Abstract:

The GABA_A receptor mainly mediates inhibitory signal transmission in mammalian central nervous systems and is the key target of sedative-hypnotics. However, the long-term use of sedative-hypnotics often leads to drug resistance, necessitating the development of novel sedative-hypnotics. This development can be achieved with novel scaffolds designed via the computer-aided drug design methods to obtain significant advantages. In this study, robust virtual screening models were established by identifying effective positive allosteric modulators of the GABA_A receptor from ChEMBL and BindingDB databases. These compounds combined with randomly extracted negative compounds were firstly applied for a 10-fold cross validation and grid search to establish machine learning models which were subsequently evaluated in an independent test set. In this step, 4 machine learning methods and 6 fingerprints were used to establish 24 models. In the test set, the CDK_LR model performed the best (MCC = 0.751) and was used for subsequent virtual screening. Two effective molecular docking models were also established based on conformation 6D6T and 6D6U, wherein the root mean square deviation (RMSD) values of redocking experiments were 1.141 and 1.505 Å (1 Å = 0.1 nm), respectively. During the virtual screening, 41112 compounds from a commercial database were scanned by machine learning, molecular docking, and molecular mechanics-generalized Born surface area models. After the screening, 16 hits were obtained, 4 of which were structurally novel positive hits verified by whole-cell patch-clamp electrophysiology experiments. The compound GPR120 was verified experimentally at both the cell and animal levels. In cortical neurons recombinantly expressing α1β2γ2-type receptors, at 10 and 50 µmol∙L⁻¹, GPR120 could potentiate GABA EC_3-10 current by 71.5% and 163.8%, respectively. Total decomposition contribution analysis and point mutation experiment showed that the key binding site between GPR120 and the GABA_A receptor is H102, similar to that of the positive drug Diazepam. To further verify GPR120 function at the animal level, locomotor activity and loss of righting reflex (LORR) tests were performed. GPR120 inhibited the locomotor activity of mice, which recovered after 6 h, indicating that GPR120 is a moderate sedative. In the pentobarbital sodium-induced righting reflex hour test, GPR120 (20 mg∙kg⁻¹) significantly shortened the time to start LORR and prolonged its duration compared with the saline control group. In summary, using integrated virtual screening methods, GPR120 was identified as a moderate sedative with a novel scaffold.

Key words: Machine learning, Ion channel, Molecular docking, Virtual screening, GABA_A receptor