig. 1C), which along with the Asn 237 -Ser 243 loop, generates a unique catalytic interface (Fig. 1D). These structural differences result in a larger catalytic interface with higher solvent accessibility in comparison to other structurally characterized GH43 ABNs, supporting the fact that ARN2 efficiently degrades branched arabinans (22), the natural substrate of these enzymes. Indeed, the majority of ABNs characterized to date display tight specificity for linear arabinan chains, exhibiting very low activity against branched forms of the polysaccharide.
Structural Basis for the Exo-operating Mode of ARN3-The exo-acting GH43 ABNs, CjArb43A and ARN3, attack the same substrate, arabinan; but they release different products, which are arabinotriose for CjArb43A (8) and arabinose for ARN3 (18). The exo-activity of CjArb43A is associated with strong affinity for the free O5 atom of the non-reducing terminal arabinose at the Ϫ3 subsite (8). The removal of two residues that make hydrogen bonds with the hydroxyl group at the C5 position in the ligand-complexed structure (Asp 35 and Gln 316 ) induced the enzyme to degrade in an endo mode (8). The residues forming the Ϫ3 subsite of ARN3 are Asp 37 and Ser 329 , similar to Asp 35 and Gln 316 of CjArb43A. Therefore, to test if those residues also contribute to the exo action mode of ARN3, three mutants were produced, D37W, S329P, and D37W/ S329P. However, all of them presented a similar cleavage pattern to that observed for the WT protein, releasing arabinose from arabinooligosaccharides (results not shown). If ARN3 releases arabinose in a similar manner as CjArb43A, there would be unique residues recognizing the chain end after the Ϫ1 subsite, but it is not the case because all residues near the Ϫ2 subsite (His 39 , Thr 55 , and Trp 96 in ARN3) are fully conserved in endo-acting ABNs (Fig. 5A). SaAraf43A, an ABF from Streptomyces avermitilis (50), also releases arabinose as ARN3; however, ARN3 cannot be considered an ABF because it does not cleave pNP-Araf (18). Moreover, the exo-acting manner of SaAraf43A was attributed to the strict pocket architecture of the Ϫ1 subsite (50), whereas ARN3 has a substrate-binding channel with additional Ϫ2 and Ϫ3 subsites (Fig. 4A).
The crystal structure of ARN3 revealed an intriguing structural feature at the aglycone-binding region that is a long insertion in the loop connecting the second and third blades of the ␤-propeller (Fig. 5A). This loop typically comprises 4 or 6 residues in other GH43 ABNs (Fig. 5B); however, in ARN3, it is 28 residues long (Arg 203 -Ala 230 ). It is partially disordered (Arg 207 -Asp 227 ) in the crystal structure of ARN3, but the proximity with positive subsites suggest a role as a physical barrier in the substrate-binding channel, implying that the exo-activity of ARN3 is related to the reducing end of the arabinan chain. All known exo-acting GH43 enzymes attack the non-reducing end of the polysaccharide (8, 12, 13, 50 -52) and few exo-enzymes that attack the reducing end have been reported to date. For instance, the exo-oligoxylanase (EC 3.2.1.156) from Bacillus halodurans C-125 (Rex) is a GH8 enzyme that cleaves xylooligosaccharides to release xylose from the reducing end (53). The exo-activity of this enzyme is determined by a physical barrier and chemical recognition. The ϩ2 subsite is blocked by a kink in the loop before the ␣10 helix and His 319 , at the tip of this loop, forms a direct hydrogen bond with the ␤-hydroxyl group of the xylose at the ϩ1 subsite, the reducing end of the polysaccharide (54). Molecular dynamics simulations of ARN3 in the native state and complexed with substrate showed no structural changes in this loop, indicating a naturally occluded aglycone region of the substrate binding channel. Moreover, in the substrate-complex model, the Asp 227 side chain is hydrogen bonded to the O1 atom from the hydroxyl group of arabinose (Fig. 5A), which limits the substrate channel to the ϩ1 subsite and explains the cleavage pattern with arabinose as final product. Based on these observations, we propose that elongation of this loop results in topographical adaptations of the catalytic interface, which limits the channel to the ϩ1 subsite and determines the exo-mode of action of this novel enzyme. This ability to attack reducing ends of the substrate within a GH43 member is of extreme relevance for outright chemical recognition and synergism in arabinan degradation. To date, all characterized GH43 exo-acting enzymes cleave from the non-reducing end of the polysaccharide and the use of both exo-enzymes that preferentially attack either the reducing or the non-reducing end is an important synergism observed for exocellulases (55).
Redesigning the Action Mode of ARN3-According to our structural analysis, the Arg 203 -Ala 230 loop is related to the action mode of the ruminal single-domain enzyme (ARN3). To test this hypothesis, we redesigned the catalytic interface of the protein, replacing this long loop insertion by the amino acid segment corresponding to classical endo-acting ABNs (sequence "SRGEEP"). The chimera SRGEEP-ARN3, heterologously produced, was fully functional and presented a nativelike structure. The chimera preserved the temperature and pH optimum (results not shown), but increased the catalytic efficiency in more than 50% (Fig. 5C). The loop replacement exposed additional subsites of the substrate binding channel, increasing the affinity for the substrate, as observed by the lower K m of the chimera (Fig. 5C). Moreover, extraction of the physical barrier represented by the loop favored the turnover of the substrate, indicated by the higher k cat . The main result of the loop exchange is the modification in the cleavage pattern. SRGEEP-ARN3 generates principally arabinobiose and arabinotriose as the product from linear arabinan (Fig. 5D), whereas cleavage by the WT enzyme produces exclusively arabinose (Fig. 5E). Arabinoheptaose was also used as substrate for the WT enzyme and the SRGEEP-ARN3 chimera. As expected, ARN3 successively removed arabinose from the oligosaccharide, sequentially producing hexaose, pentaose, tetraose, triose, and biose (Fig. 5F). The chimera also produced a range of oligosaccharides (Fig. 5G), but the ratio of the products formed at the beginning of the reaction indicates that the enzyme randomly cleaved the oligosaccharide. Moreover, differently from the WT enzyme, di-, tri-, and tetrasaccharides instead of monoand disaccharides, accumulated in late stages of the reaction with SRGEEP-ARN3 (Fig. 5, F and G, respectively). Therefore, the replacement of the Arg 203 -Ala 230 region by a short segment in the chimera SRGEEP-ARN3 clearly led to a change in the enzyme cleavage pattern, indicating that this long loop is responsible for the exo-activity of this ruminal single-domain enzyme.

A Mechanistic Model for GH43 ABN Activation-GH43
ABNs operate by a reaction mechanism with inversion of anomeric configuration using the canonical three acidic residues, Asp 32 (the base), Glu 233 (the acid), and Asp 170 (the pK a modulator) (Fig. 4F). Asp 32 is deprotonated to activate the catalytic water that promotes a single-displacement attack on the anomeric carbon (56) and Glu 223 protonates the glycosidic oxygen making the aglycone a better leaving group. For that, the Glu 223 pK a is maintained higher than usual by Asp 170 (12). However, in contrast to classical glycoside hydrolases, ABNs contain a fourth key residue for catalysis, an interfacing histidine (His 314 in TpABN) that mediates the interaction with the metallic cofactor (Fig. 4F). This residue is strategically located between the two catalytic residues (3.1 and 5.1 Å from Asp 32 and Glu 223 , respectively) and interacts via its N⑀2 atom with calcium ion (or sodium ion) located at the hydrated pocket. The rotameric configuration of the histidine ring is maintained perpendicular to the catalytic acid residues by interaction with the ion, preventing a specific interaction with these residues that could disrupt the geometry of the active site. This network of interactions and the fact that histidine mutation abolishes the catalytic activity of GH43 enzymes (11) support the crucial role of this residue in the catalysis. Moreover, the replacement of His by Gln, which should preserve the ability to make hydrogen bonds with the substrate, abolished the enzymatic activity (11), indicating a particular correlation between the metallic cofactor and histidine. These observations suggest that the positively charged ion might have a hyper-polarizing effect on the imidazole side chain of His 314 via the N⑀2 atom resulting in an electrophilic character of the His 314 -N␦1 atom. Consequently, it might influence the charge distribution in the active site microenvironment and the protonation states of catalytic acidic residues, contributing to the functional state of catalytic residues. This hyperpolarizing effect also explains the fact that the His 3 Gln mutation terminated the catalytic activity of BsArb43B (11), because the amide side chain of Gln is not polarizable by metallic ions. Therefore, these findings indicate that the ion-induced hyperpolarization of histidine is essential to maintain an intact active site configuration for catalysis.
New Routes for Arabinan Degradation-This study revealed the molecular bases of the unique functional properties for outright chemical recognition and synergism in arabinan degradation of the ruminal ABNs, ARN2, and ARN3. ARN2 efficiently degrades naturally abundant branched substrates due to profound topographical adaptations in the active site interface conferring higher accessibility, whereas all known orthologs have preference for linear substrates. ARN3 is the first GH43 member with the ability to attack the reducing ends of the substrate. This rare exo-acting operation mode is achieved by imposing a physical barrier to the aglycone subsites, as con-firmed by structure-guided redesign of the active site, which converted the ruminal exo-enzyme into a classical endo-acting glycosidase. The combined use of exo-enzymes that preferentially attack either the reducing or the non-reducing ends of the substrate is an important synergism already observed for exocellulases and now also viable for breakdown of hemicellulosic polysaccharides. Both enzymes with their unique action modes provide new routes for arabinan degradation that can impact on technologies for conversion of plant cell-wall polysaccharides into energy and chemicals.