An in silico integrated pipeline for affinity and intrinsic activity evaluation of estrogen receptor alpha putative ligands

The ability of some xenobiotics, characterized by endocrine disrupting activity (endocrine active substance, EAS), to bind to the ligand binding domain (LBD) of the nuclear sex hormones, included the alpha estrogen receptor (ERα), is one of the considered endpoints of the H2020 Euromix project. In the Euromix framework, in silico predictions are used to classify a wide range of chemicals into cumulative assessment groups (CAG). To this goal, an integrated pipeline of in silico methods was developed, combining different (Q)SAR models with molecular docking and low-mode molecular dynamics (LM) simulations. A consensus of scores, derived from the combination of all the impelmented models, can be considered an useful prioritization tool. To develop and calibrate our pipeline, we selected 52 training compounds, experimentally associated to relative binding affinity (RBA) and relative activity (RA) with respect to 17β-estradiol. Since different (Q)SAR models specific to ERα were used, a consensus score was set to synthesized the results. Molecular docking allowed us to calculate the binding free energy (ΔG) value of each ligand with respect to ERα, on the basis of three-dimensional structures of the target protein and of the investigated xenobiotic. Cooper statistics, computed for different ΔG cut-off values and plotted as ROC curves, allowed us to identify the maximum accuracy set-up for prioritization purposes. Combining the consensus results from (Q)SAR with the chemical prioritization from molecular docking, it was possible to define a global consensus score, useful for merging the two approaches. Moreover, 10 compounds predicted as good ligands were analyzed more in depth, using LM for an estimate of their intrinsic activity. At an atomistic level, the conformational changes of Erα LBD helix 12 was the main determinant of the intrinsic activity of the investigated compounds: it assumes a closed conformation when bound to agonists, whereas it is in open conformation when bound to antagonists. Our results show that the global consensus score [(Q)SAR and molecular docking] is an efficient way to prioritize a compound as an ERα binder, even if this does not provide information on intrinsic activity. The integration of low-mode molecular dynamics simulations eventually makes possible to thoroughly predict, for a restricted subset of compounds, their intrinsic activity, to increase the informativeness of our in silico pipeline, with the final CAG classification aim.