Stage-dependent evaluation of cytotoxicity, biomarker expression and self-organisation in iPSC-derived neuroectoderm allows accurate and interpretable assessment of developmental toxicity

Abstract

Today, evaluation of developmental toxicity largely depends on animal testing, despite ethical considerations, high costs, questionable human relevance and limited accuracy, raising an urgent need for alternatives. Here, induced pluripotent stem cell-based methods are considered a promising option. Especially early-stage directed neuroectodermal differentiation has enabled highly accurate classification of developmental toxicants based on whole transcriptome profiling in previous studies. However, high costs and limited consideration of later developmental stages constrain applicability. In this thesis, both limitations were addressed by first developing a targeted RNA sequencing approach based on the selection of 190 transcriptional biomarkers to provide efficient and mechanistically interpretable assessment of disturbed early neural induction, and second, extending the differentiation model for evaluation of perturbation of self-organisation into neural rosettes. After exposure to a panel of 18 validation compounds with known developmental toxicity status, Targeted RNA Expression Measurement (TREx) revealed distinct modes of perturbed differentiation, including impaired acquisition of neuroectodermal identity and activation of divergent transcriptional programs. Integration of biomarker expression and cytotoxicity in a random forest-based classification model yielded moderately accurate prediction of developmental toxicity. Comparative cytotoxicity analysis revealed pronounced stage-dependent sensitivity patterns across the validation compounds, and evaluation of neural rosette formation during the late-stage model unravelled disruption of cellular self-organisation. Notably, it was found that some compounds, which exclusively affected biomarker expression during early neuroectodermal differentiation, did not influence self-organisation and vice versa. Integration of disrupted self-organisation and late-stage cytotoxicity in a random forest-based prediction model elevated predictive performance. Finally, it was demonstrated that developmental stage-aware modelling of early embryonic neurodevelopment based on directed neuroectodermal differentiation enabled identification of distinct modes of perturbed differentiation, providing accurate, efficient, and mechanistically interpretable prediction of developmental toxicity.