A Substrate Specificity-determining Unit of Three Lin12-Notch Repeat Modules Is Formed in Trans within the Pappalysin-1 Dimer and Requires a Sequence Stretch C-terminal to the Third Module*

Members of the pappalysin family of metzincin metalloproteinases, pregnancy-associated plasma protein-A (PAPP-A, pappalysin-1) and PAPP-A2 (pappalysin-2), regulate the bioavailability of insulin-like growth factors (IGFs) by specific proteolytic inactivation of IGF-binding proteins (IGFBPs). PAPP-A cleaves IGFBP-4 and IGFBP-5, whereas PAPP-A2 cleaves only IGFBP-5. The pappalysins contain three Lin12-Notch repeat (LNR1–3) modules, previously considered unique to the Notch receptor family in which they function to regulate receptor cleavage. In contrast to the Notch receptor where three LNR modules are tandemly arranged, LNR3 is separated by more than 1000 residues from LNR1–2 in the pappalysin sequence. Each of the three LNR modules of PAPP-A is required for proteolysis of IGFBP-4, but not IGFBP-5. However, we here find that a C-terminal truncated variant of PAPP-A, which lacks LNR3 and therefore activity against IGFBP-4, cleaves IGFBP-4 when co-expressed with a PAPP-A variant, which is mutated in the active site. This suggests that LNR3 from the inactive subunit interacts in trans with LNR1–2 of the truncated PAPP-A subunit to form a functional trimeric LNR unit. We also show that formation of such a functional LNR unit depends on dimerization, as dissociation of a mutated non-covalent PAPP-A dimer results in reduced activity against IGFBP-4, but not IGFBP-5. Using PAPP-A/PAPP-A2 chimeras, we demonstrate that PAPP-A2 LNR1–2, but not LNR3, are functionally conserved with respect to IGFBP proteolysis. Additionally, we find that a sequence stretch C-terminal to LNR3 and single residues (Asp1521, Arg1529, and Asp1530) within this are required for LNR functionality.

PAPP-A is the founding member of the pappalysin family (12,13) within the metzincin superfamily of metalloproteinases (14,15), also including its only known homologue, PAPP-A2, which cleaves IGFBP-5, but not IGFBP-4 (16). The 1558 3 -residue PAPP-A2 shares 46% of its residues with the 1547-residue PAPP-A (17), and they display the same domain organization with a laminin G-like domain and a proteolytic domain in the N-terminal part (18), a central region of ϳ500 residues with unknown domain composition, and a C-terminal part with five complement control protein (CCP) modules that mediate cell surface adhesion of PAPP-A (19,20). The pappalysins also contain three Lin12-Notch repeat (LNR) modules, previously considered unique to the Notch receptor family. But in contrast to the Notch receptors, which invariably contain three tandemly arranged LNR modules, two of the LNR modules of the pappalysins (LNR1-2) are located together within the sequence of the proteolytic domain, and the third (LNR3) is located C-terminal to the CCP modules, separated by more than 1000 residues from LNR1-2.
In the Notch receptor, several studies indicate that the LNR modules are involved in the regulation of ligand-induced receptor cleavage by metzincin metalloproteinases of the ADAM family (21,22). This cleavage is critical for the subsequent intramembrane processing of the receptor by ␥-secretase and hence Notch signaling (23,24). Although the underlying mechanism is unknown, mutation or deletion of the LNR modules has been found to cause ligand-independent cleavage of the Notch receptor (25,26), whereas removal of the ligand-binding sites results in receptors that are resistant to proteolysis (27,28). It is therefore believed that the LNR modules function to restrain access to the cleavage site when ligand is not bound and that receptor ligand binding relieves this effect.
Recent data demonstrated that each of the three LNR modules is required for PAPP-A proteolysis of IGFBP-4, but not IGFBP-5 (29). It was therefore speculated that the three LNR modules of PAPP-A, although separated in the primary structure, interact with each other to determine PAPP-A substrate specificity. We here report data suggesting that within the disulfide-linked PAPP-A homodimer, such a functional LNR unit is formed in trans by the interaction between LNR1-2 of one subunit and LNR3 of the other subunit. By means of PAPP-A/ PAPP-A2 chimeras, we show that the LNR1-2 modules, but not LNR3, are functionally conserved. We also identify a region C-terminal to LNR3, and several charged residues within this region, as critical for PAPP-A LNR functionality.  )) and triple (PA(C381A/ C652A/C1130A)) point mutants of the 1547-residue human PAPP-A polypeptide were obtained using three pBluescript II SKϩ vectors each containing PAPP-A cDNA corresponding to residues 1-407 (pB1-407), 408 -988 (pB408 -988), or 989 -1547 (pB989 -1547) (12). Mutations were introduced by overlap extension PCR (30) using Pfu DNA polymerase (Promega), and the mutated cDNAs were swapped into the HindIII/BspEI, BspEI/KpnI, or the KpnI/XbaI sites of pPA-BspEI (12), a modified version of the PAPP-A wild-type construct, pcDNA3.1-PAPP-A containing a BspEI site silently introduced at residue 407. Outer primers, derived from pBluescript II SKϩ were 5Ј-TAATACGACTCACTATAGGG-3Ј and 5Ј-AATTAACCCTCACTAAAGGG-3Ј.
The internal primers used to generate mutated or chimeric fragments all had an overlap of 15-20 bp. The resulting PCR products for chimeric constructs were swapped into the BspEI/ XbaI or KpnI/XbaI sites of pPA-BspEI.
Cell Culture and Expression of Recombinant Proteins-Human embryonic kidney 293T cells (293tsA1609neo) (32) were maintained in high-glucose Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 2 mM glutamine, nonessential amino acids, and gentamicin (Invitrogen). Cells were plated onto 6-cm tissue culture dishes and transfected 18 h later by calcium phosphate co-precipitation using 10 g of plasmid DNA (33), or in the case of co-transfections 10 g of plasmid DNA of each PAPP-A variant, prepared by QIAprep Spin Kit (Qiagen). After 48 h, the culture media were harvested, cleared by centrifugation, and replaced by serum-free medium to be harvested after another 48 h.
Western Blotting-For immunovisualization, culture media containing recombinant wild-type or mutant protein were separated by non-reducing SDS-PAGE (5-15% Tris glycine gels) and blotted onto a polyvinylidene difluoride membrane (Millipore). The blots were dried and blocked in 2% Tween 20 diluted in 50 mM Tris, 500 mM sodium chloride, 0.1% Tween 20, pH 9.0 (TST). After washing and equilibration in TST, the membrane was incubated with primary antibody (polyclonal anti-PAPP-A/pro-MBP) (34) diluted in TST containing 1% fetal calf serum for 1 h at room temperature . Blots were washed in TST, incubated with secondary peroxidase-conjugated anti-rabbit IgG (P217, DAKO) diluted in TST containing 1% fetal calf serum for 1 h at room temperature, and again washed in TST. Proteins were detected by enhanced chemiluminescence (ECL, Amersham Biosciences) and images were captured with a Kodak Image Station 1000.
Enzyme-linked Immunosorbent Assay (ELISA)-The levels of recombinant wild-type PAPP-A, PAPP-A/PAPP-A2 chimeras, or PAPP-A mutants or truncated variants in culture medium were measured by a sandwich ELISA. 96-Well plates (Nunc) were coated with anti-PAPP-A/pro-MBP (34), blocked by incubation with phosphate-buffered saline (PBS) containing 2% bovine serum albumin. After sample incubation, a PAPP-Aspecific monoclonal antibody (mAb 234-5 or 234-2 (35), which recognizes the N-terminal portion and the C-terminal portion of PAPP-A, respectively, 4 followed by peroxidase-conjugated anti-mouse IgG (P260, DAKO), was used for detection. Antibodies were diluted in PBS containing 0.01% Tween 20 (PBST) and 1% bovine serum albumin. PBST was used for washing. A double monoclonal ELISA was used for detection of the PAPP-A dimer. In this assay, plates were coated with mAb 234-5 and sample incubation (100 l containing 25 nM of PAPP-A or PAPP-A mutant) was followed by incubation (1 h at 37°C) with PBS to which 0 -1 M NaCl was added. Biotinylated mAb 234-5, followed by incubation with peroxidase-conjugated avidin (P0347, DAKO), was used for detection. Washing after incubation with peroxidase-conjugated avidin was carried out using PBST supplemented with 200 mM NaCl to avoid nonspecific binding. Otherwise, washing and blocking was as described above. Dilution series of the PAPP-A/pro-MBP complex purified from pregnancy serum (34) were used in both ELISAs to establish standard curves.
Measurement of Proteolytic Activity-The proteolytic activity of wild-type PAPP-A or PAPP-A mutants was compared by incubating with purified 125 I-labeled substrates, as described for IGFBP-4 (3, 12) and IGFBP-5 (3). The expression levels of wild-type PAPP-A and PAPP-A mutants were determined by ELISA (as described above) and the amount of proteinase was adjusted to equimolar levels by dilution of the culture medium. Briefly, reactions (total sample volume of 20 l) were carried out in 50 mM Tris-HCl, 100 mM NaCl, 1 mM CaCl 2 , pH 7.5, at an enzyme:substrate ratio of 1:160 (0.1 nM proteinase and 16 nM substrate), in the presence (IGFBP-4) or absence (IGFBP-5) of 50 nM IGF-II (Diagnostic Systems Laboratories). Following incubation for 2 h at 37°C, reactions were stopped by addition of 10 mM EDTA and stored at Ϫ20°C. Cleavage products were separated by 10 -20% SDS-PAGE and visualized by autoradiography. The degree of cleavage was determined by quantification of band intensities using a Typhoon imaging system (Amersham Biosciences), and background levels (mock signal) were subtracted. The degree of cleavage by wild-type PAPP-A under these conditions (2 h incubation) was ϳ60%. The affect on proteolysis of PA(1133-1547) was analyzed by the addition of ϳ100 nM to the reaction. To assay the proteolytic activity of dissociated wild-type PAPP-A or the PA(C381A/C652A/ C1130A) mutant dimer, a 96-well plate coated with mAb 234-5 was incubated (1 h at 37°C) with 100 l of culture medium containing 25 nM proteinase, washed in PBST, and incubated (1 h at 37°C) with PBS with or without an additional 0.5 M NaCl. Following washing and equilibration in 50 mM Tris-HCl, 100 mM NaCl, 1 mM CaCl 2 , pH 7.5, reaction mixtures containing radiolabeled substrate (as described above) were added directly to the wells (total sample volume of 50 l). After incubation for 2 h at 37°C, 10 l of reaction sample was stopped by the addition of 10 mM EDTA and stored at Ϫ20°C. When assaying the IGFBP-4 proteolytic activity of PAPP-A/PAPP-A2 chimeric proteins, proteinase levels were adjusted to have the same activity against IGFBP-5 (40% cleavage after 2 h incubation), and were tested for IGFBP-4 proteolytic activity by time course experiments (reaction mixtures as described above using a total sample volume of 70 l). Samples of 10 l were stopped (as described above) at each time point (from 0 to 24 h). The initial cleavage rate of the chimeras was determined as the inclination of the linear phase of curves from the IGFBP-4 time course experiments (% cleavage of IGFBP-4 plotted as a function of time), and plotted relative to the wild-type PAPP-A IGFBP-4 cleavage rate.
Chemical Cross-linking-Proteins contained in serum-free culture medium (ϳ20 g/ml) and dialyzed against 50 mM Hepes, 100 mM NaCl, pH 7.5, were incubated with 0.05 or 0.5 mM EGS (ethylene glycolbis(succinimidyl succinate), Pierce) dissolved in dimethyl sulfoxide (Me 2 SO) for 30 min at room temperature. Samples (30 l) of the reaction mixtures were quenched with 5 l of 1 M Tris-HCl, 1 M glycine, pH 7.5, and mixed with SDS-PAGE loading buffer with reductant (20 mM dithiothreitol). Cross-linked adducts were separated by 5-15% SDS-PAGE and visualized by Western blotting as described above.
Size Exclusion Chromatography-Samples (600 l) of PAPP-A wild-type and the PAPP-A C381A/C652A/C1130 mutant contained in diluted (1:2) culture medium were loaded onto a Superose 6 10/300 GL column (Amersham Biosciences) equilibrated with 50 mM Tris-HCl, 50 mM NaCl, 2 mM CaCl 2 , pH 7.5. The flow rate was 0.3 ml/min and 0.3-ml samples were collected. The amount of protein contained in each fraction was measured by ELISA (as described above).
Native PAGE-Proteins were loaded on a native 5-10% Tris glycine gels (36) prepared without SDS, electrophoresed for 3 h at 10 mA, and visualized by Western blotting as described above using an anti-c-Myc antibody (mAb 9E10) as primary antibody and secondary peroxidase-conjugated anti-mouse IgG (P260, DAKO). To allow detection of PAPP-A by anti-c-Myc mAb, PA-P2(1529 -1547), containing PAPP-A2 residues 1550 -1558 followed by the c-Myc epitope in the C terminus was used.
Surface Plasmon Resonance Analysis-Surface plasmon resonance experiments were carried out on a BIAcore T100 instrument (BIAcore AB, Uppsala, Sweden) using series S CM5 sensor chips and coupling reagents supplied by the manufacturer. Immobilization of anti-c-Myc (9E10) was performed by first activating the surface of the sensor chip with a mixture of coupling reagents EDC/NHS followed by injection of purified mAb 9E10 (30 g/ml in 10 mM sodium acetate, pH 5.0) at 25°C. Remaining activated groups were blocked by 1 M ethanolamine. The level of mAb 9E10 immobilization corresponded to 10,000 resonance units. Purified c-Myc-tagged IGFBP-4 (3) (10 g/ml) or IGFBP-5 (3) (5 g/ml) diluted in 10 mM HEPES, 150 mM NaCl, 1 mM CaCl 2 , 0.05% Tween 20, pH 7.4 (HBS), were injected over the sensor chip for 4 min at a flow rate of 10 l/min at 25°C. For IGFBP-4, IGF-II (10 g/ml) diluted in HBS was subsequently injected for 1 min at a flow rate of 10 l/min. After onset of the dissociation phase, PAPP-A variants (PA(E483A), PA(E483A/1-1477), PA(E483A/ D1521A), PA(E483A/R1529A), PA(E483A/D1530A), or PA(1133-1547) were injected. All PAPP-A variants were purified by immunoaffinity chromatography using two PAPP-Aspecific monoclonal antibodies (mAbs 234-2 and 234-5) immobilized to CNBr-activated Sepharose 4B (GE Healthcare), dialyzed into HBS, concentrated by ultrafiltration, adjusted to a concentration of 0.15 mg/ml, and injected over the sensor surface at a flow rate of 10 l/min, at 25°C. The effect of PA(1133-1547) on PA(E483A) binding was analyzed by the injection of mixtures containing equal amounts or a 10-fold molar excess of PA(1133-1547). Recorded signals were subtracted from the

LNR Modules Interact in Trans within the PAPP-A Dimer-
Based on analogy to the tandemly arranged Notch receptor LNRs, we hypothesize two scenarios of PAPP-A LNR interactions. In one model, PAPP-A is folded so that LNR3 interacts with LNR1-2 of the same polypeptide. Alternatively, as PAPP-A is a dimeric protein, LNR1-2 from one subunit interacts with LNR3 of the other subunit within the PAPP-A dimer. To distinguish between these two possibilities, cells were cotransfected to express a C-terminal truncated variant of PAPP-A lacking LNR3, PA(1-1477) (29), and a proteolytically inactive PAPP-A variant with the active-site glutamic acid substituted with alanine, PA(E483A) (12). As expected, when either PAPP-A variant was expressed alone, no activity against IGFBP-4 could be detected in the culture medium (Fig. 1A,  lanes 3 and 4), and activity against IGFBP-5 was detected only with culture medium containing the truncated PAPP-A variant, PA(1-1477) (Fig. 1B, lane 3). However, upon co-expression, proteolytic activity against IGFBP-4 was rescued, corresponding to the formation within the dimer of one proteolytic domain with activity toward IGFBP-4 (Fig. 1A, lane 5). This is most likely explained by an interaction of LNR3 from the inactivated subunit (E483A) with LNR1-2 from the truncated PAPP-A subunit, PA(1-1477), suggesting that the three LNR modules interact in trans within the PAPP-A dimer (Fig. 1C).
Dimerization, but Not Intersubunit Disulfide Bond Formation, Is Required for PAPP-A LNR Functionality-The above findings lead us to analyze the functional implications of disrupting the covalently associated PAPP-A dimer. An expression construct encoding a PAPP-A mutant, PA(C381A/C652A/ C1130A), which is unable to form the Cys 1130 -Cys 1130 disulfide bond responsible for covalent dimerization (37), was constructed. To avoid possible unspecific polymerization through Cys 381 or Cys 652 , known to be unpaired (38), alanine substitutions of these residues were included in the mutant. This PAPP-A mutant displayed wild-type level activity against IGFBP-4 ( Fig. 2A, lane 3), suggesting that covalent dimerization is not required for LNR functionality. To test if the PAPP-A mutant forms a non-covalent dimer, we subjected it to chemical cross-linking. Upon reducing SDS-PAGE, the cross-linked mutant was found to co-migrate with cross-linked wild-type PAPP-A (Fig. 2B), demonstrating that the mutant is able to dimerize non-covalently. This finding was confirmed by size exclusion chromatography, in which PA(C381A/C652A/ C1130A) and PAPP-A wild-type co-eluted (data not shown).
When the PA(C381A/C652A/C1130A) mutant was immobilized on plastic by means of a monoclonal antibody, increasing ionic strength dissociated the non-covalently associated dimers (Fig. 3A), whereas no effect was observed on wild-type PAPP-A dimers (Fig. 3B). In accordance with this observation, the detectable IGFBP-4 proteolytic activity of the immobilized and dissociated PAPP-A mutant was found to be extensively compromised (Fig. 4), whereas the IGFBP-5 proteolytic activity was only moderately reduced (data not shown), corresponding to the expected subunit dissociation. This experiment shows that non-covalent dimerization is required and sufficient for LNR functionality, further suggesting that an LNR unit is formed in trans between the subunits of the PAPP-A dimer.
The Pappalysin LNR Modules Differ in Functional Conservation-The PAPP-A homologue, PAPP-A2, which cleaves IGFBP-5, but not IGFBP-4, does not dimerize covalently, although the PAPP-A2 residue corresponding to PAPP-A Cys 1130 is also a cysteine (16). However, when analyzed by native PAGE, PAPP-A2 migrated similar to wild-type PAPP-A (Fig. 5A), suggesting that it exists as a non-covalently associated dimer. A weak band, corresponding to monomer, revealed some dissociation of the PAPP-A2 subunits under the experimental conditions employed.
To investigate whether the LNR modules of PAPP-A2 are functionally conserved within the context of the PAPP-A dimer, LNR1-2, LNR3, and LNR1-3 of PAPP-A were replaced with the corresponding PAPP-A2 sequence. All variants were expressed as covalent dimers (Fig. 5B) and displayed PAPP-A wild-type level IGFBP-5 proteolytic activity (Fig. 5C), suggesting that the integrity of the proteolytic domain is maintained intact. When all three LNR modules of PAPP-A2 were inserted into PAPP-A, or when LNR3 alone was replaced by the PAPP-A2 sequence, very little activity against IGFBP-4 could be detected (Fig. 5D, lanes 3 and 4). In contrast, PAPP-A containing PAPP-A2 LNR1-2 showed wild-type level IGFBP-4 proteolytic activity (Fig. 5D, lane 5), demonstrating that full LNR functionality can be obtained with LNR1-2 of PAPP-A2 and LNR3 of PAPP-A. Thus, LNR1-2 are functionally conserved between PAPP-A and PAPP-A2 with respect to IGFBP proteolysis, but LNR3 is not, and this module appears to be critical for LNR unit function.
Identification of Two Regions Required for LNR Functionality-As a strategy to identify other regions that might influence the function of the LNR modules, we took further advantage of the fact that PAPP-A2 cleaves IGFBP-5, but not IGFBP-4, and constructed a set of chimeric PAPP-A/PAPP-A2 proteins with progressively less PAPP-A2 C-terminal sequence (Fig. 6A). All preparations of chimeras were adjusted to contain the same activity against IGFBP-5 prior to analysis of IGFBP-4 cleavage. This experiment identified a region (residues 1064 -1098) near the PAPP-A dimerization cysteine (Cys 1130 ), where a shift between no activity against IGFBP-4 and reduced activity occurred (Fig. 6B). Interestingly, PAPP-A/PAPP-A2 chimeras with no IGFBP-4 proteolytic activity were found to migrate as monomers in SDS-PAGE (Fig. 6C, lanes 1-3), whereas chimeras active against IGFBP-4 formed covalent dimers similar to  wild-type PAPP-A (Fig. 6C, lanes 4 -6). Further analysis of this region showed that a chimera with this sequence stretch of 35 residues replaced by the PAPP-2 sequence had no activity  against IGFBP-4, whereas chimeras with smaller parts of the sequence replaced displayed wild-type level or slightly reduced IGFBP-4 proteolytic activity (Fig. 7A). This indicates that substitution in this region, possibly located near a hypothetical dimerization interface, influences the arrangement of the dimer and therefore in trans interactions of the LNR modules.
Surprisingly, of the initial set of chimeras (Fig. 6, A and B), the chimera PA-P2(1504 -1547), which contains LNR3 of PAPP-A, still shows a marked reduction in cleavage of IGFBP-4. Only PA-P2(1529 -1547), in which the last 19 C-terminal residues of PAPP-A are replaced with the PAPP-A2 sequence, showed wild-type level activity against IGFBP-4. Construction and analysis of two additional chimeras revealed that the stretch of 24 residues located immediately C-terminal to LNR3 is required for cleavage of IGFBP-4 (Fig. 7B). Thus, in addition to LNR3 itself, this region is important for LNR functionality. Further analysis by single amino acid substitution of charged residues identified three residues C-terminal to LNR3, Asp 1521 , Arg 1529 , and Asp 1530 , as critical for proteolysis of IGFBP-4 by PAPP-A (Fig. 8). At these three positions, substitution with alanine caused a complete loss (Arg 1529 and Asp 1530 ) or a substantial reduction (Asp 1521 ) in proteolytic activity against IGFBP-4, but did not affect activity against IGFBP-5. Importantly, no IGFBP-4 proteolytic activity was restored when these mutants were co-expressed with the truncated PAPP-A variant, PA(1-1477) (data not shown), implying a direct role of these C-termi-  nal residues in permitting the in trans interactions of LNR3, or a direct participation in the binding of IGFBP-4 to PAPP-A. To analyze whether these mutants showed altered substrate binding, IGFBP-4/IGF-II or IGFBP-5 was immobilized to a BIAcore chip via an antibody. The PAPP-A mutants, further active siteinactivated by means of a Glu to Ala substitution at position 483 to prevent cleavage following binding to the immobilized substrate, were injected after the onset of the dissociation phase (Fig. 9). The data clearly demonstrate, in a qualitative manner, that PA(E483A) and double mutants PA(E483A/D1521A), PA(E483A/R1529A), and PA(E483A/D1530A) all bind equally well to IGFBP-5 (Fig. 9, A and B). However, a substantial loss of binding to IGFBP-4 was observed for all double mutants, but not for PA(E483A) (Fig. 9, C and D). By similar analysis, we found that PA(1-1477), the C-terminal truncated variant of PAPP-A lacking LNR3 (Fig. 1), showed binding to IGFBP-5, but not IGFBP-4 (Fig. 9, B and D).
Analysis of C-terminal PAPP-A Fragments-The expression of a series of PAPP-A variants truncated N-terminal to the five CCP modules and LNR3, respectively, was attempted. Only the construct including CCP1 (PA(1133-1547) expressed well (Fig.   10A), but this variant was unable to rescue the IGFBP-4 proteolytic activity of PA(1-1477) upon addition to the reaction mixture in molar excess (data not shown). Likewise, the addition of a molar excess of PA(1133-1547) to wildtype PAPP-A did not have any detectable effect on the proteolysis of neither IGFBP-4 nor -5 (data not shown). In BIAcore experiments, we observed no difference in interaction of PA(1133-1547) with IGFBP-4 or -5, and PA(1133-1547) did not inhibit substrate binding of PA(E483A) when injected simultaneously at equal concentrations or a 10-fold molar excess (data not shown).
The expression of four additional PAPP-A variants truncated N-terminal to the dimerization cysteine, Cys 1130 , was attempted. The points of truncation were selected taking knowledge of the disulfide pairing of PAPP-A (37) into account. The constructs encoding PA(1108 -1547) and PA(937-1547) expressed well; the latter appeared to be able to dimerize (Fig. 10B). Similarly, none of these mutants were able to rescue the IGFBP-4 proteolytic activity of PA(1-1477) upon addition to the reaction mixture (data not shown). Upon co-expression with PA(1-1477), however, a covalent heterodimeric complex appeared to be formed with PA(937-1547), but not with any other of these C-terminal fragments ( Fig. 10C and data not shown). In accordance with this observation, only PA(937-1547) was able to rescue the IGFBP-4 proteolytic activity of PA(1-1477) following co-expression (Fig. 10D), supporting our observation that dimerization is important for the in trans interaction of LNR3.

DISCUSSION
We here show that the three LNR modules of the metalloproteinase PAPP-A determine its specificity by in trans LNR interactions (Fig. 1), most likely by the formation of a functional LNR unit between LNR1-2 and LNR3. Such a functional LNR unit, formed between LNR1-2 and LNR3 from different subunits within the PAPP-A dimer, is required for proteolysis of IGFBP-4, but not IGFBP-5. We also show that formation of the Cys 1130 -Cys 1130 intersubunit disulfide bond is not required for LNR functionality (Fig. 2). The Cys 1130 -mutated PAPP-A forms a non-covalent dimer, which upon dissociation loses its activity against IGFBP-4 (Fig. 4). We therefore also conclude that although formation of the intersubunit disulfide bond is not required, formation of a functional LNR unit depends on dimerization.
We further show that PAPP-A2, which cleaves IGFBP-5, but not IGFBP-4, forms dimers (Fig. 5A), although its subunits are not disulfide-linked (16). PAPP-A, in which the LNR1-2 modules were replaced by the corresponding PAPP-A2 sequence, unexpectedly showed wildtype IGFBP-4 activity (Fig. 5D), suggesting that LNR1-2 of PAPP-A2 are able to form a functional unit with LNR3 of PAPP-A. This demonstrates that LNR1 and LNR2 are functionally conserved between PAPP-A and PAPP-A2. However, when LNR3 of PAPP-A was replaced with PAPP-A2 sequence, the activity against IGFBP-4 was almost completely abolished (Fig. 5D).
By the analysis of PAPP-A/ PAPP-A2 chimeras (Fig. 6), we further defined a region C-terminal to LNR3, which appears to be required for LNR functionality. We here identified three charged residues (Asp 1521 , Arg 1529 , and Asp 1530 ), which are critical for proteolytic activity against IGFBP-4 (Fig. 8). Furthermore, PAPP-A mutants with these residues substituted into alanine were unable to rescue the IGFBP-4 activity of the PAPP-A truncation variant PA(1-1477) in co-expression experiments (data not shown), emphasizing the importance of these residues. Although these charged residues are conserved in PAPP-A2, this region of PAPP-A could not be replaced by the corresponding PAPP-A2 sequence without a substantial loss of IGFBP-4 activity (Fig. 7B). Thus, LNR3 and the C-terminal sequence stretch may be structurally interdependent and both required for formation of the LNR unit. Additionally, by the analysis of the chimeric proteins we identified a region located between residues 1064 and 1098 (Fig. 7A), which appears to be important for PAPP-A dimerization and formation of the Cys 1130 -Cys 1130 intersubunit disulfide.
What is the role of LNR3 in discriminating between the two PAPP-A substrates? The LNR3 module and/or the sequence stretch C-terminal to LNR3 may participate directly in substrate binding to IGFBP-4. It is also possible that an IGFBP-4 substrate binding region may be formed in LNR3 or elsewhere on the trimeric LNR unit, with LNR3 and/or the following sequence stretch being required for its formation. Both of these scenarios are in agreement with the existence in PAPP-A of a classical protease substrate-binding exosite (39). Using surface plasmon resonance analysis, we were able to demonstrate a substrateproteinase interaction between IGFBP-4/IGF-II and an active site-inactivated variant of PAPP-A, PA(E483A), but not selected PAPP-A variants also mutated in the C-terminal sequence stretch or lacking LNR3 (Fig. 9, C and D). In contrast, all variants analyzed showed binding to IGFBP-5 (Fig. 9, A and  B). However, a C-terminal fragment of PAPP-A, PA(1133-1547) (Fig. 10A), was unable to bind IGFBP-4 and also unable to compete with PA(E483A) for substrate binding. This experiment emphasizes the requirement for LNR unit formation, regardless of whether LNR3 and/or the C-terminal sequence stretch participates directly or indirectly in substrate binding. We attempted the expression of a series of C-terminal PAPP-A variants, all containing LNR3 and the sequence stretch C-terminal to LNR3 (Fig. 10, A and B). Although not all variants were expressed, we found that only variant PA(937-1547), which formed a dimer with PA(1-1477) (Fig. 10C), was able to rescue the proteolytic activity of PA(1-1477) against IGFBP-4 (Fig.  10D). This experiment suggests that LNR3 cannot access LNR1-2 to form a functional LNR unit unless dimerization occurs, and it further supports, in an independent manner, that such dimerization requires the sequence stretch between residues 1064 and 1098, contained within PA(937-1547).
Compared with the Notch receptor family where three copies of the LNR module invariably are arranged consecutively, an interaction in trans of LNR1-2 and LNR3 within the PAPP-A dimer appears to compensate for the interrupted arrangement of the LNR modules in the pappalysins. Although the spatial interrelationship of the three Notch LNR modules is yet unknown, it is tempting to speculate that a similar trimeric LNR unit is formed in both protein families. Accordingly, a high degree of conservation of the short linker sequence between the Notch LNR modules of different species as well as gain-of-function mutations located in these, has suggested that the Notch LNR modules structurally depend on each other (40). This is supported by the finding that several hydrophobic residues are affected when a single LNR module was examined by NMR in the presence of a neighboring LNR module (41), indicating the existence of hydrophobic interactions between the three LNR modules. Furthermore, it has been suggested that residues from adjacent Notch LNR modules might contribute to the coordination of calcium ions (41), and calcium depletion has been found to cause activation and subunit dissociation of the Notch receptor (25). PAPP-A IGFBP-4 proteolytic activity likewise depended on calcium ions (29), indicating that calcium plays an analogous stabilizing role for the LNR modules of both proteins. This is further supported by the finding that the individual substitution of conserved acidic residues within each of the LNR modules, thought to coordinate calcium ions, causes the loss of activity against IGFBP-4 without any change in activity against IGFBP-5 (29).
Recently, the structure of ulilysin, an archaeal protein that shares sequence similarity with PAPP-A, but only encompasses the proteolytic domain, was solved (42). Ulilysin does not contain LNR modules and displays broad substrate specificity, as it cleaves several extracellular matrix proteins and all the IGFBPs, except for IGFBP-1, at several sites. This indicates that the additional modules and domains of PAPP-A function to control the specificity of the proteolytic domain. Another example of a metzincin with changed substrate specificity upon removal of domains is ADAMTS-4, in which the deletion of a C-terminal spacer domain resulted in broader substrate specificity. However, the isolated proteolytic domain of ADAMTS-4 displayed reduced proteolytic activity (43). It is tempting to speculate that association of LNR3 with LNR1-2 may function to control substrate access to the active site. Such an alternative model is speculative, but would be in accordance with models of LNR function within the Notch receptor, where the modules are thought to play a role in restraining the proteolytic cleavage of the receptors by the metzincin proteinase ADAM17/TACE (21,23,26). Further investigation is necessary to delineate details of the mechanism by which the proposed LNR unit regulates PAPP-A substrate specificity.
In conclusion, our data suggest that a specificity determining LNR unit is formed in trans by an association of LNR1-2 and LNR3 from different subunits of the PAPP-A dimer, and that it depends on PAPP-A dimerization. Identified charged residues located C-terminal to LNR3 appear to be required for the function of such a LNR unit. Our findings further suggest that the pappalysin and Notch LNR modules have a similar intermodular relationship.