Structural insight into distinct mechanismsof protease inhibition by antibodiesYan Wu*, Charles Eigenbrot*†, Wei-Ching Liang*, Scott Stawicki*, Steven Shia†, Bin Fan†, Rajkumar Ganesan†,Michael T. Lipari†, and Daniel Kirchhofer†‡

Departments of *Antibody Engineering and †Protein Engineering, Genentech, Inc., 1 DNA Way, South San Francisco, CA 94080

Edited by John Kuriyan, University of California, Berkeley, CA, and approved November 1, 2007 (received for review August 30, 2007)

To better understand how the relatively flat antigen-combining sitesof antibodies interact with the concave shaped substrate-bindingclefts of proteases, we determined the structures of two antibodiesin complex with the trypsin-like hepatocyte growth-factor activator(HGFA). The two inhibitory antibodies, Ab58 and Ab75, were gener-ated from a human Fab phage display library with synthetic diversityin the three complementarity determining regions (H1, H2, and H3) ofthe heavy chain, mimicking the natural diversity of the human Igrepertoire. Biochemical studies and the structures of the Fab58:HGFA(3.5-Å resolution) and the Fab75:HGFA (2.2-Å resolution) complexesrevealed that Ab58 obstructed substrate access to the active site,whereas Ab75 allosterically inhibited substrate hydrolysis. In bothcases, the antibodies interacted with the same protruding element(99-loop), which forms part of the substrate-binding cleft. Ab58inserted its H1 and H2 loops in the cleft to occupy important substrateinteraction sites (S3 and S2). In contrast, Ab75 bound at the backsideof the cleft to a region corresponding to thrombin exosite II, which isknown to interact with allosteric effector molecules. In agreementwith the structural analysis, binding assays with active site inhibitorsand enzymatic assays showed that Ab58 is a competitive inhibitor,and Ab75 is a partial competitive inhibitor. These results providestructural insight into antibody-mediated protease inhibition. Theysuggest that unlike canonical inhibitors, antibodies may preferen-tially target protruding loops at the rim of the substrate-binding cleftto interfere with the catalytic machinery of proteases without requir-ing long insertion loops.

catalysis � enzyme � phage display

Proteases hydrolyze peptide bonds of their substrate(s) result-ing in substrate degradation (e.g., extracellular matrix deg-

radation) or conversion of substrate into the biologically activeform (e.g., hepatocyte growth factor). Proteases participate in avast array of biological processes. For instance, the chymotryp-sin-type serine proteases (Clan PA, family S1), which constitutethe largest and biologically most diverse protease family, par-ticipate in processes such as food digestion, immune reactions,tissue regeneration, blood coagulation, and fibrinolysis. Manydiseases are associated with deregulated protease activity and,therefore, the therapeutic potential for targeting proteases issignificant. Many specific as well as relatively nonspecific pro-tease inhibitors are currently used in disease managementranging from cardiovascular disease to cancer (1).

Because specificity is a highly desired property of a therapeuticprotease inhibitor, antibodies are very promising as therapeuticagents, particularly when targeting the �270 extracellular pro-teases in the human genome (2). However, antibodies generallyhave a planar or concave shaped antigen-binding site (paratope),which seems ill suited to interact with the concave shapedsubstrate-binding cleft of proteases. In contrast, many naturallyoccurring protease inhibitors present a convex shaped feature,like an exposed loop, to the protease cleft to interfere withcatalysis in a substrate-like manner (the standard mechanism)(3). Similarly, the heavy chain antibodies from camels (HCAbs),which lack a light chain, seem ideally adapted for interacting withthe concave cleft. They have a relatively long and protruding

complementarity determining region (CDR) H3 loop (H3) thatinserts into the substrate-binding cleft of lysozyme and othernonproteolytic enzymes, blocking catalysis (4–6). Most conven-tional anti-lysozyme antibodies do not bind into the cleft and arenonblocking. Intriguingly, Farady et al. (7) recently described anantibody that inhibits the chymotrypsin-type serine proteasematriptase by inserting a very long H3 loop (19 residues) into thecleft. Although the lengths of H3 loops are highly variable, theaverage length, 9 residues for mouse and 12 residues for humansequences (8), might be insufficient for active site insertion andcanonical inhibition.

Conceptually, antibodies could inhibit protease activity in adirect manner by binding at or near the active site to blocksubstrate access or indirectly by binding to regions that areallosterically linked to the active site region. Several antibodiesthat block protease activity have been described, but relativelyfew were studied in detail (7, 9–13). Mutagenesis studies showedthat the binding sites of anti-factor VIIa, anti-thrombin, anti-matriptase, and anti-urokinase antibodies are located at or nearthe active site of the enzymes (7, 11–13). However, a detailedunderstanding of the underlying molecular inhibition mecha-nisms has been hampered by the lack of structural informationabout the antibody-protease interface. To our knowledge, thereis no deposited structure of a protease (EC 3.4; hydrolases actingon peptide bonds) in complex with a function-blocking antibody.

These studies raised the question of whether inhibition ofcatalysis by conventional antibodies requires insertion of a longH3 loop into the substrate-binding cleft. Alternatively, couldantibodies inhibit catalysis through other mechanisms? In thisstudy, we attempted to answer these questions by using hepato-cyte growth-factor activator (HGFA) as a model system, becausestructures of this serine protease (family S1) as well as sensitivesubstrate assays were available (14, 15). The serum-derived34-kDa active HGFA consists of a protease domain disulfidelinked to the 35-residue light-chain (16). It efficiently cleavesprohepatocyte growth factor (pro-HGF) into the functionallycompetent two-chain hepatocyte growth factor (HGF) leadingto activation of the HGF/Met signaling pathway during tissueregeneration and in cancer growth (17–19). The N-terminalKunitz domain (KD1) of the endogenous HGFA inhibitor-1(HAI-1) (15, 20) binds into the HGFA active site in a substrate-like manner (14).

Author contributions: Y.W. and D.K. designed research; C.E., W.-C.L., S. Stawicki, S. Shia,B.F., R.G., M.T.L., and D.K. performed research; Y.W., C.E., W.-C.L., and D.K. analyzed data;and Y.W., C.E., and D.K. wrote the paper.

Conflict of interest statement: The authors are employed by Genentech, Inc.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.pdb.org (PDB ID codes 2R0K and 2R0L).

‡To whom correspondence should be addressed. E-mail: dak@gene.com.

This article contains supporting information online at www.pnas.org/cgi/content/full/0708251104/DC1.

© 2007 by The National Academy of Sciences of the USA

19784–19789 � PNAS � December 11, 2007 � vol. 104 � no. 50 www.pnas.org�cgi�doi�10.1073�pnas.0708251104

To generate anti-HGFA antibodies, we used an antibodyphage library with synthetic diversity in heavy chain CDRsmimicking natural Ig diversity (21). Two phage antibodies, Ab58

and Ab75, inhibited both macromolecular and synthetic peptidecleavage and were studied in detail. Competition binding studies,enzyme kinetics, and the structures of the two Fab:HGFAcomplexes provided extensive insight into the molecular basis oftheir inhibitory mechanisms. The results suggested that antibod-ies are able to efficiently perturb the catalytic machinery by usingdistinct mechanisms, without the requirement for uncommonlylong H3 loops.

ResultsIdentification of Anti-HGFA Phage Antibodies. To identify anti-HGFA antibodies, we screened human synthetic Fab phagelibraries built on a single and well defined human framework(modified from Trastuzumab) with amino acid diversity atselected positions in the H1, H2, and H3 loops and lengthdiversity in H3 (7–19 residues). Fourteen unique HGFA-bindingclones were reformatted to full-length IgGs for further charac-terization. Two antibodies, Ab58 and Ab75, displayed distinctinhibitory properties (see below). Ab58 and Ab75 had dissimilarheavy chain CDR sequences, whereas the light chain CDRsequences were identical, as expected [supporting information(SI) Fig. 6]. Both antibodies bound to HGFA in a specificmanner as indicated by ELISA experiments with structurallyrelated proteases (SI Fig. 7), and competition binding assayssuggested that their binding sites on HGFA were overlapping(data not shown). Surface plasmon resonance (SPR) experi-ments showed that Ab58 bound HGFA with high affinity (KD0.6 � 0.1 nM), whereas that of Ab75 was 24-fold lower (KD14.6 � 1.5 nM). Both antibodies inhibited HGFA-mediatedprocessing of pro-HGF with potencies consistent with theirrespective binding affinities (Fig. 1 A and B).

Enzyme Kinetics. The antibodies were studied in enzymatic assaysby use of the synthetic para-nitroanilide substrate SpectrozymeFVIIa (pNA substrate). HGFA enzymatic activity was onlypartially inhibited by Ab75 (Fig. 1C). Eadie–Hofstee plotsshowed that the inhibition mechanism was competitive in thatAb75 increased the Km

app but not V maxapp values (Fig. 1D). In

accordance with partial inhibition, the slopes (�Kmapp) ap-

proached a finite limit at high [Ab75]. The obtained Ki value of29.2 � 4.7 nM was in reasonable agreement with the bindingaffinity determined by SPR. Ab58 was also a competitiveinhibitor (data not shown), but, unlike Ab75, it completelyinhibited substrate hydrolysis (Fig. 1C Inset). Fitting the inhibi-tion data to the equation for tight-binding competitive inhibitionsystems (22), a Ki

app value of 0.23 � 0.03 nM was obtained, whichwas consistent with the KD � 0.6 nM from SPR studies. Theresults showed that Ab58 is a pure competitive inhibitor, whereas

Fig. 1. Inhibition of HGFA catalytic activity by Ab58 and Ab75. (A and B)Cleavage of 125I-pro-HGF by HGFA in the presence of 3-fold serial dilutions ofAb58 (A) and Ab75 (B). The cleavage products HGF �- and �-chain wereanalyzed by SDS/PAGE (reducing conditions) and x-ray film exposure. C,control (no antibody). The last lane in B contained 125I-pro-HGF only. (C) Partialinhibition of Spectrozyme FVIIa hydrolysis (expressed as HGFA fractionalactivity vi/vo) by Ab75 and complete inhibition by Ab58 (Inset). (D) Eadie–Hofstee plot of HGFA inhibition by Ab75 (2–0.008 �M in 3-fold dilution steps;filled circles, ‘‘no antibody’’ control) shows competitive inhibition. Vmax � 0.99�M pNA/min and Km � 0.24 mM for control; V max

app � 1.00 �M pNA/min andKm

app � 1.21 mM for 2 �M Ab75 (triangles).

Fig. 2. Effects of active-site inhibitors on antibody binding to HGFA. (A) Surface plasmon resonance measurements of HGFA binding to immobilized antibodyafter coinjection of HGFA with benzamidine or Hfac-221 or KD1. (B) Competition binding ELISA measuring binding of HGFA to biotinylated KD1 in the presenceof increasing antibody concentrations.

Wu et al. PNAS � December 11, 2007 � vol. 104 � no. 50 � 19785

BIO

CHEM

ISTR

Y

Ab75 is a partial competitive inhibitor (see SI Fig. 8 for a schemefor equilibria of partial competitive inhibition systems).

Effects of HGFA Active-Site Occupancy on Antibody Binding. We usedSPR to measure antibody binding to HGFA in the presence ofreversible inhibitors that occupy increasingly larger portions ofthe active-site region. Benzamidine, which only fills the S1pocket of trypsin-like serine proteases, did not interfere withbinding of either antibody to HGFA (Fig. 2A). The largerinhibitor Hfac-221, which occupies the S1 and S2 sites, stronglyinterfered with Ab58 binding, but only marginally interferedwith Ab75 binding (Fig. 2 A). The Kunitz inhibitor KD1, whichinteracts with the extended active-site region (14), interferedwith Ab75 binding and even more strongly with Ab58 binding(Fig. 2 A). Conversely, in a competition binding ELISA, increas-ing concentrations of Ab58 strongly inhibited KD1 binding toHGFA, whereas Ab75 showed incomplete inhibition of KD1binding (Fig. 2B) akin to the partial inhibition in synthetic pNAsubstrate assays (Fig. 1C).

Structure of the Fab58:HGFA and Fab75:HGFA Complexes. TheFab58:HGFA structure (SI Fig. 9 and SI Table 1) revealed thatFab58 H2 and H1 loops bound in the HGFA substrate-bindingcleft, occluding substrate subsites S2 and S3, respectively, but notS1 (Fig. 3 A and B). The S1 pocket was not obstructed and wasavailable for benzamidine binding, whereas the H2 loop causeda steric clash with Hfac-221 occupying the S2 subsite (Fig. 4). TheH3 and L3 loops contacted the distal edge of the HGFA 99-loop,which formed a part of one side of the substrate-binding cleft(Fig. 5A). Thus, Phe-97 from the protruding HGFA 99-loopinserts into the cleft between the variable domain of the heavychain (VH) and the light chain (VL), whereas the H2 and H1loops occupied the enzyme cleft [a table comparing the hereinused chymotrypsinogen numbering with the continuous HGFAnumbering can be found in Shia et al. (14)]. In contrast, camelantibodies exemplified by cAb-Lys-3 insert the tip of the long H3loop into the substrate-binding cleft of lysozyme (23) (Fig. 5B).For cAb-Lys-3, this occludes access to the Asp and Glu residuesof the active site, whereas its H1 and H2 loops bind outside thecleft. There are also cAb-Lys-3 contacts involving the extendedH3 region characteristic of camelid VH domains, which serves asa highly abbreviated form of the VL domain of a conventionalantibody.

Fab75 uses all CDR loops except L2 to bind to a relatively flatepitope adjacent to, but not including, the substrate-binding cleft(Fig. 3 A and C). The epitope is centered on Leu-93 of theprotruding 99-loop, which like Phe-97 in the Fab58 complex,inserts into the cleft between the VH and VL domains (SI Fig. 10and SI Table 2). Because the Ab75 epitope is less convex thanthat of Ab58, it naturally presents itself to a greater proportionof Ab75’s antigen-combining surface. This arrangement explainsthe slightly greater total surface area in the Ab75 epitope(�1,020 Å2 vs. �890 Å2 for Ab58). Superposition ofFab75:HGFA and KD1:HGFA showed that there was neither anoverlap between the KD1- and Fab75-binding sites, nor a stericconflict between the bound KD1 and Fab75 (Fig. 3 A and C).This observation is consistent with an allosteric mechanism bywhich Ab75 inhibits enzymatic activity and KD1 binding. The 60-and 99-loops are sandwiched between the Fab75 epitope andKD1 contact site and thus well positioned to mediate theallosteric coupling (Fig. 3C). Furthermore, the Fab75 epitopehas significant overlap with thrombin exosite II (Fig. 5C), anelectropositive region that interacts with thrombin regulators,

Fig. 3. Structures of the Fab58:HGFA and Fab75:HGFA complexes. (A) TheFab58 contact region (green, 4-Å cutoff) and the Fab75 contact region (or-ange, 4-Å cutoff). The two residues, Phe-97 and Asn-98, common to bothregions are magenta. The KD1-binding region from the KD1:HGFA structure(14) is delineated by black dots and corresponds to the red surface area in Band C. The positions of the substrate binding subsites S1, S2, and S3 (white)were inferred by analogy to subsites of related proteases and from thecomplex of HGFA with KD1 (14). Note that Fab58 does not obstruct the S1pocket. (B) The Fab58:HGFA complex. Fab58 VH and VL are teal and light cyan,respectively. The KD1- and Fab75-binding regions are red and orange, respec-tively. (C) The Fab75:HGFA complex superimposed with the KD1:HGFA com-plex. The orientation of HGFA is exactly as in B. The Fab75 VH and VL are darkand light blue, respectively, and KD1 is gray. There is no overlap between theFab75 (orange) and the KD1 (red) binding sites.

Fig. 4. Steric clash of Fab58 H2 loop and the small molecule antagonistHfac-221. Hfac-221 (yellow) modeled according to a similar inhibitor in Oliveroet al. (36), occupies the S1 subsite of HGFA (beige) as it forms H bonds withAsp-189. CDR loops H2 and H1 from Fab58 (blue) are shown, and H2 has anextensive steric overlap with Hfac-221. The circled lower part of Hfac-221approximates benzamidine, a smaller inhibitor that does not affect Ab58binding.

19786 � www.pnas.org�cgi�doi�10.1073�pnas.0708251104 Wu et al.

such as heparin. Whereas electrostatic interactions mediated bya cluster of Arg and Lys residues are critical for the heparin–thrombin interaction, they are less important for Fab75 bindingto HGFA, because there are only three Arg or Lys residueswithin the Fab75 epitope, and they are located at its periphery(Arg-241, Lys-87, and Arg-61) (Fig. 5C).

For Fab75:HGFA, there is a significant intermolecular crys-tal-packing artifact involving the HGFA substrate-binding re-gion, where Arg-390� (from the light chain of a neighboringHGFA) presents its guanidinium moiety to the Asp-189 sidechain almost exactly the same way as the P1 Arg from KD1 doesin the KD1:HGFA complex. This interaction presents elementsakin to both P1 and P2 of a substrate mimic and may thereby limita structural indication of the allosteric influence of Fab75 on theenzyme active site (SI Fig. 11). Solution-binding studies based onsize exclusion chromatography only identified Fab75:HGFAcomplexes of 1:1 molar composition and did not show anyhigher-order Fab75:HGFA complexes, strongly suggesting thatthe active-site interactions seen in the structure are crystalpacking artifacts (SI Fig. 12).

DiscussionThis study provides structural insight into the underlying mo-lecular mechanisms by which antibodies inhibit protease cata-lytic activity. The anti-HGFA antibodies described herein, Ab58and Ab75, were generated from an antibody phage library withsynthetic diversity in the heavy chain CDR loops reflecting thenatural Ig diversity (21). Both antibodies displayed competitiveinhibition kinetics yet, as revealed by the Fab:HGFA crystalstructures, used quite distinct inhibitory mechanisms. Thus,whereas Ab75 was a partial competitive inhibitor and alloster-ically influenced the active-site environment, Ab58 directlycompeted with substrate binding by steric hindrance.

Most Ab58 contacts were with residues of well exposed loops(60- and 99-loops) that form one side of the canyon-like HGFAsubstrate-binding cleft, with additional contacts on the oppositeside (170-, 220-, and 140-loops). Inhibition of catalysis wascaused by insertion of the H2 and H1 loops into the cleft, therebyoccupying the S2 and S3 sites of HGFA, which are critical forinteraction with P2 and P3 residues of substrates. The predicted

steric clashes with the Kunitz domain inhibitor KD1 and with thesmall molecule antagonist Hfac-221 agreed with results fromcompetition binding studies and with the competitive inhibitionof substrate hydrolysis in enzymatic assays. It is intriguing thatthe binding region identified by alanine scanning mutagenesis ofthe anti-matriptase scFv E2 (7) has close resemblance to thestructural epitope of Ab58, including binding residues on the99-loop. However, unlike Ab58, the scFv E2 inhibits matriptaseby insertion of a very long H3 loop (19 residues) into thesubstrate-binding cleft. In contrast, the H3 loop of Ab58 does notinsert into the cleft at all but rather partners with the L3 loop toembrace Phe-97 where it protrudes at the periphery of theHGFA cleft. Thus, compared with the interaction of convexcamelid HCAbs with the enzyme cleft (4), the interaction ofAb58 with HGFA is inverted, as Ab58 uses the concave VH/VLcleft to interact with a convex structural feature of the enzyme,i.e., the protruding Phe-97 of the 99-loop, with additionalcontacts in the cleft and along the 60- and 99-loops made by theH2 and H1 loops (Fig. 5 A and B).

In contrast to Ab58, Ab75 bound to the ‘‘backside’’ of the 60-and 99-loops, away from the substrate-binding cleft. The struc-tural epitopes of Ab58 and Ab75 are mutually exclusive exceptfor two residues in common, Phe-97 and Asn-98, explaining whythe antibodies inhibit each other from binding to HGFA.Intriguingly, the location of the Fab75 epitope closely corre-sponds to thrombin exosite II, a positively charged region thatinteracts with allosteric effector molecules, such as heparin andthrombomodulin (24). Thrombin exosite II and the relatedheparin-binding exosites on coagulation factors IX (25) and X(26) comprise a cluster of functionally important arginine andlysine residues (25–29), which is not the case for the Ab75epitope.

Nevertheless, the close correspondence of the exosites andAb75 epitope suggests that the underlying allosteric mechanismsmay be related. Biochemical studies showed that exosite bindingof an anti-thrombin antibody, heparin, or prothrombin frag-ment-2 induced conformational changes at the active-site region,which is mediated at least in part by the 60- and 99-loops (12, 30,31). Although they are not part of the ‘‘activation domain,’’ thesetwo surface loops appear to have some conformational f lexibility

Fig. 5. Comparison of Ab58 with camelid antibody and Ab75 epitope with thrombin exosite II. (A) Ab58 CDR loops H2 (magenta) and H1 (yellow) are in theHGFA (beige) substrate-binding cleft and approach the active site (His-57, red) but do not occupy the S1 subsite while occluding the S2 and S3 subsites,respectively. The Ab58 H3 (red) and L3 loop (orange) surround Phe-97 from HGFA. (B) The camelid antibody cAb-Lys-3 (Protein Data Bank entry 1JTT) (green)is oriented according to the VH of Ab58 in A. The long HCAb-Lys-3 H3 loop (red) occupies the substrate cleft of lysozyme (mesh in gray) and contacts the catalyticAsp-52, whereas CDRs H1 and H2 contact the periphery. (C) Correspondence of Fab75 epitope (orange) on HGFA (beige) with thrombin exosite II. Exosite II ofthrombin (Protein Data Bank entry 2C8Y) is indicated by blue spheres for Arg and Lys residues found to be important in the interactions with oligosaccharidesand prothrombin Kringle-2 observed in different thrombin crystal structures (27–29). Note that compared with thrombin, the basic residues are underrepre-sented in the Fab75-binding region of HGFA (HGFA residues Arg-61, Lys-87, and Arg-241 are indicated).

Wu et al. PNAS � December 11, 2007 � vol. 104 � no. 50 � 19787

BIO

CHEM

ISTR

Y

(31, 32), which may predispose them to conformational changesupon exosite binding by regulators. Because the Fab75 epitopeincludes residues from both the 60- and 99-loops, we proposethat these two loops, which also form one side of the substrate-binding cleft, are critical in the allosteric inhibition of substrateturnover. Such a proposition agrees with the experimental datafrom inhibition assays. At the distal side in respect to theAb75-binding site, these loops shape the S2 and S3 sites impor-tant for interactions with P2 and P3 residues of synthetic andmacromolecular substrates. It would also explain why the KD1canonical inhibitor and Ab75, whose nonoverlapping bindingsites are on opposite sides of these loops, inhibit each other frombinding to HGFA.

The Ab75-induced allosteric influences were more detrimen-tal to cleavage of the macromolecular (pro-HGF) than thesynthetic (pNA) substrate. A similar trend was observed with anallosteric peptide inhibitor of factor VIIa, which binds near thefactor VIIa 60-loop (33). A possible explanation is that syntheticsubstrates (occupying only S1–S3 subsites) are inadequate sur-rogates of natural protein substrates, which require precisealignment of the scissile peptide and interactions beyond theS1–S3 sites. In addition, interference by Ab75 with possiblepro-HGF exosite interactions and/or with pro-HGF-inducedchanges in the substrate-binding cleft cannot be ruled out.

The exact structural cause of Fab75’s allosteric effect onHGFA enzymatic activity is not readily apparent from ouranalysis. We observe only a small difference between the HGFAactive-site regions complexed with Fab75 or KD1 (�1 Å cen-tered at residue His60a). A larger structural effect of Fab75binding might have been observed if the intermolecular packingcontacts at the S1 and S2 subsites were not present. Suchinteractions tend to enforce a canonical conformation of theactive site, as has been observed for x-ray structures of allostericinhibitors with thrombin (27).

Our results reveal two different ways by which conventionalantibodies can inhibit catalysis. Unlike the camelid HCAbs orendogenous canonical inhibitors, the inhibition mechanisms ofAb75 and Ab58 do not involve insertion of a single long loop intothe substrate-binding cleft. Ab75 and Ab58 both act by recog-nizing parts of one edge of the cleft, but one applies an allostericeffect while binding outside the cleft, and the other places its H2and H1 loops in the cleft. Because neither the 60- nor the 99-loopof HGFA is extraordinary among S1-family proteases, it is likelythat analogous features across a wide range of protease targetscan be exploited by conventional antibodies acting as inhibitors.Such a mechanism also offers the benefit of not exposing theantibody to possible proteolytic degradation. We note that theantibody 26–2F directed against a nonproteolytic enzyme, an-giogenin (pancreatic RNase superfamily), similarly embracesloops at one edge of the active site with its concave shapedparatope, thereby blocking substrate access (34). Therefore,interactions between the convex edges of substrate-binding cleftsand the flat or concave paratopes of conventional antibodiesoffer a robust general mechanism for inhibition of a broad arrayof enzymes.

Materials and MethodsAntibody Phage Display. The VH libraries were kindly providedby V. Lee and G. Fuh (Genentech, Inc.). Synthetic antibodylibraries displayed bivalent Fab fragments on M13 phage, andthe diversity was generated by use of oligo-directed mutagen-esis in three CDRs of the heavy chain. The details of the VHlibraries were described in ref. 21. Nunc 96-well MaxiSorpimmunoplates (Nunc) were coated overnight at 4°C withHGFA (10 �g/ml) and blocked for 1 h with PT buffer (PBS,0.05% Tween 20) supplemented with 1% BSA (PT buffer withBSA making PTB buffer). The antibody phage libraries wereadded and incubated overnight. The plates were washed with

PT buffer and bound phage were eluted with 50 mM HCl and500 mM NaCl for 30 min and neutralized with an equal volumeof 1 M Tris base. Recovered phage were amplified in Esche-richia coli XL-1 blue cells. During subsequent selection rounds,incubation of antibody phage with the antigen-coated plateswas reduced to 2–3 h, and the stringency of plate washing wasgradually increased (35).

Enzymatic Assays. Pro-HGF activation assays with active site-titrated HGFA (Val-373-Ser-655) were carried out essentiallyas described (15) by using serial dilutions of antibody incu-bated with 0.8 nM HGFA and 25 �g/ml of 125I-pro-HGF. Forchromogenic substrate assays with Spectrozyme FVIIa (Meth-anesulfonyl-D-cyclohexylalanyl-butyl-arginine-paranitroani-lide) (American Diagnostica), 3.9 nM HGFA was incubatedfor 40 min in 96-well plates with increasing concentrations ofantibodies in HBSA buffer [20 mM Hepes (pH 7.5), 150 mMNaCl, 5 mM CaCl2, 0.5 mg/ml BSA]. After addition ofSpectrozyme FVIIa (0.2 mM � Km), the linear rates of theincrease in absorbance at 405 nm were measured on a kineticmicroplate reader. To obtain the Ki

app value for Ab58, the datawere fitted to the equation for tight binding competitiveinhibition systems (22, 36). Enzyme kinetic measurements forAb75 were carried out with 2.4 nM HGFA incubated withAb75 (2–0.008 �M in 3-fold dilutions) in HBSA buffer for 40min. Various concentrations of Spectrozyme FVIIa wereadded, and the linear rates of absorbance increase at 405 nmwere measured. Eadie–Hofstee plots of the data obtained (vversus v/[S]) were indicative of competitive inhibition. Apply-ing the equations for partial competitive inhibition systems(37) to the herein used steady-state conditions (substituting Kmfor Ks), the values for � and Ki were obtained from 1/� slopevs. 1/[Ab75] replots of 1/v versus 1/[S] plots (37).

Binding Experiments with Anti-HGFA Antibodies. Anti-HGFA anti-body Fabs were reformatted into human IgG1 by cloning the VLand VH regions of individual clones into LPG3 and LPG4vectors, respectively (35). The full-length antibodies were tran-siently expressed in Chinese hamster ovary cells and purified ona protein A column.

SPR measurements were carried out on a BIAcore-3000instrument (GE Health Care). Rabbit anti-human IgG werechemically immobilized on CM5 biosensor chips, and theanti-HGFA antibodies were captured to give �250 responseunits (RU). For kinetics measurements, twofold serial dilu-tions of HGFA (250–0.9 nM) were injected in PT buffer at25°C with a f low rate of 30 �l/min. Association rates (kon) anddissociation rates (koff) were obtained by using a simpleone-to-one Langmuir binding model (BIAcore EvaluationSoftware version 3.2), and the equilibrium dissociation con-stants (KD) were calculated (koff/kon). Identical conditionswere used for experiments with active-site inhibitors. KD1 wasexpressed and purified as described (15). The Hfac-221 is areversible HGFA inhibitor obtained from Alan Olivero (Ge-nentech). In HGFA enzymatic assays, Ki values of 0.59 � 0.05�M for Hfac-221 and 4.6 � 0.2 mM for benzamidine weredetermined. Inhibitors were preincubated with a fixed con-centration of HGFA (2–2.5 nM for Ab58 and 5–10 nM forAb75) and injected over sensorchips with immobilized anti-bodies to measure inhibition of HGFA binding.

In a competition binding ELISA, 96-well Nunc MaxiSorpplates coated with HGFA (1 �g/ml) were incubated with in-creasing concentrations of anti-HGFA antibodies in PTB bufferfor 2 h followed by addition of 1 nM biotinylated KD1, which wasdetected by adding streptavidin-HRP conjugates.

Crystallography. Fab58 and Fab75 were expressed in E. coli andpurified by using protein G-Sepharose. Fab:HGFA complexes

19788 � www.pnas.org�cgi�doi�10.1073�pnas.0708251104 Wu et al.

were purified by size exclusion chromatography. Crystals ofFab75:HGFA [crystallized by using 20% PEG 10000, 0.1 MHepes (pH 7.5)] and Fab58:HGFA [crystallized by using 1 MNa�/K� tartrate, 0.2 M Li2SO4, 0.1 M CHES (pH 9.5)] weresupplemented with 20% glycerol before cryopreservation. Datawere collected at 100 K at beamline 5.0.1 (Fab75:HGFA) or 5.0.2(Fab58:HGFA) at Advanced Light Source (ALS), reduced byusing HKL2000 (38) with elements of CCP4 (39), solved bymolecular replacement [PHASER (40)], and refined by using

CNX (Accelrys) and Refmac5 (41). Data reduction and modelrefinement statistics appear in SI Table 3.

We thank M. Franklin, Y. Chen, L. Rouge, and I. Bosanac. The ALS isoperated by Lawrence Berkeley National Laboratory, on behalf of theU.S. Department of Energy, Office of Basic Energy Sciences. TheBerkeley Center for Structural Biology receives support from the Officeof Biological and Environmental Research (Department of Energy) andthe National Institute of General Medical Sciences (National Institutesof Health).

1. Turk B (2006) Nat Rev Drug Discov 5:785–799.2. Overall CM, Blobel CP (2007) Nat Rev Mol Cell Biol 8:245–257.3. Laskowski M, Jr, Kato I (1980) Annu Rev Biochem 49:593–626.4. De Genst E, Silence K, Decanniere K, Conrath K, Loris R, Kinne J,

Muyldermans S, Wyns L (2006) Proc Natl Acad Sci USA 103:4586–4591.5. Desmyter A, Transue TR, Ghahroudi MA, Thi MH, Poortmans F, Hamers R,

Muyldermans S, Wyns L (1996) Nat Struct Biol 3:803–811.6. Lauwereys M, Arbabi Ghahroudi M, Desmyter A, Kinne J, Holzer W, De

Genst E, Wyns L, Muyldermans S (1998) EMBO J 17:3512–3520.7. Farady CJ, Sun J, Darragh MR, Miller SM, Craik CS (2007) J Mol Biol

369:1041–1051.8. Wu TT, Johnson G, Kabat EA (1993) Proteins 16:1–7.9. Xuan JA, Schneider D, Toy P, Lin R, Newton A, Zhu Y, Finster S, Vogel D,

Mintzer B, Dinter H, et al. (2006) Cancer Res 66:3611–3619.10. Shimomura T, Miyazawa K, Komiyama Y, Hiraoka H, Naka D, Morimoto Y,

Kitamura N (1995) Eur J Biochem 229:257–261.11. Dickinson CD, Shobe J, Ruf W (1998) J Mol Biol 277:959–971.12. Colwell NS, Blinder MA, Tsiang M, Gibbs CS, Bock PE, Tollefsen DM (1998)

Biochemistry 37:15057–15065.13. Petersen HH, Hansen M, Schousboe SL, Andreasen PA (2001) Eur J Biochem

268:4430–4439.14. Shia S, Stamos J, Kirchhofer D, Fan B, Wu J, Corpuz RT, Santell L, Lazarus

RA, Eigenbrot C (2005) J Mol Biol 346:1335–1349.15. Kirchhofer D, Peek M, Li W, Stamos J, Eigenbrot C, Kadkhodayan S, Elliott

JM, Corpuz RT, Lazarus RA, Moran P (2003) J Biol Chem 278:36341–36349.16. Miyazawa K, Shimomura T, Kitamura A, Kondo J, Morimoto Y, Kitamura N

(1993) J Biol Chem 268:10024–10028.17. Itoh H, Naganuma S, Takeda N, Miyata S, Uchinokura S, Fukushima T,

Uchiyama S, Tanaka H, Nagaike K, Shimomura T, et al. (2004) Gastroenter-ology 127:1423–1435.

18. Tjin EP, Derksen PW, Kataoka H, Spaargaren M, Pals ST (2004) Blood104:2172–2175.

19. Uchinokura S, Miyata S, Fukushima T, Itoh H, Nakano S, Wakisaka S, KataokaH (2006) Int J Cancer 118:583–592.

20. Shimomura T, Denda K, Kitamura A, Kawaguchi T, Kito M, Kondo J, KagayaS, Qin L, Takata H, Miyazawa K, Kitamura N (1997) J Biol Chem 272:6370–6376.

21. Lee CV, Liang WC, Dennis MS, Eigenbrot C, Sidhu SS, Fuh G (2004) J MolBiol 340:1073–1093.

22. Morrison JF (1969) Biochim Biophys Acta 185:269–286.23. Decanniere K, Transue TR, Desmyter A, Maes D, Muyldermans S, Wyns L

(2001) J Mol Biol 313:473–478.24. Bode W (2005) J Thromb Haemost 3:2379–2388.25. Yang L, Manithody C, Rezaie AR (2002) J Biol Chem 277:50756–50760.26. Rezaie AR (2000) J Biol Chem 275:3320–3327.27. Arni RK, Padmanabhan K, Padmanabhan KP, Wu TP, Tulinsky A (1994) Chem

Phys Lipids 67:59–66.28. Carter WJ, Cama E, Huntington JA (2005) J Biol Chem 280:2745–2749.29. Li W, Johnson DJ, Esmon CT, Huntington JA (2004) Nat Struct Mol Biol

11:857–862.30. Bock PE (1992) J Biol Chem 267:14974–14981.31. Neuenschwander PF, Williamson SR, Nalian A, Baker-Deadmond KJ (2006)

J Biol Chem 281:23066–23074.32. van de Locht A, Bode W, Huber R, Le Bonniec BF, Stone SR, Esmon CT,

Stubbs MT (1997) EMBO J 16:2977–2984.33. Dennis MS, Roberge M, Quan C, Lazarus RA (2001) Biochemistry 40:9513–

9521.34. Chavali GB, Papageorgiou AC, Olson KA, Fett JW, Hu G, Shapiro R, Acharya

KR (2003) Structure (London) 11:875–885.35. Liang WC, Dennis MS, Stawicki S, Chanthery Y, Pan Q, Chen Y, Eigenbrot

C, Yin J, Koch AW, Wu X, et al. (2007) J Mol Biol 366:815–829.36. Olivero AG, Eigenbrot C, Goldsmith R, Robarge K, Artis DR, Flygare J,

Rawson T, Sutherlin DP, Kadkhodayan S, Beresini M, et al. (2005) J Biol Chem280:9160–9169.

37. Segel IH (1993) in Enzyme Kinetics: Behavior and Analysis of Rapid Equilibriumand Steady-State Enzyme Systems (Wiley, New York), pp 161–226.

38. Otwinowski Z, Minor W (1997) Methods Enzymol 276:307–326.39. Collaborative Computational Project Number 4 (1994) Acta Crystallogr

D50:760–763.40. McCoy AJ, Grosse-Kunstleve RW, Storoni LC, Read RJ (2005) Acta Crystal-

logr D 61:458–464.41. Murshudov GN, Vagin AA, Dodson EJ (1997) Acta Crystallogr D 53:240–

255.

Wu et al. PNAS � December 11, 2007 � vol. 104 � no. 50 � 19789

BIO

CHEM

ISTR

Y