Phenotypic variability within isogenic population of lymphocytes

我们发现表型变异可以由信号转导级联中关键组分的异质表达驱动。这种可变性在模拟细胞对药物反应的变化时具有实际意义。特别是，我们发现信号转导级联的拓扑结构解释了为什么近端信号成分（如Src）的小药物抑制剂以数字方式（即以全有或全无的方式）起作用，而远端信号成分（如Mek）的抑制剂以类似方式（即以连续方式）起作用。我们用一种新的方法（称为“细胞间变异性分析”）扩展了我们对细胞信号的表型变异性的发现：这种方法依赖于临床前和临床环境中原代细胞的单细胞磷酸化谱分析，以确定哪些生物成分（受体、激酶、磷酸酶、转录因子）在功能后果方面受到定量限制。我们通过展示对IL-2和IL-7的反应如何在单个原代T细胞中相互排斥来说明这种方法的强度。一个基于生化建模和贝叶斯优化的计算模型被引入，以测试共享但有限的受体链的隔离如何在细胞因子信号传导中产生这种翻转。该研究为T细胞等基因群体中效应细胞和记忆细胞之间的转变提供了机制解释（Cotari et al., Science Signaling, 2013）。同时，我们引入并发布了一个计算机程序（名为ScatterSlice），使实验人员能够分析流式细胞术数据中的细胞间变异性（Cotari等）。Science Signaling, 2013)。这种方法已在许多临床环境中得到应用（Palomba等人，PLoS One, 2014； Kitano等人，Cancer Immunol Res, 2014）。与此同时，我们一直在美国国立卫生研究院实施大规模细胞术（所谓的CyTOF）。CyTOF能够实现大量抗体的多路复用（通常一次40个），同时绕过了传统的基于荧光细胞术的光谱重叠问题。我们验证并优化了多个抗体小组，以分析多种免疫系统：小鼠和人类骨髓的一般概况，小鼠和人类T细胞群的深度概况，人类中性粒细胞和人类B细胞。我们与美国国立卫生研究院的临床研究人员合作，在XMEN、ALPS、狼疮（PBMC）和黑色素瘤（TIL）的背景下分析患者样本。此外，我们还开发了一种脉冲追踪IdU（一种被提取并插入增殖细胞中的试剂）和监测小鼠白细胞分化动力学的方法。最后，我们引入了一种基于机器学习的方法来自动识别正在考虑的白细胞之间的分化簇：该方法用于定义表型与阳性临床结果相关的新T细胞群。我们的目标是更好地描述免疫反应的细胞复杂性，更好地分类疾病和治疗状态，以及更好地建模。最后，我们正在开发一个平台来解决白细胞多样性（它的起源和功能意义）。我们建立了一个定制的机器人组织培养系统，系统地组装体外免疫反应（使用小鼠或人类的主要样本），并监测其时间动态。我们的系统通常生成500个条件（作为细胞含量，激活条件，药物扰动和时间点的卷积），这些条件以其可溶性含量（细胞因子分泌）和细胞组成（通过CyTOF和FACS进行单细胞分析）为特征。然后，我们应用人工智能领域的工具来反卷积这些免疫反应的组合复杂性。我们的目标是确定新的免疫特征和特征，最好地分类免疫反应，然后在免疫反应模型中验证这些特征（针对疫苗和/或肿瘤）。

We found that phenotypic variability can be driven by the heterogenous expression of key components within signal transduction cascade. Such variability has practical importance when modeling how cells vary in their drug response. In particular, we found that the topology of signal transduction cascades explain why small-drug inhibitors of proximal signaling components (e.g. Src) acted digitally (i.e. in an all-or-none manner) while inhibitors against distal signaling components (e.g. Mek) acted analogously (i.e. in a continuous manner) We expanded our findings on the phenotypic variability of cell signaling with a new methodology (termed "cell-to-cell variability analysis): such method relies on single-cell phospho-profiling of primary cells in preclinical and clinical settings to identify which biological components (receptor, kinase, phosphatase, transcription factor) is quantitatively limiting in terms of functional consequences. We illustrated the strength of this approach by showing how response to IL-2 and IL-7 are mutually exclusive within individual primary T cells. A computational model, based on biochemical modeling and Bayesian optimization, was introduced to test how sequestration of a shared but limited receptor chain could generate such flip-flop in cytokine signaling. This study provided a mechanistic explanation for the transition between effector and memory cells within an isogenic population of T cells (Cotari et al., Science Signaling, 2013). Concomitantly, we introduced and distributed a computer program (named ScatterSlice) that enables experimenters to analyze the cell-to-cell variability in their Flow Cytometry data (Cotari et al. Science Signaling, 2013). Such methodology has found applications in many clinical settings (Palomba et al., PLoS One, 2014; Kitano et al., Cancer Immunol Res, 2014). In parallel, we have been implementing mass cytometry (so-called CyTOF) at the NIH. CyTOF enables the multiplexing of large sets of antibodies (typically 40 at once) while bypassing issues of spectral overlap of classical fluorescence-based cytometry. We validated and optimized multiple antibody panels to profile multiple immunological systems: general profile of bone marrow in mouse and human, deepvprofile of T cell populations in mouse and human, human neutrophils and human B cells. We collaborated with clinical investigators at the NIH to profile patients' samples in the context of XMEN, ALPS, Lupus (PBMC) and melanoma (TIL). Moreover, we developed a method to pulse-chase IdU (a reagent that gets picked up and inserted in proliferating cells) and monitor the kinetics of differentiation of leukocytes in mice. Finally, we introduced a machine-learning-based method to automatically identify clusters of differentiation amongst leukocytes under consideration: this method was applied to define a new T cell population whose phenotype correlates with positive clinical outcomes. We are pursuing our goal to better characterize the cellular complexity of immune responses, towards better classification of disease and therapeutic states, and better modeling. Finally, we are developing a platform to address leukocyte diversity (its origins and its functional significance). We built a custom-made robotic tissue culture system to systematically assemble immune responses ex vivo (using primary samples from mouse or human) and to monitor its time dynamics. Our system generates typically 500 conditions (as a convolution of cell contents, activation conditions, drug perturbations and time points) that get characterized for their soluble content (cytokine secretion) and cell composition (single-cell profiling by CyTOF and FACS). We then apply tools from the field of artificial intelligence to deconvolve the combinatorial complexity of these immune responses. Our goal is to identify new immune signatures & features that best classify immune responses, then to validate these signatures in models of immune responses (against vaccines and/or tumors).