Human iPS/ES Cell-Based Models for Predictive Neural Toxicity and Teratogenicity

描述（由申请人提供）：该提案汇集了人类多能干细胞生物学（Thomson）、组织工程（Murphy）和机器学习（Page）领域的领先专家，以开发用于预测发育神经毒性的改进的人类细胞模型。在从人类多能干细胞（ES和iPS细胞）中衍生出大脑的许多基本细胞成分方面取得了巨大进展，但这些进展尚未应用于预测毒理学。脑的主要成分来自不同的胚胎学起源，包括神经板（神经元、少突胶质细胞和星形胶质细胞）、卵黄囊髓样祖细胞（小胶质细胞）、迁移性中胚层成血管细胞（内皮细胞）和神经嵴（血管平滑肌和周细胞）。由于它们的不同起源，这些成分具有非常不同的诱导信号历史。这意味着在相同的条件下一次导出它们是不可能的。出于这个原因，我们将分别将人类多能干细胞分化为大脑皮层的主要神经、神经胶质和血管成分的早期前体，冷冻保存前体，随后将它们联合收割机组合在3D水凝胶组件中以允许增加的生理相互作用和成熟。具体来说，我们将嵌入承诺的前体内皮细胞，周细胞，和小胶质细胞到水凝胶显示的肽基序的组合，促进毛细血管网络的形成。然后，我们将用神经和神经胶质前体覆盖该间充质层，以模拟头部间充质和神经上皮之间的正常相互作用，并促进大脑皮层极化层的形成。在药物暴露后，我们将使用高度多重的深度RNA测序来评估这些脑神经血管组装体的基因表达的时间变化。然后，使用安全药物和已知的神经/发育毒素从美国国立卫生研究院临床收藏，华盛顿大学致畸信息系统数据库，和美国环保署的毒性参考数据库作为训练集，我们将开发机器学习算法来预测神经毒性盲药物已知已失败的后期动物试验或人体临床试验。这种预测性的发育神经毒性模型将在广泛使用的液体处理机器人和测序仪上实施，并将很容易适应DARPA正在开发的平台。人类多能干细胞的发育潜力、可调水凝胶的模块化性质以及机器学习工具的辨别能力也使得所提出的一般方法易于适用于全身其他组织类型的预测毒性模型。 公共卫生相关性：该项目将开发人类神经组织的三维结构，以便在临床试验之前更好地预测药物的神经毒性。为了实现这一目标，人类多能干细胞生物学专家将在实验室中培养所需的神经成分，组织工程师专家将把这些细胞组装成多细胞结构，机器学习专家将利用药物暴露后基因表达的变化来预测测试化合物是否有毒。

DESCRIPTION (provided by applicant): This proposal brings together leading experts in human pluripotent stem cell biology (Thomson), tissue engineering (Murphy), and machine learning (Page) to develop improved human cellular models for predicting developmental neural toxicity. Dramatic progress has been made in the derivation of many of the basic cellular components of the brain from human pluripotent stem cells (ES and iPS cells), but these advances have yet to be applied to predictive toxicology. The major components of the brain are derived from diverse embryological origins, including the neural plate (neurons, oligodendrocytes, and astrocytes), yolk sac myeloid progenitors (microglia), migratory mesodermal angioblasts (endothelial cells), and neural crest (vascular smooth muscle and pericytes). Because of their diverse origins, these components have very different inductive signaling histories. This means that deriving them all at once under the same conditions is not currently possible. For this reason, we will differentiate human pluripotent stem cells to early precursors of the major neural, glial, and vascular components of the cerebral cortex separately, cryopreserve the precursors, and subsequently combine them in 3D hydrogel assemblies to allow increased physiological interactions and maturation. Specifically, we will embed committed precursors for endothelial cells, pericytes, and microglia into hydrogels displaying combinations of peptide motifs that promote capillary network formation. We will then overlay this mesenchymal layer with neural and glial precursors to mimic the normal interactions between the cephalic mesenchyme and the neural epithelium, and promote the formation of the polarized layers of the cerebral cortex. After drug exposure, we will assess temporal changes in gene expression by these cerebral neural- vascular assemblies using highly multiplexed, deep RNA sequencing. Then, using safe drugs and known neural/developmental toxins from the NIH Clinical Collection, the University of Washington Teratogen Information System Database, and the EPA's Toxicity Reference Database as training sets, we will develop machine learning algorithms to predict neural toxicity of blinded drugs known to have failed in late stage animal testing or human clinical trials. This predictive, developmental neural toxicity model will be implemented on liquid handling robots and sequencers in widespread use, and will be readily adaptable to platforms being developed in complementary efforts by DARPA. The developmental potential of human pluripotent stem cells, the modular nature of the tunable hydrogels, and the discriminatory power of machine learning tools also makes the general approaches proposed readily applicable to predictive toxicity models for other tissue types throughout the body. PUBLIC HEALTH RELEVANCE: This project will develop three-dimensional constructs of human neural tissue to better predict the neural toxicity of drugs prior to clinical trials. To accomplish this, experts in human pluripotent stem cell biology will grow the required neural components in the laboratory, experts in tissue engineers will assemble those cells into multicellular constructs, and experts in machine learning will use changes in gene expression after drug exposure to predict whether a test compound is toxic.