A High-Content Pseudotime Framework for Morphological Profiling of Dengue Virus Inhibitors from a Drug-Repurposing Library

Abstract

Dengue virus (DENV) causes dengue fever, a globally prevalent, epidemic-prone disease with no FDA-approved antivirals. Additionally, the available vaccine for DENV can increase the risk of severe dengue fever for those who have never had a DENV infection due to antibody-dependent enhancement. Thus, there is an urgent need to identify DENV antivirals. Drug repurposing, high-content screening, and machine learning were leveraged to efficiently identify antivirals likely to inhibit DENV replication. Using high-content screening in a DENV infection model, we characterized the morphological patterns of NS4B and envelope (E) protein expression over time and developed a viral pseudotime model that predicts infection progression to aid in phenotypic interpretation of drug effects. We then developed a single cell infection classifier for antiviral efficacy and performed high-throughput drug screening of 960 compounds. We identified four concentration-dependent inhibitors of DENV with nanomolar-to-submicromolar potencies: esomeprazole, GW4064, pralatrexate, and LY411575. With our pseudotime machine learning model trained on data from the time-series assay, we scored the concentration-dependent inhibitors to map their mechanism of action onto time points in the viral life cycle. Our most notable results show that LY411575, a gamma secretase inhibitor, exhibited an IC of 72 nM and reduced percent infection to near the level of the mock infection control. Pseudotime inferences generated mechanistic predictions that were supported by time-of-addition experiments, indicating that LY411575 arrests DENV infection before the development of late-stage infection associated morphologies, while retaining activity when added after inoculation.