Cross-species Vascular Endothelial Growth Factor (VEGF)-blocking Antibodies Completely Inhibit the Growth of Human Tumor Xenografts and Measure the Contribution of Stromal VEGF*

To fully assess the role of VEGF-A in tumor angiogenesis, antibodies that can block all sources of vascular endothelial growth factor (VEGF) are desired. Selectively targeting tumor-derived VEGF overlooks the contribution of host stromal VEGF. Other strategies, such as targeting VEGF receptors directly or using receptor decoys, result in inhibiting not only VEGF-A but also VEGF homologues (e.g. placental growth factor, VEGF-B, and VEGF-C), which may play a role in angiogenesis. Here we report the identification of novel anti-VEGF antibodies, B20 and G6, from synthetic antibody phage libraries, which block both human and murine VEGF action in vitro. Their affinity-improved variants completely inhibit three human tumor xenografts in mice of skeletal muscle, colorectal, and pancreatic origins (A673, HM-7, and HPAC). Avastin, which only inhibits the tumor-derived human VEGF, is ∼90% effective at inhibiting HM-7 and A673 growth but is <50% effective at inhibiting HPAC growth. Indeed, HPAC tumors contain more host stroma invasion and stroma-derived VEGF than other tumors. Thus, the functional contribution of stromal VEGF varies greatly among tumors, and systemic blockade of both tumor and stroma-derived VEGF is sufficient for inhibiting the growth of tumor xenografts.

To fully assess the role of VEGF-A in tumor angiogenesis, antibodies that can block all sources of vascular endothelial growth factor (VEGF) are desired. Selectively targeting tumor-derived VEGF overlooks the contribution of host stromal VEGF. Other strategies, such as targeting VEGF receptors directly or using receptor decoys, result in inhibiting not only VEGF-A but also VEGF homologues (e.g. placental growth factor, VEGF-B, and VEGF-C), which may play a role in angiogenesis. Here we report the identification of novel anti-VEGF antibodies, B20 and G6, from synthetic antibody phage libraries, which block both human and murine VEGF action in vitro. Their affinity-improved variants completely inhibit three human tumor xenografts in mice of skeletal muscle, colorectal, and pancreatic origins (A673, HM-7, and HPAC). Avastin, which only inhibits the tumor-derived human VEGF, is ϳ90% effective at inhibiting HM-7 and A673 growth but is <50% effective at inhibiting HPAC growth. Indeed, HPAC tumors contain more host stroma invasion and stroma-derived VEGF than other tumors. Thus, the functional contribution of stromal VEGF varies greatly among tumors, and systemic blockade of both tumor and stroma-derived VEGF is sufficient for inhibiting the growth of tumor xenografts.
Targeting angiogenesis, the process of new blood vessel formation, has been validated as an effective approach for cancer therapy by the recent Food and Drug Administration approval of bevacizumab (Avastin TM antibody), a blocking antibody against vascular endothelial growth factor A (VEGF-A or VEGF), 5 for the treatment of metastatic colorectal cancer (1,2). Although many pro-or anti-angiogenic factors have been implicated in tumor angiogenesis (3,4), the efficacy of a specific blocking antibody against VEGF in suppressing cancer progres-sion emphasizes the key role of VEGF. However, as VEGF is a member of a family that also includes VEGF-B, VEGF-C, VEGF-D, and placental growth factor (PlGF) with overlapping activities in three tyrosine kinase receptors, VEGFR1 (Flt-1), VEGFR2 (KDR, or Flk-1 in mouse), and VEGFR3 (5,6), the questions whether VEGF-A has a nonredundant role in tumor angiogenesis and whether blocking VEGF-A is sufficient to inhibit tumor growth are fundamental.
Systemic blockade using specific blocking antibodies against tumorderived VEGF (7) or genetic deletion of VEGF in tumor cells (8 -10) severely suppressed tumor growth in mouse models. However, residual growth or resistance of human tumor xenografts in nude mice treated with antibodies against human VEGF has been reported (7,11). One possibility is that residual tumor angiogenesis and growth are supported by murine VEGF, which is produced by host stromal cells recruited into the tumor (11). Significant VEGF expression has been demonstrated by fibroblast and immune cells that surround and invade the tumor mass (12)(13)(14)(15)(16), although the significance and extent of the contribution of such stroma-derived VEGF to tumor growth remain unclear. Other possible explanations for incomplete tumor growth inhibition following treatment with anti-VEGF antibodies include the following: insufficient binding affinity or tumor penetration; VEGF-related molecules (e.g. PlGF and VEGF-C) that may act independently or in concert with VEGF-A in promoting angiogenesis (17); VEGF-independent angiogenesis pathways; and "vascular mimicry" by tumor cells (18). To address these issues, some of us (11) and others (19) have used the soluble VEGF receptor extracellular domain (ECD) fragment, such as VEGFR1 Ig domain 1-3 Fc fusion (VEGFR1 D1-3 -Fc) or VEGF trap (VEGFR1 D2 -VEGFR2 D3 -Fc), and showed more complete inhibition of human xenografts relative to antibodies against tumor-derived VEGF. It was reasoned that these receptor decoys can block both human and murine VEGF (11) or that the receptor fragments have higher affinity than the available anti-h-VEGF antibody (19). However, these receptor fragments may also block other VEGF family members, i.e. VEGF-B and PlGF. Although gene deletion of VEGF-B (20) or PlGF (21) was well tolerated in embryonic and postnatal developments, suggesting that these factors are not essential for developmental angiogenesis, recent studies have shown that both VEGF-B and PlGF can contribute to pathological angiogenesis (21). Therefore, the specific contribution of VEGF-A cannot be defined by using these receptor decoys. As for affinity, direct comparison of anti-VEGF antibody with receptor decoy was not meaningful because of their different specificities. Other inhibitors such as inhibitors against VEGFR2 also cannot reveal the specific role of VEGF-A because VEGF-C and VEGF-D can also activate VEGFR2.
To clarify these issues, antibodies that can directly and specifically block VEGF of both tumor and stromal sources would be the tool to examine whether blocking VEGF alone is sufficient for complete inhibition of tumor growth or whether the binding affinity of the antibody affects the potency and efficacy. Furthermore, a comparison between cross-reactive antibodies and antibodies that block only tumor-derived VEGF may reveal the level of dependence on stromal VEGF for tumor growth. VEGF-A is a member of the homodimeric cysteine knot protein family growth factor (6). Two tyrosine kinase receptors mediate the signal activity of VEGF, VEGFR1 (Flt-1) and VEGFR2 (KDR), with the latter as the main receptor mediating most of the endothelial cell action as clearly demonstrated by VEGF variant and homologues that only can bind VEGFR2 (22,23). The role of VEGFR1 is less clear, but it has high affinity for VEGF-A, VEGF-B, and PlGF, with the latter two being VEGFR1-specific ligands. Structural and functional studies of VEGF receptor binding domain in complex with Avastin Fab (Fab-12) (24,25) and affinity-improved Avastin, Y0317 (26), and the primary binding domain of VEGFR1, the second Ig domain (VEGFR1 D2 ) (27), revealed that the VEGF epitope for Avastin and for VEGFR1 or VEGFR2 have sufficient overlap for mutual exclusion. Human and murine VEGFs are 85% identical in sequence. More than 10 hybridoma antibodies against human VEGF were generated, but none can bind murine VEGF, which is expected as mice remove self-reactive antibodies. The high homology of murine VEGF with rat or hamster VEGF also made it difficult to generate the desired monoclonal antibodies in these animals.
To identify antibodies that cross-react with murine and human VEGFs, we employed phage-displayed synthetic antibody libraries, which were generated by randomizing the complementarity determining regions (CDRs) of heavy chain in the humanized anti-ErbB2 antibody 4D5 (h4D5-8) template in a manner that mimicked the diversity of natural human antibodies (28). Humanized 4D5-8 contains the commonly used framework (V H -3/V k -1); thus antibodies isolated from the libraries would mimic natural human antibodies. Another main advantage of antibody generation with phage libraries is its in vitro selection process that should be indifferent to the cross-species conservation (29,30). In this study, we report the identification of cross-human and murine VEGF binding clones, G6 and B20. We examined the affinityimproved G6 and B20 variants for their binding affinities, specificities, and VEGF blocking activities in comparison to the hybridoma-derived anti-h-VEGF antibody, A4.6.1 (or the humanized version, Avastin), and we identified variants suitable for in vivo studies, G6-31 and B20-4.1. In inhibiting h-VEGF action in vitro, B20-4.1 was equipotent to Avastin, and G6-31 was equipotent to Y0317. The cross-species anti-VEGF antibodies examined the specific role of VEGF in the tumor progression using mouse models.

MATERIALS AND METHODS
Binding Specificity of G6 and B20-Competitive ELISAs were used to determine the relative binding affinity of phage clones displaying anti-VEGF Fab to h-VEGF and m-VEGF as described (28). Briefly, Fab phage were incubated with serial dilutions of h-VEGF or m-VEGF in PBS with 0.5% BSA and 0.05% Tween 20 (PBT) for 2 h, and then unbound Fab phage were captured with VEGF-coated 96-well Maxisorp ELISA plate (Nunc, Roskilde, Denmark) and detected by anti-M13 phage antibody conjugated with horseradish peroxidase (HRP) (Amersham Biosciences) followed by substrate. EC 50 values, the concentrations of VEGF incubated with Fab phage in solution that were effective in binding 50% of Fab phage to block their binding to immobilized VEGF were reported. Fab protein was prepared from Escherichia coli harboring the plasmid of the G6 Fab expression construct under the promoter of alkaline phosphatase and secretion leader sequence stII and was purified with the protein G affinity column as described (28). IgGs were generated with mammalian cell transfection and purified with protein A affinity column. For examining binding specificities, VEGF and various homologues were first coated on an ELISA plate at 2 g/ml in PBS and blocked with 0.5% BSA and 0.05% Tween 20. Serial dilutions of Fabs or IgGs were then incubated in VEGF variant-coated wells for 1 h and measured with anti-human antibody HRP diluted in PBT buffer. The VEGF homologues, mouse and human PlGF-2, human VEGF-C, mouse VEGF-D, human and mouse VEGF-B, were purchased from R & D Systems (Minneapolis, MN). Solution binding assays were also carried out in case some protein became unfunctional upon coating on the plate. Fabs or IgG (0.5 nM) were incubated with VEGF homologues at different concentrations for 1-2 h, and the unbound antibodies were captured with m-VEGF-coated ELISA wells and detected as above. The results were consistent with the direct binding ELISA format.
VEGF Receptor Blocking Assay-Initial blocking assays were performed by phage ELISA. Murine VEGF-coated ELISA wells were first incubated with increasing concentrations of receptor fragments for 30 min, and phage displaying Fabs were then added. Bound phage were measured as described above. Purified Fabs and IgGs were then used to confirm receptor blocking activities. VEGF receptor-immobilized plates were prepared by capturing VEGFR2 ECD (Ig domain 1-7) as Fc␥ fusion with goat anti-human IgG Fc␥ (Jackson ImmunoResearch, West Grove, PA) on the ELISA plate, whereas VEGFR1 ECD fragment (Ig domains 1-5) (VEGFR1 D1-5 ) was coated directly. Biotinylated h-VEGF 165 or m-VEGF 164 (ϳ2 nM) was first incubated with 3-fold serially diluted anti-VEGF antibodies in PBT. After 1-2 h of incubation at room temperature, the mixtures were transferred to a VEGF receptorimmobilized plate and incubated for 10 min. The unblocked VEGF-A was captured with VEGF receptor-coated wells and detected by streptavidin-HRP conjugates as described above.
Anti-VEGF Antibodies Binding Affinities-For binding affinity measurements, surface plasmon resonance measurements with BIAcore TM -3000 (BIAcore, Inc., Piscataway, NJ) were employed. Carboxymethylated dextran biosensor chips (CM5, BIAcore, Inc.) were activated with N-ethyl-NЈ-(3-dimethylaminopropyl)carbodiimide hydrochloride and N-hydroxysuccinimide according to the supplier's instructions. Human or murine VEGF was immobilized to achieve ϳ60 response units. 2-Fold serial dilutions of Fab or IgG (0.78 -500 nM) were injected in PBS with 0.05% Tween 20 (PBST) at 37 or 25°C (data not shown) at a flow rate of 25 l/min. Association rates (k on ) and dissociation rates (k off ) were calculated using one-to-one Langmuir binding model (BIAcore evaluation software version 3.2). The equilibrium dissociation constant (K d ) was derived as the ratio k off /k on .
HUVEC Assay of VEGF Inhibition-Human umbilical vein endothelial cells (HUVEC) (Clontech) were grown and assayed as described (31). Approximately 4000 HUVECs were plated in each well of the 96-well cell culture plate and incubated in Dulbecco's modified Eagle's/ F-12 medium (1:1) supplemented with 1.5% (v/v) fetal bovine serum (assay medium) for 18 h. Fresh assay medium with fixed amounts of VEGF (ϳ0.15 nM final concentration), which was first titrated as a level of VEGF that can stimulate submaximal DNA synthesis, and increasing concentrations of anti-VEGF antibodies were then added to the cells. After incubation at 37°C for 18 h, cells were pulsed with 0.5 Ci/well of [ 3 H]thymidine for 24 h and harvested for counting with TopCount Microplate Scintillation counter.
Tumor Implantation and in Vivo Treatments-HM-7, A673, and HPAC cells (American Type Culture Collection) were maintained in culture with Dulbecco's modified Eagle's/F-12 medium, supplemented with 10% fetal bovine serum. Cells were grown at 37°C in 5% CO 2 until confluent and then harvested and resuspended in sterile Matrigel at a concentration of 25 ϫ 10 6 cells/ml. Xenografts were established in 6 -8week-old female Beige Nude XID mice by dorsal flank subcutaneous injection of 5 ϫ 10 6 cells/mouse and allowed to grow. When tumors reached a volume of 500 mm 3 for HM-7 and 150 -200 mm 3 for A673 and HPAC (48 h), a cohort was randomly selected (n ϭ 10) as day 0 controls. The remaining mice were divided into groups of 10, and antibodies were administered intraperitoneally at the same dose for each group. Tumor sizes and weights were measured as described (11). All statistical analyses used Student's t test.
Histology and Staining-Tumor were fixed in 10% neutral buffered formalin for 12-16 h prior to paraffin embedding. Histologic sections of 4 -5 m thickness were stained with hematoxylin and eosin as described previously (32). Vascular density was assessed in MECA-32 (anti-PLVAP)-stained histological sections of tumors, essentially as described previously (33). For each tumor, three representative images, centered 500 m from the tumor margin, were analyzed.
ELISAs for VEGF Protein in Tumor Extract and Anti-VEGF in Mice Serum-For VEGF in tumors, 400 -800 g of excised tumor was homogenized in RIPA buffer (100 l/g) containing 20 mM Tris, pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% deoxycholic acid, 1% Triton X-100, and 0.25% SDS, and supernatants were collected. Human VEGF ELISA of high sensitivity was performed as described (34). Briefly, sandwich ELISA used monoclonal antibody 3.5F8 (Genentech) to h-VEGF as coat, and biotinylated A4.6.1 as detecting antibody. VEGF 165 standards (1-128 pg/ml in 2-fold serial dilution) and samples were measured side by side. This ELISA does not detect m-VEGF. To measure m-VEGF, sandwiched ELISA used goat anti-m-VEGF antibody (R&D Systems) as coat and the same antibody in the biotinylated form for detection; m-VEGF standards (3.9 -500 pg/ml in 2-fold serial dilutions) and tumor extract in sample buffer were measured side by side. The cross-reactivity of h-VEGF occurred when concentration was above 2000 ng/ml at Ͻ0.4%. For serum anti-VEGF (uncomplexed), ELISA plates were coated overnight with 0.5 g/ml VEGF 165 in 50 mM sodium carbonate, pH 9.6, blocked, washed, and then incubated with 2-fold serial dilutions of serum samples or anti-VEGF standards (0.20 -25 ng/ml for Y0317 and Avastin, 0.39 -50 ng/ml for B20-4.1, and 0.1-12.8 ng/ml for G6-31) in 0.05% BSA, 0.2% bovine ␥-globulin, 0.25% CHAPS, 5 mM EDTA, 0.35 M NaCl, 0.05% Tween 20 in PBS, pH 7.4 (samples buffer), for 2 h. Bound antibodies were detected by appropriate anti-IgG-HRP conjugates, e.g. anti-human Fc-HRP (Jackson ImmunoResearch) for Avastin, Y0317, and G6-31 as human IgG1, and anti-mouse IgG 2a -HRP (Pharmingen) for B20-4.1, which had mouse IgG2a constant domains. Total antibodies (complexed and uncomplexed) were measured with ELISA plates coated with appropriate anti-IgG antibodies as above. Bound antibodies were detected with the same anti-IgG antibodies conjugated with HRP. By using the same assay for B20-4.1, we observed that serum from control nude mice with no injected antibodies contained less than 0.2 g/ml mouse IgG2a.

RESULTS
G6 and B20 Bind Human and Murine VEGF-G6 and B20 were among several clones isolated from a synthetic antibody phage library with humanized anti-ErbB2 antibody, h4D5, as the template by selec-tion against m-VEGF. The binding specificity and receptor blocking activity of these phage clones displaying the antigen-binding fragment (Fab) were first examined (Fig. 1A). G6, the clone with the best apparent affinity for m-VEGF (EC 50 ϭ 0.6 nM), also showed near equivalent binding for h-VEGF at 1.4 nM (EC 50 ). B20, however, binds m-VEGF and h-VEGF with relative affinity of 44 and 300 nM, respectively. To assess which antibody clone can block VEGF binding to its receptors, VEGFR2 ECD with seven Ig-like domains (VEGFR2 D1-7 ) (Fig. 1B) or the second Ig domain of VEGFR1 ECD (VEGFR1 D2 ) (Fig. 1C) was used to block the interaction of Fab-displaying phage with m-VEGF. Strong blocking was observed with G6 and B20 by both receptor fragments, suggesting that their epitopes on VEGF overlapped with the epitopes for receptor fragments. As B20 had weak and different affinity for h-VEGF and m-VEGF, it went through a first step of affinity improvement by selection in a phage-displayed library with light chain CDR residues randomized. Variant B20-4 exhibited equal apparent affinities for both h-VEGF and m-VEGF (EC 50 ϳ15 nM) (see supplemental Fig. 1).
Next, G6 and B20-4 Fab were generated to confirm the binding specificities. G6 and B20-4 as Fabs indeed demonstrated near equal affinity for human and murine VEGF (see Table 1 for K d values), and no detectable binding to other VEGF homologues, h-and m-PlGF, h-VEGF-B (and m-VEGF-B, not shown), h-VEGF-D (Fig. 1D), and VEGF-C (data not shown). G6 also bound rat and rabbit VEGF with similar affinity as for m-VEGF (data not shown). The receptor blocking activity of G6 or B20-4 Fab was confirmed by blocking both h-and m-VEGF binding to VEGFR2 (1-7 Ig domain ECD) (Fig. 1E) or VEGFR1 (1-5 Ig domain ECD) (m-VEGF blocking for VEGFR2 binding shown, and the rest is not shown). Avastin Fab (Fab-12), in comparison, can only block h-VEGF but not m-VEGF binding to the receptor as expected (Fig. 1E). An Avastin variant with improved affinity for h-VEGF, Y0317, showed slight m-VEGF-blocking activity at high concentrations as it acquired a weak binding affinity for m-VEGF (26) (K d ϭ 333 nM, see below).
The Affinity-improved G6 and B20 Variants-To improve the affinity of G6, light chain CDR residues were mutated as described (28). Many G6 variants were isolated that differed from G6 by four or five residues in the CDR-L3, e.g. G6-8, G6-23, and G6-31. B20-4 was further improved by mutations in heavy chain CDRs to B20-4.1 (see supplemental Fig. 1). Binding affinities (K d ) of the improved variants as Fab or IgG protein were measured at 37°C with surface plasmon resonance (BIAcore TM ) on immobilized human or murine VEGF in

The binding affinities and kinetic parameters of anti-VEGF antibodies
Affinities of Fab or IgG are measured on BIAcore using sensor chip immobilized with human or murine VEGF at 37°C (see "Materials and Methods"). Each measurement represents an average of three independent assays that vary by 20 -50%. NB denotes no binding detected; ND denotes calculation not done. comparison to A4.6.1, Avastin (Fab-12 as Fab), and the affinity-improved version Y0317 (Table 1). The avidity advantage exhibited by the IgG reduced the apparent off-rates, but the overall ranking of these antibodies held. The improved G6 and B20 variants have similar affinities for human and murine VEGF. Compared with hybri-doma-derived Avastin (or Fab-12 as Fab), B20-4.1 binds h-VEGF with similar affinity (ϳ4 -8 nM as Fab and 2 nM as IgG) and kinetic profile of on-rate, k on , and off-rate, k off . Avastin was used interchangeably with A4.6.1 in the study as they have equivalent activities in vitro and in vivo (35). Among the high affinity variants, G6-23 and G6-31 bind ϳ5-10-fold weaker than Y0317 (20 pM), but G6 variants bind with nearly 20-fold faster kinetics (on-rate and off-rate) than Y0317.
As for binding specificity, G6-31 shows no detectable binding for any of the VEGF homologues (PlGF, VEGF-B, -C, and -D) as its parent ( Fig. 2A). However, B20-4.1 as IgG, but not as Fab, shows some weak interaction with m-PlGF; the avidity advantage of IgG apparently enhanced the weak interaction. To assess whether B20-4.1 binding to m-PlGF is biologically significant, we examined how well B20-4.1 inhibits m-PlGF from binding m-VEGFR1 ECD, and we saw weak inhibition at 10 M (estimated EC 50 ϳ30 -100 M) (Fig.  2B). VEGFR1, as control, shows physiologically relevant binding for the m-PlGF (EC 50 ϳ1 nM). G6-31 and B20-4.1 block both h-VEGF and m-VEGF from binding VEGFR2 and VEGFR1 with high potency (EC 50 ϳ subnanomolar) (Fig. 2, C and D, and data not shown). The weak interaction of B20-4.1 and m-PlGF should not play a significant role for its effect in vivo. We next determined the activity of G6 and B20 variants in inhibiting the VEGF-stimulated growth of HUVEC (Fig. 2, E and F). Both G6 and B20 affinity-improved variants were shown to inhibit the activity of both human and murine VEGF, whereas Y0317 or Avastin, as expected, blocked human VEGF selectively. B20-4.1 exhibited similar potency as Avastin in inhibiting h-VEGF, whereas G6-31 and G6-23 were ϳ20-fold more potent than B20-4.1 in inhibiting human and murine VEGF and ϳ20-fold more potent than Avastin in inhibiting h-VEGF, consistent with their binding affinities. G6-31 and G6-23 were equipotent to Y0317; the  (n ϭ 10). B, tumor weight reduction by different doses of antibodies was measured at day 23. Statistically significant reduction by each treatment versus control antibody (5 mg/kg) was denoted with * (p Ͻ 0.05, Student's t test). In comparing the potency of G6-31 and B20-4.1, a statistically significant (p Ͻ 0.05) advantage of the high affinity G6-31 was seen only when dosed at 0.1 mg/kg. C, tumor sections were stained for vessel-specific marker as described under "Materials and Methods" to determine the vessel density (vessel perimeter (m)] per tumor area (m 2 ), m Ϫ1 ). G6-31 and A4.6.1 treatment significantly reduced the tumor vessel density compared with control antibody (p Ͻ 0.001).
assay was unable to differentiate antibodies with affinities higher than the concentration of VEGF used in the assay (0.15 nM).
In Vivo Activity of B20-4.1 and G6-31 Compared with A4.6.1 and Y0317 in Inhibiting Human Tumor Growth in Nude Mice-To investigate whether a more complete blockade of VEGF will achieve greater tumor suppression, B20-4.1, G6-31, or isotype control IgG were injected into nude mice after the human colorectal cancer cell line HM-7 formed a sizable tumor under the mouse skin (ϳ500 mm 3 , 2 days). Treatments that could not block VEGF originating from host, A4.6.1 and Y0317, were included in the study for comparison. Antibodies were administered intraperitoneally at doses of 0.1, 0.5, or 5 mg/kg (2.5, 12.5, and 125 g/mouse) twice weekly for 3 weeks (n ϭ 10); tumor volume was measured twice weekly (Fig. 3A), and tumor weight was determined at the end of the experiment (Fig. 3B). All four antibodies inhibited tumors significantly relative to the control group in a dose-dependent manner. G6-31 was able to inhibit significantly tumor growth (ϳ70%) even at the very low dose (0.1 mg/kg). Higher doses resulted in complete inhibition. Furthermore, G6-31-treated tumor shows significant reduction of vessel density like A4.6.1, suggesting inhibition of angiogenesis as the mechanism of action (Fig. 3C).
To examine whether a correlation exists between binding affinity and potency of tumor inhibition, the low affinity antibody B20-4.1 was compared with the high affinity G6-31 antibody. The two antibodies have identical cross-species reactivity. We observed that both antibodies achieved nearly complete tumor inhibition at the high dose (5 mg/kg) (Fig. 3). However, at lower doses (0.5 and 0.1 mg/kg), B20-4.1 was less effective than G6-31 (p ϭ 0.065 and 0.007, respectively) (Fig. 3B), suggesting that higher affinity may result in greater in vivo potency. In apparent conflict with this conclusion, A4.6.1 and Y0317 were not significantly different in potency despite ϳ100-fold different binding affinities for h-VEGF and more than a 20-fold advantage for Y0317 in inhibiting h-VEGF-stimulated endothelial cell growth in vitro (Fig. 2). At all doses, tumor shrinkage in the Y0317-treated group was not significantly better than the A4.6.1-treated group (p Ͼ 0.05). The only group that showed a slight advantage of Y0317 was the high dose group (5 mg/kg), but the difference was not statistically significant (p ϭ 0.059) (Fig. 3B). One possible explanation is that stromal VEGF (m-VEGF) was up-regulated and masked the effect of blocking tumor-derived VEGF. The fact that Y0317 does have some weak binding affinity for m-VEGF whereas A4.6.1 (or Avastin) has no detectable binding to m-VEGF confounds the interpretation of the observation. Another possible explanation is that the twice-weekly dosing regimen and/or the metabolic turnover rate of VEGF did not permit the off-rate differences of these two antibodies to impact the potency. G6-31 and B20-4.1, in contrast, varied mainly in their on-rate.
Stromal VEGF Contribution Measurements-To examine the level of stromal VEGF contribution to tumor growth, we compared the efficacy of antibodies that blocked both stromal and tumor-derived VEGF, G6-31 and B20-4.1, and antibodies that only blocked tumor derived VEGF, Avastin. We added a slow-growing pancreatic tumor xenograft HPAC, which contains extensive stromal cell infiltration, 6 and another fast growing human rhabdomyosarcoma A673 xenograft (Fig. 4A). HPAC and A673 tumor inhibition was compared with HM-7 results at a saturating dose of these antibodies (5 mg/kg/week, n ϭ 10). For HM-7 inhibition, the cross-species binding antibodies (G6-31 and B20-4.1) were more efficacious than the non-cross-species antibodies (A4.6.1), but the inhibition by the latter was substantial (ϳ90%), indicating some but minimal contribution of stromal VEGF (Fig. 4B). For A673, the 6 F. V. Peale, and N. Ferrara, unpublished observations.

TABLE 2 Circulating anti-VEGF antibody measurements
Serum samples were collected 7 days after three weekly injections (day 28) (n ϭ 5) in the HPAC group. Uncomplexed IgG and total IgG (complexed plus uncomplexed) were measured (mean Ϯ S.E.) as described under "Materials and Methods." The differences between the uncomplexed and total IgG levels of all four antibodies are statistically insignificant (p Ͼ 0.05). The circulating levels (total IgG) of Avastin are significantly higher than all three other antibodies (p Ͻ 0.05), yet the levels of two anti-human/mouse VEGF antibodies, B20-4.1 and G6-31, are statistically indistinguishable (p ϭ 0. 34).

Antibodies
Uncomplexed Total g/ml  inhibition by the cross-species antibodies was not significantly better than Avastin, thus contribution by stromal VEGF was nearly nonexistent. In contrast, Avastin inhibited less than 50% of the HPAC growth, whereas both B20-4.1 and G6-31 completely inhibited the tumor growth, indicating significantly more stromal VEGF contribution in the growth of HPAC than HM-7 and A673 (Fig. 4, A and B). Y0317 inhibition of HPAC and HM-7 was also less effective than the cross-species reactive antibodies, but it surpassed Avastin with statistical significance in HPAC but not HM-7, which could be due to the fact that Y0317 also weakly binds m-VEGF. The differences in efficacy between Avastin, which clearly cannot bind m-VEGF, and the cross-species antibodies verified that host-derived VEGF does contribute to the tumor growth and, for the first time, demonstrated clearly that there were significant differences in the level of stromal VEGF contribution in the growth of different tumors. Tumor sections were next examined for the effect of the antibodies on these tumors with different contributions of stromal VEGF (Fig. 5). Macroscopically, the section of the HM-7 and HPAC tumors treated with Avastin IgG (or the equivalent A4.6.1), G6-31 (representative of B20-4.1 group), or control antibodies showed that Avastin indeed was not effective in decreasing the tumor size and increasing the area of necrosis of HPAC, whereas G6-31 treatment resulted in extensive necrosis in both HPAC and HM-7 tumor (Fig. 5A). Microscopically, HM-7 and HPAC exhibited quite different characteristics (Fig. 5B). HM-7 is a solid mass of tumor cells interspersed with large thin-walled vessels, which appeared to be lined only with endothelial cells; treatment with A4.6.1 resulted in collapse of these large vessels and appearances of small vessels next to clusters of host-derived stromal matrix and various host cell types. Treatment with G6-31 resulted in a largely necrotic mass thinly margined by the remaining HM-7 tumor cells, presumably surviving by accessing nutrients diffusing from the surrounding tissues. Tumor sections of A673 were highly similar to HM-7, as studied previously (11). HPAC tumors, however, appeared densely interlaced with host-derived cells and matrix, and small size vessels were visible. Avastin treatment did not result in marked histological differences; G6-31 treatment, however, reduced the tumor to a mostly necrotic mass, but the surviving tumor cells on the margin still maintained a high proportion of host-derived cells, similar to the control tumors. Recruitment of host cells was indeed much more a property of HPAC than HM-7.
VEGF Generated by Tumor or Stromal Cells and Antibody Levels in Serum-We next determined the level of tumor versus host stromaderived VEGF present in the tumor mass to see whether it was consistent with tumor inhibition by antibodies of different specificities. In the tumor extract of HM-7 (control group, n ϭ 5), tumor-derived h-VEGF was dominating (1802 Ϯ 35 pg/mg protein, mean Ϯ S.E.), although stroma-derived m-VEGF was detected (35 Ϯ 9 pg/mg). In A673, h-VEGF was present at a moderate level (163 Ϯ 39 pg/mg), and m-VEGF was minimal (15 Ϯ 4 pg/mg). In contrast, HPAC tumors contained nearly equivalent amounts of h-VEGF (116 Ϯ 30 pg/mg) and m-VEGF (147 Ϯ 60 pg/mg), consistent with the finding that blocking h-VEGF alone was very efficient at inhibiting the growth of HM-7 and A673 but not sufficient for HPAC inhibition (see supplemental Table  for summary).
Circulating concentrations of injected antibodies in the tumor-bearing mice were measured to correlate the observed effects to serum concentrations. Serum samples on day 7 after the 3rd week of treatment of the HPAC group were assayed with ELISA for uncomplexed antibody (VEGF-coated wells for capture and anti-Fc-HRP conjugate for detection) or total antibody (anti-Fc-coated wells for capture and anti-Fc-HRP for detection) ( Table 2), which were representative of the serum IgG levels of animals with the other tumors. The levels of Y0317, G6-31, and B20-4.1 were lower than Avastin (p Ͻ 0.05). The reason for this is unclear, but it is possible that the cross-reactive antibodies, including Y0317, either cleared faster or were bound by host VEGF in tissues because the free and total IgG levels in circulation were statistically indistinguishable. Studies are ongoing to understand the pharmacokinetics and tissue distribution of these antibodies in animals. For the current goal of examining the efficacy of cross-blocking antibodies, the equivalent circulating concentration of B20-4.1 and G6-31 antibodies supports the interpretation of the role of affinity in potency and efficacy using these two antibodies, and the lower circulating levels of G6-31 and B20-4.1 when compared with Avastin and Y0317 further emphasize the result that cross-reactive antibodies had greater efficacy than Avastin and Y0317 in two of three tumor models.

DISCUSSION
Both genetic and pharmacological approaches have been employed to inhibit VEGF and to demonstrate the key role of VEGF in tumor angiogenesis (36,37). As discussed in Introduction, due to target or species-specificity, previously developed pharmacological inhibitors do not adequately allow assessment of the contribution of VEGF-A to tumor angiogenesis (7,11,19). We have developed two phage libraryderived antibodies, G6-31 and B20-4.1, which block both human and mouse VEGF-A. These VEGF blockers achieved complete inhibition of the growth of implanted human colorectal cancer (HM-7), rhadomyosarcoma (A673), and pancreatic cancer (HPAC) cells in mice. The completeness of the tumor inhibition was remarkable as demonstrated by the extent of tumor reduction and necrosis. As the antibody itself did not kill tumor cells or inhibit endothelial cell growth stimulated with growth factors other than VEGF (e.g. basic fibroblast growth factor) (data not shown), the inhibition was a consequence of specific inhibition of VEGF. This model established that blocking VEGF-A is sufficient to fully inhibit tumor angiogenesis and tumor growth. This conclusion is consistent with recent studies of an endogenously induced pancreatic ␤-cell tumor model where specifically knocking out VEGF expression locally severely inhibited tumorigenesis (10). Our observations demonstrated that blocking all sources of VEGF at an early stage of tumorigenesis was sufficient to inhibit tumor progression and thus validated their application for assessing the role of VEGF at different stages of tumorigenesis in a variety of tumor models in mouse.
Mouse models represent powerful, often indispensable, means to understand the biology of many diseases. The contribution of host stroma-derived VEGF in the growth of different tumors, however, is a confounding factor in the interpretation of the effects of various pharmacological blockers or genetic modulation of tumor VEGF expression. Furthermore, it has been a controversial issue as to the functional significance of stromal VEGF. For example, according to a recent study, stromal VEGF does not contribute significantly to the growth of Rastransformed adult fibroblasts (38). In contrast, in a study that used a VEGF-null Ras-transformed fibrosarcoma of mouse embryonic origin, the growth of the implanted tumor in mouse was substantially inhibited by a blocking antibody (G6-23) against VEGF (murine VEGF) (39), indicating the significant functional importance of stromal VEGF. The different conclusions of the two studies could be due not only to the different potencies and efficacies of the blocking agents used but also may be related to differences in the levels of stromal VEGF involvement in the two tumors. The availability of cross-species blocking antibodies enables measurements of functional contribution of stromal VEGF directly, which is clearly demonstrated to vary significantly among human tumor types in this study.
A conclusion of the present study is that increases in affinity of VEGF antibodies above ϳ1 nM (K d at 37°C)) did not result in marked increases in efficacy in these models. However, G6-31 inhibited tumor growth better than B20-4.1 at low dose (0.1 mg/kg). In in vitro systems such as VEGF-stimulated HUVEC proliferation assays, the potency correlated well with affinity of anti-VEGF antibodies. However, when VEGF was added to the cells at concentrations above 1-2 nM, such affinity advantage of antibodies was reduced (data not shown). VEGF synthesis is finely regulated such that inactivating even a single copy of the VEGF gene results in embryonic lethality (40,41). A partial blockade of VEGF may be sufficient to block in vivo tumor angiogenesis. The effective concentration of available VEGF in vivo and the metabolic turnover rate of VEGF in tumor angiogenesis that could determine the impact of affinity of blocking antibodies are not known because the circulatory VEGF only represented a portion of VEGF activities in vivo. VEGF transcriptional activation in tumors has been shown to last an extended period of time (13), and the release of VEGF from a matrix-bound state to activate angiogenesis is highly regulated (42). It is likely that a low level of VEGF is generated, continuously or in pulses, throughout the process of tumorigenesis, which could vary significantly from tumor to tumor. As the stromal VEGF contributes to tumor angiogenesis, the environment of tumor growth could also influence the dynamics of VEGF recruitment to fuel tumor angiogenesis. The impact of the affinity of VEGF-blocking antibodies thus could vary significantly among different tumors and tumor models. The particularly high VEGF levels in HM-7 extracts might have reduced the impact of the affinity of the blocking antibody. It would be important to continue to assess the impact of binding affinity in relation to the potency and efficacy in different tumor models and stages, e.g. orthotopically implanted tumor, metastasized tumor, and endogenously induced tumor models. The cross-species blocking anti-VEGF antibodies are suitable to test these hypotheses.
Ever since the initial proposal that blocking angiogenesis may affect tumorigenesis (43), decades of research have shown its validity and, at the same time, recognized the complexity of tumor angiogenesis (44). Now that specific VEGF blockade has been shown to significantly suppress tumor progression in patients (2), it is important to be able to establish preclinical models that more closely reflect the complexity of tumor progression. The cross-reactive antibodies described in the present study provide tools to further understand this approach of therapy. For example, tumor types and progression stages that indeed escape VEGF blockade may now be identified and analyzed so that factors and mechanisms that contribute to tumor escape can be studied; combination treatment can be explored. Furthermore, the specific role of VEGF in other pathological and physiological processes can be dissected using these antibodies in mouse models.