Structure of antibody hypervariable loops
`reproduced by a conformational
`search algorithm
`
`Robert E. Bruccoleri*, Edgar Haber* & Jiri Novotny*
`
`Cellular and Molecular Research Laboratory,
`Massachusetts General Hospital & Harvard Medical School, Boston,
`Massachusetts 02114, USA
`
`_sM������������������-LflTERSTQNATLJRE����� ��� �NA_ T_U_R_E_v _o _L_. 3_3_5_6_o _c_TO_B_E_R _1_98_8
`11. Triezenberg, S. J., LaMarco, K. L. & McKnight, S. L. Genes Dev. 2, 730-742 (1988).
`wild-type GAL4 or GAL4-Bl7-the latter activator is GAL4
`12. Triezenberg, S. J., Kingsbury, R. C. & McKnight, S. L. Genes Dev. 2, 718-729 (1988).
`(1-147) fused to an acidic activating region encoded by an
`R. Cell 52, 435-445 (1988).
`13. O'Hare, P. & Goding, C.
`Escherichia coli genomic fragment2• Figure 2 also shows that
`14. Preston, C. M., Frame, M. C. & Campbell, M. E. M. Cell 52, 425-434 (1988).
`15. McKnight, J. L. C., Kristie, T. M. & Roizman, B. Proc. narn. Acad. Sci. U.S.A. 84, 7061-7065
`GAL4-VP1 6 worked, albeit about 10-fold less efficiently, when
`(1987).
`the UASG was positioned at -1180 or +1850. Using a different
`16. West, R., Yocum, R. & Ptashne, M. Malec. cell. Biol 4, 2467-2478 (1984).
`17. Giniger, E. & Ptashne, M. Proc. natn. Acad. Sci. U.S.A. 85, 383-386 (1 988) .
`cell line (CHO-DUKX (ref. 6)), GAL4-VP16 worked at about
`18. Lin, Y.-S. et al. Cell 54, 659-664 (1988).
`the same level of efficiency from all three positions (Fig. le).
`19. Gill, G. & Ptashne, M. Narure 334, 721-724 (1988).
`20. Ptashne, M. Nature (1988).
`Activity was not detected in either cell line for either GAL4 or
`21. Cato, A., Miksicek, R., Schuz, G., Arnemann, J. & Beato, M. EMBOJ. 5, 2237-2240 (1986).
`GAL4-Bl7 when the UASG was positioned at -1180 or +1850
`22. Ellis, L. et al. Cell 45, 721-732 (1986).
`23. McGeady, M. L. et al. DNA 5, 289-298 (1986).
`and in no case did GAL4 (1-147) activate, confirming previous
`results6• The start sites of the induced RNAs for all constructs
`were identical to those of the glucocorticoid-receptor-stimulated
`transcripts initiated from the wild-type MMTV L TR (not
`shown). Immunoprecipitation and western blotting of the
`various activator proteins revealed that all were produced at
`equal levels (within a factor of two or three) in both cell types
`(data not shown). In all of these experiments the reporter bears
`a binding site for the mammalian activator nuclear factor (NFl)
`near the TAT A sequence, and we do not know what contribu­
`tions, if any, this protein made to the activation we observed.
`It is possible, for example, that the differences observed between
`the two cell lines reflect different concentrations of this auxiliary
`factor.
`GAL4-VP16 activated only slightly more efficiently from a
`UASG (which includes four GAL4 binding sites) than from a
`single strong binding site (17 mer, Fig. 1 d), indicating that
`activation may be near-maximal from the single site. In addition,
`GAL4-VP1 6, bound at position -110, activates transcription
`about 10-fold more efficiently than the glucocorticoid receptor
`(GR) bound to the multiple sites found in the wild-type MMTV
`promoter (Fig. ld). Finally, Fig ld shows that GAL4-VP16
`bound at -110 elicits a level of CAT activity that is about that
`produced by a CAT gene driven by an adjacent simian virus 40
`(SV40) or Rous sarcoma virus (RSV) promoter. For reasons we
`do not understand, using an SV40 promoter as a measure, we
`find that wild-type.GAL4 is substantially less active than repor­
`ted by Webster et ar.7.
`Why is GAL4-VP16 such a powerful activator? It is highly
`unlikely that the distinctive properties of the molecule can be
`ascribed to, for example, increased DNA binding. A variety of
`experiments suggest that GAL4 (1-147) forms stable dimers
`(Carey, M., Kakidani, H. and M.P., in preparation) and that
`this fragment, as well as other GAL4 derivatives, efficiently fill
`the GAL4 binding sites in vivo17• Instead, we imagine that the
`acidic region of VP16 interacts with unusual avidity with some
`component of the transcriptional apparatus, perhaps the TATA­
`binding factor TFIID (ref. 18). In other experiments, we found
`that GAL4-VP16, like native VP16, inhibited transcription
`initiating at promoters lacking GAL4 binding sites. As described
`elsewhere19, we believe this inhibitory effect (which we have
`called squelching19•20) is an unavoidable consequence of
`expression of a strong activating region. If so, the continuous
`expression at high levels of a powerful activator would be
`harmful to cells, but such an activator could profitably be used
`by a virus that expressed it transiently20•
`We are grateful to Steve McKnight for suggesting this experi­
`ment. We thank T. Maniatis for helpful comments. This work
`was supported by grants from the American Cancer Society and
`National Institutes of Health to M.P. I.S. is a K.M. Hunter
`P.D.F. of the National Cancer Institute of Canada.
`
`The antigen-combining site of antibody molecules consists of six
`
`separate loops supported by a conserved fl-sheet framework; anti­
`
`
`body specificity arises from length and sequence variation of these
`
`
`
`'hypervariable' loops1 and can be manipulated by transferring sets
`
`of loops between different frameworks2• Irregular loops are the
`
`most difficult parts of protein structure to understand and to model
`
`
`correctly3--<>. Here, we describe two computer experiments where
`
`
`all the hypervariable loops were deleted from X-ray structures of
`
`
`mouse immunoglobulins and reconstructed using the conforma­
`tional search program CONGEN 7• A protocol was developed for
`
`
`reconstruction of the hypervariable loops in McPC 603 antibody.
`
`
`Calculated loop conformations were generated and a model of the
`
`combining site was built from selected low-energy conformations.
`
`We then modelled hypervariable loops in another antibody
`molecule, HyHEL-5. Both models agreed well with the known
`
`crystal structures. Our results hold out promise for the success of
`
`future modelling studies of complete antigen-combining sites from
`amino acid sequences.
`Current techniques of loop-structure modelling, and anti­
`body-binding site modelling in particular8-13, involve a variety
`of methods. Protein conformation is determined by values of
`backbone and side-chain torsional angles <P (C-N-Ca-C), rf!
`(N-Ca-C-N) and x (N-Ca-Cf3-Cy, Ca-Cf3-Cy-C8 and so on).
`Three-dimensional modelling involves finding specific values
`for all these torsional degrees of freedom. A particularly promis­
`ing approach is therefore to use automatic computer algorithms
`that uniformly sample the complete conformational space of a
`polypeptide chain segment7'14-16• Theoretically, all the backbone
`and side-chain conformations compatible with the rest of the
`protein structure can be generated. The lowest free-energy con­
`formation should correspond to the naturally occuring one. In
`practice, technical problems associated with an exhaustive
`sampling of conformational space restrict searches to short
`polypeptide segments only. A successful protocol for hypervari­
`able loop reconstruction could later be extended to automatic
`modelling of combining sites of unknown three-dimensional
`structures solely from their amino acid sequences.
`We used X-ray atomic coordinates of
`the mouse
`immunoglobulin McPC 603 (ref. 17) as our starting structure,
`and defined endpoints of the hypervariable loops as shown in
`Fig. 1. These six loops, 46 amino acid residues altogether, were
`deleted from the structure. The shortest loop, Hl, contained five
`
`Received 25 May; accepted 2 September 1988
`I. Ma, J. & Ptashne, M. Cell 48, 847-853 (1987).
`2. Ma, J. & Ptashne, M. Cell SI, 113-119 (1987).
`3. Brent, R. & Ptashne, M. Cell 43, 729-736 (1985).
`4. Hope, I. & Struhl, K. Cell 46, 885-894 (1986).
`5. Gill, G. & Ptashne, M. Cell SI, 121-126 (1987).
`6. Kakidani, H. & Ptashne, M. Cell 52, 161-167 (1988).
`7. Webster, N., Jin, J. R., Green, S., Hollis, M. & Chambon, P. Cell 52, 169-178 (1988).
`A., Giniger, E., Maniatis, T. & Ptashne, M. Nature 332, 850-853 (1988).
`8. Fischer, J.
`9. Ma, J., Przibilla, E., Hu, J., Bogorad, L & Ptashne, M. Nature 334, 631-633 (1988).
`IO. Giniger, E. & Ptashne, M. Nature 330, 670-672 (1987).
`
`*Present address: The Squibb Institute for Medical Research. Princeton, New Jersey 08543-4000,
`USA
`
`1 of 5
`
`BI Exhibit 1077
`
`

`

`Fig. I Stereoscopic view of the hyper­
`variable loops of McPC 603 with the
`underlying J3-sheets. By tight non­
`covalent association, two curved /3-
`sheets from the VL and VH domains
`create
`a
`twisted,
`elliptical J3-
`barrel'·21·22. The polypeptide back­
`bone of this barrel can be approxi­
`mated by mathematical hyperboloidal
`surfaces22 and the long axis of the
`least-squares fitted surface becomes a
`convenient reference axis of the J3-
`barrel. The fitted axis is drawn as a
`white vertical line, the /3-sheet back­
`bones are drawn in blue lines (the VL
`domain sheet in front, the VH domain
`sheet behind and the six hypervariable
`loops (heavy-chain loops Hl, residues
`28-32; H2,
`residues 50-58; H3,
`residues I02-109; light-chain loops Ll,
`residues 26-37; L2, residues 56-6I; L3,
`residues 97-102; consecutive numbering system) are in red and green. The three shorter loops, HI, L2 and L3, drawn in red, have lower
`average axis coordinates, so they are closer to the centre of the barrel than the other three loops, H2, H3 and L1, drawn in green. In our
`construction protocol, the low loops were constructed before the high loops.
`chosen on the basis of the relative positions of the loops in the
`residues; the longest, Ll, contained 12 (see Fig. 1 legend for
`loop nomenclature).
`local frame of reference provided by the ,B-sheets of the VL-VH
`domain interface21•22 (Fig. 1). The tight non-covalent interaction
`The CONGEN search algorithm7 uses a selectable angular
`of these sheets creates an elliptical ',B-barrel' whose long axis
`grid to sample the torsional degrees of freedom in a given
`polypeptide segment, using the amino- and carboxy-terminals
`nearly coincides with the two fold axis of VL-VH domain
`as fixed endpoints. In the searches, both cis and trans praline
`pseudosymmetry. In this reference frame, the three shorter loops
`peptide bonds are considered. Each sampling run generates a
`(L2, L3 and Hl) are placed lower (closer to the midpoint of
`set of loop conformations which satisfy the fixed endpoint
`the ,B-barrel interface) than the others. They do not interact with
`condition (the modified Go and Scheraga chain-closure
`each other and provide a natural basis for the construction of
`algorithm)18•19 and have acceptable potential energies20• These
`the remaining 'high' loops, H2, H3 and Ll. These 'low' loops
`were built first, H2 next, and loops H3 and Ll, which interact
`energies are used to distinguish between 'good' and 'bad' con­
`closely, were built last.
`formations and depend critically on short-range interactions
`between loops and, consequently, on the order of loop construc­
`From the set of conformations generated for each loop, 'the
`best one' was selected to be incorporated into the final model
`tion. Our order of loop reconstruction L2-Hl-L3-H2-H3-Ll was
`
`Table I Description of McPC 603 combining site reconstruction
`
`Loop
`L2
`
`HI
`
`L3
`
`H2
`
`H3
`
`L1
`
`Run and
`conformation
`number
`56LMCP7-75*
`56LMCP7-89
`28HMCPI-44*
`28HMCPI-43
`97LMCP7-42*
`97LMCP7-95
`50HMCP5-8I
`50HMCP5-75
`50HMCP5-7I
`50HMCP5-76
`50HMCP5-I 70*
`50HMCP5-I75
`50HMCP27-68I *
`50HMCP27-874
`I02HMCP17-I76*
`I02HMCP17-408
`26LMCP23-1344*
`26LMCP23-1348
`
`Total
`conformations
`found
`159
`
`46
`
`306
`
`576
`
`9I2
`
`511
`
`2,287
`
`Energy
`(kcal)
`-14.6
`-16.1
`-I7.I
`-I0.7
`-20.8
`-21.8
`-I8.8
`-I8.5
`-I8.2
`-I7.9
`-17.8
`-I7.8
`-39.6
`-33.8
`-64.5
`-62.2
`-49.0
`-46.9
`
`Root mean square (A)
`Total
`Backbone
`1.6
`1.9
`2.4
`2.0
`0.7
`1.7
`1.6
`0.7
`1.4
`0.8
`3.0
`1.5
`2.I
`2.0
`2.2
`2.2
`2.I
`2.I
`2.2
`2.I
`1.6
`1.0
`1.0
`1.7
`2.4
`1.9
`1.5
`4.0
`2.9
`1.1
`2.7
`0.9
`4.0
`3.5
`3.8
`4.4
`
`Surface
`cA2)
`
`360
`397
`344
`329
`320
`375
`lOI
`I26
`IOO
`I23
`80
`96
`550
`584
`369
`394
`799
`810
`
`Six polypeptide chain segments (see Fig. I) were deleted from the McPC 603 structure and were reconstructed in separate computer runs, in the order listed in the
`table. In the L2 loop, residues 61-59 were searched in the C-> N direction and the Go-Scheraga chain-closing algorithm7'19 was applied to residues 56-58. In the HI
`loop residues 28 and 32 were searched and the chain-closing was applied to residues 29-31. In the L3 loop, residues 97-99 were searched successively and chain-closing
`applied to residues 100-102. In the H3 loop the residues were searched successively from the amino-terminus and the chain-closing routine was applied to residues
`107-109. The H2 and LI loops were too long to be constructed as one segment and 'real space renormalization'30 was applied. In H2, the N-terminal tripeptide
`Ala-Ser-Arg was searched first (see the top six conformations listed in the table). From the six lowest-energy conformations, all within a I kcal interval, the one with
`the smallest surface was taken as the starting conformation for the search over the rest of the loop (see the lower two conformations listed in the table). In the LI loop,
`three separate runs constructed dipeptides Ser-Gin, Ser-Leu and Leu-Asn. In the second and third runs, the four lowest-energy conformations from the previous run
`were all included as starting conformations. The C-terminal di peptide Lys-Asn was then constructed separately. Finally, the lowest-energy conformations of the peptides
`Ser-Glu-Ser-Leu-Leu-Asn (residues 26-31) and Lys-Asn (residues 36-37) were included as starting conformations into the search over the rest of the loop. The angular
`grid on which the torsional spaces were searched was 30° and the non-bonded interactions were computed within the cut-off distance of 5 A, with dielectric constant,
`'= 50 (the three 'low' loops of Fig. I), or using cut-off distance of 8 A, with distance-dependent dielectric constant (the three 'high' loops of Fig. I). The lowest
`conformations of each run are ,given; *, indicates those included in the model in Fig. 2.
`
`2 of 5
`
`BI Exhibit 1077
`
`

`

`566
`
`LETIERSTO NATURE
`
`NATURE VOL. 335 6 OCTOBER 1988
`
`Table 2 Description of HyHEL-5 combining site reconstruction
`
`Loop
`L2
`
`Hl
`
`L3
`
`H2
`
`H3
`
`L1
`
`Run and
`conformation
`number
`49LHYHEL55-76*
`49LHYHEL55-89
`49LHYHEL55-20
`28HHYHEL55-21 *
`28HHYHEL55-22
`90LHYHEL56-21 *
`90LHYHEL56-18
`50HHYHEL56-4*
`50HHYHEL56-3
`100HHYHEL55-l 1 *
`100HHYHEL55-2
`26LHYHEL55-11
`26LHYHEL55-7*
`
`Total
`conformations
`found
`100
`
`37
`
`33
`
`6
`
`14
`
`13
`
`Energy
`(kcal)
`-57.6
`-56.3
`-55.2
`-26.1
`-22.9
`-20.4
`-5.5
`-0.3
`2.5
`-47.3
`-46.7
`-35.1
`-32.8
`
`Root mean square (A)
`Total
`Backbone
`0.8
`1.7
`2.3
`1.3
`1.1
`1.9
`1.8
`1.1
`1.2
`1.9
`4.1
`1.1
`4.4
`1.4
`3.1
`2.1
`3.2
`2.1
`2.7
`1.0
`1.5
`0.9
`2.0
`0.4
`1.8
`0.6
`
`Surface
`(A2)
`268
`294
`286
`489
`498
`357
`368
`205
`214
`359
`381
`367
`362
`
`Amino acid sequences of variable domains of McPC 603 and HyHEL-5 light and heavy chains were aligned to maximize sequence homology (43% identical residues
`in the VH domain and 54% identical residues in the VL domain). Endpoints of HyHEL-5 hypervariable loops were then determined as residue positions coinciding
`with the endpoints of McPC 603 hypervariable loops determined previously (see Fig. 3 legend). The six HyHEL-5 hypervariable loops defined in this way (32 residues
`altogether) were deleted from the HyHEL-5 crystallographic structure and reconstructed using the protocol developed for construction of the McPC 603 combining
`site. The order of loop construction, Go-Scheraga chain-closure, and so on was as described in Table I. More specifically, the L2 loop residues 54-52 were searched in
`the C-> N direction and the chain-closing algorithm7"19 was applied to residues 49-51. In the HI loop the residues 28 and 32 were searched and chain-closing applied
`to residues 29-3 I. In the L3 loop, the residues 90 and 91 were searched successively and chain-closing applied to residues 92-94. In the H2 loop the N-terminal
`tetrapeptide 50-53 was searched first and the chain-closing was applied to residues 54-56. In the H3 loop residue 100 was searched and chain-closing applied to residues
`101-103. In Ll, the residues 26 and 30 were searched and chain-closing applied to residues 27-29. Because of the short lengths of the HyHEL-5 loops H2 and Ll,
`whole loops could be constructed in single search runs. The angular grid on which the torsional spaces were searched was 30° and the non-bonded interactions were
`computed using cut-off distance of 8 A, with the distance-dependent dielectric constant used for all six loops. The lowest-energy conformations of each run are given;
`* indicates those included in the model displayed in Fig. 3.
`
`of McPC 603 as follows. All the conformations were ranked by
`their empirical potential energies20; the two lowest-energy con­
`formations were examined in detail. If calculated energies of
`these two structures differed by about 2 kcal (three times the
`Boltzmann factor kT, indicating that both conformations are
`well populated at room temperature), the one with the smaller
`solvent-exposed surface23 was selected. Otherwise, the lowest­
`energy structure was selected. The H2 and Ll loops were too
`long to be constructed in a single computer run. Instead, selected
`results of searches over shorter sections of the loop were included
`into new searches over the rest of the loop as starting conforma­
`tions (see Table 1). The success of our reconstruction was gauged
`against the original X-ray structure, by evaluating root-mean­
`square (r.m.s.) deviations between the computed and the crystal
`conformations.
`Details of individual loop constructions in McPC 603 are
`described in Tables 1 and 3. Overall, the best r.m.s. conforma­
`tions were not often the lowest energy ones and therefore could
`not be selected in a true modelling experiment where the final
`structure would be unknown. Nevertheless, there were low­
`energy conformations generated which still matched the native
`structure fairly well. The r.m.s. agreement of our model, con­
`structed from the few selected lowest-energy conformations, is
`2.4 A total and 1.7 A backbone r.m.s. shifts from the X-ray
`structure (Fig. 2). The model shows close agreement with the
`crystal; for example, the rare cis peptide bond conformation of
`Pro LlOl was accurately reproduced. Two isolated regions show
`large disagreements with the crystal structure: the tyrosine side
`chain ring H103, which had been misplaced by rotation of about
`160° around its Xi torsional angle (Fig. 2d) and the backbone
`of the middle of the L1 loop, residues 32-35 (Fig. 2b ).
`Each of the six McPC 603 loops was very closely matched
`by at least one of the computed conformations. If these 'best'
`r.m.s. constructions could be assembled into a model, its shift
`from the X-ray structure would be 1.7 A, 1.3 A on the backbone
`only. This level of accuracy approaches the limits of atomic
`resolution of the McPC 603 structure17 (0.5- 1.0 A).
`The reconstruction protocol developed for McPC 603 hyper­
`variable loops was next applied to another antibody structure,
`HyHEL-5, to check its generality. The six HyHEL-5 hypervari­
`able loops (32 residues altogether) were reconstructed essentially
`
`as described for the McPC 603 loops; the only change to the
`above construction protocol was a consistent use of the distance­
`dependent dielectric constant, e = R (R is distance between the
`two interacting charges), in the evaluation of electrostatic inter­
`actions. As for McPC 603, each of the hypervariable loops was
`matched closely by at least one calculated conformation. The
`model of the HyHEL-5 combining site, made from selected
`lowest-energy conformations showed a 2.6 A r.m.s. shift from
`the X-ray structure, 1.4 A r.m.s. from the backbone. Thus, our
`construction scheme developed on the McPC 603 molecule
`could be used, with only small technical modifications, on a
`different molecule to give an equally good result.
`Figure 3 shows details of our HyHEL-5 model. Loop back­
`bones show good agreement with the X-ray structure, except
`for the H2 loop residues H51-H54. Most of the side chains
`match the X-ray structure equally well, the exceptions being
`Tyr HlOl and Arg L92. The poor agreement in the H2 modelling
`is due to inaccurate positioning of the side chain Phe H29 during
`the Hl loop construction. The shift in the Phe ring position,
`
`Table 3 CONGEN reconstruction of combining sites in antibodies
`McPC 603 and HyHEL-5
`
`Root mean square
`to crystal (A)
`
`Loop
`HI in McPC 603
`HI in HyHEL-5
`H2 in McPC 603
`H2 in HyHEL-5
`H3 in McPC 603
`H3 in HyHEL-5
`L1 in McPC 603
`L1 in HyHEL-5
`L2 in McPC 603
`L2 in HyHEL-5
`L3 in McPC 603
`L3 in HyHEL-5
`All loops, McPC 603
`All loops, HyHEl-5
`
`Length
`5
`5
`9
`7
`8
`4
`12
`5
`6
`6
`6
`5
`46
`32
`
`total
`1.7
`1.8
`2.1
`3.1
`2.9
`2.7
`3.0
`1.8
`1.9
`1.7
`1.4
`4.1
`2.4
`2.6
`
`backbone
`0.7
`1.1
`1.6
`2.1
`1.1
`1.0
`2.6
`0.6
`1.6
`0.8
`0.8
`1.1
`1.7
`1.4
`
`CPU time*
`4 hours
`3 hours
`5 days
`20 minutes
`7 days
`2 hours
`7 days
`40 minutes
`8 hours
`37 hours
`5 hours
`12 hours
`20 days
`2.25 days
`
`*Time on central processing unit, a Micro Vax II.
`
`3 of 5
`
`BI Exhibit 1077
`
`

`

`a
`
`b
`
`c
`
`d
`
`Fig. 2 Stereoscopic view of the Mc PC 603 anti­
`gen combining site, as reconstructed by CON­
`GEN (see Table 1). Light lines trace the crystallo­
`graphic structure, heavy lines are CONGEN­
`generated conformations. a, Comparison of all
`six hypervariable loops in the X-ray structure and
`those constructed by CONGEN. The loops are
`clockwise from the top: H2, H-chain residues
`50-58 (Ala-Ser-Arg-Asn-Lys-Gly-Asn-Lys-Tyr);
`L3, L-chain residues 97-102 (Asp-His-Ser-Tyr­
`Pro-Leu); Ll, L-chain residues 26-37 (Ser-Gln­
`Ser-Leu - Leu-Asn-Ser-Gly-Asn-Gln- Lys-Asn);
`L2, L-chain residues 56-61 (Gly-Ala-Ser-Thr­
`Arg-Glu); H3, H-chain residues 102-109 (Tyr­
`Tyr-Gly-Ser-Thr-Trp-Tyr-Phe); and HI, H-chain
`residues 28-32 (Thr-Phe-Ser-Asp-Phe). b, Back­
`bones of the framework residues outside the
`hypervariable loops are traced through a-carbons
`only, hypervariable loop backbones are drawn as
`N, Ca and C atoms. Orientation of the combining
`site was as above. The middle part of the L1 loop,
`where there is the greatest discrepancy between
`the crystallographic and calculated structures, is
`poorly visible and disordered in the crystal17• c,
`A detailed picture of the L3 loop (total r.m.s.
`shift between the crystallographic and calculated
`conformations 1.4 A, backbone shift 0.8 A). Note
`the cis proline residue LlOl. d, A detailed picture
`of the H3 loop (total r.m.s. shift between the
`crystallographic and calculated conformations
`2.9 A, backbone shift 1.1 A). Note the displace­
`ments of the side-chain ring Tyr H 103 and, to a
`lesser extent, Trp HI07. These inaccuracies are
`largely responsible for the unsatisfactory r.m.s.
`value.
`
`4 of 5
`
`BI Exhibit 1077
`
`

`

`-5�����·��������--������LETTERSTQ NATLJRf ���������-N_A _TU_ R_E_ v_ o_L_._3_3 5_ 6�o_c T_o_B_E_R _t_9s_s
`
`a
`
`b
`
`Fig. 3 Stereoscopic view of the HyHEL-5
`antigen combining site, as reconstructed by
`CONGEN (see Table 2). Light lines trace the
`crystallographic structure, heavy lines are
`CO NG EN-generated conformations. a, Com­
`parison of all six hypervariable loops in the
`X-ray structure and as constructed by CON­
`GEN. The loops are, clockwise from the top:
`H2, H-chain residues 50-56 (Glu-Ile-Leu­
`Pro-Gly-Ser-Gly); L3, L-chain residues 90-94
`(Trp-Gly-Arg-Asn-Pro); Ll, L-chain residues
`26-30 (Ser-Ser-Ser-Val-Asn); L2, L-chain
`residues 49-54 (Asp-Thr-Ser-Lys-Leu-Ala);
`H3, H-chain residues 100-103 (Asp-Tyr-Asp­
`Phe); and Hl, H-chain residues 28-32 (Thr­
`Phe-Ser-Asp-Tyr). b, Backbones of
`the
`framework residues outside the hypervariable
`loops are traced through a-carbons only,
`hypervariable loop backbones are drawn as
`N, Ca and C atoms. Orientation of the com-
`bining site is as above.
`
`although small, interfered with the correct placement of Pro
`H54 because of atomic volume overlaps and resulted in mis­
`placement of the middle part of this loop. Discrepancies like
`these provide us with important stimuli for future work. We
`expect that improvements in energy ranking (for example, a
`better representation of solvent-modified electrostatic interac­
`tions) will allow selection of better-quality conformations and
`construction of more accurate models. Empirical free-energy
`functions24-27 should be particularly useful. Likewise, canonical
`structures developed by Chothia and Lesk13, could be used to
`improve on the loop selection process.
`
`The conceptual basis for the construction of polypeptide loops
`on a fixed backbone framework is only beginning to be under­
`stood. A priori, there was no guarantee that local potential
`energies of the loops and their local environments would lead
`to accurate models of the structure. The success of the construc­
`tion protocol may imply that local interactions in protein struc­
`ture28, particularly loops29, are more important than generally
`believed.
`This work was supported by grants from National Institute
`of Health and the Office for Naval Research. The program
`CONGEN is available by request from R.E.B.
`
`Received 7 March; accepted 22 August 1988.
`l. Wu. T. T. & Kabat, E. A. J. exp. Med 132. 211-250 ( 1970).
`2. Jones, P. T., Dear, P. H., Foote, J., Neuberger, M. S. & Winter, G. Nature 321, 522-525
`(1986).
`3. Richardson. J. Adv. Protein Chem. 34, 167-339 (1981).
`4. Sibanda, B. L. & Thornton, J. Nature 316, 170-174 (1985).
`5. Rose, G. D., Gierasch. I. M. & Smith. J. A. Adv. Protein Chem. 37, 1-50 (1985).
`6. Leszczynski, J. F. & Rose. G. D. Science 234, 849-855 (1986).
`7. Bruccoleri, R. E. & Karplus, M. Biopolymers
`26, 137-168 (1987).
`8. �abat, E. A. & Wu, T. T. Proc. natn. Acad. Sci. U.S.A. 69, 960-964 (1972).
`9. Chothia, C. et al. Science 233, 755-758 (1986).
`10. de la Paz, P., Sutton, B. J., Darsley, M. J. & Rees,
`A . R. EMBO J. 5, 415-425 (1986)
`R. M., Wang, H., Shenkin, P. S., Yarmush, D. L. & Levinthal, C. Proteins 1, 342-362
`11. Fine,
`(1986).
`12. Smith-Gill, S. J. et al. J. molec. Biol. 194, 713-724 (1987).
`13. Chothia, C. & Lesk, A. M. J. molec. Biol. 196, 901-917 (1987).
`14. Shih, H. L., Brady, J. & Karplus, M. Proc. natn. Acad. Sci. U.S.A. 82, 1697-1700 (1985)
`
`IS. Snow, M. E. & Amzel, M. L. Proteins I, 276-279 (1986)
`16. Moult, J. & James, M. N. G. Proteins l, 146-163 (1986).
`17. Satow, Y., Cohen, G. H., Padlan, E. A. & Davies, D.R. J. molec. Biol. 190, 593-604 (1986).
`A
`. Macromolecules
`18 Go, N. & Scheraga, H.
`3, 178-187 (1970).
`19. Bruccoleri, R. E. & Karplus, M. Macromolecules
`18, 2767-2773 (1985).
`20. Brooks, B. et al.]. comput. Chem. 4, 187-217 (1983).
`21. Chothia, C., Novotny, J .• Bruccoleri, R. E. & Karplus, M. J. molec. Biol. 186, 651-663 (1985).
`22. Novotny. J. & Haber. E. Proc. natn. Acad. Sci. U.S.A. 82, 4592-4596 (1985).
`23 Lee. B. K. & Richards, F. M. J. molec. Biol. SS, 379-400 (1971).
`24. Rashin, A. A. Biopolymers
`23, 1605-1620 (1984).
`25. Novotny, J., Bruccoleri, E. R. & Karplus. M. J. molec. Biol. 177, 787-818 (1984).
`26. Eisenberg, D. & Mclachlan, A. Nature 319, 199-203 (1986).
`27. Ooi, T., Oobatake, M., Nemethy, G. & Scheraga, H. A. Proc. natn. Acad. Sci. U.S.A. 84,
`3086-3090 (1987).
`28. Chou, P. Y. & Fasman, G. Biochemistry
`16, 222-244 (1974).
`29. Dyson, H.J. et al. Nature 318, 480-483 (1985).
`A
`. Proc. natn. Acad. Sci. U.S.A. 79, 5107-5110
`30. Pincus, M. R., Klausner, R. D. & Scheraga, H.
`(1982).
`
`5 of 5
`
`BI Exhibit 1077
`
`