Controlling Stem Cell Fate through Computational Modeling

摘要 自从诱导多能干细胞被发现以来，再生医学的一个主要目标就是 用患者来源的细胞替换受损或患病的组织。原则上，应该可以生成 临床使用的功能组织，通过特定改变命运的定向分化方案 多能细胞。不幸的是，与定向差异化相关的挑战很多。首先，我们做到了 没有完全了解参与发育调节通路的所有信号机制。我们 因此，不能总是确定概括这些信号的分子因素的正确组合 在实验室里。即使所需的因素是已知的，也不清楚什么剂量、时间或组合 将产生一种有效的差异化。大多数协议都是通过反复试验来开发的；需要数年时间才能 优化；并导致所需细胞类型的不完整、低效或异质混合。显然，在那里 如果我们希望生产出现实的基于细胞的疗法，那么改善分化过程是一个迫切需要。 在这篇文章中，我提出了一种控制人类干细胞命运的新方案。这种方法，我们 都是我们实验室的先驱，其工作原理是允许计算模型设计差异化协议。至 建立模型后，我们使用时间推移显微镜来监测关键发育标志物在 分化的时间进程。这些数据提供了对细胞命运决定的机械描述 单元格分辨率。为了训练模型，我们然后应用一系列系统的扰动来进行微分 通过活细胞显微镜监测细胞，并将这些新的测量结果整合到工作模型中。 最后，我们使用该模型作为预测工具，通过执行数千个虚拟实验来识别一组 预计会以规定的方式改变干细胞命运的扰动。模型预测得到验证 通过单细胞转录图谱，这些数据被用来反复精炼模型的预测 权力。作为原则的证明，我们将使用这种方法来产生人类肺的精确组合 祖细胞可用于治疗肺部疾病，如肺纤维化、肺气肿或 间质性肺病。在短期内，我们的方法将使我们能够操纵多能细胞来获得 分化细胞命运的精确组合--克服干细胞治疗中的一大障碍。在漫长的岁月里 术语，这种方法可以更普遍地用于控制其他类型的细胞的命运，包括细菌或 癌症细胞，并可能有助于设计改进的药物输送方案。

ABSTRACT Since the discovery of induced pluripotent stem cells, a major goal for regenerative medicine has been to replace damaged or diseased tissues with patient-derived cells. In principle, it should be possible to generate functional tissues for clinical use through directed differentiation protocols that specifically alter the fate of pluripotent cells. Unfortunately, there are many challenges associated with directed differentiation. First, we do not fully understand all of the signaling mechanisms that participate in developmental regulatory pathways. We therefore cannot always identify the correct combination of molecular factors that will recapitulate these signals in a laboratory setting. Even if the required factors are known, it is unclear what doses, timing, or combinations of factors will produce an efficient differentiation. Most protocols are developed by trial-and-error; take years to optimize; and lead to incomplete, inefficient, or heterogeneous mixtures of the desired cell types. Clearly, there is a critical need for improving differentiation procedures if we expect to produce realistic cell-based therapies. In this essay, I propose a novel solution for controlling the fate of human stem cells. The approach, which we have pioneered in our lab, works by allowing a computational model to design the differentiation protocol. To build the model, we use time-lapse microscopy to monitor the expression of key developmental markers over the time course of differentiation. These data provide a mechanistic description of cellular fate decisions at single-cell resolution. To train the model, we then apply a series of systematic perturbations to differentiating cells, monitor the cells by live-cell microscopy, and integrate these new measurements into the working model. Finally, we use the model as a predictive tool by performing thousands of virtual experiments to identify a set of perturbations that are predicted to alter stem cell fate in a prescribed way. Model predictions are validated through single-cell transcription profiling, and these data are used to iteratively refine the model’s predictive power. As proof of principle, we will use this approach to produce precise combinations of human lung progenitor cells that could be used to treat pulmonary diseases such as pulmonary fibrosis, emphysema, or interstitial lung disease. In the near term, our approach will enable us to manipulate pluripotent cells to acquire a precise combination of differentiated cell fates—overcoming a major hurdle in stem cell therapy. In the long term, this approach could be used more generally to control the fate of other cell types including bacteria or cancer cells, and may help to design improved drug delivery protocols.