Methods for merging data sets in electron cryo-microscopy

Abstract

Recent developments have resulted in electron cryo-microscopy (cryo-EM) becoming a useful tool for the structure determination of biological macromolecules. For samples containing inherent flexibility, heterogeneity or preferred orientation, the collection of extensive cryo-EM data using several conditions and microscopes is often required. In such a scenario, merging cryo-EM data sets is advantageous because it allows improved three-dimensional reconstructions to be obtained. Since data sets are not always collected with the same pixel size, merging data can be challenging. Here, two methods to combine cryo-EM data are described. Both involve the calculation of a rescaling factor from independent data sets. The effects of errors in the scaling factor on the results of data merging are also estimated. The methods described here provide a guideline for cryo-EM users who wish to combine data sets from the same type of microscope and detector.

Keywords: RELION; cryo-EM; merging of data; single-particle processing.

Figure 1

Schematic overview of two methods for scaling cryo-EM data. In Method 1, micrographs are scaled (pink arrow) before particles are extracted (blue arrow). In Method 2, particles are extracted and then scaled, with a optional cropping step if required.

Figure 2

Flowchart of the data-merging process. The first step for both methods is obtaining independent 3D reconstructions to calculate the scaling factor between data sets. Method 1 rescales the data at the level of micrographs and Method 2 uses extracted particles. It is essential to redo the CTF estimation (Method 1) or to apply the scaling factor to defocus values to recalculate the CTF (Method 2). Once the particles from the two data sets have been merged (‘Join’ job in RELION), further processing can be carried out using standard procedures. Asterisks (*) indicate where scripts are provided to perform different steps in the process: 1*, determine_relative_pixel_size.py; 2*, rescale_particles.py; 3*, scale_ctf.sh; 4*, boxscaler.py.

Figure 3

Scaling factor between data sets. (a) Independent 3D maps obtained from data set I (yellow) and data set II (cyan) vary in volume size owing to differences in the nominal pixel sizes. Superposition of such maps in Chimera shows that correlation between the reconstructions improves when the pixel size of map 2 (cyan) is scaled to fit the pixel size of map 1 (yellow). (b) Fourier shell correlation (FSC) between maps 1 and 2, before and after scaling. The correlation increases at 1.28 Å per pixel.

Figure 4

Merging data sets of the polymerase module of CPF from yeast using Method 1. (a) Reconstructions of the polymerase module of CPF are shown before and after joining the data. The global and local resolution of the final 3D structure (EMDB-3908) improves after combining the data sets. Maps from data sets I and II represent the particle contribution of each data set to the final structure. Local resolution was determined using RELION for 3D reconstructions of particles from data sets I and II alone (at the final/scaled pixel size, i.e. 1.40 Å per pixel) and from combining data sets after scaling data set II using a relative pixel size of 1.28 Å per pixel. Resolutions are given according to the gold-standard criteria. (b) Fourier shell correlation plots for gold-standard refinements. (c) Example density for data sets I and II and combined data sets. The final 3D structure shows finer details in the main chain and side chains when compared with the individual data sets, as indicated by green arrows.

Figure 5

Merging data sets of the yeast post-catalytic (P complex) spliceosome using Method 2. (a) The global and local resolution of the final 3D structure improves after combining the data sets. Local resolution was determined using RELION for 3D reconstructions of particles from data sets I and II alone (at the final/scaled pixel size) and from combining data sets after scaling data set II using a relative pixel size of 0.880 Å per pixel. Resolutions are given according to the gold-standard criteria. (b) Example density for data sets I and II and combined data sets.

Figure 6

Effect of scaling. (a) Correlation between the 3D reconstructions calculated from data set I and data set II (at 1.120 Å per pixel) using various pixel sizes for data set II. The correlation was calculated using Chimera as described in the text. (b) Gold-standard FSC for spliceosome reconstructions for data set I alone (light grey), data set II alone (black) and from combining data as described in the text (red). Additional curves come from mis-scaling data set II using the indicated pixel size to 1.120 Å per pixel, then merging with data set I, refining and post-processing. These curves are coloured the same as the corresponding points in (a). Resolutions are given according to the gold-standard criteria.