Statistical analysis of multi-endpoint phenotypic screening increases sensitivity of planarian neurotoxicity testing.

There is an urgent need for high-throughput screening (HTS) models to replace, refine, and/or reduce ("3Rs") vertebrate toxicity testing. Replacing in vivo animal studies is challenging for neurotoxicity and developmental neurotoxicity (DNT), where the functional relevancy of adverse outcomes needs to be assessed on the whole organism. We previously screened the NTP 87-compound library (NTP87), consisting of known and suspected developmental neurotoxicants, and showed that planarian HTS can identify known (developmental) neurotoxicants. Because analysis methods can impact screening results, and our original analysis used the lowest-observed-effect-level (LOEL) only, we hypothesized that the use of state-of-the-art statistical analysis would increase the sensitivity of planarian HTS to identify neurotoxicity and DNT. Using the original NTP87 planarian data, we quantified 8 additional behavioral endpoints for a total of 26 readouts on days 7 and 12 of exposure, evaluated at 5 log-scale concentrations (10 nM-100 µM). Benchmark concentration (BMC) modeling replaced LOEL analysis. We also calculated a concentration-independent multi-readout summary measure using weighted Aggregate Entropy, providing insight into systems-level toxicity. Finally, we compared the planarian BMC data to in vitro and developing zebrafish data from independent screens of the NTP87 library that were analyzed using the same BMC pipeline. Planarian and developing zebrafish screens showed similar sensitivity. Regenerating planarian hits helped correctly identify known neurotoxicants of the NTP87 library. Hierarchical clustering showed that organismal, neuron outgrowth, and neuron firing models were the main contributors to the NTP87 DNT battery's information content, emphasizing their relevance for DNT testing.