Binding free energy prediction in strongly hydrophobic biomolecular systems Landry Charlier, a Claude Nespoulous,w b Se´bastien Fiorucci, a Serge Antonczak a and Je´rome Golebiowski* a Received 4th July 2007, Accepted 3rd September 2007 First published as an Advance Article on the web 17th September 2007 DOI: 10.1039/b710186d We present a comparison of various computational approaches aiming at predicting the binding free energy in ligand–protein systems where the ligand is located within a highly hydrophobic cavity. The relative binding free energy between similar ligands is obtained by means of the thermodynamic integration (TI) method and compared to experimental data obtained through isothermal titration calorimetry measurements. The absolute free energy of binding prediction was obtained on a similar system (a pyrazine derivative bound to a lipocalin) by TI, potential of mean force (PMF) and also by means of the MMPBSA protocols. Although the TI protocol performs poorly either with an explicit or an implicit solvation scheme, the PMF calculation using an implicit solvation scheme leads to encouraging results, with a prediction of the binding affinity being 2 kcal mol 1 lower than the experimental value. The use of an implicit solvation scheme appears to be well suited for the study of such hydrophobic systems, due to the lack of water molecules within the binding site. Introduction Accurate prediction of binding free energy, a value directly connected to the binding affinity or the binding constant between molecular systems is one of the most challenging tasks for computational chemists and biochemists. 1 Such calculations provide a way of making the connection between macroscopic measurements and microscopic or atomic-scale behaviours based on structural data. Nowadays, state-of-the- art free energy calculations, by means of statistical simulations can offer quite an efficient answer to this problem, helping in the prediction of the behaviour of specific targets dedicated to a given activity. The knowledge of the affinity between some molecular species prior to experiments would represent a clear advance for experimentalists, with the condition that the prediction relies on a strongly validated protocol. It follows that a good knowledge of the performance, in terms of success and drawback of the various calculations techniques for each type of system (neutral, charged, hydrophobic, hydrophilic...) are of paramount importance for further work dedicated to the prediction of binding affinities. For classical systems where the balance between hydrophilic and hydrophobic interactions is equilibrated, numerous free energy calculation examples are found. Although ranking between ligands, obtained by means of relative binding free energies, are generally properly represented, absolute binding free energy calculations still remain difficult. Successful abso- lute binding free energy calculations have been reported for the T4 lysozyme and some of its mutants. 2,3 Other calculations involved FKPB, a protein from the immunophilin group that has an affinity to FK506 drugs and antibiotics. The binding free energy between FKBP and 4-hydroxy-2-butanone has been obtained with the MM-PB(GB)SA protocol. 4 The affinity has been computed with an over binding of 2.9 kcal mol 1 with respect to experiment. The same protein bound to 8 structurally related ligands has also been studied by means of state-of-the-art free energy perturbation approaches, either with or without simulation bias, such as distance restraint. Both free and restrained calculations lead to a small under- binding (B1 or 2 kcal mol 1 ), although the binding hierar- chies are correctly predicted. 5,6 Also, the use of umbrella sampling calculations to obtain a potential of mean force for FKBP bound to FK506 and hydroxy-butanone led to accurate prediction of the binding energetics, contrarily to the MM/ PB-SA approach. 7 The biotin–strepatavidin system is also a thoroughly studied system. Here, the interaction is mostly hydrophilic, with a large network of hydrogen bonds forming a rigid lattice accompanied by optimal van der Waals contacts. These strong interactions lead to an absolute binding free energy of 18.3 kcal mol 1 , one of the largest affinities in natural biomolecular systems. 8 This large affinity is well represented by means of both FEP and TI approaches, with an accuracy of B2 kcal mol 1 . 9 Another work aiming to predict the different contributions to the binding energetics was less optimistic, since the calculated binding affinities were in the correct range but showed very high uncertainties that the authors attribute to large protein reorganization and ligand flexibility. 10 a LCMBA, Faculte ´ des sciences de Nice – Sophia Antipolis, Centre National de la Recherche Scientifique, UMR 6001, Universite´ de Nice-Sophia-Antipolis, UFR Sciences, Parc Valrose, 28, avenue Valrose, 06108 Nice Cedex 2, France. E-mail: jerome.golebiowski@unice.fr; Fax: +33 4 92 07 61 25; Tel: +33 4 92 07 61 03 b NOPA BOG, UMR1197, INRA – Universite ´ Paris XI, Domaine de Vilvert, 78352 Jouy en Josas Cedex, France w Present address: UR1199 Prote´omique, Campus SupAgroINRA, 34060 Montpellier Cedex 2, France. This journal is  c the Owner Societies 2007 Phys. Chem. Chem. Phys., 2007, 9, 5761–5771 | 5761 PAPER www.rsc.org/pccp | Physical Chemistry Chemical Physics