Development of computational methods for exploration of cryo-electron tomograms

低温电子断层扫描（CET）具有独特的对天然环境下大分子复合物进行三维成像的能力，是现代结构生物学中一种重要的成像方式。利用透射电子显微镜（TEM）从顺序倾斜的样品中获得的三维（3-D）密度近似于其投影。在这里，我们建议开发计算方法，以提高从低温电子层析数据中获得的三维重建的准确性。首先，我们将实现一种新的算法，用于将二维TEM图像配准到一个共同的坐标系。特别是，这种方法侧重于较小的区域的标本体积，而不是整个成像区域。这种方法可以补偿光束引起的样品变化，这是目前从CET获得高分辨率的最大挑战。由于配准算法不依赖于基准标记，因此对于使用聚焦离子束制备的样品的分析将特别重要，并且它可以进一步实现层析重建的自动化。其次，我们将实现一种新的迭代方法，从注册的二维图像中重建三维体。三个特征表明，该方法将产生比目前使用的方法更精确的重建：(i)非均匀傅立叶变换允许高度精确的插值和使用适当考虑实验几何形状的度量；（ii）正则化避免了对低信噪比数据的过拟合（如CET）；（iii）试样形状的实际空间限制允许部分恢复实验中无法获得的结构数据（缺失的楔形）。第三，我们将把这种重建方法整合到现有的框架中，用于描述特定大分子复合物的单个拷贝的亚层析图的统计分析。子层析图分析包括相干对准和平均，以获得比单个噪声层析图分辨率高得多的平均密度，以及根据所描绘复合物的结构差异对子层析图进行分类。具体来说，我们将使用不同的三维重建进行对齐/分类和平均。最后，开发的方法将被广泛地记录下来，并将编写合适的教程，以最大限度地提高算法在科学界的价值。总之，本提案中开发的方法将使我们能够更详细地了解复合物在其原生环境中的结构及其构象变化。

Cryo-electron tomography (CET) has the unique ability to image macromolecular complexes in their native environment in three dimensions, making it an important imaging modality in modern structural biology. The three-dimensional (3-D) density of the frozen-hydrated sample is approximated from its projections, which are acquired from the sequentially tilted specimen using a transmission electron microscope (TEM). Here, we suggest developing computational methods that improve the accuracy of 3-D reconstructions derived from cryo-electron tomographic data. Firstly, we will implement a novel algorithm for registration of the two-dimensional (2-D) TEM images to a common coordinate system. In particular, this approach focuses on smaller areas of the specimen volume rather than the whole imaged area. This approach allows compensation for beam-induced alteration of the sample, currently arguably the biggest challenge for obtaining high-resolution from CET. Since the registration algorithm does not rely on fiducial markers it will be of particular importance for the analysis of samples prepared using a focused ion beam and it enables further automation of tomographic reconstruction. Secondly, we will implement a novel iterative method for reconstruction of the 3-D volume from the registered 2-D images. Three features suggest that the methodology will yield substantially more accurate reconstructions than approaches currently used: (i) nonuniform Fourier transforms allows highly accurate interpolation and the use of a metric that considers the geometry of the experiment appropriately; (ii) regularization avoids overfitting to noise for data with low signal-to-noise ratios as in CET; (iii) real space constraints on the shape of the specimen allow partial restoration of structural data that are inaccessible in the experiment (missing wedge). Thirdly, we will integrate this reconstruction methodology into existing frameworks for the statistical analysis of subtomograms depicting single copies of specific macromolecular complexes. Subtomogram analysis involves coherent alignment and averaging to obtain averaged densities with much higher resolution than the individual noisy tomograms, as well as classification of subtomograms according to structural differences of the depicted complexes. Specifically, we will use distinct 3-D reconstructions for alignment/classification and averaging. Finally, the developed methodology will be extensively documented and suitable tutorials will be compiled to maximize the value of the algorithms in the scientific community. In summary, the methodology developed in this proposal will enable much more detailed insights into the structures of complexes in their native settings as well as their conformational changes.