`
`25. Oiiier, C. D. Tectonics and Landforms (Longman. New York. 1981)
`26. Fitch. F. J & Miller. J A. Spec. Pubis geof. Soc. s Afr 13, 247 - 266 (1984)
`27. Cox. K. G. J. Petrol. 21, 629- 650 (1980).
`28. England, P. C. & Molnar. P. Geology (in the press)
`29. Petri. S. & Fulfar6. V. J. Geo/ogia do Brasil (Editora da Universidade de S8o Paulo. 1983)
`30. Tankard. A. J. et al. Crustal Evolution of Southern Africa (Springer. New Yori<. 1982).
`31 . Craddock, C. (ed.) Antarctic Geoscience (University of Wisconsin Press. Madison. 1982)
`
`32. King. L. C. South African Scenery 2nd edn (Oliver and Boyd, Edinburgh, 1951).
`33. King, L. C. The Morphology of the Earth (Oliver and Boyd, Edinburgh, 1962).
`34. Partridge, T. C. & Maud, R R. S. Afr. geo/. J 90, 179- 208 (1987)
`35. Cahen. L., Snelling, N. J. , Delhal, J. & Vail, J. R. The Geochronology and Evolution of Africa (Clarendon,
`Oxford. 1984).
`36. Brown. R. W. 6th Int. Cont. Fission Track Dating Abs tr . Vol. Universite de Franche-Com te. Besarn;:on.
`1988)
`
`Conformations of immunoglobulin
`hypervariable regions
`Cyrus Chothia*t, Arthur M. Lesk**, Anna Tramontano*, Michael Levitt*§,
`Sandra J. Smith-Gillll, Gillian Air11, Steven Sheriff#**, Eduardo A. Padlan#,
`David Davies#, William R. Tuliptt, Peter M. Colmantt, Silvia Spinellitt,
`Pedro M. Alzaritt & Roberto J. Poljaktt
`* MRC Laboratory of Molecular Biology, Hills Road, Cambridge CB2 2QH, UK
`:j: European Molecular Biology Laboratory, Meyerhofstrasse 1. Postfach 1022.09, D-6900 Heidelberg, FRG
`§ Department of Cell Biology, Stanford University Medical School, Stanford. California 94305, USA
`t Christopher Ingold Laboratory. University College London. 20 Gordon Street. London WC1H OAJ, UK
`II National Institute of Cancer and # National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda,
`Maryland 20892, USA
`,.; Department of Microbiology, University of Alabama, Birmingham, Alabama 35294, USA
`tt CSIRO Division of Biotechnology, 343 Royal Parade. Parkville 3052. Australia
`++ Unite d'lmmunologie Structurale, Departement d'lmmunologie, lnstitut Pasteur, 25 rue du Dr Roux. 75724 Paris, France
`
`On the basis of comparative studies of known
`antibody structures and sequences it has been
`argued that there is a small repertoire of main(cid:173)
`chain conformations for at least five of the six
`hypervariable regions of antibodies, and that the
`particular conformation adopted is determined by
`a few key conserved residues. These hypotheses
`are now supported by reasonably successful pre(cid:173)
`dictions of the structures of most hypervariable
`regions of various antibodies, as revealed by
`comparison with their subsequently determined
`structures.
`
`the amino-acid sequences of
`relationships between
`THE
`immunoglobulins and the structures of their antigen-binding
`sites are important for understanding the molecular mechanisms
`of the generation and maturation of the immune response and
`for designing engineered antibodies. Antigen-binding sites are
`formed by six loops of polypeptide, the hypervariable regions;
`three from the variable domain of the light chain (VL) and three
`from the variable domain of the heavy chain (VH), denoted Ll,
`L2, L3, and HI, H2, H3, respectively (Fig. la). Within the
`domains, the loops are connected to a j3-sheet framework whose
`2
`structure is conserved 1
`• The specificity and affinity of the bind(cid:173)
`•
`ing sites are governed by the structures of the six hypervariable
`4
`regions 3
`.
`'
`Two models can be proposed for the relationship between
`the amino-acid sequence and structure of the binding-site loops.
`In one model, different sequences produce different conforma(cid:173)
`tions for both the main chain and side chains of the loops.
`Because hypervariable regions have different sequences in
`different antibodies, this model implies that each region adopts
`a different conformation in different antibodies. In the other
`
`** Present address: The Squibb Institute for Medical Research, PO Box 4000,
`Princeton, New Jersey 08543-4000, USA.
`
`NATURE · VOL 342 · 21128 DECEMBER 1989
`
`model, antibodies have only a few main-chain conformations
`or 'canonical structures' for each hypervariable region. Most
`sequence variations would only modify the surface provided by
`the side chains on a canonical main-chain structure. Sequence
`changes at a few specific sets of positions would switch the main
`chain to a different canonical conformation.
`Canonical structure model
`Experimental evidence indicates that the canonical structure
`model describes the relationship between amino-acid sequence
`9
`and structure for at least five of the six hypervariable regions 5
`.
`-
`Kabat et al. 5 found conserved residues at sites within certain
`sets of hypervariable regions and suggested that they had a
`structural role. Padlan and Davies 6
`, and more recently de la
`Paz et al. 7
`, showed that some of the hypervariable regions
`in the immunoglobulins of known structure have the same
`main-chain conformation in spite of several differences in
`sequence.
`Chothia and Lesk 8 identified the residues that through pack(cid:173)
`ing, hydrogen bonding, or the ability to assume unusual values
`of the torsion angles ¢, r/! or w, are primarily responsible for
`the main-chain conformations of the hypervariable regions in
`the structures then known-the Fab fragments of NEW (ref.
`10), McPC603 (ref. 11), KOL (ref. 12) and J539 (ref. 13) and
`the VL domains of REI (ref. 14) and RHE (ref. 15). The
`conformations are determined by the interactions of a few
`residues at specific sites in the hypervariable regions and, for
`certain loops, in the framework regions. Hypervariable regions
`that have the same conformations in different immunoglobulins
`have the same or very similar residues at these sites (Fig. 1 and
`Table 1). Examination of the amino-acid sequence of the anti(cid:173)
`body D 1.3 showed that its hypervariable regions are the same
`size as those in known structures and contain the same or similar
`residues at the sites responsible for known conformations 9
`• On
`the basis of these observations the atomic structure of the VL-VH
`dimer of 01.3 was predicted before its experimental determina(cid:173)
`tion. Comparison of this predicted structure with the preliminary
`crystal structure showed that the conformations of four of the
`hypervariable regions had been predicted correctly; the confor(cid:173)
`mation of L3 was significantly different from that predicted, and
`
`877
`
`© 1989 Nature Publishing Group
`
`1 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`L1 Regionst
`
`Canonical
`Structure
`
`1
`
`2
`
`3
`
`4
`
`TABLE 1 Sequences and conformations of V K and VH hypervariable regions of known structure
`
`Protein
`
`26
`
`27
`
`28
`
`29
`
`30
`
`31
`
`e
`
`32
`
`2
`
`25
`
`33
`
`71
`
`J539
`HyHEL-5
`NQ10
`
`REI
`01.3
`HyHEL-10
`NC41
`
`McPC603
`
`4-4-20
`
`s
`s
`s
`s
`s
`s
`s
`s
`s
`
`s
`s
`s
`Q
`G
`Q
`Q
`
`Q
`
`s
`s
`s
`0
`N
`s
`0
`s
`s
`
`v
`v
`v
`
`I
`v
`
`s
`N
`R
`
`I
`H
`G
`s
`
`v
`
`K
`N
`N
`T
`
`N
`
`H
`
`s
`s
`
`G
`
`N
`
`N
`
`G
`
`K
`
`N
`
`N
`
`T
`
`s
`y
`y
`v
`y
`N
`A
`
`y
`
`L
`M
`M
`
`y
`y
`y
`
`y
`y
`F
`y
`
`A
`A
`A
`
`A
`A
`A
`A
`s
`s
`
`v
`
`Total no. of sequences known for L1 regions: human. 95; mouse, 299.
`Canonical structure
`1
`2
`4
`3
`Human sequences that fit (%)
`60
`5
`5
`Mouse sequences that fit (%)
`25
`20
`10
`
`15
`
`L2 Regions
`
`Canonical
`Structure
`
`Protein
`
`50
`
`51
`
`52
`
`48
`
`64
`
`REI
`McPC603
`J539
`01.3
`HyHEL-5
`HyHEL-10
`NC41
`NQ10
`4-4-20
`
`E
`G
`
`y
`0
`y
`w
`0
`K
`
`A
`A
`
`T
`T
`A
`A
`T
`v
`
`s
`s
`s
`T
`s
`s
`s
`s
`s
`
`G
`G
`G
`G
`G
`G
`G
`G
`G
`
`Total no. of sequences known for L2 regions: human, 69; mouse, 183.
`1
`Canonical structure
`Human sequences that fit (%)
`95
`Mouse sequences that fit (%)
`95
`
`L3 Regions
`
`Canonical
`Structure
`
`2
`
`3
`
`Protein
`
`91
`
`92
`
`93
`
`94
`
`95
`
`96
`
`REI
`McPC603
`01.3
`HyHEL-10
`NC41
`4-4-20
`NQ10
`
`J539
`
`HyHEL-5
`
`y
`0
`
`s
`H
`s
`w
`w
`w
`
`Q
`H
`w
`N
`y
`T
`s
`T
`
`G
`
`s
`s
`s
`s
`s
`H
`s
`y
`
`R
`
`y
`L
`R
`y
`w
`w
`L
`
`L
`y
`T
`w
`v
`N
`p
`N
`
`p
`p
`p
`p
`
`p
`
`p
`
`90
`
`Q
`N
`H
`Q
`Q
`Q
`Q
`Q
`Q
`
`Total no. of sequences known for L3 regions: human, 52; mouse, 152.
`1
`2
`Canonical structure
`3
`Human sequences that fit (%)
`90
`2
`Mouse sequences that fit (%)
`80
`1
`
`10
`
`Hl had a very different fold from that predicted 9
`• (We report
`below that the refined conformation of Dl.3 corresponds more
`closely to the predicted structure.)
`An examination of the library of the known immunoglobulin
`sequences shows that many immunoglobulins have hyper(cid:173)
`variable regions that are the same size as those in the known
`structures and contain the same or closely related residues at
`the sites responsible for the known conformations 8
`. These
`observations indicate that for at least five of the hypervariable
`regions there is only a small repertoire of canonical main-chain
`conformations and that the conformation actually present can
`often be predicted from the sequence by the presence of specific
`residues.
`for
`the canonical structure model
`The accuracy of
`immunoglobulin binding sites depends on (1) the correct deter-
`
`878
`
`ruination of the sets of residues responsible for the observed
`conformations and (2) changes in the identity of residues at
`other sites not significantly affecting the conformations of the
`canonical structures. The model can be tested, refined and
`extended by using it to predict the atomic structures of binding
`sites in immunoglobulins before their structures have been deter(cid:173)
`mined by X-ray crystallography.
`We have now tested the canonical structure model by using
`it to predict the structures of four immunoglobulins before their
`structures had been experimentally determined. These
`immunoglobulins are HyHEL-5 (ref. 16), HyHEL-10 (ref. 17),
`NC41 (ref. 18) and NQlO (S.S., P.M.A. and R.J.P., manuscript
`in preparation). The analysis of the amino-acid sequences of
`these immunoglobulins indicated that 19 of their 24 hypervari(cid:173)
`able regions should have conformations close to known canoni-
`
`NATURE · VOL 342 · 21/28 DECEMBER 1989
`
`© 1989 Nature Publishing Group
`
`2 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`H1 Regionst
`
`Canonical
`Structure
`
`1
`
`1'
`
`Protein
`
`26
`
`27
`
`28
`
`29
`
`30
`
`31
`
`32
`
`34
`
`94
`
`McPC603
`KOL
`J539
`01.3
`Hyl-EL-5
`NC41
`NQ10
`4-4-20
`
`NEW
`HyHEL-10
`
`G
`G
`G
`G
`G
`G
`G
`G
`
`G
`G
`
`F
`y
`y
`F
`
`s
`0
`
`T
`I
`0
`s
`T
`T
`T
`T
`
`T
`s
`
`s
`s
`s
`T
`s
`T
`s
`s
`s
`T
`
`0
`s
`K
`G
`0
`N
`s
`0
`
`N
`0
`
`y
`y
`y
`y
`y
`
`y
`
`0
`0
`
`M
`M
`M
`v
`I
`M
`M
`M
`y
`w
`
`R
`R
`R
`R
`R
`R
`R
`G
`
`R
`N
`
`Total no. of sequences known for H1 regions: human. 50: mouse, 321.
`1
`Canonical structure
`Human sequences that fit (%)
`50
`Mouse sequences that fit (%)
`80
`
`H2 Regions§
`
`Canonical
`Structure
`
`1
`
`2
`
`3
`
`4
`
`Protein
`
`52a
`
`b
`
`53
`
`54
`
`55
`
`71
`
`NEW
`01.3
`HyHEL-10
`
`HyHEL-5
`NC41
`
`KOL
`J539
`NQ10
`
`McPC603
`4-4-20
`
`y
`G
`y
`
`G
`N
`
`0
`D
`G
`
`N
`y
`
`H
`0
`s
`
`s
`T
`
`(•)
`G
`s
`s
`(•)
`K
`N
`
`G
`G
`G
`
`(•)
`G
`G
`
`s
`G
`s
`
`y
`y
`
`p
`T
`
`0
`p
`s
`
`N
`N
`
`K
`K
`
`G
`p
`
`A
`L
`
`R
`R
`R
`
`R
`R
`
`Total no. of sequences known for H2 regions: human. 54; mouse. 248.
`4
`2
`1
`3
`Canonical structure
`15
`40
`1
`Human sequences that fit (%)
`15
`20
`40
`15
`5
`Mouse sequences that fit (%)
`
`The residues listed here (single-letter code) are those that form the hypervariable regions and those in the framework regions that are important for the observed conformations
`. The hypervariable regions are taken as those outside the framework ,B-sheet8
`. Except for H2. they are similar to. but not identical with the regions that show
`of these regions8
`high sequence variations and which Kabat et al. 26 use to define hypervariable regions. The sequences are grouped so that those that have the same main-chain conformation.
`or canonical structure, are adjacent. The canonical structure numbers used below refer to the conformations shown in Fig. 1. The residues in the hypervariable and framework
`regions that are mainly responsible for these conformations" are indicated by an asterisk. The classification and sequence requirements of the H2 conformations have been
`. For each hypervariable region the number of human and mouse sequences listed by Kabat et al.26 are given. We
`revised in the light of work described here and elsewhere28
`also give the percentage of these sequences that are the same size as the known canonical structures and have the same residues at the positions marked by an asterisk.
`t Canonical structure ·4 is illustrated in Fig. 4. Although the size of the known L1 structures varies between 6 and 13 residues. they have closely related folds with residues
`26-19 and 32 packed against the framework in the same conformation8
`. The remaining residues form a turn or loop on the surface (Figs 1 and 4). The ends of the long loops
`have some flexibility. There are another 25% of the human sequences and 20"Ai of the mouse sequences that have one more residue than structure 2. or one fewer than structure
`4, and whose sequences satisfy the requirements listed above. It is expected that these differ only in the conformations of the tips of the surface loops.
`t The H1 hypervariable regions with canonical structure 1 have very similar conformations: the r.m.s. differences in the coordinates of their main-chain atoms are 0.3-0.8 A.
`The H1 regions in NEW and HyHEL-10 only partly satisfy the sequence requirements for structure 1 and have a distorted version of its conformation.
`§The H2 region here comprises residues 52a- 55. The region with high sequence variation is 50-65 (ref. 26). In the known structures the main-chain conformation of 50-52
`and 56- 63 do not differ significantly8 (Fig. lb). (•), The residues at positions 55 or 54 in the canonical structures 2. 3 and 4 have residues with positive values for q, and i/J,
`and usually, but not in all cases, Gly, Asn or Asp is found at these sites. For a sequence to match that of canonical structure 2, 3 or 4 the presence of these residues at sites
`54 or 55 is required.
`
`cal structures. We then compared the predicted structures of
`these hypervariable regions with the subsequently determined
`structures. Another immunoglobulin structure, 4-4-20 (ref. 19)
`has recently been reported. We did not have the opportunity to
`predict the structure of 4-4-20 before its experimental determina(cid:173)
`tion, and we discuss here only how its hypervariable regions
`have the conformations expected from the known canonical
`structures. Also, we report that the refined conformation of DL3
`(ref. 20) corresponds more closely to the predicted structure.
`
`Model building procedure
`The main-chain conformations of the hypervariable regions in
`the V K and VH domains of known structure are shown in Fig.
`L The residues responsible for these conformations are listed
`
`NATURE - VOL 342 · 21/ 28 DECEMBER 1989
`
`in Table I. Each hypervariable region in the immunoglobulins
`of unknown structure was examined to determine ( 1) whether
`it has the same size as any homologous hypervariable region of
`known structure and (2) whether its sequence contains the set
`of residues responsible for a known conformation. Except for
`L3 in HyHEL-5, all the light-chain regions correspond to a
`known canonical structure, as do all the HI regions and the H2
`region in HyHEL-10 (Table 1). The conformation predicted for
`the H2 regions in NC41 and HyHEL-5 was based on the analysis
`of the H2 region in the preliminary structure of J539 (ref. 8).
`In all three of these antibodies the H2 region is a four-residue
`turn with Gly at the fourth position and the predicted conforma(cid:173)
`tion is that almost always found for such turns 2 1
`. (Below we
`present a more accurate analysis of H2 regions.) For H3 regions
`in HyHEL-5, HyHEL-10, NC41 and NQlO, no prediction of
`
`879
`
`© 1989 Nature Publishing Group
`
`3 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`conformation could be made on the basis of the known canonical
`structures.
`The sequences of the VL and VH domains were compared to
`see which of the known framework structures have sequences
`close to those of the unknown structures. From the comparisons
`
`a
`
`q~c'.,~
`O" ~1-- C
`L3 ~H2
`
`N
`VL
`
`VH
`
`Antibody
`
`Antigen binding site
`
`b
`
`VK
`
`of the hypervariable and framework regions, one known VL and
`one known VH structure were taken as the starting points(cid:173)
`parents-for the model of the predicted structure. If the confor(cid:173)
`mation predicted for a hypervariable region was not present in
`the parent, but was present in another known structure, the
`hypervariable region in the parent was replaced by that in the
`other structure. Side chains in the parent that were different
`from those in the unknown structure were replaced 22
`23 and the
`•
`resulting model subjected to a very limited energy refinement24
`•
`HyHEL-5, HyHEL-10 and NC41 hypervariable regions
`The atomic structures of Fab fragments of immunoglobulins
`HyHEL-5, HyHEL-10 and NC41 in complexes with their protein
`18
`antigens were determined by X-ray crystallography 16
`• The
`-
`resolution of the X-ray data used to determine the structures
`and value of the residual (R) after refinement are (complex,
`resolution, R value): HyHEL-5-lysozyme, 2.5 A, 20%; HyHEL-
`10-lysozyme, 3.0 A, 24%; and NC41-neuraminidase, 2.9 A,
`19%. These structures have been determined at medium resol(cid:173)
`ution. The tracing of the polypeptide chain of the hypervariable
`regions is unambiguous, although the orientation of some of
`
`lie
`
`2
`
`26
`
`3
`
`L3
`
`~:;o
`90Lv·········'~-97
`GI~~ .. .-!
`1
`
`L2
`
`5~52
`Ile 489;!(0 •
`
`95
`Pro
`
`97
`
`2
`
`3
`
`VH
`
`"\~'
`
`Phe29 ~
`
`H1
`
`FIG. 1 a, Antigen-binding sites of immuno(cid:173)
`globulins are formed by six loops of poly(cid:173)
`peptide, three from the VL domain L1, L2
`and L3 and three from the VH domain H1, H2
`and H3 (wavy Jines). These loops are attached
`to strands (D) of a conserved {3-sheet. b,
`Canonical structures for the hypervariable
`regions of V K and VH domains. In each drawing
`the region is viewed so that the accessible
`surface is at top and the framework region
`below. The main-chain conformation and some
`of the side chains that determine this confor(cid:173)
`mation are shown. For the definition of the
`hypervariable regions used here, see Table 1.
`The immunoglobulins in which the different
`canonical structures occur are listed in Table 1.
`
`~-- .....
`
`~
`52~ . · } . . 56
`50~··.· ..
`4
`
`58
`
`1
`
`2
`
`3
`
`© 1989 Nature Publishing Group
`
`NATURE · VOL 342 · 21128 DECEMBER 1989
`
`H2
`
`880
`
`4 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`the peptide groups is uncertain. Most side chains are unequivo(cid:173)
`cally placed.
`Figure 2 shows the predicted and observed structures of each
`of the hypervariable regions, superposed by a least-squares fit
`of their main-chain atoms. Table 2a gives, for each predicted
`and observed hypervariable region, the r.m.s. difference in posi(cid:173)
`tion of the main-chain atoms.
`The main-chain conformations of the predicted and observed
`hypervariable regions are very similar (Fig. 2, Table 2a ). The
`only exception is the Hl region of HyHEL-5. Although the
`observed and predicted conformations of residues 26-29 and
`32 are the same, residues 30 and 31 are in quite different
`positions. The recently refined structure of Fab 1539 (T. N. Bhat,
`E.A.P. and D.D., manuscript in preparation) shows that these
`differences were inherited as a result of an error in the original
`determination of the 1539 structure used as the parent for this
`region. Rebuilding the predicted model with the refined 1539
`structure puts residues 30 and 31 in the correct position and
`gives an r.m.s. difference between the predicted and observed
`Hl regions of 0.6 A.
`Given the medium resolution of the structures used to derive
`the models and of the experimental structures, the agreement
`of the predicted and observed loop conformations is excellent.
`
`Relative positions of the hypervariable regions
`Figure 3 shows the positions of the hypervariable regions relative
`to each other and to the framework for the predicted and
`observed structures. To produce this figure the predicted and
`observed structures were superposed by a least-squares fit of
`framework residues. In Table 2b, the differences in position of
`the hypervariable regions are reported.
`Small differences in the relative positions of the hypervariable
`regions in the predicted and observed structures might be expec(cid:173)
`ted because of two factors not corrected for in the model build-
`
`ing. First, the predicted structures were built using parent V
`domains that have some residues in the framework and VL-VH
`interface that are different from those in the final predicted
`structure. These differences have small but significant effects on
`the main-chain structure of the individual domains and the way
`25
`26
`they pack together 2
`• Second, differences between the pre(cid:173)
`•
`'
`dicted and observed structures could arise from the effects of
`the association with the antigen 18
`• In other proteins, ligand
`binding can result in the movement of close-packed segments
`of polypeptide relative to each other by 1-2 A, and the ends of
`loops are able to move somewhat more 27
`.
`The differences in the positions of the predicted and observed
`H2 regions in HyHEL-5 and NC41 (Table 2b) are larger than
`expected from these factors. The interactions that the H2 regions
`make with the rest of the VH domain were therefore examined.
`
`Residue 71 and position and conformation of H2
`At the same time as the structures of HyHEL-5 and NC41
`became available, the refinement of the atomic structure of 1539
`was completed (T. N. Bhat, E.A.P. and D.D., manuscript in
`preparation). The conformation of H2 in the refined structure
`is not like that in HyHEL-5 and NC41 but is the same as that
`in KOL. This was quite unexpected. The main determinant of
`the conformation of small turns is usually the position of glycine
`residues 21
`: in KOL, Gly occurs at position 54, and in 1539,
`HyHEL-5 and NC41, Gly occurs at position 55.
`An examination of the environments of the H2 regions 28
`shows that in KOL and 1539 the side chain of framework residue
`Arg 71 packs between, and forms hydrogen bonds to, HI and
`H2. In HyHEL-5, residue 71 is Ala, and here the cavity that
`would be created by this smaller side chain is filled by the
`insertion of a residue from the H2 region-Pro 52a. In KOL
`and 1539 the side chain at position 52a is on the surface. The
`relative movement of position 52a involves a change in the
`
`~.MWr~,
`
`HyHEL-10l1
`
`NC41L1
`
`HyHEL·5L1
`
`NO•OL1
`
`Hy HEL-5 L2
`
`Hy HEL-10 l2
`
`NC41 L2
`
`N010 L2
`
`HyHEL-10l3
`
`NC41 L3
`
`N01 0 L3
`
`Hy HEL-5 H1
`
`Hy HEL-10 H1
`
`NC41 H1
`
`N010 H1
`
`FIG. 2 The predicted (broken line) and observed
`(continuous line) conformations of the hypervari(cid:173)
`able regions. The structures have been super(cid:173)
`posed by a least-squares fit of their main-chain
`atoms. Residue numbers and the r.m.s. difference
`in position of the superposed atoms are given in
`Table 2a. Predicted and observed side-chain con(cid:173)
`formations are shown for all regions except H1
`and L3 in NQ10 where they obscure the main chain.
`After our prediction of the NC41 structure several
`revisions were made to the sequence and some
`of the differences can be seen here.
`
`Hy HEL-5 H2
`
`Hy HEL-10 H2
`
`NC41 H2
`
`N010 H2
`
`NATURE · VOL 342
`
`21/28 DECEMBER 1989
`
`881
`
`© 1989 Nature Publishing Group
`
`5 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`TABLE 2 Differences in structure of predicted and observed hypervariable regions
`
`(a) Differences in local conformation (A)
`
`Hypervariable
`region
`
`L1:26-32
`L2:49-53
`L3:90-97
`H1 :26- 32
`H2 :52-56
`
`R.m.s. difference in atomic positions of main-chain
`atoms after optimal superposition
`HyHEL-10
`NC41
`
`NQ10
`
`HyHEL-5
`
`0.8
`0.9
`
`1.4
`1.1
`
`1.1
`0.8
`0.3
`1.3
`1.0
`
`0.7
`0.4
`0.5
`0.9
`0.7
`
`0.4
`0.9
`0.6
`0.3
`0.4
`
`(b) Differences in position relative to framework (/>.)
`
`Protein
`
`HyfEL-5
`
`HyHEL-10
`
`NC41
`
`NQ10
`
`Range of differences in positions of Ca atoms after
`superposition of frameworks residues (A)
`Hypervariable
`Original
`New structure
`region
`prediction
`library
`
`L1
`L2
`H1
`H2
`
`L1
`L2
`L3
`H1
`H2
`
`L1
`L2
`L3
`H1
`H2
`
`L1
`L2
`L3
`H1
`H2
`
`0.8-2.1
`1.2-2.3
`0.7-2.1
`0.5-2.1
`
`1.5-2.6
`1.0-2.0
`1.8- 3.0
`1.0-2.1
`0.4-1.9
`
`2.0- 3.8
`1.6-1.6
`0.8- 4.1
`3.0-7.2
`
`0.7-1.6
`0.6-.13
`0.8-1.5
`1.3- 3.5
`0.6-2.9
`
`1.4-2.4
`1.1- 1.8
`1.6 - 2.3
`1.3-2.0
`2.1- 4.4
`
`0.4- 2.7
`0.5-1.4
`0.6-1.5
`0.6- 1.2
`0.6-0.9
`
`Superposition of a are illustrated in Fig. 2. The differences in the positions of the
`hypervariable regions in the original predictions and the observed structures of b are
`shown in Fig. 3. The predictions with a new structure library involve rebuilding the
`HyHEL-5 model using the refined VL J539 and the VH NC41 structures, and rebuilding
`the NC41 model using VL McPC603 and VH HyHEL -5: see text
`
`conformation of H2. It also tilts the H2 loop so that the positions
`of residues at the top of the loop in HyHEL-5 differ in position,
`relative to those in KOL and J539, by ~4.5 A. In NC41, the
`residue at position 71 is Leu and that at position 52a is Thr,
`and the shift is not as large as that in HyHEL-5.
`This analysis implies that the conformation and position of
`four residue H2 regions are determined by the packing against
`the VH framework and by the identity of the residue at position
`71 in particular, and not by the position in the sequence of H2
`of Gly (or Asn or Asp) as was believed previously8
`• Thus VH
`domains of HyHEL-5 and NC41 should provide better parents
`for each other than the other structures do, for they contain
`similar determinants for the position of H2. The predicted
`structure of HyHEL-5 using the observed structure of VH of
`NC41 and the predicted structure of NC41 using the observed
`structure of VH of HyHEL-5 were thus rebuilt. In these new
`predicted structures the large differences between the predicted
`and observed positions of H2 are not present: most residues
`differ by no more than 2.1 A and none differs by more than
`3.0 A (Table 2b ).
`The results of this analysis were used in the prediction of the
`structure of the H2 region of the antibody NQlO.
`
`Hypervariable regions of NQ10 and D1.3
`The amino-acid sequence of the hypervariable regions and
`associated framework sites of NQlO are given in Table 1. For
`five of the hypervariable regions, the size and the residue con(cid:173)
`servation at the relevant sites clearly indicate particular canoni(cid:173)
`cal structures, and a model of the VL-VH dimer of NQlO was
`made using the procedure described above. For H2, after the
`
`882
`
`analysis described above, canonical structure 2 was expected
`because of the Arg at position 71 (Table 1).
`Recently a crystal structure has been determined for NQlO
`(S.S., P.M.A. and R.J.P., manuscript in preparation). This struc(cid:173)
`ture is determined to a resolution of 2.8 A and the present
`residual is 21 %. The tracing of the chain in the hypervariable
`regions is unambiguous. Figure 2 shows the predicted and
`observed structures of each of the hypervariable regions, super(cid:173)
`posed by a least-squares fit of their main-chain atoms. Table 2a
`gives the r.m.s. differences in position of the main-chain atoms.
`Figure 3 shows the relative positions of the predicted and
`observed hypervariable regions.
`There is close agreement between the predicted and observed
`main-chain conformations of the hypervariable regions: the
`r.m.s. differences in position are between 0.3 and 0.9 A (Table
`2a). There is also close agreement in the relative positions of
`the hypervariable regions (Fig. 3 and Table 2b ). No residues
`differ by more than 2. 7 A in position and all but two are within
`1.1 A.
`Recently, the atomic structure of antibody Dl.3 complexed
`with lysozyme has been refined 20
`. The comparison of the pre(cid:173)
`liminary experimental structure with the prediction had shown
`differences for two hypervariable regions 9
`. The L3 regions had
`differences associated with Pro 95 having a cis-peptide in the
`predicted structure and a trans-peptide in the observed. The Hl
`regions differed because the predicted structure had this region
`folded into framework, whereas in the observed it folded out
`into the solvent. The refinement of the experimental structure
`has resulted in the rebuilding of these two regions 20
`, and their
`conformations now agree with the original predictions.
`
`Canonical structures in immunoglobulin 4-4-20
`The atomic structure of immunoglobulin 4-4-20 has recently
`been determined at a resolution of 2.7 A (ref. 19). The amino(cid:173)
`acid sequence of the hypervariable regions and associated
`framework sites of 4-4-20 are given in Table 1. Four of the
`hypervariable regions, L2, L3, Hl and H2, have the size and
`the residues at specific sites that clearly indicate particular
`canonical structures (see Table 1). These four hypervariable
`regions do have the expected main-chain conformations. In Fig.
`4 the four hypervariable regions, superposed on examples
`of the same canonical structures taken from other immuno(cid:173)
`globulins, are shown. The r.m.s. difference in the atomic
`
`Hy HEL - 5
`
`L2i6
`);L1
`
`Hy HEL - 10
`
`L2v
`
`f (J
`
`L3
`
`L1
`
`j "'
`
`0 H2
`
`NQ10
`
`L2
`
`uf · .. _.-...
`L3~
`
`j
`H1
`j H2
`
`FIG. 3 The relative positions of the hypervariable regions in the predicted
`and observed structures. The positions of the hypervariable regions shown
`are those given by the superposition of the VL-VH framework regions of
`the observed and predicted structures. The Ca atoms of residues in the
`observed structure are joined by solid lines and those in the predicted
`structure by broken lines. The size of the differences in position are given
`in Table 2b.
`
`NATURE · VOL 342 · 21/ 28 DECEMBER 1989
`
`© 1989 Nature Publishing Group
`
`6 of 7
`
`Celltrion, Inc., Exhibit 1001
`
`
`
`ARTICLES
`
`the canonical structure model. It implies that a good estimate
`can be made of the extent to which hypervariable regions in
`other immunoglobulins have main-chain conformations close
`to those that are now known. This can be done by inspecting
`the known sequences to see whether their hypervariable regions
`correspond in size to the known canonical structures, and
`whether they contain one of the sets of residues that produce
`the observed conformations.
`In Table 1 we give the results of examining the immuno(cid:173)
`globulin sequences collected by Kabat et al. 29
`. About 90% of
`the hypervariable regions in V K domains, and -70% of the H 1
`and H2 regions in VH domains, are expected to have conforma(cid:173)
`tions close to those found in the immunoglobulin structures now
`known. These estimates are conservative in that they include
`only hypervariable regions that match the sequence require(cid:173)
`ments exactly, that is, those that contain the particular residues,
`or any combination of the small range of closely related residues,
`that are found in the known structures at the sites marked by
`asterisks in Table 1.
`
`Discussion
`The wide occurrence of the known canonical structures, the use
`of the much larger library of parent structures that will soon
`become available, and the more detailed understanding of the
`determinants of structure that will emerge from the analysis of
`the differences between predicted and observed structures,
`should allow, for many immunoglobulins, the prediction of the
`structure of five of the hypervariable regions with accurate local
`conformation and errors in position of 2 A or less.
`Predictions of the structures of the H3 regions in the
`immunoglobulins discussed here could not be made on the basis
`of the previously known structures. It was possible to predict
`correctly the H3 conformation in immunoglobulin Dl.3 (ref. 9).
`In general, however, the various genetic mechanisms that pro(cid:173)
`duce the H3 regions usually result in medium or large surface
`loops with very different sequences and patterns of interaction.
`For such hypervariable regions it may be possible to predict
`their structure using the conformational search algorithms that
`35
`have been developed by several groups 30
`•
`-
`The results presented here have interesting implications for
`the molecular mechanisms involved in the generation of anti(cid:173)
`body diversity. For at least five of the six hypervariable regions
`of most immunoglobulins there seems to be only a small reper(cid:173)
`toire of main chain conformations, most of which are known
`from the set of immunoglobulin structures so f