It has been recognized for many protein–ligand complexes that certain regions of the binding surface, often called ‘hot spots’, contribute a disproportionate amount to the binding free energy (Hajduk et al.
). Such regions are more likely to bind small drug-like compounds with high affinity than the rest of the binding site, and hence their identification is important for drug design (DeLano, 2002
; Vajda and Guarnieri, 2006
). Both NMR (nuclear magnetic resonance) and X-ray crystallography techniques have been used to find ‘hot spots’ of proteins. Using NMR the 15
N-labeled protein is screened against a library of small probe compounds (Hajduk et al.
; Vajda and Guarnieri, 2006
). Applications to a variety of proteins demonstrate that the ‘hot spots’ bind a variety of small molecules, and a high ‘hit rate’ is a good predictor of druggability. Indeed, a high correlation was observed between the number of different probes binding to a site, and the ability to identify high-affinity druglike ligands that bind there (Hajduk et al.
). The X-ray technique, known as Multiple Solvent Crystal Structures (MSCS) method, is based on solving the structure of the protein in aqueous solutions of various probe compounds, primarily organic solvents (Mattos and Ringe, 1996
). Each structure shows a few organic molecules associated with the protein surface in the first shell of water molecules. The power of the method arises from superimposing a number of structures solved in different solvents. Most probes generally cluster in the binding site, and the overlapping probe clusters form ‘consensus’ sites (CSs) that delineate the functionally most important subsites. As demonstrated by applications to porcine elastase (Allen et al.
; Mattos and Ringe, 1996
; Mattos et al.
) and thermolysin (English et al.
), some probes may also bind at crystal contacts or in small buried pockets, but the large CSs occur in the ‘hot spots’ of the binding site.
Since the identification of ‘hot spots’ by NMR or X-ray crystallography is very expensive, it is important to explore whether similar information can be obtained by computations. Computational mapping methods place molecular probes on the protein surface in order to explore its binding properties. A number of methods identify potential binding sites (An et al.
; Laurie and Jackson, 2005
; Glaser et al.
). Some early methods such as GRID (Goodford, 1985
) and Multiple Copy Simultaneous Search (MCSS) (Caflisch et al.
; Miranker and Karplus, 1991
) have been developed to find favorable binding positions for specific molecules or functional groups rather than to identify ‘hot spots’. Both methods result in many energy minima, and it is difficult to determine which of the minima are actually relevant (Mattos and Ringe, 1996
In this article, we describe the FTMAP algorithm specifically developed to reproduce NMR and X-ray mapping results using a number of organic probe molecules. For each probe the algorithm generates 2000 bound positions using rigid body docking, refines the positions by energy minimization, clusters the resulting conformations and ranks the clusters on the basis of this average free energy. The docking step is based on the fast Fourier transform (FFT) correlation approach which helps to efficiently sample billions of probe positions on dense translational and rotational grids, but can use only sums of correlation functions for scoring and hence, is generally restricted to very simple energy expressions. The novelty of FTMAP is that we were able to incorporate and represent on grids a detailed energy expression which includes attractive and repulsive van der Waals terms, electrostatic interaction energy based on Poisson–Boltzmann calculations, a cavity term to represent the effect of non-polar enclosures and a structure-based pairwise interaction potential. The energy expression is key to the accuracy of FTMAP, and the algorithm matches or exceeds the accuracy of our earlier mapping method (Dennis et al.
; Silberstein et al.
), although FTMAP requires only about one-sixth of the CPU time. Here, we show that FTMAP reproduces the X-ray mapping results for elastase very well (Allen et al.
; Mattos et al.
). The second application presented is to renin, a long-standing pharmaceutical target for the treatment of hypertension, with the first renin inhibitor approved in 2007 (Rahuel et al.
; Wood et al.
). For renin we do not have experimental mapping results, and our goal here is to show that mapping can reliably identify the ‘hot spots’ that substantially contribute to the free energy of ligand binding and hence should be the primary targets of drug design efforts. Two more applications are described in Supplementary Material
. In addition, the FTMAP server page provides mapping results for both unbound and ligand-bound structures of 10 drug target proteins (http://ftmap.bu.edu/