CA2129219C - Humanised antibodies - Google Patents

Abstract

CDR-grafted antibody heavy and light chains comprise acceptor framework and donor antigen binding regions, the heavy chains comprising donor residues at at least one of positions 6, 23 and/or 24, 48 and/or 49, 71 and/or 73, 75 and/or 76 and/or 78 and 88 and/or 91. The CDR-graftd light chains comprise donor residues at at least one of positions 1 and/or 3 and 46 and/or 47 or at at least one of positions 46, 48, 58 and 71. The CDR-grafted antibodies are preferably humanised antibodies, having non human, e.g. rodent, donor and human acceptor frameworks, and may be used for in vivo therapy and diagnosis. A
generally applicable protocol is disclosed for obtaining CDR-grafted antibodies.

Description

212~21~
HUMANISEI~ ANTIBODIES
This application is a division of our co-pending application Serial No.

2,037,607 filed March 6, 1991.

Field of the Invention S The present invention relates to humanised antibody molecules, to processes for their production using recombinant DNA technology, and to their therapeutic uses.

The term "humanised antibody molecule" is used to describe a molecule having an antigen binding site derived from an immunoglobulin from a non-human species, and rem~ining immunoglobulin-derived parts of the molecule being derived from a human immunoglobulin. The antigen binding site typically comprises complementarity determining regions (CDRs) which determine the binding specificity of the antibody molecule and which are carried on app-ropriate framework regions in the variable domains. There are 3 CDRs (CDRl, CDR2 and CDR3) in each of the heavy and light chain variable domains.

In the description, reference is made to a number of publications by number.
The publications are listed in numerical order at the end of the description.

Back~round of the Invention Natural immunoglobulins have been known for many years, as have the various fragments thereof, such as the Fab, (Fab')2 and Fc fragments, which can be derived by enzymatic cleavage. Natural immunoglobulins comprise a generally Y-shaped molecule having an antigen-binding site towards the end of each upper arm. The remainder of the structure, and particularly the stem of the Y, mediates the effector functions associated with immunoglobulins.

Natural immunoglobulins have been used in assay, diagnosis and, to a more limited extent, therapy. However, such uses, especially in therapy, were hindered until recently by the polyclonal nature of natural immunoglobulins.
A significant step towards the realisation of the potential ' ' - 2 - 21~9213 Of ;mm-lnoglobulins as therapeutic agents was the discovery of procedures for the production of monoclonal antibodies (MAbs) of defined specificity (l).

However, most MAbs are produced by hybridomas which are fusions of rodent spleen cells with rodent myeloma cells. They are therefore essentially rodent proteins.
There are very few reports of the production of human MAbs.

Since most available MAbs are of rodent origin, they are naturally antigenic in humans and thus can give rise to an undesirable immune response termed the EAMA (Human Anti-Mouse Antibody~ response. Therefore, the use of rodent MAbs as therapeutic agent~ in humans is inherently limited by the fact that the human subject will mount an immunological response to the MAb and will either remove it entirely or at least reduce its effectiveness. In practice, MAbs of rodent origin may not be used in patients for more than one or a few treatments as a ~AMA
response soon develops rendering the ~Ab ineffective as well as giving rise to undesirable reactions. For instance, OKT3 a mouse IgG2a/k MAb which recognises an antigen in the T-cell receptor-CD3 complex has been approved for use in many countries throughout the world as an immunosuppressant in the treatment of acute allograft rejection [Chatenoud et al (2) and Jeffers et al (3)]. However, in view of the rodent nature of this and other such MAbs, a significant HAMA response which may include a major anti-idiotype component, may build up on use. Clearly, it would be highly desirable to ~;mi n; sh or abolish this undesirable ~AMA response and thus enlarge the areas of use of these very useful antibodies.

Proposals have therefore been made to render non-human MAbs less antigenic in humans. Such techniques can be generically termed ~humanisation" techniques. These ~ 3 _ 2~ 2~2~

techniques typically involve the use of recombinant DNA
technology to manipulate DNA se~uences encoding the polypeptide chains of the antibody molecule.

Early methods for humanising MAbs involved production o~
chimeric antibodies in which an antigen binding site comprising the complete variable domains of one antibody is linked to constant domains derived from another antibody. Methods for carrying out such chimerisation procedures are described in EP0120694 ~Celltech Limited), EP0125023 (Genentech Inc. and City of Hope), EP-A-O 171496 (Res. Dev. Corp. Japan), EP-A-O 173 494 (Stanford University), and WO 86fO1533 (Celltech Limited). This latter Celltech application (WO 86/01533) discloses a process for preparing an antibody molecule having the variable domains from a mouse MAb and the constant ~om~; n~
from a human immunoglobulin. Such humanised chimeric antibodies, however, still contain a significant proportion of non-human amino acid sequence, i.e. the complete non-human variable domains, and thus may still elicit some HAMA response, particularly if ~m; n; stered over a prolonged period tBegent et al (ref. 4)~.

In an alternative approach, described in EP-A-0239400 (Winter), the complementarity determining regions (CDRs) o~ a mouse MAb have been grafted onto the framewor~
regions o~ the variable domains of a human immunoglobulin by site directed mutagenesis using long oligonucleotides.
The present invention relates to humanised antibody molecules prepared according to this alternative approach, i.e. CDR-grafted humanised antibody molecules. Such CDR-grafted humanised antibodies are much less likely to give rise to a HAMA response than humanised chimeric antibodies in view of the much lower proportion of non-human amino acid se~uence which they contain.

_ 4 - ~ ~292~

The earliest wor~ on humanising MAbs by CDR-grafting was carried out on MAbs recognising synthetic antigens, such as the NP or NIP antigens. ~owever, examples in which a mouse MAb recognising lysozyme and a rat MAb recognising an antigen on human T-cells were hllmAn;sed by CDR-grafting have been described by Verhoeyen et al (5) and Riech~nn et al (6) respectively. The preparation of CDR-grafted antibody to the antigen on human T cells is also described in WO 89/07452 (Medical Research Council).

In Riechmann et al/Medical Research Council it wa~ found that transfer of the CDR regions alone ras defined by Kabat refs. (7) and (8)] was not sufficient to provide satisfactory antigen binding activity in the CDR-grafted product. Riechmann et al found that it was necessary to convert a serine residue at position 27 of the human sequence to the corresponding rat phenylalanine residue to obtain a CDR-grafted product having improved antigen binding activity. This residue at position 27 of the heavy chain is within the structural loop adjacent to CDR1. A further construct which additionally contained a human serine to rat tyrosine change at position 30 of the heavy chain did not have a significantly altered binding activity over the humanised antibody with the serine to phenylalanine change at position 27 alone. These results indicate that changes to residues of the human sequence outside the CDR regions, in particular in the structural loop adjacent to CDR1, may be necessary to obtain effective antigen binding activity for CDR-grafted antibodies which recognise more complex antigens. Even so the binding affinity of the best CDR-grafted antibodies obtained wa~ still significantly less than the original MAb.

Very recently Queen et al (9) have described the preparation of a humanised antibody that binds to the . - 5 - 2~292 .

interleukin 2 receptor, by combining the CDRs of a murine MAb (anti-Tac) with human immunoglobulin framework and constant regions. The human framework regions were chosen to m~i m; se homology with the anti-Tac MAb sequence. In addition computer modelling was used to identify framework amino acid residues which were li~ely to interact with the CDRs or antigen, and mouse ~;no acids were used at these positions in the humanised antibody.

In WO 90/07861 Queen et al propose four criteria for designing humanised immunoglobulins. The first criterion is to use as the human acceptor the framewor~ from a particular human immunoglobulin that is unusually homologou~ to the non-human donor immunoglobulin to be humanised, or to use a consensus framework from many h~ n antibodies. The second criterion is to use the donor amino acid rather than the acceptor if the human acceptor residue is unusual and the donor re~idue i~ typical for human sequences at a specific residue of the framewor~.
The third criterion is to use the donor framework amino acid residue rather than the acceptor at positions immediately adjacent to the CDRs. The fourth criterion is to use the donor amino acid residue at framewor~
positions at which the amino acid is predicted to have a side chain atom within about 3 ~ of the CDRs in a three-dimensional immunoglo~ulin model and to be capable of interacting with the antigen or with the CDRs of the humanised immunoglobulin. It is proposed that criteria two, three or four may be applied in addition or alternatively to criterion one, and may be applied singly or in any combination.

WO 90/07861 describes in detail the preparation of a single CDR-grafted humanised antibody, a humanised antibody having specificity for the p55 Tac protein of the ~ 6 - 2~292~ ~

IL-2 receptor. The combination of all four criteria, as above, were employed in designing this humanised antibody, the variable region frameworks of the human antibody Eu (7) being used as acceptor. In the resultant hllm~nised antibody the donor CDRs were as defined by Kabat et al (7 and 8) and in addition the mouse donor residues were used in place of the human acceptor residues, at positions 27, 30, 48, 66, 67, 89, 91, 94, 103, 104, 105 and 107 in the heavy chain and at positions 48, 60 and 63 in the light chain, of the variable region frameworks. The humanised anti-Tac antibody obtained is reported to have an affinity for p55 of 3 x 109 M-1, about one-third of that of the murine MAb.

We have further investigated the preparation of CDR-grafted humanised antibody molecules and have identified a hierarchy of positions within the framework of the variable regions (i.e. outside both the Kabat CDRs and structural loops of the variable regions) at which the amino acid identities of the residues are important for obtaining CDR-grafted products with satisfactory binding affinity. This has ena~led us to establish a protocol for obtaining satisfactory CDR-grafted products which may be applied very widely irrespective of the level of homology between the donor ;~munoglobulin and acceptor framework. The set of residues which we have identified as being of critical importance does not coincide with the residues identified by Queen et al (9).

Summary of the Invention Accordingly, in a first aspect the invention provides a CDR-grafted antibody heavy chain having a variable region domain comprising acceptor framework and donor antigen binding regions wherein the framework comprises donor residues at at least one of positions 6, 23 and/or 24, 48 and/or 49, 71 and/or 73, 75 and/or 76 and/or 78 and 88 and/
or 91.

-' _ 7 _ 21292 In preferred embodiment~, the heavy chain framewor~
comprises donor residues at positions 23, 24, 49, 71, 73 and 78 or at positions 23, 24 and 49. The residue~ at positions 71, 73 and 78 of the heavy chain framework are preferably either all acceptor or all donor residues.

In particularly preferred embodiments the heavy chain framework additionally comprises donor residues at one, some or all of positions 6, 37, 48 and 94. Also it is particularly preferred that residues at positions of the heavy chain framework which are commonly conserved across species, i.e. positions 2, 4, 25, 36, 39, 47, 93, 103, 104, 106 and 107, if not conserved between donor and acceptor, additionally comprise donor residues. ~ost preferably the heavy chain framework additionally lS comprises donor residues at positions 2, 4, 6, 25, 36, 37, 39, 47, 48, 93, 94, 103 r 104, 106 and 107.

In addition the heavy chain framework optionally comprises donor residues at one, some or all of positions:
1 and 3, 72 and 76, 69 (if 48 is different between donor and acceptor), 38 and 46 (if 48 is the donor residue), 80 and 20 (if 69 is the donor residue), 67, 82 and 18 (if ~7 is the donor residue), 91, 88, and any one or more of 9, 11, 41r 87, 108, 110 and 112.

In the first and other aspects of the present invention reference is made to CDR-gra~ted antibody products comprising acceptor framework and donor antigen binding regions. It will be appreciated that the invention is widely applicable to the CDR-grafting of anti~odies in ~c - 8 - ~ 2~2~
. ~

general. Thus, the donor and acceptor antibodies may be derived from ~n;mAls of the same species and even same antibody class or sub-class. More usually, however, the donor and acceptor antibodies are derived from ~n;~l S of different species. Typically the donor antibody is a non-human antibody, such as a rodent MAb, and the acceptor antibody i5 a human antibody.

In the first and other aspects of the present invention, the donor antigen binding region typically comprises at 102 least one CDR from the donor antibody. Usually the donor antigen binding region comprises at least two and preferably all three CDRs of each of the heavy chain and/or light chain variable regions. The CDRs may comprise the Kabat CDRs, the structural loop CDRs or a 15 composite of the Kabat and structural loop CDRs and any combination of any of these. Preferably, the antigen binding regions of the CDR-grafted heavy chain variable domain comprise CDRs corresponding to the Kabat CDRs at CDR2 (residues 50-65) and CDR3 (residues 95-100) and a 20 composite of the Kabat and structural loop CDRs at CDRl (residues 26-35).

The residue designations given above and elsewhere in the present application are n1lmhered according to the Kabat numbering ~refs. (7) and (8)]. Thus the residue 25 designations do not always correspond directly with the linear numbering of the amino acid residues. The actual linear amino acid se~uence may contain fewer or additional aMino acids than in the strict Ka~at numbering corresponding to a shortening of, or insertion into, a 30 structural component, whether framewor~ or CDR, of the basic variable domain structure. For example, the heavy chain variable region of the anti-Tac antibody described by Queen et al (9) contains a single amino acid insert (residue 52a) after residue 52 of CDR2 and a three amino -~ 9 ~ 212~2~
~
acid insert (residues 82a, 82b and 82c) after framework residue 82, in the Rabat numhering. The correct Kabat numbering of residues may be determined for a given antibody by alignment at regions of homology of the sequence of the antibody with a "st~n~Ard" Kabat numbered sequence.

The invention also provides in a second aspect a CDR-grafted antibody light chain having a variable region domain comprising acceptor framework and donor antigen binding regions wherein the framework comprises donor residues at at least one of position~ 1 and/or 3 and 46 and/or 47. Preferably the CD~ grafted light chain of the second aspect comprises donor residues at positions 46 and/or 47.

The invention also provides in a third aspect a CDR-grafted antibody light chain having a variable region domain comprising acceptor framework and donor antigen binding regions wherein the framework comprises donor residues at at least one of positions 46, 48, 58 and 71.

ZO In a preferred embodiment o~ the third aspect, the framework comprises donor residues at all of positions 46, 48, 58 and 71.

In particularly preferred embodiments of the second and third aspects, the framewor~ additionally comprises donor residues at positions 36, 44, 47, 85 and 87. Similarly positions of the light chain framework which are commonly conserved across species, i.e. positions 2, 4, 6, 35, 49, 62, 64-69, 98, 99, 101 and 102, if not conserved between donor and acceptor, additionally comprise donor residues.
Most preferably the light chain framework additionally comprises donor residues at positions 2, 4, 6, 35, 36, 38, 44, 47, 49, 62, 64-69, 85, 87, 98, 99, 101 and 102.

~292~

.

In addition the framewor~ of the second or third aspects optionally comprises donor residues at one, some or all of positions:
1 and 3, 63, 60 (if 60 and 54 are able to form at potential saltbridge), 70 (if 70 and 24 are able to form a potential saltbridge)r 73 and 21 (if 47 is different between donor and acceptGr), 37 and 45 (if 47 is different between donor and acceptor), and any one or more of 10, 12, 40, 80, 103 and 105.

Preferably, the antigen binding regions of the CDR-grafted light chain variable domain comprise CDRs corresponding to the Kabat CDRs at CDR1 (residue 24-34), CDR2 (residues 50-5~) and CDR3 (residues 89-97).

The invention further provides in a fourth aspect a C~-grafted antibody molecule comprising at least one CDR-grafted heavy chain and at least one CDR-grafted light chain according to the first and second or first and third aspect~ of the invention.

The humanised antibody molecules and chains of the present in~ention may comprise: a complete antibody molecule, having full length heavy and light chains; a fragment thereof, such as a Fab, (Fab')2 or FV fragment; a light chain or heavy chain monomer or dimer; or a single chain antibody, e.g. a single chain FV in which heavy and light chain variable regions are joined by a peptide lin~er; or any other CD~-grafted molecule with the same specificity as the original donor antibody. Similarly the CD~-grafted heavy and light chain variable region may be combined with other antibody d~m~; n5 as appropriate.

2~2921 ~
, ~
Also the heavy or light chains or humanised antihody molecules of the present invention may have attached to them an effector or reporter molecule. For instance, it may have a macrocycle, for chelating a heavy metal atom, or a toxin, such as ricin, attached to it by a covalent bridging structure. Alternatively, the procedures of recombinant DNA technology may be used to produce an immunoglobulin molecule in which the Fc fragment or C~3 domain of a complete immunoglobulin molecule has been replaced by, or has attached thereto by peptide linkage, a functional non-immunoglobulin protein, such as an enzyme or toxin molecule.

Any appropriate acceptor variable region framework sequences may be used having regard to class/type of the donor antibody from which the antigen binding regions are derived. Preferably, the type of acceptor framework used is of the same/similar class/type as the donor antibody.
Conveniently, the framewor~ may be chosen to m~;m; se/
optimise homology with the donor antibody sequence particularly at positions close or adjacent to the CDRs.
However, a high level of homology between donor and acceptor sequences is not important for application of the present invention. The present invention identifies a hierarchy of framework residue positions at which donor residues may be lmportant or desirable for obt~;n;ng a ~DR-grafted antibody product having satisfactory binding properties. The CDR-grafted products usually have binding affinities of at lea~t 105 M-1, preferably at lea~t about 108 M~1, or especially in the range 108-1012 ~-l In principle, the present invention i8 applicable to any combination of donor and acceptor antibodies irre~pect~ve o~ the level of homology between their sequences. A protocol for applying the invention to any particular donor-acceptor antibody pair is given hereinafter. Examples of human frameworks which may be 12 ~ ~1292~9 used are KOL, NEWM, REI, EU, LAY and POM (refs. 4 and 5) and the like; for instance KOL and NEWM for the heavy chain and REI for the light chain and EU, LAY and POM for both the heavy chain and the light chain.

Also the constant region domains of the products of the invention may be selected having regard to the proposed function of the antibody in particular the effector functions which may be required. For example, the constant region domains may be human IgA, IgE, IgG or IgM
domains. In particular, IgG human constant region domains may be used, especially of the IgGl and IgG3 isotypes, when the humanised antibody molecule is intended for therapeutic uses, and anti~ody effector functions are required. Alternatively, IgG2 and IgG4 isotypes may be used when the humanised antibody molecul~ is intended for therapeutic purposes and anti~ody effector functions are not required, e.g. for simple blocking of lymphokine activity.

However, the remainder of the antibody molecules need not comprise only protein sequences from immunoglobulins.
For instance, a gene may be constructed in which a DNA
sequence encoding part of a human immunoglobulin chain is fused to a DNA sequence encoding the amino acid sequence of a functional polypeptide such as an effector or Z5 reporter molecule.

Preferably the CDR-grafted antibody heavy and light chain and antibody molecule products are produced by recombinant DNA technology.

Thus in further aspects the invention also includes DNA
sequences coding for the CDR-grafted heavy and light chains, cloning and expression vectors cont~;n;ng the DNA
sequences, host cell~ transformed with the DNA sequences ~ ~ 13 2~292~

and processes for producing the CDR-grafted chains and antibody molecules comprising expressing the DNA sequences in the transformed host cells.

The general methods by which the vectors may be constructed, transfection methods and culture methods are well ~nown per se and form no part of the invention. Such methods are shown, for instance, in references 10 and 11.

The DNA sequences which encode the donor amino acid sequence may be obtained by methods well known in the art. For example the donor coding sequences may be obtained by genomic cloning, or cDNA cloning from suitable hybridoma cell lines. Positive clones may be screened using appropriate probes for the heavy and light chain genes in question. Also PCR cloning may be used.

DNA coding for acceptor, e.g. human acceptor, sequences may be obtained in any appropriate way. For example DNA
sequences coding for preferred human acceptor fra~ewor~s such as KOL, REI, EU and NEWM, are widely available to wor~ers in the art.

The standard techniques o~ molecular biology may be used to prepare DNA sequences coding for the CDR-grafted products. Desired DNA sequences may be synthesised completely or in part using oligonucleotlde synthesis techniques. Site-directed mutagenesis and polymerase chain reaction (PCR) techniques may be used as appropriate. For example oligonucleotide directed synthesis as described by Jones et al (ref. 20) may be used. Also oligonucleotide directed mutagenesis of a pre-exising variable region as, for example, described by Verhoeyen et al (ref. 5) or Riechmann et al (ref. 6) may be used. Also enzymatic filling in of gapped ~ - 14 - ~29~

oligonucleotides using T4 DNA polymerase as, for example, described by Queen et al (ref. 9) may be used.

Any suitable host cell/vector system may be used for expression of the DNA sequences coding for the CDR-grafted heavy and light chains. Bacterial e.g. E. coli, and other microbial systems may be used, in particular for expression of antibody fragment~ such as FAb and (Fab')2 fragments, and especially FV fragments and ~ingle chain antibody fragments e.g. single chain FVs. Eucaryotic e.g. m~ lian host cell expression systems may be used for production of larger CDR-grafted antibody products, including complete antibody molecules. Suitable m~m~ n host cells include C~O cells and myeloma or hybridoma cell line~.

Thus, in a further aspect the present invention provides a process for producing a CDR-grafted antibody product comprising:

(a) producing in an expression vector an operon having a DNA sequence which encodes an antibody heavy chain according to the first aspect of the invention;
and/or (b) producing in an expression vector an operon having a DNA sequence which encodes a complementary antibody light chain according to the second or third aspect z5 of the invention;

(c) transfecting a host cell with the or each vector o~ part (a) and/or part (b); and (d) culturing the trans~ected cell line to produce the CDR-gra~ted antibody product A
,.~

- 15 - 212921~
.--The CDR-grafted product may comprise only heavy or light chain derived polypeptide, in which case only a heavy chain or light chain polypeptide coding sequence is used to transfect the host cells.
For production of products comprising both heavy and light chains, the cell line may be transfected with two vectors, the first vector may contain an operon encoding a light chain-derived polypeptide and the second vector cont~; n; ng an operon encoding a heavy chain-derived polypeptide.
Preferably, the vectors are identical, except in so far as the coding sequences and selectable marker~ are concerned, so as to ensure as far as possible that each polypeptide chain is equally expressed. Alternatively, a single vector may be used, the vector including the sequences encoding both light chain- and heavy chain-derived polypeptides.

The DNA in the coding sequence~ for the light and heavy chains may comprise cDNA or genomic DNA or both.
However, it is preferred that the DNA sequence encoding the heavy or light chain comprises at least partially, genomic DNA, preferably a fusion of cDNA and genomic DNA.

The presen~ invention i5 applicab~e to antibodies of any appropriate specificity. Advantageously, however, the invention may be applied to the humanisation of non-human antibodies which are used for in vivo therapy or diagnosis. Thus the antibodies may be site-specific antibodies such as tumour-specific or cell surface-speciflc antibodies, suitable for use in in vivo therapy or diagnosis, e.g. tumour imaging. Examples of cell surface-specific antibodies are anti-T cell antibodies, such as anti-CD3, and CD4 and adhesion molec~les, suc~ as CR3, ICAM and E~AM. The antibodies may have specificity for interleukins (including lymphokines, growth factors and stimulating factors), hormone~ and other biologically active compounds, and receptors for any of these. For ~.- ' 212~2~ ~

example, the antibodies may have specificity for any of the following: InterferonsC~,~, ~ or~, IL1, IL2, I~3, or IL4, etc., TNF, GCSF, GMCSF, EPO, hGH, or insulin, etc.

The the present invention also includes therapeutic and diagnostic compositions comprising the CDR-grafted products of the invention and uses of such compositions in therapy and diagnosis.

Accordingly in a further aspect the invention provides a therapeutic or diagnostic composition comprising a CDR-grafted antibody heavy or light chain or molecule according to previous aspects of the invention in combination with a pharmaceutically acceptable carrier, diluent or excipient.

Accordingly also the invention provides a method of therapy or diagnosis comprising administering an effective amount of a CDR-grafted antibody heavy or light chain or molecule according to previous aspects of the invention to a human or ~ n; m~ 1 subject.

A preferred protocol for obtaining CDR-grafted antibody heavy and light chains in accordance with the present invention is set out below together with the rationale by which we have derived this protocol. This protocol and rationale are given without prejudice to the generality of the invention as hereinbefore described and defined.

Protocol It is first of all necessary to sequence the DNA coding for the heavy and light chain variable regions of the donor antibody, to determine their amino acid sequences.
It is also necessary to choose appropriate acceptor heavy and light chain variable regions, of known amino acid sequence. The CDR-grafted chain is then designed 212~21~

starting from the basis of the acceptor sequence. It will be appreciated that in some cases the donor and acceptor ~m;no acid residues may be identical at a particular position and thus no change of acceptor framework residue is required.

1. As a first step donor residues are su~stituted for acceptor residues in the CDRs. For this purpose the CDRs are preferably defined as follows:

~eavy chain - CDRl: residues 26-35 1~ - CDR2: residues 50-65 - CDR3: residues 95-102 ~ight chain - CDRl: residues 24-34 - CDR2: residues 50-56 - CDR3: residues 8~-97 The positions at which donor residues are to be substituted for acceptor in the framework are then chosen as follows, first of all with respect to the heavy chain and subsequently with respect to the light chain.

2. ~eavy Chain 2.1 Choose donor residues at all of positions 23, 24, 49, 71, 73 and 78 of the heavy chain or all of positions 23, 2 4 and 49 (71, 73 and 78 are always either all donor or all acceptor).

2. 2 Check that the following have the same ~;no acid in donor and acceptor sequences, and if not preferably choose the donor: 2, 4, 6, 25, 36, 37, 39, 41, 48, 93, g4, 103, 104, 106 and 107.

.

- 18 _ 212~9 .

2.3 To further optimise affinity consider choosing donor residues at one, some or any of:

i. 1, 3 ii. 72, 76 iii. If 48 is different between donor and acceptor se~uences, consider 69 iv. If at 48 the donor residue is chosen, conslder 38 and 46 v. If at 69 the donor residue is chosen, consider 80 and then 20 vi. 67 vii. If at 67 the donor residue is chosen, consider 82 and then 18 viii . 9 1 ix. 88 x. 9, 11, 41, 87, 108, 110, 112 3. Liqht Chain 3.1 Choose donor at 46, 48, 58 and 71 3.2 Check that the following have the same amino acid in donor and acceptor sequences, if not preferably choose donor:

2, 4, 6, 35, 38, 44, 47, 49, 62, 64-69 inclusive, 85, 87, 98, 99, 101 and 102 3.3 To further optimise affinity consider choosing donor residues at one, some or any of:

i. 1, 3 ii. 63 ''' ' -l9-2l2~2~

iii. 60, if 60 and 54 are able to form potential saltbridge iv. 70, if 70 and 24 are able to form potential saltbridge v. 73, and 21 if 47 is different between donor and acceptor vi. 37, and 45 if 47 is different between donor and acceptor vii. 10, 12, 40, 80, 103, 105 Rationale In order to transfer the binding site of an antibody into a different acceptor framework, a number of factor~ need to be considered.

1. The extent of the CDRs The CDRs (Complementary Determin;ng Regions) were defined by Wu and ~abat (refs. 4 and 5) on the basis of an analysis of the variability of different regions of antibody variable regions.
Three regions per domain were recognised. In the light chain the sequences are 24-34, 50-56, 89-97 (numbering according to ~abat (ref. 4), Eu Index) inclusive and in the heavy chain the sequences are 31-35, 50-65 and ~5-102 inclusive.

When antibody structures became available it became apparent that these CDR regions corresponded in the main to loop regions which extended from the ~ barrel framewor~ of the light and heavy variable domains. For Hl there was a discrepancy in that the loop was from 26 to 32 inclusive and for H2 the loop was 52 to 56 and for L2 from 50 to 53. ~owever, with the exception of ~1 the CDR regions encompassed the loop regions and extended into the ~ strand 2~ 29~ ~

framewor~s. In Hl residue 26 tends to be a serine and 27 a phenylalanine or tyro~ine, residue 29 is a phenylalanine in most case~.
Residues 28 and 30 which are surface residue~
exposed to solvent might be involved in antigen-binding. A prudent definition of the ~1 CDR therefore would include residues 26-35 to include both the loop region and the hypervariable residues 33-35.

It is of interest to note the example of Riechm~nn et al (ref. 3), who used the residue 31-35 choice for CDR-Hl. In order to produce efficient antigen binding, residue 27 also needed to be recruited from the donor (rat) antibody.

2. Non-CDR residues which contribute to antiqen bindinq By examination of available X-ray structures we have identified a number of residues which may have an effect on net antigen binding and which can be demonstrated by experiment. These residues can be sub-divided into a number of groups.
2.1 Surface residues near CDR tall numbering as in Kabat et al (ref. 7)].
2.1.1. Heavy Chain - Key residues are 23, 71 and 73.
Other residues which may contribute to a lesser extent are 1, 3 and 76. Finally 25 is usually conserved but the murine residue should be used if there is a difference.
2.1.2 Light Chain - Many residues close to the CDRs, e.g. 63, 65, 67 and 69 are conserved. If conserved none of the surface residues in the light chain are likely to have a ma~or effect.
However, if the murine residue at these positions ~ 2~2~2~

is unusual, then it would be of benefit to analyse the likely contribution more closely.
Other residues which may also contri~ute to ~inding are 1 and 3, and also 60 and 70 if the residues at these positions and at 54 and 24 respectively are potentially able to form a salt bridge i.e. 60 + 54; 70 + 24.
2.2 Packing residues near the CDRs.
2.2.1. Heavy Chain - Rey residues are 24, 49 and 78.
Other key residues would be 36 if not a tryptophan, 94 if not an arginine, 104 and 106 if not glycines and 107 if not a threonine.
Residues which may make a further contribution to stable packing of the heavy chain and hence improved affinity are 2, 4, 6, 38, 46, 67 and 69. 67 packs against the CDR residue 63 and this pair could be either both mouse or both human. Finally, residues which contribute to packing in this region but from a longer range are 18, 20, 80, 82 and 86. 82 packs against 67 and in turn 18 packs against 82. 80 packs against 69 and in turn 20 packs against 80. 86 forms an ~ bond network with 38 and 46. Many of the mouse-human differences appear minor e.g.
~eu-Ile, but could have an minor impact on correct packing which could translate into altered positioning of the CDRs.
2.2.2. Light Chain - Key residues are 48, 58 and 71.
Other key residues would be 6 if not glutamine, 35 if not tryptophan, 62 if not phenylalanine or tryosine, 64, 66, 68, 99 and 101 if not glycines and 102 if not a threonine. Residues which make a further contribution are 2, 4, 37, 45 and 47.
Finally residues 73 and 21 and 19 may make long distance packing contributions of a minor nature.

~ 22 21~2~9 _ -2.3. Residues at the variable domain interface ~etween heavy and light chains - In both the light and heavy chains most of the non-CDR interface residues are conserved. If a conserved residue is replaced by a residue of different character, e.g. size or charge, it should be considered for retention as the murine residue.
2.3.1. Heavy Chain - Residues which need to be considered are 37 if the residue is not a valine but is of larger side chain volume or has a charge or polarity. Other residues are 39 if not a glutamine, 45 if not a leucine, 47 if not a tryptophan, 91 if not a phenylalanine or tyrosine, 93 if not an alanine and 103 if not a tryptophan. Residue 89 is also at the interface but is not in a position where the side chain could be of great impact.
2.3.2. Light Chain - Residues which need to be considered are 36, if not a tyrosine, 38 if not a glutamine, 44 if not a proline, 46, 49 if not a tyrosine, residue 85, residue 87 if not a tyrosine and 98 if not a phenylalanine.
2.4. Variable-Constant region interface - The elbow angle between variable and constant regions may be affected by alterations in packing of key residues in the variable region against the constant region which may affect the position of VL and VH with respect to one another.
Therefore it is worth noting the residues likely to be in contact with the constant region. In the heavy chain the surface residues potentially in contact with the variable region are conserved between mouse and human antibodies therefore the variable region contact residues may influence the V-C interaction. In the light chain the amino acids found at a number of the constant ' ~ ~ 23 2~ 2~2~
_ -region contact points vary, and the V ~ C regions are not in such close proximity as the heavy chain. Therefore the influences of the light chain V-C interface may be minor.
2.4.1. Heavy Chain - Contact residues are 7, 11, 41, 87, 108, 110, 112.
2.4.2. Light Chain - In the light chain potentially contacting residues are 10, 12, 40, 80, 83, 103 and 105.

The above analysis coupled with our considerable practical experimental experience in the CDR-grafting of a number of different antibodies have lead us to the protocol given a~ove.

The present invention is now described, by way of example only, with reference to the accompanying Figures 1 - 13.

Brief Description of the Figures Figure 1 shows DNA and amino acid sequences of the OKT3 light chain;
Figure 2 shows DNA and amino acid sequences of the OKT3 heavy chain;
Figure 3 shows the alignment of the OKT3 light variable region amino acid sequence with that of the light variable region of the human anti~ody REI;
Figure 4 shows the alignment of the OKT3 heavy variable region amino acid sequence with that of the heavy variable region of the human antibody KOL;
F-gure 5 sh~wc the hea~.~ v~riable region am-ino aGid sequences of OKT3, KOL and various corresponding CDR grafts;
Figure 6 shows the light variable region amino acid sequences of OKT3, REI and various corresponding CDR grafts;

~- 24 - 2129i~-~ 3 Figure 7 shows a graph of binding assay results for various grafted OKT3 antibodies' Figure 8 shows a graph of blocking assay results for various grafted OKT3 antibodies;
Figure 9 shows a similar graph of blocking assay results;
Figure 10 shows similar graphs for both binding assay and blocking assay results;
Figure 11 shows further similar graphs for both binding assay and blocking assay results;
10Figure 12 shows a graph of competition assay results for a m;n;m~lly grafted OKT3 antibody compared with the OKT3 murine reference standard, and Figure 13 shows a similar graph of competition assay results comparing a fully grafted OKT3 antibody 15with the murine reference standard.

~ - 25 - 2 ~ 2 ~ 2 ~ 9 DE~TTRn ~ESCRIPTION OF ~MBODI~ENTS OF TEE lNv~Nl~lON

EXAMPL~ 1 MATERIAL AND METHODS

1. INCOMING C~LLS
~ybridoma cells producing antibody OKT3 were provided by Ortho (seedlot 4882.1) and were grown up in antibiotic free Dulbecco's Modified Eagles Medium (DMEM) supplemented with glutamine and 5~ foetal calf serum, and divided to provide both an overgrown supernatant for evaluation and cells for extraction of RNA. The overgrown supernatant was shown to contain 250 ug/mL murine IgG2a/kappa antibody. The supernatant was negative for murine lambda light chain and IgGl, IgG2b, IgG3, IgA and IgM heavy chain. 20mL of supernatant was assayed to confirm that the antibody present was ORT3.

2. MOLECULAR BIOLOGY PROCEDURRS
Basic molecular biology procedures were as described in Maniatis et al (ref. 9) with, in some cases, minor modifications. DNA sequencing was performed as described in Sanger et al (ref. ll) and the Amersham International Plc sequencing handbook. Site directed mutagenesis was as described in Kramer et al ~ref. 12~ and the Anglian Biotechnology Ltd.
handbook. COS cell expression and metabolic labelling studies were as described in Whittle et al (ref. 13) ~ - 26 - 2~29~
.

3. RESEARCH ASSAYS
3.1. ASSEMBLY ASSAYS
Assembly assays were performed on supernatants from transfected COS cells to determine the amount of intact IgG present.
3.1.1. COS CELLS TRANSF~CTED WITH MOUSE OKT3 GENES
The assembly assay for intact mouse IgG in COS
cell supernatants was an ELISA with the following format:
96 well microtitre plates were coated with F(ab' )2 goat anti-mouse IgG Fc. The plates were washed in water and samples added for 1 hour at room temperature. The plates were wa~hed and F(ab') 2 goat anti-mouse IgG F(ab' )2 (HRPO con~ugated) was then added. Substrate was added to reveal the reaction. UPC10, a mouse IgG2a myeloma, was used as a standard.
3.1.2. COS AND CHO CELLS TRANSFECTED WITH CHIMERIC OR
CDR-GRAFT~D ORT3 GENES
The assembly assay for chimeric or CDR-grafted antibody in COS cell supernatants was an ELISA
with the following ~ormat:
96 well microtitre plates were coated with F(ab' )2 goat anti-human IgG Fc. The plates were washed and samples added and incubated for 1 hour at room temperature. The plates were washed and monoclonal mouse anti-human ~appa chain was added for 1 hour at room temperature.
The plates were washed and F(ab')2 goat anti-mouse IgG Fc (HRPO conjugated) was added. Enzyme substrate was added to reveal the reaction.
Chimeric B72.3 (IgG4) (ref. 13) was used as a standard. The use of a monoclonal anti-kappa chain in this assay allows grafted antibodies to be read from the chimeric standard.

~ 27 - 2129~

3.2. ASSAY FOR ANTIGEN BINDING ACTIVITY
Material from COS cell supernatants was assayed for OKT3 antigen binding activity onto CD3 positive cells in a direct assay. The procedure was as follows:
~UT 78 cells (human T cell line, CD3 positive) were maintained in culture. ~onolayers of HUT 78 cells were prepared onto 96 well ELISA plates using poly-L-lysine and glutaraldehyde. Samples were added to the monolayers for 1 hour at room temperature.
The plates were washed gently using PBS. F(ab' )2 goat anti-human IgG Fc (~RPO conjugated) or F(ab' )2 goat anti-mouse IgG Fc (~RPO con~ugated) was added as appropriate for humanised or mouse samples.
Substrate was added to reveal the reaction.
The negative control for the cell-based assay was chimeric B72.3. The positive control was mouse Orthomune OKT3 or chimeric ORT3, when available.
This cell-based assay was difficult to perform, and an alternative assay was developed for CDR-grafted OKT3 which was more sensitive and easier to carry out.
In this syste~ CDR-grafted ORT3 produced by COS
cells was tested for its ability to bind to the CD3-positive ~PB-ALL (human peripheral blood acute lymphocytic leukemia) cell line. It was also tested for its ability to block the binding of murine OKT3 to these cells. Binding was measured by the following procedure: ~PB-ALL cells were harvested from tissue culture. Cells were incubated at 4~C for 1 hour with various dilutions of test antibody, positive control antibody, or negative control antibody. The cells were washed;~nce and incubated at 4~C for 1 hour with an FITC-labelled goat anti-human IgG (Fc-212~2~9 . ~ - 28 -specific, mouse absorbed). The cells were washed twice and analysed by cytofluorography. Chimeric OKT3 was used as a positive control for direct binding. Cells incubated with mock- transfected COS cell supernatant, followed by the FITC-labelled goat anti-human IgG, provided the negative control.
To test the ability of CDR-grafted OKT3 to block murine OKT3 binding, the ~PB-ALL cells were incubated at 4~C for l hour with various dilutions of test antibody or control antibody. A fixed saturating amount of FITC OKT3 was added. The samples were incubated for l hour at 4~C, washed twice and analysed by cytofluorography.
FITC-labelled ORT3 was used as a positive control to determine m~;mum binding. Unlabelled murine OKT3 served as a reference standard for blocking. Negative controls were unstained cells with or without mock-transfected cell supernatant.
The ability of the CDR-grafted ORT3 light chain to bind CD3-positive cells and block the binding of murine OKT3 was initially tested in combination with the chimeric OKT3 heavy chain. The chimeric OKT3 heavy chain is composed of the murine OKT3 variable region and the human IgG4 constant region. The chimeric heavy chain gene is expressed in the same expression vector used for the CDR-grafted genes. The CDR-grafted light chain expression vector and the chimeric heavy chain expression vector were co-transfected into COS cells. The fully chimeric OKT3 antibody (chimeric light chain and chimeric heavy chain) was ~ound to be fully capable of binding to CD3 po~itive cells and blocking the binding of murine OKT3 to these cells.
3.3 DETERMINATION OF RELATIVE BINDING AFFINITY
The relative binding affinities of CDR-grafted anti-CD3 monoclonal antibodies were determined by competition binding (ref. 6) using the ~PB-AL~
human T cell line as a source of CD3 antigen, and fluorescein-conjugated murine ORT3 (Fl-O~T3) of known binding affinity as a tracer antibody. The binding affinity of Fl-OKT3 tracer antibody was determined by a direct binding assay in which increasing amounts of Fl-OKT3 were incubated with ~PB-ALL (5xl05) in PBS with 5% foetal calf serum for 60 min. at 4~C. Cells were washed, and the fluorescence intensity was determined on a FACScan flow cytometer calibrated with quantitative microbead standards (Flow Cytometry St~n~rds, Research Triangle Park, NC). Fluorescence intensity per antibody molecule (F/P ratio) was determined hy using microbeads which have a predetermined number of mouse IgG antibody binding sites (Simply Cellular bead~, Flow Cytometry ' Standards). F/P equals the fluorescencs intensity of beads saturated with Fl-OKT3 divided by the n~ ~r of binding sites per bead. The amount of bound and free Fl-OKT3 was calculated from the mean fluorescence intensity per cell, and the ratio of bound/free was plotted against the number of moles of antibody bound. A linear fit wa~
used to determine the affinity of binding (absolute value of the slope).
For competitive binding, increasing amounts of competitor antibody were added to a sub-saturating dose of Fl-OKT3 and incubated with 5x105 ~PB-AL~ in 200 ml of PBS with 5% foetal calf serum, for 60 min at 4~C. The fluorescence intensities of the cells were measured on a FACScan flow cytometer calibrated with quantitative microbead standards.
The concentrations of bound and free Fl-OKT3 were calculated. The affinities of competing anti-* Trade-mark _ 30 _ 2~29~1~
.~
bodies were calculated from the equation [X]-[ORT3] = (l/Kx) - (l/Ka), where Ra i8 the affinity of murine OKT3, Kx is the affinity of competitor X, [ ] is the concentration of competitor antibody at which bound/free binding is R/2, and R is the m~;m~l bound/free binding.

4. cDNA LIBRARY CONSTRUCTION
4.1. mRNA PREPARATION AND cDNA SYNTHESIS
OKT3 producing cells were grown as described above and 1.2 x 109 cells harvested and mRNA extracted using the guanidinium/LiCl extraction procedure.
cDNA was prepared by priming from Oligo-dT to generate full length cDNA. The cDNA was methylated and EcoRl linkers added for cloning.
4.2. LIBRARY CONSTRUCTION
The cDNA library was ligated to pSP65 vector DNA
which had been EcoRl cut and the 5' phosphate groups removed by calf intestinal phosphatase (EcoRl/CIP3. The ligation was used to transform high transformation efficiency Escherichia coli (E.coli) HB101. A cDNA li~rary was prepared.
3600 colonies were screened for the light chain and 10000 colonies were screened for the heavy chain.

5. SCREENING
E.coli colonies positive for either heavy or light chain probes were identified by oligonucleotide screening using the oligonucleotides:
5' TCCAGATGTTAACTGCTCAC for the light chain, which is complementary to a sequence in the mouse kappa constant region, and 5' CAG~G&CCAGTGGATGGATAGAC
for the heavy chain which is complementary to a se~uence in the mouse IgG2a constant CHl domain region. 12 light chain and 9 heavy chain clones 212~213 were identified and taken for second round screening. Positive clones from the second round of screening were grown up and DNA prepared. The sizes of the gene inserts were estimated by gel electrophoresis and inserts of a size capable of contA; n i ng a full length cD,NA were subcloned into M13 for ~A sequencing.

6. DNA SEQUENCING
Clones representing four size classes for both heavy and light chains were obtained in M13. DNA
sequence for the 5' untranslated regions, signal sequences, variable regions and 3' untranslated regions of full length cDNAs ~Figures l(a) and 2(a)] were obtained and the corresponding amino acid sequences predicted [(Figures l(b) and 2(b)]- In Figure l(a) the untranslated DNA
regions are shown in uppercase, and in both Figures 1 and 2 the signal sequences are underlined.

20 7. CONSTRUCTION OF cDNA EXPR SSION VECTORS
Celltech expression vectors are based on the plasmid pEE6hCMV (ref. 14). A polylinker for the insertion of genes to be expressed has been introduced after the major immediate early 2 5 promoter/enhancer of the human Cytomegalovirus (hCMV). Marker genes for selection of the plasmid in transfected eukaryotic cells can be inserted as BamH1 cassettes in the unique Bam~1 site of pEE6 hCMV; for instance, the neo marker to provide pEE6 hCMV neo. It is usual practice to insert the neo and gpt markers prior to insertion of the gene of interest, whereas the GS
marker is inserted last because of the presence of internal EcoRl sites in the ca~sette.

~ 32 _ 2 1 ~ ~ 2 ~ ~
,, The selectable markers are expressed from the SV40 late promoter which also provides an oriqin of replication so that the vectors can be used for expression in the COS cell transient expression system.
The mouse sequences were excised from the M13 based vectors described above as EcoRl fragments and cloned into either pEE6-hCMV-neo for the heavy chain and into EE6-hCMV-gpt for the light chain to yield vectors pJA136 and pJA135 respectively.

8. EXPRESSION OF cDNAS IN COS CELLS
Plasmids pJA135 and pJA136 were co-transfected into COS cells and supernatant from the transient expression experiment was shown to contain assembled antibody which bound to T-cell enriched lymphocytes. Metabolic labelling experiments using 35S methionine showed expression and assembly of heavy and light chains.

9. CONSTRUCTION OF CHIMERIC GENES
Construction of chimeric genes followed a previously described strategy ~Whittle et al (ref.
13)]. A restriction site near the 3' end o~ the variable domain sequence is identified and used to attach an oligonucleotide adapter coding for the remainder of the mouse variable region and a suitable restriction site for attachment to the constant region of choice.
9.1. LIG~T CXAIN GENE CONSTRUCTION
The mouse light chain cDNA sequence contains an Aval site near the 3' end of the variable region [Fig- l(a)]. The majority of the se~uence of the variable region was isolated as a 396 bp.
EcoR1-Aval fragment. An oligonucleotide adapter was designed to replace the remainder of the 3' sr- ~ 292~ ~

region of the variable region from the Aval site and to include the 5' residues of the human constant region up to and including a unique Narl site which had been previously engineered into the constant region.
A Hindlll site was introduced to act as a marker for insertion of the linker.
The linker was ligated to the VL fragment and the 413 bp EcoR1-Narl adapted fragment was purified from the ligation mixture.
The constant region was isolated as an Narl-BamHl fragment from an M13 clone NW361 and was ligated with the variable region DNA into an EcoR1/BamH1/ClP pSP65 treated vector in a three way reaction to yield plasmid JA143. Clones were isolated after transformation into E.coli and the linker and junction sequences were confirmed by the presence of the Hindlll site and by DNA
sequencing.
9.2 LIGHT CHAIN GENE CONSTRUCTION - VERSION 2 The construction of the flrst chimeric light chain gene produces a fusion of mouse and human amino acid sequences at the variable-constant region junction. In the case of the OKT3 light chain the amino acids at the chimera junction are:
........ Leu-Glu-Ile-Asn-Arq/ -/Thr-Val-Ala -Ala VARIABLE CONSTANT
This arrangement of sequence introduces a potential site for Asparagine (Asn) lin~ed (N-lin~ed) glycosylation at the V-C junction.
Therefore, a second version of the chimeric light chain oligonucleotide adapter was designed in which the threonine (Thr), the first ~;no acid of the human constant region, was replaced with the equivalent amino acid from the mouse constant region, Alanine (Ala).

' _ 34 _ 21292~9 . --An internal ~indlll site was not included in this adapter, to differentiate the two chimeric light chain genes.
The variable region fragment was isolated a~ a 376 bp EcoRl-Aval fragment. The oligonucleotide linker was ligated to Narl cut pNW361 and then the adapted 396bp constant region was isolated after recutting the modified pN~361 with EcoRl. The variable region fragment and the modified constant region fragment were ligated directly into EcoRl/ClP treated pEE6hCMVneo to yield p~A137.
Initially all clones e~m; ned had the insert in the incorrect orientation. Therefore, the insert was re-isolated and recloned to turn the insert round and yield plasmid pJA141. Several clones with the insert in the correct orientation were obtained and the adapter sequence of one was confirmed by DNA sequencing 9.3. ~EAVY CHAIN GENE CONSTRUCTION
g.3.1. CHOICE OF HEAVY CHAIN GENE ISOTYPB
The constant region isotype chosen for the heavy chain was human IgG4.
9.3.2. GENE CONSTRUCTION
The heavy chain cDNA sequence showed a Banl site near the 3' end of the variable region [Fig. 2(a)].
The majority of the sequence of the variable region was isolated as a 426bp. EcoRl/ClP/Banl fragment. An oligonucleotide adapter was designated to replace the remainder of the 3' region of the variable region from the Banl site up to and including a unique ~indIII site which had been previously engineered into the first two ~;no acids of the constant region.
The linker was ligated to the V~ fragment and the EcoRl-~indlll adapted fragment wa~ purified from the ligation mixture.

2~ 9 . ~

The variable region was ligated to the constant region by cutting pJA91 with EcoRl and Hindlll removing the intron fragment and replacing it with the VH to yield pJA142. Clones were isolated after transformation into E.coli JM101 and the lin~er and junction sequences were confirmed by DNA sequencing. (N.B. The Hindlll site is lost on cloning).

10. /CONSTRUCTION OF CHIMERIC EXPRESSION VECTORS
10.1. neo AND gpt VECTORS
The chimeric light chain ~version 1) was removed from pJA143 as an EcoRl fragment and cloned into EcoRl/ClP treated pE~6hC~Vneo expression vector to yield pJA145. Clones with the insert in the correct orientation were identified by restriction mapping.
The chimeric light chain (version 2) was con~tructed as described above.
The chimeric heavy chain gene was isolated from pJA142 as a 2.5Kbp EcoRl/BamHl fragment and cloned into the EcoRl/Bcll/ClP treated vector fragment of a derivative of pEE6hC~Vgpt to yield plasmid pJA144.
10.2. GS SEPARAT~ VECTORS
GS versions of pJA141 and pJA144 were constructed by replacing the neo and gpt cassettes by a Bam~l/Sall/ClP treatment of the plasmids, isolation of the vector fragment and ligation to a GS-containing fragment from the plasmid pRO49 to yield the light chain vector pJA179 and the heavy chain vector pJA180.
10.3. GS SINGLE VECTOR CONSTRUCTION
Single vector constructions containing the cL
(chimeric light), cH (chimeric heavy) and GS genes on one plasmid in the order c~-cH-GS, or cH-cL-GS

~ ~29~ ~
~ - 36 -.

and with transcription of the genes being head to tail e.g. cL>cH>GS were constructed. These plasmids were made by treating pJA179 or pJA180 with B~l/ClP and ligating in a Bglll/~indlll hCMV promoter cassette along with either the Hindlll/BamHl fragment from pJA141 into pJA180 to give the cH-cL-GS plasmid pJA182 or the Hindlll/BamHl fragment from pJA144 into pJA179 to give the cL-cK-GS plasmid pJA181.

11. EXPRESSION OF CHIMERIC GENES
11.1. EXPRESSION IN COS CELLS
The chimeric antibody plasmid pJA145 (cL) and pJA144 (cH) were co-transfected into COS cells and supernatant from the transient expression experiment was shown to contain assembled antibody which bound to the HUT 78 human T-cell line.
Metabolic labelling experiments using 35S
methionine showed expression and assembly of heavy and light chains. ~owever the light chain mobility seen on reduced gels suggested that the potential glycosylation site was being glycosylated. Expression in COS cells in the presence of tunicamycin showed a reduction in size of the light chain to that shown for control chimeric antibodies and the OKT3 mouse light chain. Therefore JA141 was constructed and expressed. In this case the light chain did not show an aberrant mobility or a size shift in the presence or absence of tunicamycin. This second version of the chimeric light chain, when expressed in association with chimeric heavy (cH) chain, produced antibody which showed good b; n~; ng to ~UT 78 cells. In both cases antigen binding was equivalent to that of the mouse antibody.

212~2~ ~

11.2 EXPRESSION IN CHINESE HAMSTER OVARY (C~O) CELLS
Stable cell line5 have been prepared from plasmids PJA141/pJA144 and from pJA179/pJA180, pJA181 and pJA182 by transfection into CHO cells.

5 12. CDR-GRAFTING
The approach taken was to try to introduce sufficient mouse residues into a human variable region framework to generate antigen binding activity comparable to the mouse and chimeric antibodies.
12.1. VARIABLE REGION ANALYSIS
From an examination of a small database of structures of antibodies and antigen-antibody complexes it is clear that only a small number of antibody residues make direct contact with antigen. Other residues may contribute to antigen binding by positioning the contact residues in favourable configurations and also by inducing a stable packing of the individual variable domains and stable int~raction of the light and heavy chain variable domains.
The residues chosen for transfer can be identified in a number of ways:
(a) By examination of antibody X-ray crystal structures the antigen binding surface can be predominantly located on a series of loops, three per domain, which extend from the B-barrel framework.
(b) By analysi~ of antibody variable domain sequences regions of hypervariability [termed the Complementarity Det~rmining Regions (CDRs) by Wu and Kabat (ref. 5)]
can be identified. In the most but not all cases these CDRs correspond to, but extend a short way beyond, the loop regions noted above.

38 - 212~2~
_ -(c) Residues not identified by (a) and (b) may contribute to antigen binding directly or indirectly by affecting antigen b;n~;ng site topology, or by inducing a stable packing of the individual variable dom~; n~
and stabilising the inter-variable domain interaction. These residues may be identi~ied either by superimposing the sequences for a given antibody on a known structure and looking at key residues for their contribution, or by sequence alignment analysis and noting "idiosyncratic" residues followed by examination of their structural location and likely effects.
12.1.1. LIGHT C~AIN
Figure 3 shows an alignment of sequences for the human framework region RE1 and the OKT3 light variable region. The structural loops (LOOP) and CDRs ~KABAT) believed to correspond to the antigen binding region are marked. Also marked are a number of other residues which may also contribute to antigen binding as described in 13.1(c).
Above the sequence in Figure 3 the residue type indicates the spatial location of each residue side chain, derived by examination of resolved structures from X-ray crystallography analysis.
The key to this residue type designation is as follows:
N - near to CDR (From X-ray Structures) P - Packing B - Buried Non-Packing S - Surface E - Exposed I - Interface * - Interface - Packing/Part Exposed ? - Non-CDR Residues which may require to be left as Mouse sequence.

_ 3g _ 2~2~2~
~ . ~

Residues underlined in Figure 3 are amino acids.
REl was chosen as the human framework because the light chain i8 a kappa chain and the kappa variable regions show higher homology with the mouse sequences than a lambda light variable region, e.g. KOL (see below). RE1 was chosen in preference to another kappa light chain because the X-ray structure of the light chain has been determined so that a structural ~m; nAtion of individual residues could be made.
12.1.2. HEAVY C~AIN
Similarly Figure 4 shows an alignment of sequences for the human framework region ROL and the OKT3 heavy variable region. The structural loops and CDRs believed to correspond to the antigen binding region are marked. Also marked are a number of other residues which may also contribute to antigen binding as described in 12.1~c). The residue type key and other indicators used in Figure 4 are the same as those used in Figure 3.
KOL was chosen as the heavy chain framework because the X-ray structure has been determined to a better resolution than, for example, NEWM and also the sequence alignment of OKT3 heavy variable region showed a slightly better homology to KOL
than to NEWM.
12.2. ~ESIGN OF VARIABLE GENES
The variable region domains were designed with mouse variable region optimal codon usage [Grantham and Perrin (ref. 15)] and used the B72.3 signal sequences [Whittle et al (ref. 13)]. The sequences were designed to be attached to the constant region in the same way as for the chimeric genes described above. Some constructs contained the ~Kozak consensus sequence" t~ozak (ref. 16)] directly linked to the 5' of the signal ' ~ - 40 - 2~2~2t~

sequence in the gene. This sequence motif is believed to have a beneficial role in translation initiation in eukaryotes.
12.3. GENE CONSTRUCTION
To build the variable regions, various strategies are available. The sequence may be assembled by using oligonucleotides in a manner similar to Jones et al (ref. 17) or by simultaneously replacing all of the CDRs or loop regions by oligonucleotide directed site specific mutagenesis in a manner similar to Verhoeyen et al (ref. 2).
Both strategies were used and a list of constructions is set out in Tables 1 and 2 and Figures 4 and 5. It was noted in several cases that the mutagenesis approach led to deletions and rearrangements in the gene being remodelled, while the success of the assembly approach was very sensitive to the quality of the oligonucleotides.

13. CONSTRUCTION OF EXPRESSION VECTORS
Genes were isolated from M13 or SP65 based intermediate vectors and cloned into pEE6hCMVneo for the light chains and pEE6hC~Vgpt for the heavy chains in a manner similar to that for the chimeric genes as described above.

~ 41 - 212921~
ABLE 1 CD~_GRAFTED GENE C~N~1~UCTS
DE MOUSE SEQUENCE METHOD OF KOZAK
CONTENT CONSTRUCTION SEQUENCE
+
_ _ _ _ LIGHT CHAIN ALL HUMAN FRAMEWORK REl 121 26-32, 50-56, 91-96 inclusive SDM and gene assembly + n.d.
121A 26-32, 50-56, 91-96 incluslve Partial gene assembly n.d. +
+1, 3, 46, 47 121B 26-32, 50-56, 91-96 inclusive Partial gene assembly n.d. +
+ 46, 47 10 221 24-24, 50-56, 91-96 inclusive Partial gene assembly + +
22LA 24-34, 50-56, 91-96 inclusive Partial gene assembly + +
+1, 3, 46, 47 221B 24-34, 50-56, 91-96 inclusive Partial gene assemb~y + +
+1, 3 15 221C 24-34, 50-56, 91-96 inclusive Partial gene assembly + +

HEAVY CHAIN ALL HUMAN FRAMEWORK KOL
121 26-32, 50-56, 95-lOOB inclusive Gene assembly n.d. +
131 26-32, 50-58, 95-lOOB inclusive Gene assembly n.d. +
141 26-32, 50-65, 95-lOOB inclusive Partial gene assembly + n.d.
20 321 26-35, 50-56, 95-lOOB inclusive Partial gene assembly + n.d.
331 26-35, 50-58, 95-lOOB inclusive Partial gene assembly +
Gene assembly +
341 2~-35, 50-65, 95-lOOB inclusive SDM +
Partial gene assembly +
25 34LA 26-35, 50-65, 95-lOOB inclusive Gene assembly n.d. +
+6, 23, 24, 48, 49, 71, 73, 76, 78, 88, 91 (+63 human) 341B 26-35, 50-65, 95-lOOB inclusive Gene assembly n.d. +
+ 48, 49, 71, 73, 76, 78, 88, 91 (+63 + human) KEY
n.d. not done SDM Site directed mutagenesis Gene assembly Variable region assembled entirely from oligonucleotides Partial gene Variable region assembled by combination of restriction assembly fragments either from other genes originally created by SDM
and gene assembly or by oligonucleotide assembly of part of the variable region and reconstruction with restriction fragments from other genes originally created by SDM and gene assembly ~12~2~ ~

14. EXPRESSION OF CDR-GRAFT~D GENES
14.1. PRODUCTION OF ANTIBODY CONSISTING OF GRAETED LIGHT
(gL) CHAINS WIT~ MOUSE HEAVY ~mH) OR CHIMERIC
HEAVY (cH) CHAINS
All gL chains, in association with mH or cH
produced reasonable amounts of antibody.
Insertion of the Kozak consensus sequence at a position 5' to the ATG (kgL constructs) however, led to a 2-5 fold improvement in net expression.
1 Over an extended series of experiments expression levels were raised from approximately 2QOng/ml to approximately 50Q ng/ml for kgL/cH or kgL/mH
combinations.
When direct binding to antigen on HUT 78 cells was 15' measured, a construct designed to include mouse sequence based on loop length (gL121) did not lead to active antibody in association with mH or cH.
A construct designed to include mouse sequence based on Kabat CDRs (gL221) demonstrated some weak binding in association with mH or cH. However, when framework residues 1, 3, 46, 47 were changed from the human to the murine OKT3 equivalents based on the arguments outlined in Section 12.1 antigen binding was demonstrated when both of the new constructs, which were termed 12lA and 22lA
were co-expressed with cH. When the ef~ects of thesa residues were examined in more detail, it appears that residues 1 and 3 are not major contri~uting residues as the product of the g~22lB
gene shows little detectable binding activity in association with cH. The light chain product of gL221C, in which mouse sequences are present at 46 and 47, shows good ~inding activity in association with cH.

- 2~29~1~
~ - 43 -.

14.2 PRODUCTION OF ANTIBODY CO~SISTING OF GRAFTED HEAVY
(gH) CHAINS WITH MOUSE LIGHT (mL) OR CHIMERIC
LIGHT (cL~ C~AINS
Expression of the gH genes proved to be more difficult to achieve than for gL. First, inclusion of the Rozak sequence appeared to have no marked effect on expression of g~ genes.
Expression appears to be slightly improved but not ' to the same degree as seen for the grafted light chain.
Also, it proved difficult to demonstrate production of expected quantities of material when the loop choice (amino acid 26-32) for CDR1 is used, e.g. gH121, 131, 141 and no conclusions can be drawn about these constructs.
Moreover, co-expression of the gH341 gene with cL
or mL has been variable and has tended to produce lower amounts of antibody than the cH/cL or mH/mL
combinations. The alterations to gH341 to produce g~34lA and gH34lB lead to improved levels o~ expression.
This may be due either to a general increase in the ~raction o~ mouse sequence in the variable region, or to the alteration at position 63 where the residue is returned to the human amino acid Valine (Val) from Phenylalanine (Phe) to avoid possi~le internal packing problems with the rest of the human framework. This arrangement also occurs in gH331 and g~321.
When gH321 or gH331 were expressed in association with cL, antibody was produced but antibody binding activity was not detected.
When the more conservative gH341 gene was used antigen binding could be detected in association with cL or mL, but the activity was only marginally above the background level.

. .

., 2~2~2~
~ - 44 -.

When further mouse residue5 were substituted based on the arguments in 12.1, antigen binding could be clearly demonstrated for the antibody produced when kgH34lA and kgH34lB were expressed in asso~iation with cL.
14.3 PRODUCTION OF FULLY CDR-GRAFTED ANTIBODY
The kgL221A gene was co-expressed with kgH341, kgH34lA or kg~34lB. For the combination kg~22lA~kgH341 very little material was produced in a normal COS cell expression.
For the combinations kgL22lA/kgH34lA or kgH22lA/kgH34lB amounts of antibody similar to gL/c~ was produced.
In several experiments no antigen binding activity could be detected with kgH221A/g~341 or kgH22lA/kg~341 com~binations, although expression levels were very low.
Antigen binding was detected when kgL221A/kg~341A
or kgH22lA/kgH34lB combinations were expressed.
In the case of the antibody produced from the kgL22lA/kgH34lA combination the antigen binding was very similar to that of the chimeric antibody.

An analysis of the above results is given below.

15. DISCUSSION OF CDR-GRAFTING RESULTS
In the design of the fully humanised antibody the aim was to transfer the m;n;mllm number of mouse amino acids that would confer antigen binding onto a human antibody framework.
15.1. LIGHT C~AIN
15.1.1. EXTENT OF THE CDRs For the light chain the regions defining the loops known from structural studies of other antibodies to contain the antigen contacting residues, and _ 45 _ 2~2~219 . ~

those hypervariable sequences defined by Kabat et al (refs. 4 and 5) as Complementarity Determ;n;ng Regions (CDRs) are equivalent for CDR2. For CDRl the hypervariable region extends from residue~
24-34 inclusive while the structural loop extends from 26-32 inclusive. In the case of OKT3 there is only one amino acid difference between the two options, at amino acid 24, where the mouse sequence is a serine and the human framework REl has glutamine. For CDR3 the loop extends from residues 91-96 inclusive while the Kabat hypervariability extends from residues 89-97 inclusive. For OKT3 amino acids 89, 90 and 97 are the same between OKT3 and R~1 (Fig. 3). When constructs based on the loop choice for CDRl (gL121) and the Kabat choice (gL221) were made and co-expressed with mH or c~ no evidence for antigen binding activity could be found for gL121, but trace activity could be detected for the gL221, suggesting that a single extra mouse residue in the grafted variable region could have some detectable effect. Both gene constructs were reasonably well expressed in the transient expression system.
15.1.2. FRAMEWORK RESIDUES
The r~m~i n; ng ~ramework residues were then further ~;ned, in particular amino acids known from X-ray analysis of other antibodies to be close to the CDRs and also those amino acids which in OKT3 showed differences from the consensus framework for the mouse subgroup (subgroup VI) to which OKT3 shows most homology. Four positions 1, 3, 46 and 47 were identified and their possible contribution was examined by substituting the mouse amino acid for the human amino acid at each position.
Therefore gL221A (gL221 + DlQ, Q3~, L46R, L47W, ~ ~ - 46 - ~ 2~ ~

see Figure 3 and Table 1) was made, cloned in E~6hCMVneo and co-expressed with cH (pJA144). The resultant antibody was well expressed and showed good bi n~; ng activity. When the related genes gL221B (gL221 + DlQ, Q3V) and gL221C (gL221 +
L46R, L47W) were made and similarly tested, while both genes produced antibody when co-expressed with c~, only the gL221C/cH combination showed good antigen binding. When the gL12lA (gL121 DlQ, Q3V, L46R, L47W) gene was made and co-expressed with cH, antibody was produced which also bound to antigen.
15.2. HEAVY CHAIN
15.2.1. EXTENT OF THE CDRs For the heavy chain the loop and hypervariability analyses agree only in CDR3. For CDR1 the loop region extends from residues 26-32 inclusive whereas the Kabat CDR extends from residues 31-35 inclusive. For CDR2 the loop region is from 50-58 inclusive while the hypervariable region covers amino acids 50-65 inclusive. Therefore humanised heavy chains were constructed using the framewor~ from antibody KOL and with various com~inations of these CDR choices, including a shorter choice for CDR2 of 50-56 inclusive as there was some uncertainty as to the definition of the end point for the CDR2 loop around residues 56 to 5B. The genes were co-expressed with mL or cL
initially. In the case of the gH genes with loop choices for CDR1 e.g. gH121, gH131, gH141 very little antibody was produced in the culture supernatants. As no free light chain was detected it was presumed that the antibody was being made and assembled inside the cell but that the heavy chain was aberrant in some way, possibly incorrectly folded, and therefore the antibody was 2~ 2'~21~

being degraded internally. In some experiment~
trace amounts of antibody could be detected in 35S
labelling studies.
As no net antibody was produced, analysis of these constructs was not pursued further.
When, however, a combination of the loop choice and the Kabat choice for CDRl was tested (mouse amino acids 26-35 inclusive) and in which residues 31 (Ser to Arg), 33 (Ala to Thr), and 35 (Tyr to His) were changed from the human residues to the mouse residue and compared to the first series, antibody was produced ~or gH321, kgH331 and kgH341 when co-expressed with cL. Expression was generally low and could not be markedly improved by the insertion of the Kozak consensus sequence 5' to the ATG of the signal sequence of the gene, as distinct from the case of the gL genes where such insertion led to a 2-5 fold increase in net antibody production. However, only in the case of gH341JmL or kgH341/cL could marginal antigen binding activity be demonstrated. ~hen the kgH341 gene was co-expressed with kgL22lA, the net yield of antibody was too low to give a signal above the background level in the antigen binding Z5 assay.
15 . 2 . 2 . FRAMEWORK RESIDUES
As in the case of the light chain the heavy chain frameworks were re-examined. Possibly because of the lower initial homology between the mouse and human heavy variable domains compared to the light chains, more amino acid positions proved to be of lnterest. Two genes kgH34lA and kgH34lB were constructed, with 11 or 8 human residues respectively substituted by mouse residues compared to gH341, and with the CDR2 residue ~3 returned to the human amino acid potentially to improve dom~ i n packing. Both showed antigen binding when combined with c~ or kgL221A, the kgH34lA gene with all 11 changes appearing to be the superior choice.
5 15.3 INTERIM CONCLUSIONS
It has been demonstrated, therefore, for OKT3 that to transfer antigen binding ability to the humanised antibody, mouse residues outside the CDR
regions defined by the Kabat hypervariability or structural loop choices are required for both the light and heavy chains. Fewer extra residues are needed for the light chain, possibly due to the higher initial homology between the mouse and human kappa variable regions.
Of the changes seven (1 and 3 from the light chain and 6, 23, 71, 73 and 76 from the heavy chain) are predicted from a knowledge of other antibody structures to be either partly exposed or on the antibody surface. It has been shown here that residues l and 3 in the light chain are not absolutely required to be the mouse sequence; and for the heavy chain the gH341B heavy chain in combination with the 22lA light chain generated only weak binding activity Therefore the Z5 presence of the 6, 23 and 24 changes are important to maintain a binding affinity similar to that of the murine antibody. It was important, therefore, to further study the individual contribution of othe other 8 mouse residues of the kgH34lA gene compared to kg~341.

16. FURTHER CDR-GRAFTING ~XPERIMENTS
Additional CDR-grafted heavy chain genes were prepared substantially as described above. With reference to Table 2 the further heavy chain genes were based upon the gh341 (plasmid pJAl78) and ~' _ 49 2~2~2~

gH341A (plasmid pJA185) with either mouse O~T3 or human KOL residues at 6, 23, 24, 48, 49, 63, 71, 73, 76, 78, 88 and 91, as indicated. The CDR-gra~ted light chain genes used in these further experiments were gL221, gL22lA, gL22lB and gL22lC
as described above.

2~ 2~2~

1. gH341 and derivatives 5 OKT3vh Q K A I G F T K S A A Y
gH341 E S S V A F R N N L G F JA178 gH341A Q K A I G V T K S A A Y JA185 gH341E Q K A I G V T K S A G G JA198 gH341* Q K A I G V T K N _ G F JA207 10 gH341* Q K A I G V R N N _ G F JA209 gH341D Q K A I G V T K N L G F JA197 gH341* Q K A I G V R N N L G F JAl99 gH341C Q K A V A _ R N N L G F JA184 gH341* Q S A I G V T K S A A Y JA203 15gH341* E S A I G V T K S A A Y JA205 gH341B ~ S S I G V T K S A A Y JA183 gH341* ~ S A I G V T K S A G F JA204 gH34L* E S A I G V T K S A G F JA206 gH341* Q S A I G V T K N _ G F JA208 2. gL221 and derivatives OKT3vl Q V R W
25 GL221 ~ Q L L DA221 gL22lA Q V R W DA22lA
gL221B Q V L L DA221B

REl D Q L L

MURINE RESIDUES ARE UNDERLINED

51 2~2~ 9 The CDR-grafted heavy and light chain genes were co-expressed in COS cells either with one another in various combinations but also with tha corresponding murine and chimeric heavy and light chain genes substantially as described above. The resultant antibody products were then assayed in binding and blocking assays with HPB-ALL cells as described above.

The results of the assays for various grafted heavy chains co-expressed with the gL221C light chain are given in Figures 7 and 8 (for the JA184, JA185, JA197 and JA198 constructs - see Table 2), in Figure g (for the JA183, JA184, JA185 and JA197 constructs) in Figure 10 (for the chimeric, JA185, JA199, JA204, JA205, JA207, JA208 and JA209 constructs) and in Figure 11 (for the JA183, JA184, JA185, JA198, JA203, JA205 and JA206 constructs).

The basic grafted product without any human to murine changes in the variable framewor~s, i.e. gL221 co-expressed with gh341 (JA178), and also the "fully grafted" product, having most human to murine changes in the grafted heavy chain framework, i.e. gL221C
co-expressed with gh341A (JA185), were assayed for relative binding affinity in a competition assay against murine OKT3 reference standard, using ~PB-ALL cells. The assay used was as described a~ove in section 3.3. The results obtained are given in Figure 12 for the basic grafted product and in Figure 13 for the fully graf~ed product. These results indicate that the basic grafted product has neglibible binding ability as compared with the OKT3 murine reference st~n~rd; whereas the "fully grafted" product has a binding ability very similar to that of the OKT3 murine reference standard.

The binding and blocking assay results indicate the following:

~ - 52 _ 2~2~2~9 The JA198 and JA207 constructs appear to have the best binding characteristics and S;mi lar binding abilities, both substantially the same as the chimeric and fully grafted gH341A products. This indicates that position~
88 and 91 and position 76 are not highly critical for maintaining the OKT3 binding ability; whereas at lea~t some of positions 6, 23, 24, 48, 49, 71, 73 and 78 are more important.

This is borne out by the finding that the JA209 and JA199, although of similar binding ability to one another, are of lower binding ability than the JA198 and JA207 constructs. This indicates the importance of having mouse residues at positions 71, 73 and 78, which are either completely or partially human in the JA199 and JA209 constructs respectively.

Moreover, on comparing the results obtained for the JA205 and JA183 constructs it is seen that there is a decrease in binding going from the JA205 to the JA183 constructs.
This indicates the importance of retA; n; ng a mouse re~idue at position 23, the only position changed between JA205 and JA183.

These and other results lead us to the conclusion that of the 11 mouse framewor~ residues used in the gH341A (JA185) construct, it is important to retain mouse residues at all of positions 6, 23, 24, 48 and 49, and possibly for m~;ml1m binding affinity at 71, 73 and 78.

Similar Experiments were carried out to CDR-graft a number of the rodent antibodies including antikodies having specificity for CD4 (OKT4), ICAM-1 (R6-5), TAG72 (B72.3), and TNF~<(6lE71, 101.4, hTNF1, hTNF2 and hTNF3).

- 53 ~

~ EXAMPLE 2 CDR-~RAFTING OF A MURINE ANTI-CD4 T CELL
RECEPTOR ANTIBODY, OKT4A
Anti OKT4A CDR-gra~ted heavy and light chain genes were prepared, expressed and tested substantially as described above in Example 1 ~or CDR-grafted OKT3.
The CDR grafting of OKT4A is described in detail in Ortho patent application PCT/GB90/02015 of even date herewlth entitled "Humanised Antibodies". A number of CDR-grafted OKT4 antibodies have been prepared.
Presently the CDR-grafted OKT4A of choice is the combination of the grafted light chai~ LCDR2 and the grafted heavy chain HCDR10 T~E LIGHT CHAIN
The human acceptor framewor~ used for the grafted light chains was RE1. The preferred LCDR2 light chain has human to mouse changes at positions 33, 34, 38, 49 and 89 in addition to the structural loop CDRs. Of these changed positions, positions 33, 34 and 89 fall withln the preferred extended CDRs of the present invention (positions 33 and 34 in CDRl and position 89 in CDR3).
The human to murine changes at positions 38 and 49 corresponds to positions at which the amino acid residues are preferably donor murine amino acid res-idues in accordance with the present invention.
A comparison o~ the amino acid sequences of the donor murlne light chain variable domain and the ~E1 human ac~eptor light chain variable further reveals that the murine and human residues are identical at all of positions 46, 48 and 71 and at all of positions 2, 4, 6, 35, 36, 44, 47, 62, 64-69, 85, 87, 98, 99 and 101 and 102.
However the amino acid residue at position 58 in LCDR2 is A' ~ _ 54 _ 2~292~
.

the human RE1 framework residue not the mouse OKT4 residue as would be preferred in accordance with the present invention.

T~E E~EAVY C~AIN
The human acceptor framewor~ used for the grafted heavy chains was KOL.
The preferred CDR graft HCDR10 heavy chain has human to mouse changes at positions 24, 35, 57, 58, 60, 88 and 91 in addition to the structural loop CDRs.
of these positions, positions 35 (CDRl~ and positions 57, 58 and 60 (CDR2) fall within the preferred extended CDR~
.of the present invention. Also the hllm~n to mouse change at position 24 corresponds to a po~ition at which the amino acid residue is a donor murine residue in ac~ordance with the present invention. Moreover, the human to mouse changes at positions 88 and 91 correspond to position~ at which the ~;no acid re~idues are optionally donor murine residues.
Moreover, a comparison of the murine OKT4A and human KOL
Xeavy chain variable amino acid sequences reveals that the murine and human residues are identical at all of positions 23, 49, 71, 73 and 78 and at all of positions 2, 4, 6, 25, 36, 37, 39, 47, 48, 93, 94, 103, 104, 106 and 107.
Thus the OKT4A CDR-grafted heavy chain ~CDR10 corresponds to a particularly preferred embodiment according to the present invention.

~ _ 55 _ ~ 1~ 92~3 EXAMPI~ 3 CDR-GRAFTING OF AN ANTI-MUCIN SPECIFIC MURINE
ANTIBODY, B72.3 The cloning of the genes coding for the anti-mucin specific murine monoclonal antibody B72.3 and the preparation of B72.3 mouse-human chimeric antibodies has been described previously (ref. 13 and WO 89/01783).
CDR-grafted versions of B72.3 were prepared as follows.
(a) B72.3 Liqht Chain CDR-grafting of this light chain was accomplished by direct transfer of the murine CDRs into the framework of the human light chain RE1.
The regions transferred were:

CDR Number Residues The activity of the resulting grafted light chain was assessed by co-expression in COS cells, of 20genes for the combinations:
B72.3 c~/B72.3 c~
and B72.3 c~/B72.3 gL

Supernatants were assayed for antibody concentration and for the ability to bind to microtitre plates coated with mucin. The results obtained indicated that, in combination with the B72.3 c~ chain, B72.3 cL and B72.3 gL
had similar binding properties.

Comparison o~ the murine B72.3 and REI light chain ~m; no acid se~uences reveals that the residues are identical at positions 46, 58 and 71 but are different at position 48.

~ 56 _ 2 ~29 .

Thus changing the human residue to the donor mouse residue at position 48 may further improve the binding characteristics of the CDR-grafted light chain, (B72.3 gL) in accordance with the present invention.

(b) B72.3 heavy chain i. Choice of framework At the outset it was necessary to make a choice of human framework. Simply put, the question was as follows: Was it necessary to use the framework regions from an antibody whose crystal structure was known or could the choice be made on some other criteria~
For B72.3 heavy chain, it was reasoned that, while knowledge of structure was important, transfer of the CDRs fro_ mouse to human frameworks ~ight be facilitated if the overall homology between the donor and receptor frameworks was m~x; m; sed.
2D Comparison of the B72.3 heavy chain sequence with those in Kabat (ref. 4) for human heavy chains showed clearly that B72.3 had poor homology for ROL and NEWM
(for which crystal structures are available) but was very homologous to the heavy chain for EU.
On this basis, EU was chosen for the CDR-grafting and the following residues transferred as CDRs.

CDR Number Residues ~ 212~21 ~ ~ - 57 -.

Also it was noticed that the FR4 region of EU was unlike that of any other hl~an (or mouse) antibody. Consequently, in the grafted heavy chain genes this was also changed to produce a Uconsensus'' human sequence. (Preliminary experiments showed that grafted heavy chain genes cont~; n; ng the EU FR4 sequence exprsssed very poorly in transient expression systems.) ii. Results with grafted heavy chain gene~
Expression of grafted heavy chain genes containing all human framework regions with either gL or c~ genes produced a grafted antibody with little ability to bind to mucin. The grafted antibody had about 1%
the activity of the chimeric antibody.
In these experiments, however, it was noted that the activity of the grafted antibody could be increased to~ 10~ of B72.3 by exposure to pHs of 2-3.5.
This observation provided a clue as to how the activity of the grafted antibody could be improved without acid treatment. It was postulated that acid exposure brought about the protonation of an acidic residue (pRa of aspartic acid = 3.86 and of glutamine acid = 4.25) which in turn caused a change in structure of the CDR loop~, or allowed better access of antigen.
From comparison of the se~uences of ~72.3 (ref. 13) and EU (refs. 4 and 5), it was clear that, in going from the mouse to human frameworks, only two positions had been changed in such a way that acidic residues had been introduced. These ' ~ - 58 - 2~ 213 positions are at residues 73 and 81, where K to E and Q to E changes had been made, respectively.
Which of these positions might be important was determined by ex~mi n; ng the crystal structure of the ~OL antibody. In RO~
heavy chain, position 81 is far removed from either of the CDR loops.
Position 73, however, is clo5e to both CDRs 1 and 3 of the heavy chain and, in this position it was pos~ible to envisage that a K to E change in this region could have a detrimental effect on antigen binding.
iii. Framework changes in B72.3 g~ qene On the basis of the above analysis, E73 was mutated to a lysine (~. It was found that this change had a dramatic effect on the ability of the grafted Ab to bind to mucin. Further the ability of the grafted B72.3 produced by the mutated gH/gL
combination to bind to mucin was similar to that of the B72.3 chimeric antibody.
iv. Other ~ramework changes In the course of the above experiments, other changes were made in the heavy chain framework regions. Within the accuracy of the assays used, none of the changes, either alone or together, appeared beneficial.
v. Other All assays used measured the ability of the grafted A~ to bind to mucin and, as a whole, indicated that the single framework change at position 73 is sufficient to generate an antibody with similar binding properties to B72.3.

~12~

Comparison of the B72.3 murine and EU heavy chain sequences reveals that the mouse and human residues are identical at positions 23, 24, 71 and 78.
Thus the mutated CDR-grafted B72.3 heavy chain corresponds to a preferred em~odiment of the present invention.

2 1292 1~
- ~ - 60 -ANTIBQDY
A murine antibody, R6-5-D6 (EP 0314863) having speci~icity for Intercellular Adhesion Molecule 1 (ICAM-1) was CDR-grafted substa~tially as described above in previous examples. This work is described in greater detail in published European application EP-A-0 528 951.
The human EU framewo~k was used as the acceptor framework for ~oth heavy and light chains. The CDR-grafted antibody currently of choice is provided by co-expression of grafted light chain gL22lA and grafted heavy chain gH341D which has a binding affinity for ICAM 1 of about 75~ of that of the corresponding mouse-human chimeric antibody.
LIGHT C~AIN
gL221A has murine CDRs at positions 24-34 (CDRl), 50-56 (CDR2) and 89-97 (CDR3). In addition several framewor~
residues are also the murine amino acid. These residues were chosen after consideration of the possible contribution of these residues to domain pac~ing and stability of the conformation of the antigen ~inding region. The residues which have been retained as mouse are at positions 2, 3, 48 (?), 60, 84, 85 and 87.
Comparison of the murine anti-ICAM 1 and human EU light chain amino acid sequences reveals that the murine and human residues are identical at positions 46, 58 and 71.
~EAVY C~AIN
gH341D has murine CDRs at positions 26-35 (CDRl), 50-56 (CDR2) and94-lOOB (CDR3). In addition murine residues were used in g~341D at positions 24, 48, 69, 71, 73, 80, 88 and 91. Comparison of the murine anti-ICAM 1 and human EU heavy chain amino acid sequences are identical at positions 23, 49 and 78.
.. . .

, . -- 61 - ~12~2 EXAMP~ 5 CDR-Graftinq of murine anti-TNFa antibodies A num~er of murine anti-TNFa monoclonal antibodies were CDR-grafted substantially as described above in previous examples. These antibodies include the murine monoclonal antibodies designated 61 E71, hTNFl, hTNF3 and 101.4 A
brief summary of the CDR-grafting of each of these antibodies is given below.

6lE71 A similar analysis as described above (Example 1, Section 12.1.) was done for 61E71 and for the heavy chain 10 residues were identified at 23, 24, 48, 4g, 68, 69, 71, 73, 75 and 88 as residues to potentially retain as murine. The human frameworks chosen for CDR-grafting of this antibody, and the hTNF3 and 101.4 antibodies were RE1 for the light chain and KOL for the heavy chain.
Three genes were built, the first of which contained 23, 24, 48, 49, 71 and 73 [gH341(6)] as murine residues. The second gene also had 75 and 88 as murine residues tgH341~8)] while the third gene additionally had 68, 69, 75 and 8~ as murine residues tgH341(10)]. Each was co-expressed with gL221, the ~;n;~ll~ grafted light chain ~DRs only). The gL221/gH341(6) and gL221/q~341(8) antihodies both bound as well to TNF as murine 6lE71.
The gL221/gH341(10) antibody did not express and this combination was not taken further.
Subsequently the gL221/gH341(6) antibody was assessed in an ~929 eell eom~et~tion ascay in whi~h the antibody competes against the TNF receptor on L929 cells for binding to TNF in solution. In this assay the gL221/gH341(6) antibody was approximately 10~ as active as murine 6lE71.

~ - 62 - ~ 2 ~ ~

hTNF1 hTNF1 is a monoclonal antibody which recognises an epitope on human TNF-~ . The E~ human framework was used for CDR-grafting of both the heavy and light variable dom~; n~ .

~eavy Chain In the CDR-grafted heavy chain (ghTNF1) mouse CDRs were used at positions 26-~5 (CDR1), 50-65 (CDR2) and 95-102 (CDR3). Mouse residues were also used in the framewor~s at positions 48, ~7, 69, 71, 73, 76, 89, 91, 94 and 108.
Comparison of the TNF1 mouse and EU human heavy chain residues reveals that these are identical at positions 23, 24, 29 and 78.

~ight Chain In the CDR-grafted light chain (gLhTNF1) mouse CDRs wre used at po~itions 24-34 (CDR1), 50-56 (CDR2) and 89-97 (CDR3). In addition mouse residues wer~ used in the framewor~s at positions 3, 42, 48, 49, 83, 106 and 108.
Comparison of the hTNF1 mouse and ~U human light chain residues reveals that these are identical at positions 46, 58 and 71.

The grafted hTNF1 heavy chain was co-expressed with the chimeric light chain and the binding a~ility of the product compared with that of the chimeric-light chain/chimeric heavy chain product in a TNF binding assay.
The grafted heavy chain product appeared to have bin~;ng ability for TNF slightly ~etter than the fully chimeric product.

Similarly, a grafted heavy chain/grafted light chain product was co-expressed and compared with the fully chimeric product and found to have closely similar b;n~;ng propertie~ to the latter product.

P~" "

hTNF3 hTNF3 recognises an epitope on human TNF- ~. The sequence of hTNF3 shows only 21 differences compared to 61E71 in the light and heavy chain variable regions, 10 in the light chain (2 in the CDRs at positions 50, 96 and 8 in the framework at 1, 19, 40, 45, 46, 76, 103 and 106) and 11 in the heavy chain (3 in the CDR regions at positions 52, 60 and 95 and 8 in the framework at 1, 10, 38, 40, 67, 73, 87 and 105). The light and heavy chains of the 6lE71 and hTNF3 chimeric antibodies can be exchanged without loss of activity in the direct binding assay. ~owever 61E71 is an order of magnitude less able to compete with the TNF receptor on L929 cells for TNF-a compared to hTNF3. Based on the 6lE71 CDR gra~ting data gL221 and gH341(+23, 24, 48, 4~ 71 and 73 as mouse) genes have been built for hTNF3 and tested and the resultant grafted antibody binds well to TNF-a, but competes very poorly in the L929 assay. It is possible that in this case also the framework residues identified for OKT3 programme may improve the competitive binding ability of this antibody.

101.4 101.4 is a further mlrine monoclonal antibody able to recognise human TNF-a. The heavy chain of this antibody shows good homology to KOL and so the CDR-grafting has been based on REl for the light chain and ROL for the heavy chain. Several grafted heavy chain genes have been constructed with conservative choices for the CDR's (gH341) and which have one or a small number of non-CDR
residues at positions 73, 78 or 77-79 inclusive, as the mouse amino acids. These have been co-expressed with cL
or gL221. In all cases binding to TNF equivalent to the chimeric antibody is seen and when co-expressed with cL
the resultant antibodies are able to compete well in the L929 assay. However, with gL221 the resultant antibodies - 64 _ 2~ 29~ 1~

are at least an order of magnitude less able to compete for TNF against the TNF receptor on L929 cells.

Mouse residues at other positions in the heavy chain, for example, at 23 and 24 together or at 76 have been demonstrated to provide no improvement to the competitive ability of the grafted antibody in the L929 assay.

A num~er of other antibodies including antibodies having specificity for interleukins e.g. IL1 and cancer markers such as carcinoembryonic antigen (CEA) e.g. the monoclonal antibody A5B7 (ref. 21), have been successfully CDR-grafted according to the present invention.
It will be appreciated that the foregoing examples are given by way of illustration only and are not intended to limit the scope of the claimed invention. Changes and modifications may be made to the methods described whilst still falling within the spirit and scope of the invention.

~ - 65 _ 2~2~

References 1. Kohler & Milstein, Nature, 265, 295-497, 1975.

2. Chatenoud et al, (1986), J. Immunol. 137, 830-838.

3. Jeffers et al, (1986), Transplantation, 41, 572-578.

4. Begent et al, Br. J. Cancer 62: 487 (1990).

5. Verhoeyen et al, Science, 239, 1534-1536, 1988.

6. Riechmann et al, Nature, 332, 323-324, 1988.

7. Kabat, E.A., Wu, T.T., Reid-Miller, M., Perry, H.M., Gottesman, R.S., 1987, in Sequences of Proteins of Immunological Interest, US Department of ~ealth and Human Services, NIH, USA.

8. Wu, T.T., and Rabat, E.A., 1970, J. Exp. Med. 132 211-250.
.

9. Queen et al, (1989), Proc. Natl. Acad. Sci. USA, 86, 10029-10033 and WO 90/07861 10. Maniatis et al, Molecular Cloning, Cold Spring Harbor, New York, 1989.

11. Primrose and Old, Principles of Gene Manipulation, Blac~well, Oxford, 1980.

12. Sanger, F., Nic~len, S., Coulson, A.R., 1977, Proc.
Natl. Acad. Sci. USA, 74 5463 ~ - 66 - ~12~21~

13. Kramer, W., Drutsa, V., Jansen, H.-W., Rramer, B., Plugfelder, M., Fritz, H.-J., 1984, Nucl. Acids Res.
12, 9441 14. Whittle, N., Adair, J., Lloyd, J.C., Jenkins, E., Devine, J., Schlom, J., Raubitshek, A., Colcher, D., Bodmer, M., 1987, Protein Engineering 1, 499.

15. Sikder, S.S., Akolkar, P.N., Raledas, P.M., Morrison, S.L., Rabat, B.A., 1985, J. Immunol. 135, 4215.

16. Wallick, S.C., Rabat, E.A., Morri~on, S.L., 1988, J. Exp. Med. 168, 1099 17. Bebbington, C.R., Published International Patent Application W0 89 / 010 3 6.

18. Granthan and Perrin 1986, Immunology Today 7, 160.

19. Rozak, M., 1987, J. Mol. Biol. 196, 947.

20. Jones, T.P., Dear, P.H., Foote, J., Neuberger, M.S., ~inter, G., 1986, Nature, 321, 522 21. Harwood et al, Br. J. Cancer, 54, 75-82 (1986).

Claims (14)

1. An antibody molecule having affinity for a predetermined antigen and comprising a composite heavy chain and a complementary light chain, said composite heavy chain having a variable domain comprising acceptor antibody heavy chain framework residues and donor antibody heavy chain antigen-binding residues, said donor antibody having affinity for said predetermined antigen, wherein, according to the Kabat numbering system, in said composite heavy chain, amino acid residues 5, 8, 10, 12 to 17, 19, 21, 22, 40, 42 to 44, 66, 68, 70, 74, 77, 79, 81, 83 to 85, 90, 92, 105, 109, 111 and 113 at least are acceptor residues and amino acid residues 23, 24, 31 to 35, 49 to 58 and 95 to 102 at least are donor residues provided that the antibody molecule does not have affinity for the p55 chain of the human interleukin 2 receptor.

2. The antibody molecule of claim 1, wherein amino acid residues 26 to 30 and 59 to 65 in said composite heavy chain are additionally donor residues.

3. The antibody molecule of claim 1 or claim 2, wherein at least one of amino acid residues 1, 3, and 76 in said composite heavy chain are additionally donor residues.

4. The antibody molecule of claim 1 or claim 2, wherein at least one of amino acid residues 36, 94, 104, 106 and 107 in said composite heavy chain are additionally donor residues.

5. The antibody molecule of claim 4, wherein at least one of amino acid residues 2, 4, 6, 38, 48, 67 and 69 in said composite heavy chain are additionally donor residues.

6. The antibody molecule of claim 1, claim 2 or claim 5, wherein amino acid residues 7, 9, 11, 18, 20, 25, 37, 39, 41, 45, 47, 48, 72, 75, 80, 82, 86 to 89, 91, 93, 103, 108, 110 and 112 in said composite heavy chain are additionally acceptor residues.

7. The antibody molecule of claim 1, claim 2 or claim 5, wherein said complementary light chain is a composite light chain having a variable domain comprising acceptor antibody light chain framework residues and donor antibody light chain antigen-binding residues, said donor antibody having affinity for said predetermined antigen, wherein, according to the Kabat numbering system, in said composite light chain, amino acid residues 5, 7 to 9, 11, 13 to 18, 20, 22, 23, 39, 41 to 43, 57, 59, 61, 72, 74 to 79, 81, 82, 84, 86, 88, 100, 104, 106 and 107 at least are acceptor residues and amino acid residues 24 to 34, 46, 48, 50 to 56, 58, 71 and 89 to 97 at least are donor residues.

8. The antibody molecule of claim 7, wherein amino acid residues 1, 3 and 47 in said composite light chain are additionally donor residues.

9. The antibody molecule of claim 7 wherein amino acid residues 36, 44, 47, 85 and 87 in said composite light chain are additionally donor residues.

10. The antibody molecule of claim 7, wherein at least one of amino acid residues 2, 4, 6, 49, 62, 64 to 69, 98, 99, 101 and 102 in said composite light chain are additionally donor residues.

11. The antibody molecule of claim 7, wherein at least one of amino acid residues 1, 3, 10, 12, 21, 40, 60, 63, 70, 73, 80, 103 and 105 in said composite light chain are additionally donor residues.

12. A therapeutic or diagnostic composition comprising the antibody molecule of claim 1, claim 2, claim 5, claim 8, claim 9, claim 10 or claim 11 in combination with a pharmaceutically acceptable carrier, diluent or excipient.

13. A method for producing a recombinant antigen binding molecule having affinity for a predetermined antigen comprising the steps of:
[1] determining the amino acid sequence of the variable domain of the heavy chain of a donor antibody which has affinity for said predetermined antigen;
[2] determining the amino acid sequence of the variable domain of the heavy chain of a non-specific acceptor antibody;
[3] providing a composite heavy chain for an antibody molecule, said composite heavy chain having acceptor framework residues and donor antigen binding residues wherein, according to the Kabat numbering system, amino acid residues 5, 8, 10, 12 to 17, 19, 21, 22, 40, 42 to 44, 66, 68, 70, 74, 77, 79, 81, 83 to 85, 90, 92, 105, 109, 111 and 113 at least are acceptor residues and amino acid residues 23, 24, 31 to 35, 49 to 58 and 95 to 102 at least are donor residues;
[4] associating the heavy chain produced in step [3] with a complementary light chain to form an antibody molecule;
[5] determining the affinity of the antibody molecule formed in step [4] for said predetermined antigen;
[6] if the affinity determined in step [5] is not equivalent to that of the donor antibody, providing a heavy chain as described in [3] above but in which amino acid residues 71, 73 and 78 are additionally donor residues;
[7] associating the heavy chain produced in step [6] with a complementary light chain to form an antibody molecule;
[8] determining the affinity of the antibody molecule formed in step [7] for said predetermined antigen;

[9] if the affinity determined in step [8] is not equivalent to that of the donor antibody, providing a heavy chain as described in [6] above but in which amino acid residues 26 to 30 are additionally donor residues;
[10] associating the heavy chain produced in step [9] with a complementary light chain to form an antibody molecule;
[11] determining the affinity of the antibody molecule formed in step [10] for said predetermined antigen;
[12] if the affinity determined in step [11] is not equivalent to that of the donor antibody, providing a heavy chain as described in [9] above but in which at least one of amino acid residues 1, 3, and 76 are additionally donor residues;
[13] associating the heavy chain produced in step [12] with a complementary light chain to form an antibody molecule;
[14] determining the affinity of the antibody molecule formed in step [13] for said predetermined antigen;
[15] if the affinity determined in step [14] is not equivalent to that of the donor antibody, providing a heavy chain as described in [12] above but in which at least one of amino acid residues 36, 94, 104, 106, 107 are additionally donor residues;
[16] associating the heavy chain produced in step [15] with a complementary light chain to form an antibody molecule.
[17] determining the affinity of the antibody molecule formed in step [16] for said predetermined antigen;
[18] if the affinity determined in step [17] is not equivalent to that of the donor antibody, providing a heavy chain as described in [15] above but in which at least one of amino acid residues 2, 4, 6, 38, 48, 67 and 69 are additionally donor residues; and [19] associating the heavy chain produced in step [18] with a complementary light chain to form an antibody molecule.

14. The method of claim 13, further comprising the steps of:

[1] determining the amino acid sequence of the variable domain of the light chain of said donor antibody which has affinity for said predetermined antigen;
[2] determining the amino acid sequence of the variable domain of the light chain of a non-specific acceptor antibody;
[3] providing a composite light chain for an antibody molecule, said composite light chain having acceptor framework residues and donor antigen binding residues wherein, according to the Kabat numbering system, amino acid residues 5, 7 to 9, 11, 13 to 18, 20, 22, 23, 39, 41 to 43, 57, 59, 61, 72, 74 to 79 to 81, 82, 84, 86, 88, 100, 104 and 106 to 109 at least are acceptor residues and amino acid residues 24 to 34, 46, 48, 50 to 56, 58, 71 and 89 to 97 at least are donor residues;
[4] associating the light chain produced in step [3]
with a complementary heavy chain produced by the method of claim 13 to form an antibody molecule;
[5] determining the affinity of the antibody molecule formed in step [4] for said predetermined antigen;
[6] if the affinity determined in step [5] is not equivalent to that of the donor antibody, providing a light chain as described in [3] above but in which amino acid residues 1, 2, 3 and 47 are additionally donor residues;
[7] associating the light chain produced in step [6]
with a complementary heavy chain produced by the method of claim 13 to form an antigen-binding molecule;
[8] determining the affinity of the antigen-binding molecule formed in step [7] for said predetermined antigen;
[9] if the affinity determined in step [8] is not equivalent to that of the donor antibody, providing a light chain as described in [6] above but in which amino acid residues 36, 44, 47, 85 and 87 are additionally donor residues;

[10] associating the light chain produced in step [9]
with a complementary heavy chain produced by the method of claim 13 to form an antibody molecule;
[11] determining the affinity of the antibody molecule formed in step [10] for said predetermined antigen;
[12] if the affinity determined in step [11] is not equivalent to that of the donor antibody, providing a light chain as described in [9] above but in which at least one of amino acid residues 2, 4, 6, 49, 62, 64 to 69, 98, 99, 101 are additionally donor residues; and [13] associating the light chain produced in step [9]
with a complementary heavy chain produced by the method of claim 13 to form an antibody molecule.