0022-1 767/93/1515-2623$02.00/0
`The Journal of Immunology
`Copyright 0 1993 by The American Association of Immunologists
`
`Vol 151, 2623-2632, No. 5, September 1, 1993
`Printed in U.S.A.
`
`Humanization of an Antibody Directed Against
`
`Leonard C. Presta,’* Steven J. Lahr,* Robert 1. Shields,+ James P. Porter,*
`Cornelia M. Corman,§ Brian M. Fendly,* and Paula M. Jardieut
`
`Departments of *Protein Engineering, ‘Immunology, *Medicinal and Analytical Chemistry, and $Cell Genetics, Genentech,
`Inc., South San Francisco, CA 94080
`
`ABSTRACT. IgE antibodies bind to specific high-affinity receptors on mast cells, leading to mast cell degranulation
`and release of mediators, such as histamine, which produce symptoms associated with allergy. Hence, anti-lgE
`antibodies that block binding of IgE to its high-affinity receptor are of potential therapeutic value in the treatment
`of allergy. These antibodies must also not bind to IgE once it is bound to the receptor because this would trigger
`histamine release. This study describes the humanization of a murine antibody, MaEll, with these characteristics.
`Variants of the humanized antibody were evaluated to probe the importance of framework residues on antibody
`binding and to determine which charged residues in the CDR interacted with IgE. We found that only five changes
`in human framework residues were required to provide for binding comparable to that of the original murine
`antibody. Journal of Immunology, 1993, 151 : 2623.
`
`I gE antibodies bind to a specific high-affinity receptor
`
`(FccRI) (1,2) on mast cells and basophils via their Fc
`(3),
`region (constant domains Cc2, Cc3, and Cc4)
`leading to mast cell degranulation and release of mediators
`that produce symptoms associated with allergy (4, 5). IgE
`also binds to a low-affinity receptor (FccRII and CD23) (6)
`on B lymphocytes (7, 8) and on cells involved in inflam-
`mation, leading to IgE-mediated cytotoxicity and phago-
`cytosis (9). Hence, anti-IgE antibodies that block binding
`of IgE to its receptors may be of therapeutic value.
`Herein we report the humanization of a murine antibody,
`MaEll, directed against IgE that prevents binding of free
`IgE to FccRI on mast cells but does not bind to FceRI-
`bound IgE. The latter characteristic is important because
`this antibody will not trigger histamine release by cross-
`linking IgE-loaded FccRI on mast cells. However,
`as a
`therapeutic the murine antibody would not be the molecule
`of choice because clinical use of non-human antibodies has
`
`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`fundamental problems. First, non-human
`identified three
`antibodies cause a human immune response that can reduce
`therapeutic value of the non-human antibody (10-14). Sec-
`ond, therapeutic efficacy is reduced by the relatively rapid
`clearance of the non-human antibody compared with hu-
`man ones (15). Finally, non-human antibodies generally
`show only weak recruitment of effector functions
`(e.g.,
`antibody-dependent cell-mediated cytolysis), which may
`be desirable or essential for efficacy (16, 17).
`One approach to overcoming these problems involves
`production of “humanized” antibodies that significantly re-
`duce the amount of non-human sequence in the molecule.
`This technique, pioneered by Winter et al. (16, IS), involves
`transplantation of the non-human Ag-binding loops (CDR)2
`onto a human antibody framework.
`In addition to CDRs,
`select non-human framework residues must also be incor-
`porated into the humanized antibody
`to maintain proper
`CDR conformation (19) or because they interact directly
`with the antigen (20). The humanized antibody, when prop-
`5% non-
`erly constructed, will contain approximately
`human residues, most of which will be in the CDRs. Al-
`though several humanized antibodies have been reported
`(21-28), only recently has the clinical efficacy
`of these
`
`’ Abbreviations used in this paper: CDR, complementarlty-determining re-
`gion.
`
`Received for publication February 4, 1993. Accepted for publication May 21,
`1993.
`The costs of publication of this article were defrayed in part by the payment of
`page charges. This article must therefore be hereby marked adverrisement
`in
`accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
`’ Address correspondence and reprint requests to Dr. Leonard Presta, Depart-
`ment of Protein Engineering, Cenentech, Inc., 460 Point San Bruno Blvd.,
`South San Francisco, CA 94080.
`
`2623
`
`Pfizer v. Genentech
`IPR2017-01488
`Genentech Exhibit 2044
`
`

`

`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`2624
`
`molecules begun to be evaluated (15).
`Most humanized antibodies have been designed by com-
`paring the sequence of the murine antibody of interest to a
`database of human antibody sequences and choosing the
`homologous to that of the murine
`human antibody most
`antibody (23-27). In contrast, we have used a framework
`derived from consensus sequences of human VL and VH
`subgroups. This provides for use of the most
`common
`framework found among human IgG antibodies and elim-
`inates possible idiosyncracies present in
`any individual
`framework, both of which would, hopefully, reduce the
`chance of an immunogenic response against the humanized
`antibody. This same framework has been previously used
`for other humanized antibodies (21, 22).
`Thirteen variants of the humanized antibody were eval-
`uated to probe the importance of framework residues on
`antibody binding. We found that only five changes in hu-
`man framework residues were required to provide for bind-
`ing comparable to that of the original murine antibody. We
`also used an additional IO variants to ascertain whether any
`of the charged CDR residues were important IgE-binding
`determinants.
`
`Materials and Methods
`Construction of humanized antibody
`
`The murine anti-human IgE mAb, MaE11, was generated,
`cloned, and sequenced at Genentech (unpublished data). To
`construct the first F(ab)
`variant of humanized MaE11,
`F(ab)-1, site-directed mutagenesis (29) was performed on
`a deoxyuridine-containing template
`containing a human
`K-subgroup I light chain and human subgroup Ill heavy
`chain (VH-CHI) in a pUC119-based plasmid, pAK2 (21).
`F(ab)-2 was then constructed from a F(ab)-l template. All
`other humanized F(ab) variants were constructed from a
`template of F(ab)-2. Plasmids were
`transformed
`into
`Escherichia coli strain JMlOl (30) for preparation of
`double- and single-stranded DNA. For each variant both
`light and heavy chains were completely sequenced using
`the dideoxynucleotide method. DNA encoding light and
`heavy chains was then subcloned into a derivative of the E.
`coli F(ab) expression plasmid, pAK19 (31). These deriv-
`atives lack the hinge
`cysteines that form the interheavy
`chain disulfides in F(ab'), fragments. F(ab) fragments, as
`opposed to full-length IgG antibodies, facilitated the anal-
`ysis of a moderately large number of variants because E.
`coli expression could be used rather than mammalian cell
`culture techniques. Once the best variant was determined,
`it was subsequently subcloned into a plasmid encoding a
`full-length human IgGl (see below).
`The expression plasmids were transformed into E. coli
`strain "294
`(32), and a single colony was grown in 5-ml
`2YT-100 pg/ml carbenicillin for 5-8 h at 37°C. The 5-ml
`
`
`
`
`
`HUMANIZATION OF A N ANTI-lgE ANTIBODY
`
`culture was added to 100 ml AP5-100 pg/ml carbenicillin
`and allowed to grow for 16 h in a 500-ml shaker flask at
`37°C. The culture was centrifuged at 4,000 X g and the
`supernatant removed. After freezing for 1 h, the pellet was
`resuspended in 5-ml cold 10 mM Tris-1 mM EDTA; SO pl
`0.1 M benzamidine (Sigma, St. Louis) was added to inhibit
`proteolysis. After gentle shaking on ice for 1 h, the sample
`was centrifuged at 10,000 X g for 15 min. The supernatant
`was applied to a protein A-Sepharose CL-4B (Pharmacia)
`column (0.5 ml bed volume) then washed with 10 m13 M
`KC1-100 mM Tris, pH 8.0, and eluted with 2.5 ml 100 mM
`acetic acid, pH 2.8, into 0.5 ml 1 M Tris, pH 8.0. The F(ab)
`was then buffer exchanged into PBS using a Centricon-30
`to a final volume of 0.5 ml.
`(Amicon) and concentrated
`SDS-PAGE gels of all F(ab)s were run to ascertain purity.
`F(ab) concentrations were determined using an 0.1%
`of 1.0. The extinction coefficient was determined by using
`the concentration of protein from an amino acid analysis of
`purified F(ab)-2 and the A2x0 for this same sample.
`Selected F(ab) fragments were analyzed directly by liq-
`uid chromatography/mass spectrometry to confirm their
`molecular weight. Samples were injected into a packed cap-
`illary liquid chromatography system (33) and analyzed di-
`rectly with a Sciex API 3 mass spectrometer. The higher
`charge states of human growth hormone (m.w. = 22,256.2),
`acquired using the same instrument parameters as those
`used for the samples, were used for calibration.
`For generation of human IgGl versions of humanized
`MaE11, the heavy and light chains were subcloned sepa-
`rately into previously described pRK plasmids (34). Ap-
`propriate heavy and light chain
`plasmids were cotrans-
`fected into an adenovirus-transformed human embryonic
`kidney cell line, 293 (39, using a high efficiency procedure
`(35, 36). Media was changed to serum free and harvested
`daily for up to 5 days. Antibodies were purified from the
`pooled supernatants using protein A-Sepharose CL-4B
`(Pharmacia). The eluted antibody was buffer exchanged
`into PBS by G25 gel filtration, concentrated by ultrafiltra-
`tion using a
`Centriprep-30 or Centricon-100 (Amicon),
`sterile filtered using a Millex-GV (Millipore), and stored at
`4°C. The concentration of antibody was determined using
`total Ig-binding ELISA. The concentration of the standard
`was determined by amino acid composition analysis.
`
`Soluble receptor assay
`A 96-well assay plate
`(Nunc) was coated with 0.05 ml 1
`pglml FceRI a-chain IgG chimeric receptor (Genentech;
`unpublished data)
`in coating buffer (50 mM carbonate/
`bicarbonate, pH 9.6) for 12 h at 4 4 ° C . The wells were
`aspirated and 250 pl blocking buffer (PBS, 1% BSA, pH
`7.2) was added and incubated for 1 h at 4°C. In a separate
`assay plate the samples and reference murine MaEll were
`titered from 200 to 0.001 pg/rnl by 1:4 dilutions with assay
`
`

`

`Journal
`
`Immunology
`
`2625
`
`VH domain
`. . . . . . . . . .
`DVQLQESGPGLVKPSQSLSLACSVTGYSITS[GYSWNlWIRQF
`. . . . . . . . . .
`
`EVQLVESGGGLVQPGGSLRLSCAVSGYSITS[GYSWNlWIRQA
`
`EVQLVESGGGLVQPGGSLRLSCAASGFTF-SKDYAMSIWVRQA
`20
`30
`10
`1
`4 0
`
`. . .... . . . .
`
`PGNKLEWMG[SITYDGSSNYNPSLKNlRISVTRDTSQNQFFL
`
`.. .. ..
`
`. . . . . . .
`
`PGKGLEWVA[SITYDGSTNYADSVKG]RFTISRDDSKNTFYL
`
`PGKGLEWVA[VISNGSDTYYADSVKG]RFTISRDDSKNTLYL
`80
`50
`60
`7 0
`
`.. ..
`
`KLNSATAEDTATYYCAR[GSHYFGHWHFAVlWGAGTTVTVSS
`
`QMNSLRAEDTAVYYCAR[GSHYFGHWHFAVlWGQGTLVTVSS
`
`. .
`
`. .........
`
`MaEll
`
`F(ab)-2
`
`hum111
`
`MaEll
`
`F(ab)-2
`
`hum111
`
`MaEll
`
`F(ab)-2
`
`hum111
`
`MaEll
`
`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`QMNSLRAEDTAVYYCAR[DSRFF---”DVIWGQGTLVTVSS
`abc
`
`lOOabcd
`110
`90
`VL domain
`. . . . . . . .
`
`DIQLTQSPASLAVSLGQRATISC[KASQSVDYDGDSYMNlWYQQKP
`
`.... .
`
`F(ab)-2 DIQLTQSPSSLSASVGDRVTITC[RASQSVDYDGDSYMNlWYQQKP
`
`humKI
`
`DIQMTQSPSSLSASVGDRVTITC[RASQSVDIS--SYLNlWYQQKP
`20
`30 abcd
`40
`10
`1
`
`.. .
`
`. .. .
`
`GQPPILLIY[AASYLGSIEIPARFSGSGSGTDFTLNIHPVEE
`
`. .....
`
`F(ab)-2
`
`GKAPKLLIY[AASYLESlGVPSRFSGSGSGTDFTLTISSLQP
`
`MaEll
`
`humK1
`
`MaEll
`
`F(ab)-2
`
`GKAPKLLIY[AASSLESlGVPSRFSGSGSGTDFTLTISSLQP
`60
`50
`80
`7 0
`
`EDAATFYC[QQSHEDPYT]FGAGTKLEIK
`
`. .
`
`....
`
`EDFATYYC[QQSHEDPYT]FGQGTKVEIK
`
`.
`
`.
`
`
`
`humKI
`
`EDFATYYC[QQYNSLPYTlFGQGTKVEIK
`100
`90
`FIGURE 1 . Amino acid sequences of murine MaE11, hu-
`[F(ab)-2], and human consensus
`manized MaEll variant 2
`sequences of heavy chain subgroup Ill (humlll) and light
`chain K subgroup I (hurnd) (42). Murine residues are itali-
`cized. The definition of CDR residues of Kabat et al. (42) are
`included within brackets; the definition by Chothia et al. (19)
`are in boldface. Kabat et al. (42) numbering used with inser-
`tions shown as a, b,
`c.
`
`Computer graphics models of murine and humanized
`F(ab)s
`Sequences of the VL and VH domains (Fig. 1) were used
`to construct a computer graphics model
`of the murine
`MaEll VL-VH domains; this model was used to determine
`which framework residues should be incorporated into the
`humanized antibody. Models of the humanized variants
`were also constructed to verify correct selection of murine
`framework residues. Construction of the models was per-
`formed as described previously (21, 40).
`
`Results
`Expression and purification of humanized M a E l l
`F(ab)s and antibodies
`Shaker flask expression levels of F(ab)s were usually 0.5-1
`mg/L of culture. Full-length antibody recoveries ranged
`
`buffer (0.5% BSA and 0.05% Tween 20, PBS, pH 7.2) and
`an equal volume of 10 ng/ml biotinylated IgE (37) was
`added and the plate incubated for 2-3 h at 25°C. The FceRI-
`coated wells were washed three times with PBS and 0.05%
`Tween 20 (Sigma) and 50 pl from the sample wells were
`transferred and incubated with agitation for 30 min at 25°C.
`Fifty pl/well of 500 pglml Streptavidin-HRP (Sigma), di-
`luted 15000 in assay buffer, was incubated for 15 min with
`agitation and then the plate was washed as before. Fifty
`pl/well of Microwell Peroxidase Substrate (Kirkgaard &
`Perry Laboratories) was added and color was developed for
`30 min. The reaction was stopped by adding an equal vol-
`ume of 1 N HCI, and the absorbance measured at 450 nm.
`The concentration at 50% inhibition was calculated by plot-
`ting percentage of inhibition versus concentration of block-
`ing antibody with a
`nonlinear four-parameter curve
`fit
`using the Kaleidagraph data analysis application (Synergy
`Software).
`
`FACS-based binding assays
`The ability of the antibody to inhibit FITC-conjugated (38)
`IgE binding to the a-chain of the high-affinity FceRI re-
`ceptor expressed on CHO 3D10 cells (39) was determined
`by flow cytometry. FITC-conjugated IgE (40 nM) was pre-
`incubated with the antibody (0.3-1 X
`M) at 37°C for
`30 min in FACS buffer (PBS, 0.1% BSA, and 10 mM so-
`dium azide, pH 7.4). The complex was then incubated with
`5 X lo5 CHO 3D10 cells at 4°C for 30 min. The cells were
`washed three times with FACS buffer and mean channel
`fluorescence at 475 nm measured on an FACScan flow
`cytometer (Becton Dickinson). MaEl (Genentech), a mu-
`rine anti-human IgE mAb that does not block IgE binding
`to the FceRI a-chain, was used as a positive control and
`MOPC21 (Cappel), a murine monoclonal that does not rec-
`ognize IgE, was used as a negative control.
`
`Binding of antibodies to IgE-loaded FceRI
`Antibody binding to human IgE associated with the
`a-subunit of FceRI expressed on CHO 3D10 cells (39) was
`determined by preincubating 5 X lo5 CHO 3D10 cells with
`10 pg/ml human IgE for 30 min at 4°C. Cells were washed
`three times followed by a 30-min incubation with varying
`concentrations of either murine anti-human IgE mAbs
`MaEl or MaEll or
`the humanized mAb variant 12.
`MOPC21 (murine IgG1) was used as a control for the mu-
`rine mAbs, whereas humanized 4D5 mAb
`(21) (human
`IgG1) was used as a control for humanized variant 12.
`Binding of murine mAbs was detected with a FITC-
`conjugated F(ab‘), goat anti-mouse IgG (10 pglml). Hu-
`manized mAb binding
`was detected with a FITC-
`conjugated F(ab’)2 goat anti-human IgG (50 pglml), which
`had been affinity purified on an IgE column to minimize
`cross-reactivity to IgE.
`
`

`

`2626
`
`Table I
`Humanized M a E l l Flab) variants
`
`HUMANIZATION OF AN ANTI-lgE ANTIBODY
`
`Changes from F(ab)-2”
`
`VL
`
`VH
`
`Purnose
`r - - -
`
`Concentration at 50% inh.
`(ng/ml)
`
`
`
`F(ab)-X F(ab)-X
`
`Mean
`
`SDb
`
`”
`
`F(ab)-2 MaEll
`
`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`Variant
`
`F(ab)-1
`
`F(ab)-1 Bd
`F(ab)-2
`F(ab)-3
`
`F(ab)-4
`F(ab)-5
`F(ab)-6
`F(ab)-7
`F(ab)-8
`
`F(ab)-9
`
`F(ab)-1 0
`
`F(ab)-ll
`
`F(ab)-l2
`
`MaEl1
`
`Leu 4 Met
`Arg 24 Lys
`Glu 55 Gly
`Gly 57 Glu
`
`Leu 4 Met
`Met 33 Leu
`Leu 4 Met
`
`Glu 55 Gly
`Gly 57 Glu
`
`Ala 13 Val
`Val 19 Ala
`Val 58 //e
`Leu 78 Val
`Val 104 Leu
`
`Val 24 Ala
`/le 37 Val
`Thr 57 Ser
`Ala 60 Asn
`Asp 61 Pro
`Val 63 Leu
`Gly 65 Asn
`Phe 78 Leu
`Leu 78 Phe
`
`Val 24 Ala
`Phe 78 Leu
`/le 37 Val
`
`Ala 60 Asn
`Asp 61 Pro
`Val 48 Met
`Ala 49 Gly
`Ala 60 Asn
`Val 63 Leu
`Phe 67 /le
`Ile 69 Val
`Met 82 Leu
`Leu 82c Ala
`Ala 60 Asn
`Asp 61 Pro
`Val 63 Leu
`Phe 67 /le
`Ala 60 Asn
`Asp 61 Pro
`Phe 67 /le
`
`Straight CDR swap
`
`>100,000
`
`>16.0‘
`
`>560
`
`Packing; CDR-L1
`
`Packing; CDR-L1
`Packing; CDR-H1
`Packing; CDR-H1, H2
`VL-VH interface
`Unusual Gly 55-X-Clu 57 MaE11 sequence
`
`CDR-H2; Ala 60 Asn at VL-VH interface
`
`Repack F(ab)-2 interior as in murine MaE11
`
`98,000
`6083
`9439
`
`6770
`9387
`17,537
`8622
`5799
`
`1224
`
`842
`
`1279
`508
`
`349
`733
`4372
`107
`523
`
`102
`
`130
`
`16.0
`1.0
`1.6
`
`1.1
`1.6
`2.9
`1.4
`1.0
`
`0.20
`
`0.14
`
`547
`34
`53
`
`38
`52
`24
`48
`32
`
`6.8
`
`4.7
`
`CDR-HZ; packing of Leu 63 and /le 67 41 6
`
`66
`
`0.07
`
`2.3
`
`CDR-HZ; packing of
`
`Val 63 and //e 67 501
`
`84
`
`0.08
`
`1 79
`
`63
`
`0.03
`
`2.8
`
`1 .o
`
`a Murine residues are italicized; residue numbers are according to Kabat et al. (42).
`Mean and SD of three soluble receptor assays.
`A F(ab)-WF(ab)-2 ratio >16 means that this variant exhibited no binding even at the highest F(ab) concentrations used.
`Changes from F(ab)-1.
`
`from 30 to 50 pg/L based on ELISA. As noted previously
`(31, 41), the human consensus sequence used allows pu-
`rification of F(ab) fragments from E. coli periplasmic ex-
`tracts on protein A. F(ab) was always present primarily in
`the periplasmic extract but detectable at a low level in the
`media. Mass spectrometry was performed on selected
`F(ab)s to confirm their m.w.: F(ab)-2, expected M , 48,303,
`measured MI 48,306; F(ab)-9 expected M, 48,329, meas-
`uredM148,332; F(ab)-10 expected M, 48,285, measured M,
`48,286. These values are within the expected error limit of
`the system (0.01%).
`
`Design of humanized MaEll antibodies
`In contrast to other investigations that have used human
`sequences closest to the murine Ig of interest (23-27), our
`humanized antibodies use a human consensus sequence.
`This consensus sequence consists of a framework based on
`human VH subgroup 111 and VLK subgroup I (42).
`First, we constructed F(ab)-1
`in which only the six
`
`CDRs, as defined by Kabat et al. (42), were grafted onto
`the human framework-all
`framework residues were re-
`tained as human. This variant is best described as a straight
`CDR swap. F(ab)-1 showed no detectable inhibition of IgE
`binding to its receptor (Table I). Even when one framework
`residue (H67), subsequently found
`to be important for
`maintenance of binding, was replaced with the correspond-
`ing murine framework residue, IgE binding was still not
`restored (F(ab)-1B; Table I). The failure of such “CDR
`swap” variants to bind their antigens has been reported pre-
`viously (21, 25).
`F(ab)-2 was the first variant based on modeling. In ad-
`dition to the six murine CDRs, several murine framework
`residues were incorporated into the human framework (Fig.
`1). The definition of CDRs provided by Kabat et al. (42),
`i.e., based on sequence variability, were used except for
`CDR-H1 and CDR-H2. CDR-H1 definitions based on se-
`quence variability (42) and on crystallographic studies of
`antigen-antibody complexes (19) differ significantly (Fig.
`
`

`

`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`Journal of Immunology
`
`2627
`
`FIGURE 2. Framework residues altered in various human-
`ized MaE11 variants. VL and VH backbone traces are grey,
`VL CDRs are red, and VH CDRs are blue. Selected framework
`residue side chain atoms are represented as spheres. Leu L4,
`Met L33, Glu L55, Gly L57, Val H24, Ile H37, and Phe H78
`are yellow. Ala H60 and Asp H61 are purple. Val H63 and
`Phe H67 are green.
`
`1). We therefore redefined CDR-H1 to include both defi-
`nitions, i.e., residues H26H3.5. The definition of CDR-H2
`based on sequence variability contains more residues than
`that based on antibody-antigen crystal structures (Fig. 1).
`Because none of the crystal structures reported to date show
`antibody-antigen contacts for antibody residues H60-H65,
`we used the CDR-H2 definition provided by Chothia et al.
`(19). Hence, in F(ab)-2 a shorter version of CDR-H2 (res-
`idues H50-H58) was used compared with F(ab)-1.
`In F(ab)-2 VH, 28 human consensus residues were al-
`tered to murine, including only three murine framework
`residues: Val H24, Zle H37, and Phe H78 (murine residues
`in all F(ab)s are italicized). In F(ab)-2 VL, 9 human con-
`sensus residues were changed to murine, including only one
`murine framework residue, Leu LA. F(ab)-2 is the variant
`with the minimal number of changes to the human frame-
`work which, in our judgment, should be required for main-
`tenance of binding.
`An additional 10 variants were constructed to 1) test the
`effects of buried residues on CDR conformation, 2) deter-
`mine whether the models had been successful in directing
`which framework residues should be included in the hu-
`manized MaE11, and 3) evaluate the importance of an un-
`usual sequence present in the murine MaEll (Table I). To
`test the effects of buried residues on CDR conformation,
`F(ab)-3 to F(ab)-7 were constructed in which murine res-
`idues were changed back to human ones. In F(ab)-3 the
`buried murine VL residues Leu LA (framework residue) and
`provement in binding compared with F(ab)-2 (Tables I and
`Met L33 (CDR-L1) were exchanged for human sequence
`Met LA and Leu L33 to determine their effect on CDR-L1.
`11; Fig. 3). In F(ab)-9, which exhibited a fivefold better
`binding than F(ab)-2, two residues in CDR-H2 (as defined
`Chothia et al. (19) have proposed that the residue at position
`L33 may be important in maintaining proper conformation
`by Kabat et al. (42)) were changed to murine residues: Ala
`of CDR-L1. The models suggested that the side chain at LA
`H6OAsn and Asp H61 Pro. The Pro substitution could have
`might also affect CDR-L1 conformation by its interaction
`altered the CDR-H2 conformation and/or rigidity and Am
`H60 is anticipated to be buried at the VL-VH
`interface,
`with the side chain at L33. However, binding of F(ab)-3 was
`possibly interacting with Asp L1 (Fig. 2).
`only slightly impaired over F(ab)-2. In F(ab)-4 only posi-
`tion L4 was altered and no significant difference in binding
`F(ab)-10, which also exhibited improved binding rela-
`tive to F(ab)-2, was a variant in which all buried residues
`compared with F(ab)-2 was found. Together, F(ab)-3 and
`(defined as residues with accessible surface area less than
`F(ab)-4 show that, at least for this humanized antibody, the
`5% that of the free amino acid) in both VL and VH domains
`side chains at LA and L33 have minimal affect on binding
`were those of the murine MaE11. In essence F(ab)-10 can
`and presumably on the conformation of CDR-L1.
`be considered as murine MaEll in which only exposed,
`The models also suggested that framework residue H24
`non-CDR residues in VL and VH were changed to human
`could affect the conformation of CDR-H1 and framework
`sequence. This type of humanized variant was suggested by
`residue H37 could affect the VL-VH interface (Fig. 2). Sub-
`stitution of the murine with the human
`residue at H24
`the recent proposal of Padlan (43). However, the possibility
`remained that the improved binding exhibited by F(ab)-10
`(F(ab)-5) or H37 (F(ab)-7) showed minimal reduction in
`was due to only one or a few residues. For example, F(ab)-
`binding. In contrast, replacing the murine Phe at framework
`
`-10 contains Ala H60 Asn that, based on F(ab)-9 (Table I),
`position H78 with the human Leu (F(ab)-6) affected a larger
`could alone account for much of the improved binding of
`reduction in binding. Our models suggest that this side
`chain is influencing the conformation of CDR-H1 and/or
`F(ab)-10.
`Two additional variants were evaluated to test this pos-
`CDR-H2 (Fig. 2). H78 has not previously been considered
`sibility. For these two variants, instead of using F(ab)-2 as
`important in maintaining the conformation
`of either
`the basis, we used F(ab)-9 because this variant showed a
`CDR-H1 or CDR-H2 (19).
`fivefold improved binding (Table I). According to the mod-
`In F(ab)-9 to F(ab)-12 human residues were replaced
`with murine. AU four variants exhibited substantial im-
`els, the side chains at H63 and H67 could affect the con-
`
`

`

`2628
`
`OF
`
`
`
`
`
`HUMANIZATION ANTIBODY
`
`
`
`AN ANTI-lgE
`
`Table II
`Humanized M a E l l lgG1 variants
`
`Variant'
`
`Concentration at 50% inh.
`(ndml)
`
`Mean
`
`S D"
`
`
`
`7569
`3493
`1118
`
`1449 226
`
`1042
`
`172
`
`53
`
`
`
`449
`
`IgC 1-2
`1264
`
`IgG 1-9 0.46
`
`IgG 1-1 0
`IgC 1-1 2
`
`MaEll
`" lgC1-2 represents full-length lgGl molecule, variant 2
`I, Mean and SD of five soluble receptor assays.
`
`0.1 5
`0.1 9
`
`16.9
`7.8
`2.5
`1 .o 3.2
`
`affect of only a single residue, i.e., H67.
`F(ab)-8 was constructed to examine an unusual sequence
`in MaE11 VL: Gly L55-Ser L56-GIu L57. In both murine
`and human VLK sequences L57 is conserved as Gly and in
`Variant X
`Variant X
`~-
`human VLK subgroup I Glu predominates at L55 (42).
`lgC1-2
`M a E l l
`F(db)-2 used the human sequence at these two positions and
`1 .o
`F(ab)-8 the murine. This had no effect on binding to IgE
`(Table I).
`Once we determined which F(ab) variant provided bind-
`ing closest
`to murine MaE11, we generated full-length
`IgGl molecules. The binding of these molecules relative to
`variant 2 or murine MaEll (Table 11) was comparable to the
`relative binding exhibited by the F(ab) fragments (Table 1).
`Note, however, that for variants 2 and 10 the ratio of the
`F(ab) relative to murine MaEll was twofold higher than for
`the corresponding IgGl relative to murine MaEIl. This
`may be due to error in ascertaining the protein concentra-
`tion or differences in the avidity of the IgGl form of these
`variants compared with the other variants.
`
`Downloaded from
`
`http://www.jimmunol.org/
`
` by guest on February 22, 2018
`
`loo] 80
`
`I
`
`0
`.01
`
`.1
`
`1
`
`Binding of MaEll to IgE-loaded FceRI
`Murine MaEll prevents binding of free IgE to FceRI on
`mast cells but does not trigger histamine release by binding
`to IgE-loaded FceRI (unpublished data). As shown in Fig-
`ure 4, both murine MaE11 and humanized variant 12, as
`well as the negative isotype control antibody MOPC21 and
`negative isotype control humanized 4D5 (21), did not bind
`IgE-loaded F ~ E R I on CHO 3D10 cells. In contrast, the mu-
`rine MaEl antibody, which binds to IgE but does not pre-
`vent IgE binding to FceRI, bound to the IgE-loaded FceRI.
`Unlike the human lgGl control (humanized 4D5), the mu-
`rine lgG1 isotype (as represented by MOPC21) exhibits a
`nonspecific background binding of approximately 10% on
`Antibody (uM)
`these cells.
`MaEll did not give staining above
`the
`FIGURE 3. Murine MaEll and humanized F(ab)s block
`MOPC21 control and humanized variant 12 did not give
`FITC-lgE binding to CHO 3D10 cells expressing FceRI
`staining above the humanized 4D5 control (Fig. 4). The
`a-chain. Percentage of inhibition by murine mAb MaEl1 (O),
`failure to detect binding of MaEl1 or humanized variant 12
`F(ab)-2 (O), F(ab)-9 (O), F(ab)-11 (A), F(ab)-lL (A), and the
`not due to displacement of
`to IgE on 3D10 cells was
`negative control humanized mAb 4D5 (21) (H) measured by
`receptor-bound IgE by the antibodies. Using a polyclonal
`FACS. Data points are the average of three experiments, ex-
`anti-IgE reagent, the amount of IgE detected on these cells
`cept for rnAb 4D5 (single experiment).
`after a 1-h incubation with MaEll or humanized variant 12
`in the absence of
`was equivalent to the level measured
`antibodies (data not shown).
`
`formation of CDR-H2 (Fig. 2). H63 is part of CDR-H2, as
`defined by Kdbat et al. (42), but not as defined by Chothia
`et al. (19), whereas H67 is defined as a framework residue
`under both CDR definitions. In F(ab)-11 H63 and H67 were
`the murine residues Leu and Ile, respectively. In F(ab)-12
`only H67 was changed to the murine Ile (H63 remained as
`the human Val). In both the soluble receptor and cell-based
`assays these variants exhibited binding that was at least as
`good as F(ab)-10
`and better than their parent, F(ab)-9-
`(Tables I and 11; Fig. 3). This suggests that the improved
`binding of F(ab)-10 was not due to repacking of the VH
`domain interior with murine residues,
`but was due to the
`
`CDR residues important in IgE binding
`The sequence of the MaEll CDRs shows a preponderance
`of charged side chains (Fig. 1). CDR-L1 contains three Asp
`residues, whereas CDR-L3 possesses His, Glu, and Asp; in
`CDR-H3 there are three His residues. The models of murine
`and humanized MaEll pointed to the spatial proximity of
`all of these charged residues (Fig. 5). In contrast, the lone
`Asp H54 in CDR-H2 is spatially separated from the other
`charged residues. Though we did not attempt an exhaustive
`
`

`

`Immunology
`
`80 -
`
`Discussion
`
`2629
`
`60 -
`
`40 -
`
`We have described herein the humanization of the murine
`antibody MaE11, which is targeted against human IgE. De-
`sign of a functional antibody was dependent on substitution
`of several murine framework residues into
`the human
`framework. In addition, we mapped the charged CDR res-
`idues and found some of them to be important in the
`antibody-IgE interaction.
`In agreement with previous studies (21,23-25), variants
`1 to 12 show that framework residues can have a significant
`effect on antibody function. This is underscored when con-
`sidering F(ab)-1, which is a straight CDR swap in which
`only the six CDRs were transplanted onto the human frame-
`work. No consideration was given to altering framework
`residues, all of which were retained as human. Table I
`shows that, even at high concentrations, F(ab)-1 did not
`inhibit IgE binding
`to its receptor. Inclusion of murine
`framework residue PheH78 (F(ab)-1B) still did not restore
`inhibitory activity.
`At least two possibilities might account for the lack of
`binding of F(ab)-1 to IgE. The first involves residues LA and
`Antibody (uglml)
`L33. In F(ab)-2 these residues were Leu L4 and Met L33;
`FIGURE 4. Murine MaEll and humanized variant
`12 do
`in F(ab)-1 they were MetL4 and Met L33. Conceivably, the
`not bind to
`IgE-loaded CHO 3D10 cells expressing FceRI
`a-chain. Percentage of binding by murine mAb MaEll (O),
`Met L4-Met L33 combination in F(ab)-1 might disturb the
`humanized mAb variant 12 (A), positive control murine mAb
`conformation of CDR-Ll. However, we isolated the Met
`MaEl (O), negative control antibody murine MOPC21 (A),
`L4-Met L33 combination from the other changes in F(ab)-1
`and the negative control humanized mAb 4D5 (21) (0) meas-
`(F(ab)-4; Table I) and this variant showed minimal affect
`ured by FACS. O n an arithmetidinear scale, mean channel
`on binding (Table I).
`fluorescence values at 10 pglml were MOPC21 7.3, MaEl
`Asecond possible problem in F(ab)-l involves CDR-H2.
`32.1, MaEll 6.4, hu4D5 4.7, and huMaEll 4.6. All three
`The buried hydrophobic side chains at positions H63 and
`murine mAbs were murine isotype IgG1, and both human-
`H67 could affect the conformation of CDR-H2 (Fig. 2). Our
`ized mAbs were human isotype IgGl. Data points are the
`variants contain four combinations at positions H63 and
`average of three experiments.
`H67: Leu and ZZe (murine MaEll and F(ab)-11), Val and
`Phe (F(ab)-2), Leu
`and Phe (F(ab)-1), and Val and Ile
`(F(ab)-12).