Zebrafish-based high throughput platform for human developmental toxicity assessment

ABSTRACT

Humans are in contact with chemicals from the zygote stage, continuing the exposure during embryo and foetus development. Exposure to chemicals during embryonic development may result in drastic effects ranging from abnormal growth of anatomical structures to lifelong mental disabilities. To avoid these health risks, it has become a priority to identify teratogenic threats in early phases of pharmaceutical drug and industrial chemical development. In vitro teratogenicity studies have shown significant improvements in terms of predictivity, reaching accuracies of around 90 % for some validation studies. However, its use is still challenging due to their reductionist nature, which obviates complex biological processes present in intact organisms, potentially reducing predictivity compared to whole-organism models. On the other hand, teratogenicity in vivo studies, which have traditionally relied on mammalian models such as rodents and rabbits, suffer from elevated experimental costs and low throughput. In addition, in the era of the 3Rs principles implementation there is a strong interesting in the seek for new experimental methods to reduce animal testing. Thus, the establishment of rapid, reliable and cost-effective methodologies for detecting developmental toxicity of chemical substances is a pressing need for both the scientific community and regulatory agencies. In this line, the zebrafish model has emerged as an alternative to traditional preclinical models for predicting teratogenicity and other potential chemical-induced toxicity hazards due to their small size, rapid development, transparency, and developmental pathways similarities with mammalian. Here, we will present the establishment of an innovative automated high-throughput screening platform for teratogenic drugs. The platform was validated by testing on zebrafish larvae a library of 31 known compounds classified as teratogens and non-teratogens in mammals and by assessing 16 phenotypical parameters related to embryo development. Our results show high correlation with results obtained in murine models, detecting 20 teratogenic compounds and 11 non-teratogenic compounds, with a 94.44% sensitivity, 90.91% specificity and 87.10% accuracy. Moreover, the model shows a high correlation with humans: 87.50% sensitivity, 81.82% specificity and 74.19% accuracy, increasing the prediction level reported in rodents. Importantly, based on the results obtained in this validation study, we have developed a deep learning algorithm, able to discriminate all the defined phenotypic parameters, sort larvae as positive or negative for qualitative phenotypes and extract values from quantitative ones. 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.

Zebrafish-based high throughput platform for human developmental toxicity assessment

DOWNLOAD POSTER