P07-04

Development of accurate in silico screening protocol based on protein structural fluctuation and drug binding mode

Hiroto TERADA *, Kei MORITSUGU

Graduate School of Science, Osaka Metropolitan University
( * E-mail: sg24479j@st.omu.ac.jp )

Structure-based in silico screening is widely used as an important tool in drug discovery. It is often the case that a single crystal structure is used as a receptor, then questions arise about how suitable the receptor structure is for a diverse compound library and whether the structures obtained from molecular dynamics (MD) simulations are useful for improved accuracy relative to using a single structure. In this study, we examined comprehensive evaluations of how to utilize the crystal structure ensemble, what kinds of MD simulations are good to explore structural variations, how to process these structural ensembles using structural clustering, and how to quantify the binding affinity of each inhibitor using available docking scores.
In this study, we used as two test systems, Type III inhibitors of MEK1, a phosphorylation enzyme involved in the MAPK signaling pathway, and Type I inhibitors of EGFR, a receptor tyrosine kinase involved in signaling pathways related to cell proliferation. From all co-crystal structures available in the KLIFS database, we chose the inhibitors with experimental IC50 values for evaluating the in silico screening methods. To evaluate the usefulness of the MD simulation, we performed 1-microsecond simulations of both MEK1 (PDB: 1s9j) and EGFR (4jrv), and for both apo and drug-bound holo forms. The structural clusterings to obtain a set of representative structures were then calculated by various patterns. Comprehensive docking simulations were then carried out using Autodock Vina for all inhibitors against the obtained receptor structures. Various patterns of binding affinity scores were attempted by combining thus derived docking scores, and the prediction accuracy was examined.
The correlation coefficient between the experimental IC50 values and the binding affinity scores of the inhibitors showed that using a single co-crystal structure underwent inaccurate prediction. In contrast, averaging the docking scores derived from multiple structures significantly improved the correlations. Additionally, it was found that using MD structures of the holo form resulted in better correlations rather than the apo form. The contact analysis of the inhibitor during MD simulation showed that the lead skeleton appropriately constrained the movement of the receptor protein's binding site, allowing the generation of structures that accommodate various compound modifications. This highlights the strength of the MD simulation in holo form to generate suitable receptor structures for docking simulation. Furthermore, assuming the case of no crystal structure available, the protocol of making structural modeling, docking the lead skeleton of the inhibitors, and performing the associated holo-form MD simulations was conducted, indicating comparable accuracy with the result using a co-crystal structure.