13. Stoeckel, K., Schwab, M. & Thoenen, H. Brain Res. 89, l-14 (1974).
`14. Angeletti, R. H. & Bradshaw, R. A. Proc. natn. Acad. Sci. U.S.A. 68, 2417-2420 (1971).
`15. Scott, J. et al. Nature 302, 538-540 (1983).
`16. Ullrich, A., Gray, A., Berman, C. & Dull, T. J. Nature 303, 821-825 (1983).
`17. Darling, T. L. J. et al. Cold Spring Harb. Symp. quant. Biol. 48, 427-434 (1983).
`18. Menescini-Chen, M. G., Chen, J. S. & Levi-Montalcini, R. Archs ital. Biol. 116, 53-84 (1978).
`19. Letourneau, P. C. Devi Biol. 66, 183-196 (1978).
`20. Gundersen, R. W. & Barrett, J. N. Science 206, 1079-1080 (1979).
`21. Levi.Montalcini, R. Prog. Brain Res. 45, 235-258 (1976).
`22. Renehan, W. E. & Munger, B. L. J. comp. Neurol. 246, 129-145 (1986).
`23. Iggo, A. Br. med. Bull. 33, 97-102 (1977).
`
`24. Iggo, A. & Andres, K. H. A. Rev. Neuroscl 5, 1-31 (1982).
`25. Davies, A. M. & Lumsden, A.G. S. J. comp. Neurol 223, 124-137 (1984).
`26. Davies, A. M. & Lumsden, A.G. S. J. comp. Neural 253, 13-24 (1986).
`27. Korsching, S. & Thoenen, H. Proc. natn. Acad. Sci. U.S.A. 80, 3513-3516 (1983).
`28. Bandtlow, C., Heumann, R., Schwab, M. & Thoenen, H. EMBO J. (in the press).
`29. Rush, R. A. Nature 312, 364-367 (1984).
`30. Finn, P. J., Ferguson, I. A., Renton, F. J. & Rush, R. A. J. Neurocytol. 15, 169-176 (1986).
`31. Lumsden, A.G. S. & Davies, A. M. Nature 306, 786-788 (1983).
`32. Lumsden, A.G. S. & Davies, A. M. Nature 323, 538-539 (1986).
`33. Lowry, 0. H., Rosenbrough, N. J., Farr, A. L. & Randall, R. J. J. biol. Chem. 193, 265-275
`0951).
`
`Three-dimensional structure of a complex of
`antibody with influenza virus neuraminidase
`P. M. Colman, W. G. Laver*, J. N. Varghese, A. T. Baker, P.A. Tulloch,
`G. M. Airt & R. G. Webster*
`CSIRO Division of Protein Chemistry, 343 Royal Parade, Parkville, 3052, Australia
`*John Curtin School of Medical Research, Australian National University, Canberra 2601, Australia
`t Department of Microbiology, University of Alabama, Birmingham, Alabama 35294, USA
`:j: St Jude Children's Research Hospital, Memphis, Tennessee 38101, USA
`
`The structure of a complex between influenza virus neuraminidase and an antibody displays features inconsistent with the
`inflexible 'lock and key' model of antigen- antibody binding. The structure of the antigen changes on binding, and that of
`the antibody may also change; the interaction there! ore has some of the character of a handshake.
`
`WE report here the first analysis by X-ray diffraction of the
`three-dimensional structure of a viral antigen complexed with
`an antibody Fab fragment. The antigen is the influenza virus
`enzyme neuraminidase (reviewed in ref. 1).
`Sufficient data on the chemical2 and spatial3
`6 structure of
`-
`antibodies emerged in the 1970s to provide a basis for under(cid:173)
`standing how variation in antibody structure occurs, but the
`question of how different antibodies aci;:ommodate different
`macromolecular epitopes remains unresolved. The antigen bind(cid:173)
`ing fragments (Fab) of immunoglobulins consist of a light chain
`(L) and the N-terminal half of the heavy (H) chain. On each
`chain there are two globular domains of ~100 amino acids, the
`N-terminal domain in each chain being variable (VL and VH)
`and the C-terminal domain conserved (CL and CHl) in their
`amino-acid sequences. VL and VH domains are associated with
`each other, forming a variable module and CL and Cttl
`form the constant module. Three complementarity-determin(cid:173)
`ing regions (CDRl, 2 and 3) from each of the VL and VH
`domains are clustered at the extremity of the Fab arms, and
`
`they determine the binding specificity of an immunoglobulin
`molecule.
`The three-dimensional structures of five free Fab fragments 7
`11
`-
`and one in complex with lysozyme 12 have been reported. Four
`, McPC6039 and J539 10
`of the structures, New7
`, Kol8
`, are well
`refined and form the basis of the following generalizations which
`also appear valid for the other two, Sl0/1 (ref. 11) and Dl.3
`(ref. 12). The pairing of VL and VH domains is determined by
`amino acids from both the framework region and the CDRs.
`Most of the surface area buried in the interface derives from
`conserved residues and this observation explains the largely
`conserved geometry of VL-VH pairing13
`• Small differences in
`pairing presumably result from interactions among the CDR
`amino acids 14
`• Analysis of the VL-VH contact surface shows it
`to be unlike other interfaces between f:J-sheets 15
`• Amino acids
`from the outermost strands of the two sheets fold into the
`interface where they contribute the bulk of the buried surface.
`CL -Cttl dimers are similarly conserved in their association
`pattem16
`18
`• More variable is the so-called elbow angle17
`•
`
`Fig. I Schematic diagrams of chain fold in N2 and
`N9 neuraminidase viewed down the 4-fold axis (a)
`and perpendicular to this axis ( b ). The symmetry axis
`is bottom right in a and standing vertical at the left
`rear in b. The view of the neuraminidase in b is similar
`to that shown in Fig. 5. Tagged residues and adjacent
`chain segments are referred to in the text. N2 number(cid:173)
`ing is used. In a, the side chains of amino acids 368-370
`point towards the viewer, whilst that of Arg 371 points
`away and into the catalytic site located above and to
`the right of Ca 371. Mutations at positions 367, 369,
`370, 400, and 432, as detailed in Fig. 2, abolish the
`binding of NC41 antibody to neuraminidase, whereas
`mutations at 368 and 329 reduce that binding. A muta·
`tion at 220, which falls outside the NC41 epitope, has
`no effect on NC41 binding to neuraminidase (see Fig.
`2). In a, solid chain segments show regions in contact
`with the NC41 antibody; contact assignments in the
`broken solid segments are tentative.
`
`b
`
`(cid:141) (cid:21)(cid:29)(cid:28)(cid:27) (cid:50)(cid:69)(cid:88)(cid:89)(cid:86)(cid:73)(cid:4)(cid:4)(cid:52)(cid:89)(cid:70)(cid:80)(cid:77)(cid:87)(cid:76)(cid:77)(cid:82)(cid:75) (cid:43)(cid:86)(cid:83)(cid:89)(cid:84)
`
`1 of 6
`
`BI Exhibit 1100
`
`

`

`90
`
`11400
`1/100 1/200
`NC 41 lgG dilution
`
`1/800
`
`Fig. 2 Neuraminidase inhibition (NI) curves of monoclonal anti(cid:173)
`body NC41 acting on N9 neuraminidase (NA) and a number of
`variants selected with different monoclonal antibodies to N9
`neuraminidase. Assays used fetuin as a substrate, as described58
`•
`The amino-acid sequence changes of the variants are:•, 434 (432
`in N2 numbering) K-+N; 400 (400) N-+K; 371 (370) S-+L; 370
`(369) A-+ D; 368 (367) S-+ N; 0, 369 (368) I-+ R; !::., 331 (329)
`N-+ D (designated as OX2 in text); 0, 222 (220) R-+Q (this curve
`is identical to wild-type). The amino-acid sequence changes were
`determined by sequencing cDNA of the viral RNA strand encoding
`the neuraminidase as described59
`• The changes are shown on the
`3-o structure of the monomer in Fig. 1.
`
`describing the angle between the local axes of symmetry in the
`V and C modules.
`The identification of antigenic regions on protein molecules19
`has been attempted by a variety of methods20-23 but the structural
`basis of antigenicity remains unclear. Current models of antigen(cid:173)
`antibody association are based on classical ideas of lock and
`key complementarity. But it has been suggested24 that the anti(cid:173)
`genic regions of proteins are located in the more flexible chain
`segments, which may adopt a configuration that allows antibody
`binding. It is not known whether binding to antigen has any
`effect on the structure of an antibody25, nor whether, if such an
`effect exists, it has biological significance. For the Fab-lysozyme
`complex12, the parts of the lysozyme that make contact with the
`antibody are not the more flexible regions of free lysozyme. No
`deformation of the antigen nor any structural change in the Fab
`was observed12.
`The antigen in this study is the neuraminidase of influenza
`virus. It is a tetramer of subunits with relative molecular mass
`60,000 (Mr 60 K) with circular symmetry, and is attached by a
`stalk to the viral membrane. Neuraminidase 'heads' can be
`liberated from the virus by proteolysis26. There is antigenic
`variation in the neuraminidase between different strains of virus.
`Antibodies against neuraminidase do not neutralise virus infec(cid:173)
`tivity, but do modify the disease in favour of the host27
`• The
`three-dimensional structures of neuraminidase heads of sub(cid:173)
`types28 N2 (refs 29, 30) and N9 (A.T.B., J.N.V., W.G.L., G.M.A.
`and P.M.C., manuscript in preparation) are known and a sche(cid:173)
`matic of the chain fold is shown in Fig. 1. Neuraminidase of
`subtype N9 from an avian influenza virus, G70c31 , has been
`used in this study. Its structure is very similar to the N2 enzyme
`as expected from the sequence homology of 50% (ref. 32).
`We previously reported electron and low resolution X-ray
`diffraction studies ofFab-neuraminidase complexes33 and, more
`
`recently, the crystallization ofanother complex (with NC41 Fab)
`which diffracts X-rays to beyond 3 A resolution34. The NC41
`antibody can suppress the yield of virus from infected cells and
`therefore
`allows
`the
`selection of antigenic variants.
`Neuraminidase-inhibition curves (Fig. 2) show that some
`variants with single sequence changes are not inhibited at all
`by NC41 antibody whereas others are partially inhibited. The
`location of these sequence changes on the three-dimensional
`structure of neuraminidase is shown in Fig. 1.
`Protomer structure
`Electron microscopy of negatively stained protomers of complex
`show tetrameric neuraminidase heads with four Fab fragments
`attached (data not shown). Their image appearance is very
`similar to that of protomers of N9 neuraminidase complexed
`with the Fab fragments of antibodies 32/3 and NC35 (ref. 33)
`and may be described as a square box ( 100 x 100 x 60 A) with
`four antennae (Fab molecules) attached to one surface on the
`outer corners. The overall shape of the protomer from the X-ray
`diffraction study (Fig. 3) is consistent with the electron micro(cid:173)
`scope images. The Fab arms subtend an angle of 45° to the
`plane of the tetrameric neuraminidase head, slightly foreshor(cid:173)
`tened in the electron microscope image33.
`Surface loops of the neuraminidase in contact with the CDRs
`are 368-370, 400-403, 430-434 and parts of 325-350. Of these,
`the conformation of the 325-350 loop is still tentative. It was
`the most difficult region of the original N2 neuraminidase struc(cid:173)
`ture29
`to
`interpret and remains ambiguous
`in
`the N9
`neuraminidase structure (A.T.B. et al., manuscript in prepar(cid:173)
`ation). Such instances of ill-defined structure are typical of
`surface loops of polypeptides subject to thermal motion or
`statistical disorder.
`Early indications on the distribution of temperature factors
`in the uncomplexed N2 neuraminidase, based on crystallo(cid:173)
`graphic refinement of that structure at 2.2 A resolution and an
`R-factor of 0.293, show above-average thermal parameters in
`chain segments 109·111, 141-142, 221-222, 243-247, 305-309,
`316-319, 327-331, 338-343, 365-369, 399-402, 429-437 and461-
`469 (J.N.V. and P.M.C., unpublished data). Elevated B values
`around residue 330 may correlate with inaccurate modelling as
`discussed above. All the segments listed here are surface loops.
`Note that only one of the hotter loops is underneath the
`neuraminidase head. We emphasize that the temperature factor
`analysis is preliminary, and- indeed relates to N2, not N9,
`neuraminidase.
`The monoclonal variant of N9 neuraminidase known as OX2
`(Asn 329-+ Asp 329, N2 numbering) crystallizes isomorphously
`with wild-type N9-NC41 Fab complex34. NC41 inhibits the
`OX2 neuraminidase variant less than it does wild-type (Fig. 2).
`Refinement of the N9-NC41 Fab structure using data collected
`from the OX2-NC41 Fab complex yields an R-factor of 0.36
`for a model with tight geometry. This result is comparable with
`the current R-factor against the native dataset and confirms our
`earlier conclusion34 that the binding of NC41 Fab to wild-type
`and OX2 N9 neurarninidases is isosteric. Either some bonding
`'glue' is absent or repulsive forces have been introduced by the
`mutation (Fig. 2), but there is no detectable rearrangement of
`the antibody on the antigen at this stage. The precise location
`of residue 329 (N2 numbering) in the N2 and N9 structures is
`unclear. Refinement of the OX2-NC41 Fab structure should
`eventually reveal the structural basis for the reduced binding of
`NC41 to OX2 compared with wild-type N9.
`The third CDRs on both heavy and light chains are not yet
`completely modelled, and the conformation of the 325-350
`region on the antigen remains uncertain, so we cannot identify
`the amino acids that interact in the interface. All the CDRs,
`with the possible exception of light-chain CDRl appear to make
`contact with the epitope. A preliminary assessment is that there
`are no more contacting residues than the 16 or 17 observed in
`the Fab-lysozyme structure12.
`
`(cid:141) (cid:21)(cid:29)(cid:28)(cid:27) (cid:50)(cid:69)(cid:88)(cid:89)(cid:86)(cid:73)(cid:4)(cid:4)(cid:52)(cid:89)(cid:70)(cid:80)(cid:77)(cid:87)(cid:76)(cid:77)(cid:82)(cid:75) (cid:43)(cid:86)(cid:83)(cid:89)(cid:84)
`
`2 of 6
`
`BI Exhibit 1100
`
`

`

`Quaternary structure of the Fab
`The amino-acid sequence of the Fab is known (G.M.A., unpub(cid:173)
`lished data). The assignment of heavy and light chains in the
`Fab structure
`is consistent with side-chain densities at
`homologous positions on the two chains, such as heavy-chain
`(H) Trp 47 and light-chain (L) Leu 46, H Val 37 and L Tyr 36,
`and, H Trp 103 and L Phe 98. Heavy-atom binding sites also
`show an asymmetry which is consistent with these observations.
`The quaternary structures of known Fab fragments were com(cid:173)
`pared as described in Table 1. The lower left triangle of entries
`indicates that CL -CHI pairing is a function of immunoglobulin
`chain subtype since y1: A chains (as in Kol and New) show a
`consistent mode of association as do a or y 2a: K chains (as in
`M603 and NC41). Small but significant alterations in V-domain
`pairing are observed between Kol, New and M603 (Table 1,
`upper right). This variation in V L -V H association has been
`attributed to the fact that the CDRs contribute partially to the
`dimer interface between the two domains in the V-module14,
`although in the case of Kol we suggest that intimate crystal
`contacts involving the CDRs may contribute.
`Comparison of these three Fahs with NC41 complexed with
`neuraminidase shows, in each case, a significantly larger
`difference in VL-VH association. Compared with Kol, the dis(cid:173)
`tance between H CDR2 and L CDR3 is larger and the relative
`positions of the CD Rs around the antigen-binding surface are
`altered by up to 4 A. A movement of this magnitude is of the
`order of the distance between adjacent a-carbon atoms on a
`polypeptide and may be considered as dramatic as a sequence
`insertion or deletion in a structurally critical part of the molecule.
`As well as the bulk movement of domains described here, the
`CDR loops may be flexible and able to adapt further to an
`epitope.
`Building an atomic model into the penultimate electron
`density map (that in which no phase information from the Fab
`structure had been included) required the rotation of H Trp 47
`180° around Ca-C/3 compared with its position in Kol. M603
`and New are the same as Kol at this point. The positions of the
`loops at residue 40 on both heavy and light chains also required
`remodelling. Therefore the reorganization of the interface
`through sliding of domains has consequences for the domain
`structure of the VL and VH.
`The sequence of the NC41 VL and VH domains shows that
`those framework amino acids on both the light chain (Tyr 36,
`Gin 38, Pro 44, Tyr 87, Phe 98) and the heavy chain (Val 37,
`Gin 39, Leu 45, Trp 47, Phe 91, Ala 93, Trp 103) that are buried
`in the interface15 are conserved as expected. There is nothing
`in the primary sequences of the CDRs of NC41 to suggest a
`cause of the unusual domain association. An independent analy(cid:173)
`sis of the NC41 sequence confirms this view (C. Chothia, per(cid:173)
`sonal communication).
`Two other Fab structures are known to fit the normal pattern
`of V L - V H association, the free antineuraminidase monoclonal
`antibody Sl0/1 (ref. 11) and the complexed anti-lysozyme
`monoclonal antibody Dl.3 (ref. 12). Among light-chain dimers
`the picture is less clear. Although several examples ofVL -Vwlike
`association have been observed3·35-37, there are two clear excep(cid:173)
`tions38-39, one of which is said to be caused by the CDR residue
`L 91, or, in our view, perhaps by a His at residue 38.
`If changes in the amino-acid sequence of the CDRs (which
`are only a small part of the interface) can modulate the pairing
`pattern, binding of the six CD Rs to a macromolecular surface
`might similarly perturb the V L - V H interface. Without knowing
`the three-dimensional structure of the free NC41 Fab we cannot
`definitely conclude that the antibody structure has changed on
`binding antigen, but we believe that this explanation is more
`plausible than the alternative, which requires that special
`sequences in the CDRs alone have determined the pairing
`pattern.
`Other data support conformational changes in antibodies
`when antigen or hapten binds40-44
`• Changes in circular dichro-
`
`Table 1 Quaternary structure comparisons of Fabs
`
`NEW
`M603
`KOL
`NC41
`
`NEW
`
`12.7 (2.3)
`2.5 (O.O)
`11.5 (0.4)
`
`M603
`4.2 (0.7)
`
`12.4 (2.4)
`2.1 (0.7)
`
`KOL
`6.5 (0.4)
`5.0 (0.3)
`
`10.8 (2.1)
`
`NC41
`12.2 (l.O)
`8.8 (1.3)
`12.1 {0.7)
`
`Comparison of four different Fab quaternary structures. Upper right
`triangle: VL domain of the row Fab was mapped into the VL domain
`of the column Fab using the Ca coordinates of residues 33-39, 43-47,
`84-90, 98-104 (Kabat2 numbering). These amino acids were chosen
`because they define the interface between VL and VH domains15
`• The
`calculated transformation was applied to the V H domain of the row Fab
`and the additional rotation in degrees (and translation in A) to optimize
`overlap of the two V H domains. in question was calculated and is given
`above. Ca coordinates used for overlap of VH domains were 34-40,
`44-48, 88-94 and 103-109. Lowerlefttriangle: CL domain of the column
`Fab was mapped into the CL domain of the row Fab using·the C
`coordinates of residues 118-124, 131-137, 160-162 and 174-178 (Kabat!
`numbering). These amino acids were chosen because they define the
`interface between CL and CHI domains16
`• The calculated transformation
`was applied to the CHI domain of the column Fab and the additional
`rotation in degrees (and translation in A) to optimize overlap of the
`two CHI domains are shown. Ca coordinates used for overlap of CH1
`domains were 121-125, 139-145, 172-179 and 186-190. Amino acids
`from the interface were used because if this set is expanded to include
`other parts of the domain structure, the correlation between small
`rotation angles and similar chain classes that is observed for the C
`modu.le is less good. The largest r.m.s. distance between domain structure
`alignments is 0.7 A. A similar comparison was done for the V modules
`of the M603 (antiphosphocholine) and uncomplexed SlO/l (antine(cid:173)
`uraminidase) Fab fragments. No measurable difference in the way the
`V L and V H domains associate in those antibodies was detected. A
`comparison elsewhere of Kol and New V modules 8 yields a value of
`9°. The value reported here (6.5°) is smaller because it is uninfluenced
`by structural differences outside of the interface zone.
`
`ism40 or circular polarization of fluorescence42 indicate an
`altered environment for aromatic residues after antigen binds
`to antibody, and kinetic data support a bi-molecular process
`43
`with distinct conformational states for bound and free Fab41
`44
`•
`•
`•
`Both of these aspects are embodied in the sliding VL-VH model
`presented here.
`Direct structural evidence has been presented for hapten(cid:173)
`induced structural changes in a Bence-Jones protein45
`46
`• In that
`•
`case hydrophobic ligands can apparently penetrate the VL -VL
`interface and signal their presence to the C-module. Strain
`introduced in the structure of the C-module can be relaxed by
`reducing and re-oxidizing the disulphide bond between the
`C-terminal regions of the two light chains46
`• Table 1 indicates
`that in the neuraminidase-NC41 complex no significant changes
`in the quaternary structure of the C-module have occurred.
`Analysis of the pseudo-symmetry axes of the V and C modules
`shows that the angle between these axes, the elbow angle, is
`-150°, which is intermediate in the observed range of 130-180°
`for other Fab structures. The sense of the bending is the same
`as in all other Fab fragments, with VH and CH 1 domains closer
`together than the V L and CL domains.
`Conformational changes in the antigen
`The neuraminidase upper surface loop involving residues 367-
`371 was remodelled to better fit the density as observed in the
`complex. Adjacent parts of the loop which were also omitted
`from the phasing process (residues 364-366 and 372-373) fit the
`density as if they are rigidly attached to the N9 core. The Ca
`atoms of residues 368-371 have been moved by 1 A or more
`from their position in N9 neuraminidase (Fig. 4). Least-squares
`refinement of the structure to R = 0.35, gave a mean shift of
`0.4 A in all Ca atoms with a standard deviation of 0.2 A. The
`current position of the 370 loop (Fig. 4) shows displacements
`of more than 1.2 A for Ca atoms 369-371 compared with their
`
`(cid:141) (cid:21)(cid:29)(cid:28)(cid:27) (cid:50)(cid:69)(cid:88)(cid:89)(cid:86)(cid:73)(cid:4)(cid:4)(cid:52)(cid:89)(cid:70)(cid:80)(cid:77)(cid:87)(cid:76)(cid:77)(cid:82)(cid:75) (cid:43)(cid:86)(cid:83)(cid:89)(cid:84)
`
`3 of 6
`
`BI Exhibit 1100
`
`

`

`Fig. 3 Stereo images of the Ca skeleton of the
`protomer, showing four Fab molecules attached
`to one
`tetrameric neuraminidase
`'head'.
`Neuraminidase, purple; heavy chain, green;
`light chain, blue. View with the protomer 4-fold
`axis vertical, but tipped towards the viewer. The
`crystals of N9-NC41 Fab belong to the space
`group P4212 with a= 167 A, c = 124 A (ref. 34).
`There is one quarter of a protomer per asym(cid:173)
`metric unit and 70% of the cell volume is
`occupied by solvent. Data were collected photo(cid:173)
`graphically to 3 A resolution. In all, 67,976
`measurements of 25,241 independent reflections, representing 76% of the data in the 3 A sphere, were merged with an R-factor on intensities
`of0.12. Phasing was initiated through two heavy atom derivatives (potassium tetra-chloro-platinate and diamino-dinitro-platinum). An envelope
`was computed60 on the basis of a 5 A resolution electron-density map allowing the location of the neuraminidase tetrameron the crystallographic
`4-fold axis to be determined first by inspection and subsequently by correlation methods in real space and rigid body least-squares refinement61
`,
`resulting in a correlation coefficient of 0.204 between Fobs and Fcaic for data in the range 5-3.5 A. It was now clear which surface regions of
`the neuraminidase were in contact with the remaining electron density in the image, and, in subsequent use of neuraminidase as partial
`structure in the phasing process, those regions were excised. A second electron density map, now at 3.5 A resolution, was phased on the two
`derivatives (one to 5 A and one to 3.5 A), the neuraminidase partial structure, and solvent flattening from a redetermined envelope. The four
`domains of the Fab fragment were recognised in this map and fitted independently by real-space correlation methods with the known Fab
`structure Kol8. Rigid-body least-squares refinement with five groups, neuraminidase, VH, CHI, VL and CL produced shifts of up to l.5° in
`group orientations and led to an R-factor of 0.425 and a correlation coefficient of 0.308 for data between 5 and 4 A resolution. The structure
`is based on a 3 A electron density map, which is phased in addition to the information listed above with the structure of the Fab fragment
`excluding the complementarity-determining loops. Thus there is no structural prejudice in the image of those parts of the antigen or antibody
`that are near, or in, the bonding interface. Further refinement62 has since reduced the residual to 0.35 for data between 6 and 3 A resolution.
`The r.m.s. error in the atomic positions is of the order of 0.5 A at this stage.
`
`Fig. 4 Stereo image of the electron density
`map (2Fb-Fc, R = 0.35) of the neuraminidase
`loop 368-372 in the complex with NC41 Fab.
`Residues here are labelled 868-872. The yellow
`model shows the current fit of the loop in the
`complex structure. The purple model is the
`position of this loop in uncomplexed N9
`neuraminidase.
`
`Fig. 5 Stereo image of one quarter
`of the protomer, showing the Fab
`bound to the side of the active site
`cavity on the enzyme. The enzyme
`active centre is facing the viewer,
`below and to the left of the antibody
`binding site, and is coloured yellow.
`The neuraminidase perspective is as
`in Fig. lb. Neuraminidase, green;
`heavy chain, purple; light chain,
`blue.
`
`(cid:141) (cid:21)(cid:29)(cid:28)(cid:27) (cid:50)(cid:69)(cid:88)(cid:89)(cid:86)(cid:73)(cid:4)(cid:4)(cid:52)(cid:89)(cid:70)(cid:80)(cid:77)(cid:87)(cid:76)(cid:77)(cid:82)(cid:75) (cid:43)(cid:86)(cid:83)(cid:89)(cid:84)
`
`4 of 6
`
`BI Exhibit 1100
`
`

`

`position in free N9 neuraminidase. Thus the antigen is distorted
`by interaction with the antibody.
`inhibit
`Most monoclonal anti-neuraminidase antibodies
`neuraminidase activity against fetuin, but only some of these
`also inhibit activity against the trisaccharide sialyl-lactose47.
`Although NC41 Fab cannot sterically block entry of a trisac(cid:173)
`charide into the active site (Fig. 5), NC41 does inhibit this
`activity (R.G.W., manuscript in preparation). We suggest that
`one possible mechanism for enzyme inactivation is the small
`displacement of Arg 371, which points directly into the catalytic
`site. We cannot yet rule out other interpretations, such as the
`effect of the antibody on steering substrates and products into
`or out of the active site. It has been shown that changing Arg 3 71
`for Lys by site-specific mutagenesis results in the loss of90-95%
`of enzyme activity, although the mutated protein is correctly
`folded (M. R. Lentz, R.G.W. and G.M.A., manuscript in prepar(cid:173)
`ation).
`Deformation of antigens by antibodies has been proposed to
`explain a variety of experimental data, including single-hit
`kinetics for neutralization of polio virus48 and altered binding
`affinity for second antibodies binding noncompetitively to the
`antigen49. The capacity of anti-apomyoglobin antibodies to,
`effectively, expel the haem from myoglobin has long been
`known50. There are also many reports of antibodies inhibiting
`enzymes51. Our results directly demonstrate that conformational
`changes can be induced in antigens by antibodies. Furthermore,
`the change observed here, although small, correlates with the
`inactivation of the neuraminidase towards sialyl-lactose. Other
`anti-neuraminidase antibodies
`that
`inhibit neuraminidase
`activity only towards large substrates and, like NC41, can select
`antigenic variants, probably inhibit enzyme activity by blocking
`access of large substrates to the active site. We are testing the
`hypothesis that such antibodies do not distort the active site of
`the enzyme by an X-ray diffraction study of the N9-NC10 Fab
`complex34. Unlike NC41, this antibody does not inhibit enzyme
`activity on sialyl-lactose. Antigen distortion is likely to be but
`one of a number of possible mechanisms for virus neutraliz(cid:173)
`ation52.
`Conclusions
`The structure of a complex between an antibody and influenza
`virus neuraminidase shows features inconsistent with a rigid
`'lock and key' model for antibody-antigen interactions. In con(cid:173)
`trast to the findings from a lysozyme-antibody complex12 we
`observe (1) an unusual VL-VH pairing in the V module of the
`Fab, and (2) local perturbation of the antigen at the centre of
`the epitope. The interaction therefore has some of the character
`of a handshake. NC41 Fab has not yet been crystallized, and
`we can only suppose on the basis of sequence data that its
`structure is similar to the common pattern among all six other
`Fab structures in the literature. We propose that, during antigen
`binding, the VL and VH domains of the antibody slide at their
`interface. The extent of sliding is small, but sufficient to move
`the CDRs by at least 3 A. If such a structural transition has
`occurred, it is presumably of low energy and therefore low cost
`to the binding affinity for the antigen. The observation that
`V L - V H dimers associate in a way not found in associations of
`other /3-sheet structures has led to the suggestion that this novel
`association might produce static or dynamic properties impor(cid:173)
`tant for antigen binding13
`• The ability of domains to slide could
`expand the range of specificities of a single antibody species.
`Cross-reactivity of monoclonal antibodies with proteins unre(cid:173)
`lated to the immunizing antigen53 could occur by different extent,
`or direction, of domain sliding.
`Diversity of antibody specificities is generated by libraries of
`germline gene sequences, by somatic events and by combination
`of different H and L chains54. If the V module can generate
`diversity through V L - V H sliding, this presents yet another
`mechanism for fine-tuning the specificity of an antibody. We
`do not suggest that all antigen-antibody interactions will involve
`
`reoryanization of the V L - V H interface. The lysozyme-Fab com(cid:173)
`plex 2 apparently does not. Thus the only biological significance
`we attach to V L - V H sliding is that, in some instances, it may
`facilitate the formation of a functional immune complex.
`A necessary, but apparently insufficient, condition to induce
`sliding is the interaction of the epitope with CDRs from both
`H and L chains. It may be that the larger radius of curvature
`of neuraminidase, compared with lysozyme, restricts the capac(cid:173)
`ity of the Fab to bind in its ground state.
`Sequence data on the T-cell receptor are consistent with the
`a- and /3-chains beingimmunoglobulin-like55·56. We note further
`that those sequences contain the characteristics of VL and VH
`sequences that have been attributed to the novel association of
`VL and VH domains15. If the principles ofrecognition of antigens
`by T- and B-cell receptors are indeed similar57, then induced fit
`of antigen to receptor might also be a feature of cellular
`immunity.
`As the two reported antigen-antibody complex structures are
`rather different, more data on similar systems are required to
`establish the importance of VL-VH sliding to immune rec(cid:173)
`ognition.
`We thank Paul Davis for computing assistance, Bert van
`Donkelaar and Janet Newman for technical assistance, Cyrus
`Chothia for communicating results in advance of publication
`and the Australian Overseas Telecommunications Commission
`for donating international dialling facilities. Some of the OX2-
`NC41 data were collected at SSRL and we thank Paul
`Phizackerly for assistance. This work was supported by the US
`National Institutes of Health.
`Note added in proof: The NC41 VH belongs to the recently
`characterized VH family IX (Winter, E., Radbuch, A. and
`Krawinkel, V. EMBO J. 4, 2861-2867 (1985)).
`
`Received 11 November 1986; accepted 3 February 1987.
`I. Colman, P. M. & Ward, C. W. Cu". Top. Microbiol Immun. 114, 178-255 (1985).
`2. Kabat. E. A., Wu, T. T., Bilofsky, H., Reid-Miller, M. & Perry, H. Sequences of Proteins
`of Immunological lntereJt (US Department of Health and Public Services, Washington
`DC, 1983).
`3. Schiffer, M., Girling, R. L., Ely, K. R. & Edmundson, A. B. Biochemistry 12, 4620-4631
`(1973).
`4. Poljak, R. J. et al Proc. natn. Acad. Sci U.S.A 70, 3305-3310 (1973).
`5. Segal, D. M. et al Proc. natn. Acad. Sci U.S.A 71, 4298-4302 (1974).
`6. Matsushima, M. et al. J. molec. Biol 121, 441-459 (1978).
`7. Saul, F., Amzel, L. M. & Poljak, R. J. J. biol Chem. 253, 585-597 (1978).
`8. Marquart, M., Deisenhofer, J., Huber, R. & Palm, W. J molec. Biol 141, 369-392 (1980).
`9. Satow, Y., Cohen, G. H., Padlan, E. A. & Davies, D. R. J molec. Biol 190, 593-604 (1986).
`10. Suh, S. W. et al Proteins: Structure, Function and Genetics, 1, 74-80 (1986).
`11. Colman, P. M. &. Webster, R G. in Biological Organisation: Macromolecular Interactions
`at High Resolution (ed. Burnett, R.) 125-133 (Academic, New York, 1987).
`12. Amit, A.G.