The Impact of Molecular Dynamics Sampling on the Performance of Virtual Screening against GPCRs

The formation of ligand–protein
complexes requires simultaneous
adaptation of the binding partners. In structure based virtual screening,
high throughput docking approaches typically consider the ligand flexibility,
but the conformational freedom of the protein is usually taken into
account in a limited way. The goal of this study is to elaborate a
methodology for incorporating protein flexibility to improve the virtual
screening enrichments on GPCRs. Explicit-solvated molecular dynamics
simulations (MD) were carried out in lipid bilayers to generate an
ensemble of protein conformations for the X-ray structures and homology
models of both aminergic and peptidergic GPCRs including the chemokine
CXCR4, dopamine D3, histamine H4,
and serotonin 5HT6 holo receptor complexes.
The quality of the receptor models was assessed by enrichment studies
to compare X-ray structures, homology models, and snapshots from the
MD trajectory. According to our results, selected frames from the
MD trajectory can outperform X-ray structures and homology models
in terms of enrichment factor and AUC values. Significant changes
were observed considering EF1% values: comparing the original CXCR4, D3, and H4 targets and the additional
5HT6 initial models to that of the best MD frame resulted
in 0 to 6.7, 0.32 to 3.5 (10×), 13.3 to 26.7 (2×), and 0
to 14.1 improvements, respectively. It is worth noting that rank-average
based ensemble evaluation calculated for different ensemble sizes
could not improve the results further. We propose here that MD simulation
can capture protein conformations representing the key interacting
points of the receptor but less biased toward one specific chemotype.
These conformations are useful for the identification of a “consensus”
binding site with improved performance in virtual screening.