Frealign: An Exploratory Tool for Single-Particle Cryo-EM

Abstract

Frealign is a software tool designed to process electron microscope images of single molecules and complexes to obtain reconstructions at the highest possible resolution. It provides a number of refinement parameters and options that allow users to tune their refinement to achieve specific goals, such as masking to classify selected regions within a particle, control over the refinement of specific alignment parameters to accommodate various data collection schemes, refinement of pseudosymmetric particles, and generation of initial maps. This chapter provides a general overview of Frealign functions and a more detailed guide to using Frealign in typical scenarios.

Keywords: Asymmetry; Classification; Contrast transfer function; High resolution; Masking; Refinement.

Figure 1

Schematic overview of Frealign functions, input and output data. Some of the Frealign commands mentioned in the text are shown in Courier font above the functions or files they relate to. Features and files that are optional and not always used are shown with dashed borders. Frealign is run by a number of shell scripts that prepare input and output data, manage parallel execution and perform iterations when more than one cycle is run.

Figure 2

Template for the mparameters file containing all the control parameters required to run Frealign. Each keyword has an assigned value or string (text) and, in most cases, a comment line explaining how the parameter should be set. The control parameters are divided into different sections to help users customize the parameters according to their environment and project. Parameters listed in the expert section usually do not need to be changed until refinement and classification have converged. These parameters allow users to tune their refinement to improve resolution but they require some experience to be set correctly.

Figure 3

Example of a startup parameter file containing the required data for 20 particles. Each line lists a micrograph number that the particle originates from, as well as defocus values and astigmatic angle determined for this micrograph. The defocus information can vary from particle to particle if more accurate information is available, for example by using CTFTILT (Mindell and Grigorieff, 2003). If micrograph numbers are not known, they can be set to a constant number larger than 0 or to the particle number.

Figure 4

Example of a full Frealign alignment parameter file, after running Frealign with the startup file in Fig.3, containing Euler angles (PSI, THETA, PHI) and x,y translations (SHX, SHY), as well as micrograph numbers (FILM), magnification (MAG) and defocus (DF1, DF2, ANGAST) information, occupancies (OCC), log likelihoods (LogP) and scores (SCORE). The SIGMA column lists estimates of the standard deviation of the noise present in the particle images while CHANGE lists the change in the score compared with the previous refinement cycle.

Figure 5

Example of an initial map generated from a reconstruction calculated using randomly assigned Euler angles. The dataset contained images of L protein of vesicular stomatitis virus (VSV-L) and led to a 3.8-Å reconstruction of this multi-enzyme (Liang et al., 2015). To initiate the startup procedure, the particle images was binned 3-fold to an effective pixel size of 3.711 Å and cropped to generate particle images of 60 × 60 pixels. A subset of 11,671 particles with an underfocus ranging between 1.7 and 2.5 µm were selected form the complete dataset (356,211 particles) and 40 rounds of multi-resolution search and refinement were performed according to the scheme described in Section 4.5, using five classes and starting with a resolution limit of 40 Å that was gradually increased to 10 Å resolution. FBOOST was set to T for this startup procedure. Reconstructions at each stage are shown together with the calculated FSC curves, starting with the initial map obtained with randomly assigned angles (labeled “Initial”). The reconstruction with the best FSC curve from each round was selected to seed the next round of refinement and classification. The FSC curve gradually improves from round to round until the final round 4 in which only 25 refinement cycles were run and the resolution was limited to 14 Å. The FSC curve for the final round indicates a resolution of about 9.5 Å according to the 0.143 criterion (Rosenthal and Henderson, 2003) indicated by the horizontal gray line, thus significantly exceeding the resolution limit used in the refinement (indicated by the vertical gray line) and therefore reflecting an unbiased resolution estimate. The FSC curves in earlier rounds likely reflects some bias as the resolution limit during refinement exceeded to resolution indicated by the FSC. The final published map is shown in the last panel for comparison.

Figure 6

Example of a classification scheme using 2D and 3D masking. The dataset consisted of 80S ribosomes prepared with the Taura syndrome virus internal ribosome entry site (TSV IRES) and elongation factor 2 (eEF2) (Abeyrathne et al, 2016). The complete dataset of 1,105,737 images of 80S•IRES•eEF2 complex was initially aligned against a density map calculated from the atomic model of the non-rotated 80S ribosome bound with 2 tRNAs (PDB: 3J78 (Svidritskiy et al., 2014)). This initial alignment was performed on data with a pixel size of 1.64 Å and limited to 20 Å resolution, resulting in a 3.5-Å resolution reconstruction. After five cycles of refinement the data were 2x binned (by Fourier cropping using the resample.exe tool, new pixel size = 3.28 Å) and subjected to classification into 15 classes using a 3D mask that contained the IRES, eEF2 and head domain of the small subunit. Six of the resulting classes (312,692 particle images) contained density for the IRES and eEF2 and were further classified into eight classes. For this classification, a 2D mask was applied around the ribosomal A site to include IRES pseudoknot I and eEF2 domain IV. The figure shows this mask as a sphere which, when projected according to the orientation of a particle, results in a 2D mask correctly placed on the region of interest. In the case of the 80S•IRES•eEF2 complex, this focused classification resulted in the separation of different translocation states of the IRES, catalyzed by eEF2, as shown schematically below each reconstruction. The states containing clear density for the IRES and eEF2 are highlighted in color.