Discovery of Anticancer Hybrid Molecules by Supervised Machine Learning Models and in Vitro Validation in Drug Resistant Chronic Myeloid Leukemia Cells.

Abstract

The concept of hybrid drugs for targeting multiple aberrant pathways of cancer, by combining the key pharmacophores of clinically approved single-targeted drugs, has emerged as a promising approach for overcoming drug-resistance. Here, we report the design of unique hybrid molecules by combining the two pharmacophores of clinically approved BCR-ABL inhibitor (ponatinib) and HDAC inhibitor (vorinostat) and results of in vitro studies in drug-resistant CML cells. Robust 2D-QSAR and 3D-pharmacophore machine learning supervised models were developed for virtual screening of the hybrid molecules based on their predicted BCR-ABL and HDAC inhibitory activity. The developed 2D-QSAR model showed five information rich molecular descriptors while the 3D-pharmacophore model of BCR-ABL showed five different chemical features (hydrogen bond acceptor, donor, hydrophobic group, positive ion group, and aromatic rings) and the HDAC model showed four different chemical features (hydrogen bond acceptor, donor, positive ion group, and aromatic rings) for potent BCR-ABL and HDAC inhibition. Virtual screening of the 16 designed hybrid molecules identified FP7 and FP10 with better potential of inhibitory activity. FP7 was the most effective molecule with predicted IC50 using the BCR-ABL based 2D-QSAR model of 0.005 μM and that of the HDAC model of 0.153 μM, and that using the BCR-ABL based 3D-pharmacophore model was 0.02 μM and that with HDAC model was 0.014 μM. In vitro study (dose-response relationship) of FP7 in wild type and imatinib-resistant CML cell lines harboring Thr315Ile or Tyr253His mutations showed growth inhibitory IC50 values of 0.000 16, 0.0039, and 0.01 μM, respectively. This molecule also showed better biocompatibility when tested in whole blood and in PBMCs as compared to ponatinib or vorinostat.