The crystal structures of the complexes between bovine beta-trypsin and ten P1 variants of BPTI

The high-resolution X-ray structures have been determined for ten complexes formed between bovine beta-trypsin and P1 variants (Gly, Asp, Glu, Gln, Thr, Met, Lys, His, Phe, Trp) of bovine pancreatic trypsin inhibitor (BPTI). All the complexes were crystallised from the same conditions. The structures of the P1 variants Asp, Glu, Gln and Thr, are reported here for the first time in complex with any serine proteinase. The resolution of the structures ranged from 1.75 to 2.05 A and the R-factors were about 19-20 %. The association constants of the mutants ranged from 1.5x10(4) to 1.7x10(13) M-1. All the structures could be fitted into well-defined electron density, and all had very similar global conformations. All the P1 mutant side-chains could be accomodated at the primary binding site, but relative to the P1 Lys, there were small local changes within the P1-S1 interaction site. These comprised: (1) changes in the number and dynamics of water molecules inside the pocket; (2) multiple conformations and non-optimal dihedral angles for some of the P1 side-chains, Ser190 and Gln192; and (3) changes in temperature factors of the pocket walls as well as the introduced P1 side-chain. Binding of the cognate P1 Lys is characterised by almost optimal dihedral angles, hydrogen bonding distances and angles, in addition to considerably lower temperature factors. Thus, the trypsin S1 pocket seems to be designed particularly for lysine binding.     

Non-Boltzmann thermodynamic integration (NBTI) for macromolecular systems: relative free energy of binding of trypsin to benzamidine and benzylamine

The relative free energies of binding of trypsin to two amine inhibitors, benzamidine (BZD) and benzylamine (BZA), were calculated using non-Boltzmann thermodynamic integration (NBTI). Comparison of the simulations with the crystal structures of both complexes, trypsin-BZD and trypsin-BZA, shows that NBTI simulations better sample conformational space relative to thermodynamic integration (TI) simulations. The relative binding free energy calculated using NBTI was much closer to the experimentally determined value than that obtained using TI. The error in the TI simulation was found to be primarily due to incorrect sampling of BZA's conformation in the binding pocket. In contrast, NBTI produces a smooth mutation from BZD to BZA using a surrogate potential, resulting in a much closer agreement between the inhibitors' conformations and the omit electron density maps. This superior agreement between experiment and simulation, of both relative binding free energy differences and conformational sampling, demonstrates NBTI's usefulness for free energy calculations in macromolecular simulations.     

Refined 2.3 A X-ray crystal structure of bovine thrombin complexes formed with the benzamidine and arginine-based thrombin inhibitors NAPAP, 4-TAPAP and MQPA. A starting point for improving antithrombotics.

Well-diffracting crystals of bovine epsilon-thrombin in complex with several "non-peptidic" benzamidine and arginine-based thrombin inhibitors have been obtained by co-crystallization. The 2.3 A crystal structures of three complexes formed either with NAPAP, 4-TAPAP, or MQPA, were solved by Patterson search methods and refined to crystallographic R-values of 0.167 to 0.178. The active-site environment of thrombin is only slightly affected by binding of the different inhibitors; in particular, the exposed "60-insertion loop" essentially maintains its typical projecting structure. The D-stereoisomer of NAPAP and the L-stereoisomer of MQPA bind to thrombin with very similar conformations, as previously inferred from their binding to bovine trypsin; the arginine side-chain of the latter inserts into the specificity pocket in a "non-canonical" manner. The L-stereoisomer of 4-TAPAP, whose binding geometry towards trypsin was only poorly defined, is bound to the thrombin active-site in a compact conformation. In contrast to NAPAP, the distal p-amidino/guanidino groups of 4-TAPAP and MQPA do not interact with the carboxylate group of Asp189 in the thrombin specificity pocket in a "symmetrical" twin N-twin O manner, but through "lateral" single N-twin O contacts; in contrast to the p-amidino group of 4-TAPAP, however, the guanidyl group of MQPA packs favourably in the pocket due to an elaborate hydrogen bond network, which includes two entrapped water molecules. These thrombin structures confirm previous conclusions of the important role of the intermolecular hydrogen bonds formed with Gly216, and of the good sterical fit of the terminal bulky hydrophobic inhibitor groups with the hydrophobic aryl binding site and the S2-cavity, respectively, for tight thrombin active site binding of these non-peptidic inhibitors. These accurate crystal structures are presumed to be excellent starting points for the design and the elaboration of improved antithrombotics.     

Affinity chromatography of serine proteases on the triazine dye ligand Cibacron Blue F3G-A

The interaction between complement component factor B and the triazine dye ligand Cibacron Blue F3G-A coupled to a cross-linked agarose matrix (Blue Sepharose) was found to involve the Bb part of the molecule, and to be inhibited by benzamidine. Human, chicken and rainbow trout factor B which had bound to Blue Sepharose could subsequently be eluted with benzamidine. Other serine proteases (C2, factor II, factor IX, trypsin, chymotrypsin, proteinase 3) also bound to Blue Sepharose but only those belonging to the trypsin family could be eluted with benzamidine. Trypsin treated with the active-site inhibitor phenylmethylsulfonyl fluoride did not bind to Blue Sepharose and pretreatment of Blue Sepharose with benzamidine did not influence binding of proteases. We conclude that trypsin-like serine proteases can be purified on Blue Sepharose and that the interaction of these serine proteases with Blue Sepharose involves the active site of the enzyme.     

A molecular dynamics study of the three-dimensional model of human synovial fluid phospholipase A2—transition state mimic complexes

Different modes of binding of transition state mimics: amide, phosphonate and difluoro ketone, to human synovial fluid phospholipase A₂ (HSF PLA₂) are studies by molecular dynamics simulations computed in solvent. The results are analysed in the light of primary binding sites. Hydrogen bonding interaction plays an important role for amino acids such as Gly32, Val30, and Glu55, apart from the well known active site residues viz Asp48, Gly25, Gly29, Gly31, His27, His47, Lys62, Phe23, Asn114 and Tyr112. In addition, the hydrogen bonding interaction between Sn-1 tetrahedral phosphonate group of amide and difluoro ketone inhibitors and crystallographic water molecules (H₂O 523, H₂O 524 and H₂O 401) seems to have a significant role. Many of the active site charged residues display considerable movement upon ligand binding. The structural effects of ligand binding were analyzed from RMS deviations of Cα in the resulting energy-minimized average structures of the receptor–ligand complexes. The values of the RMS deviations differ among the HSF PLA₂s, in a pattern that is not the same for the three complexes. This suggests that ligands with different pharmacological efficacies induce different types of conformational changes of the receptor. Our active-orientation model is, at least qualitatively, consistent with experimental data and should be useful for the rational design of more potent inhibitors.     

Binding mode prediction for a flexible ligand in a flexible pocket using multi-conformation simulated annealing pseudo crystallographic refinement

We describe multi-conformation simulated annealing-pseudo-crystallographic refinement (MCSA-PCR), a technique developed for predicting the binding mode of a flexible ligand in a flexible binding pocket. To circumvent the local-minimum problem efficiently, this method performs multiple independent cycles of simulated annealing with explicit solvent, "growing" the ligand in the binding pocket each time. From the ensemble of structures, a pseudo-crystallographic electron density map is calculated, and then conventional crystallographic refinement methods are used to best fit a single, optimal structure into the density map. The advantage of the MCSA-PCR method is that it provides a direct means to evaluate the accuracy and uniqueness of the calculated solution, provides a measure of ligand and protein dynamics from the refined B-factors, and facilitates comparison with X-ray crystallographic data. Here, we show that our MCSA-PCR method succeeds in predicting the correct binding mode of the VSV8 peptide to the major histocompatibility complex (MHC) receptor. Importantly, there is a significant correlation between the experimentally determined crystallographic water molecules and water density observed in the pseudo map by MCSA-PCR. Furthermore, comparison of different approaches for extracting a single, most probable structure from the calculated ensemble reveals the power of the PCR method and provides insights into the nature of the energetic landscape.     

Induced‐fit or preexisting equilibrium dynamics? Lessons from protein crystallography and MD simulations on acetylcholinesterase and implications for structure‐based drug design

Crystal structures of acetylcholinesterase complexed with ligands are compared with side-chain conformations accessed by native acetylcholinesterase in molecular dynamics (MD) simulations. Several crystallographic conformations of a key residue in a specific binding site are accessed in a simulation of native acetylcholinesterase, although not seen in rotomer plots. Conformational changes upon ligand binding thus involve preexisting equilibrium dynamics. Consequently, rational drug design could benefit significantly from conformations monitored by MD simulations of native targets.     

Polar hydrogen positions in proteins: Empirical energy placement and neutron diffraction comparison

A method for the prediction of hydrogen positions in proteins is presented. The method is based on the knowledge of the heavy atom positions obtained, for instance, from X-ray crystallography. It employs an energy minimization limited to the environment of the hydrogen atoms bound to a common heavy atom or to a single water molecule. The method is not restricted to proteins and can be applied without modification to nonpolar hydrogens and to nucleic acids. The method has been applied to the neutron diffraction structures of trypsin ribonuclease A, and bovine pancreatic trypsin inhibitor. A comparison of the constructed and the observed hydrogen positions shows few deviations except in situations in which several energetically similar conformations are possible. Analysis of the potential energy of rotation of Lys amino and Ser, Thr, Tyr hydroxyl groups reveals that the conformations of lowest intrinsic torsion energies are statistically favored in both the crystal and the constructed structures.     

X-Ray absorption studies of Zn2+ binding sites in bacterial, avian, and bovine cytochrome bc1 complexes

Binding of Zn2+ has been shown previously to inhibit the ubiquinol cytochrome c oxidoreductase (cyt bc1 complex). X-ray diffraction data in Zn-treated crystals of the avian cyt bc1 complex identified two binding sites located close to the catalytic Qo site of the enzyme. One of them (Zn01) might interfere with the egress of protons from the Qo site to the aqueous phase. Using Zn K-edge x-ray absorption fine-structure spectroscopy, we report here on the local structure of Zn2+ bound stoichiometrically to noncrystallized cyt bc1 complexes. We performed a comparative x-ray absorption fine-structure spectroscopy study by examining avian, bovine, and bacterial enzymes. A large number of putative clusters, built by combining information from first-shell analysis and metalloprotein databases, were fitted to the experimental spectra by using ab initio simulations. This procedure led us to identify the binding clusters with high levels of confidence. In both the avian and bovine enzyme, a tetrahedral ligand cluster formed by two His, one Lys, and one carboxylic residue was found, and this ligand attribution fit the crystallographic Zn01 location of the avian enzyme. In the chicken enzyme, the ligands were the His121, His268, Lys270, and Asp253 residues, and in the homologous bovine enzyme they were the His121, His267, Lys269, and Asp254 residues. Zn2+ bound to the bacterial cyt bc1 complex exhibited quite different spectral features, consistent with a coordination number of 6. The best-fit octahedral cluster was formed by one His, two carboxylic acids, one Gln or Asn residue, and two water molecules. It was interesting that by aligning the crystallographic structures of the bacterial and avian enzymes, this group of residues was found located in the region homologous to that of the Zn01 site. This cluster included the His276, Asp278, Glu295, and Asn279 residues of the cyt b subunit. The conserved location of the Zn2+ binding sites at the entrance of the putative proton release pathways, and the presence of His residues point to a common mechanism of inhibition. As previously shown for the photosynthetic bacterial reaction center, zinc would compete with protons for binding to the His residues, thus impairing their function as proton donors/acceptors.     

Stereochemistry of binding of the tetrapeptide acetyl-Pro-Ala-Pro-Tyr-NH2 to porcine pancreatic elastase. Combined use of two-dimensional transferred nuclear Overhauser enhancement measurements, restrained molecular dynamics, X-ray crystallography and molecular modelling

A nuclear magnetic resonance study of the conformation of the tetrapeptide acetyl-Pro-Ala-Pro-Tyr-NH2 bound to porcine pancreatic elastase is presented. From two-dimensional transferred nuclear Overhauser enhancement measurements, a set of 23 approximate distance restraints between pairs of bound ligand protons, indicative of an extended type structure, is derived. The structure of the bound tetrapeptide is then refined from two different starting structures (an extended beta-strand and a polyproline helix) by restrained molecular dynamics, in which the interproton distances are incorporated into the total energy of the system in the form of effective potentials. Convergence to essentially the same average restrained dynamics structures is achieved. The refined structures are then modelled into the active site of elastase by interactive molecular graphics. The determination of the anchor point of the bound tetrapeptide on the enzyme was aided by a simultaneous crystallographic study which, despite the fact that only electron density for a Pro-X dipeptide fragment was visible, enabled both the approximate position and orientation of binding to be determined. It is found that the tetrapeptide is bound in the S' binding site in the reverse orientation found in other serine protease-inhibitor complexes and is stabilized both by hydrogen-bonding and by van der Waals' interactions.     

X-ray investigation of the binding of 1,10-phenanthroline and imidazole to horse-liver alcohol dehydrogenase

We have studied the binding of two inhibitor molecules, imidazole and 1,10-phenanthroline, to liver alcohol dehydrogenase by crystallographic methods. X-ray data for the imidazole complex were collected to 0.29-nm resolution and for the 1,10-phenanthroline complex to 0.45-nm resolution. In both cases we found only one peak in the difference electron density maps close to the active zinc atom. The peak corresponding to 1,10-phenanthroline overlaps the site of the density of the zinc-bound water in the apoenzyme and the imidazole density partly overlaps this density. We can not discern any additional peaks close to the zinc atom which would correspond to new positions of bound water. We thus conclude that both these inhibitors bind to the catalytic zinc atom and that upon binding they displace the water molecule that is firmly bound to this zinc atom in the apoenzyme. We do not see any structural changes in the remaining part of the molecule.     

Screening and docking chemical ligands onto pocket cavities of a protease for designing a biocatalyst

A detailed study of the trypsin surface has been carried out to gain insight into its biological functions and interactions which helped to determine the binding specificity. Twenty-four cavity pockets were automatically identified on trypsin from PDB file entry 1AUJ using CASTp (Computed Atlas of Surface Topography of proteins). Molecular docking was exploited as an efficient in silico screening tool for studying protein–ligand interactions. A systematic docking study using Autodock 3.05 has been performed on the five largest binding pockets in trypsin. A set of ten putative chemical ligands was used to dock into selected binding pockets. Docking of ligands into the five largest pockets in trypsin showed that 1,10-phenanthroline and ethanolamine preferentially bound at pocket 24 and benzamidine at pocket 22. Thermodynamically, we also found that ethanol, propanol, propandiol and phosphoethanolamine preferentially bound at pocket 21 whereas p-aminobenzamidine, phenylacetic acid and phenylalanine interacted mainly at pocket 20 based on their lowest interaction free energy.     

Role of His505 in the soluble fumarate reductase from Shewanella frigidimarina

The X-ray structure of the soluble fumarate reductase from Shewanella frigidimarina [Taylor, P., Pealing, S. L., Reid, G. A., Chapman, S. K., and Walkinshaw, M. D. (1999) Nat. Struct. Biol. 6, 1108-1112] clearly shows the presence of an internally bound sodium ion. This sodium ion is coordinated by one solvent water molecule (Wat912) and five backbone carbonyl oxygens from Thr506, Met507, Gly508, Glu534, and Thr536 in what is best described as octahedral geometry (despite the rather long distance from the sodium ion to the backbone oxygen of Met507 (3.1 A)). The water ligand (Wat912) is, in turn, hydrogen bonded to the imidazole ring of His505. This histidine residue is adjacent to His504, a key active-site residue thought to be responsible for the observed pK(a) of the enzyme. Thus, it is possible that His505 may be important in both maintaining the sodium site and in influencing the active site. Here we describe the crystallographic and kinetic characterization of the H505A and H505Y mutant forms of the Shewanella fumarate reductase. The crystal structures of both mutant forms of the enzyme have been solved to 1.8 and 2.0 A resolution, respectively. Both show the presence of the sodium ion in the equivalent position to that found in the wild-type enzyme. The structure of the H505A mutant shows the presence of two water molecules in place of the His505 side-chain which form part of a hydrogen-bonding network with Wat48, a ligand to the sodium ion. The structure of the H505Y mutant shows the hydroxyl group of the tyrosine side-chain hydrogen-bonding to a water molecule which is also a ligand to the sodium ion. Apart from these features, there are no significant structural alterations as a result of either substitution. Both the mutant enzymes are catalytically active but show markedly different pH profiles compared to the wild-type enzyme. At high pH (above 8.5), the wild type and mutant enzymes have very similar activities. However, at low pH (6.0), the H505A mutant enzyme is some 20-fold less active than wild-type. The combined crystallographic and kinetic results suggest that His505 is not essential for sodium binding but does affect catalytic activity perhaps by influencing the pK(a) of the adjacent His504.     

Lipid-protein interactions in lipovitellin

The refined molecular structure of lipovitellin is described using synchrotron cryocrystallographic data to 1.9 A resolution. Lipovitellin is the predominant lipoprotein found in the yolk of egg-laying animals and is involved in lipid and metal storage. It is thought to be related in amino acid sequence to segments of apolipoprotein B and the microsomal transfer protein responsible for the assembly of low-density lipoproteins. Lipovitellin contains a heterogeneous mixture of about 16% (w/w) noncovalently bound lipid, mostly phospholipid. Previous X-ray structural studies at ambient temperature described several different protein domains including a large cavity in each subunit of the dimeric protein. The cavity was free of any visible electron density for lipid molecules at room temperature, suggesting that only dynamic interactions exist with the protein. An important result from this crystallographic study at 100 K is the appearance of some bound ordered lipid along the walls of the binding cavity. The precise identification of the lipid type is difficult because of discontinuities in the electron density. Nonetheless, the conformations of 7 phospholipids and 43 segments of hydrocarbon chains greater than 5 atoms in length have been discovered. The conformations of the bound lipid and the interactions between protein and lipid provide insights into the factors governing lipoprotein formation.     

Crystallographic analysis of the binding of NADPH, NADPH fragments, and NADPH analogues to glutathione reductase

The binding of the substrate NADPH as well as a number of fragments and derivatives of NADPH to glutathione reductase from human erythrocytes has been investigated by using X-ray crystallography. Crystals of the enzyme were soaked with the compounds of interest, and then the diffraction intensities were collected out to a resolution of 3 A. By use of phase information from the refined structure of the native enzyme in its oxidized state, electron density maps could be calculated. Difference Fourier electron density maps with coefficients Fsoak - Fnative showed that the ligands tested bound either at the functional NADPH binding site or not at all. Electron density maps with coefficients 2Fsoak - Fnative were used to estimate occupancies for various parts of the bound ligands. This revealed that all ligands except NADPH and NADH, which were fully bound, showed differential binding between the adenine end and the nicotinamide end of the molecule: The adenine end always bound with a higher occupancy than the nicotinamide end. Models were built for the protein-ligand complexes and subjected to restrained refinement at 3-A resolution. The mode of binding of NADPH, including the conformational changes of the protein, is described. NADH binding is clearly shown to involve a trapped inorganic phosphate at the position normally occupied by the 2'-phosphate of NADPH. A comparison of the binding of NADPH with the binding of the fragments and analogues provides a structural explanation for their relative binding affinities. In this respect, proper charge and hydrogen-bonding characteristics of buried parts of the ligand seem to be particularly important.     

Synthesis and characterization of biotinylated forms of insulin-like growth factor-1: topographical evaluation of the IGF-1/IGFBP-2 AND IGFBP-3 interface

Activation of the insulin-like growth factor-1 (IGF)-1 receptor signaling pathways by IGF-1 and IGF-2 results in mitogenic and anabolic effects. The bioavailability of the IGFs is regulated by six soluble binding proteins, the insulin-like growth factor binding proteins (IGFBPs), which bind with approximately 0.1 nM affinity to the IGFs and often serve as endogenous antagonists of IGF action. To identify key domains of IGF-1 involved in the interaction with IGFBP-2 and IGFBP-3, we employed IGF-1 selectively biotinylated on residues Gly 1, Lys 27, Lys 65, and Lys 68. All monobiotinylated species of IGF-1 exhibited high affinity ( approximately 0.1-0.2 nM) for IGFBP-2 and IGFBP-3 in solid-phase-binding assays. However, different labeling intensities were observed in ligand blot analysis of IGFBP-2 and IGFBP-3. The N(epsilon)(Lys65/68)(biotin)-IGF-1 (N(epsilon)(Lys65/68b)-IGF-1) probe exhibited the highest signal intensity, while N(alpha)(Gly1b)-IGF-1 and N(epsilon)(Lys27b)-IGF-1 demonstrated significantly lower signals. When taken together, these results suggest that, once bound to IGFBP-2 or IGFBP-3, the biotin moieties of N(alpha)(Gly1b)-IGF-1 and N(epsilon)(Lys27b)-IGF-1 are inaccessible to NeutrAvidin-peroxidase, the secondary binding component. Ligand blots using IGF-1 derivatized with a long chain form of the N-hydroxysuccinimide biotin (NHS-biotin) to yield N(alpha)(Gly1)(LC-biotin)-IGF-1 and N(epsilon)(Lys27)(LC-biotin)-IGF-1 demonstrated increased signal intensity compared with their NHS-biotin counterparts. In BIAcore analysis, IGFBP-2 and IGFBP-3 bound only to the N(epsilon)(Lys65/68b)-IGF-1-coated flowcell of a biosensor chip, confirming the inaccessibility of Gly 1 and Lys 27 when IGF-1 is bound to IGFBP-2 and IGFBP-3. These data confirm the involvement of the IGFBP-binding domain on IGF-1 in binding to IGFBP-2 and IGFBP-3 and support involvement of the IGF-1R-binding domain in IGFBP binding.     

Surface-simulation synthesis of the substrate-binding site of an enzyme. Demonstration with trypsin.

From the X-ray co-ordinates of bovine trypsin and its complexes with substrate analogues (benzamidine) and with soya-bean trypsin inhibitor, a peptide (TP) was designed and synthesized by surface-simulation synthesis, a concept previously introduced by this laboratory, to mimic the binding site of trypsin. Also, a control peptide (CTP) was synthesized that contained all the amino acids present in the TP peptide, except that their order was randomized. The radioiodinated TP peptide bound specifically to adsorbents of benzamidine, whereas the control CTP peptide exhibited no binding activity. Conjugates to succinyl (3-carboxypropionyl)-lysozyme of the TP peptide, control CTP peptide and other unrelated peptides were examined by a radiometric binding assay for the ability to bind soya-bean trypsin inhibitor and human alpha 1-antitrypsin. Conjugates of the TP peptide exhibited considerable binding activity to adsorbents of soya-bean trypsin inhibitor or alpha 1-antitrypsin. None of the other peptide conjugates possessed any binding activity. Action of the active-site-directed reagents phenylmethanesulphonyl fluoride and di-isopropyl phosphorofluoridate on free TP and CTP peptides resulted in the modification of a serine residue in the TP peptide whereas the CTP peptide remained unaltered. The TP peptide, either in the free form or as a conjugate on succinyl-lysozyme, had no enzymic activity on protein substrates or on tosylarginine methyl ester. These findings indicated that the binding activity of an enzyme was well mimicked by the surface-stimulation peptide but that reproduction of the catalytic activity was not obtained.     

Refined crystal structure of the triphosphate conformation of H-ras p21 at 1.35 A resolution: implications for the mechanism of GTP hydrolysis.

The crystal structure of the H-ras oncogene protein p21 complexed to the slowly hydrolysing GTP analogue GppNp has been determined at 1.35 A resolution. 211 water molecules have been built into the electron density. The structure has been refined to a final R-factor of 19.8% for all data between 6 A and 1.35 A. The binding sites of the nucleotide and the magnesium ion are revealed in high detail. For the stretch of amino acid residues 61-65, the temperature factors of backbone atoms are four times the average value of 16.1 A2 due to the multiple conformations. In one of these conformations, the side chain of Gln61 makes contact with a water molecule, which is perfectly placed to be the nucleophile attacking the gamma-phosphate of GTP. Based on this observation, we propose a mechanism for GTP hydrolysis involving mainly Gln61 and Glu63 as activating species for in-line attack of water. Nucleophilic displacement is facilitated by hydrogen bonds from residues Thr35, Gly60 and Lys16. A mechanism for rate enhancement by GAP is also proposed.     

Evaluation of protein-protein association energies by free energy perturbation calculations

The association energy upon binding of different amino acids in the specificity pocket of trypsin was evaluated by free energy perturbation calculations on complexes between bovine trypsin (BT) and bovine pancreatic trypsin inhibitor (BPTI). Three simulations of mutations of the primary binding residue (P(1)) were performed (P(1)-Ala to Gly, P(1)-Met to Gly and P(1)-Met to Ala) and the resulting differences in association energy (DeltaDeltaG(a)) are 2. 28, 5.08 and 2.93 kcal/mol for P(1)-Ala to Gly, P(1)-Met to Gly and to Ala with experimental values of 1.71, 4.62 and 2.91 kcal/mol, respectively. The calculated binding free energy differences are hence in excellent agreement with the experimental binding free energies. The binding free energies, however, were shown to be highly dependent on water molecules at the protein-protein interface and could only be quantitatively estimated if the correct number of such water molecules was included. Furthermore, the cavities that were formed when a large amino acid side-chain is perturbed to a smaller one seem to create instabilities in the systems and had to be refilled with water molecules in order to obtain reliable results. In addition, if the protein atoms that were perturbed away were not replaced by water molecules, the simulations dramatically overestimated the initial state of the free energy perturbations.