A machine learning-based screen of marine natural products to identify new leads for the treatment of Acanthamoeba eye infection

点击翻译按钮获取中文摘要

PROJECT SUMMARY Painful blinding keratitis is caused by the free-living amoeba Acanthamoeba and can occur in healthy individuals wearing contact lenses. Acanthamoeba can exist as a trophozoite or cyst and both stages are able to cause Acanthamoeba keratitis. While effective therapies, such as chlorhexidine gluconate and polyhexamethylene biguanide, exist to treat Acanthamoeba keratitis, the trophozoites can encyst in the ocular tissue to resist current therapies. Infection recurrence occurs in approximately 10% of cases due to the lack of efficient drugs that can kill both trophozoites and cysts. Therefore, discovery of therapeutics that are effective against both stages is a critical unmet need to avert blindness. The urgency of the issue is underscored by the NIAID's listing of acanthamoebiasis as an Emerging Infectious Disease. Current efforts to identify new anti-Acanthamoeba compounds rely primarily upon trophocidal assays that target the trophozoite stage of the parasite. Standard cysticidal assays are laborious and depend on manual observation of compound-treated cysts. Considering the manual and low-throughput approaches used in the cysticidal assays, we hypothesized that any development in automation and miniaturization could significantly increase the throughput of cysticidal drug screens to yield new cysticidal compounds. We adapted and trained a YOLOv3 machine learning object-detection neural network to recognize A. castellanii trophozoites and cysts in microscopy images. We utilized this trained neural network as a tool to count excysted trophozoites in compound-treated wells to determine if a compound was cysticidal. We validated this novel screen with literature-relevant cysticidal and non-cysticidal reference compounds by determining their minimum cysticidal concentrations. Our machine learning-based cysticidal assay improved throughput, demonstrated high specificity and an exquisite ability to identify non-cysticidal compounds. We combined this cysticidal assay with our bioluminescence-based trophocidal assay to screen about 9,000 structurally unique marine microbial metabolites against A. castellanii. Our preliminary screen identified a marine microbial metabolite that was both trophocidal and cysticidal. Based on these data, we propose to utilize our machine learning-based high-throughput cysticidal assay to 1) screen >20,000 marine microbial natural products against trophozoites and cysts of a reference strain, and isolate, dereplicate and assign the structures of the active compounds, 2) evaluate susceptibility of trophozoites and cysts of a reference strain, and relevant mammalian cells to purified compounds, and 3) confirm trophocidal and cysticidal activities of less toxic purified compounds against multiple genotypes of Acanthamoeba. The goal of this work will be to identify 1-3 molecules that are potent inhibitors of A. castellanii trophozoites and cysts. To successfully achieve the aims, we rely on our collaboration that combines the unique expertise of Dr. Debnath (PI) in Acanthamoeba parasite biology and Dr. Fenical in marine natural products drug discovery (Co-Investigator). Drs. Debnath and Fenical's expertise and experience has potential to elevate our drug discovery platform to a translational level.