The Evolution of Predictive Toxicology: Improving Predictivity Using New-Approach Methodologies

A pharmaceutical’s approval for commercial distribution is contingent on submission of pharmacological and toxicological safety data as defined by regulatory agencies such as the US FDA and ICH. Guidances state that such data can come from in vivo or in vitro studies. The current paradigm works well for collecting critical data about, e.g., a drug’s pharmacological effects and mechanism of action (MoA). However, increasing evidence points to current methods’ inadequacy for predicting a drug’s risks to human patients. Such limitations can increase risks for drug-program attrition.

The push to reduce animal use during toxicity studies has spurred on advances with in vitro technologies, including iPSCs, 3D tissue cultures (organoids), microphysiological systems (organ-/tissue-on-a-chip systems), in silico predictive tools, and quantitative structural activity relationship (QSAR) models. Those are considered to be new-approach methodologies (NAMs). Regulatory agencies often evaluate in vitro NAMs for predictive success. In vivo techniques also can be classified as NAMs if they could improve predictive outcomes and/or help scientists to replace, reduce, and refine animal studies. A shifting regulatory landscape has permitted flexibility in testing paradigms but also has raised obstacles to validating toxicological data collected by NAMs.

Emerging Applications
Studies of drug-induced toxicity must range across several categories. Hepatoxicity, nephrotoxicity, cardiotoxicity, genotoxicity, neurotoxicity, mitochondrial toxicity, and potential for phospholipidosis (PLD) all are serious problems — and leading causes of drug-program termination. Program failure during late-stage clinical trials is costly, especially when they involve animal studies.

Assays based on human cells provide a cost-effective way to predict drug toxicity while decreasing analytical costs and overcoming species differences — a factor that contributes to 40–50% of predictive error during animal-based assessment. In 2017, Tomida et al. demonstrated that drug-induced PLD, a characteristic of many drugs with a cationic amphiphilic structure, can be detected in spheroid cultures of human hepatocytes (1). The system was highly sensitive and showed much potential for predicting drug-induced PLD.

Repositioning established drugs has become an attractive approach to drug discovery and development. The method enables drug makers to decrease costs and developmental risks because safety profiles and pharmacokinetic data already have been established for the products at hand. Using an FDA-approved drug library, Voss et al. identified adefovir dipivoxil (ADV) for potential repositioning to autoimmune indications (2). NAMs revealed ADV to be a potent inhibitor not only of T-cell proliferation and activation, but also of Th1, Th2, and Th17 cytokine production. The drug’s inhibition of T-cell proliferation was observed in both preactivated and freshly stimulated T cells.

Another NAM involves recognition of phenotypic similarities across chemicals to predict toxicity mechanisms. A compound database with morphologically relevant data could help to establish correlations between mechanistic and phenotypic similarity.

NAMs show promise for large molecules, as well. Microphysiological systems have obtained translationally relevant insights into the efficacy of AAV vectors in clinical-like conditions. In 2021, Achberger et al. used retina-on-a-chip systems to study AAVs. They found that dose-dependent cell tropism observed in rat models does not translate to human retinal cells (3). In silico predictive tools helped Yang et al. to design a multiepitope vaccine against SARS-CoV-2. The candidate has good structural and physicochemical properties that could withstand known and potential mutations in the virus’s RNA (4).

Continuing Needs
Although in vitro toxicology methods have made great strides in estimating effects of chronic drug exposure and determining safe dosing for clinical trials, several needs must be addressed. Current shortcomings include inadequate tools for predicting rare drug-induced liver injuries and proarrhythmia risks. Analysts also must develop approaches that are more in line with human biology to increase understanding of drug MoA and exposure. As barriers in microphysiological systems are overcome, and as QSAR modeling is enhanced to predict risk profiles and off-target responses more effectively, drug discovery costs can be minimized while still maintaining regulatory compliance.

References
1 Tomida T, Ishimura M, Iwaki M. A Cell-Based Assay Using HepaRG Cells for Predicting Drug-Induced Phospholipidosis. J. Toxicol. Sci. 42(5) 2017: 641–650.

2 Voss L, et al. Screening of FDA-Approved Drug Library Identifies Adefovir Dipivoxil As Highly Potent Inhibitor of T Cell Proliferation. Front. Immunol. 11, 2021: 616570.

3 Achberger K, et al. Human Stem Cell–Based Retina on Chip As New Translational Model for Validation of AAV Retinal Gene Therapy Vectors. Stem Cell Rep. 16(9) 2021: 2242–2256.

4 Yang Z, et al. An In Silico Deep Learning Approach to Multi-Epitope Vaccine Design: A SARS-CoV-2 Case Study. Sci. Rep. 11(1) 2021: 1–21.

Tanya Victor, PhD, MBA, is an associate product manager at Enzo Life Sciences, 10 Executive Boulevard, Farmingdale, NY 11735; [email protected]; https://www.enzolifesciences.com.