Technology Development for 3D Electron Microscopy

在与GATAN公司的合作中，我们报道了一种用于高通量透射电子显微镜的新型多标本成像系统的开发，该系统绕过了人工标本加载所涉及的耗时步骤。鲍勃·莫里森和我设计了这个系统；所有的硬件构建都是在Gatan UK完成的，而所有的软件都是在我的实验室完成的。这种墨盒为基础的加载系统，称为加特林，允许顺序检查多达100个标本在显微镜下的室温电子显微镜使用快速和自动标本交换机制。用于加特林操作和自动数据采集的软件已在我们内部程序AutoEM的更新版本中实现。在目前的系统实施中，将95个标本送入显微镜并收集每个标本的总览图像所需的时间约为13小时。感兴趣的区域是从每个标本的低倍率图谱生成中识别出来的，随后可以使用可以从远程接口控制的全自动数据采集程序从这些区域获得无限数量的更高倍率的图像。我们预计Gatling的可用性将大大加快生物学、材料科学和纳米技术中需要快速筛选和图像分析多个标本的各种应用的数据采集速度。利用电子晶体学和单粒子电子显微镜确定生物大分子三维结构的策略利用强大的工具来平均从结构均匀标本的二维投影图像中获得的信息。相比之下，电子层析成像方法经常用于研究异质的三维结构，一类物体，如整个细胞，其中图像平均策略不适用。然而，诸如细胞和病毒之类的复杂实体包含许多大分子的多个副本，这些大分子可以单独地进行3D平均。我们已经开发了一个完整的框架，用于通过电子断层扫描得出的体积的对准、分类和平均，该框架具有计算效率，并有效地解释了有限角度电子断层扫描固有的楔形缺失。将缺失的数据建模为互反空间中的乘法掩模，我们已经表明，缺失的楔形的影响可以在所有对齐和分类操作中无缝地考虑。我们使用调和分析中的卷积定理来解决对齐问题，从而消除了需要穷举角搜索的方法的需要，并且采用不需要使用外部参考的迭代方法来对齐和分类。我们还证明，我们的方法可以成功地应用于幻影体的3D分类和平均，以及实验获得的GroEL层析图，其中分析结果可以与预期结果进行定量比较。另一个重点领域是图像处理和分割。以往使用非线性各向异性方法、基于小波的方法和滤波的研究已经证明了在各种二维和三维数据集上图像去噪的价值。现有方法通常考虑干净数据（或假设干净数据可用），人为地在干净数据中添加不同类型的噪声，然后假设已知噪声的统计量，使用各种算法对噪声数据进行去噪。此外，由于噪声模式的变化，单个去噪算法可能无法在使用各种样本和采集条件收集的不同数据集上表现一致。由于低辐射、大量的生物结构、排列和重建中的错误以及缺失的楔形存在，可能会发生变化。我们已经研究了使用变换域去噪技术和特征提取来改进病毒和细胞的冷冻电子断层图的定量解释。我们已经评估了各种去噪算法在特定大分子复合物空间分布的检测、聚类和计算方面的相对优点。与之前的大多数去噪分析相反，我们处理的是在互反空间中有有限信息缺失楔的3D体积，我们既没有噪声类型的先验知识，也没有其统计数据。此外，据我们所知，使用拟合优度（GOF）标准的去噪算法的定量分析尚未应用于电子断层扫描中的噪声分析。在我们的方法中，我们使用了四个指标进行分析，包括基于Kullback-Leibler （KL）距离的GOF测试、傅立叶环相关性和单图像信噪比，以迭代获得给定3D体积的最佳去噪算法。使用这些方法，我们表明，当小心使用去噪时，对于高吞吐量的复杂3D数据集的自动解释是一个非常强大的工具

In a collaboration with GATAN Inc., we have reported the development of a novel, multi-specimen imaging system for high-throughput transmission electron microscopy that circumvents time-consuming steps involved in manual specimen loading. Bob Morrison and I designed the system; all of the hardware construction was done at Gatan UK, while all of the software was carried out in my lab. This cartridge-based loading system, called the Gatling, permits the sequential examination of as many as 100 specimens in the microscope for room temperature electron microscopy using mechanisms for rapid and automated specimen exchange. The software for the operation of the Gatling and automated data acquisition has been implemented in an updated version of our in-house program AutoEM. In the current implementation of the system, the time required to deliver 95 specimens into the microscope and collect overview images from each is about 13 hours. Regions of interest are identified from a low magnification atlas generation from each specimen and an unlimited number of higher magnifications images can be subsequently acquired from these regions using fully automated data acquisition procedures that can be controlled from a remote interface. We anticipate that the availability of the Gatling will greatly accelerate the speed of data acquisition for a variety of applications in biology, materials science and nanotechnology that require rapid screening and image analysis of multiple specimens. Strategies for the determination of 3D structures of biological macromolecules using electron crystallography and single particle electron microscopy utilize powerful tools for the averaging of information obtained from 2D projection images of structurally homogeneous specimens. In contrast, electron tomographic approaches have often been used to study the 3D structures of heterogeneous, one-of-a-kind objects such as whole cells where image averaging strategies are not applicable. Complex entities such as cells and viruses, nevertheless, contain multiple copies of numerous macromolecules that can individually be subjected to 3D averaging. We have deeloped a complete framework for alignment, classification, and averaging of volumes derived by electron tomography that is computationally efficient and effectively accounts for the missing wedge that is inherent to limited angle electron tomography. Modeling the missing data as a multiplying mask in reciprocal space we have shown that the effect of the missing wedge can be accounted for seamlessly in all alignment and classification operations. We solve the alignment problem using the convolution theorem in harmonic analysis, thus eliminating the need for approaches that require exhaustive angular search, and adopt an iterative approach to alignment and classification that does not require the use of external references. We also demonstrated that our method could be successfully applied for 3D classification and averaging of phantom volumes as well as experimentally obtained tomograms of GroEL where the outcomes of the analysis can be quantitatively compared against the expected results. Another area of focus has been on image processing and segmentation. Previous studies using nonlinear anisotropic methods, wavelet based methods and filtering have already demonstrated the value of image denoising in various 2D and 3D datasets. The existing methods usually consider clean data (or assume that clean data is available) and artificially add different types of noise to the clean data and then denoise the noisy data assuming that the statistics of noise is known using various algorithms. Furthermore, a single denoising algorithm may not perform uniformly well on diverse datasets that have been collected using a variety of specimens and acquisition conditions due to variations in noise patterns. Variations may occur due to low radiation, a lot of biological structure, errors in alignment and reconstruction and the presence of the missing wedge. We have investigated the use of transform-domain denoising techniques and feature extraction to improve quantitative interpretation of cryo electron tomograms of viruses and cells. We have evaluated the relative merits of a variety of denoising algorithms on the detection, clustering, and computation of the spatial distribution of specific macromolecular complexes. As opposed to most previous denoising analyses, we are working with 3D volumes that have a finite missing wedge of information in reciprocal space, and we have neither an apriori knowledge of the type of noise nor its statistics. Moreover, quantitative analysis of denoising algorithms using goodness-of-fit (GOF) criteria as we implemented have not yet been applied, to our knowledge, for noise analysis in electron tomography. In our approach, we have used four metrics for analysis including the Kullback-Leibler (KL) distance based GOF test, Fourier ring correlation and single-image SNR to iteratively obtain the optimal denoising algorithm for a given 3D volume. Using these methods, we show that denoising, when used with care is an enormously powerful tool for the automated interpretation of complex 3D data sets at high throughput