PROTEINS: Structure, Function, and Genetics 35:89–100 (1999) Molecular Dynamics Simulations of the Ras:Raf and Rap:Raf Complexes Jun Zeng,1,2 Herbert R. Treutlein,2 and Thomas Simonson1* 1Laboratoire de Biologie Structurale (CNRS), IGBMC Illkirch (C.U. de Strasbourg), France 2Ludwig Institute for Cancer Research, P.O. Royal Melbourne Hospital, Melbourne, Victoria, Australia ABSTRACT The protein Raf is an immediate downstream target of Ras in the MAP kinase signalling pathway. The complex of Ras with the Rasbinding domain (RBD) of Raf has been modelled by homology to the (E30D,K31E)-Rap1A:RBD complex, and both have been subjected to multiple molecular dynamics simulations in solution. While both complexes are stable, several rearrangements occur in the Ras:RBD simulations: the RBD loop 100–109 moves closer to Ras, Arg73 in the RBD moves towards Ras to form a salt bridge with Ras-Asp33, and Loop 4 of the Ras switch II region shifts upwards toward the RBD. The Ras:RBD interactions (including the RBD-Arg73 interaction) are consistent with available NMR and mutagenesis data on the Ras: RBD complex in solution. The Ras switch II region does not interact directly with the RBD, although indirect interactions exist through the effector domain and bridging water molecules. No large-scale RBD motion is seen in the Ras:RBD complex, compared to the Rap:RBD complex, to suggest an allosteric activation of Raf by Ras. This may be because the Raf kinase domain (whose structure is unknown) is not included in the model. Proteins 1999;35:89–100. r 1999 Wiley-Liss, Inc. Key words: signal transduction; homology modelling; protein structure INTRODUCTION The protein Ras functions as a molecular switch, relaying signalling events from the cell surface to the nucleus, regulating cell growth and differentiation.1,2 Ras cycles between an active form in which guanosine triphosphate (GTP) is bound and an inactive form where guanosine diphosphate (GDP) is bound. Only the active form binds to downstream targets of Ras, such as c-raf-1,3 A-raf,3 B-raf,3 PI-3 kinase,4,5 and RalGEF/RalGDS.6,7 Oncogenic mutants of Ras exist predominantly in the active form, resulting in constitutive Ras signalling and abnormal cell growth; such mutants are found in more than 30% of human cancers.1 The protein Raf is an immediate downstream target of Ras in the MAP kinase pathway. Activation of Raf by Ras occurs partly or entirely through recruitment of Raf to the cellular membrane by farnesylated Ras. Evidence also exists for an additional allosteric mechanism, whereby Ras binding induces conformational changes in Raf that promote activation.8 Such changes would presumably be r 1999 WILEY-LISS, INC. Ras-specific, i.e., homologous proteins such as Rap that do not activate Raf significantly would not be able to induce them. Over the last decade, extensive mutagenesis studies have shown that critical binding and regulatory interactions between Ras and Raf occur through a 76-residue ‘‘Ras-binding domain’’ (RBD: residues 55–131) of Raf and two specific regions of Ras.9–11 The latter regions, termed the effector domain (residues 32–40) and the Switch II region (residues 56–76), have different conformations in the crystal structures of GTP- and GDP-bound Ras, and presumably act as the molecular switches in the signal transduction.9 A detailed analysis of Ras-RBD interactions will, therefore, be of value to design antagonists of Raf, with potential anti-cancer applications, and to understand other Ras-dependent signalling pathways, such as those involving PI-3 kinase and RalGDS. To this end, we report molecular dynamics simulations of the Ras:RBD complex in solution, which complement existing structural studies. The solution structure of the RBD of human Raf-1 has been determined by 3-D heteronuclear NMR spectroscopy,12 and the structure of Ras has been determined both by crystallography13–16 and NMR.17 The exact structure of the Ras:RBD complex is not known. However, the structure of a complex between RBD and Rap1A, a protein extremely homologous to Ras, has been solved by X-ray crystallography.18 More recently, the structure of the complex between the RBD and the more ‘‘Ras-like’’ double mutant (E30D,K31E)-Rap1A was solved19: the latter protein is known as Raps. The sequence identity between Raps and Ras is 50% (Fig. 1), and the sequence identity for residues in the interface region is 95%. The binding constant of Raps:Raf is within a factor of two of that of Ras:Raf. Therefore, the Ras:RBD structure is expected to be very similar to those of Raps:RBD and Rap:RBD. Indeed, many of the direct Rap-RBD interactions seen in the crystal structures could also be inferred from NMR experiments on the Ras:RBD complex in solution.12 We report molecular dynamics simulations of both the Ras:RBD and Raps:RBD complexes in solution. The former Grant sponsor: Australian National Health and Medical Research Council; Grant number: C.J. Martin Fellowship 967362; Grant sponsor: IDRIS Supercomputer Center of the Centre National de la Recherche Scientifique. *Correspondence to: T. Simonson, IGBMC, BP163, 67404 Illkirch, France. E-mail: email@example.com Received 5 October 1998; Accepted 8 December 1998 90 J. ZENG ET AL. Fig. 1. Sequence alignment between Ras and Raps. The sequences were aligned using the CLUSTALX program.48 Fully conserved residues are highlighted in grey. The effector domain and the switch II region are indicated. Helices and ␤-strands are indicated below the sequences. The sequences RASH and RAPA correspond to the proteins simulated here. complex is initially model-built based on the latter. Both complexes are shown in Figure 2. The simulations then investigate the stability of the model and the differences in structure and fluctuations between the two complexes, providing insights into the role of various groups in Ras-Raf recognition. In the Rap:RBD crystal structures, the Rap effector domain (residues 32–40) forms a tight network of electrostatic interactions with a 10-residue ␤-sheet (residues 59–68) and a 12-residue helix (residues 78–89) in the RBD. In the Raps structure, Glu31 and Asp33 interact directly with 84Lys of the RBD.† However, the role of the Ras Switch II region in the Ras-Raf interaction, if any, is not evident from the Rap:RBD crystal structures. This region has low sequence identity between Ras and Rap, particularly in the loop L4 (residues 60–65; Fig. 1). This loop is next to the GTP ␥-phosphate, and its residue Q61 has been proposed as a general base for catalyzing GTP hydrolysis, ‘‘switching’’ Ras from its active state to its inactive state.20–22 Mutants at Ala59, Asn61, and Asp63 of the loop L4 are found in many types of oncogenic Ras proteins.1 Moreover, although the switch II region does not interact directly with the RBD in the Rap:RBD crystal structures, Ras mutants D57A and A59T have significantly reduced Raf binding affinity, and are incompetent for signalling in vivo.23 The possible existence of Ras switch II interactions with the RBD is thus important for understanding the recognition of Raf by Ras, as well as interactions of Ras with other downstream targets. The present simulations of the Ras:RBD and Rap:RBD complexes can be compared to the experimental structures of Ras and the RBD alone, as well as to earlier simulations of the RBD24 and Ras21,25 alone. Binding to Ras and/or Rap may induce conformational changes in Raf, possibly leading to an allosteric activation of Raf.8 Such a mechanism would presumably involve changes in the kinase domain of Raf, whose structure is not known, and which is therefore not included in the simulations. This limits our ability to investigate such a mechanism. Nevertheless, the simula- †In what follows, Ras and Raps residues are designated with their name first and number last; RBD residues are designated with the opposite order. SIMULATIONS OF RAS:RAF AND RAP:RAF COMPLEXES 91 Fig. 2. Stereo view of the C␣ trace of the Raps:RBD and Ras:RBD complexes. The RBD in both complexes is colored grey. The Raps:RBD complex is shown in thick lines and the Ras:RBD complex in thin lines. The effector domain, the loop L4 (Switch II region), and the helix H2 (Switch II region) in Ras/Raps are colored red, green, and blue, respectively; the rest of Raps/Ras is black. tions were analyzed to detect possible conformational changes induced in the RBD by Ras or Rap binding. Recent experiments have demonstrated that a second Ras-binding domain, the Cysteine Rich Region (CRR; residues 139–184) of Raf could also bind to Ras, possibly to the switch II region.26 The present model of the Ras:RBD complex has also been used in a docking study of the ternary Ras:RBD:CRR complex, as well as in free energy simulations of the binding to Ras of a Raf point mutant, presented elsewhere.24,28 at a position near residue 62 of Ras/Raps, containing the entire RBD and half of Ras/Raps. The complexes were solvated by 2,243 water molecules. The heavy atoms of Ras/Raps in the outermost 3 Å (‘‘buffer’’ region) were restrained by weak harmonic forces to their starting positions. The partial charges of charged side chains in the buffer region were reduced by scaling factors, to mimic solvent screening, following Simonson et al.31 The scaling factors are related to the electrostatic potential produced by each charged side chain at the center of the model, calculated from continuum electrostatics. Specifically, the scaling factor is the ratio between the potentials calculated with the spherical MD region surrounded either by vacuum or by high-dielectric solvent. With this scheme, the potential near the center of the model accurately reproduces the potential that would be observed in bulk solvent, even though the actual MD model is surrounded by vacuum. The residues scaled were Lys88, Asp92, Lys117, Lys 147, Arg102, Lys104, Arg149, and Glu153 in the case of Ras, and Arg2, Lys149, Lys151, Lys117, Arg102, and Lys104 in Raps. A few residues outside the 29 Å boundary were included in the model to preserve continuity of the polypeptide chain. Charges of ionized residues in this outermost region were scaled by 80, the dielectric constant of bulk water. These are Asp47, Glu49, and Arg97 in Ras, and Asp47, Asp92, and Glu45 in Raps. In one Ras:RBD simulation (S2), the charges of residues 104–153 of Ras were scaled to zero, providing a test of the simulations’ robustness with respect to the exact scaling protocol. Water molecules were described by a modified TIP3P water model,32,33 with internal geometry constrained by the SHAKE algorithm.34 The CHARMM22 force field was used for all other atoms.29 No electrostatic cutoff was applied. Long-range interactions were treated efficiently by using a multipole approximation for interactions be- MATERIALS AND METHODS Starting Structures The starting structure of the Raps:RBD complex was taken from the (E30D,K31E)-Rap1A:RBD X-ray structure in the PDB (entry 1gua). The initial structure of the Ras:RBD complex was built from the Raps:RBD complex by first superimposing the C␣ atoms of Ras onto those of Raps via least-squares fit. Since the effector domain has the same sequence in Ras and Rap (Fig. 1), its coordinates were then taken explicitly from the Raps:RBD crystal structure. GTP and magnesium were also positioned as in the Raps:RBD complex. The resulting structure was then optimized by 400 steps of conjugate gradient minimization using the CHARMM22 force field,29 during which the backbone atoms shifted by just 0.3 Å. Finally, 70 ps of molecular dynamics simulation were carried out with the protein atoms harmonically restrained, to further relax bad contacts. Molecular Dynamics Setup Molecular dynamics (MD) simulations were carried out at a constant temperature of 25°C with stochastic boundary conditions30 and a 1.5 femtosecond time-step. The model included a spherical region of 29 Å radius, centered 92 J. ZENG ET AL. tween groups of atoms more than 12 A apart.35 Atoms in the buffer region experienced friction and random forces mimicking the coupling to a thermal bath. Simulations were performed using the CHARMM36 program. B-Factor Calculation To compare the crystallographic B-factors with the MD results, the contribution of overall rotation and translation of the protein, absent in the simulations, must be added to the MD data. We proceed as follows. The rigid body motion of the protein can be represented by two (unknown) symmetric tensors T and L37; the mean square amplitudes of translational motion in the direction of a unit vector l and of libration about l are given by 兺兺T ll ij j j (1) 兺 兺 L ll. (2) i j and ij i j i j respectively, where the li, lj are the cartesian components of l. The contribution of overall protein rotation and translation to the B-factor of a protein atom is38 BRT ⫽ 1⁄3 Trace (GL ⫹ T) ⫽ 1⁄3 [(r 22 ⫹ r 23 )L11 ⫹ (r 12 ⫹ r 32 )L22 ⫹ (r 12 ⫹ r 22 )L33 ] ⫺ 2r2r3 L23 ⫺ 2r1r3 L13 ⫺ 2r1r2 L12 ⫹ t 2. (3) where G is an array involving the coordinates of the atom (r1 · r2 · r3 ), and t 2 represents the isotropic mean-square amplitude of translational vibration, t 2 ⫽ 1⁄3 (T11 ⫹ T22 ⫹ T33 ). Fig. 3. a: Rms deviation (Å) of the Raps Switch II region from its starting structure (solid line) and final structure (dashed line) during the 512 ps Raps:RBD simulation. b: Occupancy vs. time of conformations sampled by the Raps switch II region (defined by clustering based on C␣ rms deviations; see text). (4) To determine the unknown tensors T and L, the difference ⌬B between the atomic B-factors from the X-ray data and the MD simulation are first computed. These differences should provide reasonable estimates of the BRT; therefore, the tensor elements are adjusted so that the BRT in Eq. 3 optimally reproduce the ⌬B. Finally, the resulting BRT are added to the MD B-factors. The same procedure was used in a simulation study of Ribonuclease S.39 The above procedure is circular, in the sense that the X-ray and MD data are used jointly to determine the BRT, then the BRT are added to the MD B-factors, and the results are compared back to the Xray B-factors. This is unavoidable, since experimental values of the BRT are not available for this system. To obtain such experimental values, the X-ray refinement of the structure would have to be performed using a sophisticated model of the disorder in which overall rotation-translation is treated explicitly, as has been done for a few proteins.37,40,41 An alternative procedure, used in many early MD studies, consists in neglecting the BRT component of the experimental Bfactors altogether. However, crystallographic studies have shown37,40,41 (and the present study confirms; see below) that the BRT are far from negligible, and so the present fitting procedure is preferable. The procedure is valid, because the determination of T and L involves only twelve adjustable coefficients (the tensors are symmetric), which are fit to several hundred atomic B-factors (here, backbone atoms of buried residues in the complexes were used in the fit). The B-factors are then calculated and compared for approximately 1,000 atoms, including side chains. RESULTS Simulation of the Raps:RBD Complex The average Raps:RBD simulation structure is very similar to the crystal complex. The rms deviation from the X-ray structure increases during the first half of the 510-ps simulation (Fig. 3a), then stabilizes at 0.9 Å (nonhydrogen atoms; 0.7 Å for backbone atoms). The potential energy of 93 SIMULATIONS OF RAS:RAF AND RAP:RAF COMPLEXES TABLE I. Atomic Fluctuations and Deviations From Starting Structures (Å) Rmsd. backbonec Rmsd. side chains Rmsd. heavy atoms Rmsf. heavy atomsd B-factor. heavy atoms Rmsd. backbone Rmsd. side chains Rmsd. heavy atoms Rmsf. heavy atoms B factor. heavy atoms Ras MDa RBD Both 0.76 1.20 1.00 0.62 1.03 1.84 1.49 1.07 0.88 1.49 1.22 0.87 10.1 0.53 0.94 0.76 0.65 12.7 30.2 0.67 1.23 0.99 1.14 34.1 20.1 0.60 1.07 0.86 0.94 23.1 Ras MD ⫹ TLb RBD Both Raps 0.86 1.09 1.03 0.85 19.3 37.5 0.90 21.4 1.23 39.6 27.9 1.07 30.2 19.9 0.85 19.9 Experiment RBD 0.82 1.10e 17.6 32.0e 0.82 1.10e 17.6 32.0e Both 0.84 18.6 0.84 18.6 aAveraged over the Raps:RBD and Ras:RBD simulation trajectories. overall rotation-translation correction (see Materials and Methods). cRms deviation from starting structure. dRms fluctuation around average structure. eFrom NMR spectroscopy on the RBD in solution.12 bIncluding the system converges over this 250-ps time segment (data not shown). The rms deviation from the final structure stabilizes during the last 150 ps. Details are listed in Table I. The main conformations sampled in the simulation were determined by clustering the trajectory based on rms deviations of the C␣s. The switch II region in Raps (residues 56–76) displays the greatest conformational variability. Five main conformations are sampled in this region, separated by small energy barriers, as shown by the frequent transitions between them (Fig. 3b). In the predominant conformation, the ␣ carbon of Raps-Ala66 is shifted upwards 2.2 Å, slightly distorting the Raps ␣2 helix. In the RBD, significant changes (and large fluctuations) occur in two loop regions (Fig. 2). The loop L3 (residues 100–109) moves 2.0 Å down towards the Raps switch II region, probably due to long-range interactions between its charged amino acids (106Lys, 108Lys, and 109Lys) and the switch II region. The loop L2 (residues 73–76) maintains a backbone position similar to the crystal structure, but different sidechain rotamers of residues 73Arg, 74Asn, and 76Met were sampled in the simulation, giving a large rms fluctuation (1.6 Å) for these residues. In the crystal structure, crystal contacts occur in these two loop regions, resulting in smaller observed B-factors. In the Raps:RBD interface region, six salt bridges seen in the crystal complex (84Lys/Asp33, 84Lys/Glu31, 59Arg/Glu37, 67Arg/ Glu37, and 89Arg/Asp38) are preserved during the entire simulation. The simulation samples a solvent structure between the Raps Loop 4 in the switch II region and the GTP ␥-phosphate that differs from the Raps:RBD crystal structure. In the latter, a water molecule next to the ␥-phosphate forms a channel connecting the GTP to bulk solvent. In Ras, a water molecule in this position is thought to attack the ␥-phosphate in GTP hydrolysis. The channel, bounded by the Thr35-C␥ and the Ala59-C␤, is just large enough to allow a water molecule to enter, and inorganic phosphate to leave the GTP pocket.18 This water is also found in the simulation, with a B-factor of 17.5 Å2 (including the protein translation/libration contribution: see below, Methods), close to the crystallographic value of 19.0 Å2. However, the simulation places a second water molecule between the ␥-phosphate and O␥1 of Thr61. The latter water molecule is more mobile than the first, with a B-factor of 28.5 Å2. The presence of the second water molecule could be due to the fully solvated simulation conditions. Another structural difference is the direct formation of a hydrogen bond between the Tyr32-OH and an oxygen atom of the ␥-phosphate, which were separated by a water molecule in the crystal structure. The existence of such a hydrogen bond in unbound Ras (free in solution) is suggested by NMR data.42 The average rms fluctuation of nonhydrogen atoms around the mean structure is calculated to be 0.94 Å (Table I). This gives an average B-factor of 23.1 Å2, comparable to the crystallographic value of 18.6 Å2.19 However, the experimental B-factors include overall rotation and translation of the protein, which is not present in the simulations. By the method outlined above (see Materials Methods), this contribution is estimated to be 8.7 Å2 for Raps, 5.6 Å2 for the RBD, and 7.1 Å2 for all heavy atoms. The amplitudes of overall rotation and translation are thus estimated to be 0.44 Å and 0.30 Å, respectively, close to the values observed in other proteins of similar size (e.g., 0.64 Å and 0.24 Å in ribonuclease A crystals).37,40,41 Adding these to the intramolecular fluctuations from the simulation, the overall B-factors are calculated to be 21.5 Å2 for Raps, 39.6 Å2 for the RBD, and 30.2 Å2 for all heavy atoms. While the Raps value is close to the crystallographic result, the RBD value is much larger. It is in fair agree- 94 J. ZENG ET AL. Fig. 4. Comparison between atomic rms fluctuations (Å) in the Raps:RBD simulation and experiment.19 Simulation results with (dots; MD/TL) and without (x: MD) rotation-translation contribution. Results averaged over entire residues. a: Raps: b: RBD. ment with the flexibility of the uncomplexed RBD in solution, suggested by the rms deviation from the mean of the 30 structures in the experimental NMR ensemble12 (1.1 Å, corresponding to an effective B-factor of 32.0 Å2 ). This is consistent with the geometry of the simulation model, in which the inner part of Raps is involved in the interface with the RBD and the outer part is in the restrained buffer region (where the restraint force constants are actually determined by the experimental B-factors). Most of the RBD surface, on the other hand, is exposed to solvent. Thus, the Raps flexibility is similar to that in the crystal environment, while that of the RBD is solution-like. Similarly, the variations of the calculated Raps B-factors along the polypeptide chain (Fig. 4) are in good agreement with the crystallographic values, except Fig. 5. a: Rms deviation (Å) of the Ras Switch II region from its starting structure (solid line) and final structure (dashed line) during the 635 ps Ras:RBD simulation S1. b: Occupancy vs. time of conformations sampled by the Ras switch II region (defined by clustering based on C␣ rms deviations; see text). for residues Glu62 and Gln63 (whose larger calculated values arise from the sampling of multiple sidechain orientations in the simulation), while the calculated RBD values are larger. Residues Ala57 and Asp131 at the two Raps chain termini have large rms fluctuations of approximately 1.4 Å and 1.6 Å, respectively, typical of protein chain termini in solution.25 The RBD structure in the complex, averaged over the MD simulation, was compared to the average NMR structure of the RBD in solution. Both structures were slightly minimized first, to remove stereochemical distortion associated with the averaging. While the RBD in the complex folds in a similar manner to the RBD alone in solution, some local conformational changes occur. In the complex, the ‘‘Ras-binding helix’’ extends from residue 84 to 89: in solution, two distinct conformations are seen for residues 87–90: one where the helix extends to residue 89 as in the SIMULATIONS OF RAS:RAF AND RAP:RAF COMPLEXES 95 Fig. 6. Snapshots of the interface between Ras and the RBD during the Ras:RBD simulation S1. Orientation as in Figure 2. The backbone as well as charged sidechains in the interface are shown. Residues 59Arg, 67Arg, 73Arg, and 89Arg are dark blue: residues Asp33, Glu37, and Asp38 are red. Snapshots corresponding to the first 100 ps of simulation have their backbone colored black; similarly for the 73Arg side chain. The downwards drift of the RBD backbone around 73Arg can be seen. complex, and one where residues 87–90 form a ␤ turn.12,24 The backbone rms deviation between the MD complex and the NMR structure, averaged over non-loop RBD residues, is 1.2 Å. Deviations in the surface loops are larger, ranging from 2.9 to 3.6 Å. These loops (residues 63–66, 73–77, 90–95, 100–110, and 122–125) are very flexible and only moderately conserved throughout evolution, based on alignment of six RBD sequences (human, pig, rat, C. elegans, D. melanogaster, X. laevis: not shown). most of the analysis of the Ras:RBD complex below. In addition, the RBD loop 72–76 undergoes a 1.3–1.7 Å shift going from conformation 2 to conformations 3–6, with a salt bridge forming between RBD-73Arg and Ras-Asp33. In the Raps:RBD simulation, in contrast, the separation between the RBD-73Arg and Raps-Asp33 always remained greater than 5.0 Å, with a water molecule bridging the 73Arg-NH and the Asp33-OD. Principal component analysis43 of S1 (not shown) indicates that a significant portion of the RBD motion corresponds to an overall rotation towards the switch II region of Ras, correlated with an upwards motion of the switch II ␣2 helix towards the RBD. The motion of the switch II region and the RBD are thus coupled, possibly through long-range electrostatic interactions. The stability of the interface structure during the three simulations is illustrated in Figure 6 by a set of snapshots from the trajectories. Hydrogen bonds between Ras and the RBD are listed with their occupancies in Table II. Three salt bridges, RBD-59Arg/Ras-Glu37, 67Arg/Glu37, and 89Arg/Asp38, and two backbone hydrogen bonds, Glu37-O/69Val-N and Ser39-N/67Arg-O, in the core of the interface remain stable during all three simulations. Mutations of charged amino acids in or around the interface will form or disrupt Ras-Raf salt bridges, and consequently affect the Ras-Raf binding affinity. The relationship between the Ras:RBD structure and experimental mutagenesis results is summarized in Table III. Some variations are seen between the three simulations (Table II). In Ras, Asp31 and Glu33 flank the effector domain and are exposed to bulk solvent. Their charged sidechains form stable salt bridges to RBD-84Lys throughout S1 and S3: these salt bridges are partially broken during S2. While Glu33 directly contacts RBD-73Arg during S1 and S3, it is separated from it by a water Simulations of the Ras:RBD Complex The Ras:RBD structure was built and solvated as described above (see Materials and Methods), and three simulations were carried out, lasting 635, 600, and 600 ps, respectively (referred to below as S1, S2, and S3). These provide a dynamic model of the Ras:RBD complex, whose experimental structure is not known. Structure and dynamics of the Ras:RBD complex Table I reports deviations from the starting model structure (which is similar to the Raps:RBD complex; see Materials and Methods). In simulation S1, the average deviation is 0.88 Å for backbone atoms and 1.22 Å for all heavy atoms, larger than the values from the Raps:RBD simulation, as expected. Figure 5a shows the rms deviation as a function of time. From the clustering protocol used above, six main conformations were sampled during S1 (Fig. 5b). The six conformations are mainly distinguished by the C␣ positions in the Ras switch II region. The first 225 ps sampled two well-defined conformations, each for approximately 100 ps. These are unstable, and the structure then shifts about 1.3 Å (C␣ deviations) into a set of four conformations that differ by only 0.5–0.8 Å from each other and are separated by low energy barriers. The stable 225–635 ps trajectory segment of S1 was used for 96 J. ZENG ET AL. TABLE II. Occupancy of Selected Hydrogen Bonds During Simulations Hydrogen bond partners RBD Ras 84Lys—NZ 84Lys—NZ 73Arg—NH1 73Arg—NH2 59Arg—NH2 67Arg—NH1 67Arg—N 68Thr—OG 89Arg—NH1 89Arg—NH2 89Arg—NH2 89Arg—NH1 WATa 69Val—N 67Arg—O GTP167—O2G Glu31—OE Glu33—OD Glu33—OD Glu33—OD Glu37—OE Glu37—OE Ser39—OG Asp38—OD Ser39—O Ser39—O Asp38—OD WATa Asp38—OD Glu37—O Ser39—N Tyr32—OH Hydrogen bond partners Ras Ras Gly60—O Arg68—NH1 Arg68—NH2 Ser17—OG aBridging Tyr96-OH Tyr71—OH Tyr71—OH Asp57—OD1 Simulation run S1 S2 S3 1.0 0.91 0.61 0.75 1.00 1.00 1.00 1.00 1.00 0.98 1.00 1.00 1.00 0.97 0.94 0.97 0.29 0.73 0.00 0.01 1.00 1.00 1.00 0.92 1.00 1.00 1.00 1.00 1.00 0.99 0.98 0.99 0.98 1.00 0.58 0.87 1.00 1.00 1.00 0.91 1.00 1.00 1.00 1.00 1.00 0.99 0.97 0.51 Simulation run S1 S2 S3 1.00 0.23 0.30 1.00 0.74 0.00 0.98 1.00 0.25 0.00 0.58 1.00 water molecule. molecule during S2, similar to the Raps:RBD simulation and the crystal structure. Tyr32-OH directly bonds to the GTP ␥-phosphate during S1 and S2; this hydrogen bond was only stable for half of S3. The conformation of the Ras Loop 4 varies significantly between simulations. While the hydrogen bonds Gly60-O/Tyr96-OH and Asn61-NE/Gly13-O are present in S1, they are less stable in S2 and completely broken in S3. Stable hydrogen bonds between the guanidinium group of Arg68 and Tyr71-OH are only observed in S2. The best-defined hydrogen bond in the Ras switch II region is Ser17-OG/Asp57-OD. This hydrogen bond strongly links the Switch II region and the helix ␣1 in all three simulations. The Asp57Ala mutation, which decreases Ras-Raf binding, may break this linkage, and consequently cause a conformational change in Ras (see Table III). The solvation around the Ras effector domain and switch II region is stable and similar in the three simulations (Fig. 7a). A chain of water molecules connects Asp57 of the Ras switch II region, Asp37 and Asp38 of the effector domain, and Arg89 of the RBD. In the GTP pocket, 2–3 water molecules are observed between loop L4 and the GTP ␥-phosphate (Fig. 7b). While two water molecules consistently bind to the GTP ␥-phosphate as in the Raps: RBD simulation, a third water molecule only enters the GTP pocket after 420 ps of simulation S2. The atomic rms fluctuations of the Ras:RBD interface are in good agreement with both the observed and simulated B-factors of the Raps:RBD complex (Figs. 4 and 8). The RBD does not interact directly with the switch II region in either the Ras:RBD or the Raps:RBD complexes. The effect of switch II mutations on Ras-RBD binding TABLE III. Relative Binding Affinity of Ras, Raf Mutants† Mutation Affinity Ras E31K D33xb E37K D38N S39P ⫺ ⫺ ⫺ ⫺ ⫺ Y40K D57A ⫺ ⫺ A59T Raf ⫺ Binding site Raf 84K 84K 59R,67R 89R 89R 64N 89R Comment Breaks salt bridge Breaks salt bridge Breaks salt bridge Breaks salt bridge Lost 89R-S39 hydrogen bonds Lost hydrogen bond Long-range effect/conformation change Conformation change References 23 23 23 23 23 23 23 23 Ras R59A N64M R67A T68A V70R K84A K84E K84L K84D A85R A85D V88R V88K R89D ⫺ ⫺ ⫺ ⫺ ⫹ ⫺ ⫺ ⫺ ⫺ ⫹ ⫺ ⫹ ⫹ ⫺ R89L ⫺ R89K ⫺ E37 R41 E37 D38 E31/D33 E31/D33 E31/D33 E31/D33 E31/D33 E31/E33 E38 E31/D33 E31/D33 D38/S39 Breaks salt bridge Lost hydrogen bond Breaks salt bridge Lost hydrogen bond Forms salt bridge Breaks salt bridge Breaks salt bridge Breaks salt bridge Breaks salt bridge Forms salt bridge Breaks salt bridge Forms salt bridge Forms salt bridge Breaks salt bridge and hydrogen bonds D38/S39 Breaks salt bridge and hydrogen bonds D38/S39 Increased desolvation cost/lost hydrogen bonds 46 a 11,46 a a,c 46 a a,c 11 11 11 a 11 a,c a,c a,c a,c 11 a 46 a 24,28 a †⫹ and ⫺ indicate increased or decreased binding affinity, respectively. affinity measured for the Ras:RBD complex. bMutation to any uncharged residue. cM. Fridman and H. Maruta, unpublished data. aBinding would thus be expected to be minor. Nevertheless, two mutations at the switch II N-terminus of Ras, i.e., Asp57Ala and Ala59Tyr, have been shown to decrease the binding affinity significantly (Table III). These residues are buried beneath the effector domain of Ras, and interact with the RBD indirectly through a water channel shown in Figure 7a. The channel was also found in the Rap:RBD crystal structures (see above), with a water molecule forming a bridge between RBD-89Arg89 and Ras-Asp38,19 having a B-factor of 17.9 Å2. The average B-factor of the channel water molecules is calculated to be 20.1 Å2 from the simulations (including protein translation/libration, as appropriate for comparison to the X-ray value). The RBD structure in the complex, averaged over the MD simulation, was compared to the average NMR structure of the RBD in solution. As in the Rap:RBD complex above, the ‘‘Ras-binding helix’’ extends from residue 84 to 89 in the Ras:RBD MD complex, whereas this region occupies a mixture of two conformations in solution. The backbone rms deviation between the MD complex and the NMR structure, averaged over non-loop RBD residues, is 1.5 Å. Deviations in the surface loops are larger, ranging from 3.2 to 4.4 Å. SIMULATIONS OF RAS:RAF AND RAP:RAF COMPLEXES 97 Fig. 7. Stereo view of the solvation structure around the effector domain of Ras, the Ras switch II region, and the GTP pocket, from the Ras:RBD simulation S1. a: Solvation structure around the N-terminus of the Ras switch II region. The lower two ␤-strands belong to Ras, the upper two to the RBD. b: Solvation structure in the GTP pocket. The magnesium ion is shown as a sphere. Comparison to the Raps:RBD complex bonds directly to Ras-Asp33, with two N-O␦ distances of 2.9 Å and 2.6 Å. The newly formed 73Arg/Asp33 salt bridge appears to drive the backbone rearrangement of the RBD-72-77 loop. In Ras, the Loop 4 of Switch II (residues 62–65) shifts away from its starting position; the ␣2 helix moves by approximately 2.0 Å and shortens by two residues, from 64–75 to 66–75. The solvation structure in and near the GTP pocket (shown in Fig. 7b) is similar to the Raps:RBD simulation, with a water channel bounded by Thr35 and Ala59, and two water molecules bound to the GTP ␥-phosphate. The B-factors of these two waters (including protein translation/ The overall orientation of the RBD is slightly different from that in the Raps:RBD complex, as shown in Figure 2. However, all the Rap-RBD salt bridges are preserved in Ras:RBD. Two loop regions of the RBD (residues 100–109 and residues 72–77) have different conformations from the Raps:RBD simulation (and the crystal structure), as well as large fluctuations (Fig. 8). Both regions move toward Ras during the simulations; e.g., the C␣ atom of Asn74 moves 4.0 Å during S1. In the starting structure, RBD73Arg was more than 5.0 Å away from Ras-Asp33; after 225 ps of S1, the RBD-73Arg guanidinium group hydrogen 98 J. ZENG ET AL. Fig. 8. Comparison between atomic rms fluctuations (Å) in the Ras:RBD simulation S1 and experimental data for Raps:RBD.19 Simulation results with (dots) and without (x) rotation-translation contribution. Results averaged over entire residues. a: Ras; b: RBD. libration) are calculated to be 10.4 Å2 and 22.1 Å2, respectively, slightly smaller than the Raps:RBD X-ray and simulation results. Energy component analysis Interaction energies were calculated between different residues of the RBD and charged side chains in the effector domain and the switch II region of Ras. Atoms within 6 Å of an Arg-C, or 4 Å of an Asp-C␥ or Glu-C␦, were taken to define a first interaction shell. (No Lys occurs in the regions considered.) All six charged residues in the effector domain have an RBD salt bridge partner in their first interaction shell. Corresponding interaction energies are between ⫺111 and ⫺83 kcal/mol, dominated by electrostatics. Mutation of any single salt bridge partner will destroy one such interaction, and the unbound state will be overwhelmingly stabilized by solvation of the remaining, unbalanced partner. This is confirmed by experimental mutagenesis data (Table III). Joint mutation of both salt bridge partners, on the other hand, is likely to have a much smaller, possibly negligible effect on binding.44 Solvent is present in the first interaction shells of the effector domain, with coordination numbers of 3–7 and interaction energies of ⫺35.0 to ⫺66.0 kcal/mol. This gives on average ⫺10.0 kcal/mol interaction energy per water molecule, typical of short-ranged charge-water interactions. No RBD amino acids are found in the first interaction shells of the Ras Switch II region. Asp57 is buried inside Ras, interacting with three water molecules (Fig. 7a). Charged residues in the segment 60–73 are entirely exposed to solvent, experiencing interaction energies of ⫺10.0 kcal/mol per water molecule. The interaction energy differences between Ras and Raps interacting with the RBD were also calculated. Several large differences are seen for residues conserved between Ras and Raps. First, although the sequence of the effector domain (residues 30–42) is identical in Ras and Raps, Ras-Asp33 forms a salt bridge to RBD-73Arg not found in Raps. Thus, Ras-Glu31 and Asp33 contribute interaction energies more negative by ⫺28 and ⫺52 kcal/ mol, respectively. The contribution of Ras-Arg68 is increased by 72 kcal/mol in the Ras:RBD complex, as a result of a structural difference: Arg68 hydrogen bonds to RasTyr71 in the Ras:RBD complex (Table II), but forms a salt bridge to Raps-Asp92 in the Raps:RBD complex. Several large differences arise because of sequence differences between Ras and Raps. For example, His27 in Ras has an unfavorable interaction with the RBD residues 89Arg and 87Lys, and contributes ⫹153 kcal/mol to the RBD:Ras interaction energy. Some other differences are seen that are expected to be largely compensated by changes in the unbound state. For example, Ras-Glu63 contributes ⫺236 kcal/mol to the interaction energy, compared to ⫺34 kcal/mol for its homologue Gln63 in Raps. At the same time, Glu63 is expected to interact strongly with solvent in the unbound state, destabilizing the bound form. The extent of this destabilization is expected to be of the same order of magnitude as, but somewhat smaller than, the Glu63/Glu63 interaction energy difference (e.g., the solvation free energy difference between Gln and Glu is on the order of 80 kcal/mol, favoring Glu45 ). Overall, all the significant changes in the interaction energy between Ras-RBD and Raps-RBD are due to charged amino acids, underlining the importance of longrange electrostatics for Ras-RBD recognition. CONCLUDING DISCUSSION The present simulations probe the structure and dynamics of the Raps:RBD complex in solution, and provide a dynamical model of the Ras:RBD complex, whose structure has not been determined experimentally. The model combines a high quality force field with an accurate treatment of electrostatics, where long-range interactions SIMULATIONS OF RAS:RAF AND RAP:RAF COMPLEXES are treated with a multipole approximation and bulk solvent is represented by scaling factors obtained from Poisson-Boltzmann calculations on the fully solvated complexes. Repeated simulations and comparison to both the Rap:RBD crystal structures and NMR data on the Ras: RBD complex suggest that the present Ras:RBD model is stable and representative of the actual complex in solution. Structural differences between the Ras:RBD and Raps: RBD complexes occur mainly in three regions. In the Ras:RBD case, the RBD loop 100–109 moves closer to Ras; 73Arg in the RBD loop 72–77 moves towards Ras to form a salt bridge with Ras-Asp33, and Loop 4 of the Ras switch II region shifts up towards the RBD. Loop 4 has a low sequence identity between Ras and Raps; in particular, the charged Glu63 is present in Ras but not Raps. This additional charge may help drive the conformational change, as well as attracting RBD-73Arg. These structural differences are reflected by the interaction energy differences between Ras and Raps: Glu31 and Asp33 of Ras interact much more strongly with the RBD than in the Raps:RBD case, and Glu63 of the Ras switch II (Gln in Raps) stabilizes the Ras:RBD complex by an additional ⫺142 kcal/mol. This residue may also ‘‘pull’’ the 100–109 loop downwards through a long-range electrostatic interaction, since the latter loop contains several positive charged residues, i.e., 106Lys, 108Lys, and 109Lys. The sequence of the effector domain in the interface with the RBD is identical in Ras and Raps. The conformation of this region is very similar throughout the Raps:RBD and Ras:RBD simulations, and very stable, with an average B-factor and rms atomic fluctuation of 19.7 Å2 and 1.0 Å, respectively (from the Ras:RBD simulation S1). The RasRBD interactions seen in the simulations are consistent with those observed by NMR spectroscopy in solution. In the NMR experiment, the RBD residues with significantly perturbed amide chemical shifts were in the N-terminal region (i.e., residues 56–61, 65–69, 78–91), which directly contacts the effector domain of Ras in the simulations. The detailed correspondence between chemical shift perturbations and Ras-RBD interactions is almost perfect (see ref.28 ). The direct contact between 73Arg of the RBD and Asp33 of Ras predicted from our simulations can also be inferred from a large broadening of the experimental resonance of a 73Arg side chain proton, even though this residue was not originally classified as an interacting one.12 Experimental measurements of the Ras-RBD binding affinity change associated with point mutations of 73Arg are desirable to further validate this interaction. The solution structure and dynamics of the Ras:RBD complex obtained from the molecular dynamics simulations are also consistent with the available mutagenesis data (Table III). The simulations confirm the Ras:RBD interface structure suggested by the Rap:RBD crystal structures. Salt bridges connect the effector domain of Ras and the RBD, and binding is correspondingly sensitive to mutations in this region. The Ras switch II region does not interact directly with the RBD, and most of its side chains are exposed to solvent, so that mutations in this region should not affect Ras-RBD binding strongly. The Asp57Ala 99 mutation at the Ras N-terminus does have a significant effect on Ras-Raf binding, even though Asp57 does not directly interact with the RBD. The simulations suggest several possible mechanisms for this effect. First, Asp57 is connected with the Ras effector domain (Asp38, Tyr40) through bridging water molecules, and its mutation may therefore disrupt the conformation of the effector domain. Second, the energy decomposition described above indicated that 89Arg of the RBD and Asp57 have a large interaction energy (⫺22 kcal/mol); loss of this direct interaction may contribute to the binding free energy change. Third, the hydrogen bond between Asp57 and Ser17 links the Ras switch II region to the Ras ␣1 helix. The Asp57Ala mutation is expected to break this linkage, possibly leading to a conformational change of the effector domain. Some differences in the solvation of the active site were observed, compared to the crystal structures. In addition to the water molecule observed between the Loop L4 of Switch II and the GTP ␥-phosphate in the crystal structures of both Ras and the Rap:RBD complex, the present simulations (of both Raps:RBD and Ras:RBD) place a second water molecule, and possibly a third, inside the GTP pocket. These water molecules are expected to play a role in GTP hydrolysis and, consequently, in Ras-Raf signalling. Comparing the RBD in solution to the RBD in the complexes, significant local conformational differences are seen, mainly in surface loops far from the interface. These loops are very flexible and only moderately conserved throughout evolution. It remains to be seen whether these motions are involved in the signalling mechanism. Comparing the Rap:RBD and Ras:RBD complexes, no large-scale shift of the RBD relative to Ras or Rap is seen (Fig. 2) that would provide evidence for a putative allosteric activation of Raf by Ras. This could be a consequence of the limited size of the model (29 Å radius) and simulation time, and possibly of the absence of the Raf kinase domain. ACKNOWLEDGMENTS J.Z. gratefully acknowledges a C.J. Martin fellowship (967362) awarded by the Australian National Health and Medical Research Council. The CHARMM program was kindly provided by Professor Martin Karplus. REFERENCES 1. Barbacid M. Ras genes. Annu Rev Biochem 1987;56:779–827. 2. Lowy D, Willumsen B. Function and regulation of ras. Annu Rev Biochem 1993;62:851–891. 3. Daum G. The ins and outs of Raf kinases. Trends Biochem Sci 1994;19:474–480. 4. Wittinghofer A, Herrmann C. Ras-effector interactions, the problem of specificity. FEBS Lett 1995;369:52–56. 5. Vanhaesebroeck B, Leevers S, Panayotou G, Waterfield M. Phosphoinositide 3-kinase: a conserved family of signal transducers. Trends Biochem Sci 1997;22:267–272. 6. Geyer M, Herrmann C, Wohlgemuth S, Wittinghofer A, Kalbitzer HR. Structure of the Ras-binding domain of RalGEF and implications for Ras binding and signalling. Nature Struct Biol 1997;4: 694–699. 7. Huang L, Weng X, Hofer F, Martin GS, Kim S. Three-dimensional structure of the Ras-interacting domain of RalGDS. Nature Struct Biol 1997;4:609–615. 8. McCormick F. Ras biology in atomic detail. Nature Struct Biol 1996;3:653–655. 100 J. ZENG ET AL. 9. Milburn MV, Tong L, DeVos AM, Brünger A, Yamaizumi Z, Nishimura S, Kim S-H. Molecular switch for signal transduction: structural differences between active and inactive forms of protooncogenic Ras proteins. Science 1990;247:939–945. 10. Chuang E, Barnard D, Hettich L, Zhang X-F, Avruch J, Marshall MS. Critical binding and regulatory interactions between Ras and Raf occur through a small, stable N-terminal domain of Raf and specific Ras effector residues. Mol Cell Biol 1994;14:5318–5325. 11. Barnard D, Diaz B, Hettich L, Chuang E, Zhang X-F, Avruch K, Marshall M. Identification of the sites of interaction between c-Raf-1 and Ras-GTP. Oncogene 1995;10:1283. 12. Emerson SD, Madison VS, Palermo RE, Waugh DS, Scheffler JE, Tsao K-L, Kiefer ST, Liu SP, Fry DC. Solution structure of the Ras-binding domain of c-Raf-1 and identification of its Ras interaction sites. Biochemistry 1995;30:6911–6918. 13. deVos A, Tong L, Milburn M, Matias P, Jancarik J, Noguchi S, Nishimura S, Miura K, Ohtsuka E, Kim S. Three-dimensional structure of an oncogene protein: catalytic domain of human c-H-ras p21. Science 1988;239:888–893. 14. Brunger A, Milburn M, Tong L, deVos A, Jancarik J, Yamaizumi Z, Nishimura S, Ohtsuka E, Kim S. Crystal structure of an active form of Ras protein, a complex of a GTP analog and the H-Ras p21 catalytic domain. Proc Natl Acad Sci USA 1990;89:4849–4853. 15. Pai E, Krengel U, Petsko G, Goody R, Kabsch W, Wittinghofer A. H-Ras p21 protein complex with guanosine triphosphate: human cellular Harvey-Ras gene (truncated) expressed in Escherichia coli. EMBO J 1990;9:2359. 16. Tong L, deVos A, Milburn M, Kim S. Crystal structures at 2.2 å resolution of the catalytic domains of normal ras protein and an oncogenic mutant complexed with GDP. J Mol Biol 1991;217:503– 516. 17. Kraulis P, Domaille P, Campbell-Burk S, Aken TV, Laue E. Solution structure and dynamics of ras p21-GDP determined by heteronuclear three- and four-dimensional NMR spectroscopy. Biochemistry 1994;33:3515–3531. 18. Nassar N, Horn G, Herrmann C, Scherer A, McCormick F, Wittinghofer A. The 2.2 å crystal structure of the Ras-binding domain of the serine/threonine kinase c-Raf-1 in complex with Rap1A and a GTP analogue. Nature 1995;375:554–560. 19. Nassar N, Horn G, Herrmann C, Block C, Janknecht R, Wittinghofer A. Ras/Rap effector specificity determined by charge reversal. Nature Struct Biol 1996;3:723–729. 20. Langen R, Schwins T, Warshel A. On the mechanism of guanosine triphosphate hydrolysis in Ras p21 proteins. Biochemistry 1992; 31:8691. 21. Frech M, Darden T, Pederson L, Foley C, Charifson P, Anderrson M, Wittinghofer A. Role of glutamine-61 in the hydrolysis of GTP by p21H-ras: an experimental and theoretical study. Biochemistry 1994;33:3237–3244. 22. Muegge I, Schweins T, Langen R, Warshel A. Electrostatic control of GTP and GDP binding in the oncoprotein p21-ras. Structure 1996;4:475–489. 23. Shirouzu M, Koide H, Fuita-Yoshigaki J, Oshio H, Toyama Y, Yamasaki K, Fuhrman SA, Villafranca E, Kaziro Y, Yokovama S. Mutations that abolish the ability of Ha-Ras to associate with Raf-1. Oncogene 1994;9:2153–2157. 24. Zeng J, Treutlein HR, Simonson T. The conformation of the Ras-binding domain of Raf studied by molecular dynamics and free energy simulations. Proteins 1998;31:186–200. 25. Diaz J, Wroblowski B, Engelborghs Y. Molecular dynamics simulation of the solution structures of Ha-ras-p21 GDP and GTP complexes: flexibility, possible hinges, and levers of the conformational transition. Biochemistry 1995;34:12038–12047. 26. Luo Z, Diaz B, Marshall M, Avruch J. An intact Raf zinc finger is required for optimal binding to processed Ras and for Rasdependent Raf activation in situ. Mol Cell Biol 1997;17:46–53. 27. Reference deleted in proofs. 28. Zeng J, Fridman M, Maruta H, Treutlein HR, Simonson T. Protein-protein recognition: an experimental and computational study of the R89K mutation in raf and its effect on ras binding. Protein Sci 1998;8:50–64. 29. Mackerell A, Bashford D, Bellott M, Dunbrack R, Evanseck J, Field M, Fischer S, Gao J, Guo H, Ha S, Joseph D, Kuchnir L, Kuczera K, Lau F, Mattos C, Michnick S, Ngo T, Nguyen D, Prodhom B, Reiher W, Roux B, Smith J, Stote R, Straub J, Watanabe M, Wiorkiewicz-Kuczera J, Yin D, Karplus M. An all-atom empirical potentil for molecular modelling and dynamics study of proteins. J Phys Chem B 1998;102:3586–3616. 30. Berendsen H, Postma J, van Gunsteren W, DiNola A, Haak J. Molecular dynamics with coupling to an external bath. J Chem Phys 1984;81:3684–3690. 31. Simonson T, Archontis G, Karplus M. Continuum treatment of long-range interactions in free energy calculations. Application to protein-ligand binding. J Phys Chem B 1997;101:8349–8362. 32. Jorgensen WL, Chandrasekar J, Madura J, Impey R, Klein M. Comparison of simple potential functions for simulating liquid water. J Chem Phys 1983;79:926–935. 33. Neria E, Fischer S, Karplus M. Simulation of activation free energies in molecular systems. J Chem Phys 1996;105:1902–1921. 34. Ryckaert J, Ciccotti G, Berendsen H. Numerical integration of the cartesian equations of motion for a system with constraints. J Comput Phys 1977;23:327–341. 35. Stote R, States D, Karplus M. On the treatment of electrostatic interactions in biomolecular simulation. J Chim Phys 1991;88: 2419–2433. 36. Brooks B, Bruccoleri R, Olafson B, States D, Swaminathan S, Karplus M. Charmm: a program for macromolecular energy, minimization, and molecular dynamics calculations. J Comp Chem 1983;4:187–217. 37. Sternberg M, Grace D, Phillips D. Dynamic information from protein crystallography. J Mol Biol 1979;130:231–253. 38. Schomaker V, Trueblood K. Acta Cryst B 1968;24:63–76. 39. Simonson T, Brünger AT. Thermodynamics of protein-peptide binding in the ribonuclease S system studied by molecular dynamics and free energy calculations. Biochemistry 1992;31:8661– 8674. 40. Diamond R. On the use of normal modes in thermal parameter refinement: theory and application to the bovine pancreatic trypsin inhibitor. Acta Cryst 1990;A46:425–435. 41. Howlin B, Moss D, Harris G. Segmented anisotropic refinement of bovine ribonuclease A by the application of the rigid-body TLS model. Acta Cryst 1989;A45:851–861. 42. Geyer M, Schweins T, Herrmann C, Prisner T, Wittinghofer A, Kalbitzer H. Conformational transitions in p21ras and its complexes with the effector protein Raf-RBD and the GTPase activating protein GAP. Biochemistry 1996;35:10308–10320. 43. Brooks C, Karplus M, Pettitt M. Proteins: a theoretical perspective of dynamics, structure and thermodynamics. Adv Chem Phys 1987;71:1–259. 44. Hendsch Z, Tidor B. Do salt bridges stabilize proteins? A continuum electrostatics analysis. Prot Sci 1994;3:211–226. 45. Archontis G, Simonson T, Moras D, Karplus M. Specific amino acid recognition by aspartyl-tRNA synthetase studied by free energy simulations. J Mol Biol 1998;275:823–846. 46. Block C, Janknecht R, Herrmann C, Nassar N, Wittinghofer A. Quantitative structure-activity analysis correlating Ras/Raf interaction in vitro to Raf activation in vivo. Nature Struct Biol 1996;3:244–251. 47. Reference deleted in proofs. 48. Thompson J, Higgins D, Gibson T. CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Res 1994;22:4673–4680.