High-throughput high-resolution microscopy for phenotypic drug discovery applications

摘要：多药耐药(MDR)是导致癌症化疗失败的主要原因。 健康问题。通常，多药耐药癌具有侵袭性、转移性和预后不良的特点。此外，MDR 癌症对导致常规程序性细胞死亡的治疗具有高度抵抗力，如化疗 和辐射。为了对抗细胞凋亡介导的多药耐药，新的药物发现正在被引导 用于诱导细胞凋亡抑制过程的治疗，如坏死性下垂、自噬、细胞凋亡、 甲氧西林和铁性下垂。当在药物发现的早期阶段优化新的化学分子时， 需要解决两个关键问题：a)药物杀死癌细胞的能力；b)通过 这种药物可以杀死癌细胞。目前，常规生化分析和高清晰度成像是 研究这些过程的唯一方法，但它们耗时、成本高，并且需要熟练的专家， 因此，它们的用途仅限于少数实验室。 我们提出了一种改变范式的表型筛选工具，它可以实时识别细胞死亡机制 使用高分辨率广域显微镜结合深度学习。首先，我们建议开发一种低成本的 宽视场全息显微镜，具有多波长照明，包括紫外线(UV)，使高 内容筛选无需外部标记，分辨率高，超过衍射极限。我们将实现 这是通过将无透镜全息显微镜与基于微粒阵列的成像衬底相结合来实现的，这将 让我们在每一次测试中都能看到数千个活细胞。紫外线照射可提供有关原子核的额外信息 细胞的其他细胞器，即使它被很少使用。利用癌细胞的时间推移图像，3D- 卷积神经网络将被训练来识别不同的形态特征，如收缩， 气泡、空泡和膜破裂，与不同的细胞死亡过程有关。在孵化期间 步骤，将使用预先训练的网络对细胞进行自动分类，从而实时获得结果 死亡过程。利用这种方法，可以筛选新的抗癌药物分子及其中间体。 无需进一步处理或贴标签即可实现高吞吐量。这个项目的圆满完成 将产生可用于许多不同应用的负担得起的、紧凑的、高含量的筛选工具， 除了表型筛选。特别是在这场新冠肺炎危机期间，这突显了 高通量诊断、药物筛选和治疗工具。 通过实施这一领域奖，我们将显著加强托莱多大学的研究氛围和 为本科生提供独特的生物物理、显微镜、 成像、深度学习、细胞生物学和药理学。

Abstract: Multidrug resistance (MDR) is a major cause of chemotherapy failure in cancer and a major public health concern. Often, MDR cancers are aggressive, metastatic, and have poor prognoses. In addition, MDR cancer is highly resistant to treatments that induce conventional programmed cell death, such as chemotherapy and radiation. For the purpose of combating apoptosis mediated MDR, new drug discoveries are being directed towards therapies that induce apoptosis-inhibiting processes, such as necroptosis, autophagy, paraptosis, methuosis, and ferroptosis. When optimizing new chemical molecules during the early phases of drug discovery, two key questions need to be addressed: a) the ability of the drug to kill cancer cells, and b) the mechanism by which the drug kills cancer cells. At present, conventional biochemical assays and high-definition imaging are the only methods for studying these processes, but they are time consuming, costly, and require skilled experts, thus limiting their utility to a small number of laboratories. We propose a paradigm-altering phenotypic screening tool that identifies cell death mechanisms in real time using high-resolution widefield microscopy coupled with deep learning. First, we propose to develop a low-cost widefield holographic microscope, with multi-wavelength illumination, including ultraviolet (UV), to enable high content screening without external labeling, at high resolution that exceeds the diffraction limit. We will achieve this by integrating lens-less holographic microscopy with microparticle array-based imaging substrates that will allow us to image thousands of live cells per test. UV illumination may provide extra information about nuclei and other organelles of cells, even though it is used sparingly. Using time lapse images of cancer cells, a 3D- convolutional neural network will be trained to identify different morphological features, such as shrinking, blebbing, vacuoles and membrane ruptures, associated with different cell death processes. During the incubation step, results will be obtained in real time, using the pre-trained network for automated classification of the cell death process. Using this approach, new anti-cancer drug molecules and their intermediates can be screened at high-throughput without requiring any further processing or labeling. A successful completion of this project will result in an affordable, compact, high-content screening tool that can be used for many different applications, in addition to phenotypic screening. In particular, during this Covid-19 crisis, which has highlighted the need for high throughput diagnostics, drug screening, and therapy tools. By implementing this AREA award, we will significantly strengthen University of Toledo's research climate and provide undergraduate students with a unique interdisciplinary training experience in biophysics, microscopy, imaging, deep learning, cell biology, and pharmacology.