New statistical and computational tools for optimization of planarian behavioral chemical screens

点击翻译按钮获取中文摘要

Objectives. There is an urgent need to develop high-throughput screening (HTS) non-animal models to replace, refine, and/or reduce (“3Rs”) vertebrate toxicology testing. The development of HTS models is especially challenging for neurotoxicity (NT) and developmental neurotoxicity (DNT) where the functional relevancy of adverse outcomes needs to be assessed on the whole organism level. The overarching goal of this research is to develop a non-animal organismal HTS methodology to identify NT and DNT. The specific objective is to determine whether using state-of-the-art computational approaches will increase sensitivity and specificity of planarian HTS to identify NT and DNT using the verified National Toxicology Program 87-compound library (NTP87). As a partner in the NTP Neurotoxicology Screening Strategies Initiative, we previously screened the NTP87 library consisting of known and suspected developmental neurotoxicants in the asexual freshwater planarian Dugesia japonica using 9 readouts. Using this library, we demonstrated that planarian HTS can identify known (developmental-) neurotoxicants and adds complementary value to screens in developing zebrafish. Planarians are invertebrates of intermediate neural and anatomical complexity compared to nematodes and zebrafish and have tractable, evolutionarily conserved neuronal circuits. Planarians uniquely allow for direct comparison of xenobiotic effects on the adult and developing nervous systems. For asexual D. japonica, which reproduce via fission, neuroregeneration is the sole method of neurodevelopment and shares conserved key events with vertebrate neurodevelopment. These features and our previous work demonstrate the value of planarian HTS for first-tier screening of potential neurotoxicants. We hypothesize that we can augment sensitivity and specificity of this non-animal model by re-analyzing our NTP87 data using state-of-the- art machine learning and statistical tools. Experimental approach. In Aim 1, we will re-analyze the raw data using 18 new behavioral and 10 new morphological readouts using computer vision and machine learning. Thus, in total we assay 37 readouts evaluated at 5 concentrations, in intact and regenerating organisms. In Aim 2, we will re-analyze potency including all 37 readouts using a benchmark concentration approach with empirically determined, endpoint- specific benchmark responses. This analysis will overcome the intrinsic limitations associated with lowest- observed-effect levels that was previously applied. In Aim 3, we will use a Bayesian statistical model originally developed for zebrafish embryo screens to obtain a holistic toxicity summary score and evaluate the relative importance of the different readouts for the predictive capabilities of the planarian system. Expected results. By combining non-animal organismal behavioral HTS with state-of-the-art analytical approaches, this project will bolster the development of a non-animal organismal HTS methodology that can be integrated with predictive bioinformatics to meet the urgent need to fill the DNT data gap.