`
`MODELING OF ANTIBODY COMBINING SITES
`
`3
`
`[1] Modeling of Antibody Combining Sites
`
`By EDUARDO A. PADLAN and ELVIN A. KABAT
`
`Introduction
`
`Antibodies constitute an extremely large family of closely related
`serum proteins, termed immunoglobulins, produced in vertebrates by a
`class of lymphocytes, termed B lymphocytes, which are programmed to
`respond to contact (immunization) with foreign substances (antigens) or,
`under certain circumstances, to antigens of an individual's own tissues
`(autoantibodies). Antibodies may cause the elimination of antigens by
`phagocytosis, neutralization of toxins or viruses, lysis of tissue cells through
`the complement system, precipitation with the antigen used for immuniza(cid:173)
`tion, or clumping of bacteria or red cells containing the antigen or the
`antigen adsorbed to inert particles (agglutination).
`Antibodies may be produced to almost all classes of substances, e.g.,
`proteins, polysaccharides, nucleic acids, and to more complex particles,
`e.g., pollens, infectious agents, viruses, and tissue cells. Unless there is
`some structural similarity between two antigens, one will generally . not
`react with antibody to the other; when structural similarity does exist,
`reactions with both will occur (termed cross-reactions), the antibody react(cid:173)
`ing more strongly to the antigen used for immunization and to a lesser
`degree with the other. (For exceptional, poorly understood, anomalous
`cross-reactions, see Future Prospects and Problems.)
`A second class of proteins, termed T cell receptors for antigen, is
`produced by a different class of lymphocytes, termed T lymphocytes. The
`T cell receptors for antigen do not circulate in the bloodstream but remain
`attached to the cells which synthesize them and, in conjunction with
`proteins of the major histocompatibility complex and cells of the macro(cid:173)
`phage system, can destroy tissue cells infected with viruses, tumor cells, etc.
`This is termed the cellular arm of the immune system.
`Antibodies and T cell receptors for antigen show certain structural as
`well as certain genetic similarities, but appear to have arisen at different
`evolutionary periods and by different pathways. This chapter on modeling
`will focus exclusively on the antigen-binding sites of antibodies (the anti(cid:173)
`body combining sites).
`The high degree of specificity of an antibody for the antigen used for
`immunization and the wide range of specificities that the immune system
`is capable of generating have been the subject of numerous investigations.
`The specificity of antibody:antigen interactions is accepted as being due to
`
`METHODS IN ENZYMOLOGY, VOL. 203
`
`Copyright e 1991 by Aclldemic Prea. Inc.
`All rights of reproduction in any form raerval.
`
`1 of 19
`
`BI Exhibit 1099
`
`
`
`4
`
`ANTIBODIES AND ANTIGENS
`
`[1]
`
`the complementarity of the antibody combining site structure and that of
`the antigenic determinant. The diversity of antigen-binding specificities is
`then due to variation in the combining site topography brought about by
`variation in primary and three-dimensional structure.
`Recognition of the central importance of the antibody combining site
`in immune function has prompted intensive study of the primary and
`three-dimensional structures of antibodies. Considerable three-dimen(cid:173)
`sional information has become available from X-ray crystallography,1-9
`although the number of antibody structures that will be elucidated by
`X-ray analysis can be only a very small fraction of the total number of
`different antibodies that higher organisms can produce. Other techniques
`are needed in the study of antibody combining sites and one that can make
`a significant contribution is modeling.
`Here, we review the various modeling procedures that have been ap(cid:173)
`plied to antibodies, evaluate the success of these procedures in predicting
`combining site structures, and discuss potential improvements and prob(cid:173)
`lems.
`
`Structural Background
`Antibodies are multimers of a basic unit that consists of four polypep(cid:173)
`tide chains identical in pairs (Fig. 110). There are two light (L) chains of
`about 220 amino acids and two heavy (H) chains of 450-575 amino acids.
`Both L and H chains are made up of regions of sequence homology
`(domains) of about 100-120 residues. There are two such domains in the
`L chain and four or five in the H chain. The N-terminal domains of both L
`and H chains are variable, i.e., they differ in sequence from antibody to
`antibody; the other domains are constant, i.e., they are the same in anti(cid:173)
`body chains of the same type except for single amino acid differences at a
`few positions. The variable domains of the L and H chains, V L and V H,
`
`1 R. J. Poljak, Adv. Immunol. 21, 1 (1975).
`2 D.R. Davies, E. A. Padlan, and D. M. Segal, Annu. Rev. Biochem. 44, 639 (1975).
`3 R. Huber, Trends Biochem. Sci. 1, 87 (1976).
`4 E. A. Padlan, Q. Rev. Biophys. 10, 35 (1977).
`s L. M. Amzel and R. J. Poljak, Annu. Rev. Biochem. 48, 961 (1979).
`6 D.R. Davies and H. M. Metzger, Annu. Rev. Immuno/. 1, 87 (1983).
`7 P. M. Alzari, M.-B. Lascombe, and R. J. Poljak, Annu. Rev. Immunol. 6, 555 (1988).
`8 P. M. Colman, Adv. Immunol. 43, 99 (1988).
`9 D. R. Davies, E. A. Padlan, and S. Sheriff, Annu. Rev. Biochem. 59, 439 ( 1990).
`10 E. A. Kabat. T. T. Wu, M. Reid-Miller, H. M. Perry, and K. S. Gottesman, ••Sequences of
`Proteins of Immunological Interest," 4th Ed. U.S. Department of Health and Human
`Services, Washington, D.C., 1987.
`
`2 of 19
`
`BI Exhibit 1099
`
`
`
`[1]
`
`MODELING OF ANTIBODY COMBINING SITES
`
`5
`
`so
`
`Gm( a-)
`Gt., 356
`Met 358
`36711'1 c 3
`446 425V> H
`
`Fe
`
`FIG. 1. Schematic representation of the four-chain structure of human lg01 molecule. The
`numbers on the right-hand side denote the actual residues of protein Eu [G. M. Edelman, B.
`A. Cunningham, W. E. Gall, P. D. Gottlieb, U. Rutishauser, and M. J. Waxdal, Proc. Natl.
`Acad. Sci. U.S.A. 63, 78 (1969)). The numbers of the Fab fragments on the left side are
`aligned for maximum homology: light and heavy chains are numbered according to T. T. Wu
`and E. A. Kabat, J. Exp. Med. 132, 211 (1970), and E. A. Kabat and T. T. Wu, Ann. N.Y.
`Acad. Sci. 190, 382 (1971). The heavy chains of Eu have residues 52A and 82A, B, C but lack
`the residues termed lOOA-K and 35A, B. Thus, residue 110 (the end of the heavy-chain
`variable region) is 114 in the actual sequence. Hypervariable or complementarity-determin(cid:173)
`ing regions are shown by heavier lines. V L and V H denote the light- and heavy-chain variable
`regions; CHI, CH2, and CH3 are domains of the constant region of the heavy chain; CL is the
`constant region of the light chain. The hinge region, in which the two heavy chains are linked
`by disulfide bonds, is indicated approximately. The attachment of carbohydrate is at position
`297. Arrows at positions 107 (in the light chain) and 110 (in the heavy chain) denote
`transition from variable to constant region. The sites of cleavage by papain and pepsin and
`the locations of various allotypic genetic factors [Gm( ft), Gm(a-), Inv'>] are indicated.10
`
`3 of 19
`
`BI Exhibit 1099
`
`
`
`6
`
`ANTIBODIES AND ANTIGENS
`
`[l]
`
`display significant sequence similarity, as do the constant domains10; there
`is no obvious sequence similarity between variable and constant domains.
`The site on the antibody molecule that binds to antigen is formed by the
`association of V L and V H, the Fv module. The fragment formed by the
`association of the L chain and the two N-terminal domains of the H chain
`is called the Fab, or antigen-binding, fragment (Fig. 1 ). The antigen-bind(cid:173)
`ing properties of an antibody are determined entirely by the variable
`domains; indeed, a chimeric structure, in which the variable domains have
`been linked to the constant domains of the heterologous chains, was shown
`to display exactly the same ligand-binding characteristics. 11
`A comparison of the sequences of variable domains from a variety of
`immunoglobulins revealed the existence of regions of hypervariability,
`three each in the L and H chains. 12
`•13 These hypervariable regions were
`predicted by Wu and Kabat12 to fold up three-dimensionally to form the
`walls of the antigen-binding site of the antibody (the antibody combining
`site) several years before any X-ray structures had been determined.
`X-Ray crystallographic studies have allowed the direct visualization of
`
`the three-dimensional structure of complete antibodies, 14•15 although only
`at low resolution. Structures at high resolution, however, have become
`available for a number of antibody fragments, including several Fabs,
`which bear the antigen-binding sites. These studies have revealed the high
`degree of structural similarity of the homologous domains and, as would be
`expected, the similarity in the three..dimensional structure was found to
`parallel the similarity in amino acid sequence. Thus, the V L and V 8
`domains were found to have very similar structures, as were the constant
`domains; constant and variable domains, on the other hand, share only a
`18 The hy(cid:173)
`basic tertiary structure and appear to be rotational isomers. 16
`-
`pervariable segments were found to exist mainly as loops that are for the
`most part exposed and located at one end of the variable domains.
`In spectacular confirmation of the prediction of Wu and Kabat, 12 the
`
`11 T. Simon and K. Rajewsky, EMBO J. 9, 1051 (1990).
`12 T. T. Wu and E. A. Kabat, J. Exp. Med. 132, 211 (1970).
`13 E. A. Kabat and T. T. Wu, Ann. N.Y. Acad. Sci. 190, 382 (1971).
`14 E. W. Silverton, M.A. Navia, and D.R. Davies, Proc. Natl. Acad. Sci. U.S.A. 74, 5140
`(1977).
`15 S. S. Rajan, K.. R. Ely, E. E. Abola, M. K.. Wood, P. M. Colman, R. J. Athay, and A. B.
`Edmundson, Mol. lmmunol. 20, 797 (1983).
`16 R. J. Poljak, L. M. Amzel, H. P. Avey, B. L. Chen, R. P. Phiz.ackerly, and F. Saul, Proc.
`Natl. Acad. Sci. U.S.A. 70, 3305 (1973).
`17 M. Schiffer, R. L. Girling, K.. R. Ely, and A. B. Edmundson, Biochemistry 12, 4620 (1973).
`18 A. B. Edmundson, K. R. Ely, E. E. Abola, M. Schiffer, and N. Panagiotopoulos, Biochemis(cid:173)
`try 14, 3953 ( 1975).
`
`4 of 19
`
`BI Exhibit 1099
`
`
`
`[l)
`
`MODELING OF ANTIBODY COMBINING SITES
`
`7
`
`hypervariable regions were indeed found to form a continuous surface at
`the tip of the Fabs. 16
`19 Furthermore, crystallographic studies of complexes
`•
`of Fabs with specific ligands unequivocably established the essential iden(cid:173)
`tity of the hypervariable surface with the antibody combining site. It is the
`chemical nature of this surface that determines the particular specificity of
`an antibody and the affinity with which it binds to its specific ligand.
`Furthermore, the insertions and deletions that are frequently found in the
`hypervariable regions, together with the variation in the amino acid resi(cid:173)
`dues in these regions, result in the variability of the combining site topogra(cid:173)
`phy, thus providing an obvious structural basis for the wide diversity of
`antigen-binding specificities. In view of the undisputed importance of the
`hypervariable regions in the binding interaction with the antigen, these
`regions are now also called complementarity-determining regions (CDRs).
`The comparison of the three-dimensional structures of variable do(cid:173)
`mains from different antibodies reveals that the nonhypervariable or
`framework regions of these domains are essentially superimposable, so that
`the structural variation is mainly confined to the hypervariable seg(cid:173)
`ments. 20.21 In addition to the close similarity in the tertiary structures of the
`homologous domains, it had also been found that the mode of association
`of the paired domains is essentially invariant. 6·22.23 Thus the antibody
`combining site can be viewed as being formed by a small number of
`segments of variable structure grafted onto a scaffolding of essentially
`invariant architecture.
`Even the hypervariable regions have been found to display a high
`degree of structural similarity, so that canonical structures have been
`advanced for several of the CDRs. For example, it was found that hyper(cid:173)
`variable regions with the same number of residues, especially those with
`significant sequence similarity, have very similar backbone conforma(cid:173)
`tions.4·20·21·24 Furthermore, Kabat et al.25 noted that certain positions in the
`CDRs did not vary but were conserved and they suggested that those
`residues play a structural role. Indeed, structural comparisons of known
`CDR structures have shown that there is a small repertoire of main-chain
`conformations for at least five of the six CDRs and that the particular
`
`19 D. M. Segal, E. A. Padlan, G. H. Cohen, S. Rudikoff, M. Potter, and D. R. Davies, Proc.
`Natl. Acad. Sci. U.S.A. 71, 4298 (1974).
`20 E. A. Padlan and D.R. Davies, Proc. Natl. Acad. Sci. U.S.A. 72, 819 (1975).
`21 C. Chothia and A. M. Lesk, J. Mo/. Biol. 196, 901 (1987).
`22 J. Novotny and E. Haber, Proc. Nall. Acad. Sci. U.S.A. 82, 4592 (1985).
`23 C. Chothia, J. Novotny, R. Bruccoleri, and M. Karplus, J. Mo/. Biol. 186, 651 (1985).
`24 P. de la Paz, B. J. Sutton, M. J. Darsley, and A. R. Rees, EMBO J. 5, 415 (1986).
`25 E. A. Kabat, T. T. Wu, and H. Bilofslcy, J. Biol. Chem. 252, 6609 (1977).
`
`5 of 19
`
`BI Exhibit 1099
`
`
`
`8
`
`ANTIBODIES AND ANTIGENS
`
`[1]
`
`conformation adopted is determined by a few key conserved residues,
`several of which are outside the CDRs.21
`•26
`It is this high degree of structural similarity among the immunoglobulin
`domains that has been the stimulus for the modeling of antibody combin(cid:173)
`ing site structures.
`
`Techniques for Modeling Antibody Combining Site Structures
`All current attempts at modeling antibody combining sites assume the
`same framework structures for V L and V 8 , as well as the same quaternary
`association of these domains, as found in the crystal structures. With.
`regard to the modeling of the CDRs, two approaches have been taken. One
`approach uses a known CDR or other loop structure as a template (homol(cid:173)
`ogy or knowledge-based modeling), the other attempts to predict the CDR
`structures solely on the basis of energetic considerations (ab initio model(cid:173)
`ing). An automated procedure that combines both approaches has been
`developed. 27
`
`Homology Modeling of Antibody Combining Sites
`The first structure-based modeling of an antibody combining site was
`done by Padlan et al.,28 who built a model of the Fv of MOPC315. In
`constructing the model the following assumptions were made: ( 1) the
`framework structures of the V L and the V 8 domains of MOPC315 were
`essentially the same as those of the corresponding domains of the
`McPC603 Fab, the structure of which had been previously determined
`crystallographically19; (2) the quaternary association of the V L and V 8
`domains of MOPC315 was the same as that found in McPC603; and (3)
`the CDRs ofMOPC315 would have the same backbone conformations as
`those CDRs from other structures which had the same number of amino
`acid residues. Sequence similarity was taken into account in building loop
`structures for which no suitable starting model was available.
`The same basic procedure was used in the modeling of the combining
`site structures of the rabbit BS-5 anti-polysaccharide antibody,29 the inu-
`
`26 C. Chothia, A. M. Lesk, M. Levitt, A.G. Amit, R. A. Mariuzza, S. E. V. Phillips, and R. J.
`Poljak, Science 233, 7 55 ( 1986).
`27 A. C.R. Martin, J.C. Cheetham, and A. R. Rees, Proc. Natl. Acad. Sci. U.S.A. 86, 9268
`(1989).
`28 E. A. Padlan, D. R. Davies, I. Pecht, D. Givol, and C. Wright, Cold Spring Harbor Symp.
`Quant. Biol. 41, 627 ( 1976).
`29 D. R. Davies and E. A. Padlan, in "Antibodies in Human Diagnosis and Therapy" (E.
`Haber and R. M. Krause, eds.), p. 119. Raven, New York, 1977.
`
`6 of 19
`
`BI Exhibit 1099
`
`
`
`[1]
`
`MODELING OF ANTIBODY COMBINING SITES
`
`9
`
`!in-binding EPC 109 mouse myeloma protein, 30 and the WJ 129 and 19 .1.2
`mouse anti-a(l--+ 6)-dextrans.31 Very similar assumptions were used by
`Feldmann and co-workers32•33 in their modeling of the Fv of the galactan(cid:173)
`binding mouse immunoglobulin J539, by Smith-Gill et al. 34 in their model(cid:173)
`ing of the Fv of the lysozyme-specific HyHEL-10 antibody, and by de la
`Paz et al.,24 who built models of three antibodies specific for an antigenic
`disulfide loop of lysozyme.
`26
`With the demonstration by Chothia and co-workers21
`35 that canoni(cid:173)
`•
`•
`cal structures exist for most of the hypervariable loops, CDR templates for
`modeling have been chosen on the basis of size as well as on the presence of
`the residues believed to be responsible for the different conformations. It
`should be pointed out that the hypervariable loops, as defined by Chothia
`and co-workers, overlap but do not entirely correspond to those defined by
`Kabat et al. 10 on the basis of sequence variation.
`An automated approach to the homology modeling of antibody Fvs has
`been successfully implemented by Levitt and co-workers. 26•36•37 In this
`approach, the sequence for which a model is desired is divided into a
`number of segments, homologous regions from the known immunoglobu(cid:173)
`lin structures are examined, and a search is made for segments that are
`most similar in size and sequence. The model is then pieced together using
`the most similar regions while inserting or deleting residues at positions
`chosen to maximize sequence similarity. Missing atoms and inserted resi(cid:173)
`dues are introduced using randomized coordinates; these are then properly
`incorporated into the model and breaks in the main chain rectified by
`limited energy minimization.
`
`Ab Initio Modeling of Antibody Combining Sites
`
`The first attempt to predict CDR structures ab initio was made by
`Stanford and Wu,38 who constructed a model of the backbone structure of
`
`30 M. Potter, S. Rudikoff, E. A. Padlan, and M. Vrana, in "Antibodies in Human Diagnosis
`and Therapy" (E. Haber and R. M. Krause, eds.), p. 9. Raven, New York, 1977.
`31 E. A. Padlan and E. A. Kabat, Proc. Natl. Acad. Sci. U.S.A. 85, 6885 (1988).
`32 R. J. Feldmann, M. Potter, and C. P. J. Glaudemans, Mo/. Immunol. 18, 683 (1981).
`33 C.R. Mainhart, M. Potter, and R. J. Feldmann, Mo/. lmmunol. 21, 469 (1984).
`34 S. J. Smith-Gill, C. Mainhart, T. B. Lavoie, R. J. Feldmann, W. Drohan, and B. R. Brooks,
`J. Mo/. Biol. 194, 173 (1987).
`35 C. Chothia, A. M. Lesk, A. Tramontano, M. Levitt, S. J. Smith-Gill, G. Air, S. Sheriff, E.
`A. Pad.Ian, D. Davies, W. R. Tulip, P. M. Colman, S. Spinelli, P. M. Alzari, and R. J.
`Poljak, Nature (London) 342, 877 ( 1989).
`36 J. Anglister, M. W. Bond, T . Frey, D . Leahy, M. Levitt, H. M. McConnell, G. S. Rule, J.
`Tomasello, and M. Whittaker, Biochemistry 26, 6058 (1987).
`37 R. Levy, 0. Assulin, T. Scherf, M. Levitt, and J. Anglister, Biochemistry 28, 7168 (1989).
`38 J.M. Stanford and T. T. Wu, J Theor. Biol. 88, 421 (1981).
`
`7 of 19
`
`BI Exhibit 1099
`
`
`
`10
`
`ANTIBODIES AND ANTIGENS
`
`[1]
`
`the combining site of MOPC3 l 5 on the basis of amino acid sequence and
`steric considerations. These authors assumed the same framework struc(cid:173)
`ture for MOPC315 as that found in Fab NEW39 and used the predictive
`method of Kabat and Wu40 to construct models of the CDRs. Backbone
`dihedral angles for tripeptides were obtained from known protein struc(cid:173)
`tures, relying mostly on P-sheet proteins, and these angles were imposed on
`the segment being constructed based on its sequence. The peptide angles
`were allowed to vary from the initial values by as much as 30° in 5°
`intervals. The large number of resulting possible structures was reduced by
`imposing the conditions that the ends of the modeled CDRs should fit onto
`the assumed framework structure and that nonbonded atoms should not
`come within allowed minimum contact distances based on normally ac(cid:173)
`cepted van der Waals radii.
`A related procedure was used by Bruccoleri et al.41 to reconstruct the
`crystallographically determined hypervariable loops of McPC603 and
`HyHEL-5. These authors generated all the possible conformations for each
`loop by imposing peptide dihedral angles that are energetically accessible
`to each residue in the sequence, sampling torsional space at 30° intervals.
`Here also, the number of possible structures was limited by imposing the
`conditions that the ends of the loops should fit onto the framework and
`that bad steric contacts are avoided.
`A different approach was taken by Fine et al., 42 who modeled four of
`the CDRs of McPC603. These authors generated a large number of ran(cid:173)
`dom conformations for the backbone of those CDRs, subjecting the con(cid:173)
`formations to molecular dynamics and energy minimization and then
`selecting those with lowest energy for further characterization of side-chain
`conformations. The random structures generated were required to fit onto
`the framework with correct geometry. In generating the conformations,
`random values were first assigned to each <P and f/I peptide angle along the
`backbone; then these angles were adjusted as minimally as possible in an
`iterative procedure to produce the desired fixed-end conditions. Side
`chains were then added to the lowest energy structures and energetically
`favorable conformations were obtained by varying side-chain torsional
`angles.
`
`Combined Knowledge-Based and ab lnitio Methods
`The approach taken by Martin et al. 21 makes use of segments from all
`known structures (not restricted to antibody structures) as the database
`
`39 F. A. Saul, L. M. Amzel, and R. J. Poljak, J. Biol. Chem. 253, 585 (1978).
`40 E. A. Kabat and T. T. Wu, Proc. Natl. Acad. Sci. U.S.A. 69, 960 (1972).
`41 R. E. Bruccoleri, E. Haber, and J. Novotny, Nature (London) 335, 564 (1988).
`42 R. M. Fine, H. Wang, P. S. Shenkin, D. L. Yarmush, and C. Levinthal, Proteins: Struct.
`Funct. Genet. 1, 342 ( 1986).
`
`8 of 19
`
`BI Exhibit 1099
`
`
`
`[1]
`
`MODELING OF ANTIBODY COMBINING SITES
`
`11
`
`from which to choose the most similar segment for the CDR loop of
`interest. A conformational search procedure41·43 is then used to obtain the
`lowest energy conformations. Each conformation is then screened and the
`one with the smallest. hydrophobic exposed area is selected. This algorithm
`has been implemented as a completely automated procedure. 27
`
`Evaluation of Predictive Methods
`
`The assessment of the success of modeling algorithms is not straightfor(cid:173)
`ward, since crystal structures are not always available to allow the compari(cid:173)
`son of model with X-ray structure. Even when crystal structures are avail(cid:173)
`able, these are frequently of too-low accuracy themselves to permit
`meaningful comparisons. Indeed, most of the X-ray analyses of combining
`site structures had been done at medium resolution, i.e., between 2 and 3
`A. In these analyses, estimates of the error in atomic position in the most
`ordered regions, i.e., the most accurately determined parts of the molecule,
`run to about 0.5 A; the error in the loop regions and for the side-chain
`atoms would be much larger. Thus, if a model agrees with the X-ray
`structure to within 1.5 A, the agreement can be said to be within the limits
`of error. In fact, higher levels of disagreement could be tolerated since the
`CDR structures, being mostly loops, could be suffering from larger errors.
`Indeed, it is not unusual that a major rebuilding of external regions is
`required when a crystallographic analysis is extended to higher resolution.
`Thus, for example, when the crystal structure of J539 Fab was extended
`from a resolution of 2.6 A44 to l.95,45 large differences were noted in the
`CD Rs. In one CDR, differences between corresponding C0 positions of as
`much as 8 A were observed.
`There are several cases where models for antibody combining sites can
`be compared with crystallographically determined structures. Crystal
`structures are available for the Fahs of McPC60J,46 J539,44·45 Dl.J,47
`NC41,48 HyHEL-5,49 HyHEL-10,50 and Gloop2 (unpublished results re(cid:173)
`ferred to by Martin et al. 27) for which models of the combining sites or of
`
`43 R. E. Bruccoleri and M. Karplus, Biopolymers 26, 137 (1987).
`44 S. W. Suh, T. N. Bhat, M. A. Navia, G. H. Cohen, D. N. Rao, S. Rudikoff, and D. R.
`Davies, Proteins: Struct. Funct. Genet. 1, 74 (1986).
`45 T. N. Bhat, E. A. Padlan, and D.R. Davies, in preparation.
`46 Y. Satow, G. H. Cohen, E. A. Padlan, and D.R. Davies, J. Mo/. Biol. 190, 593 (1986).
`47 A.G. Amit, R. A. Mariu.u.a, S. E. V. Phillips, and R. J. Poljak, Science 233, 747 (1986).
`48 P. M. Colman, W. G. Laver, J. N. Varghese, A. T. Baker, P.A. Tulloch, G. M. Air, and R.
`G. Webster, Nature (London) 326, 358 (1987).
`49 S. Sheriff, E.W. Silverton, E. A. Padlan, G. H. Cohen, S. J. Smith-Gill, B. C. Finzel, and D.
`R. Davies, Proc. Natl. Acad. Sci. U.S.A. 84, 8075 (1987).
`so E. A. Pad.Ian, E.W. Silverton, S. Sheriff. G. H. Cohen, S. J. Smith-Gill, and D.R. Davies,
`Proc. Natl. Acad. Sci. U.S.A. 86, 5938 (1989).
`
`9 of 19
`
`BI Exhibit 1099
`
`
`
`12
`
`ANTIBODIES AND ANTIGENS
`
`[1]
`
`some of the CDRs had been constructed. When the models are compared
`to the crystal structures, some of the CDR structures are found to have
`been, by and large, correctly predicted, while in others large disagreements
`between model and X-ray structure are found.
`For example, comparison of the crystal structure of the J539 Fv44 with
`the model ofMainhart et al.33 showed that the Ca root-mean-square devia(cid:173)
`tions for individual CDRs ranged from 1.1 to 4.0 A. When the modeled
`hypervariable loops ofDl.3 were compared with the crystal structure,47 the
`deviations of the model based on structural analysis ranged from 0.5 to
`2.07 A, and the deviations of the models built using conformational energy
`calculations ranged from 0.47 to 3.76 A.26 Comparing the model con(cid:173)
`structed for the Fv ofHyHEL-1034 with the X-ray structure, so deviations in
`Ca positions in the CDRs ranged from 0.44 to 4.27 A.51
`The use of canonical structures for CD Rs in the modeling of combining
`site structures has been shown to be reasonably successful. 3' When models
`of five out of the six hypervariable loops ofHyHEL-5, HyHEL-10, NC41,
`and NQ 10 antibodies were compared to the crystal structures, the differ(cid:173)
`ences between the predicted and observed structures initially ranged from
`0.4 to 1.2 A in the Ca positions, with the larger errors being found in the
`predicted loops of HyHEL-5 and NC41. When better refined crystal struc(cid:173)
`tures became available for inclusion in the database and after the structures
`of the HyHEL-5 and NC4 l hypervariable loops had been remodeled on
`the basis of these new data, the differences between predicted and observed
`ranged from 0.4 to 3.5 A for the four models.35
`The structures predicted ab initio for the hypervariable loops of
`McPC603 and HyHEL-5 by Bruccoleri et al.41 matched the X-ray struc(cid:173)
`tures fairly well. The backbone differences between predicted and observed
`structures ranged from 0.7 to 2.6 A (1.7 for all loops) for the six hypervari(cid:173)
`able loops of McPC603 and from 0.6 to 2.1 A (1.4 for all loops) for
`HyHEL-5. The predictions of Fine et al.42 of the hypervariable loop struc(cid:173)
`tures for McPC603 also yielded very reasonable results, with root-mean(cid:173)
`square deviations between crystal and modeled structures of about 1.0 A
`for backbone atoms.
`
`Future Prospects and Problems
`
`Possible Improvements
`The use of hypervariable loops from known crystal structures as tem(cid:173)
`plates in knowledge-based modeling studies emphasizes the need for more,
`
`s1 E. A. Padlan, unpublished results.
`
`10 of 19
`
`BI Exhibit 1099
`
`
`
`[1]
`
`MODELING OF ANTIBODY COMBINING SITES
`
`13
`
`and especially more accurate, combining site structures. The relative ease
`with which Fab structures are now being determined crystallographi(cid:173)
`cally52·53 could satisfy this need soon. The use of loops from proteins other
`than immunoglobulins27 could fulfill this need even sooner.
`Ab initio methods have already produced models that are in reasonable
`agreement with crystal structures. However, in view of the prohibitive
`computer time required, those methods have been restricted so far to only
`the shorter hypervariable segments. The advent of more powerful com(cid:173)
`puters will remove this restriction. Eventually, the framework regions,
`which have been largely ignored, can also be included in the computations
`to obtain a more complete and more realistic situation.
`In those cases where the structural consequences of only a single amino
`acid replacement is sought, 54·55 the magnitude of the problem is greatly
`reduced and more reliable results can be expected.
`All models can benefit from further refinement using molecular dy(cid:173)
`namics. 27·56 This procedure is still somewhat limited, however, in that the
`relative weights given to the various potential terms used in the calcula(cid:173)
`tions are usually semiempirically assigned. Other limitations arise from the
`inadequate description of the electrical properties of the protein interior
`and of the protein environment.57•58 Improvements along these lines
`should result in better prediction of structures.
`Table I shows the combining site residues which have been found to be
`involved in the contact with the ligand in the antibody-ligand complexes
`of known structure. As further additions to Table psa-c are made, each
`CDR will be further defined structurally. This information will be ex-
`
`52 M. Cygler, A. Boodhoo, J. S. Lee, and W. F. Anderson, J. Biol. Chem. 262, 643 (1987).
`53 S. Sheriff, E. A. Padlan, G. H. Cohen, and D.R. Davies, Acta Crystallogr., Sect. B 46, 418
`(1990).
`54 M. E. Snow and L. M. Amzel, Proteins: Struct. Funct. Genet. 1, 267 ( 1986).
`55 N. V. Chien, V. A. Roberts, A. M. Giusti, M. D. Scharff, and E. D. Getzoff, Proc. Natl.
`Acad. Sci. U.S.A. 86, 5532 ( 1989).
`56 L. Holm, L. Laaksonen, M. Kaartinen, T. T. Teeri, and J. K. C. Knowles, Protein Eng. 3,
`403 (1990).
`57 M. K. Gilson and B. H. Honig, Proteins: Struct. Funct. Genet. 3, 32 ( 1988).
`58 J. J. Wendoloski and J.B. Matthew, Proteins: Struct. Funct. Genet. S, 313 (1989).
`ssa L. M. Amzel, R. J. Poljak, F. Saul, J.M. Varga, and F. F. Richards, Proc. Natl. Acad. Sci.
`U.S.A. 71, 1427 (1974).
`sab E. A. Padlan, G. H. Cohen, and D. R. Davies, Ann. Inst. Pasteur/lmmunol. 136C, 271
`(1985).
`sac J. N. Herron, X.-M. He, M. L. Mason, E.W. Voss, Jr., and A. B. Edmundson, Proteins:
`Struct. Funct. Genet. S, 271 ( 1989).
`8d R. L. Stanfield, T. M. Fieser, R. A. Lerner, and I. A. Wilson, Science 248, 712 (1990).
`5
`sae W. R. Tulip, J. N. Varghese, R. G. Webster, G. M. Air, W. G. Laver, and P. M. Colman,
`Cold Spring Harbor Symp. Quant. Biol. 54, 239 (1989).
`
`11 of 19
`
`BI Exhibit 1099
`
`
`
`FR2
`-
`49
`
`so
`
`Tyr Tyr
`Asp
`Tyr
`
`Gin
`
`His
`
`Tyr
`Asn Tyr
`Gly Asn Asn
`Gly Asn
`
`His
`
`Asp
`
`Asp
`
`Tyr
`Tyr
`
`Arg
`
`Tyr Trp
`
`Tyr
`Ser 1br
`
`His De
`
`Heavy-chain residues
`
`Dl.3
`HyHEJ.....S
`HyHEJ.....10
`NEWM
`McPC603
`4-4-20
`81312
`NC41
`
`VH
`
`TABLE I
`COMBINING SITE RESIDUES IN CONT ACT WITH THE LIGAND IN COMPLEXES OF KNOWN STRUCTURE"'
`
`Light-chain residues
`
`VL
`
`242S2627ABC D E F
`
`28
`
`29
`
`30
`
`31
`
`32
`
`33 34
`
`CDRl
`
`CDR2
`
`CDR3
`
`51
`
`S2
`
`S3
`
`S4
`
`SS
`
`S6 89 90
`
`91
`
`92
`
`93
`
`94
`
`9S
`
`96
`
`97
`
`Pbc Trp Ser
`Trp Gly Arg
`Ser Asn
`
`Pro
`
`Tyr
`
`Tyr
`
`Asp
`Ser
`Gly
`
`Ser Leu
`Tyr Pro Leu
`Trp
`Pro
`Trp
`
`Val
`Tyr Ser Pro
`
`FRI
`
`FR2
`-
`30 31 32 33 34 3S 3S AB 47 SO Sl S2 A B C 53 S4 SS
`
`CDRl
`
`CDR2
`
`56
`
`51 58 59 60 61 62 63 64 6S 9S 96 91 98 99 100 A B
`
`CDR3
`c DEFGHIJK 101102
`
`1br Gly Tyr
`Dl.3
`Trp Glu
`HyHEJ.....S
`HyHEJ.....10 1br Ser Asp Tyr
`NEWM
`McPC603
`4-4-20
`81312
`NC41
`
`Arg
`Asn Tyr
`
`Tyr
`Trp
`Ala
`
`Trp Gly Asp
`
`Trp Glu
`Tyr
`
`Trp
`
`Ser
`Tyr
`Arg
`
`Ser
`Tyr Ser
`
`Ser 1br Asn
`Tyr
`Ser
`1br
`
`Arg Asp Tyr Alg
`Tyr
`
`Leu Ile Ala
`
`Gly
`Trp
`
`Asn
`
`Trp
`
`De Ser Ser
`Asn
`
`Gly
`Asn
`
`Ser Tyr
`
`Pbc
`
`Tyr
`
`Pro Pbc
`Glu Asp Asn Pbc
`
`Ser Leu
`
`•The contacts for Dl.3 are from Amit et al.,41 HyHEJ.....S from Sbaift' et a/.,49 HyHEI.-10 from Padlan et al.,'°NEWM from Am7.Cl et al .• "" McPC603 from PadJan et al .• • 4-4-20 from Hcnon et
`al.,"" Bl 312 from Stanfield et al.,"" and NC41 from Tulip et al.,.
`
`12 of 19
`
`BI Exhibit 1099
`
`
`
`[1]
`
`MODELING OF ANTIBODY COMBINING SITES
`
`15
`
`tremely useful in the prediction of which residues are probably exposed
`and available for interaction with the antigen.
`Verification of the modeled binding site structures can be provided by
`techniques, like two-dimensional nucl