=
`MAR 11 1993
`ALIHduieWEsUpsaRy
`
`L~
`
`|
`|
`
`20 February 1993
`Number 4
`—
`__ a
`
`
`
`
`
`z
`
`
`AE UN
`
`ACADEMICPRESS
`
`HarcourtBrace Jovanovich, Publishers
`Sere,
`
`JMOBAK229(4)805-1186 ISSN0022-2836
`
`7
`
`-2836(199382)229:4:1-6
`
`Pfizer Ex. 1021
`Page 1 of 31
`
`Pfizer Ex. 1021
`Page 1 of 31
`
`

`

`Journal of Molecular Biology
`Editor-in-Chief
`P. Wright
`Department of Molecular Biology, The Scripps Research Institute
`10666 N. Torrey Pines Road, La Jolla, CA 92037, U.S.A.
`
`Assistant Editor
`J. Karn
`MRCLaboratory of Molecular Biology
`Hills Road, Cambridge CB2 2QH, U.K.
`
`Founding Editor
`Sir John Kendrew
`
`Consulting Editor
`Sydney Brenner
`
`Editors
`A. R. Fersht, University Chemical Laboratory, Cambridge University, Lensfield Road, Cambridge CB2 IEW, U.K.
`M. Gotlesman, Institute of Cancer Research, College of Physicians & Surgeons of Columbia University,
`701 W. 168th Street, New York, NY 10032, U.S.A.
`P. von Hippel, Institute of Molecular Biology, University of Oregon, Eugene, OR 97403-1229, U.S.A.
`R. Huber, Max-Planck-Institut fiir Biochemie, 8033 Martinsried bei Miinchen, Germany.
`A. Klug, MRC Laboratory of Molecular Biology, Hills Road, Cambridge CB2 2QH, U.K.
`M. Yaniv, Department of Biotechnology, Pasteur Institute, 25 rue de Dr Roux, 75724 Paris Cedex 15, France.
`Associate Editors
`N.-H. Chua, The Rockefeller University, 1230 York Avenue, New York, NY 10021, U.S.A.
`F. B. Cohen, Department of Pharmaceutical Chemistry, School of Pharmacy, University of California, San Francisco,
`CA 94143-0446, U.S.A.
`D. J. DeRosier, Rosenstiel Basic Medical Sciences Research Center, Brandeis University, Waltham, MA 02254, U.S,A.
`W. A. Hendrickson, Department of Biochemistry & Molecular Biophysics, College of Physicians & Surgeons of
`Columbia University, 630 West 168th Street, New York, NY10032, U.S.A.
`IB. Holland, Institut de Génétique et Microbiologie, Batiment 409, Université de Paris XT, 91405 Orsay Cedex 05,
`France.
`B. Honig, Department of Biochemistry & Molecular Biophysics, College of Physicians & Surgeons of Columbia
`University, 630 West 168th Street, New York, NY 10032, U.S.A.
`H. EB. Huxley. Rosenstiel Basic Medical Sciences Research Center, Brandeis University, Waltham, MA 02254, U.S.A.
`V. Luzzati, Centre de Génétique Moléculaire, Centre National de la Recherche Scientifique. 91 Gif-sur-Yvette, France.
`B. Matthews. Institute of Molecular Biology, University of Oregon, Eugene, OR 97403-1229, U.S.A.
`JH. Miller, Department of Microbiology, University of California, 405 Hilgard Avenue, Los Angeles, CA 90024, U.S.A.
`M. F, Moody, School of Pharmacy, University of London, 29/39 Brunswick Square, London WCIN 1AX, U.K.
`S. Reed, Department of Molecular Biology, The Scripps Research Institute. 10666 North Torrey Pines Road, La Jolla,
`CA 92037, U.S.A.
`T. Richmond, Institut fiir Molekularbiologie und Biophysik, Eidgendssische Technische Hochschule, Hénggerberg,
`CH 8093 Zurich, Switzerland,
`;
`:
`R. Schleif, Biology Department, Johns Hopkins University, Charles & 34th Streets, Baltimore, MD21218, U.S.A.
`N. L. Sternberg, Du Pont Merck Pharmaceutical Company, Cancer and Inflammatory Diseases, P.O. Box 80328,
`E 328/148 C, Wilmington, DE 19880-0328, U.S.A.
`._
`I. Wilson, The Scripps Research Institute, 10666 North Torrey Pines Road, La Jolla, CA 92037, U.S.A.
`et
`RB. Yamamoto, Department of Biochemistry and Biophysics, School of Medicine, University of California,
`K.
`San Francisco, CA 94143-0448, U.S.A.
`a,
`M. Yanagida, Department of Biophysics, Faculty of Science, Kyoto University, Sakyo-Ku, Kyoto 606, Japan.
`Editorial Office
`;
`oes
`G. Harris, Journal of Molecular Biology, 10d St Edward's Passage, Cambridge CB2 3PJ, U.K.
`
`JOURNAL OF MOLECULAR BIOLOGY: ISSN 0022-2836. Volumes 229-234, 1993, published twice a month on the
`5th and 20th by Academic Press at 24-28 Oval Road. London NWI 7DX, England. Annualsubscription price including
`postage: £1008 U.K. and Europe and U.S.$1860elsewhere. Personal subscription rate: £252 U.K. and Europe and U.S.
`$399 elsewhere, Subscription orders should be sent to Academie Press Limited, Foots Cray, Sidcup, Kent DAI4 aoe
`U.K. (Tel: 081-3003322). Send notices of changes of address to the publisher at least 6-8 weeks in advance, including
`both old and newaddresses.
`Buyers in Canada should add GST at the current rate of 7%.
`Secondclass postage rate paid at Jamaica, NY 11431, U.S.A.
`Air freight and mailing in the U.S.A. by Publications Expediting Inc., 200 Meacham Avenue, Elmont, NY 11003, U.S.A.
`U.S.A. POSTMASTERS:send change of addresses to JOURNAL OF MOLECULAR BIOLOGY, c/o Publications
`Expediting, Inc. 200 Meacham Avenue, Elmont, NY11003, U.S.A.
`Printed in U.K.
`
`Pfizer Ex. 1021
`Page 2 of 31
`
`Pfizer Ex. 1021
`Page 2 of 31
`
`

`

`Journal of Molecular Biology—_——$<eee
`Volume 229, Number 4
`
`Contents
`
`Communications
`Hierarchy of Base-pair Preference in the Binding
`Domain of the Bacteriophage T7 Promoter
`Crossed-Stranded DNA Structures for
`Investigating the Molecular Dynamics of the
`Holliday Junction
`Specificity of the Mnt Protein. Independent
`Effects of Mutations at Different Positions in the
`Operator
`Structure of Nucleosomal DNA at High Salt
`Concentration as Probed by Hydroxyl Radical
`
`G. A. Diaz, C. A. Raskin and
`W. T. McAllister
`
`805-8114_
`
`R. D. Johnson and L. S. Symington
`
`812-820
`
`G. D. Stormo, S. Strobl,
`M. Yoshioka and J. S. Lee
`
`821-826
`
`H. L. Puhl and M. J. Behe
`
`827-832
`
`Articles
`
`Significant Dispersed Recurrent DNA Sequences in
`the Escherichia coli Genome. Several New Groups
`
`B. E. Blaisdell, K. E. Rudd,
`A. Matin and S. Karlin
`
`Duck Lactate Dehydrogenase B/e-Crystallin Gene.
`Lens Recruitment of a GC-promoter
`
`Quantitative Model of Col El Plasmid Copy
`Number Control
`
`Mammalian Heterogeneous Ribonucleoprotein Al
`and its Constituent Domains. Nucleic Acid
`Interaction, Structural Stability and Self-
`association
`
`The Gut Esterase Gene (ges-/) from the
`Nematodes Caenorhabditis elegans and
`Caenorhabditis briggsae
`Nuclear Control of the Messenger RNA Expression
`for Mitochondrial ATPase Subunit 9 in a New
`Yeast Mutant
`
`On the Location of Histones H1 and H5 in the
`Chromatin Fiber. Studies with Immobilized
`Trypsin and Chymotrypsin
`Structure of Hen Lysozymein Solution
`
`Refined 8 A X-ray Crystal Structure of Astacin,
`a Zine-endopeptidase from the Crayfish Astaces
`astacus L. Structure Determination, Refinement,
`Molecular Structure and Comparison with
`Thermolysin
`X-ray Structures of the Antigen-binding Domains
`from Three Variants of Humanized anti-p185"**?
`Antibody 41D5 and Comparison with Molecular
`Modeling
`Modeling Side-chain Conformation for Homologous
`Proteins Using an Energy-based RotamerSearch
`X-ray Analysis and Spectroscopic Characterization
`of M121Q Azurin. A Copper Site Model for
`Stellacyanin
`
`H.J. Kraft, W. Hendriks,
`W. W.de Jong, N. H. Lubsen and
`J. G. G. Schoenmakers
`
`V. Brendel and A. S. Perelson
`
`J. R. Casas-Finet, J. D. Smith, Jr,
`A. Kumar, J. G. Kim, 8. H. Wilson
`and R.L. Karpel
`
`B. P. Kennedy, E. J. Aamodt,
`F. L. Allen, M. A. Chung,
`M. F. P. Heschl and J. D. McGhee
`K. Ziaja, G. Michaelis and
`T. Lisowsky
`
`S. H. Leuba, J. Zlatanova and
`K. van Holde
`
`L. J. Smith, M. J. Sutcliffe,
`C. Redfield and C. M. Dobson
`F. X. Gomis-Riith, W. Stécker,
`R. Huber, R. Zwilling and W. Bode
`
`833-848
`
`849-859
`
`860-872
`
`873-889
`
`890-908
`
`909-916
`
`917-929
`
`930-944
`
`945-968
`
`C. Eigenbrot, M. Randal, L. Presta,
`P. Carter and A. A. Kossiakoff
`
`969-995
`
`C. Wilson, L. M. Gregoret and
`D. A. Agard
`A. Romero, C. W. G. Hoitink,
`H. Nar, R. Huber, A. Messerschmidt
`and G. W. Canters
`
`996-1006
`
`1007-1021
`
`Pfizer Ex. 1021
`Page 3 of 31
`
`Pfizer Ex. 1021
`Page 3 of 31
`
`

`

`Q. Huang,S. Liu and Y. Tang
`
`1022-1036
`
`Refined 1-6 A Resolution Crystal Structure of the
`Complex Formed between Porcine -Trypsin and
`MCTI-A, a Trypsin Inhibitor of the Squash
`Family. Detailed Comparison with Bovine
`B-Trypsin and Its Complex
`Three-dimensional Structure of the Lipoyl Domain
`from Bacillus stearothermophilus Pyruvate
`Dehydrogenase Multienzyme Complex
`Modeling the Three-dimensional Structure of RNA
`Using Discrete Nucleotide Conformational Sets
`Empirical and Structural Models for Insertions
`and Deletions in the Divergent Evolution of
`Proteins
`
`F. Dardel, A. L. Davis, E. D. Laue
`and R. N. Perham
`
`D. Gautheret, F. Major and
`R. Cedergren
`
`S. A. Benner, M. A. Cohen and
`G. H. Gonnet
`
`Three-dimensional Structure of the Glutathione
`Synthetase from Escherichia coli B at 20 A
`Resolution
`
`H. Yamaguchi, H. Kato, Y. Hata,
`T. Nishioka, A. Kimura, J. Oda and
`Y. Katsube
`
`Nicotinic Acetylcholine Receptor at 9 A
`Resolution
`
`N. Unwin
`
`C. P. M. van Mierlo, N. J. Darby,
`J. Keeler, D. Neuhaus and
`T. E. Creighton
`
`Y. Devedjiev, B. Atanasov,
`I. Mancheva and B. Aleksiev
`
`A. M. Hassell, T. N. C. Wells,
`P. Graber, A. E. I. Proudfoot,
`R. J. Anderegg, W. Burkhart,
`S. R. Jordan and M. V. Milburn
`
`K. Lewinski, Y. Hui, C. G. Jakob,
`C. R. Lovell and L. Lebioda
`
`1037-1048
`
`1049-1064
`
`1065-1082
`
`1083-1100
`
`1101-1124
`
`1125-1146
`
`1147-1149
`
`1150-1152
`
`11538-1156
`
`1157-1158
`
`1159-1162
`
`1163-1164
`
`Partially Folded Conformation of the (30-51)
`Intermediate in the Disulphide Folding Pathway
`of Bovine Pancreatic Trypsin Inhibitor. 'H and
`SN Resonance Assignments and Determination of
`Backbone Dynamies from '*N Relaxation
`Measurements
`
`Crystallization Notes
`
`Crystals of Phospholipase A, Inhibitor. The Non-
`toxic Component of Vipoxin from the Venom of
`Bulgarian Viper (Vipera ammodytes)
`
`Crystallization and Preliminary X-ray Diffraction
`Studies of Recombinant Human Interleukin-5
`
`Crystallization and Preliminary Crystallographic
`Data for Formyltetrahydrofolate Synthetase from
`Clostridiumthermoaceticum
`
`Crystallization and Preliminary X-ray
`Crystallographic Analysis of the Protease Inhibitor
`Ecotin
`
`Formation in Vivo, Purification and
`Crystallization of a Complex of the ) and ¢
`Subunits of the FjF,-ATPase of Escherichia coli
`
`Crystallization and Preliminary X-ray Analysis of
`a Truncated Tissue Metalloproteinase Inhibitor
`Ayo8-194 TIMP-2
`
`Sequence Notes
`
`A Superfamily of ATPases with Diverse Functions
`Containing Either Classical or Deviant ATP-
`binding Motif
`
`Expression of the Human Placental Folate
`Receptor Transcript is Regulated in Human
`Tissues. Organization and Full Nucleotide
`Sequence of the Gene
`
`Author Index
`
`<3aSmZaxADD.
`
`K.
`Cc.
`
`G.
`J.
`R.
`
`5.
`R.
`A.
`
`=Ss
`in, K. Y. Hwang,
`im, H. R.
`3 @.S: Lee;
`hung and S. W. Suh
`
`Cox, B. A. Cromer,
`Guss, I. Harvey, P. D.Jeffrey,
`. Solomon and D. C. Webb
`
`a .
`
`lley, G. Murphy, M. O’Shea,
`ard, A. Docherty, M. Cockett,
`Rawas and G. Davies
`
`E. V. Koonin
`
`1165-1174
`
`S. T. Page, W. C. Owen, K. Price
`and P. C. Elwood
`
`1175-1183
`
`1185-1186
`
`Pfizer Ex. 1021
`Page 4 of 31
`
`Pfizer Ex. 1021
`Page 4 of 31
`
`

`

`J. Mol. Biol, (1993) 229, 969-995
`
`X-ray Structures of the Antigen-binding Domains from Three
`Variants of Humanizedanti-p185"*** Antibody 4D5 and
`Comparison with Molecular Modeling
`
`Charles Eigenbrot}, Michael Randal, Leonard Presta, Paul Carter
`and AnthonyA. Kossiakoff
`
`Department of Protein Engineering
`Genentech, Inc., South San Francisco, U.S.A.
`
`(Received 24 July 1992; accepted 7 October 1992)
`
`The X-ray structures of 1 Fy and 2 Fab humanized anti-p185"=*? antibody fragments
`(IgG,-x) have been determined at a resolution between 2-7 A and 2-2 A. The antibodies are
`three different versions of a human antibody framework onto which the antigen recognition
`loops from a murine antibody (4D5) have been grafted. The sequencesof the three versions
`differ in the frameworkregion at positions L55, H78 and H102. The version 8 Fv fragment
`crystallizes in space group P2, with cell parameters a= 37-6 A, b = 63-4 A, c= 902 A,
`f = 98-2°, with two molecules per asymmetric unit, and has been refined against data
`10-0 A—22 A to an R-factor of 183%. Versions 4 and 7 Fabscrystallize in space group P1
`with cell parameters a = 39:2 A, b = 80-2 A, c= 861A, a = 113-1°, B = 92-7 A, y = 10264
`and two molecules per asymmetric unit. Version 4 has been refined against data
`10-0 A—2-5 A resolution to an R-factor of 17-99%. Version 7 has been refined against data
`10 A—2-7 A to an R-factor of 17-1%.
`The X-ray structures have been used to assess the accuracy ofstructural predictions made
`via molecular modeling, and they confirm the structural role of certain frameworkresidues
`and the conformations of five of six complementarity determining regions (CDRs). The
`average deviation of the model from the X-ray structures is within the range observed
`among the X-ray structures for 81% of the C? atoms. Of the hydrogen bonds commonto the
`X-ray structures, 94% of the main-chain-main-chain and 79% of the main-chain-side-
`chain ones were predicted by the model. The side-chain conformation was predicted
`correctly for 79% ofthe buried residues. The third CDR in the heavy chain is variable,
`differing by up to 8A between molecules within an asymmetric unit. The structural
`relationship between variable domainsof light and heavy chains is not significantly altered
`by the absence of constant domainsin the Fv molecule. The antigen-binding potential of an
`unusuallight chain sequence has been confirmed. Thearginineat position 66 interacts with
`thefirstlight chain CDR.but in a fashion somewhat differentthan predicted. A substitution
`of a leucine for an alanine side-chain directed between the B-sheets has only relatively small
`and local effects. There is no direct contact between tyrosines at positions 59 of the light
`chain and 102 of the heavy chain of version 8, despite their proximity and the synergism
`seen in their effects on binding affinity andbiological activity.
`Keywords: humanized antibody; X-ray structure; molecular modeling; effects of mutation;
`binding and antiproliferation
`
`
`there are fundamental immunological limits to their
`applicability. The specific tar eting
`and effector
`i
`;
`;
`;
`fanition seshilitiesore vapetirally distinct, The
`Diagnosis:end eaement of human disease wie
`menoconaantbodin“haserome,ean it reaeete
`target
`selected molecules
`an4 direct
`immune
`extreme diversity of antigenic challenge, and the
`fF
`f ae
`Fi
`h
`Unfortunatel
`second requires sequence invariance for interactions
`effector
`functions against
`them.
`Unfortunately,
`with other members of the native immune system.
`Non-human (e.g. murine) monoclonal antibodies
`iSSEe
`with highly specific and tight binding to a selected
`human antigen can be produced by hybridomatech-
`+ Author to whom all correspondence should be
`addressed.
`© 1993 Academic Press Limited
`0022-2836/93/040969-27 $08.00/0
`
`1. Introduction
`
`969
`
`Pfizer Ex. 1021
`Page 5 of 31
`
`Pfizer Ex. 1021
`Page 5 of 31
`
`

`

`970
`
`C. Eigenbrot et al.
`
`-
`Table 1
`Binding and bioactivity of murine and humanized versions of 4D5
`eeEEE
`V_ residue Vy residuewoeSS
`
`
`
`66 71 73 78 93 102 Kg (nm)ft55 Growth}NN
`
`
`
`
`
`
`4D5
`0-060
`37
`VAL
`ALA
`LEU
`ASP
`ARG
`GLY
`GLU
`verlL
`26
`102
`VAL
`ALA
`LEU
`ASP
`ALA
`GLY
`GLU
`ver2
`3
`101
`VAL
`SER
`ALA
`THR
`ALA
`GLY
`GLU
`ver3.
`1-100
`48
`
`vert=GLU arg ALA THR leu SER VAL 0-420 56
`
`
`
`
`
`
`
`
`vero=GLU arg ALA THR ALA SER VAL 0-430 oe
`
`
`
`
`
`
`
`ver6
`tyr
`arg
`ALA
`THR
`ALA
`SER
`VAL
`0159
`I
`
`ver?=GLU arg ALA THR ALA SER tyr 0-400 58
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`tyr arg ALA THR ALA SER tyr 0-088vers 54a
`
`For versions 4-8, residues which differ from version 3 are in lower case lettering.
`+ RIA assay from Kelley et al. (1992).
`{ Relative cell proliferation +15°%from Carter et al, (1992).
`
`the
`nology (Koehler & Milstein, 1975). However,
`non-antigen-binding regions of these antibodies are
`sufficiently different from their human counterparts
`that their use in humans (Waldmann, 1991) can
`itself cause an immune reaction (Jaffers et al., 1986;
`Miller ef al., 1983; Schroff et al., 1985). Principal
`among the methods being employed to cireumvent
`this phenomenon is the “humanization” of non-
`humanantibodies (Carter et al., 1992; Hale et al.,
`1988; Jones et al., 1986; Riechmann ef al., 1988).
`to
`The purpose of humanizing antibodies
`is
`prevent a human immune response and to provide
`native effector functions while retaining the highly
`specific and tight binding characteristic of the
`antigen/antibody interaction. As a first approxima-
`tion, humanization involves grafting the antigen-
`recognizing complementarity-determining regions
`(CDRsf) from a murine antibody onto a human
`antibody framework. However, it has been demon-
`strated that this step alone is not always sufficient
`(Carter et al., 1992; Queen et al., 1989; Verhoeyen et
`al., 1988) and that other key residues outside the
`hypervariable regions may play an importantrole in
`antigen binding (Kelley et al.,
`1992), whether
`through direct antigen interaction (Amit ef al., 1986;
`Bentleyet al., 1990; Padlan et al., 1989; Sheriff et al.,
`1987) or by providing structural support for a CDR
`(Chothia et al., 1989; Tramontano et al., 1990).
`The murine antibody 4D5 (IgG,-«) was raised
`against
`the extracellular domain of
`the proto-
`oncogene HER2 gene product (p185""*?) (Fendlyet
`al., 1990). Amplified expression of p185"°®? on cell
`surfaces is implicated in 25%,
`to 30% of primary
`breast and ovarian cancers in humans (Slamon et
`al., 1987, 1989), Murine 4D5 (mur4D5) specifically
`inhibits the tm vitro proliferation of human tumor
`cells overexpressing p185"°"? (Hudziak et al., 1989),
`
`
`+ Abbreviations used: CDR, complementarity-
`determining region; r.m.s., root-mean-square, L, light
`chain; H, heavy chain; PEG, polyethylene glycol; RF,
`rotation function; Pc, Patterson correlation; NCS, non-
`crystallographic symmetry; SA, simulated annealing.
`
`The humanization of mur4D5 has been described
`(Carter et al., 1992). An integral step in the human-
`ization process was the modeling of the murine and
`humanized antibodies. The accuracy of these models
`had a direct effect on the success of the humaniza-
`tion. They were used to determine if the CDRs could
`be successfully transplanted from the rodent frame-
`work to a human one and to determine if there were
`anykey framework residues in the rodent antibody
`that needed to be incorporated into the humanized
`antibody in order to maintain a CDR conformation.
`Because manyof the decisions based on the models
`could only be informed guesses, and because the
`errors inherent
`in the modeling process are well
`beyond those required for a significant functional
`impact, candidate sequences were tested in appro-
`priate assays. The candidate molecules varied
`widely in their performance in both antigen binding
`and antiproliferation assays while including only
`three to seven murine residues outside the CDRs
`(Table 1),
`The X-ray crystallographic structures reported
`here are the first of humanized antibody fragments.
`They allow discussion of several specific topics:
`(1) The accuracy of
`the predicted structures is
`determinedto assess the validity of the assumptions
`and estimates inherent
`in the modeling process.
`(2) The structural influence of residues L55, H78
`and H102 are investigated through a detailed com-
`parison among X-ray structures whose sequences
`differ at
`these light (L) and heavy (H) chain posi-
`tions.
`(3) Mur4D5 contains a x-isotype light chain
`with an aminoacid rarely found at position L66.
`Among the sequences available at
`the time the
`humanization was being performed,
`the vast
`majority had glycine, but 4D5 has arginine. Assay
`data have shown that this arginine is important for
`both antigen binding and antiproliferation (Kelley
`et al., 1992; Carter et al., 1992), and we sought to
`determine the structural
`indications of
`its role,
`especially the extent to which it interacts with the
`proximal CDR-LI. (4) The X-ray structures provide
`the only reliable information on the conformationof
`
`Pfizer Ex. 1021
`Page 6 of 31
`
`Pfizer Ex. 1021
`Page 6 of 31
`
`

`

`
`
`X-ray Structures of Humanized Antibody Fragments 971a
`
`CDR-H3, which could not be confidently predicted.
`No generalstructural classification scheme for this
`CDR has been discerned. In 4D5it is eight amino
`acids long and contains three glycines. (5) Fab frag-
`mentsof eight humanized 4D5 antibodies have been
`the subjects of detailed thermodynamic measure-
`ments, allowing us to determine the structural
`effects of sequence changes knownto haveaneffect
`on the thermodynamics of antigen binding (Kelley
`et al., 1992). (6) The X-ray structures of both Fab
`and Fv fragments from nearly identical antibodies
`allows a determination of the extent to which the
`V,/V_
`interface
`is
`influenced by the constant
`domains. Similar comparisons with other antibodies
`have shown someeffect (Bhat et al., 1990; Colman et
`al., 1989).
`(7) Despite the presence in the human
`immunesystem of nearly equal populationsof 4 and
`x light chain genes, crystallographic results thus far
`are almost exclusively of human / chains. The 3D6
`Fab structure (He et al., 1992) is the only prior
`example of an antibody fragment with a human k
`light chain. Thus,
`the humanized 4D5 antibody
`fragments add to the small repertoire of human
`structures. Bradyet al. (1992) recently reported an
`X-ray structure containing a human C.K domain.
`Compared to most current protein engineering
`studies, antibody humanization entails a large
`percentage sequence change. That this can be done
`while retaining antigen bindingaffinity is a proofof
`concept for the even more challenging technical goal
`of taking biologically active fragments from one
`protein and transferring them to another protein
`
`framework in order to produce new activities or
`make polyspecific molecules. Based on work like
`that described here and on the rate of technological
`advance, ab initio design ofspecific biological activi-
`ties on immunoglobulin structural scaffolds could be
`a near-term reality.
`
`2. Materials and Methods
`(a) Model building of mur4D5 (Carteretal., 1992)
`The model of the mur4D5 Vi, and Vy domains was
`generated using consensus co-ordinates based on the
`crystal structures of 7 immunoglobulin Fabs (Table 2)
`(Epp et al., 1975; Furey Jr et al., 1983; Marquartef al.,
`1980; Poljak et al., 1974; Satow et al., 1986; Sheriff et al.,
`1987: Suh ef al., 1986) taken from the Protein Data Bank
`(Bernstein et al., 1977). Use of a consensus main-chain
`structure served to eliminate inappropriate structural
`idiosyncrasies that might have been retained if only 1 of
`these structures had been used as a template. The V,, and
`V, domains were treated separately and only V,« strue-
`tures were used to construct the V, consensus structure
`(Eppet al., 1975; Satowet al., 1986; Sheriff ef al., 1987;
`Suhet al., 1986).
`The following procedure was performed to produce the
`mur4D5 and hu4D5 models:
`(1) Building the consensus
`structure involved assignment of secondary structure for
`all
`immunoglobulin-variable domains. Using backbone
`dihedral angles and hydrogen bonds (program HBOND;
`L. Presta), the 9 B-strands in each variable domain were
`assigned. Although secondary structure assignments are
`often listed in the header of a Protein Data Bankfile,
`these assignments are made by the individual investiga-
`
`Table 2
`Immunoglobulin residues used in superpositioning and those included in the
`consensus structure
`eeSSS
`VL « domain
`IREI
`Consensus}
`Igt
`2-11
`16-27
`33-39
`41-49
`59-77
`
`19-25
`33-38
`
`61-66
`70-75
`85-89
`
`50
`
`2FB4
`18-24
`32-37
`
`60-66
`6-74
`84-88
`
`2MCP
`19-25
`39-44
`
`G7-72
`76-81
`91-95
`
`1FBI
`19-25
`32-37
`
`0-65
`Go-74
`84-88
`
`2HFL
`19-25
`32-37
`
`60-65
`69-74
`84-85
`
`§2-91
`101-105
`
`48
`O54
`060
`rm.s§
`Vy domain
`2HFL—Consensus}
`Igt
`2FB4
`2MCP
`3FAB
`IFBJ
`a8
`17-23
`18-25
`18-25
`18-25
`18-25
`33-41
`34-39
`34-39
`34-39
`34-39
`45-51
`46-52
`46-52
`46-52
`46-52
`57-61
`57-61
`59-63
`56-60
`57-61
`66-71
`68-71
`70-73
`67-70
`68-71
`75-82
`78-84
`80-86
`77-83
`78-84
`88-94
`92-99
`94-101
`91-98
`92-0
`102-108
`0-43
`0:85
`062
`r.m.s.§
`0-73
`077
`0-92
`ool
`r.m.s\l|
`+ Four-letter code for Protein Data Bank (Bernstein et al., 1977)file.
`he erystal structures are taken from the Protein Data Bank files. Residue
`{ Residue numbers for t!
`tructure are according to Kabatet al. (1987).
`numbers for the consensus 5
`§ Root-mean-square deviation in A for (N, C*, C) atoms superimposed on 2FB4.
`|| Root-mean-square deviation in A for (N, C%, C) atoms superimposed on 2HFL.
`
`18-25
`34-39
`46-52
`57-61
`68-71
`78-84
`92-99
`
`Oo
`
`Pfizer Ex. 1021
`Page 7 of 31
`
`Pfizer Ex. 1021
`Page 7 of 31
`
`

`

`
`
`972 C. Eigenbrot et al.
`
`ce.
`Table 3
`Average bond lengths and angles for “average” (before) and energy-minimaz
`consensus (after 50 cycles) structures
`y,
`
`Standard
`=
`hx
`
`Before (A)
`After (A)
`Before (A)
`After (A)
`geometry(A)
`g
`1-449
`—-1-452 (0-004)
`4
`1-451 (0-023)
`1-451 (0-004)
`1-459 (0-012)
`—1-542 (0-005)
`1522
`1-507 (0-033)
`1-523 (0-005)
`1515 (0-012)
`_—_—‘1-231 (0-003)
`1-229
`1-160 (0-177)
`1-229 (0-003)
`1-208 (0-062)
`—_—‘1-335 (0-004)
`1335
`1-282 (0-065)
`1-337 (0-002)
`1-288 (0-049)
`1-530 (0-002)
`1526
`1-499 (0-039)
`1-530 (0-002)
`1-508 (0-026)
`()
`ae
`()
`()
`()
`124-0 (1-1)
`121-9
`125:3 (46)
`123-8 (1-1)
`1235 (4-2)
`C-N-C
`109:5 (1-6)
`1101
`1103 (28)
`109:5 (1-9)
`110-0 (4-0)
`N-C-¢
`1166 (0-8)
`1166
`117-6 (5-2)
`1166 (1-2)
`116-6 (4-0)
`C-CN
`123-3 (0-4)
`1229
`122-2 (49)
`123-4 (0-6)
`123-1 (41)
`0=C-N
`109-8 (0-6)
`1095
`110°6 (2:5)
`109-8 (0-7)
`110:3 (2:1)
`N-C*-CF
`111-1 (0-6)
`LIE-2 (22)
`111-1 (0-7)
`1114 (2-4)
`choc
`Wl
`—_—eenkkesng
`
`N-C?
`Co
`o=C
`C-N
`cue
`
`Values in parentheses are standard deviations. Note that while some bondlength and angle averages
`did not change appreciably after energy minimization,
`the corresponding standard deviations are
`reduced due to deviant geometries assuming standard values after energy minimization. Standard
`geometry values are from the AMBERforce-field (Weiner et al., 1984) as implemented in DISCOVER
`(Biosym Technologies).
`
`tors for each protein, Using the samerules for all immuno-
`globulins results in a consistent assignmentof secondary
`structure. (2) The V, (or Vy) domains were superposed on
`one another using backbone N, C?, and C atom co-ordi-
`nates (INSIGHT, Biosym Technologies). For each eate-
`gory, i.e. V, or Vy, a single crystal structure was chosen as
`the template upon which the others were superposed.
`The residues used for superpositioning the Fabs and r.m.s.
`deviations are given in Table 2. (3) In order to determine
`which positions would be includedin the consensus model,
`the distances from the template C? to the analogous C? in
`the other superposed crystal structures were calculated,
`resulting in a table of C?-C? distances for each amino acid
`position. Generally,
`if all C?-C? distances for a given
`position were <10A (1 A=0-1 nm), that position was
`included in the consensus structure. If, for a given posi-
`tion, only | crystal structure was > 1-0A, the position
`was included but the outlying crystal structure was not
`included in the next step (for this position only).
`In
`general, the 9 #-strands passedthisfilter while some of the
`loops connecting the B-strands, e.g. the CDRs, failed to
`pass. The residues included in the consensus model are
`provided in Table 2. Residue numbering of the consensus
`models conforms to that of Kabat
`ef af.
`(1987). The
`consensus structure was not sensitive to which crystal
`structure was used as the template (Table 2). (4) For each
`amino acid position, the average of the co-ordinates for
`individual N, C7, C, O, and C4 atoms was calculated. Due
`to the averaging procedure. as well as variation in bond
`length, bond angle, and dihedral angle among the crystal
`structures, this ““average’’ structure contained some bond
`
`+ The consensus structure might conceivably be
`dependent upon which crystal structure was chosen as
`the template on which the others were superimposed, As
`a test, for the Vj domain the entire procedure was
`repeated using the crystal structure with the worst
`r.m.s. deviation versus 2FB4, i.e. 2HFL, as the new
`template. The two consensus structures compared
`favorably with an r.m.s. deviation of 0-11 A for all N. C2
`and C atoms.
`
`lengths and angles that deviated from standard geometry.
`To correct this, the “average” structure was su bjected to
`50 cycles of energy minimization (DISCOVER, Biosym
`Technologies) using the AMBER (Weiner et al., 1984)
`force-field with the C® co-ordinates fixed. This restrained
`all bond lengths and angles to a standard geometry;
`statistics for bond geometries prior
`to and following
`energy minimization
`are
`provided
`in
`Table 3.
`(5) Side-chains of conserved residues were added. e.g. the
`conserved disulfide bond linking the 2 f-sheets and the
`conserved tryptophan that lies against this disulfide. For
`these conserved residues (about 60 residues total), the
`most common side-chain conformation found in the Fab
`erystal structures was adopted. The consensus structure
`could then be used to generate models of a variety of
`V,—Vyq structures with a minimum ofeffort. (6) With the
`consensus structure built, the sequence of the mur4D5 VM,
`and Vy domains was incorporated, First, for each CDR
`the sequence of mur4D5 was com pared to the sequences of
`the Fabs in the crystal structures, using the tabulations of
`Chothia et al. (1989) as a guide. Once a CDR category and
`backbone conformation were chosen, that conformation
`was incorporated into the model. Side-chain conforma-
`tions of the mur4D5 sequence were based on information
`from the Fabcrystal structures, rotamerlibraries (Ponder
`& Richards, 1987) and packing considerations. When the
`side-chain was exposed to solvent,
`the rotamer library
`was given more weight.
`Of the 6 CDRs, only CDR-H3 could notbe assigned a
`definite backbone conformation, For this CDR 3 different
`backbone conformations were used: 2 were based on a
`search ofloops of similar size and the 3rd was modeled
`using packing and solvent exposure considerations.
`(7) These 3 models (differing only at CDR-H3) were
`subsequently
`subjected
`to
`energy minimization
`(DISCOVER, Biosym Technologies) (Weineret al., 1984),
`using conjugate gradients, until the maximum derivative
`was <0-05 keal/mol-A (1 cal=4-184J). The f-strand C*
`atomswere tethered to theirinitial positions, using a force
`constant which decreased from 100 keal/A to 10 keal/A
`over 3000 cycles, and then released for another 3000
`
`Pfizer Ex. 1021
`Page 8 of 31
`
`Pfizer Ex. 1021
`Page 8 of 31
`
`

`

`
`
`X-ray Structures of Humanized Antibody Fragments 973
`
`cycles. Hydrogen bond distances between f-strands were
`also constrained for 5000 cycles.
`An alternate energy minimization protocol was applied
`using the CHARMM force-field (Brooksef al., 1983) imple-
`mented in X-PLOR (Briinger, 19906). The electrostatic
`energy term was excluded. The f-strand C* atoms were
`restrained to their initial positions by a force constant
`decreasing from 100 kcal/mol to 0 keal/mol
`in 11 steps.
`Hydrogen bonding distances were restrained in the early
`stages, but these restraints were also removedforthefinal
`cycles.
`
`(b) Modeling hudD5 from mur4D5
`A model of the humanized form of mur4D5 (hu4D5)
`was created by transferring the CDRs from the mur4D5
`model to a human consensus structure. The latter was
`built using the same methodsdescribed for generating the
`mur4D5 co-ordinates. Consensus human sequences used
`for the model were those of human V,« subgroup I and
`human V;,, subgroup IIT. These 2 subgroups are the largest
`subgroups among human sequences compiled by Kabat et
`al. (1987).
`In addition to incorporating the mur4D5 CDRsinto the
`human framework, we noted a number of differences
`between the mur4D5 and human framework sequences
`(Table 4), Each difference was individually investigated
`using the models
`in order
`to ascertain its possible
`influence on CDR conformation and/or binding to the
`pl854882 protein. When such an investigation gave an
`ambiguous result, both the possible amino acids were
`incorporated in different versions of hu4D5 (Carter et al.,
`1992), so that their effect on binding affinity and antipro-
`liferative activity could be assessed directly.
`CDR-LI
`(Table 4)
`is comprised of residues L24-L34
`andits backbone conformation was based on the crystal
`structure of REI (Epp et al., 1975). As pointed out by
`Chothia & Lesk (1987), in the available erystal structures
`the side-chain at position L29 is always hydrophobic and
`is directed in between the 2 f-sheets, thereby anchoring
`CDR-LI to the framework. Several framework side-chains
`interact with the L29 side-chain (residues L2, L25, L33
`and L71) (Chothia & Lesk, 1987). Mur4D6 is somewhat
`unusual
`in that residues L29 and L33, which are most
`often isoleucine or
`leucine in the human and murine
`crystal structures, are both valine in this
`immuno-
`globulin; both were retained as valine in the humanized
`form.
`The conformation of CDR-L2 (residues L50-L56) was
`taken from McPC603 (Satow ef al., 1986) and that of
`CDR-3 (residues L89-L97) was taken from REI (Epp et
`al., 1975). Mur4D5 CDR-L3 includes the sequence Pro
`L95-Pro L96. Proline at L95 is highly conserved but the
`proline at L96is distinctive. Nevertheless, proline could
`be accommodated at position L96 without alteration of
`the REI CDR-L3 conformation.
`The CDR-H1 (residues H26-H35) conformation was
`based oncrystal structures HyHEL-5 (Sheriff ef al., 1987)
`and MePC603 (Satow et al., 1986); the CDR-H2 (res