Tailoring in Vitro Selection for a Picomolar Affinity Human Antibody Directed against Vascular Endothelial Growth Factor Receptor 2 for Enhanced Neutralizing Activity*

Vascular endothelial growth factor (VEGF) and its receptors have been implicated in promoting solid tumor growth and metastasis via stimulating tumor-associated angiogenesis. We previously identified several fully human neutralizing anti-VEGF receptor 2 (or kinase inserting domain-containing receptor (KDR)) antibodies from a large antibody phage display library. These antibodies bind specifically to KDR, block VEGF/KDR interaction, and inhibit VEGF-induced proliferation of human endothelial cells and migration of KDR+ leukemia cells. Three of these antibodies, interestingly, share an identical heavy chain variable (VH) sequence. In this report, we constructed a new library comprising the single VH paired with the variable light chain (VL) repertoire obtained from the original naïve human library. Initial in vitro selection revealed that the single VH could pair with a number of different VL while retaining its specificity for KDR. However, a consensus VH/VL pair, clone 1121, was identified after three or four rounds of selection by tailoring the stringency of the panning conditions. Clone 1121 showed a >30-fold higher binding affinity to KDR (Kd, 100 pm) because of improvement on both association and dissociation constants and blocked VEGF/KDR interaction with an IC50 of ∼1 nm, compared with that of 3–4 nm for the parent Fab fragments. Further, clone 1121 was more potent in inhibiting VEGF-stimulated KDR phosphorylation in endothelial cells. A binding epitope mapping study on clone 1121 and one of the parent clones, 2C6, demonstrated that both antibodies interacted with the third immunoglobulin domain within the extracellular region of KDR. Several peptide phage display libraries were utilized to further examine the fine binding specificities of the two antibodies. All of the 2C6-binding peptides are cysteine-constrained, whereas clone 1121 binds to both cysteine-constrained and linear peptides. It is noteworthy that most of the 2C6-binding peptides also cross-react with clone 1121, but none of the clone 1121-specific peptides binds to 2C6, indicating that clone 1121 retained part of the original binding epitope(s) of 2C6 while gaining new binding specificity. Taken together, our observation suggests that clone 1121 may have great clinical potential in anti-angiogenesis therapy. It further underscores the efforts to identify antibodies of high affinity for enhanced biological activities.

Monoclonal antibodies (mAb), because of their high specificity toward a given target, represent a unique class of novel therapeutics as angiogenesis inhibitors (18,19). With the recent progresses in molecular engineering techniques and the availability of human antibody transgenic mice and human antibody phage display libraries, chimeric, humanized, and fully human mAb with desired specificities and properties, e.g. low immunogenecity and high affinity, can be readily produced (20,21). We previously produced a chimeric and several fully human anti-KDR antibodies and demonstrated that these antibodies were capable of blocking KDR/VEGF interaction and inhibiting VEGF-stimulated receptor activation and mitogenesis of human endothelial cells (10,22,23). In addition, these antibodies inhibited VEGF-induced proliferation of human leukemia cells in vitro and prolonged survival of nonobese diabetic-severe combined immunodeficient mice inoculated with human leukemia cells (11,24,25). We observed that three of the neutralizing human anti-KDR antibodies identified from a phage display library share an identical heavy chain variable (VH) sequence (23). In this report, we constructed a new library * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
ʈ To whom correspondence should be addressed: Dept. of Antibody Technology, ImClone Systems Incorporated, 180 Varick St., New York, NY 10014. Tel.: 646-638-5190; Fax: 212-645-2054; E-mail: Zhenping@ imclone.com. comprising the single VH paired with the light chain variable (VL) repertoire obtained from the original phage library and, together with a tailored stringent in vitro selection process, identified a consensus VH/VL pair with ϳ30-fold affinity improvement. The high affinity variant was more potent in blocking KDR/VEGF interaction and in inhibiting VEGF-stimulated receptor activation. Epitope mapping studies using peptide phage display libraries revealed that the high affinity variant gained novel epitope(s) on KDR while retaining part of the original epitope(s) to which the parent antibodies interact.

EXPERIMENTAL PROCEDURES
Cell Lines, Proteins, and Phage Display Libraries-Primary cultured HUVEC were maintained in EBM-2 medium (Clonetics, Walkersville, MD) at 37°C and 5% CO 2 as described previously (10). Porcine aortic endothelial cells transfected with KDR (PAE-KDR) was generated at ImClone Systems Incorporated (New York, NY) and maintained in Dulbecco's modified Eagle's medium containing 10% fetal calf serum. The soluble KDR-alkaline phosphatase fusion protein (KDR-AP), its extracellular Ig domain (ECD) deletion variant-AP fusions, and VEGF 165 proteins were produced as previously described (26). IMC-1C11 and MAB664, two mAb directed against KDR, were generated at ImClone Systems Incorporated as previously described (10,12,27). The human anti-KDR antibodies, clones 2C6, 1H4, and 2H2, were isolated from a large human Fab phage display library (containing 3.7 ϫ 10 10 clones) (28) by several rounds of selection against immobilized recombinant KDR protein as previously described (23). All of the peptide phage display libraries and the Escherichia coli strain K91 were kind gifts from Dr. G. P. Smith of University of Missouri-Columbia (for details see www.biosci.missouri.edu/smithgp/PhageDisplayWebsite/ PhageDisplayWebsiteIndex.html).
Generation of the Light Chain-shuffled Phage Display Library-The original human Fab phage display library (28), from which the anti-KDR antibodies 2C6, 1H4, and 2H2 were identified, was used as the source of the VL repertoire in the shuffled library. The phagemid preparation from the original library was first digested with SfiI and NotI followed by electrophoresis on an agarose gel to separate the VH gene fragments from the antibody light chain-containing backbone vector to delete the entire VH repertoire. The gene encoding the VH domain of clone 2C6/1H4/2H2 was generated by digestion of 2C6-coding phagemid with the same restriction enzymes and purified as described above. The 2C6 VH-coding gene was then ligated into the purified backbone vector to create the VL-shuffled Fab repertoire. E. coli TG1 cells were transformed with the ligation mixtures via electroporation (Electro Cell Manipulator 600, BTX Electroporation System, San Diego, CA). The transformed TG1 cells were growth with shaking in 2YT medium at 37°C for 60 min and plated onto several 150-mm 2YT agar plates containing 100 g/ml ampicillin and 2% glucose (2YTAG plates). After incubation overnight at 30°C, all of the colonies grown on the plates were scraped into 5 ml of 2YTAG medium, mixed with 1.2 ml of 50% glycerol (final concentration, 10%), aliquoted, and stored at Ϫ70°C as the library stock.
Selection of the VL-shuffled Library against KDR-The library stock (100 l) was grown to log phase in 20 ml of 2YTAG medium, rescued with M13K07 helper phage, and amplified overnight in 2YTAK medium (2YT containing 100 g/ml of ampicillin and 50 g/ml of kanamycin) at 30°C. The phage preparation was precipitated in 4% PEG, 0.5 M NaCl, resuspended in 1 ml of 3% fat-free milk/PBS containing 500 g/ml of AP protein and incubated at 37°C for 1 h to block nonspecific binding. For first round selection, KDR-AP-coated Maxisorp Star tubes (Nunc, Rosklide, Denmark) were first blocked with 3% milk/PBS at 37°C for 1 h and then incubated with the phage preparation at room temperature for 1 h. The tubes were washed 15 times with PBST (PBS containing 0.1% Tween 20) followed by 15 washes with PBS. The bound phage was eluted at room temperature for 10 min with 1 ml of a freshly prepared solution of 100 mM triethylamine (Sigma). The eluted phage were incubated with 10 ml of mid-log phase TG1 cells at 37°C for 30 min stationary and 30 min shaking. The infected TG1 cells were pelleted, plated onto several large 2YTAG plates, and incubated overnight at 30°C. All of the colonies grown on the plates were scraped into 3-5 ml of 2YTA medium, mixed with glycerol (final concentration: 10%), aliquoted, and stored at Ϫ70°C. For subsequent rounds of selection, 100 l of the phage stock was added to 20 ml of 2YTAG medium and grown to mid-log phase. The culture was rescued with M13K07 helper phage and amplified overnight in 2YTAK medium, and phage-containing supernatant was used directly (without PEG precipitation) for se-lection followed the procedure described above, with conditions of various stringency as described in Table I.
Phage-based KDR Binding and Blocking ELISA-Individual TG1 clones recovered after each round of selection were randomly picked and grown at 37°C in 96-well plates and rescued with M13K07 helper phage as described above. The amplified phage preparation was blocked with 1 ⁄6 volume of 18% milk/PBS at room temperature for 1 h and added to Maxi-sorp 96-well microtiter plates (Nunc) coated with KDR-AP or AP (1 g/ml ϫ 100 l at 4°C overnight). After incubation at room temperature for 1 h, the plates were washed three times with PBST and incubated with a mouse anti-M13 phage-HRP conjugate (Amersham Biosciences). The plates were washed five times, TMB peroxidase substrate (KPL, Gaithersburg, MD) was added, and the absorbance at 450 nm was read using a microplate reader (Molecular Device, Sunnyvale, CA).
For KDR/VEGF blocking ELISA, 100 l of the blocked phage were mixed with 100 ng of KDR-AP and incubated at room temperature for 1 h, after which the mixture were transferred to 96-well plates coated with VEGF (200 ng/well) at incubated for additional 2 h. The plates were washed five times with PBST, and the substrate for AP (p-nitrophenyl phosphate; Sigma) was added, followed by reading the absorbance at 405 nm to quantify the bound KDR-AP molecules (10) DNA Pattern Analysis and Nucleotide Sequencing-The diversity of the VL-shuffled library and the anti-KDR Fab clones recovered after each round of selection was analyzed by restriction enzyme digestion patterns (i.e. DNA fingerprints) using a frequent cutting enzyme, BstNI (10,23). DNA sequences of representative clones from each digestion pattern were determined by dideoxynucleotide sequencing. Classification and alignments of the VH and the VL genes were performed using NCBI IgBlast (www.ncbi.nlm.nih.gov/igblast) and V base (www. mrc-cpe.cam.ac.uk).
Expression and Purification of the Soluble Fab Fragments and the Full-length IgG-Phagemids of the individual selected clones were used to transform a nonsuppressor E. coli host HB2151. Expression of the Fab fragments in HB2151 and purification of the soluble Fab proteins from the periplasmic extracts of the E. coli were carried out as previously described (10,23). The IgG1 forms of clones 2C6, 1H4, and 1121 (IMC-2C6, IMC-1H4, and IMC-1121, respectively) were constructed and produced in NS0 cells under serum-free conditions following procedures previously described (11).
Quantitative KDR Binding and Blocking Assay-In the direct binding assay, various amounts of anti-KDR antibodies were added to KDR-coated 96-well Maxi-sorp microtiter plates and incubated at room temperature for 1 h, after which the plates were washed three times with PBST. The plates were then incubated at room temperature for 1 h with 100 l of a rabbit anti-human Fab antibody-HRP conjugate (Jackson ImmunoResearch Laboratory Inc., West Grove, PA). The plates were washed and developed, and the absorbance at 450 nm was read following the procedure described above for the phage ELISA. In the competitive KDR/VEGF blocking assay, various amounts of anti-KDR antibodies were mixed with a fixed amount of KDR-AP (100 ng) and incubated at room temperature for 1 h. The mixture were then transferred to 96-well microtiter plates precoated with VEGF165 (200 ng/ well) and incubated at room temperature for an additional 2 h, after which the plates were washed five times, and the substrate for AP was added, followed by reading the absorbance at 405 nm to quantify the bound KDR-AP molecules (10). IC 50 , i.e. the antibody concentration required for 50% inhibition of KDR binding to VEGF, was then calculated.
Antibody Affinity Determination by BIAcore Analysis-The binding kinetics of various antibodies to KDR was measured using a BIAcore 3000 biosensor and evaluated using the program BIA Evaluation 2.0 (Biacore, Inc., Uppsala, Sweden). The affinity constant, K d , was calculated from the ratio of dissociation rate (k off )/association rate (k on ) (10).
KDR Phosphorylation Assay-The assay was carried out on both HUVEC and PAE-KDR cells following a protocol previously described (10,22). Briefly, the serum-starved cells were incubated with various amounts of antibodies at room temperature for 30 min, followed by stimulation with 20 ng/ml of VEGF165 for additional 15 min. The cells were lysed, and KDR protein was precipitated from the cell lysates with a polyclonal anti-KDR antibody (ImClone Systems) followed by protein A-Sepharose beads (Santa Cruz Biotechnology, Santa Cruz, CA). The proteins were resolved with SDS-PAGE and subjected to immunoblotting analysis using an anti-phosphotyrosine antibody-HRP conjugate (Santa Cruz Biotechnology, Santa Cruz, CA). The signals were detected using enhanced chemoilluminescence (Amersham Biosciences).
Antibody Epitope Mapping Using KDR Ig Domain Deletion Variants-Full-length KDR (KDR(Ig1-7)) and KDR Ig domain deletion vari-ants, including KDR(Ig1-3) and KDR (Ig3), the deletion variants that contain KDR ECD N-terminal Ig domains 1-3 and 3 only, respectively, were first immobilized onto a 96-well plate (Nunc). The plate was then incubated with various anti-KDR antibodies at room temperature for 1 h, followed by incubation with a mouse anti-human Fc-HRP conjugate (for IMC-1C11, IMC-2C6, and IMC-1121) or a rat anti-mouse Fc-HRP conjugate (for MAB664 only). The plate was washed and developed as described above.
Antibody Epitope Mapping Using Peptide Phage Display Libraries-A total of eight randomized pVIII phage-displayed peptide libraries were used to dissect out the fine binding specificities of IMC-2C6 and IMC-1121, including a linear 15-mer library and 7 cysteine-constrained 15-mer libraries with the configuration of -X n CX 0 -6 CX n -(n ϭ 3-7). Two parallel selections were carried out using either the single 15-mer linear peptide library or the mixture of all of the 7 cysteine-constrained peptide libraries. Briefly, 96-well ELISA plates were coated with 600 ng of IMC-2C6 or IMC-1121 in 150 l of PBS overnight and blocked with 3% (w/v) skim milk/PBS at 37°C for 1 h. The peptide-phage library preparation (ϳ1-2 ϫ 10 11 transforming units in 100 l of PBS) were mixed with 50 l of 9% milk/PBS, added into the antibody-coated wells, and incubated at room temperature for 1 h. A nonspecific human IgG was added in the solution (1 mg/ml) to absorb IgG Fc-binding phages. The plate was washed 15 times with PBS/Tween 20 (0.1%), and the bound phages were eluted by incubation with 150 l of HCl with bovine  serum albumin (0.1 M HCl, 1 mg/ml bovine serum albumin, adjusted to pH 2.2 with glycine) at room temperature for 10 min, followed by neutralization with 25 l of 1 M Tris-HCl (pH 9.0). Recovered phages were then rescued by infecting 75 l of mid-log phase E. coli K91 cells at room temperature for 15 min, followed by incubation at 37°C for 45 min. The infected cells were amplified in 4 ml of 2YT/tet (tetracycline at 40 g/ml) medium at 37°C overnight, followed by PEG precipitation and resuspension in PBS for the subsequent round of selection. For the second and third round selections, 300 and 150 ng, respectively, of antibodies were used to coat the microwells of the 96-well plate.
Identification of Antibody-binding Peptide Phages-ELISA was used to identify peptide phage clones that bind to IMC-2C6 or IMC-1121 after three rounds of selection. Individual phage clones were inoculated in 200 l of 2YT/tet/isopropyl-␤-D-thiogalactopyranoside (1 mM) in 96well plates and grown at 37°C overnight. After centrifugation at 2,000 rpm for 10 min, the culture supernatants were mixed with 1 ⁄3 volume of 9% milk/PBS and incubated at room temperature for 60 min. Aliquots of 50 l of the mixture were transferred to 96-well plates coated with an anti-M13 phage-antibody (1 g/ml) and incubated at room temperature for 60 min to capture the phage particles on the plate. IMC-2C6 or IMC-1121 antibody were added to the plate at 1 g/ml, and the amounts of bound antibodies were detected by an anti-human Fc antibody-HRP conjugate as described. A total of two to three dozen positive clones were randomly picked and subjected to DNA sequencing using the primer 5Ј-CCTGTTGACAATTAATCATCGGCTC-3Ј.
Dose-dependent Antibody Binding and Blocking Assays-To produce phage preparations for quantitative binding and blocking assays, individual selected clones were inoculated into 5 ml of 2YT/tet and incubated at 37°C with shaking for 5 h. The cell cultures were then transferred into 20 ml of 2YT/tet/isopropyl-␤-D-thiogalactopyranoside in 250-ml flasks and cultured overnight. Phage were purified twice by PEG precipitation, and the phage pellets were resuspended in PBS at final concentration of ϳ10 13 phages/ml. The binding assay was carried out following the procedure described above, starting at 1:10 phage dilution (in 3% milk/PBS) and followed by 2ϫ titration up to 1:1280. In the blocking assay, various amounts of phages were mixed with 10 l of KDR-AP (2.5 g/ml) first and then added to 96-well plates coated with IMC-2C6 or IMC-1121. After incubated at room temperature for 1 h, the amounts of bound KDR-AP were detected by the addition of AP substrate as previously described (10,22).

RESULTS
Construction of the VL-shuffled Library-In constructing the VL-shuffled library, the single VH gene segment shared by the three anti-KDR antibodies, 2C6, 1H4, and 2H2, was subcloned into the large Fab phage display library replacing the original diverse VH repertoire. After electroporating into TG1 cells, a library of 1.5 ϫ 10 8 plaque-forming units was obtained. In this library, the single VH gene is randomly paired with, in theory, 1.5 ϫ 10 8 different VL genes, thus creating the same numbers of independent Fab binding specificities. DNA fingerprinting and nucleotide sequencing of several dozen randomly picked colonies revealed no identical patterns and nucleotide se-quences between different clones and between these clones and the three parent clones, indicating an excellent diversity of the shuffled library.
Selection of High Affinity Anti-KDR Clones-Two different selection processes were carried out in parallel using conditions described in Table I. A total of four rounds of selection were performed: the first two in Maxi-sorp star tubes and the last two in 96-well plates. After initial phage binding to immobilized KDR, the tubes or wells were washed 15 times with PBST (PBS containing 0.1% Tween 20) followed by another 15 times with PBS. Approximately 59% of the phage clones recovered after the first round selection were specific KDR binders. In the subsequent selections, conditions of varying stringency were applied. First, to select anti-KDR clones with fast association rate (or on rate, k on ), the number of input phages were reduced in each subsequent selections, and the duration of initial binding processes between the phages and the immobilized antigen was shortened from 1 h in the first round to 0.5 h in the second and the third rounds to 0.2 h in the fourth round. Further, a process aiming to select clones with improved dissociation rate (or off rate, k off ) was also applied; after the initial binding, the KDR-bound phages were subjected to extensive period of washes, from 0.5 h in the second round, 3 h in the third round to 24 h in the fourth round selection, with PBS or various amounts of KDR (from 100 to 2000 nM) in solution as a competitor (see Table I for details).
After each round selection, 96 phage clones were picked randomly and assayed for KDR binding activity. The KDRpositive rates in the PBS-washed group were 34.4, 21.1, and 81.1% after the second, third, and fourth round selections, respectively, which are comparable with that of 32.2, 22.2, and 63.3%, respectively, in the KDR-washed group (Table I). Ten to twenty clones recovered from each round selection were analyzed by DNA sequencing. In the PBS-washed group, 12 of 20 clones (60%) sequenced after the third round selection were clone 2C6, one of the parent clones. A total of seven new sequences were identified, with one sequence, designated as clone 1121, repeated once (Table I). Five different sequences were recovered after the fourth round selection; only one of ten clones (10%) sequenced was 2C6. Six of ten sequences (60%) were clone 1121, and one clone (4B4) was also seen after the third round selection (clone 3B10). In the KDR-washed group, only three different sequences were recovered after the third round selection; twelve of twenty clones (60%) were 2C6, and seven of twenty (35%) were clone 1121. All of the clones (10/10) sequenced after the fourth round selection were clone 1121 (Table I). In total, ten new anti-KDR VH/VL pairs were iden- tified from the VL-shuffled library after three or four rounds of selection ( Fig. 1 and Table I).
Sequence alignments using NCBI IgBlast and V Base of the single VH domain showed that the gene has the highest homology with the germline DP77 segment of the human VH3 family, with only one amino acid substitution from the germline sequence (Table II). Interestingly, alignments of the VL domains revealed both V and V gene families; two of the original clones, 2C6 and 2H2, along with four newly identified clones including 1121 belong to V, whereas the other clones, including the original clone 1H4 and six new clones, are from the V family. It is noteworthy that all the six V genes have the same canonical structure for L1, L2, and L3 (2-1-1), whereas the six newly selected V genes share identical length of both CDR1 and CDR2 (14 and 7 amino acids, respectively) (30, 31). Taken together, these observations underscore the structure requirement on VL genes for pairing with the fixed VH segment to maintain KDR binding specificity.
KDR binding and blocking activities of all ten positive clones were examined using phage-based ELISA. Compared with the parent clone 2C6, six new clones demonstrated moderate to strong enhancement in KDR-binding activity, with clone 1121 being the best binder ( Fig. 2A). However, only three clones, 1121, 3A1, and 4C7, were better blockers to KDR/VEGF interaction than clone 2C6 (Fig. 2B). Clone 1121 was therefore selected as the leading candidate for further studies.
KDR Binding and KDR/VEGF Blocking by the Anti-KDR Antibodies-Soluble Fab fragments of all the three parent anti-KDR clones, 2C6, 1H4, and 2H2, and the consensus clone 1121 were produced and compared quantitatively in their antigen

FIG. 2. Binding to KDR and blocking KDR/VEGF interaction by various anti-KDR phage clones recovered after three to four rounds of selection.
A total of ten different KDR-binding VH/VL pairs were recovered and assayed by phage-based ELISA. Clone 3B10 is identical to clone 4B4. A, phage supernatant obtained from overnight culture in 96-well plates were added to microwells coated with KDR-AP and incubated at room temperature for 1 h, after which the plates were incubated with a mouse anti-M13 phage antibody-HRP conjugate. The plates were washed, peroxidase substrate was added, and A 450 nm was read. B, inhibition of binding of KDR to immobilized VEGF by the anti-KDR phages. The phage supernatants were incubated with a fixed amount of KDR-AP at room temperature for 1 h, after which the mixtures were transferred to 96-well plated coated with VEGF and incubated for an additional 2 h. The amount of KDR-AP that bound to the immobilized VEGF was quantified by incubation of the plates with AP substrate and reading of A 405 nm .
binding efficiency and potency in blocking KDR/VEGF interaction. Consistent with the phage binding result ( Fig. 2A), clone 1121 Fab binds much more efficiently to KDR than all three parent Fab (Fig. 3A). It is also a much stronger blocker to KDR/VEGF interaction (Fig. 3B), with an IC 50 value of ϳ1 nM, compared with that of 3-5 nM for 2C6, 1H4, and 2H2 Fab. Three Fab fragments, 2C6, 1H4, and 1121, were converted into the full-length bivalent IgG1 format. IMC-1121 again is the strongest KDR-binder and the most potent blocker of KDR/ VEGF interaction (Fig. 4). The IC 50 values for blocking KDR binding to VEGF were 0.8, 1.2, and 6.2 nM, for IMC-1121, IMC-2C6, and IMC-1H4, respectively (Fig. 4B).
The binding kinetics of various anti-KDR antibodies were determined by surface plasmon resonance on a BIAcore instrument (Table III). As seen previously, the three parent Fab fragments bind to immobilized KDR with a similar overall affinity (K d ) of 3.1-4.2 nM, although each individual Fab showed different k on and k off rates; clone 2C6 has the fastest k on and the fastest k off , whereas clone 1H4 possesses the slowest k on and also the slowest k off among the three fragments. Clone 1121 Fab possesses an affinity of 100 pM, representing Ն30-fold improvement over the parent clones, because of improvement on both k on and, more remarkably, k off . As a bivalent IgG, IMC-1121 showed an overall affinity of 50 pM, representing a ϳ4-fold improvement over IMC-2C6, and 12-fold improvement over IMC-1H4 (Table III).
Inhibition of VEGF-stimulated KDR Phosphorylation-The biological activity of IMC-1121 and IMC-2C6 was evaluated in a VEGF-stimulated KDR phosphorylation assay using both HUVEC and PAE-KDR cells. VEGF stimulation resulted in significant KDR phosphorylation in both HUVEC and PAE-KDR cells (Fig. 5). Neither antibody alone had any effects on KDR phosphorylation. In the presence of VEGF, both antibodies neutralized KDR phosphorylation in a dose-dependent manner. Consistent with its potent blocking activity, IMC-1121 was a much more stronger (Ͼ5-25-fold) inhibitor in both cell assays than IMC-2C6 (Fig. 5).
BIAcore analysis was employed to examine whether IMC-2C6 and IMC-1121 competes with each other or with other neutralizing anti-KDR antibodies for binding to KDR ECD. In this assay, an antibody was first injected onto a KDR-coated chip at high concentrations to saturate all of the receptor immobilized on the chip, which was followed by injection of a second antibody. An increase in binding density (as measured by reference units) to the KDR-coated chip upon the injection of the second antibody indicates no competition for binding between the two antibodies, i.e. the two antibodies are most likely bind to different, nonoverlapping epitope(s) on KDR, as seen in the case between MAB664 and IMC-1C11 (Table IV; also see Ref. 26). Binding of IMC-2C6 to the KDR-coated chip blocked further binding by IMC-1121 and vice versa, suggesting the two antibodies interact with either the same or overlapping epitope(s) within KDR ECD Ig domain 3. Both IMC-2C6 and IMC-1121 also blocked subsequent binding by IMC-1C11 but not by MAB664. Because IMC-1C11 requires KDR ECD Ig domain 1-3 for binding, this observation suggests that IMC-1C11 may share an overlapping epitope(s) with both IMC-2C6 and IMC-1121 within Ig domain 3. It is noteworthy that the overall KDR binding reference units increased when IMC-1C11 was followed by either IMC-2C6 or IMC-1121, but there was no such increase when IMC-2C6 or IMC-1121 was followed by IMC-1C11. This phenomenon is most likely a reflection that both IMC-2C6 and IMC-1121 are stronger KDR binder and that they replace KDR chip-bound IMC-1C11 when injected over the chip at high concentration.
We then used a panel of peptide phage display libraries to   (Table V). Of note, all IMC-2C6-selected clones, including the three clones (clones 2D5, 1A3, and 1A5) identified from the randomized 15-mer linear peptide library, are cysteine-constrained peptides, whereas IMC-1121 binders include both linear and cysteineconstrained peptides. Eight of the 11 IMC-2C6-selected peptides have the same XCX 5 CX motif, with the consensus motif ECP(H/ P/S)X 2 RC(P/Q). IMC-1121-selected peptides were structurally more diverse; of 15 sequences obtained there were five linear peptides, four peptides with XCX 5 CX, three with XCX 4 CX, two with XCX 9 CX, and one with XCX 3 CX motif (Table V). and incubated at room temperature for 1 h. The plate was washed and then incubated with a mouse anti-human Fc antibody-HRP conjugate for additional 1 h, after which the plate was developed as described above. Antibody binding to the two Ig deletion variants are shown as relative (as percentages) to their binding to the full-length KDR ECD.

FIG. 6. Inhibition of VEGF-stimulated KDR phosphorylation in both HUVEC and PAE-KDR cells by fulllength anti-KDR antibodies.
Serumstarved cells were incubated with various amounts of antibodies at room temperature for 30 min, followed by stimulation with 20 ng/ml of VEGF165 for additional 15 min. The cells were lysed, and KDR protein was precipitated from the cell lysates with a polyclonal anti-KDR antibody followed by protein A-Sepharose beads. The proteins were resolved with SDS-PAGE and subjected to immunoblotting analysis using an anti-phosphotyrosine antibody. The signals were detected using enhanced chemoilluminescence. The numbers represent the number of reference units bound to the KDR-coated chip upon the injection of the first antibody followed by the number of reference units bound to the chip after the injection of the second antibody and are the representatives of three separate experiments.
Based on their frequency of occurrence and binding efficiency, four peptide phage clones were selected for each antibody for further analysis. The phage clones were amplified and assayed quantitatively for their efficiency in binding to both IMC-2C6 and IMC-1121 and their ability in blocking the antibodies from binding to their cognate antigen, KDR. As shown in Fig. 7 (top panels), all four IMC-1121-selected phage clones bound specifically to IMC-1121 without any cross-reactivity to IMC-2C6. On the other hand, all the four IMC-2C6-selected phage clones reacted with both IMC-2C6 and IMC-1121. Western blot analysis confirmed the same reactivity pattern of these phage clones (not shown). In another assay, the peptide phages were used to compete with soluble KDR for binding to immobilized antibodies. Consistent with the antibody binding results, all eight peptide phages were able to compete with KDR for binding to IMC-1121, but only IMC-2C6-selected phages blocked KDR/IMC-2C6 interaction (Fig. 7, bottom panels). A control cysteine-constrained phage clone, 9C10 (YCDPRCM-RESKYVIS), did not bind to IMC-2C6 and IMC-1121, nor did it block the antibodies from binding to KDR. DISCUSSION We previously identified four fully human anti-KDR antibodies from a large human naïve antibody library and demonstrated that these antibodies were able to block KDR/VEGF interaction and neutralize VEGF-induced angiogenic activity (23). Three of these antibodies share, interestingly, an identical VH sequence, suggesting that this single VH domain may pair with different VL domains while preserving KDR binding/ blocking activities. In this report, we successfully improved the binding affinity of these anti-KDR antibodies by Ͼ30-fold through VL shuffling in association with a tailored selection process. Several observations are noteworthy in our study. First, by following a tailored stringent selection procedure, we were able to identify a consensus VL/VH pair, clone 1121, that showed a picomolar binding affinity because of significant improvement on both the association (k on ) and dissociation (k off ) rates. Second, by shuffling against a large VL repertoire, we identified ten different VL sequences (in addition to the original three VL sequences) that could pair with the single VH domain while preserving KDR binding specificity. Consistent with our previous study, these different VL/VH pairs bind KDR and block KDR/VEGF interaction with varying efficiency. It is interesting to note that both the V and the V families are utilized almost equally among the anti-KDR VL/VH pairs, and of the three best KDR/VEGF blocking clones, two (clone 1121 and 4C7) belong to V, and one (clone 3A1) belongs to V. Third, our epitope mapping studies revealed that although clone 1121 competes with 2C6 for binding to certain overlapping epitope(s) on KDR ECD Ig domain 3, the new variant also gained novel binding specificity that is distinct from 2C6. Taken together, these observations suggest that in all of the anti-KDR VL/VH pairs, the single VH domain likely plays a dominate role in determining KDR binding specificity, whereas the VL domains, on the other hand, may contribute to both the binding strength and the fine epitope specificity via influencing the overall folding/presentation of the antigen-binding surface of the antibodies. It is pertinent to note that VL sharing has also been reported in a earlier study among a panel of phagederived human antibodies directed against several totally unrelated antigens (32), again indicating the dominating role of VH domains in the binding specificities of these antibodies.
A number of approaches have been developed and utilized successfully to improve antibody binding affinity. These approaches usually start with the construction of mutagenized antibody libraries and followed by an array of selections to identify variants with improved affinity. A variety of methods, including randomized mutations (33)(34)(35), designed mutations such as "hot spot" and "codon-based" mutations (guided by structural knowledge and/or computer modeling) (36 -38) and "CDR walking" (39) have been used to construct the mutagenized libraries. These libraries were then displayed on the surface of filamentous phage (40,41), bacterial (42)(43)(44), yeast (45,46), or ribosome (47)(48)(49) and selected on relevant targets under conditions specifically designed for the enrichment of variants with desired properties (i.e. high affinity) (for review, see Ref. 50). A common practice in these previous studies is that most of them were focused on selection of variant(s) with an improved k off rate. Selection for improvement on k on rate, on the other hand, was largely neglected. Because the overall binding affinity (K d ) of an antibody equals the ratio of k off over k on , we reasoned that it should be of significant benefit if we could develop a strategy that would enable us to simultaneously select variants with improvement on not only k off but also k on .
The three original neutralizing anti-KDR Fab, clones 2C6, 2H2, and 1H4, possess similar overall binding affinity (K d ); however, their individual binding kinetics, i.e. k on and k off , are significantly different. For example, clone 2C6 Fab has the fastest k on and also the fastest k off rates, whereas clone 1H4 FIG. 7. Quantitative binding to anti-KDR antibodies and blocking KDR/antibody interaction with selected peptide phage clones. Amplified phages were purified by PEG precipitation and resuspended in PBS at final concentration of ϳ10 13 phages/ml. In the binding assay, various amounts of phages were added to 96-well plates coated with an anti-M13 phage antibody to capture the peptide phage particles, followed by incubation with IMC-2C6 or IMC-1121 at room temperature for 1 h, after which the plates were incubated with a mouse anti-human Fc antibody-HRP conjugate. The plates were washed, peroxidase substrate was added, and A 450 nm was read. In the blocking assay, the phages were mixed with 25 ng of KDR-AP first and then added to 96-well plates coated with IMC-2C6 or IMC-1121. After incubation at room temperature for 1 h, the amounts of bound KDR-AP were detected by the addition of AP substrate and reading of A 405 nm . Note that all four IMC-2C6-selected phage clones (solid lines) also bind to IMC-1121 and block KDR/IMC-1121 interaction, whereas none of the IMC-1121-selected clones (dotted lines) bind to IMC-2C6 or block KDR/IMC-2C6 interaction. 9C10, the control phage clone, does not bind to IMC-2C6 and IMC-1121 nor block the antibodies from binding to KDR.
Fab has the slowest k on and k off rates (Table III). These observations indicate that VL domains have significant influence on the binding kinetics of these anti-KDR antibodies, which share an identical VH domain. In this report, we therefore constructed a large library by shuffling the single VH domain shared by the three anti-KDR antibodies against the entire VL repertoire from the original naïve human library in an attempt to identify the ideal VH/VL pair(s) with improved binding kinetics (i.e. faster k on and slower k off ). The VL-shuffled library was selected on immobilized KDR protein under conditions of increasing stringency that were designed for both k on and k off selection; the on rate selection was achieved by reducing the numbers of input phages and by shortening the initial binding period, whereas the off rate selection was accomplished by extending the washing process (up to 24 h) in the presence of a high concentration of the soluble receptor protein. This combinational selection resulted in the identification of a consensus VL/VH pair (clone 1121) with a K d of 100 pM, representing Ͼ30-fold improvement over the parent Fab fragments. It is pertinent to note that the enhanced affinity of clone 1121 was a result of improvement on both k on and k off rates.
It is generally accepted that to achieve effective tumor binding and antitumor activity, the affinity (K d ) of a mAb needs to be 10 Ϫ8 M or better (51). Although some investigators presume that high affinity antibody is desirable (52), there is considerable debate whether improving mAb affinity beyond 10 Ϫ9 M would result in a further increase in its tumor localization and biological activity (53). Recently, Adams et al. (54) employed several anti-HER2 antibodies of varying affinities derived from the same parent antibody by side-directed mutations and showed a positive correlation between the amount of antibody localized in tumor and antibody affinity up to 10 Ϫ9 M. Further increase in affinity beyond 10 Ϫ9 M did not, however, result in higher tumor localization of the antibody (55). Histological study on the tumor specimens demonstrated that antibodies with moderate affinity (between 10 Ϫ8 and 10 Ϫ9 M) penetrated deeper in to tumors and percolated more homogeneously into tumor mass than antibodies with high affinity (Ͻ10 Ϫ10 M), which were trapped around the perivascular tumor cells with much less localization in the distant tumor tissues (55). The impaired tumor penetrating capacity of antibodies with very high affinity may negatively affect their anti-tumor activity when the antibodies are used as targeting devices to deliver cytotoxic agents to tumor cells within a large tumor mass (53,55). This argument may not be relevant in the context of anti-KDR therapy, because the target antigen (i.e. KDR) is expressed on the surface of vessel-lining endothelial cells, which are in direct contact with the systemically administered antibody (i.e. no tumor penetration is required for antigen access). Antibodies with higher affinity, such as IMC-1121, should be able to bind more tightly to their target for a prolonged period of time, thus efficiently preventing the growth factors, such as VEGF, from interacting to their receptors. In this setting, high affinity antibodies are therefore likely to be more effective agents than those with low affinity, hence the preferable choice for anti-angiogenesis therapy. Here we demonstrated that the high affinity antibody, IMC-1121, is indeed more potent in inhibiting KDR/VEGF interaction and in neutralizing VEGF-stimulated KDR activation than its parent antibodies. Taken together with our previous observation that IMC-1121 was more potent than IMC-2C6 in inhibiting KDR ϩleukemia growth in an animal model (11), these results lend strong support to IMC-1121 for potential clinical evaluation as an anti-angiogenic/antitumor agent.