Available on line at www.sciencedirect.com
`
`SCIENCE ®DIRECT ®
`
`ELSEVIER
`
`Methods 36 (2005) 69–83
`
`METHODS
`
`www.elsevier.com/locate/ymeth
`
`Design of humanized antibodies: From anti-Tac to Zenapax
`¤
`, Paul R. Hinton, Shankar Kumar
`
`Naoya Tsurushita
`
`Protein Design Labs, Inc., 34801 Campus Drive, Fremont, CA 94555, USA
`
`Accepted 17 January 2005
`
`Abstract
`
`Since the introduction of hybridoma technology, monoclonal antibodies have become one of the most important tools in the bio-
`sciences, Wnding diverse applications including their use in the therapy of human disease. Initial attempts to use monoclonal antibod-
`ies as therapeutics were hampered, however, by the potent immunogenicity of mouse (and other rodent) antibodies in humans.
`Humanization technology has made it possible to remove the immunogenicity associated with the use of rodent antibodies, or at
`least to reduce it to an acceptable level for clinical use in humans, thus facilitating the application of monoclonal antibodies to the
`treatment of human disease. To date, nine humanized monoclonal antibodies have been approved for use as human therapeutics in
`the United States. In this paper, we describe procedures for antibody humanization with an emphasis on strategies for designing
`humanized antibodies with the aid of computer-guided modeling of antibody variable domains, using as an example the humanized
`anti-CD25 monoclonal antibody, Zenapax.
`© 2005 Elsevier Inc. All rights reserved.
`
`Keywords: Antibody engineering; Antigen-binding aYnity; Daclizumab; EVector functions; Immunogenicity; Molecular biology; Molecular model-
`ing; Protein expression; Sequence homology
`
`1. Introduction
`
`Monoclonal antibodies form an important class of
`human therapeutics. Since the approval of Orthoclone
`OKT3 for treatment of allograft rejection in 1986, a total
`of 18 monoclonal antibodies, including nine humanized
`antibodies (Table 1), have been approved to date by the
`Food and Drug Administration (FDA) for therapeutic
`use in the United States [1]. The utility of monoclonal
`antibodies as therapeutics was recognized soon after the
`introduction of hybridoma technology in 1975 [2]. Due
`to their high aYnity and exquisite speciWcity, monoclo-
`nal antibodies can recognize even small quantities of
`antigen in complex mixtures and neutralize the function
`of antigens responsible for the onset or maintenance of
`disease. In addition, eVector functions associated with
`
`¤
`
`Corresponding author. Fax: +1 510 574 1500.
`E-mail address: ntsurushita@pdl.com (N. Tsurushita).
`
`1046-2023/$ - see front matter © 2005 Elsevier Inc. All rights reserved.
`doi:10.1016/j.ymeth.2005.01.007
`
`the Fc region, such as antibody-dependent cellular cyto-
`toxicity (ADCC) and complement-dependent cytotoxic-
`ity (CDC), eYciently trigger immune responses that
`result in the elimination of antibody-bound cells [3]. A
`number of rodent monoclonal antibodies with potential
`clinical applications have now been generated, of which
`the majority are derived from mice.
`The development of monoclonal antibodies as human
`therapeutics, however, was hampered by the problem
`that mouse antibodies are strongly immunogenic in
`humans [4–6]. In the vast majority of clinical studies,
`potent human anti-mouse antibody (HAMA) responses
`were observed in human subjects who were administered
`mouse monoclonal antibodies. As a result, mouse anti-
`bodies were neutralized and rapidly cleared from the
`body, resulting in limited eYcacy for such antibodies.
`Although the invention of mouse–human chimeric anti-
`bodies, which are composed of mouse variable regions
`and human constant regions [7], helped reduce the
`immunogenicity of mouse monoclonal antibodies in
`
`

`

`70
`
`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`Table 1
`Humanized antibodies approved for therapeutic use in the US
`
`Generic name
`
`Product name
`
`Antigen
`
`Indication
`
`Approval
`
`Company
`
`Daclizumab
`Palivizumab
`Trastuzumab
`Gemtuzumab Ozogamicina
`Alemtuzumab
`Omalizumab
`Efalizumab
`Bevacizumab
`Natalizumab
`
`Zenapax
`Synagis
`Herceptin
`Mylotarg
`CamPath
`Xolair
`Raptiva
`Avastin
`Tysabri
`
`CD25
`RSV gpF
`HER2/neu
`CD33
`CD52
`IgE
`CD11a
`VEGF
`♡4-integrin
`Abbreviations used: RSV gpF, respiratory syncytial virus glycoprotein F; AML, acute myeloid leukemia; CLL, chronic lymphocytic leukemia;
`VEGF, vascular endothelial growth factor.
`a Conjugated to calicheamicin.
`
`Renal allograft rejection
`RSV infection
`Breast cancer
`AML
`CLL
`Asthma
`Psoriasis
`Colorectal cancer
`Multiple sclerosis
`
`1997
`1998
`1998
`2000
`2001
`2003
`2003
`2004
`2004
`
`Protein Design Labs/Roche
`MedImmune
`Genentech
`Celltech/Wyeth
`Millennium/ILEX
`Tanox/Genentech/Novartis
`Xoma/Genentech
`Genentech
`Biogen Idec/Elan
`
`humans, chimeric antibodies still induced HAMA, or
`human anti-chimeric antibody (HACA), responses since
`the mouse-derived variable regions were suYcient to
`trigger immune responses in humans [5,6,8].
`Each of the heavy and light chain variable (V) regions
`forms a domain structure, composed of three comple-
`mentarity-determining regions (CDRs 1–3) and four
`framework regions (FRs 1–4), which belongs to the
`immunoglobulin superfamily. The CDRs of the heavy
`and light chain V domains together form the antigen-
`binding site, while the framework regions constitute a
`scaVold for the antigen-binding site. The concept of
`CDR grafting [9] for generating less immunogenic anti-
`bodies originated from the hypothesis that the CDRs of
`a mouse monoclonal antibody (the donor antibody) may
`replace those of a human antibody (the acceptor anti-
`body) without aVecting the structure of the antigen-
`binding site formed by the mouse CDRs. Although
`CDR grafting was successful in some cases [10,11], most
`CDR-grafted antibodies have been found not to retain
`the antigen-binding aYnity of the parental mouse anti-
`body. This is because certain framework residues inti-
`mately interact with CDR residues in the V domains,
`thereby aVecting the structure of the antigen-binding
`site. Thus, as pointed out by Queen et al. [12], the trans-
`fer of mouse CDR residues alone into human frame-
`works may alter the structure of the CDRs, resulting in a
`loss of antigen-binding aYnity. Queen and co-workers
`went on to propose that key framework residues inter-
`acting with the CDRs, and therefore important for the
`integrity of the antigen-binding site, should be trans-
`ferred from the donor to the acceptor antibody along
`with the CDRs. To identify such residues, Queen and co-
`workers used computer-generated three-dimensional
`models of V domains. By transferring CDR residues
`together with key framework amino acids from a mouse
`antibody into human frameworks, it became possible to
`routinely generate engineered antibodies, generally
`referred to as humanized antibodies, which retain the
`binding aYnity and speciWcity of the parental mouse
`antibodies. Since the introduction of computer-guided
`
`humanization technology, a large number of humanized
`antibodies have been successfully generated [13]. Clinical
`studies have indicated that humanized antibodies are, in
`general, much less immunogenic than mouse or chimeric
`antibodies, and are safe and well tolerated in humans
`[4,14,15]. Thus, the application of mouse antibodies to
`human therapy has become feasible through the use of
`humanization technology.
`Zenapax (generic name, daclizumab) is the Wrst
`humanized antibody approved by the FDA for human
`therapeutic use in the United States. It is a humanized
`IgG1 form of the mouse monoclonal antibody anti-Tac
`[16], an anti-human IL-2 receptor ♡ chain (CD25) anti-
`body that blocks the interaction of IL-2 with IL-2 recep-
`tor and thus prevents activation of T cells. Zenapax was
`approved in 1997 for prevention of renal allograft rejec-
`tion in the United States. To date Zenapax has been
`administered to over 20,000 patients, and has been found
`to be safe and eVective, thus fulWlling the concept of
`humanized antibodies.
`In this paper, we use the humanization of anti-Tac as
`an example to describe the process of antibody human-
`ization. Since the theoretical background of the human-
`ization methodology has been discussed elsewhere [17],
`we focus on technical aspects of antibody humanization
`in this paper. It should be noted that Zenapax is the
`trade name used by Roche Pharmaceuticals. We will
`hereafter use its generic name, daclizumab, to refer to the
`humanized form of the anti-Tac monoclonal antibody.
`
`2. Antibody humanization procedure
`
`The ultimate goal of antibody humanization is to
`generate human-like V regions by transferring CDR res-
`idues and a minimal number of key framework amino
`acids from a donor mouse monoclonal antibody onto an
`acceptor human framework without losing antigen-bind-
`ing aYnity and speciWcity. The computer-guided
`antibody humanization technique requires expertise
`primarily in two areas: three-dimensional modeling of
`
`

`

`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`71
`
`Cloning and sequencing ofVH and VL I
`t
`Design of humanized VH and VL I
`t
`t
`
`Expression of humanized antibodies
`
`Confirmation of binding properties
`
`Fig. 1. Overall scheme for antibody humanization.
`
`protein structures and genetic engineering. The overall
`Xow of a typical antibody humanization project is shown
`in Fig. 1. Since the experimental techniques required for
`genetic engineering are well documented in various
`methods books, such as Molecular Cloning: A Labora-
`tory Manual [18], and have been successfully applied to
`antibody engineering, we focus in this section only on
`the key considerations for successful humanization with-
`out describing the detailed experimental procedures.
`Regarding the computer-guided design of humanized
`antibodies, we describe the principle and application of
`our procedure in detail using daclizumab as an example.
`
`2.1. Cloning and sequencing mouse V genes
`
`A humanization project typically begins by cloning
`and sequencing the V genes from a rodent hybridoma,
`which is usually of mouse origin, but may also be derived
`from rat or hamster B cells. Before initiating V gene
`cloning, however, it is important to ensure that the
`hybridoma has been subcloned, ideally by single cell sub-
`cloning using a cell sorter, rather than by limiting dilu-
`tion. Without establishing
`the clonality of
`the
`hybridoma, multiple productive heavy and/or light chain
`mRNA sequences are occasionally observed even
`though the hybridoma may appear to be producing a
`single kind of antibody. It is formally possible for a
`clonal hybridoma to express more than one kind of pro-
`ductive heavy or light chain mRNA, although it is more
`likely in this case that the hybridoma is not clonal. While
`it is possible to experimentally determine which of the
`multiple productive heavy or light chain mRNAs encode
`the mouse antibody to be humanized, for example, by
`expressing all heavy and light chain combinations and
`examining their binding to antigen, it is usually more
`expeditious to perform one round of single cell sorting if
`
`there is any doubt about the clonality of the starting
`hybridoma.
`A number of methods are available and have been
`successfully used for cloning cDNAs encoding the VH
`and VL regions of the target monoclonal antibody. The
`5⬘ RACE (rapid ampliWcation of cDNA ends) method
`using, for example, the SMART RACE cDNA AmpliW-
`cation Kit (BD Biosciences, San Jose, CA) or the Gen-
`eRacer Kit (Invitrogen, Carlsbad, CA) is a common
`choice. To synthesize gene-speciWc primers for 5⬘ RACE,
`information about the isotype (and subtype) of the pro-
`duced monoclonal antibody is helpful. Isotyping of
`mouse monoclonal antibodies can be readily performed
`using a commercially available kit, for example, the Iso-
`Strip Mouse Monoclonal Antibody Isotyping Kit
`(Roche Applied Science, Indianapolis, IN) or the Mouse
`Immunoglobulin Isotyping Cytometric Bead Array Kit
`(BD Biosciences). Depending on the isotype of the heavy
`and light chains of the target monoclonal antibody, a
`gene-speciWc primer can be designed that binds immedi-
`ately downstream of the variable region for each of the
`heavy and light chains. In the case of ♤ heavy chains, a 5⬘
`RACE primer may be designed to be speciWc for each
`subtype (♤1, ♤2a, ♤2b or ♤3 in mice) or alternatively a
`region identical or highly homologous among all the ♤
`subtypes may be chosen as a basis for designing a com-
`mon primer. The following set of 5⬘ RACE primers has
`been successfully used in our laboratory for cDNA clon-
`ing of the variable regions of mouse ♤ heavy and ♭ light
`chains: 5⬘-GCCAGTGGATAGACTGATGG-3⬘ (for
`mouse ♤1, ♤2a, and ♤2b heavy chains) and 5⬘-GAT GGA
`TACAGTTGGTGCAGC-3⬘ (for mouse ♭ light chains).
`PCR-ampliWed V gene fragments can be directly cloned
`into a plasmid vector, for example, using the Zero
`Blunt TOPO PCR Cloning Kit (Invitrogen). The
`cloned fragments are then subjected to DNA sequenc-
`ing to determine the nucleotide sequences of the VH
`and VL regions. Since nucleotide substitutions are
`occasionally introduced by PCR, it is important to
`sequence several clones to obtain correct V region
`sequences.
`It should be noted that hybridoma cells often pro-
`duce two kinds of light chain mRNAs, one encoding
`the productive light chain and another encoding a non-
`productive form. The latter type of mRNA usually car-
`ries a frameshift and/or non-sense mutation(s) in the V
`region, typically near or in CDR3. Since identical, or
`very similar, non-productive VL sequences have been
`observed in many hybridomas generated using the
`same fusion partner, it is likely that the non-productive
`light chain mRNA originated from the fusion partner
`and its gene was retained in the hybridoma. The ratio
`between productive and non-productive VL cDNAs
`observed during cloning varies substantially among
`hybridomas. If the antibody production level of a
`hybridoma is very low (less than 1 ♯g/ml), more than
`
`

`

`72
`
`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`90% of VL cDNA clones might encode the non-pro-
`ductive type. Therefore, to obtain several cDNA clones
`encoding a productive type of VL, sequencing of a few
`dozen clones might be necessary. However, if a non-
`productive type of VL is exclusively obtained during
`cDNA cloning, an alternative approach is to design a
`pair of PCR primers to identify and exclude the non-
`productive VL gene. This is typically done using CDR1
`to design a forward primer and CDR3 to design a
`reverse primer for the non-productive gene. The prim-
`ers are then used for PCR ampliWcation of the plasmid
`clones containing VL genes; clones ampliWed by the
`primers are excluded from further analysis, while the
`remaining clones are sequenced to identify productive
`VL genes.
`Although the authenticity of the cloned VH and VL
`genes can be conWrmed by demonstrating antigen bind-
`ing using recombinant mouse or chimeric forms of the
`antibody, N-terminal amino acid sequencing of the tar-
`get mouse monoclonal antibody is usually more expedi-
`ent. Sequencing of N-terminal amino acids may be
`achieved with a protein sequencer such as a Model 241
`Protein Sequencer (Hewlett–Packard, Palo Alto, CA)
`using a standard protocol, or more typically by sending
`the sample to a contract laboratory for a reasonable fee.
`Heavy and light chains may be separated before amino
`acid sequencing by polyacrylamide gel electrophoresis
`under reducing conditions, or alternatively whole anti-
`body may be subjected to sequencing. In the latter case,
`two amino acid residues, one each from the heavy and
`light chains, are usually observed in most of the
`sequencing cycles, unless the N-terminus of the heavy
`and/or light chains is blocked by conversion of gluta-
`mine to pyroglutamine [19]. The assignment of two
`amino acids detected at each position to either VH or
`VL can be performed relatively easily by comparison to
`the V region sequences in the Kabat antibody database
`[20,21]. When only one amino acid is detected in a given
`cycle, it is likely that the heavy and light chains share
`the same amino acid at that position. A typical amino
`acid sequence determination by Edman degradation
`provides at least 15–20 amino acids from the N-termi-
`nus, which is usually suYcient to conWrm the authentic-
`ity of the VH and VL sequences obtained by cDNA
`cloning.
`It is important to note that glutamine is one of the
`two most common N-terminal amino acids of mouse
`heavy chain variable regions [20]. Thus, if only the light
`chain sequence is obtained from standard amino acid
`sequencing of whole antibody, deblocking the N-termi-
`nus of the heavy chain is necessary to obtain its sequence
`[22]. Also, the N-terminus of a mature mouse light chain
`is occasionally blocked since mouse germline V segments
`belonging to subgroup IV often start with glutamine
`[20]. Rarely, both the heavy and light chains of a mouse
`antibody start with glutamine.
`
`Three-dimensional modeling
`ofV regions
`
`Selection of human
`frameworks
`
`t
`Design of humanized V regions I
`t
`
`Inspection of designed V regions
`
`Fig. 2. Procedure for designing humanized V regions.
`
`2.2. Designing humanized antibodies
`
`For successful humanization of a mouse (or other
`rodent) antibody, the set of mouse framework residues
`potentially important for maintaining the structure of
`the CDRs is Wrst identiWed and then transferred onto a
`human framework that is selected based on homology to
`the mouse framework, together with the mouse CDR
`residues. Building a reliable three-dimensional model of
`the variable regions is an important Wrst step in the
`design of a humanized antibody. The process of design-
`ing humanized V regions is outlined in Fig. 2.
`
`2.2.1. Three-dimensional modeling of V regions
`A detailed all-atom model of the heavy and light
`chain variable regions of a mouse antibody can be con-
`structed following the method described by Levitt and
`co-workers [23,24] using the algorithm ABMOD and the
`ENCAD set of programs and molecular mechanics
`energy function. The principle of the model building is as
`follows. Each of the heavy and light variable domains is
`divided into 14 structurally meaningful segments; these
`segments are ♢ strands and loop-like structures compris-
`ing the domain structure of the immunoglobulin super-
`family. The amino acid sequence of each of the 28
`segments (14 from each of the heavy and light chain var-
`iable regions) of the mouse antibody is aligned with the
`corresponding segments of antibodies of known struc-
`ture in the PDB database [25]. For this purpose, we
`choose a small number of antibodies (typically under 30)
`that best match the murine antibody to be humanized,
`and these sequences along with the murine sequences are
`then subjected to multiple sequence alignment following
`Levitt’s method [23]; in our experience, CLUSTALW
`[26,27] has also worked well for this purpose. For each of
`the 28 segments, a corresponding segment having the
`highest sequence homology is selected from the multiple
`
`

`

`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`73
`
`segments. Even though 13 crystal structures were ini-
`tially chosen for alignment in this example, only six were
`used to model the individual segments of the heavy and
`light chains. This is typical for building antibody models
`of the variable regions from the primary sequences
`alone. The resulting three-dimensional model of the anti-
`Tac variable regions is shown in Figs. 3A and B.
`The structural parameters of the selected segments are
`then combined to build a model of the variable regions. At
`this stage, the resulting structure is not considered to be
`optimal, because the selected segments are likely to belong
`to multiple antibodies with diVerent structures. Further-
`more, some of the segments, particularly the CDRs, may
`have deletions or insertions compared to the original
`mouse sequences. To obtain a reliable structure, the model
`must be subjected to multiple cycles of conjugate gradient
`energy minimization (e.g., using ENCAD, or as described
`by Press et al. [28]) to relax the structure until the energy
`and the energy gradient reach an acceptable level.
`For antibody humanization
`in our
`laboratory,
`ABMOD and ENCAD (with our own modiWcations) are
`routinely used for modeling the structures of variable
`regions; however, other model building software and
`energy functions, for example, AMBER [29], could also be
`used with similar results. In addition, there are currently
`many websites that allow one to build three-dimensional
`models from primary sequences; for instance, 3D-JIG-
`SAW
`(http://www.bmm.icnet.uk/servers/3djigsaw) and
`SWISS-MODEL (http://www.expasy.org/swissmod/) are
`available to the interested reader.
`
`2.2.2. Selection of human frameworks
`In parallel with modeling the structure of the variable
`regions, each of the mouse VH and VL region amino
`acid sequences deduced from cDNA cloning is com-
`pared to human V region sequences in the database.
`
`Table 2
`Variable region segments used for modeling of the mouse anti-Tac V regions
`
`Sequence
`
`Structure
`
`Segment used
`for modelinga
`
`VH residues
`1–10
`11–15
`16–26
`27–32
`33–43
`44–52
`52a–55
`56–66
`67–76
`77–82c
`83–88
`89–95
`96–99
`100–113
`
`QVQLQQSGAE
`LAKPG
`ASVKMSCKASG
`YTFTSY
`RMHWVKQRPGQ
`GLEWIGYIN
`PSTG
`YTEYNQKFKDK
`ATLTADKSSS
`TAYMQLSSL
`TFEDSA
`VYYCARG
`GGVF
`DYWGQGTTLTVSS
`
`♢ strand
`
`♢ strand
`♢ strand
`
`♢ strand
`
`♢ strand
`
`♢ strand
`
`VL residues
`QI
`1–2
`VLTQSPA
`3–9
`IMSASP
`10–15
`GEKVTITCSA
`16–25
`SSSISY
`26–32
`MHWFQQKP
`33–40
`GTSPKLWIY
`41–49
`TTSNLAS
`50–56
`GVPARFSGSGSG
`57–68
`TSYSLTISR
`69–77
`MEAED
`78–82
`AATYYCHQ
`83–90
`RSTYPL
`91–96
`97–107
`TFGSGTKLELK
`a IdentiWcation code in the PDB database.
`
`♢ strand
`
`♢ strand
`
`♢ strand
`♢ strand
`
`♢ strand
`♢ strand
`
`♢ strand
`
`1JHL
`1NMB
`1NGQ
`1NGQ
`1NMB
`1NGQ
`1NGQ
`1JHL
`1NMB
`1NMB
`1NMB
`1NMB
`1NMB
`1NMB
`
`1BAF
`1BAF
`1BAF
`1BAF
`1CLO
`1FOR
`1BAF
`1BAF
`1BAF
`1BAF
`1BAF
`1BAF
`1FOR
`1BAF
`
`the procedure
`illustrate
`sequence alignment. To
`described above, a three-dimensional model of the
`mouse anti-Tac variable regions was built. The crystal
`structures shown in Table 2 were used for the 28
`
`Fig. 3. (A) Three-dimensional model of the mouse anti-Tac V regions. Color scheme: white, framework residues; red, CDR residues; blue, mouse-spe-
`ciWc residues retained in the humanized form due to contact with the CDRs; yellow, substituted with human consensus residues in the humanized
`form. (B) Spatial location of an alanine at position 67 (blue) in VH, which was determined to be a key framework residue for proper formation of the
`CDR structure. Framework (white) and CDR (red) residues located within about 5 Å of the alanine at position 67 in VH are shown.
`
`

`

`74
`
`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`Currently, the Kabat database [21] provides a wide selec-
`tion of variable region sequences that are suitable for
`this purpose. Although an updated database can be
`created by including more variable region sequences
`recently introduced into GenBank and/or identiWed
`within each laboratory, inspection of the Kabat data-
`base is normally suYcient to Wnd a human framework
`sequence suitable for antibody humanization. As an
`acceptor for humanization, a human V region frame-
`work sequence with high homology to the mouse
`sequence is preferred. The Smith–Waterman algorithm
`[30], or other standard methods such as BLAST [31], can
`be used to scan the database for sequences having high,
`e.g., at least 65%, homology to the mouse sequence.
`For humanization of the mouse anti-Tac monoclonal
`antibody, the human Eu antibody [20] was chosen to
`provide VH and VL frameworks. The amino acid identi-
`ties in the framework regions between mouse anti-Tac
`and human Eu are 67% for VH and 65% for VL. The VH
`and VL amino acid sequences of the mouse anti-Tac and
`human Eu antibodies are aligned in Fig. 4. Note that
`
`although VH and VL frameworks are often chosen from
`the same human antibody, this is not essential unless a
`three-dimensional model of the designed humanized V
`region indicates conformational changes at the interface
`of VH and VL domains (discussed below in Section
`2.2.4). Indeed, many humanized antibodies have been
`successfully generated using human VH and VL frame-
`works that originated from diVerent antibodies.
`Although a human acceptor framework can generally
`be chosen from V region sequences expressed in B cells,
`i.e., cDNA-based and protein-derived sequences, the
`recent completion of the sequencing of the human
`genome allows the use of germline V segments for this
`purpose. The advantage of using a germline V segment
`is, at least in theory, to eliminate potential immunogenic-
`ity associated with somatic hypermutations in cDNA-
`based and protein-derived sequences. The use of a
`consensus framework sequence is another approach to
`this issue. When using a cDNA-based or protein-derived
`V region sequence as an acceptor for humanization, the
`identiWcation and removal of hypermutated residues in
`
`A
`
`4
`3
`2
`1
`123456789 0123456789 0123456789 0123456789 0123456789
`Mouse anti-Tac QVQLQQSGA ELAKPGASVK MSCKASGYTF TSYRMHWVKQ RPGQGLEWIG
`QVQLVQSGA EVKKPGSSVK VSCKASGYTF !SYRMHWVRQ APGQGLEW!G
`Daclizumab
`QVQLVQSGA EVKKPGSSVK VSCKASGGTF S-----WVRQ APGQGLEWMG
`Eu
`
`7
`8
`6
`5
`01223456789 0123456789 0123456789 0122223456789
`a
`abc
`Mouse anti-Tac YINPSTGYTEY NQKFKDKATL TADKSSSTAY MQLSSLTFEDSAV
`YINPSTGYTEY NQKFKDKATI TADESTNTAY MELSSLRSEDTAi
`Daclizumab
`----------- ------RVTI TADESTNTAY MELSSLRSEDTAF
`Eu
`1
`1
`1
`0
`9
`0123456789 0123456789 0123
`Mouse anti-Tac YYCARGGGVF -DYWGQGTTL TVSS
`~CAiGGGVF -DYNru2GILV TVSS
`Daclizumab
`YFCAG-----
`---EYNGGLV TVSS
`Eu
`
`B
`
`4
`3
`2
`1
`123456789 0123456789 0123456789 0123456789 0123456789
`Mouse anti-Tac QIVLTQSPA IMSASPGEKV TITCSASS-S ISYMHWFQQK PGTSPKLWIY
`DIQMTQSPS TLSASVGDRV TITCSASS-S ISYMHWYQQK PGKAPKLLJY
`Daclizumab
`DIQMTQSPS TLSASVGDRV TITC------
`-----WYQQK PGKAPKLLMY
`Eu
`
`7
`9
`8
`6
`5
`0123456789 0123456789 0123456789 0123456789 0123456789
`Mouse anti-Tac TTSNLASGVP ARFSGSGSGT SYSLTISRME AEDAATYYCH QRSTYPLTFG
`TTSNLASGVP ~FgGSGSGT EFTLTISSLQ PDDFATYYCH QRSTYPLTFG
`Daclizumab
`-------GVP SRFIGSGSGT EFTLTISSLQ PDDFATYYC- --------FG
`Eu
`
`1
`0
`01234567
`SGTKLELK
`QGTKVEVK
`QGTKVEVK
`
`Mouse anti-Tac
`Daclizumab
`Eu
`
`Fig. 4. Alignment of mouse, humanized, and human Eu acceptor amino acid sequences of anti-Tac VH (A) and VL (B). Amino acid sequences are
`shown in single-letter code. Vertical numbers show amino acid location according to Kabat [20]. CDR residues according to Kabat are underlined in
`the mouse anti-Tac VH and VL sequences. The CDR residues in the human Eu V region are omitted. Single-underlined residues in the daclizumab
`sequences are mouse-speciWc amino acids retained in the humanized form due to contact with the CDRs. Double-underlined residues are substitu-
`tions to human consensus amino acids.
`
`

`

`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`75
`
`the framework, as practiced in Queen et al.’s method [12]
`and described below in Section 2.2.3.2, also produces a
`satisfactory result in solving the potential immunogenic-
`ity problem. Therefore, the use of a cDNA-based, pro-
`tein-derived, germline, or
`consensus
`framework
`sequence may result in a similar, if not identical, human-
`ized V region sequence.
`When choosing a germline VH segment as an accep-
`tor framework, careful consideration must be given to its
`chromosomal location. Human germline VH segments
`are encoded on chromosomes 14, 15, and 16; however,
`only the VH segments on chromosome 14 are actually
`used for VDJ recombination to form functional VH
`regions, because the germline JH segments and heavy
`chain constant regions are also encoded on chromosome
`14 [32]. The use of germline VH segments on chromo-
`somes 15 or 16 in designing a humanized antibody
`should thus be avoided since this might result in unex-
`pected structural defects in the framework.
`
`2.2.3. Design of humanized V regions
`Once a three-dimensional model of the mouse vari-
`able regions has been built, and human frameworks have
`been selected, several considerations are relevant in
`designing humanized V regions. Foremost among these
`is the identiWcation of mouse framework residues that
`could potentially contribute to antigen binding. It is also
`advisable to identify atypical residues in the acceptor
`frameworks that might represent a potential source of
`immunogenicity. In addition, it is prudent to examine
`the variable region sequences for the presence of poten-
`tial N-linked glycosylation signals since carbohydrate
`groups may aVect antigen binding. Finally, selection of
`an appropriate heavy chain isotype is an important con-
`sideration in designing humanized antibodies.
`
`2.2.3.1. IdentiWcation of key framework residues. Using a
`three-dimensional structure model of the variable
`regions of the mouse antibody, mouse framework resi-
`dues that could potentially inXuence the conformation
`of the antigen-binding site or directly interact with the
`antigen, termed key framework residues, need to be iden-
`tiWed. According to the humanization method of Queen
`et al. [12], framework amino acids within about 4–6 Å of
`the CDRs are candidates for designation as key frame-
`work residues. For example, Fig. 3B shows that the side
`chain of an alanine residue at position 67 (all amino acid
`positions are numbered according to the Kabat number-
`ing system) in the anti-Tac VH is located within 5 Å of
`four CDR residues. This process can be achieved using a
`computer program,
`such as RASMOL
`(http://
`www.umass.edu/microbio/rasmol/), that calculates inter-
`atomic distances from the atomic coordinates or, though
`tedious, through manual inspection of a computer
`model. If amino acids at key framework positions diVer
`between mouse donor and human acceptor sequences,
`
`they are considered for replacement in the humanized
`form of the antibody. For this purpose, the location and
`orientation of the side chains of the candidate key
`framework residues relative to the CDR residues are
`examined, along with the extent of their solvent interac-
`tions. From such analyses, the impact on the CDR struc-
`ture of replacing a mouse residue with its human
`counterpart can be determined for each of the candidate
`key framework residues. If the impact is considered to be
`minimal, for example, a serine occurs in the human
`framework and a threonine at the corresponding mouse
`framework position appears to have a relatively small
`interaction with the CDR, it is recommended to choose
`the human amino acid. This process should be carried
`out judiciously to minimize the number of mouse-spe-
`ciWc residues that are retained in the humanized form.
`For humanization of mouse anti-Tac, following the
`criteria described above, a total of nine amino acids were
`identiWed in the human Eu frameworks for substitution
`with the corresponding mouse residues due to their
`potential contributions to the formation of the antigen-
`binding site. These amino acids are at positions 27, 30,
`48, 66, 67, 94, and 103 in the VH, and at positions 48 and
`60 in the VL (Fig. 4) [12].
`
`2.2.3.2. Potential immunogenicity. While a humanized anti-
`body designed according to the procedures described
`above would be expected to maintain the aYnity and
`speciWcity of the parental antibody, additional factors
`ought to be considered for humanization. One of the
`additional steps we employ is to Wnd atypical, or rare,
`amino acid residues in the selected human acceptor
`framework sequences [12]. This process is particularly
`important when a protein- or cDNA-derived human
`framework sequence is used as an acceptor for human-
`ization. An amino acid residue that occurs in less than
`about 10–20% of the variable region sequences belong-
`ing to the same subgroup is labeled “atypical” in our
`standard procedure. If these atypical amino acids in the
`selected human framework sequence are diVerent from
`their counterparts in its potential ancestral germline V
`segment, they are considered to be the result of somatic
`hypermutation during aYnity maturation. A somatically
`hypermutated amino acid in the human acceptor frame-
`work that has little contribution to the CDR structure is
`a potential source of immunogenicity in humans and
`therefore we replace it with the donor amino acid at that
`position if it happens to be a common amino acid in
`human V regions, or alternatively with a consensus
`amino acid residue from the same human V region sub-
`group, or as a third alternative with an amino acid at the
`corresponding location of the ancestral human germline
`V segment.
`In the case of humanized anti-Tac [12], nine amino
`acids in the Eu frameworks, at positions 89, 91, 94, 103,
`104, 105, and 107 in VH, and at positions 48 and 63 in
`
`

`

`76
`
`N. Tsurushita et al. / Methods 36 (2005) 69–83
`
`VL, were determined to be atypical and were therefore
`replaced with the corresponding amino acids from the
`donor antibody, which are common or consensus amino
`acids at those positions in human V regions. Among
`these nine positions, three amino acids, at positions 94
`and 103 in VH, and at position 48 in VL, were also iden-
`tiWed as key framework re