THEJOURNAL OF BIOLOGICAL CHEMISTRY 0 1994 by The American Society for Biochemistry and Molecular Biology, Inc Vol. 269, No. 52, Issue of December 30, pp. 32788-32795, 1994 Printed in U.S.A. Building Synthetic Antibodies as Adhesive Ligandsfor Integrins* (Received for publication, June 10, 1994, and in revised form, September 12, 1994) Jeffrey W.Smith+§,Dana HuS, Arnold Satterthwaitl, Sally Pinz-Sweeneyll, and Carlos F. Barbas IIInll From the Devartments o f PVascular (VB-1) andllMolecular Biology (MB-ll), The Scripps Research Institute, .. La Jolla, Caiifornia 92037 An antibody engineering strategy was employed to ligand binding to integrins and maintained essentially build high affinityligands and antagonists of integrins the same specificityas the parentantibody. a,,& and cyIrn& Previously, we inserted the integrin recognition motif, RGD, into the antigen binding site of a human antibody and selected the optimal flanking seIntegrin adhesion receptors are ap heterodimers that bind quencesfrom a phage-display library (Barbas, C. F., extracellular matrix and plasma-adhesive proteins(1-3). InteLanguino, L. R., and Smith, J. W. (1993)Proc. Nutl. Acad. grins direct numerouscell-matrix and cell-cell contacts and are Sci. U. S. A. 90,10003-10007). The resulting antibody, linked to organismal development and t o the progression of Fab-9, blocked the function of integrin a,& but also bound to the ligand binding site of platelet integrin several human diseases. The extensive primary structurehoa common gloanb&In thisreport, the antibody engineering effort has mology among all integrins indicates they have been extended by 1)redesigning Fab-9 to achieve spec- bal fold and probably operate via similar mechanisms. Perhaps ificity for platelet integrin aIm& 2) building non-RGD- the most important aspect of integrin function is the ligand containing antibodies that bind the ligand binding site binding event. Ligand binding triggers cell adhesion and cellof both &-integrins, and 3) testing the hypothesis that cell contact and often generates a cascade of intracellular sigpeptides derived from complementaritydetermining re- nals thatchange cellular behavior (2,4). Complications arise in gions (CDR) can be used to emulate the activity of the the studyof integrins because their natural ligandsoften bind in a non-dissociable manner (5, 6) and because most adhesive parent synthetic antibody.Thesegoalswereaccomplished by subjecting the original antibody, Fab-9,to a proteins bind multiple integrins on the cell surface. “motif optimization” (MTF). A phage library was conSince the studyof integrin function is complicated by native structed in which the residues flanking the RGD motif ligands that are difficult to work with andoften yield data that in Fab-9 weremaintained, but the RGDX sequence was are difficult to interpret,we have made an effort to “build high randomized. Thislibrary was panned on purified arm& affinity, well behaved, integrin ligands. We accomplished the to identify high affinity binders. Four function-blocking first step in this process by building Fab-9 (7). Fab-9 was obantibodies lacking RGD, but with specificity for aIrnp3, tained by redesigning a human antibody against the human were characterized. The antibody with the highest pref- immunodeficiency virus coat protein gp120 (8) using a synerence for aIrn&,MTF-10, had an adhesion sequence of thetic strategy to insert an RGD motif into thesequence of the KGDN. This sequenceis similar in primary structure to heavy chain CDR3 (Fig. 1).By incorporating this motif into a the active sequence within the disintegrin barbourin, phage library where the 6 residues flanking theRGD sequence which alsoantagonizes aIm& (Scarborough, R. M., Rose, were randomized, a series of antibodies was obtainedby “panJ. W., HSU,M. A., Phillips, D. R., Fried, V.A, Campbel1,A M., Nannizzi, L., and Charo, I. F. (1991)J.Biol. Chem. 266, ning” this library on purified integrin a,& (7). Fab-9, themost potent of these antibodies, binds toav& nearly 1000-fold better 9359-9362).MTF-10 had a 70-fold higher affinity for am& a,& and asp1.However, Fab-9 did not appear to distinthan than a&,. Through our selection strategy, we also identified several antibodies that lack RGD but still blocked guish the two &-containing integrins, a,& and aIIbPs. One of the objectives of this study was to determine whether ligand binding to both integrins withhighaffinity. Therefore, the RGD sequence is not necessary for a high further rounds of engineering and selection on the RGD motif affinity interaction with the ligand binding site of p3- within HCDR3l of Fab-9 could produce an antibody with specAlthough murine antibodiesthat integrins. Further investigation showed that the activ- ificity for either a,& or aIIb,Ps. ity of inhibitory antibodies could be emulated by syn- preferentially recognize the ligand binding site of aIB& over thetic peptides derived from the protein sequences of a,& have been raised via immunization (9, 10) andextensive the antibody’s HCDR3. CDR-derived peptides blocked screening of snake venoms has identified barbourin, a ligand that can distinguish aIIbPs and av& (ll), theapplication of the * This work was supported in part by National Institutes of Health synthetic antibody approach toward this end could supplant Grants CA56483 and AR42750 and Grant 3RT-0258 from the Tobacco- conventional immunization, where the generationof functionrelated Disease Research Program of the University of California (to J. W. S.). This is Manuscript 8365-VB from The Scripps ResearchInsti- blocking antibodies is largely fortuitous, and also obviate the tute. The costsof publication of this article were defrayedin part by the need for cumbersome screening of biological samples for spebe hereby marked cific inhibitors. A second objective of this study was to deterpayment of page charges. This article must therefore “uduertisement”in accordancewith18U.S.C.Section1734solely to mine if the CDR of an antibody is an appropriate vehicle for the indicate this fact. 6 An Established Investigator supportedby the American Heart Association and Genentech. Tiwhom correspondence should be addressed:The abbreviations used are: HCDR3, heavy chain complementarity Dept. of Vascular Biology(VB-l),The Scripps ResearchInstitute, 10666 determining region three; CDR, complementarity determining region; SPR, surface plasmon resonance; MTF,motif optimization; ELISA, enNorth Torrey Pines Rd., La Jolla, CA 92037. Tel.: 619-554-7107; Fax: zyme-linked immunosorbent assay; Fmoc, N-(9-fluorenyl)methoxycar619-554-6402. I/ A Scholar of the American Foundation for AIDS Research andthe bonyl; HPLC, high pressure liquid chromatography; PCR, polymerase recipient of an InvestigatorAward from the Cancer Research Institute. chain reaction. 32788 Integrins Against Antibodies Synthetic 32789 Ligand Binding and Inhibition Studies-The ability of synthetic antibodies to antagonize the activity of aV& and u ~ was ~ ~compared & in purified ligand-receptor binding assays. The method for these binding studies has been documented (7, 21). Ideally, the same ligand would have been usedfor inhibition studies, but fibrinogen does not bind a,p3 when Ca2+ is present (22). In contrast, Ca2+ is essential for the rapid binding of fibrinogento anb& Similar complications arisewhen vitronectin is considered becauseit does not bind identically tothe two integrins (22). Therefore, we chose to use the optimal ligand for each integrin, i.e. vitronectin for a,& and fibrinogen for arrb&. Radiolabeled MATERIALSANDMETHODS ligands were used at concentrations of 1nM, and binding was challenged Library Construction and Phage Selection of Antibodieswith unlabeled antibody. Following a 3-h incubation, free ligand was Oligonucleotides were purchased from OperonTechnologies (Alameda, removed by washing, and bound ligand was detected by gamma countCA). PCR mutagenesis was performed to construct a library of variants ing. The data from this analysis were highly reproducible with the within theHCDR3 of Fab-9 (7).PCR amplification was performed with difference betweendata points typicallybelow 11%. Data areexpressed the Fab-9 cDNA as a template using the following two oligonucleotide as the average of triplicate data points, and all experiments were reprimers: 1) CTCCTCCTCCTCCTCGACGTCCATATAATAGCAATTCC- peated at least four times. TMNNMNNMNNMNNCCCAAACGAGCACCCCACTCTCGCACA Surface Plasmon Resonance-Surface plasmon resonance (SPR) is a AATA and 2) GCAAWAACCCTCACTAAAGGG, where N isA, C, G, or means of assessing ligand affinities in real time (23, 24). SPR was T and M is A orC. The PCR product was digested withXhoI and AatII performed usingthe BIAcore plasmon resonance instrument from Pharand gel-purified. This product, 700 ng, was then ligated t o the Fab-9 macia Biotech Inc. Integrins were immobilized on the biosensor chip construct in thepComb vector (3 pg) digested with the same restriction withN-hydroxysuccinimide and N-ethyl-N"(3-diethylaminopropyl)carenzymes. Introduction of thisconstruct by electroporation into bodiimide according tothe methods outlinedby Pharmacia. The sensor Escherichia coli XL-1 Blue yielded a library of 3 x lo7 independent clones, which vary only in theRGDX region. Subsequent steps were as surface wasfirst activated with N-hydroxysuccinimideand N-ethyl-N" (3-diethylaminopropyl) carbodiimide. Integrins were coupled by injectpreviously described to produce phage-displaying antibody fragments ing 30 plof a 100pg/ml sample of purified integrin ontothe sensor chip on their surfaces (7, 12). surface. Ethanolamine was used t o block unreacted moieties on the To select antibodies withspecificity for each integrin from the phage sensor chip surface. library, integrinsa,& or aIIbp3were immobilized in Costar 3690microThe association and dissociation rate constants( k , and k - , ) for syntiter plates in 50mM Tris-HC1, 150 mM NaCl, 1 mM MgCl,, 1 mM CaCl,, thetic antibodies were obtained from BIAcore measurements asfollows. and 1 mM MnCl,, pH 7.4. Selection of phage-bearing antigen-specific antibody fragments was performed by panning for six rounds as de- To derive k - l , a pulse-containing ligand was passed throughthe sensor chip. Atthe endof the association phase, theflow was changedt o buffer scribed (12)but with thefollowing modification. To select for antibodies without ligand and the change in response unit (RU) was measured as with specificity for one or the other integrin, the second integrin was a function of time. The dissociation rate constant k-, is derived from used as a soluble competitor at 20 pg/ml during panning of the phage with the immobilized target integrin. For example,to select antibodies Equation 1. specific for aIIbPa,this receptor wasimmobilized in microtiterwells, and k-, = (In RU,IRU,)It - t, 0%. 1) avp3 was added, in solution, to the phage library during the panning step. RU, is the initial response unit due to binding of antibody, and RU, is Following selection of high affinity binders, soluble Fab was prothe response unit remainingfollowing dissociation. Time is designated duced by excision of the gIII fragment by digestion of the vector with as t. SpeI and NheI and subsequent religation as described (7, 12). Positive To obtain the association rate constant k,, an antibody was passed clones identified by simple ELISA and subsequent purification of isothrough the sensor chip containing unoccupied integrin, and the related clones was performedby affinity chromatography with goat antisponse unit as a result of binding is again measured as a function of human Fabcolumns. Dideoxy sequencing of double-stranded DNA was k , , is calculated as shownin time.Theassociationrateconstant, performed with the primer GGGAAGTAGTCCTTGACCAGGC. PeptideSynthesis-Syntheticpeptidesderived from HCDR3 se- Equation 2. quences of antibodies Fab-9, MTF-10, and MTF-32 were synthesized. k, = [((dRU/dt)/RU)-k-,/L 2)(Eq. Peptides from other antibodies were not synthesized or tested for economic reasons. Synthesis was performed on Rink's amide resin (0.5 L is the concentration of ligand. Measurements ofdRUldt are obtained mmoVg, NovaBiochem) with anAdvanced Chemtech 350 multiplepepa t several ligand concentrations. The overall K, of each binding event is tide synthesizer using Fmoc synthesis (13). Amino acids were coupled derived by simple division as shown in Equation 3. with diisopropylcarbodiimide and 1-hydroxybenzotriazole monohydrate. The side-chainFmoc amino acids were protected by pentamethKD= k - , l k , (Eq. 3) ylchroman-6-sulfonate for arginine, trityl for Cys, and butyl for Ser, Duplicate measurementsof k - , yielded identical values in all cases. In Thr, Asp, and Tyr. Each peptide was blocked at the N terminus by acetylation andat the C terminus by amidation. Peptides werecleaved the caseof k,, the difference was typically lessthan 13% across several from the resin with 10% trifluoroacetic acid/dichloromethane, and pro- ligand concentrations. Cell Adhesion and Platelet Aggregation Studies-The ability of retecting groups were removed with 5% water, 5% anisole, 2.5% ethanedicombinant antibodies t o support cell adhesion was examined with M21 thiol, 7.5% phenol in trifluoroacetic acid (14). Peptides were precipia& tatedwithdiethyletherandpurified on a C-18 column (Vydac melanoma cells, which adhere to vitronectin primarily through (25). Cell adhesion was measured essentially as described (22, 26). The 201TP1022, 2.2 x 25 cm) with a 0-100% water, acetonitrile gradient containing 1% trifluoroacetic acid. Cyclizations of cysteine peptides (1 wells of Titertek microtiter plates were coated with a range of antibody mg/ml) were carried to completion in 20% dimethyl sulfoxide, H,O at pH for 18 h a t 4 "C. M21 cells were harvestedfrom tissue culture flasks and 4.0 in 1 dayaccordingto Tam's procedure(15).Products from this placed in microtiter wells coated with the given antibody. Cells were reaction were purified via HPLC.Fast atom bombardment mass spec- allowed to adhere to immobilized antibody for 1 h. Non-adherent cells troscopy confirmed the molecular weight of the linear and cyclic pep- were removed by gentle washing, and bound cells were detected by tides. The cyclic disulfide peptides were negative for free thiol when colorimetric assay for acid phosphatase (26). Astandard curve with cells tested with Ellman's reagent (16). in suspension showed that all absorbance values were directly proporProtein Purification-Integrin aIbp3was purified from human plate- tional to cell number. lets by afflnity chromatographyon KYGRGDS-Sepharose (17). Integrin Platelet aggregation was performed with washed human platelets a,& was purified from a n octylglucoside extract of human placenta by (27). Platelets were drawn into acid/citrate/dextrose and then purified monoclonal antibody affinity chromatography (18). Vitronectin was pu- by gel filtration on Sepharose CL-2B. Aggregation was measured in rified from human plasmaby heparin-agarose affinity chromatography Tyrode's buffer, containing 2mM Ca" and 100 pglml purified fibrinogen. (19). Fibrinogen was purified from plasma by cold-ethanol precipitation Platelets were stimulated with 20 p~ ADP. Aggregation was measured (20). The Fab fragmentsof antibodies were affinity-purified from bacas a function of light transmission in a Scienco aggregometer. All agterial lysates by antibody affinity chromatography using goat anti-hu- gregation assays were performedat leasttwice with virtually identical man IgG-Sepharose. Proteins were radiolabeled with Iodogen (Pierce). results between experimental repetitions. display and identification of inhibitory peptide motifs. The prototypical RGD peptide, with sequence GRGDSP, does not greatly distinguish among integrins that bind to RGD. Here, we test the hypothesis that peptides with specificity for particular integrins can be designed based on the amino acid sequences of HCDRs in the antibodies selected from phage-display libraries. Synthetic Antibodies Against Integrins 32790 Motil Position Orlglnal Ubrary t234587891011 --VGCXXXRGDXXXCYY -?.-SI Fab-9 --VGCSFGRGDIRNCYY ‘- I S S I -VGCSFGXXXXRNCYY- - TARI.EI Amino acidsequences of HCDRJ of synthetic antibodies The protein sequencesof the randomized section of HCDR3 for Fab-9 and the MTF antibodies are shown. The residues randomized for the motif optimization are shown inbold. The protein sequence was derived from the nucleic acid sequence obtained from purified bacteriophage as described under “Materials andMethods.” Antibodies MTF-2, -10, -32, and -40 show substantial specificity for anh&Antibodies MTF-1, -12, -15 (selected by panning on a”&)and MTF-7, -13, -14, and -20 (selected by panning on all,@,)were derivedvia the competitive selection scheme described under “Materials and Methods” but either showed substantially lower binding affinity than Fab-9 or did not distinguish between a”&and allh& by more than 3-fold in ELISA. Synthetic antibody MTF Ubrary -S.5- FIG.1.Engineering and selection of synthetic anti-integrin antibodiee. Many RGD motifs are displayed at theapex of a flexible loop (42-44). Since antibodyCDRs (particularly HCDR3) often exist a s flexible loops, we placed an integrin recognition sequence into this region. To optimize antibody binding affinity, a phage library was constructed in which the residues flanking the RGD were randomized. The resulting antibodies are expressed as afusion protein on the surface of bacteriophage (12). The optimal antibody from our initial selection, Fab-9, bound a,& and allh&with high affinity. In this report,a similar strategy was used to obtain specificity for platelet integrin allh&and to identify inhibitory motifs that lack RGD. The sequences flanking the RGDI from Fab-9 were maintained to preserve the optimal presentation, but positions 5-8 were randomized in anew phage library termed MTF. Antibodies specific for al& were selected from this library by competing with purified a,& in solution and subsequently panningon purified all,,&. RESULTS Engineering Antibody Specificity by Motif OptimizationOne of the goals of this study was to determine whether antibodies that discriminate between the ligand binding pocket of the two PI-integrins could be designed. To accomplish this task, a “motif optimization”was performed starting with the scaffold of Fab-9, which binds to both receptors. The motif optimization involved the construction of a modified phage library. Residues corresponding to 1-4 and9-11 of the integrin bindingmotif in Fab-9 (Fig.1)were maintained in thenew phage library. However, residues a t positions 5-9, corresponding to RGDI of Fab-9, were randomized by oligonucleotide doping and PCR a s described under “Materials andMethods.” The complexity of this library ensureda 99% probability that allof the possible amino acid sequences were represented in the randomized 4-residue motif. This phage library was screened for antibodies thatcould distinguish a,& from allhPR using three rounds of competitive selection as described under“Materialsand Methods.” The phage antibodies resulting from this selection were initially tested for specificity by ELISA (not shown). Antibodies that displayed a substantial preference for the target integrin in ELISA were expressed as soluble Fab fragments in E. coli (7) and then screened more rigorously by measuring their ability to antagonize receptor activity in purified ligand-receptor binding assays. Threeantibodies wereobtained that showed substantial binding preference for allh&:MTFS, MTF-10, and MTF-32 (Table I). The abilityof these antibodies to interfere with ligand binding was measured with purified ligand-receptor binding assays. MTF-2, -10, and -32 had lower IC,, values for allhPJ than a,& in purified ligand-receptor binding assays. The IC,, of MTF-2 was 20-fold lower for alrh&than for a,& (Fig. 2 A ) . MTF-10 exhibited a 50-fold greater ability to antagonize cyllh& (Fig. 2B). In the sameassay, MTF-32 displayed a 15-fold preference for q r h & (not shown). All of the MTF antibodies exhibited a higher IC,, than the parent Fab-9, indicating that by gaining specificity for aIIhP3, some affinity for the integrin was sacrificed. Interestingly, no antibodies were identified with the competitive selection of thephagelibrarythat displayed CDR sequence 1-2-3-4-5-6-7-8-9-10-11’ Fab-9 MTF-2 MTF-10 MTF-32 MTF-40 C-S-F-G-R-G-D-I-R-N-C C-S-F-G-R-T-D-Q-R-I-C C-S-F-G-K-G-D-N-R-I-C C-S-F-G-R-R-D-E-R-N-C C-S-F-G-R-N-D-S-R-N-C MTF- 1 MTF-12 MTF-15 MTF-7 MTF-13 MTF-14 MTF-20 C-S-F-G-R-V-D-D-R-N-C C-S-F-G-R-A-D-R-R-N-C C-S-F-G-R-S-V-D-R-N-C C-S-F-G-K-R-D-M-R-N-C C-S-F-G-R-W-D-A-R-N-C C-S-F-G-R-Q-D-V-R-N-C C-S-F-G-R-D-D-G-R-N-C Numbers indicate motif positions. greater than 3-fold binding specificity for a& This suggests that within thecontext of the Fab-9 scaffold, it isunlikely that the adhesion motifs can be manipulated to gain substantial specificity for a,& over allh&. This is not surprising because we are unawareof any RGD-based ligands thatbind to a,& but not to allh&’ One antibody, MTF-40, survived the competitive selection process for q I h & and showed preference for this integrin by SPR (see below) but did not display specificity in the purified ligand-receptor binding assays. Plasmon resonance analysis of the affinity for this Fab showed that in Ca2+ it has a 10-fold higher affinity for allh&than a,& (see below) and also the highest affinityfor allhB3(1 x M) of any of the antibodies derived from the phage libraries. Itlikely is that thepreference of this antibody for aIlhPRwas not detected in the inhibition assays because it also retained very high affinity for a,&. The association rate of the native ligands for both integrins is so slow (22) that the ligand binding assaynot is sensitive enough to detect small differences in binding affinity if the affinity of the antibody is very high. Asimilar scenario is evident for the parent Fab-9, which showed no preference in ligand competition assays but has obvious an preference for a,& when binding is measured directly with plasmon resonance (Table 11). Since sequences corresponding to RGDI in Fab-9 were randomized in the MTF phage library (Fig.l), another advantage of this selection scheme is that it has the potential identify to amino acid sequences other than RGD that could bind to the integrin ligand binding pocket. In fact, none of the MTF antibodies contain the exact RGD motif, although in three of four cases Arg and Asp are conserved. No functional substitutions for Asp were observed in anyof the selected clones. In MTF-10, the arginine was replaced by lysine, a substitution that has been found in other ligandswith specificity for a,,&, including barbourin (11).Four other antibodies lacking RGD were also identified from this phage library that bound equally well to a,& and allhPI.These antibodies did not display a substantial preference for either a,& or all,,&in ELISA. This group of antibodies also had IC,, values 20-50-fold higher than Fab-9, consistent with a substantially lower binding affinity for integrin. The CDR sequences of these antibodies a t positions 5-9 were RWDA, RADR, and KRDM (Table I). One antibody with 32791 Synthetic Antibodies Against Integrins A .-GI _I MTF-llb-Ill0 TABLE I1 The affinities of synthetic antibodies for a,& and alIbP3as measured by surface plasmon resonance The association (k, or ken) and dissociation (k-l or k,) rate constants between synthetic antibodies and the two P,-integrins were determined in real time with surface plasmon resonance. Data collection and analysis were performed with purified integrins and Fab fragments as described under "Materials and Methods." Measurements were made in either 2 mM Ca2+or 0.2 mM Mn2+,both of which have been found to saturate theligand binding response of the p,-integrins. (2) ;:I lo nl -- 1 1 Cation I -10 I y -9 -8 Log[MTF-Z] B (10) MTF-llb-lllo KGDN 2 Y c 0 0 L O x n I -7 Receptor Antibody KonW's-l) Fab-9 1.3 x loL Ca2+(2 mM) cyII& MTF-2 7.8 x lo4 MTF-10 1.7 x lo5 MTF-32 6.8 x lo5 MTF-40 7.7 x lo5 Fab-9 1.4 x lo6 aVp3 MTF-2 1.4 X lo4 MTF-103.4 x lo3 MTF-32 1.5 x lo5 MTF-40 1.0 x lo5 Fab-9 4.0 x lo3 aIm& MTF-2 1.8 x lo4 Mn2+( 0 . 2 m ~ ) MTF-10 4.3 x lo4 MTF-32 1.1x lo5 MTF-40 4.1 x lo4 Fab-9 2.4 x lo5 a,& MTF-2 3.8 x lo4 MTF-103.6 x lo4 MTF-32 1.3 x lo5 MTF-40 1.1 x lo5 KO,&") 7.0 x 1.3 X 1.2 x 2.0 x 8.4 X 2.3 x 1.8x 10-4 10-3 10-3 10-3 10-4 10-3 10-3 1.7 x 10-3 1.0 x 10-3 1.1 x 10-3 6.6 x 1.7 x 10-4 3.6 x 10-4 1.2 x 10-4 1.5 x 10-4 6.5 x 10-5 1.0 x 10-3 1.8 x 10-3 8.7 x 10-4 9.6 x 10-4 KdU" 5.0 x 1.7 x 7.0 x 2.9 x 1.1x 10-9 10.' 10-5 10-9 10-9 1.6x 10-9 1.3 x 10-7 5.0 x 10-7 6.7 x 10.' 1.1 x 1.6 x 10" 1.0 x 10-8 8.3 x 10.' 1.0 x 10-9 3.6 x 10-9 2.8 x 10"' 2.6 x lo-' 5.0 x lo-' 6.7 X 10-9 8.8 x lo-' lute on and off rates. SPR also enabled a measure of the effect of different divalent ions on k, and k - l . Because divalent cations, like Ca2+and Mn2+,are known t o dramatically influence 10 \i#l ligand binding to &-integrins (221, wemeasured the binding of I 0 -11 -10 -9 -8 -7 the antibodies in the presence of both ions. In prior studies, it log[MTF-IO] was determined that ligand binding to either integrin could be FIG.2. Motif optimization of Fab-9 generates antibodies with maximally supported by either 2 mM Ca2+or 200 V M Mn2+and specificity for orm&. Antibodies with binding preference for arIbP3 that these ions had different effects on ligand association (22, were generated by redesign of Fab-9 as described under "Materials and 28). These conditions were adopted for SPR analysis, and an Methods" and Fig. 1. Following selection by panning and initial screening byELISA, the MTF antibodies were tested in ligand inhibition SPR "sensogram" showing the association and dissociation of studies. The ability of two of these antibodies, MTF-2 (A) and MTF-10 MTF-10 with integrin aIm& is shown in Fig. 3. The marked ( B ) ,to block ligand binding is shown. Ligand binding to avp3(0)was influence of the type of divalent ion present on the binding of measured with '251-vitronectin,and ligand binding to aImp3(0) was MTF-10 is apparentfrom the slopes of the sensograms in Fig. assessed with '251-fibrinogen.All data are expressed as the percent of control binding in the absence of inhibitor. Data points are the average 3. The association of MTF-10 with cyIIb/.?, in Ca2+(Fig. 3 A ) is of triplicate points in which the error was less than 11%of the total more rapid than inMn2+(Fig. 3B), leading to a correspondingly specific binding. Nonspecific binding was determined by competition greater slope of the binding isotherm. The quantitative differwith RGD peptides and was normally less than 10%of the total bound ence between association in the presence of the two ions is counts. actually more dramatic than displayed because a higher concentration of MTF-10 had to be used in buffer containing Mn2+ the sequence RVDD had anIC,, value comparable with that of to obtain valid binding isotherms. The association rate conFab-9 but also failed to distinguish the two P,-integrins. In stant (k,) is derived from the slope of this phase as described conjunction with the sequences of the other MTF antibodies under "Materials and Methods." To measure dissociation, the (Table I), these data indicate that position 6 in the adhesion sensor chip was placed in buffer lacking free antibody (arrows), motif (Fig. l), which corresponds to the Gly position in the and antibody dissociation from arm& is readily observed as a RGD, is highly permissive. This glycine can be substituted by descending phase in the sensogram. Val, Ala, Asn, Arg, Thr, Gln, Asp, Ser, and Trp. Consequently, a The rate constants for the other MTF antibodies in these two full steric spectrum of side chainsis tolerated in this position. ions is shown in Table 11. Regardless of the antibody tested, Although several ligands for both a,P3 and uIIbP3have been association of antibodies with uIIbPswas always faster in ea2+. identified, there hasbeen no systematic searchfor other poten- However, sequence changes in residues 5-9 did influence the tial ligands. Our data indicate that such a search should not be magnitude of the cation preference as much as 5-fold.For MTFlimited to proteins containing RGD. The potential ligands for 40, the ratio of k,Mn2+/k,Ca2+ is 0.05, but for MTF-2,the ratio is the &-integrins should be expanded to include proteins with 0.23. In contrast to aIIb&, integrin a,P3 usually binds its ligands (K/R)XD sequences. better when Mn2+is present (22). In general, this trendis also Measuring Ligand Association and Dissociation Constants maintained with the MTF antibodies. Fab-9 appears to be the for Synthetic Antibodies-To further characterize the binding one exception because it binds faster in Ca". The type of divabetween the MTF antibodies and integrins, SPR was used (23, lent ion present can affect k , as much as 10-fold. However,the 24). In contrast to ligand competition assays, where ligand on rates of both MTF-32 and MTF-40 fora$, were very similar affinity is a parameter and where association and dissociation regardless of the type of ion present. Antibody dissociation rate cannot be distinguished, SPR allows a direct measure of abso- constants also varied with the type of divalent ion present, but "1 I 32792 Synthetic Antibodies Against Integrins 600 I Control 3 a 5xlOSM 1 ', 2dO 4d0 do Time (s) 860 71300 ' 20xlOgM 50xIO9M Fab-9 Control 5x1OW c 3 a 100 20x10-W 50 '0 500 1000 1500 2000 2500 3000 Time (s) FIG.3. Measuring the binding of MTF-10to integrins with surface plasmon resonance. Surface plasmon resonance was used to measure on and off rates between MTF-10and a&. The binding was measured in buffer containing 2 m Ca2+( A )or 200 PM Mn2+( B ) .In A, a solution containing 5 p g h l of MTF-10 waspassed through the sensor chip containing immobilized aIB&. Because the association of antibody occurs more slowly in Mn2+,in B , the concentration of solution phase MTF-IO was increased to 25 pg/ml to obtain a valid binding isotherm. Arrows denote the time at which solution phase was changed to buffer lacking free MTF-10 to allow for dissociation of bound antibody from cur,,,&. RU, response unit. MTF-IO FIG.4.Fab-9and MTF-10 block platelet aggregation.The ability of Fab-9 (A) and MTF-10 ( B ) to block platelet aggregation was measured with washed human platelets. Human platelets (1 x 10') were mixed with 100 pg/ml of purified fibrinogen and 2.0 mM Ca2+in Tyrode's buffer.Thesewereplaced in a glass aggregation tube. The indicated concentration of antibody was added, and platelets were stimulated with 20 PM ADP. Aggregationwas measured using a Scienco aggregometer. rI there was no obvious correlation between the influence of ions on k , and k-, for a given antibody. Consequently, divalent ions can influence the ligand on-rate without necessarily influencing itsoff-rate. These data suggest thatligand association and dissociation with integrinsare independently regulated by divalent ions. A Comparison of Fab-9 and MTF-10 in Platelet Aggregation and Cell Adhesion-All of the initial selection and characterization of the MTF antibodies relied upon the use of purified integrins. To test the activity of the MTF antibodies in a biological assay, we measured their ability to block platelet aggregation (Fig. 4). Washed platelets were placed in Tyrode's buffer containing 2 m~ Ca2+ and100 pg/ml fibrinogen, and aggregation was stimulated by addition of 20 p~ ADP. Fab-9 wasused as a reference, and it blocked 50% of the aggregation at a concentration near 5 nM and completely blocked aggregation at a concentration of 20 nM (Fig. 4A). Only slightly more MTF-10 was requiredfor inhibiting aggregation, with half-maximal inhibition a t 20 nM Fab and complete inhibition a t 100 r m (Fig. 4B).MTF-2 and MTF-40 were slightly better at blocking aggregation than MTF-10 with IC,, concentrations of 15 and 10 nM, respectively (not shown). The IC,, for the MTF antibodies in platelet aggregation studies agree with theaffinity of each antibody as measured by SPR. The ability of these synthetic antibodies to abrogate platelet aggregation validates their potential as anti-thrombotic agents. The antibody most selective for aIh& is MTF-10. To assess its ability to interact with the sister integrin a,& in a biological membrane, cell adhesion assays were performed using M21 melanoma cells because they adhere toRGD ligands primarily via a,P3 (25). M21 cells were tested for adherence to a concentration rangeof immobilized Fab-9 orMTF-10. As shown in Fig. 5 , the cells were able to adhere to Fab-9 butwere unable to bind to MTF-10. Thus, the specificity exhibited by MTF-10 in puri- Fab-9 ." n 0.8 0.6 0.4 0.2 - MTF- 10 V 0.0 [Antibody] (ug/rnl) FIG.5. MTF-10 fails to support a,&-mediated cell adhesion. The ability of Fab-9 (0)and MTF-10 (D) to support a&,-mediated cell adhesion was compared with M21 melanoma cells. A range of each antibody was immobilizedin microtiter wells for 18 h at 4 "C. Nonspecific binding sites on the plate were blocked with 0.1% bovine serum albumin, and asuspension of cells were addedto each well. Followinga I-h incubation period, non-adherent cells were removed by gentle washing and aspiration. The relative number of cells in each well wasquantified with a colorimetric assay for lysosomal acid phosphatase using the absorbance at 405 nm (26). fied receptor assays isalso evident incell-based assays. Importantly, antibody MTF-10 will support the adhesion of Chinese hamster ovary cells expressing recombinant human aIIbP3.2 Synthetic Peptides Derived from Antibody HCBR3 Protein Sequences Faithfully Antagonize p3 Zntegrins-Phage display of random peptide libraries is becoming standard technology in protein biochemistry (29,30). Engineering synthetic antibodies by manipulating HCDR3 offers a unique vehicle for the expression and analysisof random peptide sequences.To determine if the CDR is an appropriate vehicle for identifying inhibitory peptide motifs from phage libraries, we measured theability of synthetic peptides derivedfrom the sequences of Fab-9, MTF-2, and MTF-40 to block ligand binding to selected integrins. PepJ. Smith, C. Barbas 111, and J. Loftus, manuscript in preparation. Integrins Against Antibodies Synthetic A. 32793 Lin-F9 100 I90c 80 70- T . I . , r 60 x u D - 50 - !j 40 2 30 - - 2ol . -I 1 10 ’ 0‘ I -9 -7 -8 -5 -6 [Peptide] 6. GRGDSP loor a 90 - 80 - 70 60 - 50 - 40 - 30 - -9 -7 -8 -5 -6 [Peptide] Peptide #40 C. 2 120 110 100 90 L 80 2 ,“ v ’O v c 50 g 40 r- 60 -11 -9 -10 -8 -7 -6 -5 -4 [Peptide] D. = Peptide # 2 120 110 100 90 ; 0” e tides derived from the sequence of MTF-10, which has a KGD sequence, werenot testedbecause peptides containingthis motif have beenpreviously shown to inhibit thefunction of a&, (31, 32). The RGD motif we originally inserted into the CDRcontained 2 flanking cysteine residues t o provide the opportunity for constraining theflexibility of the RGD loop by the formation ofa disulfide bond. Since it is notevident whether thisdisulfide bond actually forms in theCDR of the antibody, both cyclic and linear forms of each peptide were synthesized and tested. Cyclization was achieved as described under“Materialsand Methods” and confirmed by HPLC and mass spectroscopy. In the case of Fab-9, the linear peptides were synthesized with cysteines (sequence of CSFGRGDIRNC) and with glycine bequence of GSFGRGDIRNG) to eliminate thepotential for peptide cyclization during the inhibition assay. Importantly, both the linear (Cys and Gly forms) and cyclic forms of these peptideshadnearlyidentical efficacy in all assays (datanot shown). Consequently, we conclude that cyclization is not a determinant in binding affinity for peptides derived from the Fab-9 sequence. The ability of synthetic peptides basedon the Fab-9sequence to block ligandbinding t o av&, alIbP3, and a,& in purified ligand-receptor binding assays is shown in Fig. 6. In examining the peptides for activity, there aretwo criteria for judging their efficacy: affinity and specificity. The proto-typical RGD peptide, GRGDSP, was used as a positive control in inhibition assays because it antagonizes all three integrins (Fig. 6B). Interestingly, the CDR-derived peptide (Lin-9) inhibited ligand binding t o a,& and aIIb& butdidnot antagonize(Fig. 6A). Thus, Lin-9 is the first RGD-containing peptide, that we are awareof, with specificity for a,& over CY&.It is important to note that GRGDSP did antagonize Infact, of thethreeintegrins tested, a,& had the highest apparentaffinity for this peptide. We conclude that the synthetic peptide derived fromthe CDR3 of Fab-9 emulates the activity of the whole antibody and, like Fab-9, maintains specificity for a,& over the related integrin : Synthetic peptides based on the sequence of HCDRB in antibodies MTF-2 and MTF-40 (sequence of GCSFGRTDQRNCY, peptide 2; and GCSFGRNDSRNCY, peptide40) were also tested for their ability to block ligand bindingto a,p3 and arrb&. These twoexamples were chosen because 1) MTF-2 and MTF-40 have binding preference for aIIbP3 and 2) they provide a good comparison to see if CDR-derived peptides maintain the Antibody samerank-order potency as theFabfragments. MTF-40 has a much higher affinity than antibody MTF-2 for aIIbp3,and peptide 40 completely blocked ligand binding to anb& but only inhibited 40% of the ligand binding to a,P3 over the range of concentration tested (Fig. 6C).We were unable to test higher concentrations of peptide because several nonspe- 80 almP3(W. B , the same concentration range of peptide GRGDSP was tested in an identical assay. All data are expressed as the percent of E 50 control binding in the absence of inhibitor and are theaverage of triplicate data points. This is representative of four experiments in which m0 40 nearly identical results were obtained in each repetition. C, peptides u 30 derived from HCDR3 of MTF-40 (sequence = GCSFGRNDSRNCY)were 6m 2 0 .also synthesized in linear (filled symbols) and cyclic forms (open symA ’0 b o k ) .The ability of these peptides toblock ligand bindingto a,& (V,0) 01’ ’ -11 -10 --89 -7 -6 -5 - 4 and allbPa (0,V ) were compared. All data are expressed as the percent of control binding in the absence of inhibitor and are the average of LPsptidsJ triplicate values. This experiment is representativeof three repetitions FIG.6. Synthetic peptides derived from synthetic antibodies in which nearly identical results were obtained. D, similar measureinhibit integrin function and display the specificity of the whole ments were madefor peptides derivedfrom HCDR3 of MTF-2 (sequence antibody. Peptides derived from the CDFU protein sequence of Fab-9 = GCSFGRTDQRNCY). Linear (filled symbols) and cyclic (open sym(GSFGRGDIRNG and CSFGRGDIRNC) were synthesized in linear and bols) peptides were tested for the ability to block ligand binding to a,P3 cyclic forms as described under “Materials andMethods.” A, the linear (V,0) and allbP3(0,VI. Again, the data are the average of triplicate form of this peptide, Lin-9, was tested for its ability to block lZ5I- points, and the experiment is representative of three repetitions with vitronectin binding t o a$, (V), avP5(0)and ‘251-fibrinogenbinding to similar results. v 70 60 0 , 8 8 1 32794 Synthetic Antibodies AgainstIntegrins cific peptides began to disrupt ligand binding in this assayat concentrations of 5 x M. Peptide 2 also abolished fibrinogen binding to aImP3but did not interfere with ligand binding to a$, over the range tested (Fig. 6D). These data show that synthetic peptides derived from CDR3 sequences of MTF-2 and MTF-40 have the same integrinpreference as the parent antibodies. The peptides alsohave the same rank-order potency as IC,,of peptide 40 is 5 x the antibodies. For integrinthe M, and that of peptide 2 is 2 x M. Like peptides based on the Fab-9 sequence, the activity of peptide 2 or 40 was not enhanced by cyclization (data not shown). only a 20-fold lower IC,, for a$, than aIrbP3 (32). Collectively, these data suggest that itwill be substantially more challenging to obtain a n RGD-based antagonist with a large degree of selection for a&. Since we exhausted allof the potential natural amino acid sequences with the MTF selection strategy, it is likely that a$,-specific antagonists will require access to functional moieties not present in the20 amino acids and will have to be organically synthesized. This does not exclude the possibility that antibodies can be further engineered to gain selection for a,&. For example, the HCDR3 of Fab-9 can be recombined with libraries of light chains to build in other contact sites specific for a$,. Alternatively, five other CDRs are presDISCUSSION ent in the antibody and could be genetically manipulated to Numerous integrin ligands have been identified. These in- achieve the same end. clude extracellular matrix proteins, plasma adhesive proteins, It is somewhat surprising that our selection strategy did not cell-bound adhesive ligands, and venom proteins from snakes and identify the sequences flanking the RGD that are present in ticks (2).All of these ligands have been optimized through millions natural extracellular matrix and blood-borne adhesive proteins of years of evolution. Here, we showthat integrin ligands and an(35). This discrepancy suggests that the integrin-ligand bindtagonists that maintaina high degree of specificity can be built by ing pocket could accommodate more proteins than originally using antibodies as a scaffold. We believe this is a significant step thought. Our datashow that theRGD motif is certainly not an toward the futuredesign of biologicalligands with desired binding absolute requirement for occupation of the ligand binding site characteristics. The synthetic approach described here may evenof the P,-integrins. Extending the definition of &-ligands to tually obviate the need to randomly screen biological samples for sequences of ( w R ) X D should be considered, and this would inhibitors of protein-protein interactions and could also eliminate vastly increase the number of potential natural ligands for the cumbersome manipulations associated with conventional integrins. This is illustrated by a computer search of the Prohybridoma technology. tein Identification Resource Data base using the RTDQ seInterestingly, important differences were observed in the way that the synthetic antibodies and native ligands bind to quence of antibody MTF-32. Of 56,849 sequences in the data integrins. Native adhesive ligands for the integrins typically base, 3,026 contained RTDQ. Similar numbers were obtained have verylow association rate constants but ultimately bind in for RNDS, the active sequence within MTF-40. Consequently, a non-dissociable manner (5, 6, 22). In contrast, our engineer- the pool of potential p,-ligands is much larger than originally ing efforts generated synthetic antibodies that rapidly bound believed. The sequences we identified at motif positions 5-9 the ps integrins, with at least one interaction greater than 1x were also distinct from antibodies raised by immunization that lo6 “‘s-’. This rate is 600-fold faster than the association of bind the ligand binding site of aIm/3,(9, 10). Antibodies derived RYD or RHD motif in CDR3 and vitronectin for a$, (22). However, unlike vitronectin, all of the from immunization all have an antibody ligands bound in a completely dissociable manner. are specific for aI1,&,but our selection scheme did not identify The inability of the synthetic antibodies to bind in a non-dis- any antibodies with this sequence. It is likely that antibodies sociable manner indicates that the native ligands probably derived from immunization, where in vivo selection of the anhave ancillarycontact surfaces aside from RGD that contribute tigen binding siteoften relies upon the collective binding affinity of six CDRs, also derive binding affinity from secondary to the stabilized binding state. These hypothetical secondary contact points are also a likely means to obtain anotherlevel of contacts with integrin that are contributed by other CDRs in the antibody. It is unlikely that the antibodies described here specificity between integrins and their naturalligands. Attempts to build synthetic antibodies that could distinguish make contacts outside of HCDR3 because these contacts would the two &integrins produced three antibodies in the MTF be entirely fortuitous. The parent antibody, hivl2, does not series thatshow a substantial preference for a,&,, e.g. MTF-10 bind either integrin. Much recent effort has been placed on screening for ligand has a 70-fold lower KD for aIIbpsthan av&. Interestingly, the same selection strategy to find antibodies with preference for motifs using random peptide libraries displayed on the surface of phage (29), and some studies haveidentified integrin binding a$, failed. Fab-9, which was originally identified by panning on pure a,& does show preference for this integrin inplasmon peptides from phage libraries (36). Our approach differed substantially in that the CDR of an antibody was used asa vehicle resonance binding studies, but we did not identify antibodies withgreater specificity by optimizing the motif in Fab-9. for presentation of the peptide motif. Therefore, we were unpeptides derived from the selected Clearly, the method that must be used to identify specific an- sure as to whether synthetic tibodies leaves a slight chance that antibodies with a similar sequences would behave faithfully as high fidelityantagonists. preference for a,$, could be missed. However, it is unlikely that In fact, HCDR3-derived peptides do block integrin function. any motifs with greater than 10-fold selectivity for a,& were More importantly though, these peptides mimic thetarget present. The complexity of the library we screened was suffl- specificity of the parent antibody. Even though both p,-integrins and the highly related integrin a,& bind t o the protocient to ensure that all potential amino acid sequences were present in the motif optimization, leading to the conclusion that typical RGD sequence, GRGDSP, the RGD peptide derivedfrom within thecontext of the adhesion motif shown in Fig. 1,the set Fab-9 bound only the P,-integrins. Similar target specificity Interestingly, no was observed for peptides derived from MTF-2 and MTF-40. of all sequences that bind a$, also bind aIIbP3. natural ligands have been identified that bind a$, substan- Like the parent antibodies, peptides 2 and 40 showed specificover a,&. It is unclear whether the selection of the tially better than aIIbp3. Although peptidomimetics and cyclic ity for aIIb,p3 antibody CDR contribute more peptides have been reported to have a high degreeof specificity proper flanking residues in the (33, 34), we are unaware of any such compounds that contacts with integrinor whether these flanking residues have for aIIbp3 have good preference for a,P3. Despite the identification of sev- conformationally constrained the angle that defines the orientation of the side chains of the Arg and Asp residues. Surpriseral snake venom peptides that selectively antagonize aI&, only one such protein, cerastin, hasspecificity for a,& but has ingly, cyclizing the CDR peptides provided no advantage over Synthetic Antibodies Against Integrins 32795 A,, and Shattil, S.J. (1989) J. Biol. Chem. 264, 259-265 the linear peptide sequences in terms of affinity or selectivity. 10. 'Ibmiyama, Y., Tsubakio, T., Piotrowicz, R. S., Kurata, Y.,Loftus, J. C., and The selection of the optimal flanking sequences may add SO Kunicki, T. J. (1992) Blood 79, 2302-2312 many additional linear contact points for the integrin that CY- 11. Scarborough, R. M., Rose, J. W., Hsu, M. A,, Phillips, D. R., Fried, V. A., Campbell, A. M., Nannizzi, L., and Charo, I. F. (1991) J. Biol. Chem. 266, clization provides no additional advantage. Alternatively, the 9359-9362 minimal energy conformation of the linear peptides may ap- 12. Barbas, C. F., 111, Bain, J. D., Hoekstra, D. M., and Lerner,R. A. (1992) Proc. Natl. Acad. Sci. U. S. A . 89, 44574461 proximate the same structure as thecyclic form, in which case no difference in theaffinity of the two forms would be expected. 13. Fields, G . B., and Noble, R. L. (1990) Int. J . Pept. Protein Res. 35, 161-214 Fields, C. G., and Fields, G. B.(1990) Int. J. Pept. Protein Res.36, 14. King, D. S., Resolution of this issue must await an analysis of peptide con255-266 15. Tam, J. P., Wu, C. R., Liu, W., and Zhang, J. W. (1991)J. Am. Chem. Soc. 113, formation by NMR. 6657-6662 The findings presented here show that CDRs are a viable 16. Stewart, J.M., and Young, J. D. (1984) Solid Phase Peptide Synthesis,2nd Ed., design template for obtaining inhibitory peptides that mainpp. 116-134, Pierce Chemical Co., Rockford, IL tain thesame specificity as theparent antibody. There are now 17. Lam, S. C-T., Plow, E. F., Smith, M. A,, Andrieux, A,, Ryckwaert, J-J., Marguerie, G., and Ginsberg, M. H. (1987) J . Biol. Chem. 262,947-950 examples in the literature describing the use of CDR-derived 18. Smith, J. W., and Cheresh, D. A. (1988) J. Biol. Chem. 263,18726-18731 synthetic peptides to emulate the activity of the whole antibody 19. Yatohgo, T., Izumi, M., Kashiwagi, H., and Hayasbi, M. (1988) Cell Struct. Funct. 13, 281-292 (9, 37, 38). By optimizing with phage display, the affinity and 20. Doolittle, R. E , Schubert, D., andSchwartz, S. A. (1967) Arch. Biochem. specificity of such CDR-derived peptides could be improved. Biophys. 118,456-521 Consequently, monoclonalantibodies could be used as a start- 21. Smith, J. W., Vestal, D. J., Irwin, S.V., Burke, T. A., and Cheresh,D. A. (1990) J. Biol. Chem. 266, 11008-11013 ing point for obtaining antagonist peptides. 22. Smith, J. W., Piotrowicz, R. S., and Mathis, D. (1994)J. Biol.Chern. 269, Finally, a major advantage to building human antibodies 9fiO-967 ". ". that bind the integrin ligand binding site is that they could 23. Altschuh, D., Dubs, M-C., Weiss, E., Zeder-Lutz, G., and Van Regenmortel, M. H. V. (1992) Biochemistry 31,6298-6304 have immense therapeutic application, particularly because 24. Fagerstam, L. G., Frostell-Karlsson. A,, Karlsson, R., Persson, B., and they should circumvent a host immune response. The two p3Ronnberg, I. (1992) J. Chromatogr. 697,397-410 integrins have been implicated in numerous diseases. An an- 25. Cheresh, D. A,, and Spiro, R. C. (1987) J. Biol. Chem. 262,17703-17711 tagonist of the platelet integrin would arrest platelet aggrega- 26. Pratner, C. A., Plotkin, J., Jaye,D., and Frazier,W. A. (1991) J. Cell Biol. 112, 1031-1040 tion and could find wide application in treating thrombotic 27. Marguerie, G. A,, Edgington, T. S., and Plow, E. F. (1980)J . Bid. Chem. 265, 154-161 episodes. In fact, antagonists of aIIbp3have been effective at blocking platelet function in vivo (34,39).The fact that some of 28. Smith, J. W., and Cheresh, D. A. (1991) J. Biol. Chem. 266,11429-11432 29. Barbas, C. F., I11 (1993) Cum Opin. Biotechnol, 4, 526-530 the synthetic antibodies antagonize both &-integrins with high 30. Cull, M. G., Miller, J. F., and Schatz,P. J. (1992)Proc. Natl. Acad.Sci. U. S. A. 89,1865-1869 affinity may be extremely useful in combating certain types of M., Pierschbacher, M. D., Ruoslahti, E., Marguerie, G., andPlow, E. metastasis. For example, it is now thought that melanoma is 31. Ginsberg, (1985) J. Bzol. Chem. 260,3931-3936 often carried through the vasculature by adhesion to circulat- 32. Scarborough, R. M., Rose, J. W., Naughton, M. A,, Phillips, D. R., Nannizzi, L., Arfsten, A., Campbell, A. M., and Charo, I. F. (1993) J. Biol. Chem. 268, ing platelets (40,41). This cell-cell bridge is thought to involve 105a-1065 a,& on the tumor cell and qrn& on the platelet, and blocking 33. Isoai, A., Ueno, Y., Giga-Hama, Y., Goto, H.. and Kumagai, H. (1992) Cancer both ends of this adhesion event with the synthetic antibodies Lett. 66, 259-264 described here would seem to present a superb therapeutic 34. Zablocki, J. A., Miyano, M., Garland, R. B., Pireh, D., Schretzman, L.,Rao, S. N., Lindmark, R. J., Panzer-Knodle, S. G., Nicholson, N. S., Taite, B. B., strategy, Salyers, A. K., King, L. W., Campion, J. G., and Feigen, L. P. (1993) J. Med. Acknowledgments-Weacknowledge Yverre Bobay, Doug Cababa, Terri Jones, and Kim Green for superb technical assistance. We thank Shari Olsen for secretarial support. REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. Albelda, S.M., and Buck, C. A. (1990)FASEB J . 4, 2868-2880 Hynes, R. 0. (1992) Cell 69,ll-25 Phillips, D. R., Charo, I. F., and Scarborough, R. M. (1991) Cell 66, 359-362 Schwartz, M. A. (1993) Cancer Res. 53, 1503-1506 Muller, B.,Zenves, H.-G., Tmgemann, K., Peter, J., and Engel, J. (1993) J. Biol. Chem. 268,680G6808 Orlando, R. A,, and Cheresh, D. A. (1991) J . Biol. Chem. 266, 19543-19550 Barbas, C. F., Languino, L. R., and Smith, J. W. (1993) Proc. Natl. Acad. Sci. U. S. A . 90, 10003-10007 Barbas, C. F., 111, Bjorling, E., Chiodi,F., Dunlop, N., Cababa,D., Jones, T. M., Zebedee, S.L., Persson, M. A. A., Nara, P. L., Norrby, E., and Burton,D. R. (1992)Proc. Natl. Acad. Sci. U. S. A. 89,9339-9343 Taub, R., Could,R. J., Garsky, V. M., Ciccarone, T. M., Hoxie, J., Friedman, P. Chem. 36,1811-1819 35. Ruoslahti, E., and Pierschbacher, M. D. (1987) Science 238,491-497 36. Koivunen, E., Wang, B., and Ruoslahti, E. (1994) J. Cell Biol. 124, 373-380 37. Levi, M., Sallberg, M., Ruden, U., Herlyn, D., Maruyama, H., Wigzell, H., Marks, J., and Wahren,B. (1993) Proc. Natl. Acad. Sci. (I. S. A. 90,43744378 38. Welling, G. W., Van Gorkum, J., Damhof, R. A,, and Dnjfhout, J. W. (1991) J . Chromatogr. 648,235-242 39. Imura, Y.,Stassen, J-M., Bunting, S.,Stockmans, F., and Collen, D. (1992) Blood 80, 1247-1253 40. Karpatkin, S.,and Pearlstein, E. (1981)Ann. Intern. Med. 96, 636-641 41. Menter, D. G., Steinert, B. W., Sloane, B. F., Gundlach, N., OGara, C. Y., Marnett, L. J., Diglio, C., Walz, D.. Taylor, J. D., and Honn, K. V. (1987) Cancer Res. 47,67514762 42. Acharya, R., Fry, E., Stuart, D., Fox, G., Rowlands, D., and Brown, F. (1989) Nature 337, 704-716 and Wagner, G. (1991) Science 263, 43. Adler, M., Lazarius, R. A., Dennis, M. S., 445448 44. Leahy, D. J., Hendrickson, W. A., Aukhil, I., and Erickson,H. P. (1992) Science 258,987-991