The same and the not same

One of the most important questions that someone in the early stages of drug discovery can ask is: Do similar ligands bind in a similar manner? Ambiguities riddle this seemingly commonsense question, right from the definition of 'similar'. Countless drug discovery projects must have been made or broken because of attention or lack of attention to this central principle.

If I had to give a one word answer to the question, it would be no, not because it's the right answer, but I would be playing safe by saying that. It can be a very big mistake to assume that similar ligands will bind to the same way in a protein binding pocket, or even in the same pocket, and medicinal chemists both on the experimental and computational sides are aware what a wide and disparate range of SAR relationships similar ligands can have. This extends very much to similar binding too.

Not on one occasion have medicinal chemists taken the known binding conformation of a ligand, and then twisted and turned another 'similar' ligand into that same conformation. The two structures are then superimposed. They look so similar, with the hydrophobic and polar groups in the same places. Ergo, they must be overlapping in their binding mode. Big mistake. While this can turn out to be the case many times, it's just wrong on a philosophical basis to assume it. That's because ligands have their own personalities, and each one of them can prefer to act quite differently with a protein binding pocket. The problem is; superposition of ligands can always be justified in retrospect if your ligand shows activity. But that does not make your assumption true.

One of the better known cases concerns the search for a 'common pharmacophore' for Taxol, Epothilone, and other ligands which bind to the same pocket in tubulin. A common pharmacophore is a minimal set of common structural features which will cause bioactivity. In typical fashion, researchers compared the various parts of the molecules and twisted them to overlap with each other (PNAS paper). Based on this superposition, they then designed analogs. Result: most of the analogs turned out to be inactive. Thus, there was no 'common pharmacophore'. In this context, Snyder and others published a model (SCIENCE paper) in which, doing meticulous electron density fitting for Taxol and Epothilone, they demonstrated that each one of these molecules explores the binding pocket of tubulin in a unique way, utilizing unique interactions. Thus, the ligands are 'promiscuous'.

Many times, modeling may assume that similar ligands are binding in the same conformation. Docking programs dock them in the same conformation. What docking programs fail to take into account is mainly protein flexibility, which accounts as much for ligand binding as ligand flexibility. Then, when X-ray structures of those 'similar' ligands bound to the same protein are obtained, they reveal that either the protein underwent a conformational change and changed the binding modes of the ligands, or that even without this, the ligands bound in a dissimillar manner. Ligand binding is a 1 kcal/mol energy window game, and there's no telling how each ligand will exploit this window.

So it's timely that a Swedish team has published a paper in J. Med. Chem that tries to tackle the question: Do similar ligands bind in the same way? The team has used the Tanimoto index to measure similarity of ligands, and then has looked at many samples from the PDB to guage the binding of similar ligands to the same protein. They try to use three main criteria of difference:
1. The position of water molecules
2. The movement of backbone atoms
3. The movement of protein sidechain atoms.

The first factor significantly turned out to be the most different for their examined cases, and one which is not always paid attention to. Water is a fickle guest for a protein host, and it can mediate interactions differently even in ligands with slight structural differences. The second factor is also significant, as side chain conformational differences induced by particular ligand features can greatly change the electrostatic environment of the protein; as illustrated above from the paper, the conformation of a single Met residue changes the electrostatics. The third factor, backbone movements, is relatively unchanged and a benign variable.

The bottom line is, it may be an ok working hypothesis to assume that because your known ligand binds in a known conformation, other very similar ligands or even known actives will bind the same way. But start taking it as an obvious rule, and you can always expect trouble.