Characterization of recombinant amino-terminal NC4 domain of human collagen IX: interaction with glycosaminoglycans and cartilage oligomeric matrix protein.

The N-terminal NC4 domain of collagen IX is a globular structure projecting away from the surface of the cartilage collagen fibril. Several interactions have been suggested for this domain, reflecting its location and its characteristic high isoelectric point. In an attempt to characterize the NC4 domain in more detail, we set up a prokaryotic expression system to produce the domain. The purified 27.5-kDa product was analyzed for its glycosaminoglycan-binding potential by surface plasmon resonance and solid-state assays. The results show that the NC4 domain of collagen IX specifically binds heparin with a K(d) of 0.6 microm, and the full-length recombinant collagen IX has an even stronger interaction with heparin, with an apparent K(d) of 3.6 nm. The heparin-binding site of the NC4 domain was located in the extreme N terminus, containing a heparin-binding consensus sequence, whereas electron microscopy suggested the presence of at least three additional heparin-binding sites on full-length collagen IX. The NC4 domain was also shown to bind cartilage oligomeric matrix protein. This interaction and the association of cartilage oligomeric matrix protein with other regions of collagen IX were found to be heparin-competitive. Circular dichroism analyses of the NC4 domain indicated the presence of stabilizing disulfide bonds and a thermal denaturation point of about 80 degrees C. The pattern of disulfide bond formation within the NC4 domain was identified by tryptic peptide mass mapping of the NC4 in native and reduced states. A similar pattern was demonstrated for the NC4 domain of full-length recombinant collagen IX.

matrix protein (COMP) 1 is able to bind collagen IX and collagen II with high affinity in a Zn 2ϩ -dependent fashion (11,12). The collagen IX molecule appears to possess four binding sites for the C-terminal globular domain of COMP, one at or near each NC domain of the protein (11,12). According to a recent report, another oligomeric protein of the cartilage matrix, matrilin-3, also binds to collagen IX and collagen II in a Zn 2ϩ -dependent fashion in vitro (13) and has been reported to interact with COMP in a manner that is inhibited by cations Zn 2ϩ and Ca 2ϩ (13). Although direct evidence is lacking, it seems possible that these in vitro interactions reflect the potential of collagen IX, COMP, and matrilin-3 for forming a macromolecular complex in a controlled fashion in vivo. Interestingly, mutations in these three proteins are known to cause a similar phenotype, multiple epiphyseal dysplasia, MED (13). This form of osteochondrodysplasia is a diverse disease both genetically and phenotypically and is typically manifested in irregular epiphyses of the long bones and in most cases early onset osteoarthritis. All mutations thus far characterized in the three genes encoding collagen IX result in an in-frame deletion of at least 12 amino acids from the COL3 domain of the respective component polypeptide. Despite this intriguing finding and the existence of detailed information on mutations found in the genes encoding COMP and matrilin-3, the molecular mechanism of the pathogenesis of MED is not yet properly understood (13).
In the present work, we describe the production and characterization of the NC4 domain of human collagen IX as a recombinant protein and report on the interaction properties of this domain and its parental molecule with heparin and COMP.

Generation of the Bacterial Strain Expressing the Collagen ␣1(IX)
NC4 Domain-To generate a DNA construct for expression of the NC4 domain of the human ␣1(IX) polypeptide, oligonucleotide primers were designed to amplify the region corresponding to amino acids 24 -268 of the full-length polypeptide (NCB accession number NP_001842), thereby omitting the 23-amino acid signal peptide (14). The 5Ј-primer (CGA TCC ATG GTC ATC GAA GGT CGA GCT GTC AAG CGT CGC) contained an engineered NcoI cleavage site, and the 3Ј-primer (GAC TGA ATT CTT ATC TCT CGT CGG TGG TCT G) contained an engineered EcoRI cleavage site. Since the respective domain of the avian ␣1(IX) chain had previously been produced as a C-terminal GST fusion protein (15), a similar expression strategy was chosen here. Consequently, the 5Ј-primer contained an additional sequence to generate a cleavage site for Factor Xa protease between the fusion partner and the NC4 domain. A previously characterized viral expression construct for the full-length collagen ␣1(IX) chain (14) was used as a template in PCR. The amplification product was digested with the enzymes indicated above and ligated into the pGAT-2 bacterial expression vector in-frame with the sequences for the GST fusion tag. Following transformation into TOP10 cells (Invitrogen), a positive clone pGAT2-NC23 was obtained and found to contain the correct nucleotide sequences. An additional clone pGAT2-NC1 with a mutation converting the Cys 219 to an arginine residue was obtained. For expression of the fusion protein, the DNA constructs were transferred into Escherichia coli BL21(DE3) cells.
Expression and Purification of the NC4 Domain-The fusion protein was expressed in shaker flasks by inoculating an aliquot of the frozen cell line directly into the desired final volume of luria broth supplemented with 100 g/ml ampicillin. The cells were grown at 37°C until absorbance at 600 nm reached the value 1.0. The expression was started by adding 0.5 mM isopropyl-␤-D-thiogalactopyranoside, and the incubation was continued at 37°C for 4 -6 h. Following centrifugation, the cell pellets were stored frozen and later homogenized on ice by sonication in 0.3 M NaCl, 0.2% IGEPAL CA-630 (Sigma), and 0.05 M sodium phosphate buffer, pH 7, supplemented with 0.25 mg/ml lysozyme. Insoluble material was removed by centrifugation at 17,000 ϫ g for 40 min at 4°C, and the fusion protein was precipitated from the supernatant by adding ammonium sulfate to 30% saturation. The precipitate was collected by centrifugation at 23,000 ϫ g for 30 min at 4°C and dissolved in PBS supplemented with 1% IGEPAL-CA 630. The solution was clarified by centrifugation and incubated with a 1:20 volume of a 50% slurry of glutathione-Sepharose (Amersham Biosciences) in PBS at 4°C for 1 h. Following removal of the unbound material by centrifugation and three washing steps with PBS, the recombinant NC4 domain (rNC4) was cleaved off from the fusion partner by overnight digestion with Factor Xa protease (Amersham Biosciences) at room temperature. The insoluble fraction was then removed by centrifugation, and the rNC4 solution was diluted with 0.1 M NaCl and 0.05 M Tris buffer, pH 8, and subjected to further purification using a Hi Trap Heparin affinity column and a Hi Trap SP cation exchange column (Amersham Biosciences), eluting with an increasing NaCl concentration gradient. The size and purity of the product were determined by both SDS-PAGE analysis and electrospray ionization mass spectrometry. The identity of the purified protein was verified by amino acid analysis in an Applied Biosystems 421 analyzer and by analysis of the tryptic peptides by matrix-assisted laser desorption/ ionization time-of-flight (MALDI-TOF) mass spectrometry. The concentration of purified rNC4 was determined in a dye-binding assay (Bio-Rad) calibrated by quantitative amino acid analysis of rNC4.
Expression and Purification of Recombinant Collagen IX and COMP-Full-length collagen IX was prepared using a baculovirus expression vector system as described earlier (14). Full-length recombinant COMP and recombinant COMP T3ϩTC, containing residues 266 -755 of COMP (Swiss Protein Database accession number P35444), were prepared as recently described (11,16).
Circular Dichroism Analyses-CD spectra were measured using a JASCO J-810 instrument. Far-UV spectra were measured using a 0.1-cm path length cell for the 185-250-nm region, with a step size of 0.1 nm, scan speed of 20 nm/min, response of 0.25 s, bandwidth of 1 nm, and two accumulations per spectrum. The temperature was raised from 5 to 90°C in steps of 5°C and at a rate of 1°C/min, allowing stabilization for 2 min before scanning. Lyophilized rNC4 was dissolved in water at a concentration of 10 M. Measurements under reducing conditions were performed in water supplemented with 1 mM DTT. The temperature scan experiments were performed at the wavelength of 205 nm, raising the temperature from 10 to 90°C at a rate of 0.5°C/min. Near-UV CD spectra were measured for 105 M rNC4 in 1 mM HEPES buffer, pH 7.4, in the presence or absence of a 10-fold molar excess of 17-19-kDa heparin (Sigma), using a 0.5-cm path length cell for the 250 -320-nm region. Other parameters were as above.
Preparation of Biotinylated Small Unilamellar Vesicles-Biotinylated small unilamellar vesicles (SUV) for the surface plasmon resonance studies were prepared by a method adapted from those described earlier (17,18). A solution of egg yolk L-␣-phosphatidylcholine (Sigma) in chloroform/methanol (9:1) containing 2% biotinylated phosphatidylethanolamine (biotin DHPE; Molecular Probes Europe BV) was dried under a stream of nitrogen on a 37°C water bath and lyophilized in a vacuum for 2 h. Multilamellar vesicles were obtained by quickly hydrating the dried mixture with HBS buffer (150 mM NaCl and 20 mM HEPES buffer, pH 7.4). The solution was subjected to probe sonication until the sample became clear (5 ϫ 3 min on a cool water bath using a microtip probe), and the resulting SUV were purified by ultracentrifugation at 100,000 ϫ g for 30 min. The top third of the supernatant was collected and stored, protected from light, at 4°C for use in the surface plasmon resonance studies.
Surface Plasmon Resonance Assay-The surface plasmon resonance assays were performed on a Biacore 2000 TM instrument using either streptavidin-coated SA or hydrophobic HPA sensor chips (Biacore AB), and kinetic parameters were determined with the manufacturer's BI-Aevaluation 3.02 software using both simultaneous and separate k a and k d rate constant measurements according to the Langmuir binding model. The hydrophobic surface of the HPA sensor chips was coated with biotinylated SUV as suggested by the manufacturer. The biotinbinding protein NeutrAvidin™ (Pierce), dissolved in HBS, was injected over the biotinylated lipid surface to obtain an active surface for binding biotinylated ligands.
The biotinylated heparin was injected over the NeutrAvidin TM or streptavidin surface in a Biacore instrument, and the remaining binding sites were blocked by injecting dilute uncoupled biotin. One of the four flow paths in each sensor chip was left free of bound heparin to provide a negative control. Either rNC4 or recombinant human collagen IX (rcIX) produced in insect cells (14) was injected as the analyte at various concentrations at a flow rate of 10 or 25 l/min for 2 or 5 min in HBS buffer. To study the specificity of interaction with heparin, a constant amount of analyte was preincubated with different amounts of heparin, chondroitin 6-sulfate, or dermatan sulfate (Sigma) prior to injection. The surfaces were regenerated after each injection cycle with repeated injections of a solution containing 0.61 M NaCl and 15 mM HEPES, pH 7.4.
To study the interaction between NC4 and COMP, an aliquot of rNC4 was biotinylated at its carboxyl groups by incubation with a 10-fold molar excess of 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride solution and a 100-fold molar excess of biotin-PEO-amine (Pierce), followed by dialysis and lyophilization, and dissolved in HBS buffer containing 0.5 mM ZnCl 2 . Biotinylated rNC4 was attached to a biacore sensor chip in a similar fashion as above, but supplementing the HBS buffer with 0.5 mM ZnCl 2 (19), and the interaction between COMP and rNC4 was studied by injecting various amounts of rCOMP (11) at a flow rate of 10 l/min with a 10-min contact time. The surface was regenerated after each cycle with three injections of a solution containing 0.61 M NaCl, 15 mM HEPES, and 10 mM EDTA, pH 7.4. The kinetics of the interaction between rNC4 and COMP were determined by using COMP T3ϩTC (16), which was biotinylated by incubation with a 2-fold molar excess of biotinamidocaproate N-hydroxysuccinimide ester (Sigma) in 0.2 M NaHCO 3 , pH 8.2, followed by dialysis and lyophilization, and dissolved in HBS buffer containing 0.5 mM ZnCl 2 . Biotinylated COMP T3ϩTC was attached to the surface of a sensor chip as above, and rNC4 or rcIX was injected over the surface in the presence or absence of excess heparin at a flow rate of 40 l/min with a 2-min contact time. After each cycle, the surface was regenerated with two injections of 10 mM EDTA in HBS.
Solid-phase Heparin-binding Assays-For solid-state binding assays, rNC4 or rcIX at 1 g/ml in PBS (0.137 M NaCl, 10 mM phosphate buffer, pH 7.4, 2.7 mM KCl) were coated onto 96-well plates (Maxi-Sorp TM , Nunc, Denmark) overnight at 4°C. Further binding was blocked with 1% bovine serum albumin in PBS. Heparin-albumin-biotin (Sigma) at 5 g/ml or biotinylated heparin (see above) at 36 g/ml in PBS supplemented with 0.1% bovine serum albumin and 0.05% Tween 20 was allowed to interact with rNC4 or collagen IX surface, respectively, in the absence or presence of a nonbiotinylated competitor. The levels of bound biotinylated reagents were detected with horseradish peroxidase-conjugated NeutrAvidin TM using o-phenylenediamine dihydrochloride (Pierce) as substrate. The low background levels from wells coated with bovine serum albumin were subtracted from the experimental data.
Electron Microscopy-Bovine serum albumin was coupled to high molecular weight heparin by incubating with a 20-fold molar excess of heparin and a 200-fold excess of 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride solution for 2 h at room temperature in 0.1 M MES buffer, pH 5.0. The obtained heparin-bovine serum albumin was purified by gel filtration chromatography on Superdex 75 (Amersham Biosciences), dialyzed, conjugated with colloidal gold of 4 nm, and allowed to react with full-length collagen IX. Formed complexes were visualized by electron microscopy after negative staining with uranyl formate. Preparation of gold conjugates and electron microscopy were performed as described recently in detail (20).
Identification of the Heparin-binding Region of the NC4 Domain-The recombinant NC4 domain was reduced with DTT and alkylated with iodoacetamide. Free iodoacetamide was reduced with 0.1 M DTT and removed by dialysis into 50 mM ammonium acetate buffer, pH 4.0. A 450-g aliquot of the alkylated rNC4 was subjected to digestion with 12 units of V8 protease (Sigma) in the same buffer at room temperature for 4 h. The reaction mixture was diluted 4-fold with 0.1 M ammonium bicarbonate buffer, pH 7.8, and part of the sample was passed through a heparin-Sepharose affinity column (Amersham Biosciences). Unbound material, the material eluted with 0.5 M NaCl, and the remaining digestion mixture were desalted with ZipTip reversed phase tips C4 and C18 (Millipore Corp.) in tandem prior to analysis by MALDI-TOF mass spectrometry. On the basis of the obtained results a peptide spanning amino acids 1-19 of the mature NC4 domain was synthesized using Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistries in a 433A peptide synthesizer (Applied Biosystems), purified, inspected by MALDI-TOF mass spectrometry, and analyzed for its heparin-binding potential by heparin affinity chromatography. The peptide, designated as peptide 1-19, was also analyzed by CD under similar conditions as rNC4.
Using synthetic oligonucleotides and the QuikChange TM site-directed mutagenesis kit (Stratagene), a construct was prepared for expression of mutant rNC4 carrying a sequence NGL in place of the normal sequence KRR at positions 3-5 of the polypeptide. The mutant rNC4 was expressed as above, purified using glutathione-Sepharose and Factor Xa digestion, and subjected to heparin affinity chromatography.
Analysis of Disulfide Bonding Pattern-In order to isolate the NC4 domain from full-length collagen molecules, rcIX was digested with collagenase (Sigma), the reaction mixture was passed through a heparin-Sepharose affinity column (Amersham Biosciences), and the bound material was eluted with an increasing salt gradient and analyzed by SDS-PAGE. This material and rNC4 were then subjected to trypsin digestion with or without prior reduction with DTT and alkylation with 4-vinylpyridine. For in-liquid digestion, both the native and alkylated material were subjected to reversed phase chromatography on a C1 column, and the proteins were collected and dried. Trypsin (sequencing grade modified trypsin; Promega) in 50 l of 0.1 M (NH 4 ) 2 HCO 3 was added to the dried proteins to 5% (w/w), and digestion was carried out at 37°C for 4 h. The digestion mixtures were analyzed by MALDI-TOF mass spectrometry.
Fourier Transform Infrared Spectroscopy-Secondary structures of rNC4 and its mutated form were analyzed by Fourier transform infrared (FTIR) spectroscopy. Samples were exchanged to D 2 O and concentrated to about 20 mg/ml. Spectra were recorded at room temperature in IR cells having CaF 2 windows and 15-m Teflon spacers using a Bruker I55 spectrometer and a Hg/Cd/Te infrared detector. For each sample, 15 spectra of 64 scans were averaged, and the separately collected solution background absorption for each sample in the same cell assembly was subtracted from the final spectra. Gaussian curve fitting was used to determine the secondary structures.
Mass Spectrometry-Electrospray ionization mass spectra of the expressed rNC4 were obtained using a Micromass Q-TOF quadrupole/ time-of-flight hybrid mass spectrometer (Micromass) calibrated using myoglobin as a standard. Protein masses were calculated by deconvulation in MassLynx 3.4 (Micromass).
The NC4 peptides generated by either trypsin or V8 protease digestion were analyzed with a Biflex TM MALDI-TOF mass spectrometer (Bruker-Daltonics) in positive ion reflector mode (for peptides under 3500 Da) and positive ion linear mode (for peptides larger than 3500 Da) using ␣-cyano-4-hydroxycinnamic acid as the matrix. The MALDI spectra were externally calibrated with the standard peptides, angiotensin II and adrenocorticotropin 18 -39 or insulin.

Production of the Recombinant NC4 Domain of Human
Collagen IX-To study the structure and interactions of the NC4 domain of human collagen IX, sequences encoding amino acids 24 -268 of the human ␣1(IX) polypeptide were cloned into a prokaryotic expression vector for production as a C-terminal GST fusion protein lacking the native ␣1(IX) signal peptide. Following induction in E. coli BL21(DE3) cells, the recombinant protein was isolated from the harvested cells and subjected to initial affinity purification by glutathione-Sepharose. The GST tag was removed enzymatically, and a solution containing the suspected recombinant NC4 domain was applied to a heparin-Sepharose affinity column at physiological pH and ionic strength. A protein with a size expected for the recombinant NC4 domain bound effectively to the column and began to elute in the presence of about 0.25 M NaCl. Further purification was obtained by cation exchange chromatography. The purified protein was identified as the NC4 domain by amino acid analysis and analysis of the tryptic peptides. The expected molecular mass of 27,446 Da was obtained for the protein, designated as rNC4, by electrospray ionization mass spectrometry.
Surface Plasmon Resonance and Solid-state Analyses of the Interaction of rNC4 with Heparin-In the surface plasmon resonance analyses carried out to study the heparin-binding properties of rNC4, high molecular weight heparin was biotinylated and used to coat surface plasmon resonance sensor chips. Injection of soluble rNC4 at physiological pH and ionic strength over the heparin surface demonstrated the ability of the NC4 domain to interact with the immobilized ligand ( Fig.  1). A series of injections was used to determine the monomeric binding kinetics of the heparin interaction ( Fig. 2A), resulting in the determination of the association and dissociation rate constants k a and k d and the equilibrium dissociation constant K d (Table I). Using a hydrophobic sensor chip surface coated with biotinylated liposomes, a K d of 0.6 M was reliably obtained with low and regenerable background levels (Table I).
Relatively low 2 values indicate that the Langmuir 1:1 binding model provides a good approximation of the interaction strength ( Fig. 2A). Use of a carboxymethylated dextran-based sensor chip with a streptavidin coating gave an approximate K d of 3.2 M but with accumulating background levels.
Interaction of full-length rcIX with heparin was likewise analyzed using a hydrophobic sensor chip and biotinylated heparin (Fig. 2B). Using the Langmuir model, an averaged K d of 3.6 nM was obtained (Table I). The use of a bivalent binding model did not result in improved fitting or a change in the apparent K d (data not shown), and the estimation of the low affinity binding kinetics using the bivalent binding model was not possible.
The specificity of the interaction of rNC4 with heparin was demonstrated by incubating rNC4 with different amounts of various glycosaminoglycans (GAGs) prior to injection over heparin immobilized on a hydrophobic sensor chip (Fig. 3). When heparin was used as a competing GAG, a greater than 50% reduction in the relative binding level was obtained even at a GAG concentration of 0.5 M. By contrast, a 100-fold higher concentration of dermatan sulfate was required for a similar effect, and even a 40-fold excess of chondroitin-6-sulfate over rNC4 did not reduce the relative binding level to 50% of the maximum.
The interaction of rNC4 and rcIX with heparin was also studied by solid-state binding assay. For rcIX, biotinylated high molecular weight heparin prepared for surface plasmon resonance analyses was used. For increased sensitivity, a commercial preparation of biotinylated albumin coupled to high amount of heparin was used in the case of rNC4. As expected, results showed a concentration-dependent interaction of both rNC4 and rcIX with heparin (data not shown). In a competition assay, an excess of soluble heparin efficiently blocked the interaction of rNC4 with the tagged heparin, and a limited amount of soluble rNC4 efficiently competed with the binding of heparin to the surface (Fig. 4A). In analogy with the surface plasmon resonance analysis, dermatan sulfate demonstrated limited blocking activity, whereas chondroitin 6-sulfate (Fig.  4A) or chondroitin 4-sulfate (data not shown) could not compete with the heparin interaction. Similar results were obtained in the competition assay of the interaction of heparin with rcIX (Fig. 4B).
Electron Microscopy of Collagen IX Complexed with Heparin-In order to determine the number and approximate locations of the heparin-binding sites along collagen IX, electron -sulfate (f), or dermatan sulfate (OE) for a minimum of 60 min at room temperature prior to injection over an HPA sensor chip surface coated with heparin. The level of bound rNC4 was read 30s after the 5-min injection. At the beginning of the concentration series for each GAG, a sample of rNC4 without GAG was injected to obtain a response level representing 100% relative binding. microscopy on negatively stained complexes of full-length collagen IX and heparin-albumin conjugated with colloidal gold was performed (Fig. 5). Results suggest the presence of four binding sites for heparin on collagen IX, one located at or near each NC domain of the protein.
Identification of the Heparin-binding Region of the NC4 Domain-Alkylated rNC4 was digested with V8 protease and the resulting mixture of peptides subjected to heparin affinity chromatography. Analysis of the peptide mixture by MALDI-TOF mass spectrometry along with unbound and bound material from the affinity chromatography revealed that the extreme N-terminal peptide (amino acids 1-19, monoisotopic mass 2098.078) was able to bind to heparin, whereas other fragments were not. The N-terminal 19-amino acid fragment was subsequently prepared as a synthetic peptide (peptide 1-19), which was shown to bind to heparin-Sepharose and elute upon introduction of about a 0.15 M concentration of NaCl (i.e. a 40% smaller concentration than was required for the elution of native rNC4). CD analysis of peptide 1-19 indicated that it was largely devoid of secondary structure (data not shown).
A mutant form of rNC4, carrying a sequence NGL in place of the basic amino acid sequence KRR of the suspected heparinbinding site, was created. Using heparin affinity chromatography, it was demonstrated that this mutant rNC4 was unable to interact with heparin at physiological conditions. Comparison of the mutant rNC4 with wild-type rNC4 by far-UV CD analysis (data not shown) and by FTIR spectroscopy (Fig. 6) showed that the mutation did not affect the secondary structure of the NC4 domain. Both the mutant and the wild-type rNC4 appear to consist mainly of ␤-sheet, with smaller amounts of other types of secondary structures present (Fig. 6) (21). The possibility that freezing and lyophilization would affect the structure of rNC4 was also ruled out by FTIR measurements (Fig. 6).
Circular Dichroism Analyses-CD spectrum of rNC4 was measured at 25°C (Fig. 7A). An identical spectrum was obtained for the same sample under reducing conditions (Fig. 7B). In support of the FTIR analysis, the CD spectra at low temperatures present maximal negative ellipticity near 215 nm with no significant shoulder around 220 nm, indicating a secondary structure dominated by ␤-sheet. The thermal stability of rNC4 was assessed at a wavelength of 205 nm, increasing the temperature linearly from 10 to 90°C. A midpoint temperature of thermal denaturation of about 79°C was obtained for rNC4 ( Fig. 7A, inset). In the presence of a reducing agent, this value was lowered to about 72°C (Fig. 7B, inset). These results indicate that the notably high thermal tolerance of rNC4 is in part due to the presence of stabilizing intramolecular disulfide bonds. This conclusion was verified by analogous results indicating decreased thermal stability in a mutant form of rNC4, where Cys 219 is replaced by an Arg residue (data not shown). The presence of an isodichroic point at about 193 nm (Fig. 7) indicates that the thermal unfolding occurs in a two-state transition between native and unfolded conformation.
Far-UV CD analysis of the NC4 domain isolated from fulllength rcIX produced in insect cells gave a similar, but not identical, spectrum as that of rNC4 (data not shown). The minor difference is probably explained by the presence of Nglycans in the material produced in insect cells, a finding that was demonstrated by results obtained in peptide mapping ex- FIG. 4. Solid-state binding assay on the specificities of the interactions of rNC4 and rcIX with heparin. A, rNC4 was coated onto 96-well plates at 1 g/ml overnight and incubated with 5 g/ml heparin-albumin-biotin (ϳ50 nM) for 1 h at 4°C in the presence of no competing substance (1), 5 M rNC4 (2), 200 g/ml high molecular weight heparin (3), 200 g/ml chondroitin 6-sulfate (4), or 200 g/ml dermatan sulfate (5). Bound biotinylated heparin was detected using Neutravidin TM conjugated to peroxidase and a chromogenic substrate. Black bars represent the signals that are presented as percentages of the signal of the noninhibited reaction. White bars denote the S.D. values as percentages (n ϭ 8). B, wells were coated with 1 g/ml rcIX overnight and incubated with 36 g/ml heparin-biotin for 1 h at 4°C in the presence of no competing substance (1), 720 g/ml high molecular weight heparin (2), 720 g/ml chondroitin 6-sulfate (3), or 720 g/ml dermatan sulfate (4). The signals were detected, and the data are presented as in A (n ϭ 8).

FIG. 5. Reconstitution of complexes between collagen IX and heparin in vitro.
Collagen IX was incubated with heparin-albumingold for 30 min at 4°C, and formed complexes were visualized by negative staining with uranyl formate and electron microscopy. A, collagen IX molecules with heparin-albumin-gold conjugates (arrowheads) bound along the triple helix. The bar represents 100 nm. Closer inspection reveals heparin-albumin-gold particles at or near the noncollagenous domains NC1 (B), NC2 (C), NC3 (D), and NC4 (E). The bar represents 50 nm.
periments (data not shown) and by the heterogeneity of this material.
The structure of rNC4 was also studied by near-UV spectroscopy in the presence or absence of excess heparin. Only minimal changes in the spectrum of rNC4 were seen upon incubation with heparin, suggesting that association with heparin does not involve any major rearrangements of the tertiary structure of the NC4 domain.
Identification of the Disulfide Bonding Pattern of the NC4 Domain-In an attempt to identify the pattern of disulfide bond formation within the NC4 domain, an aliquot of rNC4 was alkylated with 4-vinylpyridine and digested with trypsin. Comparison of the pattern of the resulting peptides with that obtained by digestion of native rNC4 without alkylation indicated that all four cysteine residues of the NC4 domain appeared to be involved in disulfide bonding, with a pattern of Cys 21 -Cys 219 and Cys 175 -Cys 229 (Table II). An identical pattern was demonstrated for the NC4 domain isolated from full-length rcIX.
Surface Plasmon Resonance Analysis of the Interaction of rNC4 and rcIX with rCOMP or COMP T3ϩTC-Previous studies had demonstrated an approximate K d of 32 nM for the interaction of rcIX with rCOMP (11). Since these involved both a ligand and an analyte with multiple interaction sites, we wanted to obtain more reliable kinetic information by analyzing the interaction in a simplified system with the avidity effects diminished or eliminated. In initial experiments, rNC4 was biotinylated from its carboxyl groups and used to coat a hydrophobic sensor chip. Injection of soluble pentameric rCOMP (11) over the surface at various concentrations resulted in typical concentration-dependent association and dissociation curves (Fig. 8A), but there seemed to be concentration-dependent differences in the kinetics of the interaction that would require further analysis. Also, treatment with EDTA did not result in complete regeneration of the surface, and therefore the resulting K d values in the subnanomolar range appear not to be fully reliable, despite the low 2 values obtained with the Langmuir 1:1 model. The use of bivalent binding model did not improve the quality of the curve fitting further.
To analyze the interaction of COMP in a more reliable oneto-one situation, we used a biotinylated recombinant 57-kDa C-terminal fragment of the COMP monomer (COMP T3ϩTC) to coat a hydrophobic sensor chip. Soluble rNC4 was injected over the surface at various concentrations, and a K d of 0.23 Ϯ 0.11 M with low 2 values was obtained using the Langmuir 1:1 model (Fig. 8B). Injection of rcIX over the same surface at a 20 nM concentration gave a K d of 0.9 nM for the interaction of full-length collagen with the C-terminal fragment of COMP (data not shown).
The same sensor chip was also used to analyze the effect of excess heparin on the interaction of COMP T3ϩTC with rNC4 and rcIX. Heparin was found to inhibit both interactions at low FIG. 6. IR spectra of rNC4 in the amide I region. The spectra of lyophilized wild-type rNC4 (a), nonlyophilized wild-type rNC4 (b), and lyophilized mutated rNC4 dissolved in D 2 O (c) are shown in the amide I region between 1700 and 1600 cm Ϫ1 . All three spectra are nearly identical, indicating a similar secondary structure consisting mainly of ␤-sheet, as judged by the position (ϳ1635 cm Ϫ1 ) of the main peak of the spectra.

FIG. 7. CD spectra of rNC4 as a function of temperature.
Far-UV spectra of 10 M rNC4 under nonreducing (A) and reducing conditions (B) were measured for the 185-250-nm region at temperatures ranging from 5 to 90°C in steps of 5°C. Results obtained at selected temperatures are shown as indicated. Thermal stability of rNC4 was also studied by monitoring the CD signal at 205 nm upon heating the sample at a rate of 0.5°C/min under nonreducing (inset of A) and reducing (inset of B) conditions. Results show that upon reduction, the midpoint temperature of thermal transition of rNC4 was lowered by about 7°C.  219 and Cys 175 -Cys 229 linkages The NC4 domain was either alkylated with 4-vinylpyridine and digested with trypsin or digested in a native state. The digests were analyzed by MALDI-TOF mass spectrometry, and the measured monoisotopic masses are shown along with calculated monoisotopic masses for the NC4 tryptic peptides. Peptides with molecular mass below 800 Da are not included. The peptides containing cysteines are indicated with an asterisk. micromolar concentrations. Preincubation of 20 nM rcIX with 4 M heparin (molecular mass 6 kDa) prior to injection over the coated surface resulted in more than 99% inhibition of the interaction with COMP T3ϩTC, and 50% inhibition was achieved with a heparin concentration lower than 0.1 M (Fig.  9). Analogous results were obtained when rcIX was replaced with 7.1 M rNC4, and a similar effect was seen when a heparin preparation with a molecular mass of 17-19 kDa was used (data not shown). In general, the hydrophobic sensor chip was found to yield more consistent results than any other sensor chip in all surface plasmon resonance analyses performed.

DISCUSSION
In the present study, we have developed a method for producing and purifying high levels of the NC4 domain of human collagen IX as a recombinant protein. Analyses of the protein show that this domain has a remarkably high thermal stability, complying with the known high stability of the COL3 domain (22,23), which allows the NC4 domain to project away from the body of the collagen fibril into the extrafibrillar space. The emergence and preservation of such stable structures in the course of evolution speaks for a specific role for these domains within the cartilage matrix.
We have also shown here that the high stability of the NC4 domain is in part due to the formation of two intramolecular disulfide bridges. A similar pattern of disulfide bonding was shown to occur in the NC4 domain isolated from full-length recombinant collagen IX. This finding, together with the observed high similarity of the far-UV CD spectra of rNC4 and the NC4 domain of rcIX, suggests that the NC4 domain folds in a certain fashion that is not dependent on the presence of the collagen IX triple helix and that its cysteines do not participate in intermolecular bonding. Despite the contribution of the disulfide bonds to thermal stability, rNC4 has a stable folded structure even in the presence of a reducing agent. Thus, folding of rNC4 is presumably taking place in the cytosol of the bacteria, whereas the disulfide bridge formation must occur only after cell lysis, since the cytosol of bacterial cells is a reducing environment (24).
Our results show that the NC4 domain can now be added to the growing list of extracellular matrix proteins that interact with heparin or heparan sulfate. Pepsin-treated collagen IX, together with several other collagens residing in cartilage, is known to interact with heparin (9). Collagen XI, for example, possesses several heparin-binding sequences, two of which comply with the heparin-binding consensus sequence XB-BXBX, where B denotes a basic amino acid and X is any other amino acid (25). A sequence KRRPRF matching this consensus is present in the extreme N-terminal region of the NC4 domain of collagen IX and is indeed shown here to be the site of interaction of the domain with heparin. This sequence motif is also present in the N terminus of the mouse collagen ␣1(IX) chain (26). Comparison of this sequence of NC4 with a reported distribution of residues in known heparin-binding proteins (27) further implicates the sequence as a potential heparin-binding site. Since a synthetic peptide representing the 19 extreme N-terminal amino acids of the NC4 domain elutes from a heparin affinity column with a buffer of lower ionic strength than the full-length NC4 domain, it seems that a proper structural context is required in addition to the consensus binding site in order to obtain the full heparin affinity of the NC4 domain. In support, CD analysis of the peptide indicated a significant lack of secondary structure.
The electron microscopic analysis of collagen IX complexed with heparin verifies the previously suggested presence of at least one heparin-binding site on collagen IX outside its NC4 domain (9). Our results suggest the presence of such a site at or near each NC domain of the protein. These regions of the collagen IX polypeptides do not contain any basic sequence stretches matching the linear consensus sequence (25). However, basic residues with apparent spacings of about 20 Å, an arrangement found at the heparin-binding sites of some collagens and many other proteins (25,28), are present at the NC2 domain and at the COL2 domain not far from the NC3. In addition, most of these residues are highly conserved both between different ␣ chains and between mammalian and avian species. As in other proteins (28), a few basic residues are found amid the ones spaced at 20-Å intervals but show somewhat less conservation. In the NC1 domain, the linear spacing of the basic residues at 7-or 8-residue intervals is less apparent, although a cluster of at least 5 basic residues is found in the C-terminal half of the human, mouse, and chicken NC1 domains.
We have shown that the NC4 domain of collagen IX is clearly capable of interaction with heparin in vitro at physiological pH and ionic strength, albeit with a relatively low affinity. This affinity may be significant in vivo, however, since other components are likely to be involved. Collagen IX is known to interact with COMP via its NC domains (11,12), and structural alterations in these two proteins can lead to a similar phenotype, multiple epiphyseal dysplasia, MED (13). In addition, collagen IX reportedly interacts with matrilin-3, which is yet another protein implicated in the pathogenesis of MED (13).
Here we demonstrated an interaction between rNC4 and the C-terminal fragment of COMP with a K d of 230 nM, whereas a K 0.5 value of 71 nM was recently obtained by a solid-state assay using rcIX instead of rNC4 (16). The difference in the magnitude for rcIX and rNC4 is most likely explained by the presence of three additional COMP-binding sites in rcIX, although other differences in the experimental setup may also have contributed. On the basis of our results, it is impossible to estimate whether the affinity of the NC4 for COMP is in any way affected by the absence or presence of the adjacent collagen IX triple helix. In an earlier study, an approximate K d of 32 nM was obtained for the interaction of rcIX with pentameric rCOMP (11). In light of the recent results (16), our observation of a clearly lower approximate K d of 0.9 nM toward rcIX when using COMP T3ϩTC, instead of rCOMP, appears as an underestimation, but it may also reflect the different technical approach used here.
Somewhat surprisingly, our results show that COMP and heparin compete for the same binding site or overlapping ones in the NC4 domain. Similarly, we were able to block the interaction of full-length rcIX with the C-terminal domain of COMP using relatively low heparin levels. It is unclear whether these findings translate into a mechanism for controlling the interaction of collagen IX with COMP in vivo by means of heparan sulfate proteoglycans or whether they are an indication of two distinct functional roles for the NC4 domain in different regions of the cartilage extracellular matrix.
Using full-length recombinant collagen IX, we obtained an apparent K d of 3.6 nM for its interaction with heparin. Pepsintreated collagen IX has previously been shown to bind heparin with a K d of 7 nM (9). Both of these figures are about 2 orders of magnitude smaller than the K d obtained here for the interaction of NC4 with heparin in vitro. As suggested previously (9) and verified here, additional heparin-binding sites are present in the long helical arm of collagen IX outside the NC4 domain, explaining the difference in K d for NC4 and full-length collagen IX. Unfortunately, no experimental data are available regarding the affinities of the different collagen IX heparin-binding sites for heparan sulfate proteoglycans in situ. Collagen IX lies parallel to the surface of the collagen fibril in the cartilage extracellular matrix and is covalently attached to it via the long arm (3), and therefore the accessibility of the other binding sites may be sterically reduced, leading to preferred binding of heparan sulfate proteoglycans to the NC4 domain of collagen IX.
The functional significance of the heparin-binding potential of the NC4 domain remains unknown. Interestingly, in addition to mutations affecting the COL3 domain of collagen IX, a variant of MED can also be caused by a homozygous mutation in the gene encoding the diastrophic dysplasia sulfate transporter, possibly as a result of incomplete sulfation of proteoglycans (29). Cartilage extracellular matrix reportedly contains some heparan sulfate proteoglycans (e.g. perlecan (30)), which are potential interaction partners for collagen IX. Such small proteoglycans may fulfill a bridging function between adjacent collagen IX-containing fibrils. This would explain the importance of collagen IX in the maintenance of the long term integrity of the fibrillar network (4, 5). It has been suggested that the homologous collagens XII and XIV provide a similar bridging mechanism via their N-terminal domains, and there is some evidence that these domains modulate the biomechanical properties of tissues by controlling the organization of the fibrillar network (31,32). Another function for the collagen IX heparin-binding potential can also be hypothesized, however, since chondrocytes are known to express cell surface heparan sulfate proteoglycans (e.g. syndecan-1, -2, and -4 (33, 34)), which are believed to contribute to integrin-mediated cell attachment and signaling (35). Despite the low affinity of NC4 for heparin in vitro, collagen IX may interact with such a cell surface receptor in vivo if other components participate in the macromolecular complex. Of the other homologous proteins, collagen XIV has been shown to bind several cell types in a heparin-inhibitable fashion via the N-terminal noncollagenous domain of the molecule (36,37), and heparin is also involved in the attachment of chondrocytes to collagen XI in vitro (25). Chondrocyte survival and differentiation, on the other hand, are known to be mediated by integrins in situ (38), but integrins are thought to require a co-receptor for full activity (35). It has been demonstrated recently that chondrocytes require the presence of collagen fibrils with suitable suprastructures for maintenance of the cartilage phenotype (39). Taking the above information together, it can be hypothesized that a macromolecular complex including collagen IX, a cell surface heparan sulfate proteoglycan, and perhaps other macromolecules may act in concert with the integrin system and serve as a mechanism that provides the chondrocyte with the ability to accommodate to or resist changes in the surrounding extracellular matrix. A structural alteration in any component of this complex would compromise the integrity of the cartilage. This hypothesized role of collagen IX in cell attachment and/or signaling is in harmony with the known preferential location of collagen IX in the thinnest fibrils (40,41) that prevail in the pericellular basket surrounding the chondrocytes (42,43). Direct experimental evidence of the participation of collagen IX in the macromolecular complexes suggested above is, however, of utmost importance to validate any of the above hypotheses.