Development of a 3D-QSAR model from interaction energy maps-derived descriptors for the prediction of antifungal quinolizidines

Quinolizidines are secondary metabolites mainly occurred in Fabaceae species (Wink, 2013). This kind of alkaloids are interesting due to its demonstrated activity in different biological assays against phytopathogens and insect pests (Zamora-Natera et al., 2008; Torres & Wink, 2009). In the present work, 21 quinolizidine alkaloids were isolated by chromatographic methods from several specimens of the genus Lupinus, Genista and Ulex and characterized by spectroscopic techniques. Subsequently, each isolated compound was evaluated in vitro against the phytopathogen Fusarium oxysporum at 0.01-10 mg/mL. The half-maximal inhibitory concentration (IC50) was calculated through non-linear regressions using Graph Pad Prism 7.0. The obtained results exhibited significant IC50 values (1-100 µg/mL) in comparison to those of commercial standards. This information led to propose an initial relationship between the structure and activity, specifically for tetracyclic compounds derived from lupanine and tricyclic alkaloids derived from cytisine possessing 2-pyridone moiety. In order to reasonably explain such a trend, a three-dimensional quantitative structure-activity relationship (3D-QSAR) model was built using chemoinformatic tools. Firstly, a chemical space was established using the following chemical data input: 76 compounds with reported IC50 values on F. oxysporum (retrieved from ChEMBL database), 2 commercial fungicides (e.i., dithane and rovral), and the above-mentioned 21 isolated alkaloids. The Volsurf+ descriptors of volume and surface of each compound were calculated using the VS+ modeler, and the resulting data were correlated with the bioactivity data (IC50), which were processed and analyzed using WEKA 3.8.2. The chemical descriptors of the retrieved 76 molecules were selected as training set. Thus, the 3D-QSAR model was then built by a linear regression and the model was consequently reevaluated involving the test set (i.e., 21 isolated quinolizidine alkaloids). The resulting model exhibited good statistical quality parameters (R = 0.7026, MAE = 36.9, RMSE = 47.2). The most-influencing descriptors on the resulting model were found to be related to dipole and hydrophobic integy moments. Finally, the achieved regression equation from the externally-validated model was then used to predict the IC50 values of 100 unassessed quinolizidine alkaloids against Fusarium oxysporum. The calculated results from this model showed that cytisine-derived and 2-pyridone-derived alkaloids exhibited the highest antifungal activity (predicted IC50 = 1-50 µg/mL). In conclusion, 3D-QSAR resulted into a valuable predictive tool for addressing efficiently the research on key quinolizidine moieties among natural products or synthetic molecules with potential antifungal activity against F. oxysporum.