Sunday, December 10, 2006

Protein flexibilty and ligand interaction complementarity example


Human neutrophil elastase complexed with an inhibitor (gw475151) studied with QUANTUM normal modes analysis tools. The video shows the inhibitor (red licorice) remains in a strongly interacting position regardless of the protein most probable motion:



Most of available docking software performs free energy scoring for different ligands positions on the same rigid protein structure. The validity of such procedure can not be easily asserted. Not only different PDB Data Bank structures of the same protein are different, NMR studies show often impressive protein flexibility and thus uncertainty in the protein atoms positions. The necessity to compensate for the lack of the structure information makes scoring functions developers utilize smooth energy scores corresponding to some kind of coarse grain approximation for the protein-ligand interactions. Such an averaging leads to inability to recover fine details of interactions and hence to lack of selectivity and false postives, i.e. hits with binding constants actually lower than predicted.


The movie above helps explain this situation. While a scoring function could suggest an accurate value for the particular ligand docked in the experimentally observed position, the scores for "hits" overlapping with the protein motion could be the same good or better, but false, since the protein position extracted from the PDB data represents only a single member of the statistical ensemble. Mathematically speaking the ligand gw475151 does not only have a good interaction energy (enthalpy), but also remains complementary to the protein pocket in spite of sufficiently large protein displacements (has a low entropy loss associated with the protein degrees of freedom).


Materials and methods:The initial information about the protein-ligand structure is taken from 1h1b PDB entry. The ligand was taken out and protein was let to relax in QUANTUM continuum water (see cond-mat/0601129). After a sufficiently long molecular dynamics simulation the protein motion has been analyzed and the lowest normal mode (highest amplitude protein motion) was separated. The ligand is then placed back at a fixed position to highlight dynamic overlap between the ligand steric interactions and the protein motion.