Lysine-Specific Demethylase (LSD1) is a FAD-dependent nuclear protein that is overexpressed in several types of tumors. Pharmacological LSD1 inhibition was shown to inhibit tumor growth, proliferation and metastasis. The goal of this study was to build and validate predictive 2D and 3D QSAR models of allosteric LSD1 inhibitors to aid in the virtual screening (VS) and optimization of new anticancer agents. 93 1,2,3-triazole-dithiocarbamate derivatives, synthesized and tested as LSD1 inhibitors by Zheng et al., were clustered into training and test sets for QSAR studies according to calculated physical-chemical properties in the SYBYL-X 2.1.1 platform. IC50 values of dataset ranged from 0.39 μm to &gt;125 μm. Principal Component Analysis (PCA) and Hierarchical Clustering Analysis (HCA) were employed to assess the information of variables and to remove those that provided poor information. Compounds were then clustered according to six remaining properties (number of rotatable bonds, number of H-Bond acceptors and donors, cLogP, molecular volume and PSA). Hologram-QSAR (HQSAR), CoMFA and CoMSIA models were generated. External validation employed a test set of 19 compounds. q2 for the most robust HQSAR, CoMFA and CoMSIA models were 0.690, 0.732 and 0.692, respectively. Docking studies were performed in GOLD Dock using the crystal structure of LSD1-FAD complex (PDB: 2DW4). The 7 most potent inhibitors (IC50 &lt; 10 μm) were docked to the receptor and the poses with highest GoldScore were used as templates to build chemical similarity models in vROCS 3.2.0.4. by varying molecular shape (template molecule) and chemical features. For validation, 50 decoys were generated for each one of the 49 active molecules (IC50 &lt; 100 μm) employing DUD-E. Compounds with IC50 &gt; 100 μm were grouped as decoys, in a total of 2499 decoys. 30 conformers were generated for each active and inactive molecule employing MMFF94 force field using OMEGA 2.5.1.4. The ROC curve was calculated for all models and values of area under the curve (AUC) and score histograms were compared to define a score cut-off of true positives. Chemical features of the models with highest AUC are in agreement with the positive contributions revealed by QSAR studies. The 3 models with the lowest false-positive and false-discovery rates were employed as filters for VS. From 12 consensus hits above the cut-off, 3 were selected after excluding PAIN-like molecules and will be biologically evaluated as LSD1 inhibitors.

Search of new anti-cancer compounds through combined QSAR and virtual screening of allosteric LSD1 inhibitors

253rd National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.