New tylophorine analogs as potential antitumor agents

Tylophorine and related phenanthroindolizidine alkaloids isolated principally from Asclepiadaceae have been targets of synthetic modification because of their profound cytotoxic antitumor activity. As part of our interest in plant-derived antitumor agents, novel water-soluble phenanthrene-based tylophorine derivatives (PBTs) were designed, synthesized and evaluated for anticancer activity.; Several PBTs showed superior activity profiles with EC50 values in the sub-micromolar range, which are comparable to those of currently used antitumor drugs. A structure-activity relationship (SAR) study was also explored to facilitate the further development of this new compound class.; Subsequently, C9-substituted PBTs were designed and synthesized using 2, 3 methylenedioxy-6-methoxyphenanthrene as a common skeleton based on our prior SAR findings. The C-9 site is an ideal position for introducing more polar, water-solubility-enhancing moieties. We also extended the in vitro antitumor screening to include additional significant tumor types [A549 (lung), DU-145 (prostate), ZR-751 (breast), KB (nasopharyngeal)] as well as a multi-drug resistant cancer cell subline [KB-Vin (multi-drug resistant KB subline)]. Most of the compounds showed fairly uniform and potent cytotoxic activity with EC50≅10-7 M against both wild type and matched multi-drug resistant KB cell lines, and displayed notable selectivity toward DU-145 (prostate) and ZR-751 (breast) cancer cell lines.; A combination of QSAR modeling and database mining was used to facilitate further design and discovery of novel anticancer PBTs. MolConnZ 2D topological descriptors were applied to a dataset of 52 chemically diverse PBTs and variable selection models were generated using the kappa nearest neighbor (kappaNN) method. The derived kappaNN QSAR models have high internal accuracy, with leave-one-out cross-validated R2 (q2) values ranging between 0.6 and 0.8. The original dataset was then divided into several training and test sets to provide highly predictive models with q2 values greater than 0.5 for the training sets and R2 values greater than 0.6 for the test sets. The ten best models were capable of mining the commercially available ChemDiv Database (450,000 compounds) and resulted in 34 consensus hits. Of these 34 compounds, 10 compounds were tested and 8 were confirmed to be active with a best EC50 of 1.8 muM. These models were further validated by predicting the activity of four new PBTs compounds with reasonable accuracy and 11 consensuses hits with R 2 of 0.52. These results indicate that this approach can be successfully applied to further design and discovery of anticancer drug candidates from this compound class.