A Workflow for Visualizing Protein Conformational Dynamics from Cryo-EM Maps Using Human Asparagine Synthetase

Summary

This protocol outlines how to perform 3D variability analysis (3DVA) in CryoSPARC and variability refinement in Phenix to analyze conformational heterogeneity in cryo-EM data. Using human asparagine synthetase as an example, it outlines how to generate variability maps, visualize principal components, and refine corresponding atomic models.

Abstract

This article presents a comprehensive workflow for analyzing conformational heterogeneity in single-particle cryo-electron microscopy (cryo-EM) data using 3D variability analysis (3DVA) in CryoSPARC and variability refinement in Phenix. The protocol describes how to compute variability maps along principal components derived from particle images, visualize principal modes of structural variation, and refine corresponding ensembles of atomic models against individual variability frames. The resulting outputs include a series of 3D variability maps, refined multi-model atomic structures representing discrete conformational states, and visualization files suitable for comparative structural analysis and movie generation. This workflow is broadly applicable to cryo-EM reconstructions in the resolution range (typically ~2.0-6.0 Å) where conformational heterogeneity can be meaningfully interpreted at the backbone and side-chain levels. It is particularly suited for capturing continuous domain motions, hinge-bending movements, and coordinated local rearrangements that are not easily resolved by discrete classification alone. The combined use of variability maps and refined model ensembles enables direct structural comparison of conformational extremes, supports quantitative analyses such as root-mean-square fluctuation (RMSF)  calculations, and provides a framework for linking dynamic structural features to biochemical or functional hypotheses. Human asparagine synthetase (ASNS) is used as an example to demonstrate the practical implementation of the workflow.

Introduction

Characterizing conformational heterogeneity in macromolecular structures is an increasingly important aspect of single-particle cryo-electron microscopy (cryo-EM) analysis. The goal of this protocol is to provide a practical workflow for extracting, visualizing, and refining dominant modes of structural variability directly from cryo-EM datasets using three-dimensional variability analysis (3DVA) in CryoSPARC¹ combined with variability refinement in Phenix². This article is intended to guide readers in determining whether this computational approach is appropriate for their data and in understanding the types of outputs generated by the workflow.

Traditional experimental techniques (e.g., hydrogen-deuterium exchange, NMR spectroscopy, mass spectrometry) provide valuable information on protein motions^3,4,5 but are often limited by molecular size, sample requirements, or experimental complexity. Advances in cryo-EM have enabled high-resolution structure determination for large and heterogeneous systems, allowing analysis of conformational variability within particle populations. 3D variability analysis (3DVA), implemented in CryoSPARC¹, decomposes cryo-EM particle datasets into principal components that describe dominant, continuous modes of structural variation. When combined with variability refinement in Phenix², this approach produces atomic models corresponding to discrete positions along each principal component, enabling visualization and comparison of alternative conformations derived from experimental data.

This protocol is best suited for cryo-EM datasets reconstructed at near-atomic to intermediate resolution (typically ~2-6 Å). At higher resolutions, the workflow can capture backbone rearrangements and limited side-chain variability⁶, whereas at lower resolutions it is more appropriate for identifying domain-level or secondary-structure movements^2,7. The method is suited to capture dominant, continuous modes of motion that are broadly populated in the dataset, such as hinge motions, domain breathing, or coordinated rearrangements between domains or subunits. Adequate particle numbers and angular coverage are essential, and interpretation may be limited in flexible or weakly resolved regions where side-chain positions remain ambiguous. Compared with focused classification⁸ or extensive subclassification strategies, the 3D variability refinement workflow represents conformational heterogeneity as continuous trajectories rather than discrete classes. This makes the method particularly useful for the exploratory analysis of enzymes, molecular machines, and multi-domain complexes that are expected to sample multiple conformations.

Human asparagine synthetase (ASNS) is used as an illustrative example to demonstrate the protocol's implementation. ASNS is a glutamine-dependent amidotransferase composed of two functional domains connected by an intramolecular ammonia tunnel^9,10. Previous studies have suggested that conformational changes involving residues such as Arg-142 contribute to the modulation of this tunnel⁶. In this article, ASNS serves solely as a representative system for demonstrating how 3DVA and variability refinement generate variability maps, refined atomic model ensembles, and visualizations. Detailed mechanistic interpretation is deferred to the Results and Discussion sections. Although molecular dynamics (MD) simulations are often used as a complementary approach to explore conformational landscapes^11,12, MD is not required for this protocol. Comparisons with MD simulations may be performed as an optional downstream analysis, but the workflow described here (Figure 1) focuses exclusively on extracting and visualizing conformational heterogeneity from cryo-EM data.

Figure 1. Workflow overview. A consensus cryo-EM map is reconstructed from single-particle data and subjected to three-dimensional variability analysis (3DVA) to generate variability maps along principal components. Variability refinement (Phenix.varref) uses the consensus atomic model to generate corresponding atomic models for visualization and comparison. Please click here to view a larger version of this figure.

Protocol

1. Data preparation

1. Generation of consensus EM map
   1. Use single-particle cryo-electron microscopy (cryo-EM) to determine the consensus 3D structure of the macromolecular complex.
   2. Vitrify samples on cryo-grids and image using a high-voltage transmission electron microscope equipped with a direct electron detector.
   3. Process the collected movies to correct beam-induced motion, estimate imaging parameters, and select particle images.
   4. Perform iterative classification and refinement of the particles to generate a consensus EM map.
   5. Perform data processing using CryoSPARC (v4 or higher)¹³, since 3D variability analysis (3DVA) will be implemented.
2. Generate a consensus model
   1. Build an atomic model and refine it against the consensus map. Validate the refined model quality using standard criteria
      NOTE: For ASNS, the structure was determined using a template-based particle-picking strategy. Template derived from the X-ray structure of DON-modified human ASNS (PDB 6GQ3)⁹was used to efficiently select high-quality particles in CryoSPARC¹³. These particles underwent iterative rounds of 2D classification, Ab initio reconstruction, and heterogeneous and non-uniform refinement with C2 symmetry imposed, yielding a 3.5 Å consensus map . The atomic model was generated by using the crystal structure⁹ into the EM map, followed by real-space refinement in Phenix^14,15 and manual adjustment in Coot¹⁶, producing the final consensus ASNS structure as described⁶.

2. Performing 3D variability analysis (3DVA)

1. Use the particle stack and mask (not a final refined map) for a final refinement
   NOTE: Often, the results from a non-uniform refinement job will be used for 3DVA.
2. Create a 3D Variability job in CryoSPARC. Input particle stack and mask from the non-uniform refinement job.
3. Set Number of modes to solve: number of principal components (PCs) to 5.
   NOTE: Because macromolecular structures exist in three-dimensional space, at least three PCs are generally required to describe their major spatial modes of motion. To ensure that additional, potentially smaller but functionally relevant conformational variations were not overlooked, two more PCs were included.
4. Set Filter resolution to 4.0 Å, and Highpass resolution to 20. Leave all other parameters as default. Start the job.
   NOTE: "Filter resolution" of 4.0 Å was set to match the overall quality of the dataset while avoiding overfitting to high-frequency noise. Since the final consensus map was determined at 3.5 Å resolution , applying a slightly lower filter resolution (4.0 Å) ensured that the variability analysis focused on genuine structural movements rather than noise near the resolution limit.
5. Obtain a set of variability volumes as output (Figure 2A)

3. Performing 3D variability display

1. Create a 3D Variability Display job in CryoSPARC to visualize the continuous conformational changes identified by the 3DVA.
2. Input the particle stack and the variability volumes generated from the 3DVA job.
3. Select Simple as output mode (see below).
4. Set Filter resolution to 4.0 and set Highpass resolution to 20. Leave all other parameters at their default values. Start the job.
5. Obtain 20 variability maps as outcome (frame 0 to 19) corresponding to 5 Principal components (PC) from 0 to 4 (Figure 2B)
   NOTE: The "simple mode" in the 3D Variability Display program provides a straightforward visualization of conformational changes along each principal component. This mode generates linear trajectories between the extreme states identified by 3DVA, enabling a clear, direct interpretation of the motions within the particle population.

4. Data preparation for movies and Phenix 3D variable refinement

1. Download 3DVA variability maps: Download the zip files containing variability maps for PCs 0-4 (five zips total) from CryoSPARC.
2. Organize directories: Create a parent folder (e.g., 3DVA/). Inside, create one subfolder for each principal component (comp0, comp1, comp2, comp3, comp4).
3. Move each zip into its corresponding folder and unzip the contents.
4. Confirm that each folder contains 20 variability maps (frames 0-19) and repeat for comp1-comp4.
   NOTE: Maintain separate folders for each principal component to prevent mixing frames across PCs.

5. Prepare quick movies in UCSF Chimera (Optional) ¹⁷

1. Open the Volume Series tool: Launch Chimera, then navigate to Tools → Volume Data → Volume Series . A new Volume Series window will open.
2. Load the maps: Click Open and select the 20 maps corresponding to one principal component (e.g., comp0).
3. Adjust visualization parameters: Open Volume Viewer, set the appropriate density threshold, and assign a color to the map.
   NOTE: In this example, sky blue was used (Figure 3).
4. Preview motion: In the Volume Series window, set Play direction to Oscillate, then click Play to display a continuous transition from frame 0 → 19 → 0.
5. Repeat for additional components: Repeat Steps 5.2-5.4 for other principal components as needed.
6. Generate a movie file using command-line input: Export a movie by opening the Chimera command line and running Supplementary Coding File 1.
   NOTE: The per-component, 20-frame mrc series (e.g., compX_frame000-019.mrc) are the inputs for Phenix 3D variable refinement. Keep component folders separate (comp0-comp4) to avoid mixing frames across PCs.

6. Model generation by variability refinement in phenix

NOTE: Phenix.varref can be executed from any directory as long as the full paths to the input maps and model are specified. However, for organizational clarity, it is recommended to perform the refinement in a dedicated subdirectory for each PC.

1. Copy the final refined consensus model file into the subdirectory corresponding to each PC.
2. From within each subdirectory, run Phenix.varref from the command line using the following syntax:
   phenix.varref consensus_model.pdb *frame_*.mrc resolution=4 nproc=20 models_per_map=50
3. Set Resolution to 4.0 Å to match the effective resolution of the 3DVA maps
4. Set nproc to the number of processors available (20 in this example, but this can vary depending on available computational resources)
5. Set models_per_map to 50 models (default).
   NOTE: For each map, 50 models are generated with restraints adjusted to the resolution of the 3DVA data. The best-fitting model for each map is automatically retained. The resulting output file contains 20 refined models corresponding to the 20 variability maps. The output consists of a PDB bundle containing 20 refined models, along with individual map files.
6. Do not apply additional map sharpening during variability refinement. Use the variability maps generated by CryoSPARC (filtered at 4.0 Å) directly as inputs to phenix.varref.

7. Visualization

1. Open the PDB bundle in ChimeraX¹⁸ to load all 20 refined models simultaneously.
2. Use the Models panel to selectively display individual frames or specific models of interest for comparison and analysis (Figure 4). Alternatively, each refined model can also be opened individually in ChimeraX or Chimera.

8. Data preparation for analysis

NOTE: Running the syntax above generates a PDB bundle containing 20 variable models and each individual map. For most analyses, only two models-corresponding to frame 0 and frame 19-are required, as they represent the two extreme conformational states along the selected principal component (PC).

1. Open the PDB bundle in ChimeraX, select model 1.1 (frame 0) and model 1.20 (frame 19), and save each model as a separate PDB file (Figure 4).
2. Edit the resulting PDB files to isolate residues of interest. This can be done manually or using Phenix.pdb.tools.
3. Compare the two modified PDB files to identify conformational changes in backbone or side chains.
4. To compare the two PDB models, superimpose the modified structures using the MatchMaker function in Chimera.
   NOTE: For reliable alignment, MatchMaker typically requires at least 10-15 overlapping residues. In our case, only 6 residues were analyzed; however, successful superposition was achieved by leveraging the homodimeric nature of ASNS-while only 6 residues remained in chain A, chain B was intact and served as the reference for alignment.

Results

This section presents representative outputs generated by the 3D variability analysis (3DVA) and variability refinement workflow and provides guidance on their interpretation. Human asparagine synthetase (ASNS) is used as an illustrative dataset to demonstrate successful outcomes, common artifacts, and comparative analyses enabled by the protocol. The complete set of 3DVA-derived variability maps and the corresponding refined models generated from the ASNS datasets have been deposited in the Zenodo public repository. Det...

Discussion

Critical protocol steps and decision points
Effective use of this protocol depends on several key steps in the workflow. The quality of the particle stack and mask used for final cryo-EM refinement is critical, as 3D variability analysis (3DVA) operates directly on these inputs rather than on a post-processed map. Selection of the number of principal components (PCs) should reflect dataset size and expected complexity; too few components may underrepresent variability, whereas too many may introduc...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CryoSPARC v4.4+	Structura Biotechnology	https://cryosparc.com	Software for cryo-EM processing
Phenix v1.24+	Phenix Consortium	https://phenix-online.org	Software for Variability refinement
UCSF Chimera	Resource for Biocomputing, Visualization, and Informatics, UCSF	https://www.cgl.ucsf.edu/chimera/	Software for Molecular Visualization
UCSF ChimeraX	Resource for Biocomputing, Visualization, and Informatics, UCSF	https://www.cgl.ucsf.edu/chimerax/	Software for Molecular Visualization
CPPTRAJ	AMBER molecular dynamics software suite	https://amberhub.chpc.utah.edu/cpptraj/	Python tools for RMSF calculation
Linux Workstation (in-house assembled GPU-based system)
Workstation Case	Supermicro	https://www.supermicro.com	Workstation case: Supermicro CSE-743TQ-1200B-SQ 1200W 4U Server Super Chassis
CPU	Intel	https://www.intel.com	CPU: XEON E5-2690V3 x2
Mother Board	Supermicro	https://www.supermicro.com	Mother Board: Supermicro Extended ATX DDR4 LGA 2011 Motherboard X10DAI-O
Memory	Crucial	https://www.crucial.com	Memory: Crucial 64GB Kit (16GBx4) DDR4 2133 (PC4-2133) DR x4 ECC Registered. 288-Pin Server Memory CT4K16G4RFD4213 / CT4C16G4RFD4213 x2
Hard drive	Western Digital	https://www.westerndigital.com	HDD: WD RE 4TB Enterprise Hard Drive
RAID CONTROLLER	LSI Logic	https://www.broadcom.com	RAID CONTROLLER: LSI LOGIC MegaRAID SAS 9271-8i Kit
GPU	NVIDIA	https://www.nvidia.com	Quadro M6000 graphics card with 24 GB GDDR5 memory