Alternative Splicing in the Aggrecan G3 Domain Influences Binding Interactions with Tenascin-C and Other Extracellular Matrix Proteins*

The proteoglycans aggrecan, versican, neurocan, and brevican bind hyaluronan through their N-terminal G1 domains, and other extracellular matrix proteins through the C-type lectin repeat in their C-terminal G3 domains. Here we identify tenascin-C as a ligand for the lectins of all these proteoglycans and map the binding site on the tenascin molecule to fibronectin type III repeats, which corresponds to the proteoglycan lectin-binding site on tenascin-R. In the G3 domain, the C-type lectin is flanked by epidermal growth factor (EGF) repeats and a complement regulatory protein-like motif. In aggrecan, these are subject to alternative splicing. To investigate if these flanking modules affect the C-type lectin ligand interactions, we produced recombinant proteins corresponding to aggrecan G3 splice variants. The G3 variant proteins containing the C-type lectin showed different affinities for various ligands, including tenascin-C, tenascin-R, fibulin-1, and fibulin-2. The presence of an EGF motif enhanced the affinity of interaction, and in particular the splice variant containing both EGF motifs had significantly higher affinity for ligands, such as tenascin-R and fibulin-2. The mRNA for this splice variant was shown by reverse transcriptase-PCR to be expressed in human chondrocytes. Our findings suggest that alternative splicing in the aggrecan G3 domain may be a mechanism for modulating interactions and extracellular matrix assembly.

The aggregating proteoglycans aggrecan, versican, neurocan, and brevican form the lectican (1) or hyalectan (2) family and are major components of the extracellular matrix (ECM) 1 with important functions in many tissues. The core proteins of these proteoglycans have extended central glycosaminoglycan attachment regions of varying length that are flanked by globular domains (3)(4)(5)(6). In the cartilage proteoglycan aggrecan, the large extent of glycosaminoglycan side chain substitution and the resulting fixed charge density attracts counter-ions and water through osmotic processes. The resulting swelling pressure is crucial for the biomechanical properties of this tissue (7). The conserved N-terminal globular G1 domains anchor these proteoglycans to hyaluronan in an interaction stabilized by the link protein (8 -12). Aggrecan contains an additional globular G2 domain of unknown function between the G1 domain and the glycosaminoglycan attachment region (13). The C-terminal G3 domain is highly conserved and found in all four of these proteoglycans.
We have shown previously that the G3 domain mediates binding to other ECM molecules, e.g. tenascin-R (14,15), fibulin-1 (16), fibulin-2 (17), and fibrillin-1 (18). The G3 domain also binds sulfated glycolipids on the cell surface (19). In addition, neurocan has been reported to bind to tenascin-C (20). The ECM protein ligands for the G3 domains are all dimeric or multimeric proteins, and we have shown that they can crosslink proteoglycans from different hyaluronan/proteoglycan aggregates (17). This may well be of functional importance for the organization and assembly of the ECM during development and reparative synthesis.
The G3 domains contain an EGF module (EGF1) (21), a calcium-binding EGF module (EGF2) (22), a C-type lectin module (CLD), and a complement regulatory protein-like module (SCR) (23). The C-type lectin module is constitutively expressed and is the mediator of all the G3 interactions listed above. The other modules of the G3 domains are subject to alternative splicing (22)(23)(24)(25)(26). The expression of the SCR module is variable, but in humans this module is usually present, regardless of age (26). The EGF1 and EGF2 modules are expressed to a lower extent (25-28 and 5-8%, respectively, in humans) (22). The EGF2 module is highly conserved and uniformly expressed at low levels in several different species (22), whereas the less conserved EGF1 module is expressed to different degrees in various species (25).
These differences and the alternative splicing of the EGFlike repeats may reflect different functions for the EGF1 and EGF2 modules in different species. Expression of the EGF modules could constitute a mechanism for feedback regulation of differentiation and proliferation of the chondrocyte, for example, similar to what has been proposed for the versican G3 EGF modules (27)(28)(29)(30). In addition, alternative splicing of modules in the G3 domain could affect aggrecan glycosaminoglycan substitution and transport through the secretory pathway. The alternative splicing of the flanking EGF and SCR modules could also have a regulatory function by modulating the C-type lectin-mediated interactions.
In the present study we confirm that tenascin-C is a high affinity ligand for aggrecan, versican, and brevican as well as for neurocan. We also map the binding site on tenascin-C to the region corresponding to the binding site on tenascin-R (15). By using recombinant G3 domain splice variants, we then show that the alternative splicing of flanking modules indeed affects the affinities of the constitutively expressed C-type lectin module for different ECM ligands. The most prominent effects were found in the splice variant containing the full G3 set, i.e. EGF1, EGF2, CLD, and SCR modules, and we confirmed the expression of this splice variant in human chondrocytes. The observed variation in affinity suggests a mechanism for fine-tuning of the G3 interactions by alternative splicing.

EXPERIMENTAL PROCEDURES
Assembly of G3 Variant Constructs-We have previously produced recombinant aggrecan fragments containing part of the chondroitin sulfate chain attachment region and different variants of the aggrecan G3 domain (31). The aggrecan G3 variant sequences were released from these pcS plasmids by digestion with PstI and ligated into the episomal expression vector pCEP4-BM40-hisEK (32). The resulting expression plasmids code for the BM40 signal peptide, a hexahistidine tag, an enterokinase cleavage site, and at the C-terminal, a variant of the G3 domain of aggrecan, as shown in Fig. 1B. The CS region present in the pcS expression constructs (31) is not included in the new plasmids.
Expression and Purification of G3 Variant Proteins-The G3 variant plasmids were transfected into human embryonic kidney 293-EBNA cells (Invitrogen) using FuGENE 6 (Roche Applied Science) lipid transfection reagent. After 48 h, transfected cells were selected with 260 g/ml hygromycin and allowed to grow to confluency. For large scale continuous production of G3 variant proteins, the cells were cultured in Dulbecco's modified Eagle's medium (Sigma) for 48-h periods alternately in the presence and then the absence of 10% fetal calf serum. Serum-free medium was collected, cleared of cell debris by centrifugation for 2 min at 4000 rpm, treated with protease inhibitors (2 mM phenylmethylsulfonyl fluoride, 10 mM N-ethylmaleimide, 0.5 g/ml leupeptin, 1 g/ml antipain, 5 g/ml benzamidine-HCl, 0.5 g/ml aprotinin, 0.5 g/ml chymostatin, and 0.5 g/ml pepstatin), and stored at Ϫ20°C prior to purification.
Purification of G3 Variant Proteins-The samples were adjusted to an ionic strength of 0.5 M NaCl and 20 mM sodium phosphate (pH 8.0), applied to a nickel-loaded 5-ml HiTrap chelating HP column (Amersham Biosciences) equilibrated in 0.5 M NaCl, 20 mM sodium phosphate (pH 8.0), and eluted using 0.5 M imidazole in the equilibration buffer. Fractions containing recombinant protein were identified by dot blotting with anti-His 6 monoclonal antibodies (R&D Systems), pooled, and further purified by ion exchange chromatography on a MonoQ HR 5/5 column (Amersham Biosciences) with a 0 -1 M NaCl gradient over 20 column bed volumes in 20 mM Hepes (pH 7.5), 1 mM EDTA. The identity of the purified recombinant proteins was confirmed by mass spectrometry. Briefly, the proteins (100 pmol) were adjusted to pH 8.8, reduced (10 mM dithiothreitol, 56°C, 30 min), alkylated (10 mM iodoacetamide, 25°C, 90 min), and digested with endoprotease Lys-C (1 g, 37°C, 15 h). The resulting peptides were purified on ZipTips (Millipore) and analyzed by MALDI-TOF mass spectrometry as described previously (33).
Expression and Purification of Other Extracellular Matrix Proteins-The bacterial expression and purification of recombinant GST fusion protein fragments of tenascin-R (15,34) and tenascin-C (35) have been described. A secreted alkaline phosphatase fusion protein containing the aggrecan C-type lectin was produced in 293-EBNA cells as described (17), as was the case for fibulin-1 (36) and fibulin-2 (37). In the present study we used a recombinant fragment of mouse fibulin-1 that corresponds to the human fibulin-1A splice variant. Although it should be mentioned that this particular splice variant has only been detected at low levels in placenta (38), it contains the aggrecan G3 domain binding site and does indeed bind aggrecan G3 (16). Full-length tenascin-C purified from U251MG human glioma cells was purchased from Life Technologies. Full-length rat tenascin-R cDNA was assembled from partial cDNAs (kind gifts from Drs. Dirk Montag and Melitta Schachner, ETH Zü rich, Switzerland). First, a 4.1-kb SacI to XmnI fragment of clone J1-160/180-1 (39) was ligated into plasmid pEGF-L (34), cut with XbaI, filled in, and cut with SacI. Thereafter, the 5Ј-end of the cDNA was isolated as a 575-bp PvuII fragment from the mouse genomic clone J34AS43 2 and inserted in the PvuII site of the previous construct. This fragment codes for the first 85 amino acid residues of tenascin-R, resulting in two amino acid residues of the rat sequence being replaced by the mouse equivalents (E4D and V40A) (40). Finally, the HindIII fragment containing the full-length cDNA was isolated from the resulting plasmid and inserted into the mammalian episomal expression vector pCEP4 (Invitrogen). Sequencing of this plasmid confirmed that the cDNA fragments had been correctly assembled. After transfection of the pCEP4-tenR plasmid into 293-EBNA cells and hygromycin selection and expansion as described above, conditioned serum-free Dulbecco's modified Eagle's medium was collected and stored frozen until purification. Four hundred ml of conditioned medium was concentrated to 40 ml in a 160-ml Amicon ultrafiltration device with a PM-10 membrane. The concentrate was then dialyzed (molecular weight cut-off 10,000) against 8 liters of 10 mM Tris-HCl, 100 mM NaCl (pH 7.2). After filtration through a 0.45-m filter, the medium was pumped (1 ml/min) onto a 5-ml HiTrap Q-Sepharose column (Amersham Biosciences) equilibrated in the dialysis buffer. After washing with 30 ml of equilibration buffer, the bound protein was eluted with a 100-ml linear gradient from 0.1 to 1 M NaCl in 10 mM Tris-HCl (pH 7.2), 2 D. Montag, personal communication. A shows the structure of aggrecan (brackets indicate possible alternative splice variants). B shows the expressed recombinant G3 variants used in this paper. A legend to the symbols used for different structural modules of the globular domains is found in A. and 1-ml fractions were collected. Tenascin-R-containing fractions were identified through SDS-PAGE, Western blotting, and ligand overlay blotting with alkaline phosphatase-tagged aggrecan lectin domain (see above). Positive fractions were pooled and concentrated to 1 ml through ultrafiltration as above. The concentrated pool was then subjected to size exclusion chromatography on a Superdex-200 column (Amersham Biosciences) equilibrated in 20 mM sodium phosphate, 150 mM NaCl (pH 7.4), and 0.4-ml fractions were collected. After SDS-PAGE analysis as above and confirmatory MALDI-TOF mass spectrometry of trypsindigested bands from the gel (14% sequence coverage, resulting in more than 99% probability for tenascin-R), tenascin-R containing fractions were pooled, frozen on dry ice, and stored at Ϫ80°C.
Surface Plasmon Resonance Measurements-The extracellular matrix ligands of aggrecan G3, i.e. the tenascins and fibulins, both whole molecules and fragments, were diluted with 10 mM sodium acetate (pH 4.0) and immobilized in different flow cells of CM5 sensor chips (BIAcore). Immobilization levels were between 200 and 1000 resonance units. As a control for changes in a sample refraction index, a blank flow cell was subjected to the immobilization procedure without added protein. For affinity measurements, binding and dissociation were monitored in a BIAcore 2000 instrument (BIAcore). In control experiments for possible mass transfer limitations, the aggrecan G3 variant proteins (25 nM) were injected over the ligand surfaces at different flow rates in running buffer (10 mM Hepes-HCl, 150 mM NaCl, 2 mM CaCl 2 , 0.005% P-20 detergent (pH 7.5), 25°C). No difference in initial binding rate was observed at flow rates of 35 l/min or above, indicating no mass transfer limitation, and a flow rate of 50 l/min was used in all subsequent experiments. For affinity measurements, the purified G3 variants, LCt, Lt, Ct, E1LCt, E2LCt, and E1E2LCt, were injected in running buffer (at 50 l/min) into the flow cells at concentrations ranging between 1.5 and 50 nM with duplicate measurements at each concentration. The ligand surfaces were regenerated by injection of a 300-l pulse of 20 mM EDTA and 1.15 M NaCl in running buffer between each experiment. No binding was observed in control experiments using buffer containing 5 mM EDTA instead of 2 mM CaCl 2 . To confirm the results obtained, control experiments were performed in the reverse orientation. The different proteins were further purified and confirmed to be homogeneous by gel filtration on a Superdex 75 PC3.2/30 column in a SMART chromatography system (Amersham Biosciences) using BIAcore running buffer as eluent. Approximately equimolar amounts of the aggrecan G3 variants LCt, Ct, and E1E2LCt were immobilized on a CM5 sensor chip (382, 201, and 621 resonance units immobilized, respectively). After mass transfer and regeneration tests as above, the recombinant fragments comprising FnIII repeats 3-5 of tenascin-R and tenascin-C were injected (in duplicate) at 50 l/min over the surfaces at concentrations ranging from 1.5 to 50 and 15 to 500 nM, respectively. After x and y normalization and subtraction of blank curves, the association (k a ) and dissociation (k d ) rate constants were determined simultaneously using Marquardt-Levenberg global curve fits to the equation for 1:1 Langmuir binding in the BIAevaluation 3.1 software (BIAcore). The equilibrium dissociation constant (K D ) of each binding reaction was calculated from these values.
SDS-PAGE and Ligand Overlay Assays-Protein samples were separated by electrophoresis (41) under reducing conditions in NuPAGE 4 -12% BisTris gels (NOVEX and Invitrogen), and either stained with Coomassie Brilliant Blue R (Sigma) or transferred to nitrocellulose membranes (42) with Tris/Bicine transfer buffer (NOVEX, Invitrogen) for ligand overlay assays. Transfer was confirmed by Ponceau S (Sigma) staining. The nitrocellulose filters were blocked with 3% BSA in incubation buffer (25 mM Tris-HCl, 0.15 M NaCl, 0.1% Tween 20 (pH 7.4)) at 4°C overnight. After washing twice with 0.3% BSA in incubation buffer, the filters were incubated with recombinant G3 variant protein (0.2 g/ml in 0.3% BSA in incubation buffer) in the presence of either 5 mM CaCl 2 or EDTA. The filters were washed twice, and bound G3 variant protein was detected by incubation with anti-His 6 monoclonal antibody, again in the presence of either 5 mM CaCl 2 or EDTA, horseradish peroxidase-conjugated anti-mouse secondary antibody in calcium-or EDTA-containing incubation buffer, followed by chemiluminescence substrate (PerkinElmer Life Sciences).
Analysis of mRNA Expression-Total RNA was prepared from primary human articular chondrocytes using Tri-Reagent (Sigma). Using the GeneAmp RNA PCR kit (PerkinElmer Life Sciences) with random primers, the sample was reverse-transcribed in one cycle of 42°C for 15 min, 95°C for 105 s, and 5°C for 5 min. The product of reverse transcription was amplified from a hot start at 95°C for 105 s with two sets of primer pairs, Agg4 and Agg11, and Agg4 and Agg13 (31), in PCRs using 35 cycles of denaturation at 95°C for 45 s followed by annealing and extension at 56°C for 45 s and a final hold of 72°C for 7 min. An aliquot of each product was separated by electrophoresis on a 1.5% (w/v) agarose gel. The larger 355-bp product of the Agg4 and Agg13 reaction was cloned into the pCR vector of the TA Cloning kit (Invitrogen) and sequenced using the BigDye Terminator v1 Cycle Sequencing Kit (ABI Prism).

The Proteoglycan C-type Lectin Domains Bind Tenascin-C-
BIAcore interaction experiments revealed that recombinant proteoglycan lectin domains bound to tenascin-C in addition to tenascin-R (Fig. 2). As shown in Fig. 3A, tenascin-C and -R have very similar domain organizations, with N-terminal multimerization motifs followed by epidermal growth factor (EGF) repeats, fibronectin type III (FnIII), repeats and a fibrinogenlike (Fbg) module. The main difference lies in the number of EGF repeats and inserted alternatively spliced FnIII-repeats. By using a panel of recombinant variants of tenascin-C cover- ing the FnIII-Fbg region (Fig. 3B), we set out to map the interaction site. It is clear from the Coomassie-stained SDS-PAGE (Fig. 4A) that the TNfn-(6 -8) protein preparation contains a band of lower mass in addition to the full-length protein. The presence of a putative truncated form of the protein does not, however, affect the overlay assays, as the full-length protein is present and available for interaction. Overlay blot assays with different variants of the aggrecan G3 domain (see below) identified fragments TNfn-(1-8) and TNfn-(3-5) as binders (Fig. 4). As seen from Fig. 4E, this interaction was eliminated by the addition of EDTA. Indeed, injecting the tenascin fragments over recombinant aggrecan, brevican, neurocan, and versican lectin domains immobilized on BIAcore sensor chips revealed that all the lectican proteoglycans are capable binders to the FnIII-(3-5) region of both tenascin-C and tenascin-R (Table I). From these data it is also clear that brevican is a much weaker tenascin-C binder than the other lecticans, even though it remains an avid binder for the corresponding FnIII-(3-5) fragment of tenascin-R. In summary, tenascin-C and tenascin-R both contain conserved proteoglycan G3-binding sites, as outlined in Fig. 3C.
The Influence of the Alternative Splicing of the G3 Domain on Its Interactions with Known Ligands-Alternatively spliced variants of the human aggrecan G3 domain containing natural splice junctions were expressed in mammalian cells. The constructs based on a cDNA for the aggrecan C-terminal domain (31) were prepared in a His 6 -tagged vector derived from pCEP4 (32). They contained different combinations of the two alternatively spliced EGF-like motifs, the mammalian C-type lectinlike region, the SCR-like region, and the tail region of the G3 domain of aggrecan (Fig. 1). Stable episomal expression of the constructs was established in 293-EBNA cells, and the resulting G3 variant proteins were purified from the conditioned medium using a combination of metal chelation affinity and ion exchange chromatographic techniques.
All the recombinant proteins migrated as single bands with sizes corresponding to their predicted molecular masses (Fig.  5). After storage at 4°C, the purified Ct protein appeared as a double band on SDS-PAGE of somewhat smaller size than the 12,456.9 Da predicted from the sequence (not shown). Because the presence of the expected N-terminal peptide was confirmed by MALDI-TOF mass spectrometry of endoprotease Lys-C-digested Ct (not shown), we assumed that the lower masses of the bands were caused by C-terminal truncation. Indeed, the "tail" of aggrecan contains consensus sequences for furin-like proteases, and cleavage at these sites has been reported (43). The Ct bands were cut out from the gel and the proteins eluted, and MALDI-TOF mass spectrometry gave masses of 10,553.0 and 9998.7 for the two bands, respectively. Comparison of these masses with the recombinant protein sequence using the Find-Pept software (www.expasy.org) showed that the two bands corresponded to truncated versions of the Ct protein (amino acid residues 1-89 and 1-85 of 105, respectively). In summary, although both forms of the Ct protein lack the C-terminal end of the tail region, they both have an intact SCR repeat and a histidine tag. The change in migration distance observed upon reduction also confirmed disulfide bonding in the Ct protein (Fig. 5), and the protein eluted as a symmetrical peak in gel filtration (not shown). The Ct sample was thus considered suitable for use in the ligand interaction studies.
The interaction between the different G3 variants and known aggrecan ligands was measured by surface plasmon resonance technology. In a first set of experiments, the aggrecan G3 variants were injected over the immobilized matrix proteins on the BIAcore sensor chip, and bound material was measured over time. All of the lectin domain-containing G3 variants studied showed strong calcium-dependent binding to tenascin-C and -R, both as the whole molecule and the recombinant domains of tenascin-R FnIII- (3)(4)(5), and to fibulin-2 (Table II). However, the variant Ct, which does not contain the lectin-like domain, did not bind to any of these ligands. The G3 variant E1E2LCt, which contains both of the EGF-like motifs that can be alternatively spliced in vivo, bound more strongly than the others to both tenascin-R FnIII-(3-5) and fibulin-2 (the K D values for tenascin-R FnIII-(3-5) and fibulin-2 binding were ϳ5and 10-fold lower than those of the others). However, the K D value for the binding of E1E2LCt with either full-length tenascin-R or -C was similar to that of the other lectin-containing G3 variants, although the association and dissociation constants were very different than those of the other variants (the FIG. 3. Domain structure of tenascin-R and -C. A shows the domain structures of tenascin-R and -C, respectively. B depicts the different bacterially expressed recombinant tenascin-C fragments used in mapping the binding site for proteoglycan lectins. C summarizes the tenascin-C data from overlay (Fig. 4) and BIAcore experiments (Table I) and previous work on tenascin-R (15). The tenascin N termini and multimerization domains are shown as triangles and spiral-filled circles, EGFL repeats as diamonds, FnIII repeats as ovals, and fibrinogen globules as hexagons. Alternatively spliced FnIII repeats are shaded.
association constant for the binding to both of the tenascins was 10-fold lower, and the dissociation constant for the binding to tenascin-R was 5-10-fold lower). In a second set of experiments, we repeated the analysis in reverse orientation, i.e. with the aggrecan G3 splice variants immobilized. This confirmed the high affinity of the E1E2LCt for different ligand proteins (Table II).
Although fibulin-1 is a known ligand of the C-type lectin domain of aggrecan (16), E1E2LCt was the only one of the G3 variants that bound to immobilized fibulin-1, and this was with

TABLE I Proteoglycan lectin domain interactions with tenascin-C fragments
Recombinant tenascin-C fragments and a tenascin-R control fragment were injected over proteoglycan C-type lectin surfaces (i.e. corresponding to Lt but without the tail), and the binding parameters were determined in a BIAcore 2000 instrument as described under "Experimental Procedures." The 2 values of the fitted curves were 0. 18 -15.3, and the T values were 11.2-678 and 32.9 -496 for k a and k d , respectively. The units used are as follows: k a , M Ϫ1 s Ϫ1 ϫ 10 Ϫ3 ; k d , s Ϫ1 ϫ 10 3 ; and K D , nM. The k d value is on the border of the measurable interval (10 Ϫ5 to 10 Ϫ1 s Ϫ1 ) in the BIAcore 2000 instrument, introducing uncertainty in the K D value calculation. However, steady state affinity evaluation of the same data gave a K D of 2970 nM for the brevican lectin interaction with TNfn- (3)(4)(5). a much weaker interaction than with the other ECM ligands. This may be due to masking of its binding site during immobilization to the BIAcore chip, since in earlier studies with fibulin-1, it was the aggrecan lectin domain that was immobilized and not the fibulin fragment. Indeed, in the reverse orientation experiment described above, the interaction between fibulin-1 and LCt was also observed (Table II).
The interactions of the G3 variants with tenascin-C fragments were investigated by ligand overlay blot studies. Aliquots of bacterially expressed recombinant tenascin-C fragments (35) were separated by SDS-PAGE and blotted onto nitrocellulose filters. The filters were incubated with three of the 6-histidine-tagged G3 variant proteins, LCt, E1E2LCt, and Ct, in the presence and absence of calcium and immunoblotted with an antibody against the 6-histidine tag (Fig. 4). LCt and E1E2LCt can be seen to bind to both tenascin-C fragments 1-8 and 3-5 in the presence of 5 mM calcium. This binding is abolished in the presence of EDTA and appears to be stronger with the E1E2LCt variant than with the LCt variant. The Ct fragment, which lacks the mammalian type C lectin-like domain present in the two other variants tested, does not bind to either of these recombinant forms of tenascin-C in the presence or absence of calcium. There is some nonspecific binding of all three G3-related proteins, E1E2LCt, LCt, and Ct, to the fibrinogen-like region of tenascin-C, but this binding is not abolished in the presence of EDTA.
In solid phase assays, alkaline phosphatase-tagged rat aggrecan lectin domain was incubated with fibulin-2 or tenascin-R surfaces in the presence of varying concentrations of the different G3 variants. As shown in Fig. 6 and Table III, the different G3 variants displayed different levels of inhibitory activity. Binding of the aggrecan lectin domain was inhibited by all of the G3 variants except Ct, indicating that they are all competing for the same binding site as the lectin domain alone. The calculated IC 50 values are shown in Table III. Interestingly, the E1E2LCt variant showed considerably stronger inhibition than the other G3 variants (Fig. 6 and Table III).
Aggrecan mRNA Transcripts Containing Both Epidermal Growth Factor-like Motifs Are Expressed in Human Chondrocytes-As E1E2LCt was the only G3 variant that differed in its affinity for the ECM ligands in this study, it was important to establish whether both of the EGF motifs can be expressed together in vivo to determine whether this greater binding capability is physiologically relevant. Reverse transcriptase-PCRs using RNA from primary human chondrocytes and prim-ers Agg4 (to the chondroitin sulfate attachment region 2) and Agg11 (to the CLD region) and Agg4 and Agg13 (to EGF2) (31) each yielded two products (Fig. 7A). In the Agg4 and Agg11 reaction, the two products of ϳ380 and 494 bp corresponded to the sizes expected for products containing neither or only one of the two EGF-like motifs of the aggrecan G3 domain. However, in the more EGF2 product-specific Agg4 to Agg13 reaction, the   two products formed corresponded to the sizes expected for EGF2 alone (241 bp) and for EGF1 and EGF2 expressed together (355 bp). Cloning and sequencing of the larger product determined that both of the EGF-like motifs can be expressed together in aggrecan mRNA transcripts in primary human chondrocytes (Fig. 7B).

DISCUSSION
In an earlier study, we demonstrated versican C-type lectin binding to tenascin-R in overlay assays on brain extract (14). We were, however, unable to detect versican interaction with tenascin-C in the same assays, even though immunoblotting revealed its presence. It has since been demonstrated that the neurocan C-type lectin domain can bind tenascin-C in similar assays (20). We now confirm the neurocan interaction with tenascin-C and demonstrate that aggrecan, versican, and brevican also bind tenascin-C. In fact, aggrecan, versican, and neurocan show fairly low affinities for tenascin-C compared with tenascin-R (Table I). Brevican, on the other hand, is a very weak tenascin-C binder but an avid tenascin-R binder ( Table I). The comparatively low affinities may explain why we did not detect tenascin-C interaction in our original study (14).
By using a set of recombinant tenascin-C fragments, we mapped the binding sites of the proteoglycan lectins to FnIII repeats 3-5 ( Fig. 5 and Table I). This is in agreement with the neurocan binding to FnIII-(4 -5) reported previously (20). Interestingly, exactly the same region (FnIII- (3)(4)(5)) was identified as the lectin interaction site on tenascin-R (15). The importance of these interactions is supported by the conservation of this proteoglycan G3 interaction site in different tenascins and by the strong conservation of the region in different species. The proteoglycan C-type lectin binding to the bacterially expressed tenascin-C fragments also shows that the interaction is independent of carbohydrates, as is the case for the interaction with tenascin-R (15).
Previous studies (14 -17, 44) on the G3 interactions with ECM molecules have predominantly used recombinant C-type lectin domains in isolation. Alternative splicing of the EGF and SCR modules in the G3 domain is well established (22)(23)(24)(25)(26) and could have an effect on the interactions of the C-type lectin repeat. Although there are binding studies using full-length proteoglycan (17) or recombinant G3 domains (14,15,20), no thorough investigation of the effects of alternative splicing in the G3 domain has been published. To this end we produced a set of recombinant aggrecan G3 modules containing different combinations of the EGF1, EGF2, C-type lectin, and SCR modules.
As expected, all the variant proteins containing the C-type lectin bound to the known lectin ligands tenascin-R, tenascin-C, and fibulin-2. Surprisingly, when fibulin-1 was immobilized, interaction was only observed with the E1E2LCt construct. We know from previous work that the aggrecan C-type lectin alone can bind fibulin-1 in BIAcore experiments (16) but also that fibulin-1 fails to bind aggrecan lectin in solid phase assays while binding well to versican lectin (17). We have no conclusive explanation for these discrepancies, but we speculate that the lack of interaction reflects changes in the conformation or accessibility of fibulins-1 when covalently immobilized on the surface instead of in solution. The observed interaction between fibulins-1 and LCt in reverse orientation BIAcore experiments, i.e. with aggrecan G3 variants immobilized, supports this assumption (Table II).
Thus it does not appear that inclusion of the alternatively spliced domains has an all-or-nothing effect on G3 interaction with extracellular matrix proteins. There are, however, differences in affinities for extracellular matrix proteins. Notably, the variant containing the full complement of modules, i.e. E1E2LCt, had distinctly higher affinity for tenascin-R, tenascin-C, and fibulin-2 in solid phase assays than did the other G3 variants. The K D values obtained from BIAcore experiments confirmed these observations. As already mentioned, the E1E2LCt was the only G3 variant that showed any binding to fibulin-1. The other G3 variant proteins showed more subtle differences in binding strength.
Although the crystal structure of any proteoglycan G3 domain has not been determined, molecular modeling of the aggrecan G3 domain suggests a structure where the calcium coordination sites of the C-type lectin are located at the opposite end of the lectin domain relative to the SCR and EGF modules (45). It is possible that the presence of the latter modules in the G3 could affect the activities of the G3 by stabilizing the C-type lectin in a high affinity state, for example. Another possibility is that the EGF1 module in the E1E2LCt is positioned in such a way that it is able to interact with either a site on the G3 or on its ligand. This modulation of ligand binding affinity may be of physiological importance, e.g. in fine-tuning matrix assembly.
Other functions have also been suggested for the G3 domain. Randomly primed reverse transcriptase product of mRNA prepared from primary articular chondrocytes was amplified with primers Agg4 and Agg11, and with Agg4 and Agg13 (31), and the products were run on a 1.5% agarose gel. B, sequencing of the cloned 355-bp fragment (marked with an arrow in A) revealed that both of the EGF-like motifs were present in aggrecan expressed in human articular chondrocytes.
For example, this domain has been suggested to function as an "intramolecular chaperone," making secretion of the aggrecan proteoglycan possible. A part of the C-type lectin corresponding to exon 15 of the aggrecan gene was suggested to be responsible for this function (46). In previous work we have shown that alternative splicing of the G3 domain has no discernible effect on recombinant aggrecan secretion. In fact, even the Ct splice variant lacking the lectin domain allowed efficient secretion of aggrecan chondroitin sulfate region (31), arguing against the notion of the C-type lectin as a chaperone. Another intriguing possibility is that the alternatively spliced EGF repeats of the G3 domain could actually function as growth factors. This is to say that upon proteolytic release, for example, they would bind EGF receptors on the cell surface and be involved in a type of auto/endocrine signaling loop. Binding of tenascin-C-derived EGF modules to EGF receptor with concomitant signaling has been reported (47). Indeed, the EGF repeats in versican have been reported to affect cell proliferation and differentiation (27,28), but no similar investigations have been reported for the aggrecan EGF repeats.
The present results outline more detail in the range of ligand protein interactions for the proteoglycan G3 domains. Together with the high conservation of G3 sequences throughout the proteoglycan family and the modulation of activity in aggrecan G3 by alternative splicing, this argues strongly for significant extracellular functions in matrix assembly and organization.