Mapping of the discontinuous kininogen binding site of prekallikrein. A distal binding segment is located in the heavy chain domain A4.

Prekallikrein, the precursor to the serine proteinase kallikrein, circulates in plasma in an equimolar complex with H-kininogen. The binding to H-kininogen is mediated by the kallikrein heavy chain consisting of four "apple" domains, A1-A4, which attaches to H-kininogen with high specificity and affinity (KD = 83 nM). At least two distinct portions of the kallikrein heavy chain form this H-kininogen binding site: a proximal segment located in the NH2-terminal fragment of the heavy chain encompassing A1, and distal segment(s) located in COOH-terminal fragment spanning domains A2-A4. The proximal binding segment has been located to amino acid positions 56-86 of A1. To precisely map the distal binding segment, we have identified monoclonal antibodies directed to the COOH-terminal fragment which interfere with the H-kininogen-prekallikrein complex formation. Monoclonal antibody 13G11 binds to recombinant apple domain A4 but not to domain A3 of the prekallikrein heavy chain. Deletion mutagenesis of domain A4 narrowed down the target epitope of 13G11 to the center portion of domain A4, positions 284-331. Direct binding studies of H-kininogen to various domain A4 constructs revealed that the distal H-kininogen binding portion is located on a segment of 48 residues, which overlaps the 13G11 epitope. Hence the tight interaction of H-kininogen and prekallikrein is mediated by at least two separate sequence segments located in domains A1 and A4, respectively, of the prekallikrein heavy chain. The isolated distal binding segment significantly prolongs the partial thromboplastin time of reconstituted Williams plasma thus stressing the critical role of the prekallikrein-H-kininogen complex formation in the initiation of the endogenous blood coagulation cascade.

Plasma prekallikrein, the 86 -88-kDa precursor to the serine protease, ␣-kallikrein, participates in the inflammatory response (1), in prourokinase-dependent fibrinolysis (2), and in blood coagulation on artificial surfaces (3). Factor XII a converts prekallikrein to the active proteinase, ␣-kallikrein, on cellular and artificial surfaces via a single cleavage at position 371. This cleavage generates a heavy chain of 50 kDa and a light chain of 35 kDa (4). The light chain harbors the catalytic domain, whereas the heavy chain binds to the non-enzymatic cofactor, H-kininogen. 1 Sequence analysis has revealed that the heavy chain consists of four repetitive units, "apple" domains A1-A4, characterized by a unique disulfide loop pattern (5). Each apple domain comprises 90 -91 amino acid residues; the only other known protein containing apple domains is the heavy chain of factor XI (6). Autocatalytic cleavage of ␣-kallikrein at Lys 140 -Ala 141 generates ␤-kallikrein, with two heavy chain portions, Gly 1 -Lys 140 ("fragment N") and Ala 141 -Arg 371 ("fragment C"), connected by two disulfide bridges (5,7,8).
Prekallikrein circulates in plasma as a heterodimeric complex with the non-enzymatic cofactor, H-kininogen (K a ϭ 8.3 ϫ 10 7 M Ϫ1 ) (9). The prekallikrein binding site on H-kininogen has been mapped to a "continuous" sequence segment of 25 amino acid residues located at the extreme carboxyl-terminal part of the H-kininogen light chain (10 -12). The corresponding Hkininogen binding site on prekallikrein is located in the heavy chain region (8,13). At least two separate sequence segments, namely fragment N and fragment C of ␤-kallikrein, contribute to this binding site: an amino-terminal ("proximal") segment has been been mapped to domain A1, positions Phe 56 -Gly 86 of fragment N (14,15), whereas the precise location of the carboxyl-terminal ("distal") binding segment of fragment C remains unclear. Binding studies with synthetic peptides have indicated that at least one portion of the A4 domain might be involved in the binding (15).
The present study was undertaken to define the distal H-kininogen binding site(s) of the prekallikrein heavy chain. Employing an antibody-directed strategy, we have identified a critical segment of 48 residues of domain A4 that contributes to the binding of H-kininogen. Our results suggest that at least two separate regions of the kallikrein heavy chain, domains A1 and A4, contribute to the H-kininogen binding site, thereby mediating the interaction of the prehormone and its processing enzyme in the fluid phase and on the cell membrane.

EXPERIMENTAL PROCEDURES
Sources of Proteins and Peptides-Prekallikrein was isolated from human plasma following established protocols (16). Factor XII was from Enzyme Research Laboratories, South Bend, IN. H-kininogen was purified from human plasma (17) with minor modifications described elsewhere (18). The generation and characterization of monoclonal antibodies PKH1, PKH4, PKH6 (16), PKH19 (14), and 13G11 (8,19) has been detailed previously. Proteins were biotinylated as described (16) except that the final buffer change was made to 150 mM NaCl, 100 mM NaH 2 PO 4 , 10 mM Na 2 HPO 4 , pH 7.4, by centrifuging three times at 2,000 ϫ g, 4°C using a microconcentrator (Amicon) with a 10-kDa cut-off membrane.
SDS-Polyacrylamide Gel Electrophoresis (PAGE) and Western Blotting-Proteins and fusion proteins were resolved by PAGE in the presence of 0.1% (w/v) sodium dodecyl sulfate at 30 mA for 30 min (20). The low molecular weight electrophoresis calibration kit was from Pharmacia Biotech Inc. The resolved proteins were visualized by the silver staining technique (21) or transferred to nitrocellulose at 100 mA for 30 min by the semi-dry technique (22). The membranes were blocked with 50 mM KH 2 PO 4 , 0.2 M NaCl, pH 7.4, containing 5% (w/v) dry milk powder and 0.05% (w/v) Tween 20 (buffer A). Immunoprinting of the transferred proteins was done according to Towbin et al. (23) with modifications specified elsewhere (11). Typically the first antibody was diluted 1:1,000 in buffer A. Bound antibody was detected by a horseradish peroxidase-coupled secondary antibody against rabbit or mouse immunoglobulin, followed by the chemiluminescence detection according to the manufacturer's instructions (Amersham Corp.).
Limited Proteolysis of Prekallikrein-To produce ␣and ␤-kallikrein, 100 g of human prekallikrein and 4 g of human factor XII a (molar ratio 1:50) were dissolved in 250 l of 10 mM sodium phosphate, 150 mM NaCl, pH 7.4, and incubated for 3 days at 37°C. Aliquots of the reaction mixture were removed after 2 h and 72 h. Analysis of the reaction products by reducing SDS-PAGE revealed that prekallikrein had been completely activated to ␣-kallikrein after 2 h of incubation, and was converted to ␤-kallikrein after an additional incubation period of 70 h.
Cyanogen Bromide Cleavage of Plasma Prekallikrein-Ten g of prekallikrein was dissolved in 1.0 ml of 1 M BrCN in 70% (v/v) formic acid, and the reaction mixture was incubated overnight in the dark at 25°C (24). The cleavage products were evaporated to dryness by lyophilization, redissolved in sample buffer, and analyzed by SDS-PAGE. For amino-terminal sequence analysis, the proteins were electrotransferred to polyvinylidene difluoride sheets and microsequenced by Edman degradation using a 470A pulsed-liquid phase sequencer (Applied Biosystems).
Indirect and Competitive ELISAs-The indirect ELISA was performed as described (14). A 10 -20 g/ml stock solution of kallikrein fragments in 100 mM sodium acetate, 100 mM NaCl, pH 5.5, was used for coating. The probe was 5 g/ml H-kininogen in the same buffer, and monoclonal antibody HKH14 (4 g/ml) against the H-kininogen heavy chain (25) served as the reporter antibody. To test for the interference of unlabeled antibodies with the H-kininogen-prekallikrein complex formation, the same ELISA was employed except that biotinylated H-kininogen (0.5 g/ml) served as the reporter. For the mutual displacement of monoclonal antibodies from the prekallikrein heavy chain, the ELISA in the competitive mode was applied (16) using biotinylated antibody 13G11 (1 g/ml) as the probe.
Expression of Apple-Tissue Plasminogen Activator (tPA) Fusion Proteins in Baby Hamster Kidney Cells-The A3 (Gly 179 -Glu 271 ) and A4 (Pro 272 -Ser 262 ) domains of prekallikrein with flanking 5Ј BglII and 3Ј XhoI sites were amplified by the polymerase chain reaction (PCR) with Taq polymerase. The PCR products were cloned into the TA cloning vector pCRII (Invitrogen), and the sequence was verified by dideoxy sequencing. The DNA segments corresponding to the A3 and A4 domains were excised by digestion BglII and XhoI and cloned into the corresponding sites in the restriction site-modified and active sitemodified tPA expression vector ZpL7(S478A) (26,27). The constructs encode fusion proteins, which contain the signal sequence and propeptide of tPA, followed by the prekallikrein apple domain A3 or A4, kringle 1 and kringle 2 domains of tPA, and the active site-mutated tPA light chain. The expression plasmids were transfected into BHK cells as described previously (28). The fusion proteins were expressed in serumfree medium (Opti-MEM, Life Technologies, Inc.), and purified with a monoclonal antibody to tPA as described (27).
Partial Thromboplastin Time-To study the effect of kallikrein heavy chain fragments on the kaolin-activated partial thromboplastin time, we used the original assay (30) with modifications described previously (14). Typically 50 l of kininogen-deficient human plasma (Williams trait) reconstituted with 0.5 g of purified H-kininogen in the presence or absence of 200 g of recombinantly expressed kallikrein A4 domains were used, and the assay was performed on an Amelung coagulometer.

RESULTS
Our strategy to localize the distal H-kininogen binding segment of the prekallikrein heavy chain involved: (i) the selection of monoclonal antibodies that bind to fragment C and interfere with the H-kininogen-prekallikrein complex formation, (ii) the precise mapping of the target epitope(s) of the selected antibodies, (iii) the recombinant expression of prekallikrein heavy chain fragments encompassing the relevant epitope, (iv) the analysis of the binding capacity of the recombinant constructs for H-kininogen, and (v) the identification of the critical binding segment and its application to a biological test.
Monoclonal Antibodies Interfering with H-kininogen-Prekallikrein Complex Formation-In previous studies we have described a panel of 20 monoclonal antibodies to prekallikrein (14,16,19); of these, 11 antibodies were found to be directed to the kallikrein heavy chain. Ten antibodies have been mapped to four distinct epitope classes exposed on the prekallikrein heavy chain, arbitrarily named A-D. To test whether these antibodies interfere with H-kininogen-prekallikrein complex formation, we established a competitive ELISA where prekallikrein was immobilized on the titer plate and biotinylated H-kininogen served as the probe. Increasing concentrations of monoclonal antibodies PKH1 (epitope class A), PKH4 (B), PKH6 (C), PKH19 (D), and 13G11 (unclassified) were applied and tested for their capacity to prevent H-kininogen binding to prekallikrein (Fig. 1A). Antibody PKH6 almost completely blocked H-kininogen binding with an apparent IC 50 of 3.9 nM. Antibodies PKH1 and 13G11 prevented binding of biotinylated H-kininogen with IC 50 values of 42 and 59 nM, respectively, whereas anti-peptide antibody PKH19 was less effective with an IC 50 of 760 nM, and PKH4 did not interfere at all, IC 50 Ͼ 1 M.
Epitope Classification of Antibody 13G11-The observation that antibody 13G11 efficiently inhibited H-kininogen binding to prekallikrein prompted us to identify its epitope class. The competitive ELISA was modified such that titer plate-bound prekallikrein was probed by the biotinylated antibody 13G11 in the presence of increasing concentrations of antibodies PKH1 (epitope class A), PKH4 (B), PKH6 (C), and PKH19 (D) (Fig.  1B). As a control unlabeled antibody 13G11 was applied and found to displace its biotinylated homologue almost completely. Similarly antibody PKH1 inhibited binding of biotinylated 13G11 in a concentration-dependent manner, whereas antibodies PKH4, PKH6, and PKH19 were without effect. Hence antibodies PKH1 and 13G11 are members of the same epitope class, A, suggesting that they might recognize overlapping, if not identical target sequences in the prekallikrein heavy chain. Other members of class A epitope-specific antibodies are PKH2, PKH8, and PKH9 (16), indicating that their target sequence(s) might represent a major immunogenic epitope of the prekallikrein heavy chain.
Identification of Antibodies Directed to Fragment C-To identify antibodies that interfere with complex formation and bind to fragment C containing the distal H-kininogen binding site, we employed immunoprinting. A mixture of ␣-kallikrein and ␤-kallikrein was separated by SDS-PAGE under reducing conditions, electroblotted on nitrocellulose, and probed by the various antibodies (Fig. 2). For control a polyclonal antibody was used (lane 1), which detected multiple kallikrein fragments. All of the monoclonal antibodies bound to the kallikrein heavy chain, although with varying intensities. Antibodies PKH1 (lane 2), PKH4 (lane 3), and 13G11 (lane 6) recognized fragment C, and PKH19 (lane 5) bound to fragment N, whereas PKH6 (lane 4) decorated the heavy chain but failed to detect a heavy chain breakdown product. Hence epitope classes A and B are located on fragment C, whereas epitope class D is positioned on fragment N; the relative position of epitope class C remains unknown.
Epitope Mapping Using Cyanogen Bromide Cleavage Products of Prekallikrein-Our results indicate that epitope class A antibodies 13G11 and PKH1 might be useful probes for the localization of the distal H-kininogen binding segment because they (i) interfere efficiently with the H-kininogen-prekallikrein complex formation and (ii) bind to fragment C of the kallikrein heavy chain. To characterize their target sequence(s) more precisely, prekallikrein was chemically cleaved by cyanogen bromide. The resultant fragments were reduced and separated by SDS-PAGE; four identical blots on nitrocellulose were prepared, and the transferred proteins were probed by the antiprekallikrein antibodies described above (Fig. 3). Control antibodies PKH6 (lane 3) and PKH19 (lane 4) were applied, which either failed to recognize a cleavage product (PKH6) or produced a complex staining pattern (PKH19). Antibody 13G11 (lane 2) recognized two CNBr cleavage products of 19 and 23 kDa, respectively. Antibody PKH1 (lane 1) bound to the same fragments, although with differential intensity. These results suggest that its target sequences of PKH1 and 13G11 might be overlapping, although not necessarily identical. Edman degradation of the 19-kDa fragment, which was readily detected by both antibodies, revealed a single amino-terminal sequence of Asn-Ile-Phe-Gln-His-Leu-Ala. This heptapeptide maps to positions 185-191 located in the center portion of domain A3. Considering the size of the 19-kDa fragment and that of a single apple domain (ϳ10 kDa corresponding to 90 -91 residues), we conclude that the relevant epitopes for 13G11 and PKH1 must be located in domains A3 and/or A4 of the prekallikrein heavy chain.
Epitope Mapping Using Truncated Domain A4 Constructs-For precise epitope mapping, we chose an expression system in E. coli, where domain A4 and its truncated versions were fused to the maltose binding protein, MBP. Initial experiments indicated that MBP-A4 (Thr 257 -Arg 371 ) but not MBP-A3 (Glu 179 -Thr 268 ) or MBP alone was recognized by 13G11 and PKH1 (data not shown) indicating that post-translational modifications such as glycosylation and/or disulfide bridging are not critical to antibody binding. Apple domain constructs were engineered such that domain A4 was either truncated at the carboxyl terminus (MBP-A4.1, Thr 257 -Arg 331 ; MBP-A4.2, Thr 257 -Thr 313 ), at the amino terminus (MBP-A4.3, Gly 284 -Arg 371 ; MBP-A4.4, Thr 313 -Arg 371 ), or at both termini (MBP-A4.5, Gly 284 -Arg 331 ). The results of the antibody binding experiments are summarized in Fig. 5. Antibody 13G11 bound to the intact domain A4 and to the carboxyl-terminally shortened form, A4.1, but not to a further truncated version, A4.2. The same antibody bound to the amino-terminally truncated version, A4.3 but not to the shortened form, A4.4. Overlap analysis of the various domain A4 fragments indicate that the center portion of domain A4, Gly 284 -Arg 331 , holds the relevant target sequence of the antibody. Indeed the corresponding construct, MBP-A4.5, was recognized by antibody 13G11, whereas a shortened version thereof (Gly 284 -Thr 313 ) escaped from antibody detection (data not shown). Results with antibody PKH1 were indistinguishable from those obtained with 13G11 (data not shown). Therefore we conclude that 13G11 and PKH1 are directed to the center portion of domain A4 which covers residues 284 to 331. Our attempts failed to further narrow down the target sequences of 13G11 and PKH1 by a synthetic peptide library covering the relevant sequence segments (data not shown).
Mapping    distal H-kininogen binding segment by a combinatorial peptide library that covered the relevant portions of the binding region (data not shown).
Functional Interference of Domain A4 Fragments with Partial Thromboplastin Time-Given that domain A4 and its fragments bind to H-kininogen without conveying an enzymatic function to the resulting complex, one might expect that domain A4 constructs interfere with the physiological interaction of H-kininogen with prekallikrein in surface-dependent blood coagulation. To test this hypothesis, we employed the kaolinactivated partial thromboplastin time (aPTT) assay using a kininogen-deficient plasma (Williams trait) (Fig. 7). This plasma was reconstituted with a limited amount of purified H-kininogen, and the competition of recombinant fusion proteins MBP-A4 and its fragments with plasma kallikrein for the limited amount of H-kininogen was measured by a prolongation of the aPTT time. Recombinant MBP alone had no effect on the aPTT of reconstituted Williams plasma (column 9). MBP-A4 efficiently competed with plasma prekallikrein and caused a significant prolongation of the aPTT (column 3), as did MBP-A4.1, and less efficiently MBP-A4.2, MBP-A4.3, and MBP-A4.5, but not MBP-A4.4 (columns [4][5][6][7][8]. These findings support our notion that A4.5 is the minimum fragment which contains the distal H-kininogen binding segment and is able to block the binding of functionally intact prekallikrein molecules under the conditions of the aPTT assay. Binding Properties of Heavy Chain Variants-Our previous work (14,15) and this study suggest that a "discontinuous" binding site for H-kininogen is formed by at least two distinct portions located on domains A1 and A4 of the kallikrein heavy chain. To address if a proteolytic cleavage between domains A1 and A4 would influence the relative binding affinity for Hkininogen, we tested various forms of kallikrein including prekallikrein, where the heavy chain is contiguous with its light chain, ␣-kallikrein, which contains an intact heavy chain (Gly 1 -Arg 371 ) connected to the light chain (Ile 372 -Ala 619 ) by an interchain disulfide bridge, and ␤-kallikrein, where the heavy chain is cleaved into fragment N (Gly 1 -Lys 140 ) and fragment C (Ala 141 -Arg 371 ) held together by two disulfide loops (5). An indirect ELISA (Fig. 8) was set up where equivalent amounts of the various kallikrein forms were coated on the titer plate, and the binding of increasing concentrations of biotinylated H-kininogen was followed by the biotin-avidin-peroxidase system (note that biotinylation does not impair H-kininogen bind-ing to prekallikrein) (16). Prekallikrein bound to H-kininogen with the highest apparent affinity (half-maximum binding at 0.6 nM corresponding to 0.05 g/ml), and ␣-kallikrein binding (2.3 nM ϭ 0.2 g/ml) differed only by a factor of 4. In contrast, the apparent affinity of ␤-kallikrein for H-kininogen was considerably lower (60 nM ϭ 5 g/ml), i.e. it differed by a factor of 26 -100 from that of ␣-kallikrein and prekallikrein, respectively. Hence, the cleavage of the heavy chain region connecting critical domains A1 and A4 results in a dramatic loss of the apparent affinity of kallikrein for H-kininogen. DISCUSSION Gene multiplication and exon shuffling are the major driving forces in the evolution of "mosaic" plasma proteins. One such complex protein is prekallikrein, which most likely arose by fusion of a particular motif, the "apple" domain, to a catalytic domain, and by quadruplication of the exon set specifying the apple domain. In this way a multidomain protein was generated which subserves functions such as kinin release, initiation of the endogenous blood coagulation cascade, stimulation of the profibrinolytic pathway (1), and stimulation of neutrophils (31). Sequence analysis has revealed that prekallikrein is 58% identical on the protein sequence level with another plasma protein, factor XI (32). Prekallikrein and factor XI share their overall structure, except that the latter is a homodimer with the two disulfide-bridged heavy chains connected in the A4 domains (5,27,28). Factor XI has developed multiple interactions sites, which have been analyzed in great detail by Walsh and co-workers: (i) domain A1 binds H-kininogen and thrombin (33,34); (ii) domain A2 exposes a substrate binding site for factor IX (35); (iii) domain A3 serves as the cell binding site (36); and (iv) domain A4 provides a factor XIIa recognition site (37). Some of these interactions are not found in prekallikrein, e.g. the recognition of factor IX; others are yet ill-defined, e.g. the interaction with factor XII (8); and still others are modified, e.g. the binding of H-kininogen.
The detail analysis of the interaction of prekallikrein with H-kininogen reveals that at least two domains of the heavy chain, namely A1 (14,15) and domain A4 (this work), are involved in H-kininogen binding. Our present findings suggest a "discontinuous" binding site formed by two segments, which are almost 200 residues apart in the primary sequence. We cannot entirely exclude the possibility that the "bridging" domains, A2 and A3, also contribute directly or indirectly to H-kininogen binding although the failure of recombinantly expressed A3 to bind H-kininogen (data not shown) does not support such a view. Thus prekallikrein and factor XI, although similar in structure, seem to have evolved distinct H-kininogen interaction site(s). This notion is corroborated (i) by the observation that their corresponding binding sites on the H-kininogen light chain are overlapping although not identical (10); (ii) and by the finding that a synthetic peptide spanning positions 317 to 350 of domain A4 of factor XI is without effect on the complex formation between H-kininogen and factor XI (38). It is tempting to speculate that, in the case of factor XI, two A1 domains contributed by the two heavy chains in the dimeric molecule might suffice to secure binding to H-kininogen, whereas in the monomeric prekallikrein, two apple domains in the same chain must interact to tether this cofactor with sufficient affinity.
In this work we have used recombinant prekallikrein domains expressed in eukaroytic and in prokaryotic systems. Similar binding characteristics were observed for the constructs made in kidney cells or in E. coli, implying that the expressed sequence segments are able to spontaneously acquire a "correct" folding. This finding is unexpected because reduction and denaturation of prekallikrein result in a complete loss of its H-kininogen binding properties (16). The cysteine-rich core of apple domain A4 seems to be of critical importance to the formation of an "active" binding site. These notions are consistent with the finding that a synthetic peptide, which spans the segment Leu 262 -Gly 295 of prekallikrein domains A3 and A4 and overlaps the Gly 284 -Arg 331 segment by 12 residues, shows binding affinity for H-kininogen (15). It should be noted that in the latter study the H-kininogen was present on the surface of a microtiter plate, where its conformation may have been altered.
The precise functional role of the circulating complex of prekallikrein and H-kininogen is still obscure. Recent studies have demonstrated that these two proteins are present on the surface of cardiovascular cells such as neutrophils (39,40) and endothelial cells (41). Prekallikrein and H-kininogen, possibly in concert with other proteins such as L-kininogen and factor XI, are likely to subserve the local kinin release on endothelial cells and circulating blood cells (18,42,43). The potent biological effects exerted by kinins and their extremely short half-life in plasma (Յ15 s) demand for efficient control mechanisms to initiate, sustain, and terminate the cell-associated kinin generation. A powerful mechanism controlling kallikrein activity is provided by the endogenous inhibitors such as C1-inhibitor and ␣ 2 -macroglobulin (44). Our finding that the proteolytic cleavage of the kallikrein heavy chain drastically reduces its ability to bind H-kininogen might provide yet another regulatory mechanism. If H-kininogen is the major anchoring protein for (pre)kallikrein on cell surfaces, the loss of the intrinsic binding affinity for kininogen would result in a rapid release of the processed kallikrein from the cell surface, thereby terminating local kinin release. This notion is supported by the finding that the procoagulant activity of kallikrein is critically dependent on the integrity of its heavy chain (45). Likewise the blockage of the H-kininogen binding site in prekallikrein by synthetic peptides or monoclonal antibodies drastically reduced its procoagulant activity (14,15). We envisage that the assembly and disassembly of the prekallikrein-H-kininogen complex on the cellular H-kininogen-binding protein is a critical factor in the regulation of circumscribed kinin release.