Discovery and Optimization of Natural Product-Derived GLS1 Inhibitors via Quantum Chemical Analysis
Mio YOKOYAMA *1, Yusuke TATEISHI1, Manabu SUGIMOTO2
1Graduate School of Science and Technology, Kumamoto University
2Faculty of Advanced Science and Technology, Kumamoto University
Introduction Accumulation of senescent cells associated with aging increases the risk of diseases such as cancer and diabetes. Recent studies have shown that glutaminase 1 (GLS1) is involved in the maintenance of senescent cells. [1] Therefore, the development of GLS1-targeted inhibitors is considered important for prevention and treatment of age-related diseases. In this study, we aim to discover new GLS1 inhibitors using in silico screening through machine learning modelling using the Electronic-Structure Informatics (ESI) descriptor set our laboratory has been suggesting[2]. This descriptor set consists of descriptors reflecting quantum chemical characteristics of molecules , i.e. the electronic properties of molecules. This descriptor set does not refer to structural features of molecules so that identification of structurally novel inhibitors, including "scaffold hopping", is highly expected.
Methods A list of known GLS1 inhibitors was generated from the ChEMBL database. The list contained 260 molecules. For these compounds, quantum chemical calculations were performed to calculate 21 ESI descriptors. Using this dataset, a machine learning model was developed to predict inhibitory activity. The best-performing model was then applied to 2,647 natural products registered in the KampoDB [3] database. This procedure corresponds to "drug repositioning", which involves utilizing existing pharmaceutical ingredients for new indications. In addition, we attempted to develop a design method called "natural products remodeling" which modifies natural compounds using active fragments of GLS1 inhibitors.
Results and Discussion In our modelling procedure, the ExtraTreesRegressor model was found to be the best model for prediction with an R2 value of 0.777 for the test set. Through virtual screening and docking simulations, multiple natural compounds with high GLS1 inhibitory activity and strong binding affinity were identified. Furthermore, by analyzing known GLS1 inhibitors, we explored the partial structural motifs contributing to inhibitory activity. Based on these results we designed structures more favorable as GLS1 inhibitors by modifying the structures of natural compounds.
Conclusion We conducted a search for GLS1 inhibitor candidates by combining ESI descriptors and machine learning. The integration of in silico screening and fragment-based design led us to promising therapeutic agents targeting senescent cells with different molecular scaffold. This demonstrates a potential of combining quantum-chemistry-based molecular descriptors and natural product databases as a new type of approach in drug discovery.
[1] Y. Johmura, et al., Science, 371, 265-270 (2021).
[2] M. Sugimoto et al., Chem. Lett., 50, 849-852 (2021).
[3] R. Sawada et al., Sci. Rep., 8, 11216 (2018).