Structural and Thermodynamic Insights into Chitooligosaccharide Binding to Human Cartilage Chitinase 3-like Protein 2 (CHI3L2 or YKL-39)*

Background: Human YKL-39 is currently recognized as a biomarker for osteoarthritis. Results: Crystal structures of YKL-39 reveal that chitooligosaccharide induces local conformational changes to stabilize sugar·protein complexes and that the protein contains five binding subsites for sugars. Conclusion: YKL-39 binds to chitooligosaccharide through enthalpic reactions. Significance: Our findings suggest how YKL-39 interacts with GlcNAc moieties of the natural ligands, which may possibly activate local tissue inflammation. Four crystal structures of human YKL-39 were solved in the absence and presence of chitooligosaccharides. The structure of YKL-39 comprises a major (β/α)8 triose-phosphate isomerase barrel domain and a small α + β insertion domain. Structural analysis demonstrates that YKL-39 interacts with chitooligosaccharides through hydrogen bonds and hydrophobic interactions. The binding of chitin fragments induces local conformational changes that facilitate tight binding. Compared with other GH-18 members, YKL-39 has the least extended chitin-binding cleft, containing five subsites for sugars, namely (−3)(−2)(−1)(+1)(+2), with Trp-360 playing a prominent role in the sugar-protein interactions at the center of the chitin-binding cleft. Evaluation of binding affinities obtained from isothermal titration calorimetry and intrinsic fluorescence spectroscopy suggests that YKL-39 binds to chitooligosaccharides with Kd values in the micromolar concentration range and that the binding energies increase with the chain length. There were no significant differences between the Kd values of chitopentaose and chitohexaose, supporting the structural evidence for the five binding subsite topology. Thermodynamic analysis indicates that binding of chitooligosaccharide to YKL-39 is mainly driven by enthalpy.

Because YKL-39 is closely related in size and sequence to YKL-40, it was named following the convention for that homolog, which is based on the following three N-terminal amino acid residues, tyrosine (Tyr), lysine (Lys), and leucine (Leu), and an apparent molecular mass of 39 kDa. YKL-39 is secreted from articular chondrocytes (13). YKL-39 mRNA has been detected in lung, heart, and glioblastoma but not in brain, spleen, or pancreas (13,16). YKL-39 mRNA was also detected in macrophages that were strongly stimulated by a combination of IL-4 and TGF-␤ (17). YKL-39 is currently recognized as a specific biomarker for the activation of chondrocytes and for the progress of osteoarthritis (18 -20), a degenerative joint disease involving the degradation of articular cartilage and subchondral bone that globally affects 25% of adults aged over 65 years (21). Real time PCR and DNA microarray analyses showed that YKL-39 mRNA was significantly up-regulated in the cartilage of patients with advanced osteoarthritis (19). Moreover, the level of YKL-39 mRNA expression was positively correlated with collagen type 2 up-regulation in both early and late stages of the disease (20). YKL-39 was also found to induce an autoimmune response in patients with rheumatoid arthritis (22,23), as well as in a rheumatoid arthritis mouse model (24). A recent study in human embryonic kidney (HEK293) and human glioblastoma (U87 MG) cells showed that YKL-39-activated signal transduction was regulated through the phosphorylation of ERK1/ERK2 kinases (25). YKL-39 was later reported to enhance cell proliferation, colony formation, and type II collagen expression in mouse chondrogenic ATDC5 cells (26). It may therefore act as a novel growth/differentiation factor for articular cartilage chondrocytes, which regulate joint homeostasis in adults. However, the mechanistic details of how YKL-39 regulates cell proliferation and cell differentiation remain to be identified.
Previously, the crystal structure of an N35Q mutant of YKL-39, bound to GlcNAc 6 , was reported (27). In the work reported here, we employ both crystallographic and thermodynamic studies to evaluate the binding of chitooligosaccharides of different lengths to YKL-39. Inspection of the crystal structures of YKL-39 in the absence and presence of chitooligosaccharides clearly indicates that YKL-39 binds specifically to chitooligosaccharides and that the binding strength is dependent on the length of the chitin chain. These structural data are discussed in relation to binding affinity, subsite topology, and the thermodynamic contributions to the sugar-protein interactions.

MATERIALS AND METHODS
Gene Cloning-The nucleotide sequence of the full-length CHI3L2 gene encoding chitinase 3-like protein 2 or YKL-39 was retrieved from the GenBank TM database (accession number NM_004000), and the gene was amplified from a human cDNA template by the PCR technique (GeneScript Corp.) and cloned into the pET32a(ϩ) expression vector. The recombinant YKL-39 was expressed as a fusion protein containing a cleavable thioredoxin (Trx) fragment, followed by a hexahistidine tag at the N terminus of the YKL-39 polypeptide (28). The Trx fragment was fused to the protein to increase its solubility, and the His 6 tag was included to aid purification. Nucleotide sequences of both sense and antisense strands of the CHI3L2 fragment were confirmed by automated DNA sequencing (First Base Laboratories, Malaysia).
Recombinant Protein Expression and Purification-Recombinant YKL-39, lacking the 26-amino acid signal sequence, was expressed at high levels in Escherichia coli BL21 (DE3) (28). YKL-39-expressing cells were harvested by centrifugation, resuspended in lysis buffer (50 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM PMSF, 1 mg⅐ml Ϫ1 lysozyme and 1% (v/v) Triton X-100), and then lysed on ice using a Sonopuls Ultrasonic homogenizer with a 20-mm diameter probe. The crude supernatant obtained after centrifugation at 19,000 rpm for 1 h was filtered through a 0.45-m cutoff membrane filter (Millipore), and then applied to a HisTrap TM HP (1.0 ϫ 5.0 cm) pre-packed column (GE Healthcare), connected to an Ä KTAprime TM Plus system (GE Healthcare). The column was equilibrated with 10 column volumes of equilibration buffer (50 mM Tris-HCl, pH 8.0, 50 mM NaCl), with a constant flow rate of 1 ml⅐min Ϫ1 . After the column was thoroughly washed, bound protein was eluted with 10 column volumes of 250 mM imidazole, pH 8.0, and imidazole was then removed with a HiPrep TM 26/10 desalting column (GE Healthcare). The purified Trx/His 6 /YKL-39 fusion protein was subsequently treated with enterokinase, following the manufacturer's instruction (GenScript), to remove the fusion tag. The Trx/His 6 segment was resolved from the YKL-39 polypeptide using a HisTrap TM HP (1.0 ϫ 1.0 cm) prepacked column (GE Healthcare). Unbound fractions containing YKL-39 were pooled, concentrated, and then further purified to homogeneity using a HiLoad TM 16/60 Superdex TM 200 preparation grade gel filtration column (GE Healthcare). Fractions containing highly purified YKL-39 were combined and then exchanged into 20 mM Tris-HCl buffer, pH 8.0. The pooled fraction was then concentrated to 10 -20 mg⅐ml Ϫ1 using a Vivaspin TM membrane concentrator. Protein concentrations were determined by the Pierce TM BCA assay (Novagen, Darmstadt, Germany). Aliquots of the purified YKL-39 were flashfrozen in liquid N 2 and then stored at Ϫ30°C.
Site-directed Mutagenesis-The recombinant plasmid pET32a-(ϩ)/CHI3L2 was used as DNA template in PCR-based site-directed mutagenesis. For the W36A mutant, the forward and reverse primers were, respectively, 3Ј-GTTTGCTACTTTAC-CAACGCATCCCAGGACCGGCAGGAACC-5Ј and 5Ј-GGT-TCCTGCCGGTCCTGGGATGCGTTGGTAAAGTAGCA-AAC-3Ј. For the Y243A mutant, the forward and reverse primers were, respectively, 3Ј-GACAGAGGGCCAAGCTCCTACG-CAAATGTGGATATGCTGTGGGG-5Ј and 5Ј-CCCCACA-GCATATTCCACATTTGCGTAGGAGCTTGGCCCTCT-GTC-3Ј. For the W360A mutant, the forward and reverse primers were, respectively, 3Ј-CCTGGGAGGAGCCATGATCGC-CTCTATTGACATGGATGAC-5Ј and 5Ј-GTCATCCATGT-CAATAGAGGCGATCATGGCTCCTCCCAGG-3Ј. The underlined sequences indicate the mutated codons. Site-directed mutagenesis was performed following the QuikChange sitedirected mutagenesis protocol of Stratagene. The DpnI-treated DNA was transformed into E. coli XL1-Blue competent cells. The recombinant plasmids obtained from positive colonies were extracted using QuickClean II plasmid miniprep kits (GenScript, Piscataway, NJ) and were then re-transformed into E. coli DH5␣ cells. To verify that mutations were correct, the nucleotide sequences of the sense and antisense strands of the PCR fragment were determined by automated sequencing (First BASE Laboratories Sdn Bhn, Selangor Darul Ehsan, Malaysia). The mutant proteins were expressed and purified using the same protocol as for the wild-type protein.
Protein Crystallization-Initial crystallization screens were set up using a Screenmaker 96 ϩ 8 TM Xtal (Innovadyne Technologies Inc.) with sitting drop CrystalQuick TM plates (Greiner Bio-one, Germany). For each crystallization drop, 170 nl of freshly prepared YKL-39 (12.75 mg⅐ml Ϫ1 dissolved in 20 mM Tris-HCl, pH 8.0) was added to an equal volume of each precipitating agent from three screening kits, including Wizard I and II (Emerald BioSystems) and Crystal Screen HT TM (Hampton Research). Crystal optimization was performed using the hanging drop vapor diffusion method. After mixing equal vol-umes of protein and mother liquor, hexagonal single crystals were observed after 1 day at 25°C under the condition of 30% (w/v) PEG 3350, 0.2 M Li 2 SO 4 , 0.1 M BisTris, pH 5.5. The crystals were allowed to grow for 1 week. For crystal complexes, the YKL-39 crystals were soaked with chitooligosaccharides overnight at 25°C, using optimized concentrations (0.1 mM for GlcNAc 5 and GlcNAc 6 , 5 mM for GlcNAc 4 , 10 mM for GlcNAc 3 , and GlcNAc 2 ) prepared in the mother liquor and then flashfrozen in liquid nitrogen for subsequent x-ray diffraction analysis.
Data Collection, Processing, and Structure Determination-The YKL-39 crystals, either ligand-free or complexed with chitooligosaccharides, were exposed to 1.00 Å wavelength x-rays at the BL13B1 beamline, National Synchrotron Radiation Research Center, Taiwan. Data were collected on an ADSC Quantum 315 CCD detector. All diffraction data were indexed, integrated, and scaled using the program HKL2000 (29), and molecular replacement was employed to obtain phase information using the program MOLREP from the CCP4 suite (30). The structure of YKL-39 bound to GlcNAc 2 was solved using the previously published structure of YKL-39 in complex with GlcNAc 6 (PDB code 4AY1) as the search model (27). Other data sets, including ligand-free YKL-39 and YKL-39 in complex with GlcNAc 4 and GlcNAc 6 , were solved using the final structure of the YKL-39⅐GlcNAc 2 complex as the model for rigid body refinement. The analyses of the electron density map F obs Ϫ F cal and 2F obs Ϫ F cal and model building were carried out in COOT (31) and restrained refinement in REFMAC5 within CCP4 (32) and Phenix (33). The geometry of each final model was validated by PROCHECK (34). Evaluation of the secondary structure indicated no residues in the outlier regions of the Ramachandran plots. However, the amino acid residue at position 318 was found to be Trp instead of Arg. This discrepancy has been suggested to arise from a single nucleotide polymorphism, which is related to tissue specificity (35). The final 2F obs Ϫ F cal omit map, contoured at 1.0 , clearly showed the electron density maps for GlcNAc 2 , GlcNAc 4 , and GlcNAc 6 with full occupancy. The structures and electron density maps of all the refined structures were created and displayed by PyMOL (36) and LIGPLOT (37). Atomic coordinates and structure factors of the final models of YKL-39 have been deposited in the Protein Data Bank with PDB accession numbers 4P8U for apo-YKL-39, 4P8V for the YKL-39⅐GlcNAc 2 complex, 4P8W for the YKL-39⅐GlcNAc 4 complex, and 4P8X for the YKL-39⅐GlcNAc 6 complex.
Binding Study by Isothermal Titration Calorimetry (ITC)-Binding of chitooligosaccharides GlcNAc 2 to GlcNAc 6 to YKL-39 was investigated by ITC. This technique measures heat released or absorbed during the binding event, providing information about binding thermodynamics and yielding the stoichiometry (n), equilibrium binding association constant (K a ), enthalpy change (⌬H), Gibb's free energy (⌬G), and the entropy change (⌬S) of the reaction in a single experiment (38,39). Experiments were performed at 25°C. ITC experiments were carried out at least three times using the ITC-200 system (Microcal Inc) at 25°C with a stirring speed of 260 rpm. For experiments with GlcNAc 5 and GlcNAc 6 , 4 l of 0.25 mM chitosugar was injected into the 300-l calorimeter cell, contain-ing 20 mM potassium phosphate buffer, pH 8.0, and 10 M purified YKL-39. The injections were repeated 29 times over 140-s intervals. The background was measured by injecting the corresponding ligand into the cell containing only the buffer. Experiments with GlcNAc 2 , GlcNAc 3 , and GlcNAc 4 were performed as described above, but the concentration of each protein/sugar reaction was re-optimized as follows: 30 M YKL-39 and 0.45 mM GlcNAc 4 ; 15 M YKL-39 and 3 mM GlcNAc 3 ; and 15 M YKL-39 and 4 mM GlcNAc 2 . The ITC data were collected and analyzed using the Microcal Origin version 7.0 software. The ITC profile obtained by injecting the corresponding ligand into the reaction cell containing buffer without YKL-39 was subtracted from the corresponding data set. The resultant data were fitted by a single-site binding model in the nonlinear least square algorithm. The thermodynamic parameters, including binding stoichiometry (n), the equilibrium binding association constant (K a ), and the enthalpy change (⌬H) were subsequently evaluated. The Gibb's free energy (⌬G) and the entropy change (⌬S) were calculated from the relationship shown in Equation 1, where R is the gas constant (1.98 cal⅐K Ϫ1 mol Ϫ1 ) and T the absolute temperature in kelvin.
Binding Study by Fluorescence Spectroscopy-Each purified YKL-39 variant was titrated with different concentrations of the chitooligosaccharides in 20 mM Tris-HCl, pH 8.0, at 25°C. Changes in intrinsic tryptophan fluorescence were monitored directly in an LS-50 fluorescence spectrometer (PerkinElmer Life Sciences). The excitation wavelength was set at 295 nm, and emission intensities were collected over 300 -450 nm with excitation and emission slit widths of 5 nm. For the wild-type YKL-39 and the mutant Y243A, a fixed amount of protein (25 g) was titrated with 100 mM GlcNAc 2 , 10 mM GlcNAc 3,4 , or 0.1 mM GlcNAc 5,6 . Much higher concentrations of ligands were required for titrating the mutants W36A and W360A (100 mM GlcNAc 2,3,4 and 25 mM GlcNAc 5,6 ) were used. Each protein spectrum was corrected for the buffer spectrum. Binding curves were evaluated using a nonlinear regression function available in Prism version 5.0 (GraphPad Software), following the single-site binding model shown in Equation 2, where ⌬F is the difference between fluorescence intensity before and after titration with the sugar ligand; F max refers to the maximum emission intensity; F min is the minimum emission intensity; L 0 is the initial concentration of ligand; and K d is the equilibrium dissociation constant (micromolar).

RESULTS AND DISCUSSION
Sequence Analysis-Human mature wild-type YKL-39, lacking the 26-amino acid signal sequence, was cloned and functionally expressed in E. coli as a Trx/His 6 fusion protein (28). The amino acid sequence of the recombinant YKL-39 was identical to the 390-amino acid sequence of CHI3L2 isoform 1 (identifier Q15782-4) reported in the UniProtKB/Swiss-Prot database, with the sole exception that the amino acid at position 318 was Trp instead of Arg. The divergence of this amino acid has been suggested to arise from a genetic variation, which naturally occurs through a single nucleotide polymorphism that is tissue-specific (35). After enterokinase cleavage to remove the Trx/His 6 fragment, the recombinant YKL-39 contains seven extra N-terminal residues (AMADIGS), and the intact polypeptide has a predicted mass of 41.5 kDa.
Overall Structures of YKL-39 in the Absence and Presence of Ligands-Recombinant YKL-39, purified to homogeneity, was subjected to crystallization trials. This E. coli expressed human YKL-39 readily crystallized in PEG 3350. Initial crystallographic analysis showed that the YKL-39 crystals belong to the space group P4 1 2 1 2, with one molecule in the asymmetric unit. Four crystal structures of YKL-39 were determined, including the apo-form and the complexes with GlcNAc 2 , GlcNAc 4 , and GlcNAc 6 . Table 1 summarizes the data collection and refinement statistics of the final models of the YKL-39 structures. YKL-39 in complex with GlcNAc 2 was refined against the highest resolution data at 1.53 Å. The overall structure of YKL-39 includes two conserved domains (Fig. 1A) that are found in all GH-18 chitinases and chitinase-like proteins. The major (␤/␣) 8 TIM barrel domain (domain I) comprises eight parallel strands B1-B8 (purple), alternating with eight helices A1-A8 (cyan) as shown in Fig. 1B. It is noticeable that helices A1, A3, and A6 are broken and contain short helices: G1-1 in helix A1; G3-1 and G3-2 in helix A3; G6-1 and G6-2 in helix A6. The TIM barrel domains of GH-18 glycosyl hydrolases are known to interact specifically with chitin oligosaccharides (40 -45). The second domain is termed the small ␣ϩ␤ insertion domain (domain II) (Fig. 1A, gray) and is located between the tail of strand B7 and the start of helix A7. It is made up of six anti-parallel strands connected by a short helix that forms a typical greek key motif. The exact function of this domain is unknown, but certain amino acid residues in this domain help to stabilize the sugar⅐protein complex (41,43,46).
Detailed structural analysis revealed that YKL-39 contains two disulfide bonds in the TIM barrel domain (data not shown).   Close inspection of the GlcNAc-binding subsites of the apoprotein ( Fig. 2A, blue) in comparison with that of the protein in complex with the longest chitooligosaccharide GlcNAc 6 (Fig. 2, cyan) reveals considerable movements of loop L 1 on the surface of subsite Ϫ3, loops L 3 and L MЈ (see Fig. 1B for loop assignment) near subsites Ϫ2 and Ϫ1 and L 6 near subsites ϩ2 and ϩ3, in the structure with GlcNAc 6 . As compared with the exterior of the native protein (Fig. 2B), four key residues were found to shift significantly toward the center of the chitin-binding cleft as follows: Tyr-104 and Leu-105 (part of the bottom loop L 3 ), and Met-364 and Phe-301 (part of the top loop L MЈ ). This narrows the cleft of the sugar-bound protein around subsites Ϫ2/Ϫ1 (Fig. 2C) by 1.3 Å. Such local movements engender close contacts between the GlcNAc rings and the binding residues around the corresponding subsites. In contrast, Tyr-243 (part of loop L 6 ) rotates away from its original position, widening the cleft at this particular subsite (ϩ3) by 1.9 Å in comparison with that of the unliganded structure (Fig. 2D). This orientation of Tyr-243 increases access to subsite ϩ2 from the subsite ϩ3 direction, suggesting that longer sugars may be accommodated in this conformation. Taken together, these structural differences provide evidence that the binding of chitooligosaccharides induces local structural changes that strengthen the interactions between YKL-39 and chitosugars. and helix A7 is shown in gray. The chitooligosaccharide GlcNAc 6 is shown as ball and sticks and colored by atoms with yellow for carbon, blue for nitrogen, and red for oxygen. RE and NRE denote the "reducing" and "nonreducing" ends of the sugar chain. B, structural topology of YKL-39. The eight strands are referred to as B1 to B8, the loops as L, and the eight ␣-helices connecting the ␤-strands are referred to as A1 to A8. The helices are often segmented. Short helices are referred to by the letter G, for instance G1-1 indicating the short helix within the region of the A1 helix.
Specific Interactions of YKL-39 with Chitooligosaccharides-As in all GH-18 members, the sugar-binding cleft of YKL-39 comprises multiple subsites for GlcNAc units. The critical binding features of these subsites are aromatic residues, which bind the GlcNAc rings mainly through hydrophobic interactions, and polar amino acid side chains, which form hydrogen bonds with the saccharide chain. Inspection of the substratebinding cleft of unliganded YKL-39 (Fig. 3A) and YKL-39 in complex with GlcNAc 2 (Fig. 3B), GlcNAc 4 (Fig. 3C), or GlcNAc 6 (Fig. 3D) reveals the binding behavior of individual chitooligosaccharides. GlcNAc 2 was found at the center of the binding cleft between subsites Ϫ2 and Ϫ1 (Fig. 3B). As summarized in Table 2, subsites Ϫ2 and Ϫ1 include a large number of binding residues. Residues Phe-63, Leu-105, Phe-301, Trp-360, and Met-364 interact with the GlcNAc ring at subsite Ϫ2, whereas Tyr-32, Ser-143, Tyr-104, Ile-145, Asp-213, Tyr-269, and Trp-360 form subsite Ϫ1. The residues around the center of the binding cleft make highly ordered interactions, which suggest that an arriving chitin chain would initially interact preferentially at the center of the binding groove. For a longer chitooligosaccharide, the interaction may become extended through occupation of the neighboring weaker affinity subsites, where the four sugar rings in GlcNAc 4 were extended from subsites Ϫ3 to ϩ1 (Fig. 3C), and in the complex with GlcNAc 6 from subsites Ϫ3 to ϩ3 (Fig. 3D). This affinity gradient is supported by the B-factor being lowest for Ϫ2GlcNAc (43.63 Å 2 ), indicating high rigidity through tight interactions at the internal site. Increases in B-factor for bound sugar at the extending subsites (Ϫ1GlcNAc, 44.76; ϩ1GlcNAc, 44.75; ϩ2GlcNAc, 52.48; ϩ3GlcNAc, 81.64; and Ϫ3GlcNAc, 64.28 Å 2 ) suggest greater flexibility of the sugar rings, which make weak interactions with the binding residues at such subsites.
We further analyzed the hydrophobic interactions between chitohexaose GlcNAc 6 and aromatic/hydrophobic residues surrounding subsites Ϫ3 to ϩ3 of the YKL-39-binding groove (Fig. 4A). Three residues, Trp-36 at subsite Ϫ3, Trp-360 at subsite Ϫ1, and Trp-218 at ϩ2, were found to stack directly against the plane of the pyranose rings of the GlcNAc units, and these interactions are expected to contribute significantly to binding at these subsites. The first two residues (Trp-36 and Trp-360) are completely conserved in other GH-18 homologs and act as key binding residues (42,43,45). Fig. 4B shows a number of hydrogen bonds that help to stabilize the sugar⅐protein complexes. Such hydrogen bonds are formed either directly between the sugar rings and the binding residues  or are mediated by water molecules. As summarized in Table 2, high densities of interactions are seen at subsites Ϫ2, Ϫ1, and ϩ1, which are likely to determine the preferential binding at these sites.
Structural Comparison with Other Human GH-18 Members-Superimposition of the structure of the wild-type YKL39⅐ GlcNAc) 6 complex obtained in this study on that of YKL39⅐GlcNAc 6 complex (4AY1) reported by Schimpl et al. (27) results in a root mean square deviation of 0.22 for C␣ posi-tions over 334 atoms (Fig. 5). Thus, the two structures are essentially identical even though they are derived from crystals with different space groups. The differences between the two protein sequences are at residues 35 and 318.
In the structure by Schimpl et al. (27), Asn-35 was mutated to Gln to prevent glycosylation, generating mutant N35Q. In this study, YKL-39 was functionally expressed in E. coli as a native, nonglycosylated protein. In our structure, Asn-35 makes three salt bridges, two with Tyr-61 and Asp-75 and the third with a neighboring water molecule (Fig. 5A, left panel, inset). Gln-35 in Schimpl's structure flips to a vertical position and forms salt bridges with Tyr-61, Lys-74, and also a neighboring water molecule. Although residue 35 in both structures lies in subsite Ϫ3, it makes no contact with the GlcNAc ring at the corresponding subsite. Residue 318, Trp in native YKL-39, lies in a region of the protein distant from the chitin-binding groove. In our structure, Trp-318 forms hydrophobic interactions with Pro-325 and the stalk of Lys-340, although in structure 4AY1 Arg-318 forms a salt bridge with Asp-338 (Fig. 5A, right panel, inset). These are standard interactions between residues, in line with the view that these substitutions arise from naturally occurring polymorphisms. In our structure, all six GlcNAc rings could be fitted into subsites Ϫ3 to ϩ3 (Fig. 5B), whereas only four GlcNAc rings of chitohexaose GlcNAc 6 were fitted into subsites Ϫ2 to ϩ2 in the 4AY1 structure (Fig. 5C). The sugar rings found in both structures are well aligned in bent conformation. The Ϫ1GlcNAc rings of both sugar chains adopted the boat conformation, with the torsion angles of the C ␣ backbone being essentially identical.
The sugar-binding grooves of YKL-39, YKL-40 (42,43), and CHIT1 (47) are very similar (Fig. 6A). Nevertheless, a few significant differences may explain divergent binding features between YKL-39 and other two homologs. The most obvious variation is at subsite Ϫ4. In CHIT1 and YKL-40, this subsite has Tyr-34 as the key residue forming a hydrophobic stacking interaction with Ϫ4GlcNAc. This residue is missing in YKL-39, and the closest residue, Asp-39, is unlikely to make productive contact with a GlcNAc ring ( Table 2). As a result, no affinity at subsite Ϫ4 is likely in YKL-39. Another considerable difference is observed at subsite ϩ3. In YKL-40, this subsite contains Trp-212, which stacks directly against the plane of ϩ3GlcNAc. In

TABLE 2
Summary of the interactions of (GlcNAc) 6

with residues in each subsite of YKL-39, YKL-40 and hCHIT
Bold indicates stacking interaction residues. Bold with underlines indicates recognition residues. Bold and italic indicates hydrogen bonding residues.  contrast, Tyr-243 in YKL-39 forms a hydrogen bond at a distance of 4.5 Å with the equatorial C6-OH group of ϩ3GlcNAc, suggesting that the interaction at this position is weaker than that in YKL-40. The Ϫ3GlcNAc unit at the nonreducing end of the GlcNAc 6 in YKL-39 (Fig. 6B) is well defined and interacts with Trp-36, whereas the ϩ3GlcNAc unit is more flexible, adopting a more open position than the GlcNAc ring that forms astack with Trp-212 at subsite ϩ3 in the YKL-40 structure (Fig. 6C). Weak affinity at the reducing end of the sugar chain is indicated by the large B-factor for ϩ3GlcNAc (81.64 Å 2 ), reflecting high flexibility due to loose fitting at this location. Within the YKL-40 structure, the chitosugar occupies two positions between subsites Ϫ4 to ϩ2 (Fig. 6C, green) and subsites Ϫ3 to ϩ3 (Fig. 6B, yellow) due to the more extended binding cleft. The residue at the position homologous to that of the catalytic residue Glu-140 in the human chitinase (CHIT1) lies at bottom of subsite Ϫ1 and is Ile-145 in YKL-39 and Leu-140 in YKL-40 (Fig. 6A). Thus, neither YKL-39 nor YKL-40 have catalytic activity, although the bent conformation of the interacting sugar, rendering it susceptible to cleavage in active chitinases, is maintained in all the CLPs (41-43, 45, 47). Binding Affinities and Five Major Subsite Topology-The enthalpic changes resulting from increasing chitooligosaccharide concentrations were measured using ITC, allowing the determination of binding affinities and inherent thermody-namic parameters for YKL-39 (39,40). The heat release profiles of YKL-39 were measured during titration with discrete concentrations of each chitooligosaccharide (Fig. 7A, upper panels). Secondary plots of the injection peaks were integrated, yielding the enthalpy change as kilocalories/mol of injectant, and plotted as a function of the molar ratio of GlcNAc n /YKL-39 (Fig. 7A, lower panels). All data obtained from the binding reactions were fitted using a single-site binding model with calculated stoichiometry (n) of 1.0, indicating that one molecule of the sugar interacts within the binding cleft of YKL-39. The suggested ratio of 1:1 of sugar/protein is consistent with that found in the crystal structures (Fig. 3, B-D), each of which shows only a single chitin chain. Theoretically, two or three molecules of GlcNAc 2 could bind within the chitin-binding groove of YKL-39, based on its size; however, only a single bound GlcNAc 2 is observed. This indicates that the subsites flanking the core subsites Ϫ2 and Ϫ1 have insufficient affinity to bind a second GlcNAc 2 molecule.

Protein
Analysis of the ITC data yields the equilibrium binding constants (K d ) for GlcNAc 2 , GlcNAc 3 , GlcNAc 4 , GlcNAc 5 , and GlcNAc 6 , which are 204, 142, 1.7, 0.06, and 0.04 M, respec-   tively ( Table 3). The decrease in the K d value clearly indicates increasing binding affinity with increasing chitooligosaccharide length. These binding parameters are consistent with the structural data, which show that GlcNAc 2 only partially occupies the binding cleft at subsites Ϫ2 and Ϫ1 (Fig. 3B), whereas GlcNAc 4 occupies subsites Ϫ3 to ϩ1. GlcNAc 6 stretches along the entire binding groove, occupying the six identified subsites and having the most stable binding of the chitooligosaccharides studied. Although the K d values decrease with increasing chain length for GlcNAc 2 , GlcNAc 3 , GlcNAc 4 , and GlcNAc 5 , those of GlcNAc 5 (0.06 M) and GlcNAc 6 (0.04 M) are about equal, within the standard deviations (Table 3, ITC). The structure of the complex of YKL-39 with GlcNAc 6 (Fig. 4, A and B) suggests strong interactions from subsites Ϫ3 to ϩ2, with Trp-36 forming a hydrophobic stack against Ϫ3GlcNAc and Trp-218 against ϩ2GlcNAc, although binding at subsite ϩ3 is weaker, because such interactions are absent. These binding characteristics were verified by data obtained from fluorescence spectroscopy. Changes in the fluorescence intensities of YKL-39/oligosaccharide solutions were measured at the maximum emission wavelength of 340 nm, with excitation at 295 nm. The fluorescence intensities were found to be dependent on the oligosaccharide concentration, indicating specific binding between the oligosaccharides and the protein. One example of this progressive fluorescence enhancement on binding of GlcNAc 6 is shown as a representative fluorescence profile (Fig. 7B, left top panel), which was analyzed and transformed to binding isotherms based on Equation 2. The fluorescence intensity data for each chitooligosaccharide were fitted reasonably well by the single-site binding model of a nonlinear regression function (Fig. 7B, middle and right top panels for GlcNAc 2 and GlcNAc 3 , respectively, and lower panel from left to right for GlcNAc 4 , GlcNAc 5 , and GlcNAc 6 , respectively). The estimated K d values of YKL-39 WT are summarized in Table 4. In good agreement with ITC data, the binding strengths are in the order GlcNAc 6 Х GlcNAc 5 Ͼ GlcNAc 4 Ͼ GlcNAc 3 Ͼ GlcNAc 2 .
The values of K d obtained from fluorescence measurements are in general higher than those from ITC measurements. This may reflect different reporting of binding by the two methods. In the fluorescence assay particular tryptophan residues lining the chitin-binding cleft are quenched during titrations with sugar. However, all interactions are accounted for in the ITC assay. The difference in K d values obtained by the two methods is small for short-chain chitooligosaccharides and becomes greater with increasing chain length and tighter binding. More importantly, the K d values for GlcNAc 2 derived from ITC and fluorescence methods are slightly different. This indeed indicates that Trp-360 that stacks against the facet of two GlcNAc rings at subsites Ϫ2/Ϫ1 plays an exclusive role in the sugar-protein interactions at such central subsites.
We further determined the role of three aromatic residues, Trp-36, Tyr-243, and Trp-360, in binding ligand. Trp-36 is located at subsite Ϫ3, Trp-360 at subsites Ϫ2/Ϫ1, and Tyr-243 at subsite ϩ3 of the protein-binding cleft (see Table 2 for the summary of interactions). These residues were mutated to alanine, generating single mutants W36A, Y243A, and W360A, respectively. The effects of mutation on binding affinity were accessed in a fluorescence quenching assay. The summary of their K d values compared with the wild-type values are shown in Table 3. Mutation of Tyr-243 caused only slight increases in K d for the long-chain sugars: GlcNAc 4 , GlcNAc 5 , and GlcNAc 6 , and caused no change in the K d values for GlcNAc 2 and GlcNAc 3 . The results suggest that Tyr-243 at subsite ϩ3 plays a minor role in proteinsugar interactions and had no great influence on the overall binding properties of YKL-39.
In marked contrast, mutations of Trp-36, and Trp-360 to Ala caused dramatic loss binding f the protein toward GlcNAc 2 and GlcNAc 3 , because no binding was detected at 100 mM sugar. The titration was not performed at higher concentrations due to solubility problems. The K d values of both mutants for GlcNAc 4 , GlcNAc 5 , and GlcNAc 6 considerably increased. Notably, W360A showed greater increases in the K d values than W36A. When compared with the WT values, the K d values for the mutant W360A were 64-and 102-fold increased for GlcNAc 5 and GlcNAc 6 , respectively. These result suggested that Trp-360 is crucial for maintaining tight binding at the center of the sugar-binding cleft (Fig. 3, B and  C). No binding was observed with GlcNAc 2 and GlcNAc 3 indicating that Trp-36 is also important for sugar-ligand interactions at subsite Ϫ3. Mutation of this residue perhaps affected binding of the sugar at the neighboring subsite Ϫ2. Nevertheless, mutation of Trp-36 exhibited less severe impact on GlcNAc 4 , GlcNAc 5 , and GlcNAc 6 binding, indicating that the hydrogen bond and hydrophobic networks formed by the remaining subsites are sufficiently robust to

parameters of chitooligosaccharide binding to YKL-39 derived from isothermal microcalorimetry
The best fit parameters, including the stoichiometry of binding (n), the equilibrium association constant (K a ), and the enthalpy change (⌬H), were estimated from nonlinear regression, using a single-site binding model, of the binding thermograms shown in Fig. 7A The equilibrium dissociation constant (K d ) was an inverse value of K a . The Gibb's free energy (⌬G) and the entropy change (⌬S) were calculated from Equation  5 and GlcNAc 6 suggest insignificant differences in the binding strengths for these two chitooligosaccharides (Fig. 7A). The results of these kinetic analyses support the crystal structure observations and suggest that YKL-39 has a less extended sugar-binding cleft than its closely related homologs, which is dominated by five recognition sites, namely (Ϫ3)(Ϫ2)(Ϫ1)-(ϩ1)(ϩ2), instead of the six subsites observed in CHIT1 and YKL-40 (42,43,47).
Analysis of Thermodynamic Parameters Involving Ligand Binding-A plot of the free energy change (⌬G) as the sum of negative ⌬H and positive ϪT⌬S demonstrates how the enthalpic and entropic parameters contribute to the binding by YKL-39 of chitooligosaccharides of increasing chain lengths (Fig. 8A). The negative ⌬G values obtained for all chitooligosaccharides are clearly influenced by the dominant negative ⌬H values, ranging from Ϫ11.3 to Ϫ17.2 kcal⅐mol Ϫ1 , although the entropic term (ϪT⌬S) is consistently unfavorable, with positive values ranging from ϩ2.0 to ϩ9.1 kcal⅐mol Ϫ1 (Fig. 8B and Table  3). This indicates that the chitooligosaccharide/YKL-39-binding reactions are mainly driven by enthalpic, rather than the entropic, factors. The dominant ⌬H term represents the specificity and strength of interaction derived from hydrogen bonding and electrostatic interactions between the two components (48 -52). A mechanistic basis for this phenomenon is provided by the crystal structures of the ligand⅐protein complexes, in which the sugar rings are held by varying numbers of hydrogen bonds (Fig. 4B). The largest enthalpy change measured was for GlcNAc 5 , suggesting that sugar specificity is mediated by five binding subsites, which is consistent with the structural evidence.
Conclusions-This study provides structural and thermodynamic insights into the binding of chitooligosaccharides to human YKL-39, a specific biomarker for osteoarthritis. Four crystal structures of YKL-39, in the absence and presence of chitooligosaccharides GlcNAc 2-6 , were solved. The YKL-39 structure contains a major (␣/␤/) 8 TIM barrel domain with a small insertion domain, similar to other GH-18 chitinases and CLPs. Superimposition of the crystal structures of ligand-free and ligand-bound YKL-39 suggests that binding of a chitooligosaccharide chain induces local conformational changes around the sugar-binding cleft that strengthen the sugar-protein interactions. The crystal structures of complexes with chitin fragments show that YKL-39 interacts with its sugar counterpart mainly through hydrophobic interactions, as well as a hydrogen bond network. ITC and fluorescence quenching data suggest that the protein binds chitooligosaccharides with very high affinity and that the binding strength increases with increasing length of the ligands. Both structural and thermodynamic evidence suggest that most of the binding is achieved with a fivesubsite topology. Analysis of thermodynamic parameters suggests that all chitooligosaccharides bind to YKL-39 through enthalpy-driven reactions.  Table 4 are plotted against the lengths of chitooligosaccharides. B, comparative plot of thermodynamic parameters (free energy, binding enthalpy, and entropy) for the binding of five chitooligosaccharides.

TABLE 4 Binding parameters of chitooligosaccharide binding to YKL-39 mutants as derived from fluorescence quenching assay
The purified YKL-39 (25 g) was titrated with a different range of concentrations, and changes in fluorescence emission intensity in the course of protein⅐ligand formation were monitored by fluorescence spectroscopy. The values are obtained from fluorescence quenching assay. The K d values (in micromolar) are presented as means Ϯ S.D. and were calculated from experiments carried out in triplicate. a ND represents no detectable change in fluorescence intensity up to the highest ligand concentration used in the titration. Note the highest concentration of each ligand was limited by its solubility in the selected buffered solution. b Values presented in parentheses are those reported by Schimpl et al. (27).