Acute cell stress screen with supervised machine learning predicts cytotoxicity of excipients

Abstract

Excipients serve as vehicles, preservatives, solubilizers, and colorants for drugs, food, and cosmetics. They are considered to be inert at biological targets; however, several reports suggest that some could interact with human targets and cause unwanted effects. We investigated 40 commonly used drug excipients for cellular stress in the AsedaSciences® SYSTEMETRIC® Cell Health Screen, which was developed to estimate toxicity risk of small molecular entities (SMEs). The screen uses supervised machine learning (ML) to classify test compound cell stress phenotypes relative to a training set of on-market and withdrawn drugs. While 80% (n = 32) of the excipients did not show elevated risk in a broad, but pharmacologically relevant, concentration range (5 nM to 100 μM), we identified 20% (n = 8) with elevated risk. This group included two mercury containing preservatives, propyl gallate, methylene blue, benzethonium chloride, and cetylpyridinium chloride, all known for previously reported safety issues. All compounds were tested in parallel in an in vitro assay panel regularly used to investigate off-target effects of drug candidates. Target engagement in this assay panel confirmed risk-indicative biological activity for the same excipients, except propyl gallate, which may have a separate, interesting mechanism. We conclude that the SYSTEMETRIC Cell Health Screen, in conjunction with in vitro pharmacological profiling, can provide a fast and cost effective methodology for first line testing of SMEs, including excipients, to avoid cellular damage, particularly in the GI, where they are represented in high concentrations.

Keywords: Cell-based; Cytotoxicity; Excipients; High-content; In vitro profiling; Machine learning; Methods; Off-target; Phenotypic; Preclinical.