Interaction between Bevacizumab and Murine VEGF-A: A Reassessment

Abstract

purpose. Bevacizumab is a humanized anti-human VEGF-A monoclonal antibody (mAb) approved by the United States Food and Drug Administration for cancer therapy and used off label to treat neovascular age-related macular degeneration. Earlier studies characterized bevacizumab as species specific and lacking the ability to neutralize murine (m) VEGF-A. However, a recent study reported that bevacizumab is a potent inhibitor of hemangiogenesis and lymphangiogenesis in murine models. The authors sought to reassess the interaction between bevacizumab and mVEGF-A.

methods. The authors performed Western blot analysis, plasmon resonance by BIAcore, and endothelial cell proliferation assays to characterize the interaction between bevacizumab and mVEGF-A. They also tested whether bevacizumab had any effects in two in vivo murine models, laser-induced choroidal neovascularization (CNV) and melanoma growth.

results. Western blot detected a very weak interaction, but BIAcore detected no measurable interaction between mVEGF and bevacizumab. Bevacizumab failed to inhibit mVEGF-stimulated endothelial cell proliferation. In addition, bevacizumab was indistinguishable from the control antibody in the CNV and tumor models, whereas a cross-reactive anti–VEGF-A mAb had dramatic inhibitory effects.

conclusions. Bevacizumab has an extremely weak interaction with mVEGF-A, which fails to result in immunoneutralization as assessed by several bioassays.

It is well established that angiogenesis is essential for many physiological processes. ^{1 2} Conversely, antiangiogenesis is a validated strategy to treat a number of disorders, including solid tumors and intraocular neovascular syndromes. ^{3 4} One of the key mediators of physiological and pathologic angiogenesis is vascular endothelial growth factor (VEGF)-A. ⁵ Early studies show that VEGF-A is increased in the eye fluids of patients with ischemic retinal disorders ⁶ or in neovascular membranes of patients with age-related macular degeneration (AMD). ⁷ Two VEGF inhibitors, pegaptanib ⁸ and ranibizumab, ⁹ have been approved by the United States Food and Drug Administration (FDA) for the therapy of neovascular AMD. Bevacizumab, an anti–VEGF-A humanized monoclonal antibody (mAb), has been approved by the FDA as a treatment for metastatic colorectal and nonsquamous, non–small-cell lung cancer, in combination with chemotherapy, ¹⁰ and is also administered off label to patients with AMD. ¹¹  

Recently, Bock et al. ¹² reported that bevacizumab potently inhibits corneal angiogenesis and lymphangiogenesis. Indeed, much research supports the notion that VEGF-A plays an important role in corneal angiogenesis. As early as 1998, Amano et al. ¹³ showed that VEGF is required for wound- and inflammation-related corneal neovascularization in a rat model. Other studies showed that VEGF is also important for corneal neovascularization associated with herpes simplex virus infection or CpG oligodeoxynucleotides. ^{14 15} More recent studies implicate VEGF-A in the regulation of lymphangiogenesis in various models. ^{16 17} Therefore, the findings reported by Bock et al. ¹² may not seem surprising at first. However, bevacizumab, which is a humanized variant ¹⁸ of the anti–human (h) VEGF-A mAb A.4.6.1, originally described by our group 15 years ago, ¹⁹ has been extensively characterized as species specific and having very little or no interaction with murine (m) VEGF-A. ²⁰ The inhibitory effects of bevacizumab or its murine precursor on human tumor xenografts implanted in immunodeficient mice were found to be inversely proportional to the content of host stroma-derived mVEGF in the tumor. ²¹ In previous studies, we elucidated the molecular basis of the species-specific binding of VEGF-A to bevacizumab. ^{22 23} X-ray structure, combined with site-directed mutagenesis, demonstrated that Gly-88 in hVEGF-A, corresponding to Ser-87 in the mVEGF-A sequence, is a key residue responsible for the species-specific VEGF-A binding to bevacizumab. ²² The crystal structure of hVEGF-A in complex with the bevacizumab-Fab revealed that the interface is tightly packed, such that there is not enough room for two additional nonhydrogen atoms introduced by the Gly-88 to Ser-87 exchange. ²³ Thus, binding of bevacizumab to mVEGF is predicted to be much weaker than hVEGF-A binding. 

Therefore, we were surprised that the animal models in which Bock et al. ¹² found bevacizumab to be active are murine. We were especially puzzled by the claim that systemic administration of as little as 5 mg/kg bevacizumab elicits marked antiangiogenic effects in such models. ¹²  

Bock et al. ¹² attempt to document the binding of bevacizumab to mVEGF-A in some biochemical assays, and they imply that the in vivo bioactivity they describe is related to such interaction. However, they do not adequately explain how a binding that even in their own hands is very weak may result in potent in vivo effects. Furthermore, no quantification of the interaction between bevacizumab and mVEGF-A was provided. In addition, the lack of a positive control and, in several instances, of a negative IgG control makes it difficult to assess the magnitude of the effects reported. 

Considering that bevacizumab is widely available, the study by Bock et al. ¹² may encourage other investigators to use this reagent to probe the pathophysiological role of VEGF-A in other murine models. Therefore, we believe that it would be of considerable interest to reassess the interaction of bevacizumab with mVEGF-A, both in vitro and in vivo. 

Materials and Methods

Western Blot Analysis

Escherichia coli–expressed recombinant human VEGF-A (0.5 or 2 μg; 165-amino acid isoform; Genentech, Inc., South San Francisco, CA) or Sf 21 insect cell-expressed recombinant murine VEGF-A (164-amino acid isoform; R&D Systems, Minneapolis, MN) were reduced with 50 mM dithiothreitol and loaded on a 4% to 20% SDS-PAGE (Invitrogen, Carlsbad, CA). After completion of electrophoresis, transfer (2 hours) was performed by semidry transfer cell (Bio-Rad, Hercules, CA). Polyvinylidene difluoride (PVDF) membranes were incubated overnight with bevacizumab or mAb G6-31 at a final concentration of 10 μg/mL in blocking buffer (5% skim milk in PBST). After three washes in 0.2% skim milk in PBST, the blots were incubated with peroxidase-conjugated, Fab-specific goat anti–human IgG (ImmunoPure; Pierce, Rockford, IL) for 1 hour at a dilution of 1:10,000 and were then developed using enhanced chemiluminescence (ECL Plus Western Blotting Detection System; GE Healthcare Bio-Sciences, Little Chalfont, Buckinghamshire, UK) after 1- or 10-minute film exposure. 

Surface Plasmon Resonance Measurement of hVEGF-A and mVEGF-A Binding to Bevacizumab or mAb G6-31

Bevacizumab ¹⁸ or mAb G6-31 ²⁴ was individually immobilized through amine linkage chemistry to a Biacore chip at approximately 1000 resonance units (RU). hVEGF-A or mVEGF-A (20–50 nM) from two different sources was injected over the chip. Recombinant hVEGF₁₆₅ was E. coli expressed (Genentech, Inc.). Recombinant mVEGF₁₆₄ was from R&D Systems (493-MV/CF, Sf21 insect cell expressed) or Chemicon, Inc. (GF060, E. coli expressed). The net binding sensorgram was acquired by subtracting the binding RU with that of the same injection on a control blank chip. After the binding interaction, the chips were regenerated by a short injection of 20 mM HCl. 

Endothelial Cell Proliferation Assays

Bovine retinal microvascular endothelial cells (BRECs) were seeded at a density of 500 cells per well in 96-well plates in growth medium (low glucose DMEM supplemented with 10% calf serum, 2 mM glutamine, and antibiotics). ²² For inhibition assay, bevacizumab or mAb G6-31 was added at the indicated concentration (ng/mL) to triplicate wells. After 0.5 hour, hVEGF-A or mVEGF-A was added to a final concentration of 6 ng/mL. After 6 to 7 days, cell growth was determined by Alamar blue (BioSource; Invitrogen). Fluorescence was monitored at 530 nm excitation wavelength and 590 nm emission wavelength. 

Laser-Induced Choroidal Neovascularization

The 532-nm diode laser photocoagulation was delivered to the choroid of C57Bl/6 mice with 120 mW power for 100 ms, with a mean diameter of 100 μm. Treatment groups (n = 10) included mAb A4.6.1 (mouse monoclonal precursor of bevacizumab), ¹⁹ mAb G6-31, bevacizumab or human IgG, given twice weekly intraperitoneally at 5 mg/kg for a total of three doses over the 10-day study. Mice were injected intravenously with 0.1 mL of 500 mg/mL fluorescein Lycopersicon esculentum lectin (FL-1171; Invitrogen) through the tail vein 5 minutes before kill to mark the choroid vasculature. ²⁵ After fixation, the entire choroid with eyecup was dissected and flatmounted. All images of neovascularization sites were obtained at a 20× objective with 1.58-second exposure time with an imaging system (Axioplan2; Carl Zeiss, Oberkochen, Germany) at the FITC channel. The neovascular area was quantified by NIH ImageJ 1.37V. Each image was first color split into a single green channel image, followed by thresholding at a value of 35 to 255. The area of neovascularization was selected by freehand draw, and a pixel/length ratio of 194 pixel/wound diameter was applied to translate pixel to area. 

Tumor Experiments

Murine B16F1 melanoma cells were cultured as previously described. ²⁶ Cells were then injected subcutaneously into 8- to 12-week-old female beige nude mice at the density of 5 × 10⁶/mouse. Forty-eight hours after tumor cell inoculation (day 0), mice (n = 10) were injected intraperitoneally with control IgG or with bevacizumab at 5 or 50 mg/kg. A positive control group received mAb G6-31 at 5 mg/kg. Antibodies were administered twice weekly thereafter. Tumor volumes were measured with calipers at the indicated time points. 

Results

We performed Western blot analysis of hVEGF-A and mVEGF-A with bevacizumab as the primary antibody using conditions similar to those reported by Bock et al. ¹² Given that 2 μg represents a relatively large amount of immobilized protein, we also included 0.5 μg of hVEGF and mVEGF. In addition, we probed parallel VEGF-A blots with mAb G6-31, a cross-reactive phage-derived mAb that recognizes mVEGF-A and hVEGF-A proteins with nearly equal affinity. ²⁴ As shown in Figure 1A , after 1-minute film exposure, the bevacizumab-probed blot revealed strong immunoreactive bands of the expected size for hVEGF-A. In contrast, mVEGF-A was completely undetectable. After 10-minute exposure, when the background was clearly darker, very strong signals for hVEGF-A were visible, whereas a very faint band corresponding to 2 μg mVEGF-A could be detected. The 0.5 μg mVEGF lane gave almost no signal. The mAb G6-31 blot showed nearly equal signal intensities for mVEGF-A and hVEGF-A. Therefore, the interaction of bevacizumab with mVEGF-A by Western blot was exceedingly weak, and, under the conditions tested, a faint signal was detectable only after overexposure. 

We next tried to reproduce the binding described by Bock et al. ¹² with the use of plasmon resonance by BIAcore. However, in two independent experiments and with mVEGF-A from two different sources, we were unable to detect any significant interaction between bevacizumab and mVEGF-A. In contrast, bevacizumab bound hVEGF-A and mAb G6-31 bound hVEGF-A and mVEGF-A proteins. Figure 2depicts the results of a representative experiment. 

We have previously reported that bevacizumab does not inhibit the activity of mVEGF-A as assessed by proliferation assays using several endothelial cell types. ^{22 24} To further verify these results, we tested the effects of bevacizumab or mAb G6-31 on the proliferation of BRECs stimulated by 6 ng/mL hVEGF₁₆₅ or mVEGF₁₆₄ (Fig. 3) . In agreement with previous studies, ¹⁸ bevacizumab inhibited hVEGF-A–stimulated BREC growth with a near maximal inhibition at approximately 500 ng/mL. However, it failed to significantly inhibit BREC growth induced by mVEGF-A, even at the highest concentration tested (13,500 ng/mL). This concentration provides an antibody/antigen molar ratio of more than 650:1. In contrast, mAb G6-31 resulted in a nearly complete inhibition of both hVEGF-A– and mVEGF-A–stimulated BREC proliferation at the concentration of approximately 18 ng/mL (Fig. 3B) . 

To follow up these in vitro studies, we compared bevacizumab with mAb G6-31 in two different murine models known to be highly responsive to VEGF inhibitors: laser-induced choroidal neovascularization (CNV) ²⁷ and subcutaneously implanted B16F1 melanoma. ²⁶  

As illustrated in Figure 4 , mAb G6-31, given three times intraperitoneally over the 10-day experimental period at the dose of 5 mg/kg, resulted in approximately 90% inhibition of CNV compared with control IgG. However, bevacizumab or its murine precursor A.4.6.1, given at the same dosing regimen as mAb G6-31, was indistinguishable from the control IgG in two independent experiments. 

We also tested the effects of bevacizumab on the growth of B16F1 murine melanoma cells transplanted subcutaneously into nude mice. Because the 5-mg/kg dose had no detectable effect in the CNV model, we also tested a 10-fold higher dose of bevacizumab. However, as shown in Figure 5 , neither 5 nor 50 mg/kg twice weekly had any effect on tumor growth. Conversely, mAb G6-31 (5 mg/kg/twice weekly) resulted in a dramatic inhibition of tumor growth. 

Discussion

Bock et al. ¹² recently reported that bevacizumab has potent inhibitory effects in murine models of corneal hemangiogenesis and lymphangiogenesis. In support of their claim, they attempted to demonstrate that bevacizumab interacts with mVEGF in some biochemical assays. They show a Western blot that purportedly documents binding of bevacizumab to mVEGF-A.¹² The intensity of the mVEGF-A signal seems weaker than that observed with hVEGF-A, but the authors might still have overinterpreted such interaction. In several areas, the background of the blot shown appears darker than the mVEGF-A signal itself, consistent with prolonged exposure, to evidence a weak interaction. It is generally appreciated that one of the pitfalls of Western blot analysis is the difficulty of quantifying bound protein, combined with a lack of linearity of the intensity of an autoradiographic signal with length of exposure. Therefore, a single, prolonged exposure may provide an inaccurate representation of the relative proportions of a weak versus a strong immunoreactive signal, possibly emphasizing the weaker signal. As shown in Figure 1 , we were able to document by Western blot a very faint, barely detectable, interaction between bevacizumab and 2 μg mVEGF only after overexposing the film. More rapid exposure that evidenced strong immunoreactive bands for 0.5 and 2 μg hVEGF-A did not yield any mVEGF-A signals. 

Bock et al. ¹² also assessed the binding of bevacizumab to hVEGF-A or mVEGF-A by ELISA. The assay showed that bevacizumab bound strongly to hVEGF-A, with an EC₅₀ of approximately 10 ng/mL. However, a minimal amount of binding to mVEGF-A was detectable, but only at the highest concentration of bevacizumab (approximately 5000 ng/mL). The authors did not provide any EC₅₀ value for this binding. If one extrapolates from the single data point, however, the EC₅₀ would likely have been more than 10,000 ng/mL. Therefore, we believe that the only legitimate conclusion to be drawn from these experiments is that the binding of bevacizumab to mVEGF-A is minimal and, in any event, dramatically lower than the binding to hVEGF-A. 

The authors also attempted to verify the same interaction by plasmon resonance with BIAcore. However, they describe only a qualitative interaction and state that mVEGF-A showed “significantly different dissociation behavior” relative to hVEGF-A. Considering that plasmon resonance is exquisitely suited for quantification of antibody–antigen interactions, ²⁸ it is likely that some readers will find it surprising that Bock et al. ¹² did not provide a dissociation constant or any quantitative parameter. In previous studies, using the same technique, we found the interaction between bevacizumab and mVEGF-A to be so weak that it was difficult to determine an accurate dissociation constant. ²⁴ As described in Results, we were unable to detect any significant binding using the conditions described by Bock et al. ¹² However, considering that the difference between a very weak and an undetectable interaction can be subtle and possibly dependent on minor experimental variables, such as the batch of mVEGF-A used or the state of the protein after storage and manipulation, the reasons for such apparent discrepancy may be difficult to determine. It is also noteworthy that the BIAcore raw data shown by Bock et al. ¹² were obtained using excessively high levels of immobilized antibody because the amount of hVEGF-A bound under these conditions was approximately 600 RU. Given that the VEGF-A dimer has approximately one fourth the molecular mass of an intact IgG, the data suggest that more than 2000 RU of bevacizumab was immobilized to obtain these sensorgrams. Such high levels of immobilized protein, in addition to promoting avidity effects for bivalent analytes, ²⁹ tend to yield mass transport kinetics effects, making the relevance of such interactions to physiological binding in solution, to say the least, questionable. 

How then can a biochemical interaction between bevacizumab and mVEGF-A that, by several criteria, is so weak as to be essentially negligible underlie potent in vivo effects? Bock et al. ¹² did not provide a crucial verification of the hypothesis that such interaction may be biologically meaningful: the demonstration that bevacizumab is able to inhibit the activity of mVEGF-A. Interestingly, they show that bevacizumab blocks, as expected, the mitogenic effects of hVEGF-A in an assay measuring proliferation of lymphatic endothelial cells. ¹² However, they make no mention of the ability of bevacizumab to inhibit the activity of mVEGF-A, in this or any other bioassay. 

As illustrated in Figure 3 , with the use of an endothelial cell proliferation assay, we were unable to detect any significant inhibition of mVEGF-A by bevacizumab, even at a very high antibody/antigen molar ratio. However, bevacizumab showed the expected inhibition of hVEGF-A in the same assay. Instead, the cross-reactive mAb G6-31 equally inhibited endothelial cell proliferation stimulated by mVEGF-A and by hVEGF-A (Fig. 3B) . In vivo studies confirmed the lack of a biologically relevant interaction between bevacizumab and mVEGF-A. In laser-induced CNV and B16F1 melanoma growth, bevacizumab had no effects, even when tested at a dose as high as 50 mg/kg twice weekly. In contrast, mAb G6-31 had a dramatic inhibitory effect in both models. 

Based on the available evidence, it seems reasonable to conclude that it is unlikely the in vivo findings described by Bock et al. ¹² result from a specific immunoneutralization of mVEGF-A by bevacizumab. Whether the explanation is the inherent variability of the animal models, the nonspecific effects inadequately controlled by the reagents employed, or some unusual, perhaps model-dependent, phenomenon linked to the interaction of a humanized protein with the immune system of the mouse remains to be established. Considering that several effective inhibitors are available to investigate the role of VEGF-A in murine models, using bevacizumab in such studies may be an inappropriate use of resources and animals and may generate confusion in the field. 

 Submitted for publication September 10, 2007; accepted December 17, 2007.

 Disclosure: L. Yu, Genentech, Inc. (F, I, E, R); X. Wu, Genentech, Inc. (F, I, E, R); Z. Cheng, Genentech, Inc. (F, I, E, R); C.V. Lee, Genentech, Inc. (F, I, E, R); J. LeCouter, Genentech, Inc. (F, I, E, R); C. Campa, Genentech, Inc. (F, I, E, R); G. Fuh, Genentech, Inc. (F, I, E, R); H. Lowman, Genentech, Inc. (F, I, E, R); N. Ferrara, Genentech, Inc. (F, I, E, R)

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

Western blot analysis of recombinant hVEGF-A and mVEGF-A using bevacizumab or mAb G6-31. E. coli-expressed recombinant human VEGF-A (0.5 or 2 μg) or Sf 21 insect cell–expressed recombinant murine VEGF-A were probed with bevacizumab or mAb G6-31 (see Materials and Methods). Immunoreactive signals were detected by autoradiographic film after 1- or 10-minute exposure.

Figure 2.

 

Surface plasmon resonance measurement of hVEGF-A and mVEGF-A binding to bevacizumab or mAb G6-31. hVEGF-A (20–50 nM) or two sources of mVEGF-A were injected over the chip (arrow). Recombinant hVEGF₁₆₅ was E. coli expressed (Genentech, Inc.). Recombinant mVEGF₁₆₄ was from R&D Systems (Sf21 insect cell expressed) or Chemicon, Inc. (E. coli expressed). The net binding sensorgram shown here was acquired by subtracting the binding RU with that of the same injection on a control blank chip. After the binding interaction, the chips were regenerated by a short injection of 20 mM HCl (dashed line arrow).

Figure 3.

 

Bevacizumab does not inhibit mVEGF-A–stimulated endothelial cell proliferation. Bevacizumab or mAb G6-31 was added at the indicated concentrations (ng/mL) to triplicate wells. After 0.5 hour, hVEGF-A or mVEGF-A was added to a final concentration of 6 ng/mL. After 6 to 7 days, cell growth was determined by monitoring fluorescence at 590 nm. N/A, basal growth with no VEGF or antibody added. Error bars = SD. *P < 0.05, using ANOVA by comparing the antibody-treated group with the antibody-untreated group.

Figure 4.

 

Bevacizumab does not inhibit neovascularization in the laser-induced choroid injury mouse model. (A) Representative images of three areas of laser injury, performed 2 to 3 disc diameters from the optic nerve in the 9, 12, and 6 o’clock positions. Treatment groups included mAb A4.6.1, mAb G6-31, bevacizumab, or hIgG given twice weekly intraperitoneally at 5 mg/kg for a total of three doses over the 10-day study. (B) Choroid neovascular areas are reported as the overall mean value for each of four treatment groups. The neovascular area was quantified by NIH ImageJ 1.37V. For statistical analysis, one-way ANOVA was used, followed by the Tukey-Kramer HSD test comparing all pairs. *P < 0.05.

Figure 5.

 

Bevacizumab does not inhibit the growth of B16F1 melanoma cells in nude mice. Forty-eight hours after inoculation of murine B16F1 melanoma cells (day 0), mice (n = 10) were injected intraperitoneally with control IgG or bevacizumab at 5 or 50 mg/kg. A positive control group received mAb G6-31 at 5 mg/kg. Antibodies were administered twice weekly thereafter. Tumor volumes were measured with calipers at the indicated time points. *P < 0.05 compared with control IgG groups by Student’s t-test.