Unusual Proteolytic Activation of Pro-hepatocyte Growth Factor by Plasma Kallikrein and Coagulation Factor XIa*

Hepatocyte growth factor (HGF), the ligand for the receptor tyrosine kinase c-Met, is composed of an (cid:1) -chain containing four Kringle domains (K1–K4) and a serine protease domain-like (cid:2) -chain. Receptor activation by HGF is contingent upon prior proteolytic conversion of the secreted inactive single chain form (pro-HGF) into the biologically active two chain form by a single cleavage at the Arg 494 -Val 495 bond. By screening a panel of serine proteases we identified two new HGF activators, plasma kallikrein and coagulation factor XIa (FXIa). The concentrations of kallikrein and FXIa to cleave 50% (EC 50 ) of 125 I-labeled pro-HGF during a 4-h period were 10 and 17 n M . Unlike other known activators, both FXIa and kallikrein processed pro-HGF by cleavage at two sites. Using N-terminal sequencing they were identified as the normal cleavage site Arg 494 -Val 495 and the novel site Arg 424 -His 425 located in the K4 domain of the (cid:1) -chain. The identity of this unusual second cleavage site was firmly established by use of the double mutant HGF(R424A/R494E), which was completely re-sistant to cleavage by kallikrein and FXIa. Experiments with another mutant form, HGF(Arg 494 3 Glu), indicated that cleavage at the K4 site

Hepatocyte growth factor (HGF), 1 the ligand for the tyrosine kinase receptor c-Met, was originally identified as a soluble factor with mitogenic activity for hepatocytes (1)(2)(3)(4) and "scattering" activity for epithelial cell colonies (5). The HGF/c-Met pathway is involved in many biological processes, such as embryonal development (6,7), angiogenesis (8), tissue regeneration and tumorigenesis (reviewed in Refs. 9 and 10). The biologically active HGF is a disulfide-linked heterodimeric protein of ϳ90 kDa consisting of an ␣and ␤-chain (11). The ␣-chain is composed of an Nterminal PAN module (12) and four Kringle domains (K1-K4), whereas the ␤-chain has strong homology to the protease domain of serine proteases. What separates HGF from functionally active serine proteases are the changed residues Gln 534 (instead of His) and Tyr 673 (instead of Ser), which are part of the catalytic triad His-Asp-Ser of serine proteases.
HGF is secreted into the extracellular matrix as a single chain form (pro-HGF) that lacks biological activity (13)(14)(15)(16)(17). It requires proteolytic cleavage at the Arg 494 -Val 495 peptide bond to convert it into the active ␣/␤ heterodimer. Therefore, pro-HGF converting proteases constitute an important regulatory system in the HGF/c-Met signaling pathway. Pro-HGF has strong structural similarity to macromolecular substrates of serine proteases, particularly to plasminogen, which also contains several Kringle domains. It is therefore not surprising that all pro-HGF-converting enzymes identified so far belong to this enzyme family. Urokinase-type plasminogen activator (uPA), a serine protease known for converting plasminogen into plasmin, was shown to also have pro-HGF converting activity (15,18). There is an important difference in respect to the enzyme kinetics underlying these two uPA-mediated proteolytic processes. uPA acts as a typical catalyst in activating plasminogen, whereas it converts pro-HGF in an unusual reaction that results in the formation of a stable complex of uPA and the reaction product HGF (19). Because this reaction does not follow classic enzyme kinetics, the efficiency of HGF formation will be low, because it is limited by the absolute number of uPA molecules present. It was suggested that this type of pro-HGF activation may be involved in invasive tumor growth (19). There are three additional pro-HGF-converting enzymes, factor XIIa (FXIIa) (20), HGF-activator (HGFA) (20 -23), and the recently identified membrane-bound serine protease matriptase (24). FXIIa and HGFA both circulate in blood as zymogens and have a high overall homology in their amino acid sequences. Both activators follow classic enzyme kinetics and efficiently cleave pro-HGF at enzyme:substrate ratios of Ͻ1/ 1000 (20). HGFA is the best described pro-HGF activator and was suggested to play a role in generating active HGF during tissue regeneration (25,26), morphogenesis (27,28), and tumorigenesis (29 -31). A common feature of all known pro-HGF activators is that they also undergo proteolytic activation to become active enzymes, a process that is mediated by yet another set of proteases. Thus, the HGF/c-Met pathway appears highly regulated and, depending on the particular biological process, may involve different activating enzyme and inhibitor systems.
To identify other potential pro-HGF converting enzymes, we screened a panel of serine proteases and found that plasma kallikrein and factor XIa (FXIa) efficiently activated pro-HGF. Their pro-HGF processing is unprecedented in that they cleave pro-HGF at two sites, as opposed to the single-cleavage reaction of other known activators. By use of fragment analysis, HGF mutants, and c-Met activation assays, we were able to identify the second cleavage site and to assess its functional impact. The results raise the possibility that enzymes involved in inflammation and blood coagulation may participate in HGFdependent processes, such as vascular remodeling.

EXPERIMENTAL PROCEDURES
Reagents-Pro-HGF, expressed in Chinese hamster ovary cells in the absence of serum and purified by HiTrap-Sepharose SP chromatography, was obtained from David Kahn (Genentech, Inc.). Plasma-purified human FXIIa, human FXIa, and human plasma kallikrein were purchased from Hematologic Technologies (Essex Junction, VT) and from Enzyme Research (South Bend, IN). Recombinant tissue-type plasminogen activator was obtained from Canio Refino (Genentech, Inc.). Recombinant HGFA, which was produced by using an insect cell expression system, was provided by Jennifer Stamos (Genentech, Inc.). Complement factor C1s was from Enzyme Research and thrombin, factor IXa, and factor Xa were from Hematologic Technologies. Relipidated human tissue factor and recombinant human factor VIIa were produced as described previously (32,33). The Kunitz domain inhibitors Alzheimer's amyloid ␤-protein precursor inhibitor (APPI) (34) and KALI-DY (35) were generously provided by Mark Dennis (Genentech, Inc.). Corn trypsin inhibitor was from Hematologic Technologies, and the molecular weight markers used were SeeBlue Plus2 and MultiMark standards (Invitrogen, Carlsbad, CA).

125
I-Labeling of Pro-HGF-For 125 I-labeling, a 250-l solution of IODO-GEN (1,3,4,6-tetrachloro-3␣,6␣-diphenylglycoluril) (Pierce Chemical Co., Rockford, IL) in chloroform (0.5 mg/ml) was placed into 5-ml borosilicate glass tubes. Solvent was evaporated at room temperature under a steady stream of nitrogen gas, and the dried material was stored in a dessicator until further use. Pro-HGF (700 g) in 20 mM Hepes, 150 mM NaCl, 5 mM CaCl 2 , pH 7.5, buffer (HNC buffer) was added to the dried IODO-GEN material. 125 I-Labeled sodium solution (PerkinElmer Life Sciences, Boston, MA) was added (5 Ci/g of protein), and the reaction mixture was incubated on ice for 5 min with gentle swirling. The material was then applied onto a PD-10 column (Amersham Biosciences, Uppsala, Sweden), which had been equilibrated with 20 column volumes of HNC. Fractions containing the 125 Ilabeled pro-HGF were collected and pooled. The specific activity was 1.8 Ci/g of pro-HGF.
Pro-HGF Activation Assays-0.05 mg/ml of 125 I-labeled pro-HGF in HNC buffer was incubated with increasing concentrations (0.6 to 80 nM) of kallikrein, FXIa, and FXIIa at 37°C. After 4 h, aliquots were removed and added to sample buffer (Bio-Rad Laboratories, Hercules, CA) with or without reducing agent dithiothreitol (Bio-Rad). After a brief heating, samples (approximately 10 6 cpm/lane) were loaded onto a 4 -20% gradient polyacrylamide gel (Invitrogen, Carlsbad, CA). After electrophoresis, the dried gels were exposed on x-ray films (X-OMAT AR, Eastman Kodak, Rochester, NY) for 10 -20 min. Films were developed (Kodak M35A X-OMAT Processor), scanned (Umax S-12, Umax Data Systems, Inc., Fremont, CA), and further processed with Adobe version 6.0 Photoshop software (Adobe Systems Inc., San Jose, CA). The bands corresponding to pro-HGF were cut from the dried gels, and the radioactivity was measured on a gamma counter (Iso-Data 100 Series). The data were fit to a four-parameter equation using Kaleidagraph software (Synergy Software, Reading, PA), and the disappearance of pro-HGF was quantified by determining the enzyme concentration that produced 50% substrate conversion (EC 50 ).
For inhibitor experiments 125 I-labeled pro-HGF (0.05 mg/ml) in HNC buffer was activated by kallikrein (80 nM), FXIa (80 nM), or FXIIa (40 nM) in the presence of 250 nM inhibitor. The inhibitors used were the kallikrein-specific Kunitz domain inhibitor KALI-DY (35), the FXIaspecific Kunitz domain inhibitor APPI (34), and the FXIIa-specific inhibitor corn trypsin inhibitor (36). After 4 h the reaction was stopped, and the inhibition of HGF conversion was analyzed by SDS-PAGE under reducing conditions as described.
In experiments with the HGF mutants R494E and R424A/R494E, 0.3 mg/ml of the HGF mutants or wild-type pro-HGF was incubated with kallikrein (80 nM), FXIa (80 nM), or FXIIa (40 nM) in HNC buffer for 4 h at 37°C. Reaction aliquots were then loaded onto 4 -20% gradient gels and analyzed under reducing conditions as described. Gels were stained with Simply Blue Safestain (Invitrogen).
N-terminal Amino Acid Sequencing-Samples containing HGF were reduced in 20 l of Bio-Rad Laemmli sample buffer adjusted to pH 8.3 containing 10 mM dithiothreitol at 85°C for 5 min. Alkylation was performed by the addition of 2 l of 200 mM of N-isopropyliodoacetamide in methanol at 25°C for 20 min. Proteins were separated on Bio-Rad precast gels and electroblotted onto PE-Applied Biosystems Problott membranes in a Bio-Rad Trans-Blot transfer cell using 10 mM 3-(cyclohexylamino)-1-propanesulfonic acid, pH 11.0, 10 mM thioglycolic acid, 10% methanol as the transfer buffer for 1 h at a 250-mA constant current (37). The polyvinylidene difluoride membrane was stained with 0.1% Coomassie Blue R-250 in 50% methanol for 0.5 min and destained with 10% acetic acid in 50% methanol for 2-3 min. The membrane was thoroughly washed with water and allowed to dry for storage at 0°C.
Pyroglutamate deblocking was performed with pyroglutamate aminopeptidase. The polyvinylidene difluoride band containing HGF protein was treated with 1-2 l of methanol and blocked with 200 l of 0.5% Zwittergent 3-16 (Calbiochem) in 0.1% acetic acid on a shaker for 5 min. The protein band was washed with 0.5 ml of water to remove all traces of Zwittergent. The protein was deblocked with 1 milliunit of Pyrococcus furiosus pyroglutamate aminopeptidase (Panvera Corp., Madison, WI) in 30 l of 50 mM sodium phosphate, 10 mM dithiothreitol, 1 mM EDTA, pH 7.0, at 90°C for 1 h. The protein band was dried in a Speed-Vac and directly sequenced.
Automated protein sequencing was performed on PE-Applied Biosystems Procise 494A protein sequencers. The Procise sequencers were equipped with 6-mm diameter micro-cartridges and an on-line phenylthiohydantoin analyzer. The coupling buffer was N-methylpiperidine in N-propanol and water (25:60:15) supplied by PE-Applied Biosystems or distilled in house. Twenty-minute Edman cycles were used as described (38) with two modifications. A high pressure (3.0 p.s.i.) delivery of coupling buffer was carried out for 20 s prior to all phenylisothiocyanate deliveries. Peaks were integrated with Chromperfect (Justice Innovation software), and sequence interpretation was performed on a DEC Alpha computer (39).
Site-directed Mutagenesis, Expression, and Purification of HGF Mutants-The HGF(R494E) mutant was described by Lokker et al. (17). Using HGF(R494E) as a template, Arg 424 was altered to an Ala by site-directed mutagenesis to give HGF(R424A/R494E) using the Muta-Gene mutagenesis kit (Bio-Rad Laboratories, Hercules, CA) according to manufacturer's protocol. The mutation was verified by DNA sequencing.
Recombinant proteins were produced using Chinese hamster ovary cells in large scale transient transfection processes. Cells were grown in 1-liter spinner flasks in F-12/Dulbecco's modified Eagle's medium supplemented with Ultra-Low IgG serum (Invitrogen) and Primatone HS (Sigma). The transfection process involved formation of the DNA-cationic lipid complex for 15 min in 300 ml of basal media followed by transfer of this complex to 700 ml of cell suspension (seeded at a density of 1.2 ϫ 10 6 cells/ml). The ratio of DNA to cationic lipid as well as the cell density were optimized to achieve maximal expression of recombinant protein. After 7-12 days, the cell culture fluid was harvested and adjusted to 0.3 M NaCl. The HGF mutants were purified by loading the cell culture fluid on a 5-ml HiTrap-Sepharose SP chromatography column (Amersham Biosciences, Uppsala, Sweden) pre-equilibrated with 20 mM Hepes, pH 7.5, 0.3 M NaCl. The column was washed with the same buffer, and proteins were eluted with a gradient of 0.3 to 1.2 M NaCl in 20 mM Hepes, pH 7.5. The HGF-containing fractions were pooled and concentrated, and the HGF concentration was determined by quantitative amino acid analysis.

Proteolytic Cleavage of Pro-HGF by Plasma Kallikrein and
Coagulation Factor XIa-By use of purified 125 I-labeled pro-HGF we examined the pro-HGF-converting activity of a panel of serine proteases. No pro-HGF-processing activity was observed for complement factor C1s and the tissue factor-factor VIIa complex nor for proteases previously examined by Shimomura et al. (20), i.e. factor Xa, thrombin, factor IXa, and tissuetype plasminogen activator. However, we found that plasma kallikrein (referred to as kallikrein throughout) and coagulation factor XIa (FXIa) efficiently processed pro-HGF during a 4-h reaction period (Fig. 1). Both enzymes cleaved pro-HGF at the normal cleavage site Arg 494 -Val 495 thereby generating the ␣/␤-chain heterodimer. The N termini of the HGF ␤-chains produced by each enzyme were identical ( 495 VVNGI-PTRTN 504 ), with the differences in mass of the two ␤-chains (ϳ36 and ϳ39 kDa) being attributed to differences in the content of attached carbohydrates (40). However, unlike the previously identified pro-HGF converting enzyme FXIIa (20), kallikrein and FXIa produced a second ␣-chain fragment (␣2), whose apparent molecular mass of ϳ54 kDa was about 10 kDa lower than the normal ␣ chain (Fig. 1). Therefore, a ϳ10-kDa fragment was released either from the N or the C terminus of the ␣ chain. N-terminal sequencing showed that ␣2 and ␣ chain N termini were identical (data not shown), suggesting that the ϳ10-kDa fragment arose by a cleavage at the C-terminal portion of the ␣-chain.
The pro-HGF converting activity of kallikrein and FXIa, quantified by measuring the disappearance of the 125 I-labeled HGF single chain, was similar to FXIIa. The concentrations of kallikrein, FXIa, and FXIIa to convert 50% (EC 50 ) of pro-HGF during a 4-h incubation period were 10, 17, and 10 nM, respectively (Fig. 1d). The FXIa concentrations used throughout this study were of the naturally occurring homodimer (M r ϳ 143,000) (41). Therefore, the EC 50 value based on monomeric FXIa concentration would be 34 nM.
Our findings were in apparent contradiction to the observed lack of pro-HGF converting activity of kallikrein and FXIa reported by Shimomura et al. (20). In an attempt to understand these different results, we used inhibitors specific for kallikrein, FXIa, and FXIIa to address the possibility of whether the pro-HGF converting activity in our assays could be due to contaminating proteases. The Kunitz domain inhibitors APPI (34) and KALI-DY (35) are potent and specific inhibitors of FXIa and kallikrein, respectively, whereas corn trypsin inhibitor is specific for FXIIa. Fig. 2 shows that KALI-DY only inhibited pro-HGF activation by plasma kallikrein but not by FXIa and FXIIa. Conversely, APPI specifically interfered with FXIa-mediated pro-HGF activation, and corn trypsin inhibitor only inhibited FXIIa-mediated but not kallikrein-or FXIamediated pro-HGF activation (Fig.2). Moreover, to rule out contaminating activity by the potent plasma-derived pro-HGF activator, HGFA, we carried out an assay with recombinant HGFA. We found that neither KALI-DY nor APPI inhibited HGFA-dependent pro-HGF activation at concentrations (250 nM) that completely blocked pro-HGF activation by kallikrein and FXIa (data not shown). This experiment thus excluded HGFA as a possible contaminant. Furthermore, we employed kallikrein and FXIa preparations from two different commercial sources and used various buffer systems, only to find consistent and reproducible pro-HGF converting activities by these two enzymes. Therefore, we concluded that kallikrein and FXIa have the intrinsic ability to process pro-HGF.

Identification of the Alternative Kallikrein/FXIa Cleavage Site in the Kringle Domain 4 (K4)-
The results suggested that the unusual cleavage of the HGF ␣-chain by kallikrein and FXIa resulted in the release of a peptide of ϳ10 kDa upon reduction. Analyzing digested pro-HGF by reducing SDS-PAGE, a fragment of this size was identified and subjected to N-terminal sequencing. The sequence, 425 HIFWEPDASK 434 , was consistent with the release of a 70-residue C-terminal ␣-chain fragment (His 425 -Arg 494 ) of ϳ10 kDa as observed by SDS-PAGE. Therefore, cleavage probably occurred at the Arg 424 -His 425 peptide bond in the K4 domain of the ␣-chain. Four testable predictions ensued from this assumption. First, because the putative cleavage site resided within a loop structure in K4 that is flanked by disulfide bonds, the cleaved HGF should migrate as a single band under non-reducing conditions. Second, the side chain of the P1 residue (Arg 424 ) should be surface-exposed, because it is required to occupy the specificity pocket of kallikrein and FXIa. Third, digestion of the primary cleavage site mutant HGF(R494E) (17) with kallikrein and FXIa should produce a "long" ␤-chain having His 425 as its N-terminal residue. Fourth, modification of the presumed K4 cleavage site should abolish proteolysis.
We attempted to systematically address these predictions. First, kallikrein-and FXIa-digested pro-HGF migrated as a single band of about 90 kDa on SDS gels under non-reducing conditions (Fig. 3, inset). This agreed with the hypothesis that the cleaved ␣-chain is held together by the disulfide bonds in K4. Second, a molecular model of K4 based on the crystal structure of K1 (42) was constructed (Fig. 3). In this model, the Arg 424 side chain pointed away from the K4 backbone consistent with its accessibility for kallikrein and FXIa active sites. Third, the primary cleavage site mutant HGF(R494E) was processed by kallikrein and FXIa to produce the expected fragments, the ␣2-chain and the long ␤-chain (Fig. 4), which had the N-terminal sequence 425 HIFWEPDA 432 . The reaction was specific in that the two Kunitz domain inhibitors KALI-DY and APPI inhibited the generation of these HGF fragments by kallikrein and FXIa, respectively (data not shown). Moreover, consistent with the mutation at the Arg 494 residue, neither enzyme nor FXIIa were able to recognize this site for proteolysis anymore as indicated by the absence of the normal ␤ chain bands (ϳ35 and ϳ38 kDa) on the gels (Fig. 4). The results also demonstrated that cleavage at Arg 424 -His 425 was not contingent upon cleavage at Arg 494 -Val 495 . Fourth, a double mutant HGF(R424A/R494E) was constructed in which both P1 residues, Arg 424 and Arg 494 , were changed. We found that kallikrein and FXIa were unable to process this mutant form anymore, even at high enzyme concentrations (Fig. 5).

Phosphorylation of c-Met Receptor by Alternatively Cleaved
HGF-The activity of HGF generated by kallikrein (HGF Kallikrein ) and FXIa (HGF FXIa ) was assessed by measuring c-Met phosphorylation of A549 epithelial cells. Fig. 6 depicts the results obtained with HGF Kallikrein and HGF FXIa in which only a small portion of HGF was processed at the alternative K4 cleavage site, as exemplified by the HGF material shown in Fig. 3 (inset). HGF Kallikrein and HGF FXIa behaved like the reference HGF material generated by pro-HGF digestion with FXIIa and showed a concentration-dependent increase in c-Met phosphorylation activity. In most experiments the optimal c-Met phosphorylation activity was at concentrations 50 -100 ng/ml HGF. Additional control experiments showed that the activators kallikrein and FXIa themselves had no activity in the assays (data not shown). To specifically address the question of whether cleavage at the K4 site (Arg 424 -His 425 ) affected c-Met phosphorylation, high concentrations of FXIa were used to produce HGF in which the K4 site cleavage approached near completion (Fig. 7a). This HGF form showed c-Met phosphorylation activity that was indistinguishable from the reference material (FXIIa-digested pro-HGF) (Fig. 7b) suggesting that cleavage at Arg 424 -His 425 is without functional consequences in respect to c-Met activation. DISCUSSION This study demonstrates that the two plasma-derived serine proteases, kallikrein and FXIa, are able to convert pro-HGF into its biologically active form. By use of specific serine protease inhibitors we were able to ascertain that activation of pro-HGF was specifically mediated by kallikrein and FXIa and not by potential contaminants, such as the plasma-derived FXIIa or HGFA. FXI circulates in blood at a concentration of about 30 nM (41) and is known as an essential component of blood coagulation reactions (41,43,44). The activation of FXI zymogen by thrombin as well as FXIa enzymatic activity occurs optimally on the surface of activated platelets, which express specific FXI and FXIa binding sites (41). In addition, platelets intracellularly store FXI (41) as well as HGF (4), which are released upon platelet activation. Therefore, surface-bound FXIa would be ideally situated to activate platelet-derived HGF, which may play a role in vascular remodeling after tissue injury.
Unlike FXIa, plasma kallikrein (plasma concentration of zymogen, 0.41-0.56 M (45)) is not essential for blood coagulation (41,43,(45)(46)(47) but has other diverse functions such as the generation of the vasoactive bradykinin peptide from kininogens (45,46,48,49). Furthermore, recent evidence suggests that kallikrein is a component of multiprotein complexes assembled on cellular surfaces (45, 48 -50). Its activity toward other protein components of these complexes, such as high molecular weight kininogen and pro-uPA, produces anti-and pro-angiogenic effects (45,48,51,52). Therefore, the generation of pro-angiogenic HGF by kallikrein may represent yet another aspect of the involvement of kallikrein in angiogenic and inflammatory processes.
HGF also plays an important role in embryogenesis. Targeted disruption of the mouse HGF gene caused embryonal lethality associated with abnormal development of liver and placenta (6,7). If kallikrein or FXIa were critically involved in the embryonal pro-HGF activation process, then prekallikrein and FXI deficiencies should also result in developmental defects. However, prekallikrein (Fletcher factor) deficiency in humans is not associated with any apparent pathology (46), and homozygote FXI null mice develop normally (53), leading us to conclude that neither kallikrein nor FXIa are important pro-HGF activators during embryogenesis. Therefore, the proteolytic conversion of pro-HGF in this biological process has to be mediated by other HGFactivating enzymes, possibly by HGFA, which was suggested to play a role in morphogenesis (27,28).
An unexpected finding was that kallikrein and FXIa, unlike the other activators FXIIa (Fig. 1) and HGFA (20), processed pro-HGF at a second cleavage site. Proteolysis at this site was independent of prior cleavage at the normal, kinetically preferred Arg 494 -Val 495 site. N-terminal sequence analysis of the unique proteolytic fragments, i.e. the ␣2-chainand the ␣-chain-derived ϳ10-kDa fragment, suggested that the cleavage site was Arg 424 -His 425 in the K4 domain. The presence of Arg 424 at the P1 position agreed with the fact that the known macromolecular substrates of kallikrein and FXIa either have an Arg or Lys (in high molecular weight kininogen) as a P1 residue. Further support came from a molecular model of K4, indicating that the side chain of the Arg 424 was indeed solventexposed and therefore potentially accessible for interaction with the active center of kallikrein and FXIa, respectively. The presumptive availability of the Arg 424 -His 425 peptide bond for proteolytic cleavage was also supported by experiments with another serine protease, trypsin, which was capable of generating an HGF fragment with His 425 as the N-terminal residue (40,54). Finally, the use of two HGF mutants allowed us to firmly establish the Arg 424 -His 425 peptide bond as the cleavage site for kallikrein and FXIa. Digestion of the mutant HGF(R494E) by either enzyme produced the expected long ␤-chain (His 425 -Ser 728 ), whereas the substitution of Arg 424 with Ala in the double mutant HGF(R424A/R494E) completely abolished proteolytic cleavage.
There were no apparent functional consequences of K4 domain cleavage with respect to c-Met receptor phosphorylation. Even when conditions were chosen to complete cleavage at both sites in a quantitative manner, receptor activation was indistinguishable from HGF that was produced by FXIIa. We attributed this result to the presence of three disulfide bonds in K4, which, after proteolysis at the Arg 424 -His 425 bond, held the cleaved ␣-chain together. It is possible that cleavage may not have appreciably altered the conformational integrity of the K4 domain. In support of this view, Pediaditakis et al. (55) demonstrated that HGF, in which the receptor binding domain K1 was cleaved by factor Xa, remained competent to elicit mitogenic responses.
Although the kallikrein-and FXIa-specific proteolysis of the HGF-K4 domain did not affect receptor phosphorylation, this unusual cleavage might elicit other effects. An intriguing possibility is that reductive processes could result in the release of an HGF fragment (residues 1-424; NK3/4) within the context of pro-HGF processing by kallikrein and FXIa. The release of such a fragment would require, at minimum, the reduction of two disulfide bonds in the K4 domain (e.g. Cys 412 -Cys 452 and Cys 391 -Cys 469 ). Interestingly, it was recently shown that a disulfide reductase from tumor cells contributed to the release of the angiostatin fragment from plasminogen, which underwent concomitant proteolytic cleavage in the K5 domain (56)(57)(58). Therefore, the possible role of disulfide reductase in the release of NK3/4 from HGF deserves further attention. Another question is whether such an HGF-derived fragment would exert biological activities similar to NK4 (59), which was shown to antagonize HGF function.