{CANCER RESEARCH57, 4593-4599, October 15, 1997]
`
`Humanization of an Anti-Vascular Endothelial Growth Factor Monoclonal
`
`Antibody for the Therapy of Solid Tumors and Other Disorders
`
`Leonard G.Presta, Helen Chen, Shane J. O’Connor, Vanessa Chisholm, Y. Gloria Meng, Lynne Krummen,
`Marjorie Winkler, and Napoleone Ferrara’
`Departments of Immunology, Process Sciences, Molecular Biology, Bioanalytical Technology and Cardiovascular Research, Genentech,
`94080
`
`Inc., South San Francisco, California
`
`ABSTRACT
`
`Vascular endothelial growth factor (VEGF) is a major mediator of
`angiogenesis associated with tumors and other pathological conditions,
`including proliferative diabetic retinopathy and age-related macular
`degeneration. The murine anti-human VEGF monoclonal antibody
`(muMAb VEGF) A.4.6.1 has been shown to potently suppress angio-
`genesis and growth in a variety of human tumorcells lines transplanted
`in nude mice and also to inhibit neovascularization in a primate model
`of ischemic retinal disease. In this report, we describe the humaniza-
`tion of muMAbVEGFA.4.6.1. by site-directed mutagenesis of a human
`framework. Not only the residues involved in the six complementarity-
`determiningregions but also several framework residues were changed
`from humanto murine. Humanized anti-VEGF F(ab) and IgG1 vari-
`ants bind VEGF with affinity very similar to that of the original
`murine antibody. Furthermore, recombinant humanized MAb VEGF
`inhibits VEGF-induced proliferation of endothelial cells in vitro and
`tumorgrowthin vivo with potency andefficacy very similar to those of
`muMAb VEGFA.4.6.1. Therefore, recombinant humanized MAb
`VEGFis suitable to test the hypothesis that inhibition of VEGF-
`induced angiogenesis is a valid strategy for the treatment of solid
`tumors and other disorders in humans.
`
`INTRODUCTION
`
`It is now well established that angiogenesis is implicated in the
`pathogenesis of a variety of disorders. These include solid tumors,
`intraocular neovascular syndromessuchas proliferative retinopathies
`or AMD,? rheumatoidarthritis, and psoriasis (1, 2, 3). In the case of
`solid tumors, the neovascularization allows the tumorcells to acquire
`a growth advantage and proliferative autonomy compared to the
`normal cells. Accordingly, a correlation has been observed between
`density of microvessels in tumor sections and patient survival
`in
`breast canceras well as in several other tumors (4-6).
`The search for positive regulators of angiogenesis has yielded
`several candidates, including acidic fibroblast growth factor (FGF),
`bFGF, transforming growth factor a, transforming growth factor B,
`hepatocyte growth factor, tumor necrosis factor-a, angiogenin, inter-
`leukin 8, and others (1, 2). However, in spite of extensive research,
`there is still uncertainty as to their role as endogenous mediators of
`angiogenesis. The negative regulators thus far identified include
`thrombospondin (7), the M, 16,000 NH,-terminal fragmentof prolac-
`tin (8), angiostatin (9), and endostatin (10).
`Work doneoverthe last several years has established the key role
`of VEGFin the regulation of normal and abnormal angiogenesis(11).
`The finding that the loss of even a single VEGFallele results in
`
`embryonic lethality points to an irreplaceable role played by this
`factor in the development and differentiation of the vascular system
`(11). Also, VEGF has been shownto be a key mediator of neovas-
`cularization associated with tumors and intraocular disorders (11).
`The VEGF mRNAis overexpressed by the majority of human tumors
`examined (12-16). In addition, the concentration of VEGF in eye
`fluids is highly correlated to the presence of active proliferation of
`blood vessels in patients with diabetic and other ischemia-related
`retinopathies (17). Furthermore, recent studies have demonstrated the
`localization of VEGFin choroidal neovascular membranesin patients
`affected by AMD(18).
`The muMAbVEGFA.4.6.1 (19) has been used extensively to
`test the hypothesis that VEGFis a mediator of pathological angio-
`genesis in vivo. This high affinity MAb is able to recognize all
`VEGFisoforms (19) and has been shown to inhibit potently and
`reproducibly the growth of a variety of human tumorcell lines in
`nude mice (11, 20-23). Moreover, intraocular administration of
`muMAbVEGFA.4.6.1 resulted in virtually complete inhibition of
`iris neovascularization secondary to retinal ischemia in a primate
`model (24).
`A majorlimitation in the use of murine antibodies in human therapy
`is the anti-globulin response (25, 26). Even chimeric molecules, where
`the variable (V) domains of rodent antibodies are fused to human
`constant(C) regions,arestill capable ofeliciting a significant immune
`response (27). A powerful approach to overcomethese limitations in
`the clinical use of monoclonal antibodies is “humanization” of the
`murine antibody. This approach was pioneered by Joneset al. (28)
`and Riechmanetal. (29), whofirst transplanted the CDRsof a murine
`antibody into human V domainsantibody.
`In the present article, we report on the humanization of muMAb
`VEGFA.4.6.1. Our strategy wasto transfer the six CDRs, as defined
`by Kabat et al. (30), from muMAb VEGF A.4.6.1 to a consensus
`human framework used in previous humanizations (31-33). Seven
`framework residues in the humanized variable heavy (VH) domain
`and one framework residue in the humanized variable light (VL)
`domain were changed from human to murine to achieve binding
`equivalent to muMAb VEGFA.4.6.1. This humanized MAbissuit-
`able for clinical trials to test the hypothesis that inhibition of VEGF
`action is an effective strategy for the treatment of cancer and other
`disorders in humans.
`
`MATERIALS AND METHODS
`
`Received 5/27/97; accepted 8/16/97.
`The costs of publication ofthis article were defrayed in part by the payment ofpage
`charges. This article must therefore be hereby marked advertisement in accordance with
`18 U.S.C. Section 1734 solely to indicate this fact.
`'To whom requests for reprints should be addressed, at Department of Cardio-
`vascular Research, Genentech,
`Inc., 460 Point San Bruno Boulevard, South San
`Francisco, CA 94080. Phone: (415) 225-2968; Fax: (415) 225-6327; E-mail: Ferrara.
`Napoleone@gene.com.
`? The abbreviations used are: AMD, age-related macular degeneration; bFGF, basic
`fibroblast growth factor, VEGF, vascular endothelial growth factor; MAb, monoclonal
`antibody; muMAb, murine MAb; rhuMAb,recombinant humanized MAb; CDR,comple-
`mentarity-determining region.
`
`Cloning of Murine Mab A.4.6.1 and Construction of Mouse-Human
`Chimeric Fab. Total RNA wasisolated from hybridomacells producing
`the anti-VEGF MAb A.4.6.1 using RNAsol (Tel-Test) and reverse-tran-
`scribed to cDNA using Oligo-dT primer and the SuperScript II system (Life
`Technologies, Inc., Gaithersburg, MD). Degenerate oligonucleotide primer
`pools, based of the NH,-terminal amino acid sequences of the light and
`heavy chains of the antibody, were synthesized and used as forward
`primers. Reverse primers were based on framework 4 sequences obtained
`from murine light chain subgroup «V and heavy chain subgroupII (30).
`After PCR amplification, DNA fragments were ligated to a TA cloning
`vector (Invitrogen, San Diego, CA). Eight clones each of the light and
`Pfizer v. Genentech
`4593
`Pfizer v. Genentech
`IPR2017-01488
`IPR2017-01488
`Genentech Exhibit 2021
`Genentech Exhibit 2021
`
`

`

`HUMANIZATION OF AN ANTI-VEGF MONOCLONAL ANTIBODY
`
`heavy chains were sequenced. Oneclone with a consensus sequencefor the
`light chain VL domain and one with a consensus sequence for the heavy
`chain VH domain were subcloned, respectively, into the pEMX1 vector
`containing the human CL and CH1 domains(31), thus generating a mouse-
`human chimeric F(ab). This chimeric F(ab) consisted of the entire murine
`A.4.6.1 VH domain fused to a human CH1 domain at amino acid SerH1 13,
`and the entire murine A.4.6.1 VL domain fused to a human CL domain at
`
`centrifugation in a 1-liter centrifuge bottle at 3000 X g, and the supernatant
`was removed. After freezing for 1 h, the pellet was resuspended in 25 ml
`of cold 10 mM Tris, 1 mm EDTA,and 20% sucrose, pH 8.0. Two hundred
`fifty ml of 0.1 M benzamidine (Sigma Chemical Co., St. Louis, MO) was
`addedto inhibit proteolysis. After gentle stirring on ice for 3 h, the sample
`wascentrifuged at 40,000 < g for 15 min. The supernatant wasthen applied
`to a protein G-Sepharose CL-4B (Pharmacia Biotech, Inc., Uppsala, Swe-
`amino acid LysL107. Expression and purification of the chimeric F(ab)
`den) column (0.5-ml bed volume) equilibrated with 10 mm Tris-1 mm
`were identical to those of the humanized F(ab)s. The chimeric F(ab) was
`EDTA,pH 7.5. The column was washed with 10 ml of 10 mM Tris-1 mm
`used as the standard in the binding assays.
`EDTA,pH7.5, and eluted with 3 ml of 0.3 M glycine, pH 3.0,into 1.25 ml
`Computer Graphics Models of Murine and Humanized F(ab)s. Se-
`of 1 M Tris, pH 8.0. The F(ab) was then buffer exchanged into PBS using
`quencesof the VL and VH domains(Fig. 1) were used to construct a computer
`a Centricon-30 (Amicon, Beverly, MA) and concentratedto a final volume
`graphics model of the murine A.4.6.1 VL-VH domains. This model was used
`of 0.5 ml. SDS-PAGEgels ofall F(ab)s were run to ascertain purity, and
`to determine which framework residues should be incorporated into the hu-
`the molecular weight of each variant was verified by electrospray mass
`spectrometry.
`manized antibody. A model of the humanized F(ab) was also constructed to
`verify correct selection of murine framework residues. Construction of models
`Construction, Expression, and Purification of Chimeric and Human-
`was performed as described previously (32, 33).
`ized IgG Variants. For the generation of human IgG1 variants of chimeric
`Construction of Humanized F(ab)s. The plasmid pEMX1 used for mu-
`(chIgG1) and humanized (rhuMAb VEGF) A.4.6.1, the appropriate murine or
`humanized VL and VH (F(ab)-12; Table 1) domains were subcloned into
`tagenesis and expression of F(ab)s in Escherichia coli has been described
`separate, previously described pRK vectors (35). The DNA coding for the
`previously (31). Briefly, the plasmid contains a DNA fragment encoding a
`consensus human « subgroupIlight chain (VL«I-CL) and a consensus human
`entire light and the entire heavy chain of each variant was verified by
`subgroupIII heavy chain (VHIII-CH1) and an alkaline phosphatase promoter.
`dideoxynucleotide sequencing.
`The use of the consensus sequences for VL and VH has been described
`For transient expression of variants, heavy and light chain plasmids were
`previously (32).
`cotransfected into human 293cells (36) using a high efficiency procedure (37).
`the first F(ab) variant of humanized A.4.6.1, F(ab)-1,
`To construct
`Media were changed to serum free and harvested daily for up to 5 days.
`site-directed mutagenesis (34) was performed on a deoxyuridine-containing
`Antibodies were purified from the pooled supernatants using protein A-
`template of pEMX1. The six CDRs were changed to the murine A.4.6.1
`Sepharose CL-4B (Pharmacia). The eluted antibody was buffer exchanged into
`sequence; the residues included in each CDR were from the sequence-based
`PBS using a Centricon-30 (Amicon), concentrated to 0.5 ml, sterile filtered
`using a Millex-GV (Millipore, Bedford, MA), and stored at 4°C.
`CDRdefinitions (30). F(ab)-1, therefore, consisted of a complete human
`framework (VL « subgroup I and VH subgroupIII) with the six complete
`For stable expression of the final humanized IgG1 variant (rhuMAb
`murine CDR sequences. Plasmids for all other F(ab) variants were con-
`VEGF), Chinese hamster ovary (CHO) cells were transfected with dicis-
`structed from the plasmid template of F(ab)-1. Plasmids were transformed
`tronic vectors designed to coexpress both heavy and light chains (38).
`Plasmids were introduced into DP12 cells, a proprietary derivative of the
`into E. coli strain XL-1 Blue (Stratagene, San Diego, CA) for preparation
`of double- and single-stranded DNA.For each variant, DNA coding for
`CHO-K1 DUX B11 cell line developed by L. Chasin (Columbia University,
`light and heavy chains was completely sequenced using the dideoxynucle-
`New York, NY), via lipofection and selected for growth in glycine/
`otide method (Sequenase; U.S. Biochemical Corp., Cleveland, OH). Plas-
`hypoxanthine/thymidine (GHT)-free medium (39). Approximately 20 un-
`mids were transformed into E. coli strain 16C9, a derivative of MM294,
`amplified clones were randomly chosen and reseeded into 96-wellplates.
`plated onto Luria broth plates containing 50 yg/ml carbenicillin, and a
`Relative specific productivity of each colony was monitored using an
`ELISAto quantitate the full-length human IgG accumulated in each well
`single colony selected for protein expression. The single colony was grown
`in 5 ml of Luria broth-100 ug/ml carbenicillin for 5—8 h at 37°C. The 5-ml
`after 3 days and a fluorescent dye, Calcien AM,as a surrogate marker of
`viable cell number per well. Based on these data, several unamplified
`culture was added to 500 ml of AP5-50 g/ml carbenicillin and allowed to
`clones were chosen for further amplification in the presence of increasing
`grow for 20hin a 4-liter baffled shake flask at 30°C. AP5 media consists
`of 1.5 g of glucose, 11.0 g of Hycase SF, 0.6 g of yeast extract (certified),
`concentrations of methotrexate. Individual clones surviving at 10, 50, and
`0.19 g of MgSO, (anhydrous), 1.07 g of NH,Cl, 3.73 g of KCl, 1.2 g of
`100 nm methotrexate were chosen and transferred to 96-well plates for
`NaCl, 120 ml of 1 M triethanolamine, pH 7.4, to | liter of water and then
`productivity screening. One clone, which reproducibly exhibited high spe-
`sterile filtered through a 0.1-mm Sealkeenfilter. Cells were harvested by
`cific productivity, was expanded in T-flasks and usedto inoculate a spinner
`
`Table 1 Binding of humanized anti-VEGF F(ab) variants to VEGF?
`
`ECS0 F(ab)-X
`
`Variant
`chim-F(ab)
`F(ab)-1
`F(ab)-2
`
`F(ab)-3
`
`F(ab)-4
`
`Fcab)-5
`F(ab)-6
`F(ab)-7
`F(ab)-8
`
`Template
`Chimeric F(ab)
`Human FR
`
`Changes?
`
`F(ab)-1
`
`F(ab)-4
`F(ab)-5
`F(ab)-S
`F(ab)-5
`
`Purpose
`1.0
`Straight CDR swap
`Chimera light chain
`F(ab)-1 heavy chain
`F(ab)-1 light chain
`Chimera heavy chain
`CDR-H2 conformation
`ArgH71Leu
`Framework
`AspH73Asn
`VL-VHinterface
`LeuL46Val
`CDR-H1 conformation
`LeuH78Ala
`CDR-H2 conformation
`TleH69Phe
`CDR-H2conformation
`TleH69Phe
`CDR-H1 conformation
`LeuH78Ala
`>150
`CDR-H2conformation
`_GlyH49Ala
`F(ab)-8
`F(ab)-9
`6.4
`Framework
`AsnH76Ser
`F(ab)-8
`F(ab)-10
`3.3
`Framework
`LysH75Ala
`F(ab)-10
`F(ab)-11
`1.6
`CDR-H3conformation
`ArgH94Lys
`F(ab)-10
`F(ab)-12
`* Anti-VEGF F(ab)variants were incubated with biotinylated VEGFand then transferred to ELISA plates coated with KDR-IgG (40).
`> Murine residues are underlined; residue numbers are according to Kabat etal. (30).
`© Mean and SDare the average ofthe ratios calculated for each of the independent assays; the ECs, for chimeric F(ab) was 0.049 + 0.013 mg/ml (1.0 nm).
`4594
`
`Mean
`
`>1350
`>145
`
`2.6
`
`>295
`
`80.9
`36.4
`45.2
`9.6
`
`ECs chimeric F(ab)°
`SD
`
`0.1
`
`6.5
`4.2
`23
`0.9
`
`L2
`0.4
`0.6
`
`N
`
`2
`3
`
`2
`
`3
`
`2
`2
`2
`4
`
`2
`4
`2
`4
`
`

`

`HUMANIZATION OF AN ANTI-VEGF MONOCLONAL ANTIBODY
`
`RESULTS
`
`antibiotics (growth medium), essentially as described previously (42). For
`culture. After several passages, the suspension-adapted cells were used to
`mitogenic assays, endothelial cells were seeded at a density of 6 X 10°
`inoculate production cultures in GHT-containing, serum-free media sup-
`cells/well in 6-well plates in growth medium. Either muMAb VEGFA.4.6.1 or
`plemented with various hormonesand protein hydrolysates. Harvested cell
`rhuMAb VEGFwas then addedat concentrations ranging between | and 5000
`culture fluid containing rhuMAb VEGFwaspurified using protein A-
`Sepharose CL-4B. The purity after this step was ~99%. Subsequent
`ng/ml. After 2-3 h, purified E. coli-expressed rhWEGF,,. was addedtoafinal
`concentration of 3 ng/ml. For specificity control, each antibody was added to
`purification to homogeneity was carried out using an ion exchange chro-
`endothelial cells at the concentration of 5000 ng/ml, either alone or in the
`matographystep. The endotoxin content of the final purified antibody was
`<0.10 eu/mg.
`presence of 2 ng/ml bFGF. After 5 or 6 days, cells were dissociated by
`exposure to trypsin, and duplicate wells were counted in a Coulter counter
`F(ab) and IgG Quantitation. For quantitating F(ab) molecules, ELISA
`(Coulter Electronics, Hialeah, FL). The variation from the mean did not exceed
`plates were coated with 2 g/ml of goat anti-human IgG Fab (Organon
`Teknika, Durham, NC) in 50 mmcarbonate buffer, pH 9.6, at 4°C overnight
`10%. Data were analyzed by a four-parameter curvefitting program (Kalei-
`daGraph).
`and blocked with PBS-0.5% BSA (blocking buffer) at room temperature for
`In Vivo Tumor Studies. Human A673 rhabdomyosarcomacells (Amer-
`1 h. Standards (0.78-50 ng/ml human F(ab)] were purchased from Chemicon
`ican Type Culture Collection; CRL 1598) were cultured as described
`(Temecula, CA). Serial dilutions of samples in PBS-0.5% BSA-0.05% poly-
`previously in DMEM/F12 supplemented with 10% fetal bovine serum, 2
`sorbate 20 (assay buffer) were incubated on the plates for 2h. Bound F(ab) was
`mMglutamine, and antibiotics (20, 22). Female BALB/c nude mice, 6-10
`detected using horseradish peroxidase-labeled goat anti-human IgG F(ab)
`weeksold, were injected s.c. with 2 < 10° tumorcells in the dorsal area in
`(Organon Teknika), followed by 3,3’,5,5'-tetramethylbenzidine (Kirkegaard &
`a volume of 200 ul. Animals were then treated with muMAb VEGF
`Perry Laboratories, Gaithersburg, MD) as the substrate. Plates were washed
`A.4.6.1, rhuMAb VEGF,or a control murine MAbdirected against the
`between steps. Absorbance was read at 450 nm on a V,,,, plate reader
`gp120 protein. Both anti-VEGF MAbs were administered at the doses of
`(Molecular Devices, Menlo Park, CA). The standard curve was fit using a
`0.5 and 5 mg/kg; the control MAb wasgiven at the dose of 5 mg/kg. Each
`four-parameter nonlinear
`regression curve-fitting program developed at
`Genentech. Data points that fell in the range of the standard curve were used
`MAbwasadministered twice weekly i.p. in a volume of 100 yl, starting
`24 h after tumorcell inoculation. Each group consisted of 10 mice. Tumor
`for calculating the F(ab) concentrations of samples.
`size was determined at weekly intervals. Four weeks after tumor cell
`The concentration of full-length antibody was determined using goat anti-
`inoculation, animals were euthanized, and the tumors were removed and
`human IgG Fc (Cappel, Westchester, PA) for capture and horseradish perox-
`weighed. Statistical analysis was performed by ANOVA.
`idase-labeled goat anti-human Fc (Cappel) for detection. Human IgG1 (Chemi-
`con) was used as standard.
`VEGFBinding Assays. For measuring the VEGFbinding activity of
`F(ab)s, ELISA plates were coated with 2 ug/ml rabbit F(ab’), to human
`IgG Fe (Jackson ImmunoResearch, West Grove, PA) and blocked with
`blocking buffer (described above). Diluted conditioned medium containing
`3 ng/ml of KDR-IgG (40)in blocking buffer were incubated on the plate for
`I h. Standards [6.9—440 ng/ml chimeric F(ab)] and 2-fold serial dilutions
`of samples were incubated with 2 nm biotinylated VEGFfor 1 h in tubes.
`The solutions from the tubes were then transferred to the ELISA plates and
`incubated for 1 h. After washing, biotinylated VEGF bound to KDR was
`detected using horseradish peroxidase-labeled streptavidin (Zymed, South
`San Francisco, CA or Sigma)followed by 3,3',5,5’-tetramethylbenzidine as
`the substrate. Titration curves were fit with a four-parameter nonlinear
`regression curve-fitting program (KaleidaGraph; Synergy Software, Read-
`ing, PA). Concentrations of F(ab) variants corresponding to the midpoint
`absorbanceofthe titration curve of the standard were calculated and then
`
`Humanization. The consensus sequence for the human heavy
`chain subgroupIII and the light chain subgroup « I were used as the
`framework for the humanization (Ref. 30; Fig. 1). This framework has
`been successfully used in the humanization of other murine antibodies
`(31, 32, 43, 44). All humanized variants were initially made and
`screened for binding as F(ab)s expressed in E. coli. Typical yields
`from 500-ml shake flasks were 0.1—0.4 mg F(ab).
`Twodefinitions of CDR residues have been proposed. Oneis based
`on sequence hypervariability (30) and the other on crystal structures
`of F(ab)-antigen complexes (45). The sequence-based CDRs are
`larger than the structure-based CDRs, and the two definitions are in
`agreement except for CDR-H1; CDR-H1 includes residues H31—H35
`according to the sequence-based definition, and residues H26—H32
`according to the structure-based definition (light chain residue num-
`bers are prefixed with L; heavy chain residue numbers are prefixed
`BlAcore Biosensor Assays. VEGFbinding ofthe humanized and chimeric
`with H). We,therefore, defined CDR-H1 as a combination ofthe two,
`F(ab)s were compared using a BlAcore biosensor (41). Concentrations of
`i.e., including residues H26-H35. The other CDRswere defined using
`F(ab)s were determined by quantitative amino acid analysis. VEGF was
`the sequence-based definition (30).
`coupled to a CM-5 biosensor chip through primary amine groups according to
`The chimeric F(ab) wasusedas the standard in the binding assays.
`manufacturer's instructions (Pharmacia). Off-rate kinetics were measured by
`In the initial variant, F(ab)-1, the CDR residues were transferred from
`saturating the chip with F(ab) (35 jl of 2 um F(ab)at a flowrate of 20 l/min]
`the murine antibody to the human framework and, based on the
`and then switching to buffer (PBS-0.05% polysorbate 20). Data points from
`models of the murine and humanized F(ab)s, the residue at position
`0-4500 s were used for off-rate kinetic analysis. The dissociation rate constant
`H49 (Ala in humans) was changed to the murine Gly. In addition,
`(koe) Was obtained from the slope of the plot of In(RO/R) versus time, where RO
`F(ab)s that consisted of the chimeric heavy chain/F(ab)-1 light chain
`is the signal at t = 0 andRisthe signal at each time point.
`[F(ab)-2] and F(ab)-1 heavy chain/chimeric light chain [F(ab)-3] were
`On-rate kinetics were measured using 2-fold serial dilutions of F(ab)
`(0.0625-2 mm). The slope, K,, was obtained from the plot of In(—dR/dt)
`generated and tested for binding. F(ab)-1 exhibited a bindingaffinity
`versus time for each F(ab) concentration using the BIAcore kinetics eval-
`greater than 1000-fold reduced from the chimeric F(ab) (Table 1).
`uation software as described in the Pharmacia Biosensor manual. R is the
`Comparing the binding affinities of F(ab)-2 and F(ab)-3 suggested
`signalat time r. Data between 80 and 168, 148, 128, 114, 102, and 92 s were
`that framework residues in the F(ab)-1 VH domain needed to be
`used for 0.0625, 0.125, 0.25, 0.5, 1, and 2 mm F(ab), respectively. The
`altered to increase binding.
`association rate constant (k,,,) was obtained from the slope ofthe plot of K,
`Previous humanizations (31, 32, 43, 44) as well as studies of
`versus F(ab) concentration. At the end of each cycle, bound F(ab) was
`F(ab)-antigen crystal structures (45, 47) have shownthat residues H71
`removed by injecting 5 ul of 50 mm HCIat a flow rate of 20 l/min to
`and H73 can have a profound effect on binding, possibly by influ-
`regenerate the chip.
`encing the conformations of CDR-H1 and CDR-H2. Changing the
`Endothelial Cell Growth Assay. Bovine adrenal cortex-derived capillary
`humanresidues to their murine counterparts in F(ab)-4 improved
`endothelial cells were cultured in the presence of low glucose DMEM(Life
`Technologies, Inc.) supplemented with 10% calf serum, 2 mM glutamine, and
`binding by 4-fold (Table 1). Inspection of the models of the murine
`4595
`
`divided by the concentration of the standard corresponding to the midpoint
`absorbanceofthe standard titration curve. Assays for full-length IgG were
`the same as for the F(ab)s except that the assay buffer contained 10%
`humanserum.
`
`

`

`HUMANIZATION OF AN ANTI-VEGF MONOCLONAL ANTIBODY
`
`Variable Heavy
`
`A.4.6.1
`
`F(ab)-12
`
`humIII
`
`EIQLVOSGPELKQPGETVRISCKASGYTFINYGMNWVKQAPGKGLKWMG
`*
`*
`kk *
`keke
`*&
`*
`*
`x *
`EVQLVESGGGLVOPGGSLRLSCAASGYTFINYGMNWVROAP GKGLEWVG
`* kek &
`*
`EVQLVESGGGLVQPGGSLRLSCAASGFTFSS YAMSWVROQAPGKGLEWVS
`1
`10
`20
`30
`40
`
`A.4.6.1
`
`F(ab) -12
`
`humIII
`
`WINTYTGEPTYAADFKRRFTFSLETSASTAYLQISNLKNDDTATYFCAK
`*
`*
`kkk kkKK
`xk *
`‘TFSLDTSKSTAYLQMNSLRAEDTAVYYCAK
`nz kek
`x *
`*
`kak *
`wk KKK KKK
`VISGDGGSTYYADSVKGRFTISRDNSKNTLYLOMNSLRAEDTAVYYCAR
`50
`a
`60
`70
`80
`abc
`90
`
`A.4.6.1
`
`XPHYYGSSHWYFDVWGAGTTVIVSS
`*
`*
`F(ab)-12 XYPHYYGSSHWYFDVWGOQGTLVTVSS
`*
`*
`G----------FDYWGQGTLVTVSS
`110
`
`humIII
`
`A.4.6.1
`
`Variable Light
`DIQMTOTTSSLSASLGDRVI ISCSASODISNYLNWYQQKPDGTVKVLIY
`ak
`*
`x
`*&
`keke
`F(ab)-12 DIQMTQSPSSLSASVGDRVTITCSASODISNYLNWYOQOKPGKAPKVLIY
`*
`*
`*
`x
`
`humKI
`
`DIQMTOSPSSLSASVGDRVT ITCRASQSISNYLAWYQQKPGKAPKLLIY
`1
`10
`20
`30
`40
`
`A.4.6.1
`
`EISSLHSGVP SRFSGSGSGTDYSLTISNLEPEDIATYYCOQQYSTVPWIF
`xk
`x *
`*
`
`xk
`*
`kkk
`F(ab)-12 EISSLHSGVPSRFSGSGSGTDFTLTISSLOPEDFATYYCOOYSTVPWTF _
`humKI
`AASSLESGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQYNSLPWIF
`50
`60
`70
`80
`90
`
`A.4.6.1
`
`GGGTKLEIKR
`*
`*
`F(ab)-12 GQGTKVEIKR
`
`humKI
`
`GQGTKVEIKR
`100
`
`Fig. 1. Amino acid sequence of variable heavy and light domains of muMAbVEGF
`A.4.6.1, humanized F(ab) with optimal VEGF binding [F(ab)-12] and human consensus
`frameworks (humill, heavy subgroup III; humx/, light x subgroup I). Asterisks, differ-
`ences between humanized F(ab)-12 and the murine MAbor between F(ab)-12 and the
`human framework. CDRsare underlined.
`
`H94, human and murine sequences most often have an Arg (30). In
`F(ab)-12, this Arg was replaced by the rare Lys found in the murine
`antibody (Fig. 1), andthis resulted in binding that was less than 2-fold
`from the chimeric F(ab) (Table 1). F(ab)-12 was also compared to the
`chimeric F(ab) using the BlAcore system (Pharmacia). Using this
`technique, the K, of the humanized F(ab)-12 was 2-fold weaker than
`that of the chimeric F(ab) due to both a slower k,,, and faster koe,
`(Table 2).
`Full-length MAbs were constructed by fusing the VL and VH
`domains of the chimeric F(ab) and variant F(ab)-12 to the constant
`domains of human « light chain and human IgG] heavy chain. The
`full-length 12-IgG1 [F(ab)-12 fused to human IgG1] exhibited bind-
`ing that was 1.7-fold weaker than the chimeric IgG1 (Table 3). Both
`12-IgG1 and the chimeric IgG1 boundslightly less well than the
`original muMAb VEGFA.4.6.1 (Table 3).
`Biological Studies. rhuMAb VEGF and muMAbVEGFA.4.6.1
`were compared for their ability to inhibit bovine capillary endo-
`thelial cell proliferation in response to a near maximally effective
`concentration of VEGF,,., (3 ng/ml). In several experiments, the
`two MAbswerefoundto be essentially equivalent, both in potency
`and efficacy. The ED.9s were, respectively, 50 + 5 and 48 + 8
`ng/ml (~0.3 nm). In both cases, 90% inhibition was achieved at the
`concentration of 500 ng/ml (~3 nm). Fig. 3 illustrates a represent-
`ative experiment. Neither muMAb VEGF A.4.6.1 nor rhuMAb
`VEGFhadanyeffect on basal or bFGF-stimulated proliferation of
`capillary endothelial cells (data not shown), confirming that the
`inhibition is specific for VEGF.
`To determine whethersimilar findings could be obtained also in an
`in vivo system, we compared the two antibodies for their ability to
`suppress the growth of human A673 rhabdomyosarcomacells in nude
`mice. Previous studies have shown that muMAb VEGFA.4.6.1 has a
`dramatic inhibitory effect in this tumor model (20, 22). As shown in
`Fig. 4, at both doses tested (0.5 and 5 mg/kg), the two antibodies
`markedly suppressed tumor growth as assessed by tumor weight
`measurements 4 weeks after cell inoculation. The decreases in tumor
`weight compared to the control group were, respectively, 85 and 93%
`at each dosein the animals treated with muMAb VEGFA.4.6.1 versus
`90 and 95% in those treated with rhuMAb VEGF.Similar results were
`obtained with the breast carcinomacell line MDA-MB 435 (data not
`shown).
`
`DISCUSSION
`
`and humanized F(ab)s suggested that residue L46, buried at the
`VL-VHinterface and interacting with CDR-H3(Fig. 2), might also
`play a role either in determining the conformation of CDR-H3
`The murine MAb A.4.6.1, directed against human VEGF(42),
`and/or affecting the relationship of the VL and VH domains. When
`was humanized using the same consensus frameworksfor the light
`the murine Val was exchangedfor the human Leuat L46 [F(ab)-5],
`and heavy chains used in previous humanizations (31, 32, 43, 44),
`the binding affinity increased by almost 4-fold (Table 1). Three
`other buried framework residues were evaluated based on the
`i.e., V«I and VHIII (30). Simply transferring the CDRs from the
`murine antibody to the human frameworkresulted in a F(ab) that
`molecular models: H49, H69, and H78. Position H69 mayaffect
`exhibited binding to VEGF reduced by over 1000-fold compared to
`the conformation of CDR-H2, whereas position H78 mayaffect the
`the parent murine antibody. Seven non-CDR, frameworkresidues
`conformation of CDR-H1 (Fig. 2). When each was individually
`in the VH domain and one in the VL domain were altered from
`changed from the human to murine counterpart, the binding im-
`human to murine to achieve binding equivalent
`to the parent
`proved by 2-fold in each case [F(ab)-6 and F(ab)-7; Table 1]. When
`both were simultaneously changed, the improvement in binding
`murine antibody.
`In the VH domain,residuesat positions H49, H69, H71, and H78
`was 8-fold [F(ab)-8; Table 1]. Residue H49 wasoriginally in-
`are buried or partially buried and probably effect binding by
`cluded as the murine Gly; when changed to the human consensus
`influencing the conformation of the CDR loops. Residues H73 and
`counterpart Ala,
`the binding was reduced by 15-fold [F(ab)-9;
`H76 should be solvent exposed (Fig. 2) and hence may interact
`Table 1).
`directly with the VEGF;these two residues are in a non-CDR loop
`Wehave found during previous humanizations that residues in a
`adjacent to CDRs H1 and H2 and have been showntoplaya role
`framework loop, FR-3 (30) adjacent to CDR-H1 and CDR-H2,can
`in binding in previous humanizations (31, 32, 44). The requirement
`affect binding (44). In F(ab)-10 and F(ab)-11, two residuesin this loop
`for lysine at position H94 was surprising given that this residue is
`were changed to their murine counterparts: AsnH76 to murine Ser
`arginine in the human framework(Fig. 1). In some crystal struc-
`[F(ab)-10] and LysH75 to murine Ala [F(ab)-11]. Both effected a
`tures of F(ab)s, ArgH94 forms a hydrogen-bondedsalt-bridge with
`relatively small improvementin binding (Table 1). Finally, at position
`4596
`
`

`

`HUMANIZATION OF AN ANTI-VEGF MONOCLONAL ANTIBODY
`
`Fig. 2. Ribbon diagram of the model of humanized F(ab)-12 VL and VH domains. VL domain is shown in brown with CDRs in tan. The side chain of residue L46 is shown in
`yellow. VH domain is shown in purple with CDRs in pink. Side chains of VH residues changed from human to murine are shown in yellow.
`
`Table 2 Binding of anti-VEGF F(ab) variants to VEGF using the BlAcore system*
`
`\
`Amountof (Fab)
`Kg (nm)
`kon (M7 'S7')
`kore (87!)
`bound (RU)
`Variant
`0.91
`6.5 x 10*
`5.9 x 1075
`4250
`chim-F(ab)”
`1.8
`3.5 x 10*
`63x 107°
`3740
`F(ab)-12
`* The amount of F(ab) bound,in resonance units (RU), was measured using a BIAcore
`system when 2 yg F(ab) was injected onto a chip containing 2480 RU of immobilized
`VEGF.Off-rate kinetics (kop) were measured bysaturating the chip with F(ab) and then
`monitoring dissociation after switching to buffer. On-rate kinetics (k,,) were measured
`using 2-fold serial dilutions of F(ab). Ky, the equilibrium dissociation constant, was
`calculated as kop@kon-
`A chim-F(ab) is a chimeric F(ab) with murine VL and VH domainsfused to human CL
`and CHI heavy domains.
`
`adjacent to CDR-H3, suggests that CDR-H3plays a major role in the
`binding of the antibody to VEGF.
`The humanized version with optimal binding, 12-IgG1, exhib-
`ited only a 2-fold reduction in binding compared to the parent
`murine antibody (Table 3). An analysis of the binding kinetics of
`the humanized and chimeric F(ab)s showed that both had similar
`off-rates but that the humanized F(ab) had a 2-fold slower on-rate
`(Table 2), which accounts for the 2-fold reduction in binding.
`However, this modest reduction in on-rate did not result in any
`decreased ability to antagonize VEGFbioactivity. The two anti-
`
`
`Table 3 Binding of anti-VEGF IgG variants to VEGF?
`
`AspH101 (33, 48). Substitution of lysine for arginine might con-
`ceivably alter this salt-bridge and perturb the conformation of
`CDR-H3.
`In the VL domain, only one framework residue had to be changed
`to murine to optimize the humanization. Position L46 is at the VL-VH
`interface, whereit is buried andinteracts directly with CDR-H3(Fig.
`2). The requirement for murine valine (as opposed to human leucine)
`implies that this residue plays an importantrole in the conformation of
`CDR-H3. The necess