HIGH RESOLUTION IN SINGLE PARTICLE RECONSTRUCTION

低温电子显微镜结合单粒子分析是一种提供大分子结构和功能信息的实验技术。在co-PI实验室中，核糖体3D重建的最新进展极大地改变了冷冻电镜在理解翻译装置中的作用，并首次提供了tRNA的3D可视化以及核糖体上的原位延伸因子。三维重建的核糖体在一个明确定义的构象状态导致密度图在15埃单位分辨率。我们在初步实验中表明，通过将粒子图像的数量增加到大约60,000个，可以进一步提高分辨率。此外，我们还证明了构象变异性是主要的分辨率限制因素，并表明正确选择分辨率度量有助于选择同质数据集。该提案描述了一项为期5年的研究计划，旨在开发一个统一的软件框架，以加快处理超过100,000个粒子图像的低温电镜数据。为了实现这样的吞吐量，我们将设计一个全自动颗粒拾取系统，该系统将利用有关低温电镜数据的纹理和先验结构信息。为了限制数据集的可变性，我们将开发基于3D模板的粒子分类方案。数学和数值工具将被实施，面向结构分辨率的多阶段优化。具体而言，将识别结构重建过程中可以控制的所有参数，并将其与所使用的分辨率措施联系起来。我们期望通过解决数据收集问题和构象异质性问题，我们将能够实质性地推进单粒子重建的分辨率。我们提出的方法有望在大分子结构和相互作用的研究中产生很大的影响。

Cryo-electron microscopy augmented by single particle analysis is an experimental technique that provides information about structure and function of macromolecules. Recent developments in 3D reconstructions of the ribosome in the co-PI's lab have dramatically changed the role that cryo-EM plays in understanding of the translational apparatus, and have provided the first 3D visualizations of tRNA as well as elongation factors in situ on the ribosome. 3D reconstruction of the ribosome in a well- defined conformational state resulted in a density map at 15 Angstrom units resolution. We show in preliminary experiments that further improvements in resolution were possible by increasing the number of particle images to approximately 60,000. In addition, we demonstrate that the conformational variability is the major resolution-limiting factor and we show that a proper choice of resolution measure facilitates selection of homogenous data sets. This proposal describes a 5-years research plan that seeks to develop a unified software framework that will expedite processing of cryo-EM data in excess of 100,000 particle images. To achieve such a throughput, we will design a fully automatic particle picking system that will utilize both textural and a priori structural information about the cryo-EM data. To limit the variability in the data set we will develop particle classification schemes that are based on 3D templates. Mathematical and numerical tools will be implemented that are geared toward a multistage optimization of the structural resolution. Specifically, all the parameters that can be controlled in the process of the structure reconstruction will be identified and linked to the resolution measure used. We expect that by addressing the data collection problem and the problem of conformational heterogenity, we will be able to advance the resolution of single-particle reconstructions substantially. The methods we propose to develop promise to have a high impact in the study of macromolecular structure and interaction.