Ligand- and structural-based discovery of potential small molecules that target the colchicine site of tubulin for cancer treatment

Abstract

Small molecules targeting the colchicine site of tubulin represent an attractive cancer treatment strategy. In this study, a total of 468 models derived from 1076 diverse inhibitors binding to the tubulin colchicine site were constructed based on fingerprints using three machine learning approaches: 1) naive Bayesian (NB); 2) single tree (ST); and 3) random forest (RF). The overall predictive accuracy of the best models exceeded 85.12% for both the training and test sets. We designed an integrated virtual screening (VS) strategy for identifying new tubulin inhibitors by combining established models, molecular docking, and similarity-based analog searching. Through two rounds of VS, compound 23g was identified as a novel potent anticancer agent exhibiting activity against MDA-MB-231, HeLa, A549, HepG2, CNE2, and HCT116 tumor cell lines with IC₅₀ values of 5.45, 8.61, 7.47, 2.29, 2.91, and 4.10 μM, respectively. Compared with taxol, colchicine, and adriamycin, 23g also displayed potent cytotoxicity against the drug-resistant tumor cell lines, HepG2/ADR, A549/CDDP, and A549/TAX cells, with IC₅₀ values of 4.12, 6.58, and 6.38 μM, respectively. Further mechanistic studies revealed that 23g inhibited microtubule polymerization by binding to the colchicine site of tubulin, arrested the cell cycle at the G2/M phase, induced cell apoptosis, and exhibited potent in vitro anti-metastasis activity. Finally, molecular docking, molecular dynamics, and free energy analyses were employed to explore the detailed binding interaction between 23g and tubulin. Collectively, these findings indicated that 23g should be further investigated as a potential novel potent antitumor agent targeting tubulin.

Keywords: 23g; Machine learning; Microtubule polymerization; Tubulin; Virtual screening.