High-resolution Single Particle Analysis from Electron Cryo-microscopy Images Using SPHIRE

The overall goal of this procedure is to determine the structure of macromolecular complexes by single particle analysis of electron cryo-microscopy data using the image processing software SPHIRE. Single particle cryo-EM is becoming a mainstream technology for studying macromolecule structures at near-atomic resolution. SPHIRE is a novel software package that enables semi-automated processing of cryo-EM data in a user-friendly manner.

The main advantage with SPHIRE is the automatic validation mechanism. Most of the procedures need to be run only once, and the results are objective and reproducible. High-resolution cryo-EM can be used to place individual residues, thereby providing crucial insight into the organization and function of complex micromolecular machines.

Individuals new to this method should keep in mind that obtaining a high-resolution structure highly depends on the quality of the sample and the input data. To begin using the SPHIRE application, open the GUI from the terminal. Open the UTILITIES to display a cryo-microscopy image with e2display.

Use the built-in measurement tool to measure the longest particle axis in angstrom per pixel. Open the Project Settings. Set the CTF window size to at least the particle box size.

Set the particle box size to an even number, at least 1 1/2 times the particle diameter. Set the particle radius to at least half of the measured length. Enter the point-group symmetry, if known, and the approximate molecular mass.

Register the settings. Next, in the MOVIE menu, open Micrograph movie alignment. Set the unblur and summovie executable file paths.

Choose a raw unaligned micrograph movie, and replace the file name variable with a wildcard to set the input micrograph path pattern. Name the output directory. Fill in the number of movie frames, the microscope voltage, and the per frame exposure.

Run the process to generate dose-weighted and dose-unweighted motion-corrected average micrographs. Next, open the CTER menu, and select CTF Estimation. Set the dose-unweighted micrographs as the input.

Name the output directory, and fill in the data collection parameters. Set the lowest frequency to 0285 inverse angstroms and the highest frequency to 0.285 inverse angstroms. Run the process to generate a list of CTF parameters.

Next, open the WINDOW menu, and select Particle Picking. Set the dose-weighted micrographs as the input, and start the e2boxer utility. Perform particle picking manually or automatically.

Extract the picked particles into image stacks, and combine the stacks into a single virtual image stack in a BDB database. Once the particles have been picked and stacked, open the ISAC menu and select ISAC 2D Clustering. Set the virtual image stack in the BDB database as the input.

Estimate the number of images per class. Fill in the remaining parameters, and generate the 2D class averages. Once 2D classification is complete, use the Display Data utility to view the validated reproducible class averages.

Verify that multiple particle orientations are represented and the class averages are of satisfactory quality. Generate a new virtual image stack containing only particles belonging to the validated class averages. Next, open the VIPER menu, and use the Display Data utility to view the class averages.

Delete class averages of poor quality or showing identical views of the particle. Save the remaining class averages to a new file. Then, select Initial 3D Model RVIPER, and use the small selection of class averages as the input.

Run the command to generate an initial reproducible 3D model of the protein. Inspect the model in the molecular graphics program Chimera. Compare the model to a known structure of a homologous protein or domain from the protein of interest.

Return to the VIPER menu in SPHIRE, and select Create 3D Reference. Generate an initial 3D reference model with the correct pixel size and a 3D mask. Open the MERIDIEN menu, and select 3D Refinement.

Fill in the input image stack and the initial 3D reference file paths. Name the output directory, and set the mask of the reference model as the 3D mask. Select Apply hard 2D mask.

Set the starting resolution to between 20 and 25 angstroms, and run the 3D refinement to produce the unfiltered half-volumes. Then, view in unfiltered half-volume in Chimera. Determine a binarization threshold in which all protein densities are connected but noise artifacts are not.

Return to the MERIDIEN menu of SPHIRE, and open Adaptive 3D Mask. Enter the unfiltered half-volume input and binarization threshold. Name the output mask, and run the process to generate a soft-edged 3D mask.

Then, open Sharpening, and select both unfiltered half-volumes. Set the soft-edge half-volume 3D mask as the user-provided mask. Run the process which merges the half-volumes and then filters and sharpens the merged volume.

Open the log file to view the final resolution estimate. Generate a 3D map of the particle angular distribution of the input image stack generated during 3D refinement. Open the sharpened and filtered 3D model in Chimera.

Inspect the density map and its high-resolution features. Open the angular distribution in Chimera, and verify that the Euler angles are isotropically distributed. Open the SORT3D menu, and select 3D Variability Estimation.

Generate a 3D variability map from the virtual stack of particle images belonging to the validated class averages. Generate a binarized 3D mask from this variability map for the subsequent focused 3D clustering. Then, open 3D Clustering RSORT3D, and select the directory of the previous 3D refinement as the input.

Name the output directory, and set the global and focused 3D masks. Use at least 5, 000 to 10, 000 images per group for large data sets. Run the sorting process to generate volumes of each homogeneous 3D group.

Use Chimera to identify the structure with the highest apparent resolution. Return to the SORT3D menu in SPHIRE, and open Local Subset Refinement. Set the subset text file path to the particle ID cluster file corresponding to the highest resolution structure.

Set the 3D refinement directory to the directory containing the unfiltered half-volumes and the restarting iteration to the highest resolution iteration from the previous 3D refinement. Run the command to generate local unfiltered half-volumes. Then, create a soft-edged 3D mask from one of the local half-volumes.

Merge the local half-volumes, and sharpen the merged volume without applying a low-pass filter. Next, open the LOCALRES menu, and click on Local Resolution. Select the local half-volumes.

Then, select the local soft-edged mask. Set the overall resolution to the absolute resolution estimated when sharpening the merged local volume. Calculate the local resolution of the volume.

Next, open 3D Local Filter. Set the sharpened unfiltered local volume as the input. Select the local resolution map and soft-edged 3D mask.

Apply the local resolution filter to the sharpened volume to produce the final 3D model. Open the resulting volume and local resolution map in Chimera. Start the Surface Color tool, and choose the volume from the menu and the option to color the surface according to the local resolution.

The toxin complex TcdA1 was imaged by cryo-microscopy. All images showed clearly discernible single particles and Thon rings extending to a resolution of better than four angstroms in the power spectrum. The particles were selected using e2boxer and then were extracted into an image stack.

Two-dimensional class averages were generated by iterative stable alignment and clustering, or ISAC. These class averages were used to calculate a reproducible 3D model with RVIPER. This model agreed well with the previously-solved crystal structure of TcdA1.

Automatic 3D refinement of this model using MERIDIEN generated a near-atomic resolution map. 3D variability analysis indicated localized flexibility of the histidine-tagged N terminal region, the receptor-binding domains, and the TcB-binding domain. Local resolution calculations were performed.

The sharpened model was locally filtered and a color scale corresponding to local resolution was applied, showing topological agreement between areas of high variability and areas of of low resolution. The final 3.5-angstrom resolution cryo-EM structure of TcdA1 showed clear densities for most side chains. A density map with this level of detail can be further used for de novo atomic model building.

Using this approach, a near-atomic resolution structure can be obtained within hours or a few days, without a priori structural information. While attempting this procedure, use only high-quality data and check intermediate results. Compositional and structural heterogeneity of the cryo-EM sample might be a limiting factor for reaching near-atomic resolution.

After watching this video, you should have now a good understanding of how to use SPHIRE for cryo-EM structural determination. As you have seen, the clustering routines included in SPHIRE will allow you not only to solve the structure of the protein of interest, but also to study conformational dynamics and provide crucial mechanistic insight into complex molecular mechanisms.