Generalized additive modeling as a tool for the analysis of the time course of tail coiling behavior in zebrafish (Danio rerio) embryos – A proof-of-concept study with nicotine, a known developmental neurotoxicant

The early detectable tail coiling behavior of zebrafish (Danio rerio) embryos is receiving increasing attention in the context of (developmental) neurotoxicity testing and may be used as a rapid screening tool for compounds with unknown or suspected neurotoxic potential. The observation of this behavior over a longer period of time already offered advantages such as the possibility of detecting effects that only occur after a few hours of development. The two major parameters, duration and frequency of coiling, allow a detailed characterization of the movements. However, this approach usually leads to complex data sets, which are often heavily simplified to allow for simpler analysis of the effects on an hourly basis. In this study, the suitability of generalized additive modeling (GAM) for the analysis of coiling behavior was tested in order to obtain an integrated impression of the trends in movement patterns. To this end, nicotine, a known potent developmental neurotoxicant, was used in a proof-of-concept study. The main advantage of GAM for biological data lies in the relaxation of assumptions, such as effect monotony, data distribution and homogeneity of variances and is, therefore, more flexible in describing different trends over time. The possibility to consider replicates and individuals as additional sources of (biological) variance is a further benefit, as highly variable data are common in behavioral studies. Here, the modeling approach demonstrates a monotone reduction of movement duration as a direct consequence of nicotine exposure. Additional pathomorphological studies revealed structural damage in secondary motoneurons and skeletal muscles as potential underlying mechanisms of changes in movement patterns. The GAM proved well-suited to illustrate and analyze complex non-linear behavioral data with high natural variability. The model also allows to reliably extract no observed effect (NOEC) and lowest observed effect concentrations (LOEC) from complex data sets, which may be of relevance in a regulatory context.