Flame retardant chemicals (FRCs) commonly added to many consumer products present a human exposure burden associated with adverse health effects. Under pressure from consumers, FRC manufacturers have adopted some purportedly safer replacements for first-generation brominated diphenyl ethers (BDEs). In contrast, second and third-generation organophosphates and other alternative chemistries have limited bioactivity data available to estimate their hazard potential. In order to evaluate the toxicity of existing and potential replacement FRCs, we need efficient screening methods. We built a 61-FRC library in which we systemically assessed developmental toxicity and potential neurotoxicity effects in the embryonic zebrafish model. Data were compared to publicly available data generated in a battery of cell-based in vitro assays from ToxCast, Tox21, and other alternative models. Of the 61 FRCs, 19 of 45 that were tested in the ToxCast assays were bioactive in our zebrafish model. The zebrafish assays detected bioactivity for 10 of the 12 previously classified developmental neurotoxic FRCs. Developmental zebrafish were sufficiently sensitive at detecting FRC structure-bioactivity impacts that we were able to build a classification model using 13 physicochemical properties and 3 embryonic zebrafish assays that achieved a balanced accuracy of 91.7%. This work illustrates the power of a multi-dimensional in vivo platform to expand our ability to predict the hazard potential of new compounds based on structural relatedness, ultimately leading to reliable toxicity predictions based on chemical structure.

许多消费品中普遍添加的阻燃化学品 (FRC) 会给人类带来与不良健康影响相关的接触负担。在消费者的压力下，FRC 制造商采用了一些据称更安全的第一代溴化二苯醚 (BDE) 替代品。相比之下，第二代和第三代有机磷酸盐和其他替代化学物质可用于估计其潜在危险的生物活性数据有限。为了评估现有和潜在替代 FRC 的毒性，我们需要有效的筛选方法。我们建立了一个 61-FRC 库，在其中系统地评估了胚胎斑马鱼模型中的发育毒性和潜在的神经毒性作用。将数据与 ToxCast、Tox21 和其他替代模型的一系列基于细胞的体外测定中生成的公开数据进行比较。在 61 个 FRC 中，ToxCast 检测中测试的 45 个 FRC 中有 19 个在我们的斑马鱼模型中具有生物活性。斑马鱼检测检测到了先前分类的 12 种发育神经毒性 FRC 中 10 种的生物活性。发育中的斑马鱼在检测 FRC 结构-生物活性影响方面足够敏感，我们能够使用 13 种理化特性和 3 种胚胎斑马鱼检测建立分类模型，达到 91.7% 的平衡准确度。这项工作说明了多维体内平台的强大功能，可以扩展我们根据结构相关性预测新化合物潜在危险的能力，最终实现基于化学结构的可靠毒性预测。