Development of COVID-19 cell entry inhibitors

Division Of Basic Sciences - Nci

Investigators

Linked publications & trials

Abstract

The reliability, and consequently success of virtual screening depends heavily on the accuracy of the x-ray structure depiction of the target protein structure in solution. Crystal packing can distort conformation of flexible parts of the protein thus leading to misrepresentation of the protein structure. Thus, the first step in all screening procedures is optimization and evaluation of the structure used in the screen. Structural biology of COVID-19 proteins has been developing with an explosive speed since the start of 2020 with dozens of structures being deposited to Protein Data Bank every week. We have evaluated suitability of structures of SARS-CoV-2 NSP7, NSP8, NSP9, NSP10, NSP14, NSP15, NSP16, nucleocapsid RNA-binding domain and spike protein receptor-binding domain for virtual screening. All structures have undergone energy minimization procedures before the screens. Initially, they have been screen-tested using a diversity library of 1.92 million compounds that were available for 2-week delivery from Enamine LTD and a library of all drugs marketed in US. The screens were run on NIH High Performing Computing system Biowulf using ICM-Pro docking software from MolSoft and optimized docking protocols developed by us that allow for highly reliable predictions of the binders. Typically, we have been running 2000 parallel docking jobs that allowed screening about 10 million compounds a day for a single binding site in a fast screening mode. In total, we have evaluated 30 x-ray SARS-CoV-2 protein structures in test screens. 4-5 hits from the test libraries have been purchased for each target structure and tested for binding using recombinant COVID-19 proteins and nano differential scanning fluorometry (DSF). Structures of NSP7, NSP8 and NSP14 did not produce hits with binding scores below the threshold. For all other proteins (NSP9, NSP10, NSP15, NSP16, nucleocapsid RNA-binding domain and spike protein receptor-binding domain) we were able to select x-ray structures allowing for accurate prediction of binding, 10 structures in total. Significant efforts have been spent on generation of highly diverse virtual libraries of compounds for the screens. Novel libraries, Synthetically Accessible Virtual Inventory (SAVI) contain compounds that can be synthesized in one easy step from readily available blocks. Current database that was made public in April 2020 (https://cactus.nci.nih.gov/download/savi_download/) was built using 53 robust synthetic procedures or transforms and contains 1,748,464,003 products. Screening databases of that size currently is impractical. That is why we have generated SAVI diversity sets of 2,955,416 and 15 million compounds that were used for identification of initial leads. The leads identified from screening SAVI and Enamine diversity sets have been used for preparation of libraries of structurally similar compounds typically containing hundreds of thousands or 2-3 million compounds from entire SAVI library. We have used machine learning tools (MNA, Multilevel Neighborhoods of Atoms and QNA, Quantitative Neighborhoods of Atoms) generated by our collaborator Prof. Vladimir Poroikov and his colleagues from the Institute of Biomedical Chemistry, Moscow, Russia for generation of libraries of compounds with electronic structures similar to the leads. These libraries were subjected to screening by docking allowing for identification of compounds with improved docking scores. The later have been synthesized, and binding to the target proteins evaluated by NanoDifferential Scanning Fluorometry, fluorescence spectroscopy and Microscale Thermophoresis. Overall, over 200 compounds have been tested for binding. Currently, we have identified compounds that bind to NSP9, NSP10, NSP16, nucleocapsid RBD and spike protein RBD with micromolar affinity. Lead compounds bind to exonuclease NSP15 with sub micromolar affinity. Major focus currently is on the inhibitors of the exonuclease not only because it is essential for viral replication but also because the active site that we are targeting is conserved in all corona viruses. Consequently, newly developed inhibitors are likely to help fighting not only current, but also future pandemics. The leads for other proteins will be developed at later time as resources will permit.