MEDICALSCIENCES
`
`VEGF-Trap: A VEGF blocker with potent
`antitumor effects
`
`Jocelyn Holash*, Sam Davis, Nick Papadopoulos, Susan D. Croll, Lillian Ho, Michelle Russell, Patricia Boland, Ray Leidich,
`Donna Hylton, Elena Burova, Ella Ioffe, Tammy Huang, Czeslaw Radziejewski, Kevin Bailey, James P. Fandl, Tom Daly,
`Stanley J. Wiegand, George D. Yancopoulos, and John S. Rudge
`
`Regeneron Pharmaceuticals, Incorporated, 777 Old Saw Mill River Road, Tarrytown, NY 10591
`
`Communicated by P. Roy Vagelos, Merck & Co., Inc., Bedminster, NJ, July 2, 2002 (received for review April 19, 2002)
`
`Vascular endothelial growth factor (VEGF) plays a critical role
`during normal embryonic angiogenesis and also in the pathological
`angiogenesis that occurs in a number of diseases, including cancer.
`Initial attempts to block VEGF by using a humanized monoclonal
`antibody are beginning to show promise in human cancer patients,
`underscoring the importance of optimizing VEGF blockade. Previ-
`ous studies have found that one of the most effective ways to block
`the VEGF-signaling pathway is to prevent VEGF from binding to its
`normal receptors by administering decoy-soluble receptors. The
`highest-affinity VEGF blocker described to date is a soluble decoy
`receptor created by fusing the first three Ig domains of VEGF
`receptor 1 to an Ig constant region; however, this fusion protein
`has very poor in vivo pharmacokinetic properties. By determining
`the requirements to maintain high affinity while extending in vivo
`half life, we were able to engineer a very potent high-affinity VEGF
`blocker that has markedly enhanced pharmacokinetic properties.
`This VEGF-Trap effectively suppresses tumor growth and vascular-
`ization in vivo, resulting in stunted and almost completely avas-
`cular tumors. VEGF-Trap-mediated blockade may be superior to
`that achieved by other agents, such as monoclonal antibodies
`targeted against the VEGF receptor.
`
`The sprouting of new blood vessels, termed angiogenesis, is
`
`required to support growth in the embryo and young animal,
`as well as to allow for repair and remodeling processes in the
`adult. However, aberrant angiogenesis is also associated with a
`number of pathological conditions and diseases, including cancer
`(1, 2). Tumors, like many normal tissues, use the vasculature as
`a means to obtain oxygen and nutrients and to remove waste
`products. Although tumors can in part grow by coopting existing
`host vessels (3–6), most tumors also induce new vessel formation,
`suggesting that this neovascularization is required for their
`growth (1, 2, 7). Consequently, much effort has been directed
`toward discovering antiangiogenic agents and evaluating them as
`cancer therapeutics. Perhaps the best characterized and most
`highly validated antiangiogenic approach involves targeting the
`vascular endothelial growth factor (VEGF) pathway (1, 8–11).
`Based on numerous animal studies, the VEGF pathway is the
`only well-defined signaling pathway known to be required for
`normal development of the vasculature as well as for the
`pathologic angiogenesis that accompanies cancer and other
`disease states (8–10).
`The VEGF pathway is initiated when VEGF binds to its
`receptors on endothelial cells. The two best characterized VEGF
`receptors are termed VEGF receptor 1 (VEGFR1) and VEGF
`receptor 2 (VEGFR2). VEGFR1 and VEGFR2 are highly
`related transmembrane tyrosine kinases that use their ectodo-
`mains to bind VEGF; this binding in turn activates the intrinsic
`tyrosine kinase activity of their cytodomains, initiating intracel-
`lular signaling. Interestingly, although VEGFR1 binds to VEGF
`with substantially higher affinity, most of the biologic effects of
`VEGF seem to be mediated via VEGFR2. In animals, blockade
`of the VEGF pathway has been achieved by many different
`means, including blocking antibodies targeted against VEGF
`(12–14) or its receptors (15), soluble decoy receptors that
`
`prevent VEGF from binding to its normal receptors (16–20), as
`well as small molecule inhibitors of the tyrosine kinase activity
`of the VEGFRs (21–23). Recently, a study that compared the
`efficacy of VEGF blockade to other ‘‘antiangiogenic’’ strategies
`established that this approach is superior to many others (ref.
`11). Consistent with predictions from animal studies, blockade
`of VEGF using a humanized monoclonal antibody has emerged
`as the first and thus far only antiangiogenesis approach reporting
`promising results in human cancer patients, based on prelimi-
`nary reports from early clinical trials.† The hope is that anti-
`VEGF approaches can be generalized to many different types of
`cancer, as well as to other diseases in which pathologic angio-
`genesis contributes, such as diabetic retinopathy and psoriasis.
`The clinical promise of initial anti-VEGF approaches high-
`lights the need to optimize blockade of this pathway. Previous
`studies have found that one of the most effective ways to block
`the VEGF signaling pathway is to prevent VEGF from binding
`to its normal receptors by administering decoy VEGF receptors
`(11, 16, 17, 24). The highest-affinity VEGF blocker described to
`date is a soluble decoy receptor created by fusing the first three
`Ig domains of VEGFR1 to the constant region (Fc portion) of
`human IgG1, resulting in a forced homodimer that has picomolar
`binding affinity (16, 17). In tumor experiments, this VEGFR1-Fc
`reagent is efficacious at approximately 500-fold lower concen-
`tration than a similar VEGFR-2 construct (11). Despite its high
`affinity, the VEGFR1-Fc is not a feasible clinical candidate
`because of its poor pharmacokinetic profile; in rodent studies,
`this protein has to be administered frequently and at very high
`levels to achieve efficacious levels (16, 17, 24). In addition, the
`VEGFR1-Fc exhibits certain toxicological side effects that are
`not seen with the VEGFR2-Fc (11). These effects appear to be
`due to nonmechanism-based and nonspecific properties of this
`agent (see Discussion). By determining the requirements to
`maintain high affinity while extending in vivo half life, we were
`able to engineer a very potent high-affinity VEGF blocker that
`has prolonged in vivo pharmacokinetics and pharmacodynamics,
`lacks nonspecific toxicities, and can effectively suppress the
`growth and vascularization of a number of different types of
`tumors in vivo.
`
`Materials and Methods
`Engineering VEGF-Traps. The parental VEGF-Trap was created by
`fusing the first three Ig domains of VEGFR1 to the constant
`region (Fc) of human IgG1. VEGF-Trap⌬B1 was created by
`removing a highly basic 10-aa stretch from the third Ig domain
`of the parental VEGF-Trap. VEGF-Trap⌬B2 was created by
`removing the entire first Ig domain from VEGF-Trap⌬B1.
`VEGF-TrapR1R2 was created by fusing the second Ig domain
`
`Abbreviations: VEGF, vascular endothelial growth factor; VEGFR1, VEGF receptor 1;
`VEGFR2, VEGF receptor 2; AUC, area under the curve.
`
`*To whom reprint requests should be addressed. E-mail: jocelyn.holash@regeneron.com.
`†Yang, J., Haworth, L., Steinberg, S., Rosenberg, S. & Novotny, W. (2002) Am. Soc. Clin.
`Oncol. (abstr. 15).
`
`www.pnas.org兾cgi兾doi兾10.1073兾pnas.172398299
`
`PNAS 兩 August 20, 2002 兩 vol. 99 兩 no. 17 兩 11393–11398
`
`CELLTRION - EXHIBIT 1004
`
`

`

`of VEGFR1 with the third Ig domain of VEGFR2. All of the
`VEGF-Trap variants were produced and purified from Chinese
`hamster ovary cells.
`Pharmacokinetic Analysis of VEGF-Traps. BALB兾c mice (25–30 g)
`were injected s.c. with 4 mg兾kg of the various Traps and bled at
`1, 2, 4, 6, 24, 48, 72, and 144 hr after injection. Levels of all
`VEGF-Traps were measured by an ELISA by using human
`VEGF165 to capture and an antibody to the human Fc region as
`the reporter.
`
`Extracellular Matrix (ECM)-Binding Assay. ECM-coated plates
`(Becton Dickinson no. 35–4607) were incubated with varying
`concentrations of VEGF-Traps for 1 hr at room temperature.
`They were washed and incubated with alkaline phosphatase-
`conjugated anti-human Fc antibody (Promega, 1:4,000 in PBS ⫹
`10% BCS) for 1 hr at room temperature. Plates were washed four
`times with PBS ⫹ 0.1%Triton-X 100 and reagent buffer added
`for color development. Plates were read at 405–570 nm.
`
`VEGF-Trap-Binding Assay. Binding affinities of VEGF-Traps were
`measured by using a specific and sensitive ELISA (R&D Systems
`kit no. DVE00) for detecting free (unbound) human VEGF in
`mixtures of the VEGF-Traps (ranging in concentration from 0.1
`to 160 pM) with human VEGF165 (at 10 pM), incubated over-
`night at room temperature.
`
`Human Umbilical Vein Endothelial Cell Phosphorylation Assay. Con-
`fluent monolayers of human umbilical vein endothelial cells [Vec
`Technologies (Rensselaer, NY) passage no. 5] were serum-starved
`for 2 hr and then challenged for 5 min with vehicle or 40 ng兾ml of
`human VEGF165, alone or preincubated with VEGF-Traps at
`1.5-fold molar excess. Cells were then lysed, immunoprecipitated by
`using a VEGFR2-specific antibody, and immunoblotted with
`a phosphotyrosine-specific antibody (Upstate Biotechnology,
`4G10 mAb).
`
`VEGF-Induced Proliferation Assay. Cells that proliferate in response
`to VEGF were generated by stably transfecting NIH 3T3 cells
`with a VEGFR2兾TrkB chimeric receptor (in which the cytodo-
`main of VEGFR2 was replaced with that of TrkB, a receptor for
`brain-derived neurotrophic factor that effectively drives prolif-
`eration in these cells). Five thousand cells were plated per well
`of a 96-well plate, allowed to settle for 2 hr, incubated for 1 hr
`with VEGF-Trap variants (titrated from 40 nM to 20 pM), then
`challenged for 72 hr with human VEGF165 at a concentration of
`1.56 nM, followed by addition of [3-(4,5 dimethylthiazol-2-yl)-
`5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium,
`innersai and spectrophotometric analysis at 450兾570 nm.
`
`Acute Hypotension. Male Wistar–Kyoto rats (180–240 g) from
`Taconic Laboratories were maintained on a 12:12 light兾dark cycle
`(lights on 0600) with food and water available ad libitum. Before
`challenge with VEGF, animals were pretreated with VEGF-Traps
`or PBS as indicated, anesthetized with 1.5–2% isoflurane in oxygen,
`and the left femoral artery catheterized for direct measurement of
`systolic blood pressure through a blood pressure transducer (IITC,
`Woodland Hills, CA) into a chart recorder (Linseis, Princeton
`Junction, NJ). Animals were then injected in the right jugular vein
`with a 200-␮l bolus containing 10 ␮g of recombinant human
`VEGF165. Systolic blood pressure was measured before VEGF
`injection and every minute thereafter for 20 min. Blood pressures
`were normalized to baseline preinjection and analyzed by using
`mixed factorial ANOVAs (see supporting information on the
`PNAS web site, www.pnas.org).
`Tumor Growth Experiments. C6 glioma cells (1.0 ⫻ 106 cells兾mouse)
`and A673 rhabdomyosarcoma cells (2.0 ⫻ 106 cells兾mouse) were
`
`Engineering of VEGF-Traps with improved pharmacokinetics. (A)
`Fig. 1.
`Schematics of full-length VEGFR1 (red) and VEGFR2 (blue) are provided,
`indicating their seven Ig domains, transmembrane regions (black bars), and
`kinase domains (ovals). The parental VEGF-Trap contains the first three Ig
`domains of VEGFR1 (including the highly basic 10-aa stretch in Ig3, blue box)
`fused to the Fc portion of human IgG1. VEGF-Trap⌬B1 is identical to the
`parental VEGF-Trap, except that the basic stretch in Ig3 has been removed.
`VEGF-Trap⌬B2 is the same construct as ⌬B1, except that the first Ig domain has
`been removed. VEGF-TrapR1R2 possesses the second Ig domain of VEGFR1 and
`the third Ig domain of VEGFR2 fused to the Fc portion of human IgG1. (B) The
`four indicated VEGF-Traps were assayed in vitro for their capacity to bind to
`extracellular matrix, with only the parental VEGF-Trap and VEGF-Trap⌬B1
`demonstrating binding. (C) Pharmacokinetic analysis of the VEGF-Traps re-
`veals that the parental VEGF-Trap has the poorest profile, whereas VEGF-
`TrapR1R2 showed the best profile.
`
`obtained from American Type Culture Collection, and B16F10.9
`melanoma cells (5.0 ⫻ 105 cells兾mouse) were a generous gift from
`Charles Lin (Duke University, Durham, NC). Cells were suspended
`in serum-free medium and implanted s.c. on the shaved right flank
`of male C.B-17 SCID mice at the indicated concentrations. After
`tumor cell implantation and twice weekly thereafter for the dura-
`tion of the experiment, mice received a s.c. injection (at the nape of
`the neck) of vehicle (PBS ⫹ 0.5% glycerol), VEGF-Trap, or DC101
`(from American Type Culture Collection). After 2–3.0 weeks,
`animals were killed and tumors were measured ex vivo with calipers
`(tumor volume ⫽ length ⫻ width ⫻ height). For immunohisto-
`chemistry studies, mice were perfused with 4% paraformaldehyde,
`and tissue was processed as previously described (25).
`
`Results
`Reengineering Parental VEGF-Trap to Improve Its Pharmacokinetic
`Profile. On the basis of the previously reported high affinity of a
`soluble decoy receptor in which VEGFR1 is fused to the Fc
`portion of human IgG1 (16, 17), we produced this fusion protein
`to study its properties (see parental VEGF-Trap, Fig. 1A). Single
`s.c. injections of parental VEGF-Trap (4 mg兾kg) into mice were
`
`11394 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.172398299
`
`Holash et al.
`
`

`

`MEDICALSCIENCES
`
`Binding affinity and inhibitory properties of VEGF-Traps. (A) Affini-
`Fig. 2.
`ties of indicated VEGF-Traps for VEGF, as determined by using a binding assay
`that measures unbound VEGF (ordinate) after incubation of 10 pM of human
`VEGF165 with varying concentrations of VEGF-Traps (abscissa). (B) Inhibition of
`VEGF-induced phosphorylation of VEGFR2 in human umbilical vein endothe-
`lial cell phosphorylations using indicated VEGF-Traps at 1.5-fold molar excess,
`as revealed with immunoblotting assay. (C) Inhibition of VEGF-induced pro-
`liferation of fibroblasts containing a chimeric VEGFR2兾TrkB receptor, using
`varying concentrations of VEGF-Traps in the presence of 1.56 nM of VEGF.
`
`analyses show that VEGF-TrapR1R2 has a kD of⬇ 1–10 pM for
`VEGF121 and approximately 45 pM for placental growth factor 2
`(not shown); other VEGF isoforms and relatives have not been
`analyzed.
`To determine whether Trap binding of VEGF could potently and
`effectively block the ability of VEGF to activate its receptor, VEGF
`and Traps were added to cultured endothelial cells, and the effects
`on VEGFR2 phosphorylation were examined, revealing that both
`parental VEGF-Trap as well as VEGF-TrapR1R2 can completely
`block VEGF-induced VEGFR2 phosphorylation when added at a
`1.5-fold molar excess compared with the added VEGF, consistent
`with very high-affinity binding to VEGF (Fig. 2B). Finally, to assess
`whether these Traps would also be effective in cell-based prolifer-
`ation assays, we engineered a cell line containing a chimeric
`VEGFR2 receptor that mediates a very strong proliferative re-
`sponse to VEGF and found that both parental VEGF-Trap and
`VEGF-TrapR1R2 potently blocked VEGF-induced proliferation in
`3-day growth assays in these cells, with an IC50 at approximately an
`equimolar concentration of Trap with the added VEGF, once again
`consistent with very high-affinity binding of the Traps for VEGF
`(Fig. 2C).
`
`VEGF-TrapR1R2 Provides Long-Term Blockade of Exogenously Admin-
`istered VEGF-Induced Acute Hypotension. The above studies indi-
`cated that VEGF-TrapR1R2 was at least as impressive a blocker
`of VEGF as the parental version, but that it had far superior
`pharmacokinetic properties. To initially explore whether these
`attributes translated into superior pharmacodynamic perfor-
`mance, we compared these reagents by using an acute readout
`of VEGF responsiveness in vivo. Administration of a single bolus
`dose (10 ␮g) of recombinant VEGF165 to rats results in acute
`hypotension, with a drop of about 40% from baseline systolic
`blood pressure; this drop is maximal at 5 min and slowly rectifies
`to normal by about 30 min (Fig. 3A). To compare the pharma-
`codynamic efficacy of the VEGF-Traps in blocking this acute
`response, we preadministered the parental VEGF-Trap or
`
`performed to confirm that it indeed displayed poor pharmaco-
`kinetic properties, with a maximal concentration (Cmax) of only
`0.05 ␮g兾ml and total ‘‘area under the curve concentration’’
`(AUC) of 0.04 ␮g ⫻ days兾ml (Fig. 1C). We postulated that these
`poor pharmacokinetic properties might be due to the high
`positive charge of this protein (pI 9.4), which in turn may result
`in its deposition at the site of s.c. injection because of nonspecific
`adhesion to highly negatively charged proteoglycans that com-
`prise the extracellular matrix. To test this hypothesis, we next
`engineered several variants of the parental VEGF-Trap with
`reduced positive charges. On review of the charge density in the
`parental molecule, we noted a highly basic stretch of 10 amino
`acids in the third Ig domain of VEGFR1 (see blue box in Fig.
`1 A). To reduce the charge, this region was excised, resulting in
`a decrease in the pI of this VEGF-Trap (termed VEGF-Trap⌬B1;
`see Fig. 1 A) from 9.4 to 9.1. It was also noted that the first Ig
`domain of VEGFR1 had a basic pI, and we thus decided to test
`removal of this domain as well as the above-noted basic region,
`resulting in a protein termed VEGF-Trap⌬B2 (Fig. 1 A), with a
`further reduced pI of 8.9. Finally, because the third Ig domain of
`VEGFR2 has a lower pI than the corresponding domain of
`VEGFR1, we simply switched these domains to make a Trap in
`which the second Ig domain of VEGFR1 is directly fused to the
`third Ig domain of VEGFR2; this trap was termed VEGF-
`TrapR1R2 (Fig. 1 A) and had a pI of 8.82. Previous structural
`analyses indicated that VEGFR1 might make greater use of its
`second Ig domain in contacting VEGF, whereas VEGFR2
`instead makes greater use of its third Ig domain (26), raising the
`interesting and useful possibility that VEGF-TrapR1R2 might
`actually bind more tightly to VEGF than the parental versions.
`Combining the distinct binding regions of two different recep-
`tors to create a higher-affinity interactor has previously been
`used in the creation of a series of interleukin and cytokine
`blockers also termed Traps (A. Economides, L. Rocco Carpen-
`ter, J.S.R., V. Wong, E. Koehler-Stec, C. Hartnett, E. Pyles, T.D.,
`M. Young, J.P.F., Frank Lee, Scott Carver, Jennifer McNay,
`K.B., S. Ramakanth, R. Hatabarat, C.R., T.H., G.D.Y., and
`N. Stahl, unpublished results). Using a simple extracellular
`matrix-binding assay, we then confirmed the hypothesis that
`decreasing the positive charge of the VEGF-Traps would result
`in decreased adhesion to extracellular matrix (Fig. 1B). Binding
`to extracellular matrix in this assay was directly related to the pI
`of the Traps, with both VEGF-TrapR1R2 and VEGF-Trap⌬B2
`displaying negligible binding in this assay.
`On the basis of the above results, we next tested these various
`VEGF-Traps in vivo for their pharmacokinetic behavior. Their in
`vivo behavior followed the theoretical charge predictions as well as
`the in vitro adhesion properties. Every reduction in pI was accom-
`panied by a corresponding improvement in Cmax and AUC: VEGF-
`Trap⌬B1 had a Cmax of 1.3 ␮g兾ml and an AUC of 1.36 ␮g ⫻ days兾ml;
`VEGF-Trap⌬B2 had a Cmax of 2.65 ␮g兾ml and an AUC of 5.42 ␮g ⫻
`days兾ml; whereas VEGF-TrapR1R2 revealed the best profile with a
`Cmax of 16 ␮g兾ml and an AUC of 36.28 ␮g ⫻ days兾ml (Fig. 1C).
`Thus, VEGF-TrapR1R2 had an AUC that was almost 1,000-fold
`higher than that of the parental VEGF-Trap, raising the possibility
`that it might be a far superior pharmacologic agent, assuming it
`retained its ability to bind and block VEGF.
`
`Comparison of Parental VEGF-Trap with VEGF-TrapR1R2 in Binding,
`Phosphorylation, and Cell Proliferation Assays inVitro. Because of the
`superior pharmacokinetic properties of VEGF-TrapR1R2, we next
`compared this Trap to its parent for its ability to bind and block
`VEGF in vitro. To determine binding affinity of the Traps for
`VEGF, equilibrium binding assays were performed in which dif-
`ferent concentrations of the Traps were incubated with VEGF165,
`and the amount of unbound VEGF165 was measured, revealing that
`parental VEGF-Trap displays a kD of ⬇5 pM, whereas VEGF-
`TrapR1R2 has a binding affinity of about 1 pM (Fig. 2A). Preliminary
`
`Holash et al.
`
`PNAS 兩 August 20, 2002 兩 vol. 99 兩 no. 17 兩 11395
`
`

`

`Using blockade of VEGF-induced acute hypotension to pharmaco-
`Fig. 3.
`dynamically compare VEGF-Traps. (A) When rats were treated with VEGF-Traps at
`25 mg兾kg at 1 day before VEGF challenge, VEGF-TrapR1R2 (n ⫽ 8) completely
`blocked VEGF-induced hypotension, whereas PBS (n ⫽ 6) and parental VEGF-Trap
`(n ⫽ 6) were ineffective. ANOVA shows treatment effect, P ⬍ 0.007. (B) At a 5-fold
`lower dose (5 mg兾kg), VEGF-TrapR1R2 was still effective at 1 day (n ⫽ 4) or 3 days
`(n ⫽ 3) before the VEGF challenge. ANOVA shows treatment effect, P ⬍ 0.03.
`
`VEGF-TrapR1R2 at 25 mg兾kg, 24 hr before VEGF administra-
`tion (Fig. 3A). Consistent with what would be expected from the
`above pharmacokinetic studies, this dose of VEGF-TrapR1R2
`completely blocked VEGF-induced hypotension, whereas the
`parental VEGF-Trap had no discernable effect. Thus, although
`the parental VEGF-Trap and its VEGF-TrapR1R2 derivative are
`quite comparable in vitro (see above), the VEGF-TrapR1R2
`performs much better in vivo, presumably because of its dra-
`matically enhanced pharmacokinetic profile.
`To further characterize the length of time in which VEGF-
`TrapR1R2 remained efficacious, we waited 1, 3, and 7 days after
`injection of the Trap at 5 mg兾kg before inducing hypotension. At
`this dose, VEGF-TrapR1R2 was completely effective in blocking
`VEGF-induced acute hypotension at 1 and 3 days after a single
`bolus (Fig. 3B) but was not significantly different from controls
`at 7 days (data not shown).
`
`VEGF-TrapRiR2 Dramatically Blocks Tumor Growth inVivo. Altogether,
`the above pharmacokinetic and pharmacodynamic studies indi-
`cated that VEGF-TrapR1R2 has the potential to be a long-term
`and potent pharmacologic blocker of VEGF-mediated activities
`in vivo, far superior to that of parental VEGF-Trap. To begin to
`explore the value of VEGF-TrapR1R2 as an anticancer thera-
`peutic and to compare it to other effective agents targeting the
`VEGF pathway, we evaluated its ability to block the growth of
`a variety of tumor cell lines in s.c. tumor models in mice. Tumor
`cells were derived from diverse tissue origins and different
`species (mouse B16F10.9 melanoma, human A673 rhabdomyo-
`sarcoma, and rat C6 glioma). After implantation of tumor cells,
`mice were allowed a brief recovery period and then received s.c.
`injections of VEGF-TrapR1R2 (25 mg兾kg) or vehicle twice weekly
`for the duration of the experiment (2–3.0 weeks), after which the
`
`Fig. 4. VEGF-TrapR1R2 dramatically inhibits the s.c. growth and vascularity of
`implanted tumors from diverse tissues and species. (A) VEGF-Trap R1R2 sub-
`stantially blocked the growth of the indicated s.c. implanted tumors, at the
`indicated doses twice weekly for 2 weeks (C6 and B16F10.9) or 3.0 weeks
`(A673). Error bars represent standard error of mean, n ⫽ five mice兾treatment
`group. The differences between control tumor volumes and VEGF-TrapR1R2–
`treated tumor volumes were analyzed by using Student’s t tests and found to
`be significant at the following levels: B16F10 P ⫽ 0.01; A673 P ⫽ 0.06; C6 P ⬍
`0.0001. (B–D) Histological analysis reveals that VEGF-TrapR1R2 can effectively
`block blood vessel growth in these implanted tumors. Sections of C6 tumors
`stained with antibodies to platelet– endothelial cell adhesion molecule reveal
`that vehicle-treated animals had large tumors that were highly vascularized
`(B), whereas animals treated with 25 mg兾kg VEGF-TrapR1R2 (C) had tumors that
`were largely avascular with large areas of necrosis (N). Viable tumor appeared
`to be vascularized because of cooption of preexisting host vessels (white
`arrowheads) associated with hypodermal musculature (M) and dermis. Treat-
`ment with 2.5 mg兾kg VEGF-TrapR1R2 greatly stunted tumor growth (C) and
`resulted in large necrotic regions (N), although small pockets of vessels were
`occasionally apparent (black arrows). (Bar ⫽ 100 ␮m.)
`
`animals were killed and tumors excised and measured. VEGF-
`TrapR1R2 significantly inhibited the growth of all three types of
`tumors (Fig. 4A). In the study using C6 glioma cells, a 10-fold
`lower dose of VEGF-TrapR1R2 (2.5 mg兾kg) was tested and found
`to be equally effective at inhibiting tumor growth.
`To evaluate the effects of VEGF-TrapR1R2 on tumor-associated
`angiogenesis, the tumors from the above studies were sectioned and
`immunostained with antibodies to platelet–endothelial cell adhe-
`sion molecule, so that the vasculature could be visualized (Fig. 4
`B–D). This analysis revealed that the higher dose of VEGF-
`TrapR1R2 almost completely blocked tumor-associated angiogene-
`sis, with the stunted tumors being largely avascular, save for regions
`in which preexisting host vessels appeared to be coopted by
`surrounding tumor (see open arrowheads, Fig. 4C). The lower dose
`of VEGF-TrapR1R2, which was quite comparable at inhibiting
`tumor growth (see above), appeared to be slightly less effective at
`completely blocking tumor-associated angiogenesis, allowing for
`small pockets of tumor-associated vessels in otherwise avascular
`tumors (see black arrowheads in Fig. 4D). In contrast to the
`VEGF-Trap-treated tumors, control tumors in vehicle-treated mice
`not only were much larger (see above) but also had a very high
`vascular density (Fig. 4B).
`
`11396 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.172398299
`
`Holash et al.
`
`

`

`MEDICALSCIENCES
`
`against a wide assortment of cancers. In addition, pathological
`angiogenesis seems to contribute to a number of non-neoplastic
`diseases, such as diabetic retinopathy (31) and psoriasis (32),
`extending the potential utility of anti-VEGF therapeutics. All this
`promise highlights the need to optimize anti-VEGF approaches.
`Herein we describe the engineering of an anti-VEGF agent, termed
`VEGF-TrapR1R2. VEGF-TrapR1R2 is a derivative of perhaps the
`most potent VEGF binder known, VEGFR1. Soluble forms of
`VEGFR1 suffer from poor pharmacokinetic properties, which
`seem to correlate with their nonspecific interactions with extracel-
`lular matrix. VEGF-TrapR1R2 was engineered to have minimal
`interactions with extracellular matrix, and this property apparently
`accounts for its satisfying pharmacokinetic profile. The combina-
`tion of high-affinity and improved pharmacokinetics apparently
`contributes toward making VEGF-TrapR1R2 one of the most, if not
`the most, potent and efficacious VEGF blocker available. An
`additional advantage is that VEGF-TrapR1R2 is composed of en-
`tirely human sequences, hopefully minimizing the possibility that it
`might prove immunogenic in human patients. Despite its wholly
`human nature, VEGF-TrapR1R2 binds all species of VEGF tested,
`from human to chicken VEGF (not shown), making it a very
`versatile reagent that can be used in almost any experimental
`animal models.
`A recent study comparing numerous antiangiogenesis ap-
`proaches concluded that anti-VEGF approaches were the most
`efficacious (11). The particular anti-VEGF agent used for these
`studies was essentially equivalent to our parental VEGF-Trap
`but was delivered in an adenoviral system in which it was highly
`expressed in the livers of infected animals. In contrast to other
`anti-VEGF approaches that seem to be well-tolerated, the
`adenovirally delivered VEGF-Trap caused severe liver toxicity
`and ascites, raising the possibility that it might have some unique
`mechanism-based side effects compared with other anti-VEGF
`approaches. To explore this possibility, we made adenoviral
`versions of both the parental VEGF-Trap as well as VEGF-
`TrapR1R2 and found that, whereas adenoviral delivery of paren-
`tal VEGF-Trap reproduces the previously reported toxicities
`(11), adenoviral delivery of VEGF-TrapR1R2 did not cause these
`side effects even though much higher levels were achieved in the
`circulation. Our conclusion is that the nonspecific interactions of
`the parental VEGF-Trap with extracellular matrix contribute to
`its increased toxicity after adenoviral administration, and that
`comparable toxicity is not noted with adenoviral administration
`of the engineered VEGF-TrapR1R2.
`In addition to the anticancer findings reported here, recent
`studies have shown that various versions of the VEGF-Trap can
`efficaciously treat a cancer-associated condition in mice similar
`to liver peliosis (33), as well as noncancer-associated disease
`models, such as of diabetic retinopathy (34–36) and psoriasis
`(Y.-P. Xia, M. Detmar, G.D.Y., and J.S.R., unpublished results).
`The accompanying manuscript (29) compares the efficacy of the
`VEGF-Trap to that of several other VEGF blockers, including
`a humanized monoclonal antibody to VEGF, in a model of
`kidney cancer. Among the several VEGF blockers tested, the
`VEGF-Trap shows the best overall efficacy. In this manuscript,
`we compare the efficacy of the VEGF-Trap to that of a
`monoclonal antibody to VEGFR2 in cancer models and find that
`far lower circulating levels of VEGF-TrapR1R2 are required for
`similar efficacy. Tumors treated with highest doses of the
`VEGF-Trap are not only stunted but also strikingly avascular.
`Our description of a VEGF blocker with such superior blocking and
`pharmacologic properties seems to demand that it be tested in
`human patients suffering from diseases involving neoangiogenesis.
`Toward this end, the safety of the VEGF-Trap has recently been
`confirmed in toxicological studies in cynomologus monkeys (data
`not shown). Consequently, the VEGF-Trap is currently in human
`clinical trials for several different types of cancer.
`
`VEGF-TrapR1R2 blocks tumor growth (of subcutaneously implanted
`Fig. 5.
`B16F10.9 cells) at far lower concentrations than DC101, a monoclonal antibody
`directed to VEGFR2. Mice were treated twice weekly with the indicated dose of
`VEGF-TrapR1R2, DC101, or vehicle. After 2.5 weeks, mice were killed, and tumors
`were excised and measured. Individual tumor volumes are shown (colored bars),
`as are average tumor volumes for each treatment (black bars) ⫾ SEM, n ⫽ six
`mice兾treatment group. Differences between treatment groups were analyzed by
`using a one-way ANOVA followed by Fisher’s protected least significant differ-
`ence test. Average volume of tumors in all treatment groups is significantly
`smaller than control tumor volume (P ⬍ 0.01). Differences in tumor volume
`between the high-dose VEGF-Trap, low-dose VEGF-Trap, and high-dose DC101
`treatment groups are not significantly different, but they are significantly dif-
`ferent from those of the low-dose DC101 treatment group (P ⬍ 0.02).
`
`VEGF-TrapR1R2 Compares Favorably with Antibodies Targeting VEGFR2.
`After establishing that VEGF-TrapR1R2 was effective at blocking
`s.c. tumor growth, we undertook studies to compare its efficacy with
`other known VEGF blockers. One particularly effective and well-
`characterized blocker is a monoclonal antibody, termed DC101,
`that targets VEGFR2 (15). When equimolar doses of VEGF-
`TrapR1R2 and DC101 were compared in the B16F10 melanoma
`model, it was apparent that much higher doses of DC101 are
`required to inhibit tumor growth (Fig. 5). Furthermore, because
`antibodies have longer circulation times in mice than simple Fc
`fusion proteins, the highly efficacious dose of DC101 accumulates
`to approximately 60-fold higher serum levels than that of the
`equally efficacious low dose of VEGF-Trap: circulating levels of
`DC101 in animals treated with the 40-mg兾kg dose were 2,442 ⫾ 272
`␮g兾ml, in contrast to the circulating levels of VEGF-Trap in
`animals treated with 3.2 mg兾kg, which were 40 ⫾ 8 ␮g兾ml. Thus,
`circulating levels of VEGF-Trap that were approximately 60-fold
`lower than those of DC101 were equally efficacious in inhibiting
`tumor growth. Importantly, the favorable allometric scaling of Fc
`fusion proteins relative to antibodies (27, 28) suggests that in
`humans the circulation time for the VEGF-Trap will be much more
`comparable to that of antibodies, which in turn suggests that in
`humans the difference in efficacious doses would be further mag-
`nified and may be as great as 60-fold.
`As described in an accompanying manuscript (29), when used
`at the same dose, VEGF-Trap shows efficacy equal to or better
`than a monoclonal antibody to VEGF (30). As noted above,
`because Fc fusion proteins have much shorter circulating half-
`lives than antibodies in mice, but comparable half-lives in
`humans, the finding that the VEGF-TrapR1R2 is at least as potent
`as the monoclonal antibody in mice suggests that the efficacious
`dose of VEGF-Trap will be much lower than that of the
`monoclonal antibody in humans.
`
`Discussion
`Validation of VEGF as an important new target in the war
`against cancer comes from pioneering clinical studies using a
`humanized monoclonal antibody that binds and blocks VEGF.†
`Because anti-VEGF approaches act by blocking tumor-a