Interaction Properties of the Procollagen C-proteinase Enhancer Protein Shed Light on the Mechanism of Stimulation of BMP-1*

Procollagen C-proteinase enhancer (PCPE) is an extracellular matrix glycoprotein that binds to the C-propeptide of procollagen I and can enhance the activities of procollagen C-proteinases up to 20-fold. To determine the molecular mechanism of PCPE activity, the interactions of the recombinant protein with the procollagen molecule as well as with its isolated C-propeptide domain were studied using surface plasmon resonance (BIAcore) technology. Binding required the presence of divalent metal cations such as calcium and manganese. By ligand blotting, calcium was found to bind to the C-propeptide domains of procollagens I and III but not to PCPE. By chemical cross-linking, the stoichiometry of the PCPE/C-propeptide

trix metalloproteinases (33). Thus, different regions of PCPE may show either stimulatory or inhibitory activities to different subfamilies of metzincin metalloproteinases in the extracellular matrix. In addition, PCPE expression is implicated in the control of cell growth (34,35), as has also been reported for TIMPs (36).
The mechanism of PCP stimulation by PCPE is unknown. From kinetic studies (1,16), PCPE increases both the K m and V max for PCP/BMP-1 cleavage of the C-propeptide region from procollagen I. Furthermore, maximum enhancement is achieved at an equimolar ratio of PCPE to procollagen. This suggests that enhancement occurs at the level of the substrate and not the enzyme. Here we investigate the interactions of PCPE with the procollagen type I molecule as well as with the C-propeptide trimers that are released from procollagens I and III by PCP cleavage. We show by surface plasmon resonance technology that C-propeptide binding to PCPE is dependent on divalent metal binding to the C-propeptides. Chemical crosslinking indicates that PCPE binds to the C-propeptide trimer in a molar ratio of 1:1. Finally, PCPE binding to intact procollagen molecules appears to be tighter than to the isolated C-propeptide trimers, suggesting possible additional binding sites in other regions of the procollagen molecule, a conclusion supported by the binding of PCPE to procollagen molecules devoid of the C-propeptide region.

MATERIALS AND METHODS
Proteins and Antibodies-Recombinant human PCPE as well as the C-terminal propeptide trimer from procollagen III (CPIII) were expressed using baculovirus systems (19,37,38). Procollagen I and its C-propeptide trimer domain (CPI) were purified from the culture media of chick embryo tendon fibroblasts (39) or chick embryo tendons (40), respectively. For the preparation of pN-collagen I, procollagen I was fully cleaved with highly purified recombinant BMP-1 (1) in the absence of PCPE and then separated from both liberated C-propeptides and BMP-1 by gel filtration on a 1.6 ϫ 60-cm column of Sephacryl S200 HR (Amersham Biosciences) equilibrated with 0.4 M NaCl, 1 M urea, and 0.1 M Tris-HCl, pH 7.4 (see ref. 41). Reduced and alkylated CPIII was prepared as described (19). The binding epitopes of the monoclonal antibodies 48D34, 48B14, and 48D19 mapped to sites within residues 1-30, 31-207, and 208 -245, respectively, of the CPIII polypeptide chain as described (42).
Surface Plasmon Resonance (SPR)-Binding analysis was performed using a BIAcore Upgrade system (BIAcore AB, Uppsala, Sweden). PCPE was covalently coupled to CM5 sensor chips (research grade) via amine coupling (BIAcore AB amine coupling kit). The carboxymethylated dextran surface was activated by the injection of a mixture of 0.2 M N-ethyl-NЈ-(diethylamino-propyl)carbodiimide and 0.05 M N-hydroxysuccinimide. PCPE (the ligand) was then injected in 10 mM maleate buffer, pH 6.0. Activation time, PCPE concentration, and contact time were adjusted according to the desired extent of immobilization. The remaining N-hydroxysuccinimide esters were blocked by the injection of 1 M ethanolamine hydrochloride, pH 8.5. All immobilization steps were performed at a flow rate of 5 l/min. The immobilization of CPIII was carried out similarly except that the coupling buffer was 10 mM sodium acetate, pH 5.0.
Control flow cells were prepared by carrying out the coupling reaction in the presence of coupling buffer alone; these were used to obtain control sensorgrams showing nonspecific binding to the surface as well as refractive index changes resulting from changes in the bulk properties of the solution. Control sensorgrams were then subtracted from sensorgrams obtained with immobilized ligand to yield true binding responses.
Binding assays were performed at 25°C in 10 mM Hepes buffer, pH 7.4, containing 0.15 M NaCl and 0.005% (v/v) P20 surfactant (HBS-P buffer, BIAcore AB). CPI, CPIII, procollagen I, and pN-collagen I were dialyzed against HBS-P buffer and then injected at several concentrations and different flow rates over immobilized PCPE. The surface was regenerated with a pulse of 2 M guanidinium chloride. When possible, kinetic constants were calculated by nonlinear fitting to the association and dissociation curves according to the manufacturer's instructions (BIAevaluation 3.0 software). Apparent equilibrium dissociation constants (K D ) were then calculated as the ratio of k d /k a . Alternatively, K D values were calculated from the equilibrium resonance signal (R eq ) as a function of analyte concentration (43), R eq values being estimated by extrapolation to infinite time using plots of resonance signal as a function of the reciprocal of time (44). Apparent K D values were then calculated by nonlinear fitting to the expression R eq ϭ R max C/(K D ϩ C), where R max is the maximum binding capacity of the surface and C is the analyte concentration, using Kaleidagraph software.
Ligand Blotting-Calcium binding was studied by ligand blotting and 45 Ca autoradiography (45,46) using nitrocellulose membranes (Schleicher and Schü ll). Different concentrations of purified proteins were applied using a dot-blot apparatus. Each well was washed twice (2 ϫ 200 l) with 10 mM imidazole buffer, pH 6.8, containing 60 mM KCl and 5 mM MgCl 2 . Membranes were incubated with 45 CaCl 2 (Amersham Biosciences) at a final concentration of 2 Ci/ml for 15 min, rinsed with 50% ethanol for 5 min, dried, and exposed overnight to film (BioMax MS) in a cassette with an intensifying screen. Gelsolin (gift from A. Zapun, Institut de Biologie Structurale, Grenoble, France) was used as a positive control (47), and collagen II was used as a negative control (46).
Cross-linking-These experiments were conducted in two steps. First, purified PCPE (7.3 M) in 0.15 M NaCl, 10 mM Hepes, pH 7.4 was incubated in the dark with a 10-fold molar excess of the extended chain length (spacer arm 18.2 Å), photoactivable, heterobifunctional crosslinker SANPAH (Perbio Science; reactive toward amino groups) for 15 min at room temperature. A 700-fold molar excess of glycine was then added, and the incubation was continued for 15 min. At the end of the incubation period, excess reagent was eliminated by centrifugation filtration through a 1-ml column filled with Sephadex G-50 equilibrated with 0.15 M NaCl, 10 mM Hepes, pH 7.4. Purified CPIII (0.4 M) was then incubated for 45 min in the dark at room temperature with an ϳ3-fold molar excess of PCPE-SANPAH derivative (1.3 M) in the previously described buffer supplemented with 5 mM CaCl 2 . Photoactivation was then induced at 312 nm for 2.5 min with a 6 watt Spectroline lamp at a distance of 7 cm from the sample. After SDS-PAGE using a 3-8% gradient gel in nonreducing conditions, proteins were electroblotted for 4 h at 4°C. Duplicate samples were examined by Western blotting, one using anti-rPCPE rabbit polyclonal antiserum and the other using monoclonal antibody 48D19 against CPIII (42). Control experiments were carried out in the absence of the cross-linking reagent.

Binding of C-propeptide Trimers to PCPE-Interactions
were studied using surface plasmon resonance (BIAcore) technology. When passed over the surface of the sensor chip, CPIII was found to bind both specifically and in a concentration-dependent manner to immobilized PCPE (Fig. 1A). As the PCPE molecule was immobilized at pH 6.0 (see "Materials and Methods"), it is possible that PCPE was immobilized mostly via the basic NTR domain (calculated pI 9.21), leaving the two acidic CUB domains (calculated pI 5.77 and 4.33 for CUB1 and CUB2, respectively) essentially free to interact with the C-propeptide. This is supported by experiments in which native PCPE lacking the NTR domain was immobilized when no interaction with CPIII was observed (data not shown). Because PCPE enhancement of PCP/BMP-1 is known to be a property of the CUB domain region of PCPE (18,31,32), it is likely that immobilization in the absence of the NTR domain interferes with CPIII binding sites in the CUB domains.
The binding of CPIII to immobilized PCPE was initially fast but then slowed down without reaching saturation equilibrium (Fig. 1A). Dissociation of bound CPIII was rapid. These combined features made fitting of a kinetic model to the data, and hence the determination of reliable on-and off-rates, difficult. Using global fitting, binding curves did not fit well to the different models included in the BIAevaluation 3.0 software (1:1 Langmuir binding, bivalent analyte, heterogeneous ligand, heterogeneous analyte, conformational change) with or without mass transport. Neither injection at a higher flow rate (30 l/min) nor the use of a F1 sensor chip with a shorter dextran layer changed the shape of the sensorgrams. Furthermore, because equilibrium was not reached during the association phase, the direct use of Scatchard analysis to calculate the apparent equilibrium dissociation constant was not allowed. Instead, to calculate the equilibrium dissociation constant, association curves were extrapolated to infinite time using recip-rocal plots in order to estimate the equilibrium binding value R eq (43,44). Nonlinear regression to the standard hyperbolic expression (see "Materials and Methods"), also shown in Fig.  1A, gave an apparent K D of 369 Ϯ 50 nM and a R max of 632 Ϯ 9 resonance units for the interaction of CPIII with PCPE.
Specific, concentration-dependent binding of PCPE to CPIII could also be demonstrated in the reverse manner by injecting the whole PCPE molecule over immobilized CPIII (Fig. 2). Unlike the results with immobilized PCPE (Fig. 1A) however, these data were obtained with relatively large amounts of immobilized ligand, a condition that precluded kinetic analysis and accounted for the relatively rapid saturation of immobilized CPIII binding sites.
As shown in Fig. 1B, the kinetics for the binding of CPI to immobilized PCPE were similar to those for CPIII, although the dissociation rate was somewhat slower. This permitted evaluation of the off-rate constant k d from the dissociation phase, independent of k a but simultaneous for all curves, using the 1:1 (Langmuir) dissociation model. In this way, k d was found to be 5.87 ϫ 10 Ϫ2 s Ϫ1 (chi-square 4.18). Neither the association phase nor the dissociation phase was modified when CPI was injected at different flow rates (5 and 15 l/min) for different contact times (injected volumes 60 and 120 l) or on different levels of immobilized PCPE (383 and 1080 resonance units). This suggests that the interaction was not significantly limited by mass transport and did not involve linked reactions (e.g. conformational change following ligand-analyte binding or binding of heterogeneous analyte to a single ligand). Because satisfactory fitting of the association phase was not possible with the BIAevaluation 3.0 software, apparent K D values were calculated from estimated equilibrium values determined as for CPIII. As also shown in Fig. 1B, nonlinear regression gave an apparent K D of 169 Ϯ 23 nM and R max of 637 Ϯ 11 resonance units for the interaction of CPI with PCPE. In summary, the above results show that binding of the procollagen C-propeptide trimers (types I or III) to PCPE is of moderate affinity, with K D values in the range 150 -400 nM.
Characterization of the CPIII-PCPE Interaction-CPIII bound to immobilized PCPE whether the surface was prepared in the presence or absence of 5 mM CaCl 2 . Furthermore, CPIII binding was unaffected by pretreatment of the PCPE surface with HBS-P buffer containing 10 mM EDTA followed by reequilibration in HBS-P. In contrast, when CPIII was freed of bound divalent metal cations by extensive dialysis against HBS-P containing 10 mM EDTA and then redialyzed against HBS-P alone before injection over PCPE, the binding of CPIII to immobilized PCPE was abolished (Fig. 3). The addition of 2 mM CaCl 2 , MnCl 2 , or MgCl 2 restored the binding (Fig. 3). These data suggest that divalent metal cation binding to CPIII is essential for binding to immobilized PCPE.
To confirm the results obtained using SPR analysis, we tested directly the binding of calcium to CPIII by the ligand blotting technique (45). Procollagen I, CPI, and CPIII all showed a distinct affinity for 45 Ca unlike PCPE, which did not bind 45 Ca significantly under the experimental conditions used (Fig. 4). As expected (46), collagen II did not bind 45 Ca, whereas gelsolin, a calcium binding protein used as a positive control (47), did.
The extent of binding of CPIII to immobilized PCPE was similar in the presence of 0.15 M, 0.3 M, and 0.45 M NaCl. Furthermore, CPIII bound PCPE to a similar extent in the absence or presence of methyl-␣-mannopyranoside at concentrations up to 12.5 mM (data not shown). Thus, CPIII binding to PCPE did not appear to involve mannose-containing glycosylation sites present in either protein.
To identify potential PCPE binding sites within the C-propeptide domain, we used monoclonal antibodies directed against different regions of CPIII (42). Affinity-purified monoclonal antibodies (150 g/ml) were pre-incubated at room temperature for 70 min with CPIII (150 g/ml), and then the mixture was injected over immobilized PCPE. Using mAb 48D34 directed against an epitope within the 30 N-terminal residues of the CPIII polypeptide chain, CPIII continued to bind to immobilized PCPE, but the SPR signal was greater than with CPIII alone (Fig. 5). No antibody binding was observed in the absence of CPIII. These results indicated that the mAb-CPIII complex bound to PCPE, thus generating an increased SPR signal, and suggested that the N-terminal region of CPIII did not participate in PCPE binding. Results using mAbs directed against other regions of CPIII were less readily interpretable (data not shown).
To measure the number of molecules of PCPE bound per CPIII trimer, interactions between the two partners in the presence of a 3-fold molar excess of PCPE were stabilized by covalent cross-linking. As shown in Fig. 6, in the presence of the cross-linker an additional band of about 140 kDa was recognized by the anti-CPIII mAb, whereas without crosslinker only the CPIII trimer (ϳ90 kDa in the unreduced state) was revealed. The 140-kDa band was also recognized by the anti-PCPE antiserum and was the only high molecular mass complex to be detected by both antibodies. The apparent molecular mass of the band revealed by cross-linking corresponded to the sum of the apparent molecular masses of one CPIII trimer (ϳ90 kDa) plus one molecule of PCPE (ϳ50 kDa). Thus the binding stoichiometry appeared to be 1:1. Further BIAcore experiments showed that immobilized PCPE did not interact with reduced and alkylated CPIII (data not shown), suggesting that the C-propeptide must be in its native trimeric state to interact with PCPE.
Binding of Procollagen to PCPE-Intact procollagen I molecules were also found to bind specifically to immobilized PCPE in a concentration-dependent manner (Fig. 7). In contrast to the results obtained with isolated C-propeptide trimers (Fig. 1), however, the association phase and in particular the dissociation phase were slow, suggesting that the complex formed between the procollagen I molecule and PCPE was very stable.
Association and dissociation phases (Fig. 7) were satisfactorily fitted (chi-square 0.844) simultaneously for all curves using a 1:1 Langmuir model. Calculated on-and off-rates were 5.52 ϫ 10 4 M Ϫ1 s Ϫ1 and 6.24 ϫ 10 Ϫ5 s Ϫ1 , respectively, corresponding to an affinity constant K D of 1.13 nM and R max of 510 -740 resonance units. The addition of mass transport did not improve the fit, although a good fit was also obtained using the two-state reaction model with conformational change (chisquare 0.854). To further evaluate possible mass transport  6. Stoichiometry of the PCPE/CPIII interaction. PCPE, previously treated (ϩ) or not previously treated (Ϫ) with the photoactivable cross-linking agent SANPAH, was incubated with CPIII at a molar ratio of 3:1 (PCPE/CPIII) as described under "Materials and Methods." After electrophoresis in unreduced form and electrotransfer, proteins were detected with anti-CPIII mAb 48D19 (A) or anti-rPCPE polyclonal antiserum (B). The cross-linked CPIII/PCPE complex recognized by both antibodies is indicated by the arrow. PCPE alone has a tendency to self-polymerize.
limitations, another set of data was collected at a higher flow rate (50 l/min) which gave a similar K D value, suggesting that mass transport effects did not occur to a significant extent under the experimental conditions used. In addition, a 2-or 4-fold increase in the flow rate (15,30, and 60 l/min) at a single procollagen concentration did not change the initial rate of the association phase and confirmed that mass transport did not interfere with the calculation of the rate constants.
Because the binding of the intact procollagen molecule to PCPE (K D ϭ 1 nM) appeared to be tighter than the binding of isolated C-propeptide trimers (K D ϭ 150 -400 nM), it is likely that there might be additional sites for PCPE binding elsewhere in the procollagen molecule. To test this hypothesis, we measured the binding of pN-collagen I (i.e. procollagen lacking the C-propeptide region) to immobilized PCPE. Specific binding of pN-collagen to PCPE was observed, and the shape of the corresponding sensorgram showed similar characteristics to those obtained with the entire procollagen I molecule with a very slow dissociation rate (Fig. 8). The limited solubility of the pNcollagen I prevented binding studies at a range of concentrations. These data suggest that PCPE binds to sites in the procollagen molecule in addition to those in the C-propeptide region. DISCUSSION We report here that the C-propeptide trimers of procollagens I and III bind Ca 2ϩ . Such binding has been demonstrated previously for the C-propeptide trimer of procollagen II, also known as chondrocalcin (48,49), as well as for the collagen X molecule (46). These are in addition to the group of known Ca 2ϩ -binding extracellular matrix proteins that includes SPARC/BM-40/osteonectin, fibrillin, and COMP, for which a number of different types of Ca 2ϩ binding sites have been described (50,51). Using PROSCAN (52), we found that the C-propeptide domains of the pro-␣1 chains of mammalian procollagens I, II, and III all contain a strongly conserved sequence (Cys-47 to Cys-73 following the BMP-1 cleavage site, numbered according to the pro-␣1(I) sequence), which shows 79% similarity to the Ca 2ϩ binding C-type lectin domain motif as found, for example, in MBL. Furthermore, within this sequence residues 59 to 71 are 74% similar to the Ca 2ϩ binding EF-hand domain motif as found in SPARC (53), although the invariant Glu or Asp at position 12 is lacking. This suggests that the Ca 2ϩ binding site(s) may be localized to these sequences. In addition, the positions of ϳ50% of the acidic residues in the C-propeptide domains are strongly conserved, half of which are found in two acidic clusters in the C-terminal region (residues 180 -246). Because relatively weak calcium binding can also be attributed to glutamic acid-or aspartic acid-rich sequences (50,51), these also may represent potential Ca 2ϩ binding sites.
Possible interaction sites for PCPE in the procollagen Cpropeptide region can be proposed on the basis of recent small angle x-ray scattering studies of CPIII (37). These indicate a structure consisting of three major lobes, each of which might correspond to the C-terminal region of each component chain containing the internal disulfide bonds, plus one minor lobe corresponding to a putative N-terminal junction region (residues 1-80) containing the interchain disulfide bonds. In view of our observation here that only one molecule of PCPE binds to the procollagen C-propeptide trimer, which is also supported by enzyme kinetic studies (1, 16), we speculate that the binding site is in this N-terminal junction region. Furthermore, because interaction with PCPE did not prevent binding of mAb 48D34 to the 30 N-terminal residues of CPIII, this might further limit the possible binding site to residues 31 to 80, which contains all the interchain disulfide bonds (54) as well as the putative calcium binding site(s) (see above). Finally, the binding to the junction region is consistent with the observed lack of binding to PCPE following reduction and alkylation of CPIII, conditions that are likely to result in dissociation of the three polypeptide chains and/or changes in the three-dimensional structure.
The interaction studies described here on PCPE give insights into the mechanism by which it stimulates the activity of PCP/ BMP-1. At least two possible mechanisms can be proposed (Fig.  9). One is that PCPE facilitates dissociation of the enzyme following procollagen cleavage by competition for common binding sites in the C-propeptide region (Fig. 9A). Such a mechanism would be possible if PCPE binds only to the liberated C-propeptide trimer but not to the intact procollagen molecule; otherwise, inhibition of PCP activity might be expected. Here we find that PCPE does bind to procollagen (K D , 1 nM) and that this binding appears to be tighter than that to the isolated C-propeptide trimer (K D , 150 -400 nM). Thus the "facilitating product release" hypothesis ( Fig. 9A) seems unlikely.
The second hypothesis for PCPE action, originally proposed by Kessler and co-workers (16), is that PCPE binding to procollagen brings about a conformational change in the substrate, thereby facilitating cleavage by PCP (Fig. 9B). The observation here that PCPE binds to the C-propeptide domain as well as to additional sites in the procollagen molecule is consistent with such a mechanism, although the precise locations of the binding sites remain to be determined. It seems unlikely that PCPE binds to the N-propeptide domain in view of the length and rigidity of the procollagen molecule (PCPE shows no enhancing activity on procollagen N-proteinase (32)). A more likely possibility is that PCPE binds within the mature collagenous region not far away from the PCP cleavage site. In this way, PCPE binding to sites within both the C-propeptide and collagenous regions would involve an additional contact interface that might lead to a conformational change in the cleavage site, thus facilitating PCP action (Fig. 9B). It should be noted that the data for procollagen binding to PCPE were equally well fitted to a two-state reaction model with conformational change.
The interaction of PCPE with procollagen is reminiscent of the interactions of other CUB domain-containing proteins. These include the complement serine proteases C1r and C1s, which interact with the collagen-like regions of the C1q molecule (55). Similarly, the MBL-associated serine proteases MASP-1 and MASP-2 and also the MBL-associated protein 19 (MAp19) interact with the collagen-like region of MBL (56,57). The observation here that PCPE might interact with the collagenous region of the procollagen molecule suggests that CUB domains might be well adapted for interacting with collagen triple helices. Further experiments are necessary to determine the precise binding sites for PCPE on the procollagen molecule (both in the C-propeptide region and elsewhere) and also to characterize the interactions of PCPE with procollagen C-proteinases and their various substrates.