JoVE Journal > Biochemistry
Summary
Abstract
Introduction
Protocol
Results
Discussion
Disclosures
Acknowledgements
Materials
References
English

Automatically Generated
Biochemistry

A Robust Single-Particle Cryo-Electron Microscopy (cryo-EM) Processing Workflow with cryoSPARC, RELION, and Scipion

Published: January 31, 2022
doi:
10.3791/63387
,
1Institute for Quantitative Biomedicine,Rutgers University, 2Department of Biochemistry & Microbiology,Rutgers University

Summary

This article describes how to effectively utilize three cryo-EM processing platforms, i.e., cryoSPARC v3, RELION-3, and Scipion 3, to create a single and robust workflow applicable to a variety of single-particle data sets for high-resolution structure determination.

Abstract

Recent advances in both instrumentation and image processing software have made single-particle cryo-electron microscopy (cryo-EM) the preferred method for structural biologists to determine high-resolution structures of a wide variety of macromolecules. Multiple software suites are available to new and expert users for image processing and structure calculation, which streamline the same basic workflow: movies acquired by the microscope detectors undergo correction for beam-induced motion and contrast transfer function (CTF) estimation. Next, particle images are selected and extracted from averaged movie frames for iterative 2D and 3D classification, followed by 3D reconstruction, refinement, and validation. Because various software packages employ different algorithms and require varying levels of expertise to operate, the 3D maps they generate often differ in quality and resolution. Thus, users regularly transfer data between a variety of programs for optimal results. This paper provides a guide for users to navigate a workflow across the popular software packages: cryoSPARC v3, RELION-3, and Scipion 3 to obtain a near-atomic resolution structure of the adeno-associated virus (AAV). We first detail an image processing pipeline with cryoSPARC v3, as its efficient algorithms and easy-to-use GUI allow users to quickly arrive at a 3D map. In the next step, we use PyEM and in-house scripts to convert and transfer particle coordinates from the best quality 3D reconstruction obtained in cryoSPARC v3 to RELION-3 and Scipion 3 and recalculate 3D maps. Finally, we outline steps for further refinement and validation of the resultant structures by integrating algorithms from RELION-3 and Scipion 3. In this article, we describe how to effectively utilize three processing platforms to create a single and robust workflow applicable to a variety of data sets for high-resolution structure determination.

Introduction

Cryo-electron microscopy (cryo-EM) and single-particle analysis (SPA) enable structure determination of a wide variety of biomolecular assemblies in their hydrated state, helping to illuminate the roles of these macromolecules in atomic detail. Improvements in microscope optics, computer hardware, and image processing software have made it possible to determine structures of biomolecules at resolution reaching beyond 2 Å1,2,3. More than 2,300 cryo-EM structures were deposited in the Protein Data Bank (PDB) in 2020, compared to 192 structures in 20144, indicating that cryo-EM has become the method of choice for many structural biologists. Here, we describe a workflow combining three different SPA programs for high-resolution structure determination (Figure 1).

The goal of SPA is to reconstruct 3D volumes of a target specimen from noisy 2D images recorded by a microscope detector. Detectors collect images as movies with individual frames of the same field of view. In order to preserve the sample, frames are collected with a low electron dose and thus have a poor signal-to-noise ratio (SNR). Additionally, electron exposure can induce motion within the vitrified cryo-EM grids, resulting in image-blurring. To overcome these issues, frames are aligned to correct for beam-induced motion and averaged to yield a micrograph with an increased SNR. These micrographs then undergo Contrast Transfer Function (CTF) estimation to account for the effects of defocus and aberrations imposed by the microscope. From the CTF-corrected micrographs, individual particles are selected, extracted, and sorted into 2D class averages representing different orientations adopted by the specimen in vitreous ice. The resultant homogeneous set of particles is used as input for ab initio 3D reconstruction to generate a coarse model or models, which are then iteratively refined to produce one or more high-resolution structures. After reconstruction, structural refinements are performed to further improve the quality and resolution of the cryo-EM map. Finally, either an atomic model is directly derived from the map, or the map is fitted with atomic coordinates obtained elsewhere.

Different software packages are available to accomplish the tasks outlined above, including Appion5, cisTEM6, cryoSPARC7, EMAN8, IMAGIC9, RELION10, Scipion11, SPIDER12, Xmipp13, and others. While these programs follow similar processing steps, they employ different algorithms, for example, to pick particles, generate initial models, and refine reconstructions. Additionally, these programs require a varying level of user knowledge and intervention to operate, as some depend on the fine-tuning of parameters that can act as a hurdle for new users. These discrepancies often result in maps with inconsistent quality and resolution across platforms14, prompting many researchers to use multiple software packages to refine and validate results. In this article, we highlight the use of cryoSPARC v3, RELION-3, and Scipion 3 to obtain a high-resolution 3D reconstruction of AAV, a widely used vector for gene therapy15. The aforementioned software packages are free to academic users; cryoSPARC v3 and Scipion 3 require licenses.

Protocol

1. Creating a new cryoSPARC v3 project and importing data

NOTE: Data was acquired at Oregon Health and Science University (OHSU) in Portland using a 300 kV Titan Krios electron microscope equipped with a Falcon 3 direct electron detector. Images were collected in a counting mode with a total dose of 28.38 e−/Å2 fractioned across 129 frames, and a defocus range from -0.5 µm to -2.5 µm, at a pixel size of 1.045 Å using EPU. The sample of AAV-DJ was provided by the staff of OHSU.

  1. Open cryoSPARC v3 in a web browser and click the Projects header. Select + Add to create a new project. Title the project accordingly and provide a path to an existing directory where jobs and data will be saved.
  2. Create a workspace for the project by opening the project, clicking + Add, and selecting New Workspace. Title the workspace and click on Create.
  3. Navigate to the new workspace and open the Job Builder on the right panel. This tab displays all functions available in cryoSPARC v3. Click on Import Movies and provide the movies path, gain reference file path, and set acquisition parameters as follows: Raw Pixel Size 1.045 Å, Accelerating Voltage 300 kV, Spherical Aberration 2.7 mm, Total Exposure Dose 28.38 e–/Å^2.
  4. Click on Queue, select a lane to run the job and a workspace, and click on Create.
    NOTE: The acquisition parameters are sample and microscope dependent.

2. CryoSPARC v3 – movie alignment and CTF estimation

  1. Open Patch Motion Correction (Multi). This job requires the movies imported in step 1.3 as input. Open the import movies job card in the workspace and drag the Imported_movies output to the movies placeholder on the new job. Queue the job.
    NOTE: For more information about the cryoSPARC methods outlined in this article, see the cryoSPARC tutorial16.
  2. To perform CTF estimation, open Patch CTF Estimation (Multi). Input the micrographs generated in step 2.1 and Queue the job.
  3. To inspect the averaged and CTF-corrected micrographs and select a subset for further processing, open Curate Exposures and input the exposures obtained in step 2.2. Queue the job.
  4. After the job enters Waiting mode, click on the Interaction tab on the job card, adjust parameter thresholds, and accept or reject individual micrographs for further processing. Accept micrographs with well-matched estimated and experimental CTFs (Figure 2) and discard those with high astigmatism, poor CTF fit, and thick ice.
  5. While processing the current data, set the upper threshold of Astigmatism to 400 Å, CTF fit resolution to 5 Å, and relative ice thickness to 2. Click on Done to select the micrographs for downstream processing.

3. CryoSPARC v3 – manual and template-based particle picking

  1. Open Manual Picker, input the accepted exposures from steps 2.4-2.5, and Queue the job. Click on the Interactive tab, set the Box Size (px) to 300, and click on a few hundred particles across multiple micrographs and avoid selecting overlapping particles. Here, 340 particles across 29 micrographs were selected. When finished, click on Done Picking! Extract Particles.
    NOTE: This protocol uses manual particle picking to generate templates for automatic selection. However, other methods are also available17.
  2. To generate templates for automated particle picking, click on 2D Classification and input the particle picks generated in step 3.1. Change the number of 2D Classes to 10 and Queue the job.
  3. Open Select 2D classes. Input the particles and class averages obtained in step 3.2 and click on the Interactive tab. Select representative 2D classes with good SNR and click on Done.
    NOTE: The class averages reflect different particle views. Select class averages that reflect each view. The goal is to produce well-defined templates representing different views of the specimen for automated picking.
  4. Open Template Picker and input the 2D classes selected in step 3.3 and micrographs from steps 2.4-2.5. Set the Particle Diameter (Å) to 220 Å and Queue the job.
  5. To inspect the automated picks, open Select Particle Picks, input the particles and micrographs generated in step 3.4, and Queue the job.
  6. On the Select Particle Picks job card, click on the Interactive tab and set the Box size (px) to 300. Click on an individual micrograph, adjust the lowpass filter until particles are clearly visible, and set the Normalized Cross Correlation (NCC) Threshold to 0.41 and Power Threshold between 54000 and 227300 .
  7. Inspect several micrographs and, if needed, adjust thresholds such that most particles are selected without including false positives. When finished, click Done Picking! Extract Particles.
    NOTE: True particles typically have a high NCC and power score, indicating they are similar to the template and have a high SNR, respectively.
  8. Open Extract from Micrographs and input the micrographs and particles from step 3.7. Set the Extracted Box Size (px) to 300 and Queue the job.

4. CryoSPARC v3 – 2D classification

  1. Click on 2D Classification and input the extracted particles from step 3.8. Set the Number of 2D classes to 50 and Queue the job.
  2. To select the best 2D classes for further processing, open Select 2D classes. Input the particles and class averages obtained in step 4.1. Click on the Interactive tab and choose 2D classes based upon the resolution and the number of particles in the class (Figure 3). Do not select classes containing artifacts. After selecting, click on Done.
    NOTE: Usually, multiple rounds of 2D classification are required to remove particles, which do not converge into distinct, well-defined classes. Run as many rounds of 2D classification as needed to remove such particles from the data set (Figure 3).

5. CryoSPARC v3 – ab-initio reconstruction and homogeneous refinement

  1. To generate an initial 3D volume, open Ab-initio Reconstruction and input the particles obtained in step 4.2 or from the final 2D classification. Adjust Symmetry to icosahedral. Queue the job.
    NOTE: Symmetry is sample-dependent and should be changed accordingly. If unknown, use C1 symmetry.
  2. Open Homogeneous Refinement. Input the volume from step 5.1 and particles from 4.2 or the final 2D classification. Change the Symmetry and Queue the job. When the job is finished, inspect the Fourier Shell Correlation (FSC) curve and download the volume to examine in UCSF Chimera18.

6. Exporting particle coordinates from cryoSPARC v3 and importing them to RELION-3 using PyEM

NOTE: Particle coordinates carry information about the location of individual particles in each micrograph. Transfer of coordinates instead of particle stacks to RELION-3 allows for running refinement steps which otherwise would not be available. For example, particle polishing requires access to initial movie frames. Hence, prior to exporting particle coordinates from cryoSPARC v3 to RELION-3, import movies and perform motion correction and CTF estimation in RELION-3. See the RELION-3 tutorial19 for details.

  1. Navigate to the RELION-3 project directory and launch RELION-3.
  2. Open Import from the job-type browser and specify the path to the movies and acquisition parameters as in step 1.3.
  3. To perform motion correction, use UCSF MotionCor220 through the RELION-3 GUI, open Motion Correction and set the default parameters as in the UCSF MotionCor2 manual21. Input the path to the movies imported in step 6.2. On the Motion tab, specify the path to motioncor2 executable.
    NOTE: MotionCor2 can be run in parallel using multiple GPUs.
  4. Perform CTF estimation using CTFFIND-4.122 through the RELION-3 GUI. Open CTF Estimation and input the micrographs.star generated in step 6.3. On the CTFFIND-4.1 tab, specify the path to CTFFIND-4.1 executable and set parameters as in the RELION-3.1 tutorial19.
  5. In order to import particle stacks from cryoSPARC v3 to RELION-3, they first must be exported from cryoSPARC v3. In cryoSPARC v3, open the job card of the Select 2D class job from step 4.2 or the final 2D classification. On the Details tab, click on Export Job. Export job outputs the particles_exported.cs file.
  6. Prior to importing particle coordinates from cryoSPARC v3 to RELION-3, the particles_exported.cs file from step 6.5 must be converted to .star format. Using PyEM23, convert the particles_exported.cs file to .star format by executing the following command: csparc2star.py particles_exported.cs particles_exported.star
  7. In RELION-3, click on the Manual Picking tab and on the I/O tab, input the micrographs from CTF refinement described in step 6.4. On the Display tab, input the following parameters: Particle diameter (A): 220, Lowpass Filter (A): -1 , Scale for CTF image: 0.5. Run the job. A directory called ManualPick is generated in the RELION-3 home folder.
    NOTE: This step is performed to create a manual picking folder structure in RELION-3. While running manual picking, a single .star file containing coordinates of picked particles is created for each averaged micrograph used for picking in the RELION-3 GUI.
  8. Navigate to the folder containing the particles_exported.star file from step 6.6 and run a home-written script producing a single manualpick.star file for each averaged micrograph used for picking of cryo-SPARC v3 particles, which contributed to the final 2D classification exported in step 6.5. The resultant coordinate files are saved in the ManualPick/Movies folder.
  9. Return to RELION-3 and re-open the Manual Picking job. Click on Continue. This will display particles previously picked in cryoSPARC v3 in the RELION-3 GUI. Inspect a few micrographs to verify if the transfer of particle coordinates has been accomplished and if particles are properly selected.

7. RELION-3 – Particle extraction and 2D classification

  1. Click on Particle Extraction. On the I/O tab, input the CTF corrected micrographs from step 6.4 and coordinates from step 6.9. Click on the Extract tab and change the Particle Box Size (pix) to 300. Run the job.
  2. Perform 2D classification to further clean the particle set generated in cryoSPARC v3 to achieve a higher-resolution reconstruction. Click on 2D Classification and on the I/O tab, input the particles.star file generated in step 7.1. On the Optimisation tab, set the Number of Classes to 50 and Mask Diameter (A) to 280. Run the job.
    NOTE: The mask should encompass the entire particle.
  3. To choose the best 2D classes, click on the Subset Selection method, input the _model.star file from step 7.2, and Run the job. Select classes as described in step 4.2.
  4. Repeat steps 7.2 and 7.3 to remove non-converging particles.

8. RELION-3 – 3D refinement, mask creation, and post-processing

  1. Use the map generated in cryoSPARC v3 (step 5.2) as an initial model for 3D refinement in RELION-3. Select the Import method and set the following parameters on the I/O tab: Import Raw Movies/Micrographs: No, Raw Input Files: Movies/*.mrc.
  2. Supply the MTF file and input the movie acquisition parameters as described in step 1.3. On the Others tab, select the cryoSPARC v3 map as the input file, change Node Type to 3D reference (.mrc), and Run the job.
  3. Select 3D Auto-Refine and on the I/O tab, set Input Images as the particles.star file from step 7.3 or the last selection job. Give the cryoSPARC v3 reconstruction as the Reference Map. Click on the Reference tab and change Initial Low-Pass filter (Å) to 50 and Symmetry to icosahedral. On the Optimisation tab, change the Mask Diameter (Å) to 280 and Run the job.
  4. After the run is finished, open run_class001.mrc in UCSF Chimera.
  5. In UCSF Chimera, click on Tools and under Volume Data, select Volume Viewer. This will open a new window to adjust volume settings. Change the Step to 1 and adjust the slider until reaching the level value where the map has no noise. Record this value, as it will be used for mask creation in the next step.
  6. The map produced from auto-refinement does not reflect the true FSC, as noise from the surrounding solvent lowers the resolution. Before post-processing, create a mask to distinguish the specimen from the solvent region.
    1. Click on Mask Creation and input run_class001.mrc from step 8.3.
    2. Click on the Mask tab and adjust parameters as follows: Lowpass Filter Map (Å): 10, Pixel Size (Å): 1.045, Initial Binarization Threshold: the level value obtained in step 8.5, Extend Binary Map this many Pixels: 3, and Add a Soft-Edge of this many Pixels: 3. Run the job.
  7. Examine the mask in UCSF Chimera. If the mask is too tight, increase Extend Binary Map this many Pixels and/or Add a Soft-Edge of this many Pixels. It is important to create a mask with soft edges, as a sharp mask may lead to overfitting.
  8. Click on Post-Processing and on the I/O tab, input the half-maps created in step 8.3 and mask from 8.6. Set Calibrated Pixel Size to 1.045 Å. On the Sharpen tab, input the following: Estimate B-Factor Automatically?: Yes, Lowest Resolution for Auto-B Fit (A): 10, Use Your Own B-Factor?: No. On the Filter tab, set Skip Fsc-Weighting? to No. Run the job.

9. RELION-3 – Polishing training and particle polishing

  1. Before correcting for per-particle beam-induced motion, first use the training mode to identify optimal motion tracks for the data set. Open Bayesian Polishing and on the I/O tab, input the motion-corrected micrographs from step 6.3, particles from step 8.3, and postprocess .star file from step 8.8. Click the Training tab and set the following parameters: Train Optimal Parameters: Yes, Fraction of Fourier Pixels for Testing: 0.5, Use this many Particles: 5000. Run the job.
    NOTE: This script will produce opt_params_all_groups.txt file in the RELION-3 Polish folder containing optimized polishing parameters required for executing the following step.
  2. Once the training job has finished, click on Bayesian Polishing. Click on the Training tab and set Train Optimal Parameters? to No. Select the Polish tab and in Optimised Parameter File specify the path to the opt_params_all_groups.txt file from step 9.1. Click on Run.
  3. Repeat 3D refinement (step 8.3) and post-processing (step 8.8) with a set of polished particles.

10. RELION-3 – CTF and per-particle refinements

  1. To estimate higher order aberrations, open CTF Refinement and, on the I/O tab under Particles, select the path to the .star file containing polished particles from the recent Refine 3D job (run_data.star).
    1. Under Postprocess Star File, set the path to the output from the latest post-processing job (step 9.3).
    2. Select the Fit tab and set the following parameters: Estimate (Anisotropic Magnification): No, Perform CTF Parameter Fitting? No, Estimate Beamtilt: Yes, Also Estimate Trefoil? Yes, Estimate 4th Order Abberations? Yes. Run the job.
  2. Repeat step 10.1 using as input Particles (from Refine3D) generated in the previous job (particles_ctf_refine.star). On the Fit tab, change Estimate (Anisotropic Magnification) to Yes and Run the job.
  3. Repeat step 10.2 using as input Particles (from Refine3D) produced in the previous job (particles_ctf_refine.star). On the Fit tab, set the following parameters: Estimate (Anisotropic Magnification): No, Perform CTF Parameter Fitting?: Yes, Fit Defocus?: Per-particle, Fit Astigmatism? Per-micrograph, Fit B-factor?: No, Fit Phase-Shift: No, Estimate Beamtilt?: No, Estimate 4th Order Aberrations?: No. Run it.
    NOTE: Given the particle has sufficient contrast, the Fit Astigmatism? tab can be set to Per-particle. For this dataset, Per-Particle astigmatism refinement did not improve the quality and resolution of the map.
  4. Repeat 3D refinement with the particles from step 10.3 and on the I/O tab, set Use Solvent-Flattened FSCs? to yes. When finished running, execute a post-processing job (step 8.8) and examine the map in UCSF Chimera (step 5.2).

11. Transferring RELION-3 particle coordinates and 3D map to Scipion 3

  1. To further refine and validate the RELION-3 map, first import the volume and particles from the last post-processing job (step 10.4) to Scipion 3. Launch Scipion 3 and create a new project.
  2. On the left Protocols panel, select the Imports drop-down and click on Import Particles. Change the following parameters: Import From: RELION-3, Star File: postprocess.star, and specify acquisition parameters as in step 1.3. Click on Execute.
  3. Click on the Imports drop-down and select Import Volumes. Under Import From give the path to the RELION-3 map. Change Pixel Size (Sampling Rate) Å/px to 1.045 and Execute.

12. Scipion 3 – High – resolution refinement

  1. First, perform a global alignment. Select the Refine drop-down on the Protocols panel and click on Xmipp3 – highres24. Input the imported particles and volumes from steps 11.2 and 11.3 as Full-Size Images and Initial Volumes, respectively and set the Symmetry Group to icosahedral. On the Image Alignment tab under Angular Assignment, choose Global and set the Max Target Resolution to 3 Å, and Run the job.
  2. When the job is finished, click on Analyze Results. In the new window, click on the UCSF Chimera icon to examine the refined volume. Additionally, click on Display Resolution Plots (FSC) to see how the FSC has changed after the refinement, as well as Plot Histogram with Angular Changes to see if the Euler angle assignments have changed.
    NOTE: Depending on the resolution of the input RELION-3 structure, this step may be repeated several times with different values set for the Max Target Resolution under the Angular Assignment tab. For more information see the Scipion tutorial25.
  3. Repeat step 12.1 with a local alignment. Copy the previous job and change Select Previous Run to Xmipp3 – highres Global. On the Angular Assignment tab, change Image Alignment to Local. Set the Max Target Resolution to 2.1 Å.
  4. Examine the refined map in UCSF Chimera and analyze the change in FSC and angular assignments (step 12.2). Repeat local refinement until resolution does not improve and the angular assignments have converged, adjusting Max Target Resolution as needed.
  5. The output map from Scipion 3 can be additionally density-modified and sharpened in Phenix26.

13. Scipion 3 – Map validation

  1. Examine the local resolution of the final map generated in Xmipp3 – highres. Open Xmipp – local MonoRes27 and input the final volume from the previous job and mask generated in step 8.6. Set Resolution Range from 1 to 6 Å with a 0.1 Å interval and Execute the job.
  2. When finished running, click on Analyze Results and examine the resolution histogram and volume slices colored by resolution.
  3. To see if particles are well-aligned, open Multireference Alignability28 and input the particles and volume from step 12.3. Click on Analyze Results to display the validation plot. Ideally, all points should be clustered around (1.0,1.0).
  4. Open Xmipp3 – Validate Overfitting. Input the particles and volumes from step 12.3. When finished running, analyze results and inspect the overfitting plot. Crossing of the aligned Gaussian noise and aligned particles curves indicates overfitting.

Representative Results

We have presented a comprehensive SPA pipeline to obtain a high-resolution structure using three different processing platforms: cryoSPARC v3, RELION-3, and Scipion 3. Figure 1 and Figure 4 summarize the general processing workflow, and Table 1 details refinement protocols. These protocols were used during refinements of a 2.3 Å structure of AAV, achieving near Nyquist resolution.

Movies were first imported to cryoSPARC v3 and subsequently motion- and CTF-corrected to generate averaged micrographs. When selecting micrographs for further processing, it is important to choose those with a good CTF-fit and low astigmatism (Figure 2), as including poor-quality micrographs can hinder later processing stages, resulting in a lower resolution reconstruction. 27,364 particles were then picked and extracted from the selected micrographs. Because the diameter of the AAV is approximately 220 Å and pixel size is 1.045 Å, a box size of 300 px was used. Next, iterative 2D classification was used to remove artifacts and particles not converging to stable classes. Examples of selected and excluded 2D class averages are presented in Figure 3. It is also important to note that class averages reflecting different conformations of the specimen should be refined separately to yield multiple 3D reconstructions. In such a case, multiple ab initio starting volumes should be calculated. Here, 26,741 particles were selected and used for ab initio modeling and homogeneous refinement of a single 2.9 Å structure.

After transferring coordinates of particles picked in cryoSPARC v3 to RELION-3, we carried out four additional rounds of 2D classification until the data set converged to stable 2D classes. The above-described 2D classification removed an additional 3,154 particles from the data set. Using the structure generated in cryoSPARC v3 as an initial model, 3D refinement in RELION-3 produced a structure with a nearly equivalent resolution of 2.95 Å. Subsequent structural refinements, which included per-particle motion correction and CTF refinements, increased the resolution to 2.61 Å. A complete list of refinements we performed is presented in Table 1. The volume calculated in RELION-3 was then further refined in Scipion 3 using multiple rounds of high-resolution refinement (Xmipp3 – highres). During subsequent rounds of refinement, an additional 3,186 particles were removed from the data set, resulting in a final set of 20,401 particles, which produced a 2.3 Å reconstruction of AAV (Figure 5 and Figure 6). Thus, given the pixel size of 1.045 Å, our refinements have nearly reached the Nyquist limit. FSC curves representing structures calculated using each program are shown in Figure 6. These FSC curves indicate the resolution increase throughout the workflow. Because resolution may vary from point to point in the map, it is often more appropriate to present the distribution of local resolution estimates in the map rather than reporting a single resolution estimate according to a single criterion (e.g., 0.143 criterion) from the FSC curve. Thus, we performed such analysis using Xmipp – MonoRes in Scipion 3. Figure 7 shows a comparison of local resolution estimates for maps obtained with cryoSPARC v3 and Scipion 3. Resolution estimates at four different slices through the structures (Figure 7A,B) and resolution histograms (Figure 7C) clearly demonstrate the incremental improvement in local resolution between the maps throughout the workflow. The FSC curve calculated using the program Xmipp3 – highres in Scipion 3 indicates the Nyquist limit has been reached (Figure 6), suggesting the resolution estimate is very likely limited by undersampling29. However, MonoRes analysis presented in Figure 7C, along with a careful analysis of the EM map and map fitting with atomic coordinates of AAV (Figure 5) suggest that a more adequate resolution estimate for the map is 2.3 Å. A similar strategy reconciling the FSC and MonoRes resolution estimates have been presented earlier24,25. Because resolution estimates can be influenced by the mask used during refinement steps, it is important to ensure the mask does not exclude any part of the density. The mask used in this study overlapped with the 3D reconstructions is presented in Figure 7D. The gradual increase in resolution in the presented workflow highlights the advantage of utilizing algorithms from multiple SPA software packages to achieve a high-quality and high-resolution 3D reconstruction.

In-situ model building or fitting the map with a pre-existing atomic model can serve as the quality check for the calculated structure. We have visualized the final map in UCSF Chimera and fitted the map with a previously published atomic model (PDB ID: 7kfr)30. Figure 5 shows regions of the cryo-EM map fitted with atomic coordinates of AAV. Well-defined EM densities allow for fitting side-chains of individual amino acids, water molecules, and magnesium ions and confirm the agreement of the cryo-EM map with the atomic model.

Figure 1
Figure 1: Complete SPA workflow across cryoSPARC v3, RELION-3, Scipion 3, and Phenix 1.18. Steps completed in cryoSPARC v3, RELION-3, Scipion 3, and Phenix 1.18 are denoted with purple, orange, green, and grey boxes, respectively. The time required for completion of each step using the processing server equipped with 8 GPUs, 40 CPUs and 750 GB of RAM is specified in each individual box. Please click here to view a larger version of this figure.

Figure 2
Figure 2: Selection of micrographs for downstream processing in cryoSPARC v3. (A) Micrographs with well-matched estimated and experimental Thon rings were used for further processing, while those with high astigmatism and poor fit (B) were discarded. Micrographs with CTF-fit above 5 Å, astigmatism over 400 Å, and relative ice thickness below 2 were removed from further processing, i.e., 70/395 micrographs. Please click here to view a larger version of this figure.

Figure 3
Figure 3: Selecting 2D classes. 2D class averages containing well-defined classes are selected (A), and those with low-resolution, noise, and partial particles are rejected (B). Please click here to view a larger version of this figure.

Figure 4
Figure 4: Workflow and representative results for AAV processing across cryoSPARC v3, RELION-3, and Scipion 3. Steps completed in cryoSPARC v3, RELION-3, and Scipion 3 are denoted with purple, orange, and green arrows, respectively. Please click here to view a larger version of this figure.

Figure 5
Figure 5: High-resolution structure of AAV shows well-defined EM densities representing different secondary structure elements and individual amino acid side chains. (A) A final map of AAV. (B) A part of the map representing beta sheets fitted with atomic coordinates of AAV (PDB ID: 7kfr)30. (C) Map densities representing individual amino acids. From left to right: arginine, phenylalanine, and tryptophan. (D) High-resolution features of the map include water molecules presented in red and magnesium ions presented as green spheres. Mg2+ ion displayed in the figure is coordinated by histidine (left) and arginine residues. Please click here to view a larger version of this figure.

Figure 6
Figure 6: FSC curves from cryoSPARC v3, RELION-3, and Scipion 3 show increasing resolution across the workflow. While the FSC curve calculated using the program Xmipp3 – highres in Scipion 3 indicates the Nyquist limit has been reached, suggesting the resolution estimate is limited by undersampling29, a more adequate analysis of the map resolution is presented in Figure 7 and discussed in the Representative Results section24,25. Please click here to view a larger version of this figure.

Figure 7
Figure 7: Validating the final reconstruction in Scipion 3 using Xmipp – MonoRes. Resolution of the map is better described by presenting local resolution distributions rather than a single resolution estimate according to a single criterion from the FSC curve. (A–B) Panels A and B show different slices from maps generated in cryoSPARC v3 and Scipion 3, respectively. (C) Histograms demonstrating a systematic increase in local resolution for maps calculated in cryoSPARC v3 (pink bars) and Scipion 3 (blue bars). (D) The mask (gray) used for local resolution calculations contain all parts of the AAV densities refined in both programs. Please click here to view a larger version of this figure.

Program Refinement type Script
cryoSPARC v3 Homogeneous Refinement Homogeneous Refinement
Non-uniform Refinement Nonuniform Refinement
Heterogeneous Refinement Heterogeneous Refinement
Per-particle motion correction Local Motion Correction
RELION-3 3D refinement Refine3D
Postprocessing – B-factor Sharpening, MTF Correction Refine3D
Particle Polishing Bayesian Polishing
CTF refinement – beam tilt CtfRefine
CTF refinement – anisotropic magnification CtfRefine
CTF refinement – per-particle defocus, per-particle/micrograph astigmatism CtfRefine
Ewald sphere curvature correction Relion_reconstruct
Scipion 3 High resolution refinement Xmipp3 – highres
Phenix 1.18 Density Modification and Sharpening ResolveCryoEM

Table 1: Refinements implemented throughout the workflow. Whether certain refinements are applicable to a specific project depends on data quality and acquisition parameters. For example, the Ewald sphere curvature correction can be applied for maps that already have high resolution.

Discussion

In this article, we present a robust SPA workflow for cryo-EM data processing across various software platforms to achieve high-resolution 3D reconstructions (Figure 1). This workflow is applicable to a wide variety of biological macromolecules. The subsequent steps of the protocol are outlined in Figure 4, including movie pre-processing, particle picking and classification, and multiple methods for structure refinements (Table 1) and validation. Processing steps in cryoSPARC v3, RELION-3, and Scipion 3 have been presented, as well as methods for transferring data between the software packages. We have shown the intermediate structures obtained throughout the protocol with increasing resolution (Figure 4, Figure 6 and Figure 7).

While the methods outlined in this manuscript can be used for structure determination of different proteins and biological assemblies, it is important to note that AAV is an ideal candidate for high-resolution structure determination by cryo-EM and SPA, as its large size produces high contrast in the microscope and icosahedral symmetry yields particles with 60-fold subunit redundancy. Obtaining high-resolution reconstructions become increasingly difficult for small (i.e., less than 100 kD), dynamic, and heterogeneous samples. In order to successfully execute this protocol, it is critical that many high-quality movies are collected for processing. With poor-quality raw data, obtaining a high-resolution and high-quality reconstruction is not possible. For instance, if ice thickness is not optimal for structure determination or if particles adhere to the ice-water interface or exhibit preferred orientations, revisit grid freezing conditions.

Another important step in the workflow is particle picking and extraction. During particle picking, the box size should be approximately 1.4-2.5 times larger than the longest axis of the particle, as sufficient box size is required to capture high-resolution information spread out due to defocus. Larger box sizes, however, require longer processing times due to the increased size of files generated during particle extraction. When choosing a box size, consider particle diameter and pixel size. With many particles, the user may want to bin particles during extraction for initial processing and then re-extract full size particles for final refinements. This protocol uses manual particle picking to generate templates for automatic selection. However, cryoSPARC v3 also offers fully automated picking methods, including a blob-based picker and a Topaz wrapper, which utilizes deep learning to select particles based on previous picks. While these algorithms are very robust, a significant number of picks would need to be later removed by 2D and 3D classification.

Critical steps also include 2D and 3D classification used to remove artifacts, such as radiation-damaged particles and separate different structural forms of the specimen present in the sample, respectively. The number of 2D classes set by the user should depend upon the number of particles extracted from micrographs, contrast, and heterogeneity in the specimen, as the goal is to separate each individual view of the particle into a separate 2D class. As a general rule, add a 2D class for every 100 particles, and if processing a new sample with a large number of particles, 100 classes is a good starting point. If a low or moderate resolution is obtained even after many rounds of 2D and 3D classification, try re-extracting particles with a larger box size to see if more structural information can be obtained. While reporting the resolution of the final reconstruction, one should analyze the FSC curve obtained according to the gold standard method, along with the local resolution estimates and careful inspection of the map densities, as well as their agreement with the atomic model.

For refinements of the virus structures, the Ewald sphere curvature correction implemented in RELION-3 has demonstrated improvements at high resolution31. If refining structures of dynamic multiprotein complexes, try 3D multi-body refinement implemented in RELION-3 or focused classification with image subtraction implemented in RELION-3 and cryoSPARC v3. If heterogeneity cannot be resolved computationally, it is necessary to revisit sample preparation conditions32. Insufficient purity or inadequate preparation resulting in protein degradation will hinder the quality of the 3D reconstruction. Additionally, buffer conditions that destabilize proteins or promote aggregation severely limit the number of well-defined particles that can be used for structure calculation. Thus, to most effectively utilize the methods presented here, it is imperative to identify optimal conditions for sample stability. We recommend negative-staining electron microscopy to screen the samples prior to cryo-EM.

As cryo-EM has become the preferred method for 3D structure determination for an increasing number of structural biologists, the need for an integrative and robust workflow for image processing and structure determination becomes more apparent. CryoSPARC offers an easy-to-use, web-based graphic user interface (GUI) that allows users of all experience levels to quickly process data and calculate a 3D structure. Notably, CryoSPARC utilizes a stochastic gradient descent to perform ab initio 3D reconstruction. Furthermore, the software employs a branch and bound likelihood optimization algorithm for quick 3D map refinement7. The processing pipeline described in this article uses cryoSPARC v3 to yield an initial 3D map. The 3D reconstruction is then refined in RELION-3, a popular package that uses an empirical Bayesian approach to estimate critical parameters based on the user's data set, thereby reducing the need for expert knowledge for program operation10. Specifically, we utilize Bayesian polishing for per-particle motion correction and CTF refinements to improve resolution. Finally, the resultant structure is further-refined and validated in Scipion 311, an integrative Python shell that supports algorithms from multiple platforms, including Xmipp13, EMAN28, SPIDER12, and others. While many different software packages are available for cryo-EM users, there is currently no universal SPA platform accepted by the field. Although the SPA workflow can be fully executed in any of the three software packages described in this article, different algorithms may yield varying results. Consequently, individual steps must be customized depending on the sample and quality of the data. For example, for the current data set, 3Drefine in RELION-3 increased the resolution of the 3D reconstruction, while Nonuniform refinement in cryoSPARC v3 led to a slight resolution decrease. Thus, it is greatly beneficial to utilize a variety of programs to recalculate structures to achieve optimal quality and resolution and to facilitate validation of the reconstructions. Although, Scipion 3 contains numerous algorithms from cryoSPARC v3 and RELION-3, the most recent implementations of these programs are not immediately available in Scipion. For instance, of the programs utilized in this manuscript, only RELION-3 offers Ewald Sphere Curvature Correction through the script Relion_reconstruct. The pipeline presented in this article provides a guide for both new and experienced users to successfully use algorithms implemented in cryoSPARC v3, RELION-3, and Scipion 3 to calculate 3D structures at near-atomic resolution.

Disclosures

The authors have nothing to disclose.

Acknowledgements

We thank Carlos Oscar Sorzano for help with Scipion3 installation and Kilian Schnelle and Arne Moeller for help with data transfer between different processing platforms. A portion of this research was supported by NIH grant U24GM129547 and performed at the PNCC at OHSU and accessed through EMSL (grid.436923.9), a DOE Office of Science User Facility sponsored by the Office of Biological and Environmental Research. This study was supported by a start-up grant from Rutgers University to Arek Kulczyk.

Materials

CryoSPARC Structura Biotechnology Inc. https://cryosparc.com/
CTFFIND 4 Howard Hughes Medical Institute, UMass Chan Medical School https://grigoriefflab.umassmed.edu/ctffind4
MotionCorr2 UCSF Macromolecular Structure Group https://msg.ucsf.edu/software
Phenix Computational Tools for Macromolecular Neutron Crystallography (MNC) http://www.phenix-online.org/
PyEM Univerisity of California, San Francisco https://github.com/asarnow/pyem
RELION MRC Laboratory of Structural Biology https://www3.mrc-lmb.cam.ac.uk/relion/index.php/Main_Page
Scipion Instruct Image Processing Center (I2PC), SciLifeLab http://scipion.i2pc.es/
UCSF Chimera UCSF Resource for Biocomputing, Visualization, and Informatics https://www.cgl.ucsf.edu/chimera/
DOWNLOAD MATERIALS LIST

References

  1. Bartesaghi, A., et al. Atomic resolution Cryo-EM structure of beta-galactosidase. Structure. 26 (6), 848-856 (2018).
  2. Merk, A., et al. Breaking Cryo-EM resolution barriers to facilitate drug discovery. Cell. 165 (7), 1698-1707 (2016).
  3. Wardell, M., et al. The atomic structure of human methemalbumin at 1.9 A. Biochemical and Biophysical Research Communications. 291 (4), 813-819 (2002).
  4. PDB statistics: Growth of Structures from 3DEM Experiments Released per Year. RCSB PDB Available from: https://www.rcsb.org/stats/growth/growth-em (2021)
  5. Lander, G. C., et al. Appion: an integrated, database-driven pipeline to facilitate EM image processing. Journal of Structural Biology. 166 (1), 95-102 (2009).
  6. Grant, T., Rohou, A., Grigorieff, N. cisTEM, user-friendly software for single-particle image processing. Elife. 7, 35383 (2018).
  7. Punjani, A., Rubinstein, J. L., Fleet, D. J., Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nature Methods. 14 (3), 290-296 (2017).
  8. Tang, G., et al. EMAN2: an extensible image processing suite for electron microscopy. Journal of Structural Biology. 157 (1), 38-46 (2007).
  9. van Heel, M., Harauz, G., Orlova, E. V., Schmidt, R., Schatz, M. A new generation of the IMAGIC image processing system. Journal of Structural Biology. 116 (1), 17-24 (1996).
  10. Scheres, S. H. RELION: Implementation of a Bayesian approach to cryo-EM structure determination. Journal of Structural Biology. 180 (3), 519-530 (2012).
  11. de la Rosa-Trevin, J. M., et al. Scipion: A software framework toward integration, reproducibility and validation in 3D electron microscopy. Journal of Structural Biology. 195 (1), 93-99 (2016).
  12. Shaikh, T. R., et al. SPIDER image processing for single-particle reconstruction of biological macromolecules from electron micrographs. Nature Protocols. 3 (12), 1941-1974 (2008).
  13. Sorzano, C. O., et al. XMIPP: a new generation of an open-source image processing package for electron microscopy. Journal of Structural Biology. 148 (2), 194-204 (2004).
  14. Lawson, C. L., Chiu, W. Comparing cryo-EM structures. Journal of Structural Biology. 204 (3), 523-526 (2018).
  15. Naso, M. F., Tomkowicz, B., Perry, W. L., Strohl, W. R. Adeno-associated virus (AAV) as a vector for gene therapy. BioDrugs. 31 (4), 317-334 (2017).
  16. Cryo-EM data processing in cryoSPARC: Introductory Tutorial. at Available from: https://guide.cryosparc.com/processing-data/cryo-em-data-processing-in-cryosparc-introductory-tutorial (2020)
  17. Bepler, T., Noble, A. J., Berger, B. Topaz-Denoise: general deep denoising models for cryoEM and cryoET. Nature Communications. 11 (5208), (2020).
  18. Pettersen, E. F., et al. UCSF Chimera-a visualization system for exploratory research and analysis. Journal of Computational Chemistry. 25 (13), 1605-1612 (2004).
  19. . Single-particle processing in RELION-3.1 Available from: https://hpc.nih.gov/apps/RELION/relion31_tutorial.pdf (2019)
  20. Zheng, S. Q., Palovcak, E., Armache, J. P., Verba, K. A., Cheng, Y., Agard, D. A. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nature Methods. 14 (4), 331-332 (2017).
  21. . MotionCor2 User Manual Available from: https://hpc.nih.gov/apps/RELION/MotionCor2-UserManual-05-03-2018.pdf (2018)
  22. Rohou, A., Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. Journal of Structural Biology. 192 (2), 216-221 (2015).
  23. . UCSF PyEM v0.5 Available from: https://github.com/asarnow/pyem (2019)
  24. Sorzano, C. O. S., et al. A new algorithm for high-resolution reconstruction of single particles by electron microscopy. Journal of Structural Biology. 204 (2), 329-337 (2018).
  25. Jimenez-Moreno, A., et al. Cryo-EM and single-particle analysis with Scipion. Journal of Visualized Experiments: JoVE. (171), e62261 (2021).
  26. Adams, P. D., et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallographica Section D Biological Crystallography. 66, 213-221 (2010).
  27. Vilas, J. L., et al. MonoRes: Automatic and accurate estimation of local resolution for electron microscopy maps. Structure. 26 (2), 337-344 (2018).
  28. Sorzano, C. O., et al. A clustering approach to multireference alignment of single-particle projections in electron microscopy. Journal of Structural Biology. 171 (2), 197-206 (2010).
  29. Penczek, P. A. Resolution measures in molecular electron microscopy. Methods in Enzymology. 482, 73-100 (2010).
  30. Xie, Q., Yoshioka, C. K., Chapman, M. S. Adeno-associated virus (AAV-DJ)-Cryo-EM structure at 1.56 A Resolution. Viruses. 12 (10), 1194 (2020).
  31. Zivanov, J., et al. New tools for automated high-resolution cryo-EM structure determination in RELION-3. Elife. 7, 42166 (2018).
  32. Kulczyk, A. W., Moeller, A., Meyer, P., Sliz, P., Richardson, C. C. Cryo-EM structure of the replisome reveals multiple interaction coordinating DNA synthesis. Proceedings of the National Academy of Sciences of the United States of America. 114 (10), 1848-1856 (2017).

Tags

Cryo-EMSingle-particle AnalysisCryoSPARCRELIONScipionAdeno-associated VirusGene TherapyPatch Motion CorrectionCTF EstimationMicrograph CurationManual Particle PickingTemplate GenerationParticle Extraction

Play Video
PDF
DOI
DOWNLOAD MATERIALS LIST
Cite This Article
DiIorio, M. C., Kulczyk, A. W. A Robust Single-Particle Cryo-Electron Microscopy (cryo-EM) Processing Workflow with cryoSPARC, RELION, and Scipion. J. Vis. Exp. (179), e63387, doi:10.3791/63387 (2022).
Download Citation
Reprints and Permissions

View Video

✖

To prove you're not a robot, please enter the text in the image below

CAPTCHA Image
Play CAPTCHA Audio
Refresh Image