Increasing predictivity of compound multi-organ toxicity through high-throughput zebrafish assays

Sergio Jarque 1, Carles Cornet 1, Valentina Schiavone 1, Maria Rubio-Brotons 1, Víctor Ordoñez 1, Elías Abad2, Alex Ferrando2, Inés Broto2, Alex Martí2, Marc Fuentes2, Andrea García2, Josep Borrell2, Mireia Cavallé2, Andreu Hughet2, Jordi-Guillem Rodríguez2, Adrián Rodríguez2, Jessica García-Fernández 1, Ferran Arqué 1, Rafael Miñana-Prieto 1, Vincenzo Di Donato 1, Sylvia Dyballa 1, Javier Terriente 1 1. ZeClinics - IGTP research centre, Barcelona (SPAIN) 2.Universitat Politècnica de Catalunya · Barcelona Tech – UPC, Barcelona (SPAIN)


Current preclinical animal experiments provide lower-than-expected prediction of toxic liabilities and therapeutic effects in human patients. 

Importantly, international organizations and agencies, among them U.S. Environmental Protection Agency, have also declared their strong intention to reduce and eventually completely eliminate animal testing during the next decade.  In this landscape, new approach methodologies (NAMs) are the focus of innovative research efforts in drug discovery and regulatory hazard determination.

Among NAMs, the zebrafish is one of the most promising models since it ensures at the same time high-throughput capabilities and biological relevance of a whole organism. At larval stages it is considered as a non-animal method, which constitutes zebrafish a relevant gamer for in vitro integrated approaches and 3Rs principles application. 

Many years ago, the National Institutes of Health already ranked the zebrafish as the second most important animal model behind the mouse. On current days, several fish toxicity tests are a standard information requirement under REACH (Registration, Evaluation, Authorisation, and Restriction of Chemicals) regulation: among them the AFT (acute fish toxicity) is one of the most spreadly used.

Here we present, among others, an example of a teratogenic assessment assay using a deep learning algorithm, able to classify phenotypes and segment regions with good scores. Overall, we demonstrate that combining the experimental advantages of the zebrafish larval model with artificial intelligence allows for high-throughput, fully automated detection of compound teratogenicity, thus paving the ground for a faster and reliable human risk assessment based on NAMs.

Together, this data demonstrates that zebrafish can complement if not replace mammalian testing, while being equally protective of human health. Furthermore, it reinforces the evidence for the zebrafish to be present in the pharmaceutical preclinical regulatory phases in addition to regulated safety studies in the chemical industry.

Increasing predictivity of compound multi-organ toxicity through high-throughput zebrafish assays