Weight of evidence tools in the prediction of acute fish toxicity

• Published in: Integrated environmental assessment and management 2023-09, Vol.19 (5), p.1220-1234
• Main Authors: Belanger, Scott E., Lillicrap, Adam D., Moe, S. Jannicke, Wolf, Raoul, Connors, Kristin, Embry, Michelle R.
• Format: Article
• Language: English
• Subjects: Acute fish toxicity; Acute toxicity; Algae; Animal alternative; Assaying; Bayesian analysis; Bayesian statistics; Cytotoxicity; Environmental hazards; Fish; Fish embryo test; Hazard assessment; Juveniles; Mode of action; Probability theory; Toxicity; Uptake
• ISSN: 1551-3777; 1551-3793
• DOI: 10.1002/ieam.4581

Acute fish toxicity (AFT) is a key endpoint in nearly all regulatory implementations of environmental hazard assessments of chemicals globally. Although it is an early tier assay, the AFT assay is complex and uses many juvenile fish each year for the registration and assessment of chemicals. Thus, it is imperative to seek animal alternative approaches to replace or reduce animal use for environmental hazard assessments. A Bayesian Network (BN) model has been developed that brings together a suite of lines of evidence (LoEs) to produce a probabilistic estimate of AFT without the testing of additional juvenile fish. Lines of evidence include chemical descriptors, mode of action (MoA) assignment, knowledge of algal and daphnid acute toxicity, and animal alternative assays such as fish embryo tests and in vitro fish assays (e.g., gill cytotoxicity). The effort also includes retrieval, assessment, and curation of quality acute fish toxicity data because these act as the baseline of comparison with model outputs. An ideal outcome of this effort would be to have global applicability, acceptance and uptake, relevance to predominant fish species used in chemical assessments, be expandable to allow incorporation of future knowledge, and data to be publicly available. The BN model can be conceived as having incorporated principles of tiered assessment and whose outcomes will be directed by the available evidence in combination with prior information. We demonstrate that, as additional evidence is included in the prediction of a given chemical's ecotoxicity profile, both the accuracy and the precision of the predicted AFT can increase. Ultimately an improved environmental hazard assessment will be achieved. Integr Environ Assess Manag 2023;19:1220–1234. © 2022 SETAC Key Points A Bayesian Network (BN) model has been developed to assist prediction of acute fish toxicity using animal alternative methods. An array of lines of evidence were used as BN model inputs including data from QSARS, fish embryo tests, gill cytotoxicity tests, the threshold approach, and mode of action assignments. Data‐rich and data‐poor scenarios indicate acute fish toxicity predictions improve when more lines of evidence are used. Evaluations confirmed that the use of fish embryo test data to replace acute fish toxicity data were justified in that GHS classification or toxicity interval predictions were the same.