Journal of Computer-Aided Molecular Design 17: 129–134, 2003. © 2003 Kluwer Academic Publishers. Printed in the Netherlands. 129 On the detection of multiple-binding modes of ligands to proteins, from biological, structural, and modeling data Paul J. Lewis 1,* , Marc de Jonge 1 , Frits Daeyaert 1 , Luc Koymans 1 , Maarten Vinkers 1 , Jan Heeres 1 , Paul A.J. Janssen 1 , Eddy Arnold 2 , Kalyan Das 2 , Art D. Clark Jr. 2 , Stephen H. Hughes 3 , Paul L. Boyer 3 , Marie-Pierre de B´ ethune 4 , Rudi Pauwels 4 , Koen Andries 5 , Mike Kukla 6 , Donald Ludovici 6 , Bart De Corte 6 , Robert Kavash 6 & Chih Ho 6 1 Center for Molecular Design, J&JPRD, Janssen Pharmaceutica N.V, Vosselaar, Belgium; 2 Center for Biotech- nology and Medicine, and Chemistry and Chemical Biology Department, Rutgers University, NJ; 3 NIH-NCI HIV Drug Resistance Program, Frederick, MD; 4 Tibotec-Virco, Mechelen, Belgium; 5 Virology Dept., J&JPRD, Janssen Pharmaceutica N.V, Beerse, Belgium; 6 Chemistry Dept., J&JPRD, Spring House, PA, USA Received 1 November 2002; Accepted for publication 1 December 2002 Key words: molecular modeling, virology, HIV, X-ray crystallography, drug design, statistical analysis, PLS Summary There are several indications that a given compound or a set of related compounds can bind in different modes to a specific binding site of a protein. This is especially evident from X-ray crystallographic structures of ligand-protein complexes. The availability of multiple binding modes of a ligand in a binding site may present an advantage in drug design when simultaneously optimizing several criteria. In the case of the design of anti-HIV compounds we observed that the more active compounds that are also resilient against mutation of the non-nucleoside binding site of HIV1-reverse transcriptase make use of more binding modes than the less active and resilient compounds. Introduction Multiple binding modes of ligands to proteins have a dual nature. First, flexible ligands with rotatable frag- ments may adapt their conformation to a given binding site, at a relatively low cost in energy. Second, both the positions of the backbone and amino acid side chains of proteins can be induced to change their orientation and position in response to ligand binding; both phe- nomena may appear simultaneously because neither the ligand nor the protein is rigid. The metaphor of a flexible hand fitting into a flexible glove (or, alternat- ively, a ‘handshake’) is more applicable here than the rigid lock and key paradigm of Emil Fischer [1]. The dual aspect of multiple binding modes is re- flected in many computational protocols for structure- based drug design. For example, some 10 to 20 repres- entative low-energy conformational structures of the * To whom correspondence should be addressed. E-mail: plewi@prdbe.jnj.com. ligand may be defined at the start of the computa- tion, each of which deviates from the minimal energy conformation by a relatively small amount (e.g. no more than 5 kcal/mole). Only those structures are re- tained that can be docked into the rigid binding site without significant steric clashes with the binding site. For each docked conformation the ligand and adjacent amino acid side chains and backbone are relaxed and the overall energy of the complex is minimized. The final result can be a weighted average binding energy over the several conformers that have been docked and optimized. The average may extend over several different binding configurations of the ligand and the protein, depending upon the degree of flexibility that is allowed. A shortcoming of this approach is that it does not permit relatively large displacements of the backbone and the side chains of the binding site, which can be induced by the ligand.