Simulating Alpha/Beta selectivity at the human thyroid hormone receptor: Consensus scoring using multi-dimensional QSARAuthors: Angelo Vedani, Martin Zumstein, Markus A. Lill, and Beat Ernst
Department of Pharmaceutical Sciences, University of Basel, CH-4056 Basel, Switzerland.
We present a consensus-scoring study on the human thyroid hormone receptor α and β using two receptor-modeling concepts (software Quasar and Raptor) that are based on multi-dimensional QSAR and allow for the explicit simulation of induced fit. The binding mode of 82 agonists and indirect antagonists — spanning an activity range of seven orders of magnitude in Ki — was identified through flexible docking to the respective X-ray crystal structures (software Yeti) and represented by a 4D data set with up to four conformations per compound. The receptor surrogates for the thyroid receptor α converged at a cross-validated r2 of 0.846 and 0.919 (64 training compounds; for Quasar and Raptor, respectively) and yielded a predictive r2 of 0.812 and 0.814 (18 test compounds); the models for the thyroid receptor β resulted in a cross-validated r2 of 0.823 and 0.909 and a predictive r2 of 0.665 and 0.796, respectively. Consensus was achieved as, on the average, the calculated activities of the training set differ only by a factor 2.2 in Ki and those of the test set by a factor 2.8 when predicted by Quasar and Raptor, respectively.
Left: Stereo view of 3,5-dichloro-3'-isopropyl-thyronine bound to the thyroid hormone receptor α.
Right: Stereo view of the Raptor surrogate for the thyroid receptor β with the largest ligand of the training set depicted. The front section has been clipped to display inner (wireframe) and outer shells (smooth surface). Areas colored in brown represent hydrophobic properties; areas in red correspond to H-bond acceptors, areas in blue to H-bond donors and green reflects H-bond flip-flops. The arrows indicate key amino-acid residues that are perfectly mapped in the Raptor model (details are given in the paper).