c-Met is a well-characterized receptor tyrosine kinase for hepatocyte growth factor (HGF). Compelling evidence from studies in human tumors and both cellular and animal tumor models indicates that signaling through the HGF/c-Met pathway mediates a plethora of normal cellular activities, including proliferation, survival, migration, and invasion, that are at the root of cancer cell dysregulation, tumorigenesis, and tumor metastasis. Inhibiting HGF-mediated signaling may provide a novel therapeutic approach for treating patients with a broad spectrum of human tumors. Toward this goal, we generated and characterized five different fully human monoclonal antibodies that bound to and neutralized human HGF. Antibodies with subnanomolar affinities for HGF blocked binding of human HGF to c-Met and inhibited HGF-mediated c-Met phosphorylation, cell proliferation, survival, and invasion. Using a series of human-mouse chimeric HGF proteins, we showed that the neutralizing antibodies bind to a unique epitope in the β-chain of human HGF. Importantly, these antibodies inhibited HGF-dependent autocrine-driven tumor growth and caused significant regression of established U-87 MG tumor xenografts. Treatment with anti-HGF antibody rapidly inhibited tumor cell proliferation and significantly increased the proportion of apoptotic U-87 MG tumor cells in vivo. These results suggest that an antibody to an epitope in the β-chain of HGF has potential as a novel therapeutic agent for treating patients with HGF-dependent tumors. (Cancer Res 2006; 66(3): 1721-9)
- fully human antibodies
- therapeutic antibodies
The receptor for hepatocyte growth factor (HGF), c-Met, is a well-characterized receptor tyrosine kinase that mediates a variety of epithelial cell functions, and dysregulation of the HGF/c-Met pathway has been implicated in a wide variety of human malignancies ( 1, 2). 6 There is evidence for both paracrine and autocrine HGF activation of c-Met in human cancer; overexpression of c-Met and/or HGF is frequently observed and amplification of the c-met gene has been reported in several tumor types. In addition, several kinase-activating mutations in c-Met have been described in hereditary and sporadic human cancers ( 3). Transgenic and knock-in expression of some of these mutant alleles in mice confirms the enhanced tumorigenicity of these mutant c-Met kinases in vivo ( 4, 5). Several of these activating c-Met mutants are sensitive to ligand stimulation in vitro ( 6). Importantly, c-Met and/or HGF overexpression and especially elevated concentrations of circulating HGF correlate with tumor stage and/or poor patient prognosis ( 1, 7– 9).
HGF, the only known ligand for c-Met, is normally secreted as an inactive single-chain glycoprotein of ∼90 kDa from cells of mesenchymal origin. Extracellular activation by proteolysis generates the fully active, two-chain (mature), form of HGF ( 10). The NH2-terminal portion of HGF comprises the α-chain and contains the NH2-terminal hairpin domain and four kringle domains. The COOH-terminal β-chain of HGF shares extensive structural homology with serine proteases; however, amino acid substitutions in the catalytic triad suggest that it is proteolytically inactive ( 11). Previous studies showed that NH2-terminal fragments of HGF (e.g., NK1, NK2, and NK4) contain the c-Met receptor binding domain and can act as HGF agonists and/or antagonists, thus demonstrating that HGF binds c-Met and functions through determinants in the α-chain ( 10, 12, 13).
The HGF/c-Met pathway mediates a plethora of normal cellular activities that are at the root of cancer cell dysregulation, including proliferation, survival, migration, invasion, and branching morphogenesis ( 1, 2). Most of these cellular processes are recapitulated during epithelial-mesenchymal transitions occurring during both embryonic development and tumorigenesis ( 14). Transgenic expression of HGF in mice leads to the development of a wide spectrum of tumors, many of which express increased levels of c-Met, suggesting that autocrine activation of this axis is sufficient to trigger tumorigenesis ( 15– 17). Using a transgenic, severe combined immunodeficiency model expressing human HGF, Zhang et al. ( 18) showed that paracrine HGF expression can also drive tumor growth.
Data from human tumor biology, cellular and animal tumor models provide compelling evidence that the HGF/c-Met axis plays an important role in tumorigenesis and metastatic disease. To test this hypothesis, we generated and characterized fully human monoclonal antibodies (mAbs) that completely neutralized HGF binding to its receptor c-Met and inhibited HGF/c-Met–mediated in vitro and in vivo activities. These anti-HGF antibodies had exceptional antitumor activity in autocrine HGF/c-Met–dependent human tumor xenograft models of glioblastoma. Our results suggest that that an antagonist antibody to HGF has potential as a novel therapeutic agent for treating cancer patients.
Materials and Methods
Recombinant human d5-HGF ( 19) was purchased (R&D Systems, Inc., Minneapolis, MN) or was expressed by Chinese hamster ovary (CHO) cells and purified by heparin sulfate affinity chromatography. Protein A–purified c-Met Fc was expressed using CHO cells. Hybridoma-expressed antibodies were purified from culture supernatants by protein A affinity chromatography. Light and heavy chains were cloned and sequenced from the hybridomas and expression vectors were constructed for CHO expression.
Human IgG2 mAbs to HGF were generated using XenoMouse technology (Abgenix, Fremont, CA; ref. 20). Mice were immunized with recombinant human d5-HGF. The preparation used was native, biologically active human HGF containing about half processed (mature, two-chain) and half unprocessed (single chain) human d5 HGF. Mice with the highest anti-HGF titers were sacrificed and their draining lymph nodes were harvested. B cell–enriched lymphocytes were fused with P3 myeloma cells to create hybridomas. Supernatants from hybridoma lines were screened for disruption of HGF/c-Met binding and inhibition of cellular phosphorylation. Purified mAbs from cloned hybridomas were characterized for their ability to completely block HGF binding to c-Met and HGF-mediated c-Met phosphorylation in PC3 cells.
Kinetic Measurement Antibody-antigen KD Values
Using a BIAcore 3000 (Biacore, Inc., Piscataway, NJ), affinity analysis of each of the anti-HGF antibodies was done according to the instructions of the manufacturer. Recombinant protein G (Pierce, Rockford, IL) was immobilized on a research-grade CM5 sensor chip using the Amine Coupling kit, and ∼200 resonance units of each of the antibodies were separately attached to immobilized protein G chips. Samples composed of 0 to 100 nmol/L of human HGF were injected over the antibody surface and antibody binding kinetic variables, including ka, kd, and KD, were determined using the BIA evaluation 3.1 software.
Receptor-ligand Binding Assay
Delphia plates (Wallac, Inc., Gaithersburg, MD) were coated with human d5 HGF. Samples contained 2 nmol/L soluble c-Met Fc and 10 serial dilutions of each antibody. Samples were added to HGF-coated wells and c-Met binding was detected with biotinylated anti-c-Met (BAF358, R&D Systems) and Eu-streptavidin (Wallac). Signal enhancement and time-resolved fluorescence detection were done as recommended (Wallac). The data were evaluated by calculating the percentage inhibition compared with the maximal signal (control antibody or no antibody added) and the IC50 values were calculated using the Xlfit 4-parameter equation (Excel version 2.0.6, Microsoft, Seattle, WA). Figure 1A shows one representative experiment; similar results were obtained in several additional experiments.
Fluorescence-activated Cell Sorting Assay for Antibody-HGF Binding
A mammalian expression vector (pCep4avidin-N) was used to generate avidin-HGF fusion/chimeric proteins ( Fig. 2A and C ). This vector contains cDNA sequence encoding recombinant chicken avidin adjacent to a multiple cloning site for insertion of a specific target gene fusion partner, in this case human or mouse HGF. 7 The resulting constructs were transiently expressed in 293T cells and conditioned medium was collected. The avidin-HGF fusion/chimeric proteins were captured using biotin-coated beads (Spherotech, Inc., Libertyville, IL), and the resulting bead-protein complexes were incubated separately with each of the anti-HGF antibodies. The avidin portion of the complex was labeled by FITC antiavidin antibody (Vector Laboratories, Burlingame, CA) and the presence of human antibodies to HGF in the complex was detected by phycoerythrin-labeled goat anti-human F(ab′)2 antibody (Southern Biotech Association, Birmingham, AL). All of the expression constructs were sequenced and the functional activity of the fusion/chimeric proteins was tested in phosphorylation assays in mouse and human cells (data not shown). The chimeras were constructed as follows: chimera A composed of an NH2-terminal portion of human HGF (amino acids 32-506) and a COOH-terminal portion of mouse HGF (amino acids 507-728); chimera B composed of an NH2-terminal portion of mouse HGF (amino acids 33-506) and a COOH-terminal portion of human HGF (amino acids 507-728); chimera C composed of an NH2-terminal portion of human HGF (amino acids 32-585) and a COOH-terminal portion of mouse HGF (amino acids 586-728); and chimera D composed of an NH2-terminal portion of mouse HGF (amino acids 33-585) and a COOH-terminal portion of human HGF (amino acids 586-728).
Cells and Culture
U-87 MG, U118, PC3, and MDA-MB-435 cells were all obtained from the American Type Culture Collection (Manassas, VA) and cultured according to their recommendations. Human umbilical vascular endothelial cells (HUVEC) were obtained from Cambrex BioScience Walkersville, Inc. (Walkersville, MD).
Functional Cell Assays
c-Met phosphorylation assay. PC3 cells were plated in growth medium, and, after 24 hours, the cells were starved for 18 to 20 hours. c-Met phosphorylation was stimulated with 200 ng/mL (2.2 nmol/L) human HGF for 10 minutes. Each of the antibodies was prepared as a 10-point serial dilution series in the stimulation medium before adding to the cells. Following the 10-minute incubation, cells were immediately placed on ice and lysed. Electrochemiluminescence (ECL) Igen assays were done using biotin-labeled goat-anti-c-Met (R&D Systems) for capture and antiphosphotyrosine antibody, 4G10, as the detection antibody. Typically, a signal-to-noise ratio of 4-fold was achieved with this assay format. The IC50 values were determined as above. Figure 1B shows one representative experiment; similar results were obtained in several additional experiments.
Immunoblots of downstream signaling molecules. PC3 cells were treated as described above using HGF at 200 ng/mL with or without neutralizing anti-HGF antibodies at 100 nmol/L for 10 minutes. Cell lysates were prepared and Western blots were done using standard procedures with the following antibodies: phosphorylated c-Met, phosphorylated Gab1, phosphorylated Erk1/Erk2 (Cell Signaling Technology, Inc., Beverly, MA). Immunoprecipitations were carried out using standard procedures and captured on protein G beads. The subsequent immunoblot was probed with antibodies to c-Met (C12, Santa Cruz Biotechnology, Santa Cruz, CA), phosphatidylinositol 3-kinase (PI3K; Upstate USA, Inc., Charlottesville, VA), or SHP2 (BD Biosciences, San Jose, CA).
HUVEC proliferation assay. A scintillation proximity assay was used to measure [14C]thymidine incorporation into HUVECs. HUVECs were plated on 96-well scintillant plates (Amersham Biosciences Corp., Piscataway, NJ) and 2 days later, 50 ng/mL (0.55 nmol/L) human HGF and [14C]thymidine was added to the cells alone or mixed with 10 concentrations of each antibody. After 3 days, the radioactivity incorporated into the cell monolayer was determined and the percentage inhibition of the maximal signal was calculated. IC50 values were determined as above. Typically, a 3-fold signal-to-noise ratio was obtained with this assay. Figure 3A shows one representative experiment; similar results were obtained in several additional experiments.
U-87 MG proliferation/survival assay. U-87 MG cells were plated in medium containing 5% fetal bovine serum, and the following day the cells were treated with increasing concentrations of antibody. After 7 days, the cells were fixed with 10% trichloroacetic acid and stained with 0.4% sulforhodamine B in 1% acetic acid. The dye was solubilized and the absorbance at 540 nm was measured. The maximum inhibition caused by anti-HGF treatment was ∼50% reduction in the signal compared with untreated cells. The percentage of maximum inhibition was determined and the IC50 values were calculated. Figure 3B shows one representative experiment; similar results were obtained in several additional experiments.
MDA-MB-435 invasion assay. The MDA-MB-435 Matrigel invasion assay used Matrigel-coated FluoroBlok Invasion System inserts from BD Biosciences, Inc. (San Jose, CA). The cell attractant medium contained 50 ng/mL (0.55 nmol/L) CHO-derived HGF. Each antibody was diluted to 10 μg/mL in the cell attractant medium and triplicate samples for each condition were evaluated. The cell suspension was added to the upper chamber and incubated at 37°C for 18 hours. The cells were then stained with Cell Tracker Green (Molecular Probes, Eugene, OR) to detect the invaded cells on the bottom of the filter. The fluorescent intensities of migrated cells were quantified with the Victor 2 (1420 Multilabel Counter, Wallac) using the bottom-reading mode. Typically, a 3-fold signal-to-noise ratio was obtained with this assay. The invasion results were also examined using a Zeiss Axiovert 35 inverted epifluorescence microscope and images were recorded using a SONY Optronics camera system and video printer. Representative data are shown in Fig. 3C and D; similar results were obtained in several other experiments.
Tumor Xenograft Models
Tumor growth and measurements. Eight to 10-week-old female CD1 nude mice (Charles River Laboratories, Wilmington, MA) were used in all experiments. Mice were injected s.c. in the flank with 5 × 106 U-87 MG or U118 cells in 0.2 mL DMEM. Antibodies were administered by i.p. injection twice per week either the day after cell implantation or when tumors were further established. Tumor measurements and body weights were recorded twice per week. Tumor volume was calculated as length × width × height in mm3. Results are expressed as mean ± SE. Mice were euthanized by CO2 inhalation, and in some experiments tumors were collected and processed for histologic analysis. These studies were conducted in accordance with federal animal care guidelines and were preapproved by the Amgen Institutional Animal Care and Use Committee.
Harvest methodology and immunohistochemical analysis. Four hours before harvest, each mouse received an i.p. injection of bromodeoxyuridine (BrdUrd; 50 mg/kg) to label S-phase nuclei. Mice were then sacrificed and tumors were resected from the mouse flank, retaining a “cap” of the overlying skin. The tumors were bisected following the long axis and one-half was fixed in zinc-immunohistochemistry solution ( 21). After 40 to 48 hours in zinc-immunohistochemistry solution, the specimens were briefly rinsed in water and transferred to 70% ethanol and processed into paraffin. Sections were cut and stained for BrdUrd, caspase-3, or CD31 by immunohistochemistry.
For BrdUrd immunohistochemistry, 5 μm midsagittal tumor cross-sections were stained for the incorporated thymidine analogue BrdUrd using a rat monoclonal anti-BrdUrd antibody at a 1:200 dilution (Accurate Chemical and Scientific Corp., Westbury, NY), peroxidase localization with 3,3′-diaminobenzidine (DAB) substrate, and a light hematoxylin-only counterstain. Four widely spaced ×20 objective fields were selected in a compass-point fashion from the circumferential region of each tumor section. The number of BrdUrd-positive tumor cells was counted for each ×20 objective field of roughly 135,600 μm2.
For caspase-3 immunohistochemistry, 5 μm midsagittal tumor cross-sections were incubated with a 1:50 dilution of rabbit polyclonal antibody directed against cleaved caspase-3 (Cell Signaling Technologies), labeled with a biotinylated antirabbit secondary antibody, and stained with DAB substrate and a light hematoxylin counterstain. Caspase-3 grading was done to evaluate treatment effect by quantitation of caspase-3–positive tumor cells per viable tumor area. All tumor cells that stained positive were counted and recorded in a randomized blinded fashion. Tumor area was then acquired from a low-magnification image to measure caspase-3–positive cells per mm2 viable tumor tissue. These data were subsequently normalized by treatment day to the concurrent day control.
For CD31 immunohistochemistry, 5 μm midsagittal tumor cross-sections were blocked with CAS BLOCK (Zymed Laboratories, San Francisco, CA) and incubated with a purified rat monoclonal anti-CD31 (platelet/endothelial cell adhesion molecule 1) at 10 μg/mL (BD PharMingen, San Diego, CA). The antibody was detected with biotinylated rabbit anti-rat immunoglobulins (Vector Laboratories). Slides were quenched with peroxidase blocking solution (DAKO Corp., Carpinteria, CA) and followed with Vectastain Elite ABC kit (Vector Laboratories). Reaction sites were visualized with DAB+ substrate-chromagen system (DAKO) and lightly counterstained with hematoxylin.
Four widely spaced ×10 objective fields, selected in a compass-point fashion from the circumferential region of each tumor section, were photographed using a Nikon DXM1200 camera (A.G. Heinze, Inc., Lake Forrest, CA). The DAB-positive CD31 signal areas, corresponding to the vascular endothelium, were determined via RGB thresholding using the image analysis software MetaMorph (v6.1, Universal Imaging Corporation, Downingtown, PA) and subsequently expressed as a percentage of the tumor field image area.
Statistical analysis. Tumor volume data were statistically analyzed with repeated-measures ANOVA for tumor volumes over time using StatView software version 5.0.1. Scheffe's post hoc test was used to determine P values for repeated measures from days 9 to 30. ANOVA followed by Bonferroni/Dunn was used to calculate P values for tumor weights and for BrdUrd cell proliferation. Caspase-3 data from treated animals were normalized and compared with concurrent treatment day control using a two-tailed Student's t test for sample sets with unequal variance.
Generation and biochemical characterization of fully human mAbs that neutralize HGF. Fully human IgG2 mAbs to HGF were generated using the d5 splice variant of human HGF ( 19) as the immunogen and XenoMouse technology to generate hybridomas ( 20). Human IgG2 mAbs, from single clones of five hybridomas, with robust binding and function blocking activities for human HGF were identified and designated as 1.29.1, 1.74.1, 1.75.1, 2.4.4, and 2.12.1.
Using a BIAcore 3000, we analyzed the affinity of the antibodies to HGF. The kinetic parameters of antibody-antigen show that each antibody displays sub-nmol/L affinity constants (KD; Table 1 ).
A quantitative, solid-phase, binding-neutralization assay was developed to assess whether individual antibodies disrupted c-Met-HGF binding. Figure 1A illustrates the percentage inhibition of HGF binding to c-Met with representative data from a single experiment. Calculated IC50 values are presented in Table 1. Each mAb was able to completely block binding between HGF and c-Met with IC50 values in the range of 1 to 5 nmol/L.
We next determined whether the anti-HGF antibodies bound to full-length human and mouse HGF. In addition, we examined antibody binding to the major splice variant of human HGF, d5 HGF (which lacks five amino acids in the first NH2-terminal kringle domain; ref. 19). The HGF fusion proteins shown schematically in Fig. 2A were transiently expressed in 293T cells, and a novel fluorescence-activated cell scanning (FACS) method 6 was used to evaluate and compare the binding specificity of the antibodies to HGF. Fusion protein/antibody complexes were detected using FITC-labeled (anti-avidin) and phycoerythrin-labeled (anti-human Ig) antibodies; the FACS scatter plots ( Fig. 2B) show that each of the four antibodies tested bound the human HGF fusion proteins but none bound to the mouse HGF fusion protein despite the ∼90% identity of these two proteins. The small upward shift seen for antibody 1.75.1 was not confirmed by Western blot analysis. All five antibodies bound human but not mouse HGF based on immunoblotting of the antigen/mAb complexes on nonreducing SDS-PAGE gels (data not shown).
By generating mouse/human chimeric fusion proteins of HGF (Chimeric protein A, B, C, and D; Fig. 2C), we were able to determine what portion of human HGF was responsible for binding of anti-HGF antibodies 2.4.4 and 2.12.1. Antibody 2.12.1 bound specifically to chimera B but not to chimeras A, C, or D, suggesting that its epitope is at least partially contained within the β-portion of human HGF ( Fig. 2D, left). Antibody 2.4.4 bound to chimera B and weakly to C but not to A or D, suggesting that it also binds in the β-portion of human HGF ( Fig. 2D, right). These data indicate that 2.4.4 and 2.12.1 bind with high affinity to epitopes in the β-portion of human HGF. Furthermore, competition-binding studies between pairs of antibodies suggested that all of the anti-HGF antibodies bound to either overlapping or even identical β-chain epitopes (data not shown). Because the β-chain of HGF shares sequence and especially structural homology with other proteins, we tested whether these anti-HGF antibodies could inhibit the related protein, MSP (MSP is the closest relative of HGF sharing 45% sequence identity). We were unable to detect any inhibition of binding or function by anti-HGF on MSP/Ron (data not shown).
Anti-HGF antibodies neutralize human HGF-mediated cellular activities. We next measured inhibition of HGF-mediated c-Met phosphorylation by anti-HGF antibodies. PC3 cells (human prostate carcinoma) do not express HGF; thus, this assay mimicked a paracrine model of ligand-mediated receptor activation. When anti-HGF antibody was introduced into HGF-containing medium before its addition to the cells, c-Met phosphorylation was completely inhibited in a dose-dependent manner ( Fig. 1B). Each of the antibodies blocked HGF-mediated c-Met phosphorylation with IC50 values in the range of 0.34 to 1.3 nmol/L ( Table 1).
HGF stimulation of c-Met activates multiple downstream signaling pathways, including ras/Erk and PI3K/AKT ( 22). To assess the ability of these antibodies to inhibit signaling downstream from HGF/c-Met, PC3 cells were treated without and with HGF and antibodies 2.12.1 or 2.4.4 for 10 minutes. Equivalent amounts of cell lysates/protein were subjected to immunoblotting for phosphorylated c-Met, phosphorylated Gab1, or phosphorylated Erk1/Erk2. HGF treatment stimulated phosphorylation of substrates and downstream signaling molecules, and each of the tested antibodies showed essentially complete inhibition of the stimulation ( Fig. 1C). To detect signaling complexes, we immunoprecipitated Gab1 and then probed the immunoblots for c-Met, PI3K, or SHP2. We detected the formation of an HGF-induced Gab1 complex with PI3K and with SHP2, and each of these interactions was prevented with HGF neutralizing anti-HGF antibody 2.12.1 or 2.4.4 ( Fig. 1C).
The HGF/c-Met axis has been implicated in angiogenesis ( 23, 24); hence, we measured [14C]thymidine incorporation into HUVEC DNA as a model for inhibition of endothelial cell proliferation and survival. Figure 3A presents data on the inhibition of thymidine incorporation by each anti-HGF mAb; the IC50 values span a broad range (1-200 nmol/L; Table 1). In similar experiments, HUVECs were stimulated with HGF, vascular endothelial growth factor (VEGF), or basic fibroblast growth factor; only the HGF-mediated proliferation/survival was inhibited by any of the anti-HGF antibodies tested (data not shown).
Autocrine expression of the HGF/c-Met axis has been implicated in driving the growth of glioblastoma in humans ( 25, 26). U-87 MG human glioblastoma cells provide an appropriate cellular model as they express both HGF and c-Met, and their proliferation and/or survival are at least partially driven by an autocrine signaling loop. When the antibodies were added to U-87 MG cells in culture, four of the five antibodies inhibited U-87 MG proliferation/survival with IC50 values of 20 to 60 nmol/L. Antibody 1.75.1 was less efficient in neutralizing in this assay (IC50 ∼300 nmol/L; Fig. 3B; Table 1). Interestingly, autocrine signaling in the U-87 MG line through c-Met both in vitro and in vivo seems to be quite modest when assessed by the phosphorylation status of either c-Met or its downstream targets. We detect, at best, 2- to 3-fold decreases in pAKT, phospho-p70S6 kinase, and phospho-RbS6 in cells or tumors from animals treated with anti-HGF 2.4.4 or 2.12.1 (data not shown).
HGF induces c-Met-dependent migration and invasion in a variety of epithelial-derived tumor cell lines. Several studies have further suggested that this cellular activity is responsible for invasion of tumor cells in breast cancer ( 27, 28). We did quantitative Matrigel invasion assays using MDA-MB-435 breast carcinoma cells. Each anti-HGF antibody significantly inhibited invasion with mean inhibition values ranging from 40% to 100% ( Fig. 3C and D).
Anti-HGF mAbs inhibit autocrine HGF/c-Met–driven tumor growth. The antibodies described here do not cross-react with mouse HGF ( Fig. 2); thus, our choice of in vivo models was limited to human HGF-driven cell lines. U-87 MG and U118 cells express both human HGF and c-Met, and their proliferation and survival in culture are at least partially driven by an autocrine-signaling loop ( Fig. 3B; data not shown). To evaluate the effect of individual anti-HGF mAbs on tumor growth, we used these two independently derived human glioblastoma cell lines as s.c. xenografts. Two different protocols were used: (a) a minimal residual disease model, where anti-HGF antibody treatment begins immediately following the s.c. implantation of cells, and (b) a more therapeutically relevant, established disease model, where tumors are allowed to grow to ∼200 mm3 before treatment begins (see also ref. 29).
We evaluated the relative effect of anti-HGF antibodies on human HGF-dependent tumor growth by treating nude mice injected s.c. with U-87 MG human glioblastoma cells with one of the five mAbs. As shown in Fig. 4A , anti-HGF treatment significantly inhibited tumor growth, compared with animals treated with an isotype control antibody (P < 0.0001). At this low dose of 5 μg/mouse twice per week, only antibodies 1.29.1 and 2.12.1 completely inhibited tumor growth. As expected, tumor weights were also markedly reduced at the end of the experiment and correlated with tumor volumes (data not shown). At higher doses, each of these antibodies completely inhibited tumor growth in this model (data not shown). To further show the specificity of these anti-HGF antibodies, we have shown that antibody 2.12.1, even at very high doses, does not significantly inhibit growth of the prostate cancer cell line, 22Rv1. Although this cell line expresses levels of HGF similar to that of the glioblastoma models, it is c-Met negative and thus are not capable of being driven by an HGF/c-Met autocrine loop (data not shown).
We next determined whether treatment with this mAb could result in growth inhibition and regression of established tumors. Animals were treated with 2.12.1 twice per week beginning on day 11 when tumors had reached an average size of 180 mm3. By day 29, treatment with 2.12.1 resulted in statistically significant inhibition of tumor growth when compared with treatment with the isotype control antibody (P < 0.0001; Fig. 4B). In addition, groups treated with 30 or 10 μg of antibody per dose resulted in statistically significant tumor regression when comparing tumor volume on days 11 to 29 (P < 0.0001, unpaired t test). Seven of 10 animals in the 30 μg dose group had no measurable tumor mass at the end of the experiment. In similar experiments, antibody 2.4.4 also caused significant growth inhibition and U-87 MG tumor regression (data not shown).
In nude mice bearing established, s.c. U118 human glioblastoma tumors, treatment with antibody 2.12.1 or IgG2 control antibody began on day 11 when tumors reached an average size of 200 mm3. By day 28, treatment with antibody 2.12.1 at 300 or 100 μg resulted in statistically significant inhibition of tumor growth when compared with treatment with the isotype control (P < 0.0001; Fig. 4C). Compared to the U-87 MG tumor model, the U118 tumors were less sensitive to anti-HGF treatment; higher doses were required to achieve complete growth inhibition and none of the dose groups showed significant tumor regression. In all of the tumor xenograft experiments done, there was no evidence of overt toxicity based on body weight and overall appearance of the treated animals (data not shown).
Mechanism of tumor growth inhibition and regression. To investigate the mechanism through which HGF neutralization inhibits tumor growth, U-87 MG tumor xenografts from mice treated with antibody 2.12.1 were examined by histology. After reaching an average size of ∼350 mm3, tumors established in nude mice were harvested at various times after a single dose of 300 μg of 2.12.1 ( Fig. 5 ). Antibody 2.12.1 caused an immediate decrease in tumor cell proliferation compared with control ( Fig. 5A-C) and this inhibition was sustained for at least 72 hours ( Fig. 5C). The results from BrdUrd staining established that 2.12.1 significantly decreased the rate of proliferation of the U-87 MG tumor cells in vivo.
We quantitatively analyzed tumor cell apoptosis using caspase-3 activation on tumor sections from the same experiment described above. A single dose of antibody 2.12.1 (300 μg) caused a rapid increase in the number of tumor cells that stained positive for cleaved caspase-3, indicating an induction of apoptosis ( Fig. 5D-F). After 24 hours, the response was ∼2-fold over control (P = 0.0026) and by 48 and 72 hours the increase in apoptosis was ∼7-fold (P = 0.0003) and 9-fold (P = 0.00079) over control, respectively ( Fig. 5F).
HGF/c-Met has been implicated in the control of tumor angiogenesis ( 23, 24). To evaluate whether tumor-derived human HGF plays a role in maintenance of the tumor through the established tumor vasculature, quantitative image analysis was done on tumors that were stained with the endothelial cell marker, CD31 ( Fig. 5G-I). Administration of a single dose of 2.12.1 (300 μg) did not cause a significant decrease in the area of CD31-positive blood vessels during a time frame of at least 72 hours ( Fig. 5I).
In this article, we identified and characterized five fully human mAbs that potently neutralize human HGF. These antibodies act by binding human, but not mouse, HGF with sub-nmol/L affinities and block HGF binding to its receptor c-Met. The antibodies seem to require the HGF β-chain, likely in the region between amino acids 507 and 585, for high affinity binding. All five antibodies inhibited HGF-driven responses in cells, including c-Met phosphorylation, cell proliferation, survival, and invasion. The anti-HGF antibodies inhibited tumor growth in autocrine HGF/c-Met–driven xenograft models of glioblastoma. Treatment of tumor-bearing animals with anti-HGF antibody resulted in significant regression of established U-87 MG tumors, and inhibited tumor growth in a second, HGF/c-Met autocrine tumor model, U118. Tumor regressions were not seen in our studies with the U118 xenograft model nor by Cao et al. ( 30) using multiple anti-HGF antibodies. The mechanism responsible for the different responses of these two GBM models is not understood. Anti-HGF antibody directly inhibited tumor cell proliferation and significantly increased the proportion of apoptotic tumor cells in the U-87 MG xenograft tumors. However, anti-HGF antibody treatment did not affect the established tumor vasculature. It is possible that other angiogenesis-promoting factors secreted by the tumor (e.g., VEGF; ref. 31), mouse stromal cells (e.g., angiopoietins; ref. 29), or mouse HGF may be sufficient to maintain and/or drive tumor angiogenesis in this model.
Previous studies have used polyclonal antibodies to inhibit the HGF/c-Met axis. Recently, Cao et al. ( 30) presented evidence that a combination of three antibodies was required for neutralizing activity in cell assays and significantly inhibited tumor growth in the U118 xenograft model. In contrast, our results show that a single anti-HGF mAb to an epitope in the β-chain of HGF completely inhibits HGF-mediated activities.
We took advantage of our finding that these antibodies did not bind to mouse HGF and designed a series of mouse/human HGF chimeric proteins to identify the portion of the human protein responsible for antibody binding. It has previously been shown that HGF binds to its receptor c-Met by binding interactions with the NH2-terminal portion of HGF. Individual subdomains from the α-chain, such as NK1, NK2, and NK4, have distinct agonist and/or antagonist activities on c-Met, and they compete for binding with the full-length HGF protein ( 12, 13, 32– 34). Thus, our finding that the neutralizing epitope resides, at least in part, in the β-chain of HGF was unexpected. Recently, however, structural data showed that the COOH-terminal β-chain of HGF is also involved in the interaction with the receptor through the c-Met sema domain ( 35, 36). The emerging concept is that interactions between c-Met and HGF involve multiple points of interaction, and the binding and neutralization data of our antibodies suggest that the interaction between the β-chain of HGF and c-Met is of critical functional importance. We cannot exclude the possibility that our antibodies also interact with the NH2-terminal α-chain of HGF through conformationally close contacts between the two domains; further studies are under way to directly address this issue.
Growth inhibition and tumor regression induced by anti-HGF antibody in the U-87 MG model seems to be driven by both inhibition of proliferation and enhancement of apoptosis in the tumor cells. As is the case for many human tumors, U-87 MG and U118 cells do not express a functional PTEN tumor suppressor ( 37), and thus the PI3K/AKT survival pathway is constitutively activated (data not shown). Despite this, both of these tumor models are growth inhibited by neutralizing the ligand/receptor interaction upstream of the PTEN tumor suppressor defect. This suggests that PTEN status in tumors may not be a barrier to therapeutic response with neutralizing antibodies to HGF or other targets upstream of PTEN.
HGF/c-Met signaling has been extensively studied and the evidence for its involvement in driving a number of human malignancies is compelling. Our data suggest that a single antagonist antibody to HGF has potential as a novel therapeutic agent for treating patients with a broad spectrum of human tumors.
Grant support: Amgen or Abgenix.
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.
We thank the many contributions to this work from our colleagues at Amgen, including the members of the anti-HGF Research, Development and Product Teams, as well as the Departments of Oncology Research, Protein Sciences, Pathology, Laboratory Animal Resources, Medical Writing, and the Law Department; Jean Gudas for initiating the program and getting it off to a successful start; George Vande Woude and Donald Bottaro for critically reviewing the manuscript; and David Lacey and Glenn Begley for their leadership, guidance, and support.
Note: K. Zhang is currently in Laboratory of Genetics, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland. X-C. Jia is currently in Agensys, Inc., Santa Monica, California.
↵7 G. Elliott et al., in preparation.
- Received September 15, 2005.
- Revision received November 9, 2005.
- Accepted December 2, 2005.
- ©2006 American Association for Cancer Research.