PROTEINS: Structure, Function, and Genetics 34:197–205 (1999) Molecular Dynamics Study of the Proposed ␤-Hairpin Form of the Switch Domain From HIV1 gp120 Alone and Complexed With an Inhibitor of CD4 Binding Andreas Graf von Stosch,1 C.W. von der Lieth,2 and Jennifer Reed1* 1Department of Pathochemistry, German Cancer Research Center, Heidelberg, Germany 2Department of Spectroscopy, German Cancer Research Center, Heidelberg, Germany ABSTRACT The strong tendency of ␤-hairpin peptides to aggregate can prevent their structural resolution. The polar form of the switch peptide (LAV 15mer) at the CD4-binding domain of HIV1 gp120 is such a peptide, and NMR investigations of its interaction with a class of CD4-binding inhibitors developed in this laboratory have been hindered. Detailed knowledge of the interaction is required for the development of more potent switch inhibitors, that act by disrupting the cooperative folding transition necessary for binding to the CD4 receptor. In carrying out molecular dynamics simulation of the free peptide under polar conditions, we found that the properties of the resulting structure agree closely with those observed by circular dichroism. The same conditions, used to model the peptide/ inhibitor complex, produced a stable bimolecular structure with specific interactions between the inhibitor and side chains on the peptide, (e.g., Trp12 and the LPCR tetrad), known to control the folding transition. These help explain existing data on the relative potency of inhibitor derivatives and provide a basis for improved inhibitor design. Proteins 1999;34:197–205. r 1999 Wiley-Liss, Inc. Key words: gp120; switch domain; molecular dynamics; peptide INTRODUCTION The process involved as the human immunodeficiency virus 1 (HIV1) env glycoprotein gp120 binds to the CD4 receptor on host cells presents an interesting problem in protein structure/function from both the theoretical and the practical point of view. Previous work in this laboratory has established that a 15-residue sequence from the principal CD4-binding domain of gp120 conserves the ability to refold abruptly and cooperatively from ␤-sheet to helix, triggered by changes in medium polarity across a very narrow range.1 This phenomenon was observed in the corresponding sequence from all strains tested despite 50% variability and appears to be closely involved in the ability of gp120 to bind to the CD4 receptor, i.e., destroying the cooperativity of the refolding process—the conformational switch—also stops binding to CD4-expressing cells.1,2 Although certain aspects of the mechanism behind this process have been reported,3,4 and a nuclear magnetic r 1999 WILEY-LISS, INC. resonance (NMR) structure of the helical form could be obtained,5 all attempts to date at determining the fine structure of the ␤-sheet form observed by circular dichroism have been frustrated by the extreme tendency of the peptide to aggregate at the concentrations necessary for NMR. A more accurate idea of this form than can be obtained by CD spectroscopy is desirable, since the exogenous switch inhibitors that have been developed as possible therapeutic agents against HIV12,6 appear to interact with the ␤-conformation. Oligopeptides that have a preference for adopting a ␤-hairpin conformation in solution present considerable difficulties for researchers interested in obtaining an accurate three-dimensional structure. The two most advanced techniques, NMR and X-ray crystallography, can both be frustrated by the tendency of ␤-hairpin peptides to aggregate.5,7–18 Indeed, the few cases in which NMR structure determination of ␤-hairpins has been reported concern artificial peptides purposely designed to avoid aggregation.19–21 On the other hand, the precise structure of natural ␤-hairpins can be important, as several are implicated in pathogenic processes: the ␤-amyloid and the prion protein PrPsc, for example. The polar or unbound form of the switch domain from the env glycoprotein gp120 from HIV1 is a case in point, one where molecular modeling offers the only possibility at present of obtaining a better idea of the structure of this 15-residue sequence and its potential interaction with switch inhibitors. Molecular dynamics simulation of the 15-residue switch domain results in a stable structure that conforms to the conformational parameters established by physical measurements, e.g., circular dichroism. Parallel runs in the presence of a functional inhibitor molecule display a spontaneous and specific association between the two that should prove useful in improving inhibitor design. MATERIALS AND METHODS All calculations were carried out on a Silicon Graphics Indigo II (Extreme) workstation (R4400) coupled with an SP2-IBM parallel processing array. Molecules (the LAV Grant sponsor: Deutsche Forschungsanstalt für Luft- und Raumfahrt; Projektträger Gesundheitsforschung des BMBF. *Correspondence to: Dr. Jennifer Reed, Department of Pathochemistry, German Cancer Research Center, Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany. E-mail: firstname.lastname@example.org Received 7 April 1998; Accepted 5 October 1998 198 A. GRAF VON STOSCH ET AL. peptide and the inhibitor BM 50.0311) were constructed with the Insight II molecular modeling program package (Builder and Biopolymer modules22). The LAV peptide (LPCRIKQFINMWQEV) was constructed with unblocked N- and C-termini to correspond with the peptide used for physical measurements. (The structure of BM 50.0311 and the nomenclature used here are given in the Insert to Fig. 4a.) Energy minimization and calculations of dynamic trajectories were carried out with the Discover program (versions 2.9.5 and 94.0) as part of the Insight II program package. Structures were initially minimized using the conjugate gradient algorithm with the gradient convergence criterium set to 0.0001 kcal/mol Å. The consistent valence force field (CVFF) without any cross-terms was used22 for the peptide as well as for the peptide/inhibitor complex. The potential type and the partial charges were assigned using the automatic assignment procedure of Insight II. The total charge of peptide as well as the peptide/inhibitor complex was zero. No cutoff distance for the nonbonded interaction terms (van der Waals and Coulomb) was introduced for the initial simulation of the LAV peptide structure in the absence of inhibitor (simulation I) or with inhibitor (simulation II). No explicit solvent was included for the initial simulations. An effective dielectric constant of 4.0 was chosen as the closest approximation to the conditions obtaining during the physical measurements, as discussed in van Gunsteren and Berendsen.23 Time steps of 1 fs were chosen for both calculations. The temperatures were set to 400K (I) and 300K (II), respectively. Both simulations were run for a cumulative time of 1.02 ns with images recorded every 250 fs. All simulations were carried out under constant temperature conditions. The temperature and the total energy was monitored to check that the molecular system was equilibrated during the production of the simulation. (This was true for all simulations.) An extended conformation of the LAV peptide was chosed as starting structure in both simulation runs in vacuo. For simulation II, the inhibitor was arbitrarily placed 20–25 Å away from the closest peptide backbone atoms before the molecular dynamics run was initialized. Being aware that the chosen conformation might be dependent on the force field selected and the dielectric constant used, we repeated simulation I applying two different force fields (AMBER and CFF91), three additional dielectric constants (e⫽1,8,80), and reduced the electrostatic contributions scaling them by a factor of 0.5. Visual inspection of series of conformations from the end of the simulations demonstrated—in spite of the inherent flexibility of a peptide of this size—quite similar overall shapes under all tested simulation conditions: a double ␤-hairpin showing the first turn starting after Arg4 and a second one at Ile9 and Asn10. The second turn is more pronounced than the first one. The results confirm that the selected conformation is representative and not artificial due to the use of specific force field or other simulation conditions. Ideally all simulations should be performed using explicit solvent molecules. This would require roughly a 60 Å box with about 25,000 atoms in the primary cell. To enable proper folding of the peptide, the production time should probably cover several nanoseconds (ns). Such a simulation is not feasible even with large supercomputers within a reasonable time. Therefore, simulations III and IV were run with explicit solvent (H2O), using as starting structure a representative frame after stabilization of the in vacuo MD run with LAV peptide and the inhibitor/peptide complex, respectively. The molecules were placed in the middle of a 34 ⫻ 34 ⫻ 34 Å large box, which was filled with 1263 water molecules using the SOAK option of the Biopolymer option in Insight II. The same partial charges as for the simulation without explicit water molecules were applied. A simple point charge model is used for the solvent, where additionally all vibrations of the water molecule are allowed. No SHAKE option was used. Periodic boundary conditions were applied using a double cutoff of 12/13 Å for the evaluation of the nonbonded interactions. After a minimization of 1,000 steps using the conjugate gradient algorithm, an equilibration period of 50,000 steps was followed by a production time of 1,000,000 steps. The integration step was 1 fs, the simulation temperature 300K. Every 500 steps the coordinates were saved. RESULTS AND DISCUSSION Our aim in performing a molecular dynamics simulation of the LAV peptide/inhibitor complex was twofold; we wished to see whether the two molecules would associate spontaneously (and specifically), and we wished to find out which inhibitor/peptide interactions were important for the association. Ideally, we wanted the starting configuration to consist of the peptide in extended form and the inhibitor placed at a sufficient distance from it to ensure that no interaction existed at the beginning of the run (⬃25 Å). As a calculation of this ensemble with explicit water molecules would require prohibitive computation time, we adopted the compromise of carrying out the simulation first in vacuo and then checking the stability of the complex formed—if any—by subjecting it to a further MD simulation with explicit solvent. To test whether this compromise was capable of producing biologically relevant results, we planned to use the technique first on the LAV peptide alone. Simulation conditions that produced a conformation in the isolated peptide that corresponded to the secondary structure characteristics measured by circular dichroism spectroscopy might reasonably be expected to produce a realistic approximation of the behavior of the peptide/inhibitor ensemble. For simulation I, of the LAV peptide in vacuo, stabilization of all main-chain and angles was reached after 350 ps, although most were already established by 100 ps. Visual inspection of a representative frame showed the peptide adopted the form of a double ␤-hairpin with a turn connecting the LPCR tetrad at the N-terminus to the main structure and a second turn at Ile9-Asn10 reversing chain direction. Estimates of the secondary structure content of MOLECULAR DYNAMICS OF AN INHIBITOR/PEPTIDE COMPLEX 199 Fig. 1. Ramachandran plot of the and angles of LAV 15mer residues during the MD simulation in vacuo. Values were taken from 106 frames at 20 frame intervals after stabilization was reached. Shown are the average and angle for each internal residue surrounded by an ellipse whose major and minor axes are the standard deviations for these values. Pro2 ⫽ 䊊; Cys3 ⫽ 䊐; Arg4 ⫽ 䉭; Ile5 ⫽ 䉮; Lys6 ⫽ L; Gln7 ⫽ °; Phe8 ⫽ 䊉; Ile9 ⫽ 䊏; Asn10 ⫽ 䉱; Met11 ⫽ 䉲; Trp12 ⫽ ♦; Gln13 ⫽ ¢; Glu14 ⫽ 䉺. the polar form of the LAV 15mer derived from its far UV circular dichroism spectrum yield values of 50–60% ␤-sheet and roughly two reverse turns.1 A plot of the and angles adopted by each residue during the MD simulation (after full stabilization has been reached) places nine residues in the ␤-region of the Ramachandran map and four in the region typical for 310 helix and reverse turn, in full agreement with the physical measurements (Fig. 1). Once attained, the conformation is quite firm, as can be seen from the relatively small deviations of and angles in Figure 1. A frame from the MD simulation in vacuo after a stable structure had formed was used as the starting configuration for further simulation of the LAV 15mer peptide with solvent (H20). No significant difference was seen in the Ramachandran space on addition of solvent molecules. Figure 2 shows an overlay of the backbone trace at 20 frame intervals during the ca. 1.0-ns run. The simulation continues to depict an orthogonal double ␤-hairpin with a high degree of conservation in the main chain secondary structure. This stability, together with the close correspondence between the properties of the simulated structure and the secondary structure content estimated from the far-ultraviolet (UV) CD spectra, suggest that the MD simulation represents a reasonably realistic model for the conformation of LAV 15mer under polar conditions. The position of the two reverse turns, located at residues 5 and Fig. 2. Stereo pair of backbone traces (106, taken at 20 frame intervals) of the LAV15 peptide simulation in the presence of simulated solvent. Alignment was to the Gln7 amide nitrogen, C␣-carbon and carbonyl carbon. Note the double orthagonal ␤-hairpin form is stable throughout the run. 200 A. GRAF VON STOSCH ET AL. Fig. 3. A single frame from the MD simulation of the LAV 15mer peptide in the presence of solvent showing the typical 90° ring to ring interaction between Pro7, Phe8 and Trp12. 9, is in agreement with the location of these turns in an NMR-derived structure of the peptide in methanol, where partial ␤-structure persists (unpublished results, A. Lindemann, J. Reed). It is also in agreement with the recently published crystal structure of a truncated deglycosylated form of gp120 complexed with a CD4 fragment and an anti-CD4 Fab,24 where the LAV 15mer corresponds to the C-terminal portion of ␤-sheet 19, all of ␤-sheet 20, and the N-terminal portion of ␤-sheet 21.* The modeled structure places the conserved Trp12 side chain in a central position (Fig. 3), a location well suited for participation in the multiple interactions with neighboring side chains that charged-to-alanine experiments indicated are necessary for stabilization of the ␤-type form.3 There are some unexpected aspects of the MD structure that may be relevant to the mechanism of cooperative refolding in this peptide. The LPCR tetrad at the Nterminus of the peptide has been proposed to function as a nucleation site for the polarity-triggered refolding of the switch domain into a 310 helix; it has theoretical potential *The region around Trp427 (Trp12 in the LAV 15mer) has been predicted to be helical in the CD4-bound form of gp120.1,24 In the crystallized complex, however, this region interacts with an antiparallel ␤-sheet array of of the Fab fragment, which may serve to maintain the ␤-structure in solution. for forming a reverse turn, it adopts a well-defined turn structure in isolation that is compatible with the angles of a 310 helix25 and, in its absence, no helix is formed.3 All this would lead one to expect that the model would exhibit a turn centering on these residues. Instead, their and angles map to the extended region of Ramachandran space, and the reverse turn occurs immediately C-terminal to them. A possible explanation may be that the modeled Pro2 residue tends to exist in alignment with the interacting ring systems of Trp12 and Phe8 in the interior of the molecule (Fig. 3). Similar favorable interactions between prolines and aromatic residues have been reported to form the basis of the SH3 domain polyproline binding site.26 Whether this is representative of the situation in vivo must await NMR resolution of the peptide structure, but it is interesting that the helical nucleation site structure adopted spontaneously by the TLPCRI fragment in solution24 is not the dominant form in the modeled ␤-type structure. An MD simulation repeated using the same in vacuo conditions as before, but with the peptide in the presence of the known switch inhibitor BM50.0311 (Fig. 4), resulted rapidly in an extremely stable and specific association between the two molecules. The inhibitor is enclosed in a loop of the peptide, and structural variation in the latter is MOLECULAR DYNAMICS OF AN INHIBITOR/PEPTIDE COMPLEX 201 Fig. 4. Starting configuration of the LAV 15mer peptide and the switch inhibitor molecule for the MD simulation (without explicit solvent) of their interaction. Inset: the inhibitor substance BM 50.0311 with nomenclature used in the text. reduced from that in the free state. Again, a frame from this MD run after stabilization was used as the starting configuration for an MD simulation with solvent. The inhibitor remains in position, inserted in the center of a single ␤-hairpin (Fig. 5a–c). A cluster of favorable interactions that persist, once established, throughout the run appear to be responsible for this (Fig. 6). The principal interactions are as follows: 1. Aromatic/aromatic interactions between the indole ring of Trp12 and the heterocyclic ring of the inhibitor. This association is meta-stable, i.e., the orientation of the two rings varies throughout the run but the distance between their centers is held at 3.5–6.0 Å. Interestingly, there is no tendency after the initial collapse of the peptide for aromatic/aromatic interactions between the inhibitor ring system and the Phe8 residue. As a Trp at position 12 is strictly conserved among HIV 1 strains while the Phe is not, this is encouraging. 2. Charge interactions between the inhibitor heterocyclic C-8 carbonyl and the backbone amides of Gln13 (2.49 ⫾ 0.52 Å) and Arg4 (3.62 ⫾ 0.89 Å). 3. H-bonding between the inhibitor C-2 carbonyl and the indole N of the Trp imidazole (4.02 ⫾ 0.97 Å). This is one of the later associations to form but is not broken once established. 4. H-bonding between the inhibitor side chain amide (N-a) and the peptide backbone carbonyls of Lys6 (2.80 ⫾ 0.71 Å) and Ile9 (2.84 ⫾ 0.54 Å). 5. H-bonding between the inhibitor side chain amide (N-c) and the peptide backbone carbonyls of Ile9 (3.60 ⫾ 0.84 Å) and Lys6 (3.01 ⫾ 0.79 Å). 6. H-bonding between the inhibitor side chain carbonyl (b) and the backbone amides of Lys6 (4.14 ⫾ 0.94 Å) and Met11 (4.39 ⫾ 0.72 Å). It is not surprising that in any MD simulation without explicit water molecules favorable intermolecular H-bond and ionic interactions between the inhibitor and the peptide should occur. The close association between inhibitor and peptide persists, however, when the simulation is carried out with water ‘‘soaking.’’ Moreover, there are several aspects of the modeled complex that suggest it may be representative of the actual relation between the two molecules. The first is the remarkable rapidity with which the peptide and its ligand associate and the stability of that association when formed. More relevant is the correspondence between the favorable interactions listed above 202 A. GRAF VON STOSCH ET AL. Fig. 5. Overlaid backbone traces (106, taken at 20 frame intervals) from the MD simulation in the presence of simulated solvent of the LAV 15mer/inhbitor interaction. Alignment was to the C4-C10-C3 atoms of the heterocyclic ring in the inhibitor, not to atoms within the peptide. a: View from above the plane of the single ␤-hairpin, showing central placement of the inhibitor molecule. b: Side view, showing perpendicular orientation of the inhibitor. c: View from the open end of the hairpin. The N-terminal LPCR reverse turn can be seen in this orientation. MOLECULAR DYNAMICS OF AN INHIBITOR/PEPTIDE COMPLEX 203 Fig. 6. Stabilizing interactions between inhibitor and peptide. Frame is taken from the simulation of the inhibitor/peptide complex with solvent. Note the displacement of the Trp12 residue from the central position occupied in the absence of inhibitor, the association of the heterocyclic ring system of the inhibitor with the tryptophan imidazole, and the hydrogen bond network connecting the pseudopeptide tail of the inhibitor to the two antiparallel strands of the LAV 15mer backbone. and chemical groups found to be necessary to the activity of switch inhibitors in general. For example, all successful switch inhibitors tested so far have had a minimum of three carbonyl groups. One of these interacts in the MD simulation with a residue of the switch sequence that is totally (Trp12) conserved in HIV1 strains, while others are involved in H-bonding with the peptide backbone. Substitution of the inhibitor side chain carbonyl with a sulfur atom abolishes activity, and this carbonyl is one of the first groups to interact with the peptide. Finally, all successful inhibitors have a double ring system, often aromatic, that has been postulated to interact with the critical Trp residue. The association between the Trp indole and the inhibitor heterocyclic ring is a prominent feature of the modeled complex. The MD simulation has helped explain some aspects of the mechanism by which the switch inhibitor disrupts cooperative refolding of the LAV 15mer. A major effect is that the bulk of the inhibitor heterocyclic ring occupies the central space in what is now a single ␤-hairpin, displacing the Trp indole ring that occupies this area in the simulated free peptide. At the same time, the Trp residue is no longer in a position to maintain the multiple interactions with side chain groups that experimental data have implicated in stabilizing the ␤-form until a critical polarity is reached. This could explain why the response of the peptide to decreasing polarity in the presence of the inhibitor is a linearly proportional loss of ␤-sheet and gain of helix. It should be noted that in the inhibitor-complexed model the reverse turn at the N-terminus is centered on the LPCR tetrad, as expected from the 1H-NMR data,5,25 rather than on Ile5 and Lys6 as in the MD model of the free peptide. The and angles of these last now map to the extended region of the Ramachandran plot (Fig. 7). If the tendency to adapt an extended conformation for LPCR in the model of the free peptide is correct, an additional effect of the inhibitor may be to permit formation of the nucleation site even in the presence of the Phe and Trp side 204 A. GRAF VON STOSCH ET AL. inhibitor activity. The molecular dynamics simulation has given us some idea as to why these groups are necessary. Additionally, it has provided us with an hypothesis as to the mechanism by which the switch inhibitors disrupt the cooperativity of the ␤-sheet = helix transition; this in turn can be used to optimize inhibitor design. The most recently synthesized substances show an improvement in ID50 (cell-binding test2) of greater than two orders of magnitude. While rational drug design should ideally proceed from actual physical data, here molecular dynamics simulation has been able to function as a practical tool in a case where these could not be obtained. ACKNOWLEDGMENTS We thank Dr. Rüdiger Pipkorn for preparation of the LAV 15mer peptide and Angelika Lampe-Gegenheimer for her expert assistance with the manuscript. REFERENCES Fig. 7. Ramachandran plot of the and angles of the LAV 15mer peptides in the presence of inhibitor during the MD simulation in vacuo. Values were calculated as in Fig. 1; key as in Fig. 1. chains under high polarity conditions, thus lowering the threshold for formation of helix at the same time as it destabilizes the ␤-sheet. In two chemically unrelated families of switch inhibitor, the modified dipeptide CPF and the compound used here with its pseudo-peptide side chain, the presence of any aliphatic group larger than a methyl at a position corresponding to the N-terminus abolished activity.2 It was proposed that this might be due to the ability of methyl (but not larger) groups to establish nonbonded resonance structures, i.e., that a degree of rigidity in this area allowed inhibitor activity. Certainly the heterocyclic ring devoid of side chain has no significant activity as a switch inhibitor (unpublished results). The simulation shows that the extended chain of the inhibitor rotates as a rigid unit around the Na-C6 linkage to the ring system, adopting two well-defined rotational positions (Fig. 5b,c). In each of these two positions it can serve as an ersatz peptide backbone, forming hydrogen bond linkages to the main chain carbonyl of Ile9 and Met11 amide on one side of the loop and the Lys6 carbonyl and amide on the other. These are established early in the run and serve to fix the heterocyclic ring in a central position, where its displacement of and interaction with the Trp residue is favored. We wished to know more about the interaction between the LAV 15mer and switch inhibitors in order to gain specific information on possible improvements in inhibitor design. It had already been discovered empirically that the presence of certain functional groups was necessary for 1. Reed J, Kinzel V. A conformational switch is associated with receptor affinity in peptides derived from the CD4-binding domain of gp120 from HIVI. Biochemistry 1991;30:4521–4528. 2. Reed J, Kinzel V. Inhibitors of the conformational switch involved in CD4 binding by the env glycoprotein gp120 from human immunodeficiency virus type 1 (HIV1). Biochemistry 1994;33: 10993–10998. 3. Reed J, Kinzel V. Primary structure elements responsible for the conformational switch in the envelope glycoprotein gp120 from human immunodeficiency virus type 1: LPCR is a motif governing folding. Proc Natl Acad Sci USA 1993;90:6761–6765. 4. Graf v Stosch A, Kinzel V, Pipkorn R, Reed J. Investigation of the structural components governing the polarity-dependent refolding of a CD4 binding peptide of gp 120. J Mol Biol 1995;250:507– 513. 5. Graf v Stosch A, Jimenez MA, Kinzel V, Reed J. Solvent polaritydependent structural refolding: a CD and NMR study of a 15-residue peptide. Proteins 1995;23:196–203. 6. Graf v Stosch A, Kinzel V, Reed J. Extension of the polaritydependent ‘‘switch phenomenon’’ of the gp120 binding domain as a target for antiviral chemotherapy Biochemistry 1996; 35:411–417. 7. Hartman R, Schwaner RC, Hermans J. Beta poly(L-lysine): a model system for biological self-assembly. J Mol Biol 1974;90:415– 429. 8. Kaiser ET, Kezdy FJ. Amphiphilic secondary structure: design of peptide hormones. Science 1984;223:249–255. 9. Osterman MG, Kaiser ET. Design and characterization of peptides with amphiphilic ␤-strand structures. J Cell Biochem 1985; 29:57–72. 10. Degrado WF. Design of peptides and proteins. Adv Protein Chem 1988;39:51–124. 11. Muga A, Surewicz WK, Wong PTT, Mantsch HH. Structural studies with the uveopathogenic peptide M derived from retinal S-antigen. Biochemistry 1990;29:2925–2930. 12. Barrow CJ, Colin J, Zagorski MG, Michael G. Solution structures of beta peptide in its constituent fragments: relation to amyloid deposition. Science 1991;251:179–183. 13. Hilbich C, Kisters-Woike B, Reed J, Masters CL, Beyreuther K. Aggregation and secondary structure of synthetic amyloid ␤A4 peptides of Alzheimer’s disease. J Mol Biol 1991;218:149–163. 14. Dyson HJ, Rance M, Houghton RA, Lerner RA, Wreight PE. Folding of immunogenic peptide fragments of proteins in water solution. I. Sequence requirements of the formation of a reverse turn. J Mol Biol 1988;201:161–200. 15. Fan P, Bracken C, Baum J. Structural characterization of monellin in the alcohol-denatured state by NMR: evidence for ␤-sheet to alpha-helix conversion. Biochemistry 1993;32:1573–1582. 16. Yang JJ, Pitkeathly M, Radford SE. Far-UV circular dichroism reveals a conformational switch in a peptide fragment from the ␤-sheet of hen lysozyme. Biochemistry 1994;33:7345–7553. MOLECULAR DYNAMICS OF AN INHIBITOR/PEPTIDE COMPLEX 17. Ilyina F, Mayo KH. Multiple native-like conformations trapped via self association-induced hydrophobic collapse of the 33-residue ␤-sheet domain from platelet factor 4. Biochem J 1995;306:407– 419. 18. Mayo KH, Ilyina F, Park H. A recipe for designing water soluble, ␤-sheet-forming peptides. Protein Sci 1996;5:1301–1315. 19. Blanco FJ, Jimenez MA, Herranz J, Rico M, Santoro J, Nieto JL. A short linear peptide that folds into a native stable ␤-hairpin in aqueous solution. Nature Struct Biol 1994;1:584–590. 20. Ramirez-Alvarado M, Blanco FJ, Serrano L. De novo design and structural analysis of a model ␤-hairpin peptide system. Nature Struct Biol 1996;3:604–612. 21. Sieber V, Moe GR. Interactions contributing to the formation of a ␤-hairpin-like structure in a small peptide. Biochemistry 1996;35: 181–188. 205 22. Van Gunsteren WF, Berendsen HJC. Molekulardynamik-Computersimulation: Methodik, Anwendungen und Perspektiven in der Chemie. Angew Chem 1990;102:1020–1055. 23. Maple JR, Dinor U, Hagler AT. Derivation of force fields for molecular mechanics and dynamics from ab initio energy surfaces. Proc Natl Acad Sci USA 1988;85:5350–354. 24. Kwong PD, Wyatt R, Robinson J, Sweet RW, Sodroski J, Hendrickson WA. Structure of an HIV gp120 envelope glycoprotein in complex with the CD4 receptor and a neutralizing human antibody. Nature 1998;393:648–659. 25. Lindemann A, Kinzel V, Rösch P, Reed J. Structure of the LAV6 peptide: a nucleation site for the correct receptor-induced refolding of the CD4-binding domain of HIV1 gp 120. Proteins 1997;29: 203–211. 26. Musacchio A, Wilmanns M, Saraste M. Structure and function of the SH3 domain. Prog Biophys Mol Biol 1994;61:283–297.