Optimizing Single-Particle Analysis Workflow: Comparative Analysis of the Symmetry Parameter and Particle Quantity upon Reconstruction of the Molecular Complex

Abstract

Recent major advancements in cryo-electron microscopy (cryo-EM) have enabled high-resolution structural analysis, accompanied by developments in image processing software packages for single-particle analysis (SPA). SPA facilitates the 3D reconstruction of proteins and macromolecular complexes from numerous individual particles. In this study, we systematically evaluated the impact of symmetry parameters and particle quantity on the 3D reconstruction efficiency using the dihydrolipoyl acetyltransferase (E2) inner core of the pyruvate dehydrogenase complex (PDC). We specifically examined how inappropriate symmetry constraints can introduce structural artifacts and distortions, underscoring the necessity for accurate symmetry determination through rigorous validation methods such as directional Fourier shell correlation (FSC) and local-resolution mapping. Additionally, our analysis demonstrates that efficient reconstructions can be achieved with a moderate particle number, significantly reducing computational costs without compromising structural accuracy. We further contextualize these results by discussing recent developments in SPA workflows and hardware optimization, highlighting their roles in enhancing reconstruction accuracy and computational efficiency. Overall, our comprehensive benchmarking provides strategic insights that will facilitate the optimization of SPA experiments, particularly in resource-limited settings, and offers practical guidelines for accurately determining symmetry and particle quantity during cryo-EM data processing.

1. Introduction

Cryo-electron microscopy (cryo-EM), which has undergone remarkable advancements in recent years, has become the most acclaimed method in the field of structural biology [1,2]. Whereas X-Ray crystallography remains the gold standard for determining atomic-resolution structures of well-ordered crystalline protein [3], and nuclear magnetic resonance (NMR) excels in resolving the dynamics of relatively small biomolecules in solution, cryo-EM has emerged as a powerful complementary technique by addressing specific challenges associated with large, heterogeneous, or membrane-bound complexes [4,5,6]. Particularly, recent breakthroughs such as direct electron detectors, automated data acquisition software, and sophisticated computational algorithms have dramatically improved resolution, allowing near-atomic to atomic resolutions to become increasingly routine [2,7,8]. Traditional methods clearly have limited ability to elucidate the structures of membrane proteins, protein complexes, viruses, macromolecules, and so on [9,10]. Rather than fully replacing traditional techniques, cryo-EM provides a synergistic alternative that expands the accessible landscape of biological structure determination [7,8].

At the center of this progress is single-particle analysis (SPA), an image processing method [11,12,13]. Briefly, SPA functions by extracting individual particles from the sample micrographs, and treats each particle as a single two-dimensional (2D) piece of raw data [14]. Using this process, it reconstructs the three-dimensional (3D) structural information of the target sample using numerous particles [15,16]. Advancements in particle picking algorithms, particularly deep learning-based approaches, have significantly improved the accuracy and efficiency of this process, reducing manual intervention and accelerating throughput [17]. In addition, since the introduction of transmission electron microscopy (TEM) to structural biology, various 3D reconstruction methods, including SPA, electron tomography (ET), random conical tilt (RCT) reconstruction, and so on, have been proposed for modeling biological specimens [18,19,20]. However, due to technical limitations, applicability constraints, and considerations of convenience, SPA has become the focus of a significant amount of research in this field [21,22].

SPA has seen striking advances not only in terms of image processing speed and convenience but also with respect to the algorithms responsible for powering 3D reconstruction [15,23]. Considering the overall workflow of SPA when using software packages for data processing [23,24,25], the micrographs that were acquired undergo processes such as image frame alignment to correct for particle movement caused by the electron beam and estimation of the contrast transfer function (CTF), which is constrained by the limits of the optical components of the microscope [26,27]. Consequently, particles are selected and extracted after enhancing the signal-to-noise ratio (SNR) to some extent [28]. Recently developed Bayesian approaches and adaptive regularization methods have notably enhanced the quality and interpretability of SPA reconstructions, especially in challenging datasets with conformational heterogeneity [29]. An iterative alignment process, involving shifts and rotations, is then applied using 2D or 3D classification [24,25]. The raw data generated by these processes are selectively curated, and, based on the fundamental structural information, 3D reconstruction is performed to obtain the final 3D map [16]. This is comprehensively described in Figure 1, which shows the typical SPA workflow for protein structural analysis and the acquisition of an atomic model in cryo-EM. Critically, the accuracy of reconstruction heavily depends on proper symmetry assignment and the appropriate selection of particle quantity, as misapplication can result in artifacts, distorted structures, and incorrect biological interpretations [30].

In SPA, symmetry refers to the assumption that the imaged particle has repeated structural elements related by rotational or point group operations. Symmetry application during reconstruction effectively amplifies the signal by averaging equivalent views, improving resolution and reducing noise, particularly for highly symmetric particles such as icosahedral viruses of oligomeric enzymes [11,31]. However, incorrect symmetry assignment can lead to the misalignment of asymmetric features, produce artificial density, and obscure biologically relevant conformations [30]. Therefore, to avoid such pitfalls, best practices now involve verifying symmetry assumptions via an inspection of 2D class averages, assessing local-resolution anisotropy, and utilizing tools such as directional Fourier shell resolution (FSC) or symmetry expansion combined with focused classification.

The dihydrolipoyl acetyltransferase (E2) complex forms the structural inner core of the pyruvate dehydrogenase complex (PDC), assembling into a highly ordered icosahedral or dodecahedral geometry composed of 60 identical monomers in many bacterial species [32,33]. Each monomer consists of multiple domains including a lipoyl domain, a peripheral subunit-binding domain, and a catalytic core domain, which together facilitate higher-order self-assembly. This oligomeric arrangement underlies the rationale for applying icosahedral symmetry during SPA-based 3D reconstruction of the E2 inner core.

In this study, we confirmed the influence of variations in the symmetry options and differences in the particle quantity on the inner core component, the E2 complex, of the PDC, within the context of the general SPA procedures [33,34]. The extent to which changes in the symmetry parameters and the number of particles lead to different results with the same raw data was investigated. Our results demonstrate that successful 3D reconstruction depends on the appropriate use of symmetry parameters, and that an efficient reconstruction can be accomplished solely with a moderate number of particles, considering the computational cost-effectiveness. Zivanov et al. introduced a Bayesian framework that improves angular assignment and motion correction, which is especially critical when symmetry is imposed [35]. Nakane et al. further demonstrated that excessive or biologically unjustified symmetry can distort functionally relevant conformational features, even at sub-3 Å resolution [11]. Recent studies have underscored the importance of rational symmetry selection and particle curation in modern cryo-EM. Our work contributes by systematically benchmarking symmetry-particle number combinations and integrating current theoretical understanding into an experimentally validated pipeline. Furthermore, by reviewing recent methodological advancements and empirical guidelines from state-of-the-art SPA studies, we propose strategic considerations and practical guidelines for optimizing symmetry usage and particle count in SPA workflows, especially valuable for labs operating with limited computational resources [36]. In parallel with these developments in image processing and data curation, atomic model building methods have also seen rapid progress. Recent tools such as ModelAngelo [37], EMbuild [38], DiffModeler [39], DEMO-EM2 [40], and DEMO-EMol [41] offer automated pipelines for interpreting cryo-EM maps, particularly enhancing model accuracy for maps of moderate to lower resolution.

2. Results

2.1. General Workflow of SPA

The general workflow of SPA image processing is depicted in Figure 1 [42]. In this study, we processed a total of 5250 acquired movies by correcting beam-induced motion and estimating CTF. Subsequently, using the automated particle picking function of CryoSPARC, we performed three rounds of 2D classification from an initial set of 113,560 particles, ultimately confirming 69,946 particles as the final selection. This is followed by ab initio reconstruction to produce five models to enable us to selectively choose particles that align well in three-dimensional space. After selecting the model with well-aligned particles with sharp features and reduced background noise from among the five models [36], we designated the data corresponding to 53,702 particles as set 1. Additionally, to study the variations in the results due to variables in the cryo-EM SPA process, we restricted the information coming from particles by choosing approximately 70% of set 1, amounting to 35,911 particles, for set 2 (Figure 2). To simulate experimental conditions with limited computational resources, we generated set 2 through random downsampling of set 1. This sampling was not based on classification bias but applied uniformly across the curated particle pool. Set 1 itself was obtained through strict particle curation involving multiple rounds of 2D and 3D classification, followed by visual inspection to retain high-SNR, well-aligned particles. This design allowed us to benchmark the impact of moderate particle reduction on structural resolution and interpretability under realistic constraints. Subsequently, utilizing the same initial model, we conducted 3D classification, ultimately confirming set 1 with 43,249 particles and set 2 with 29,376 particles.

Recently, iterative refinement strategies have significantly advanced the classification process, enabling a more accurate resolution of heterogeneous structures. Techniques such as multi-body refinement and 3D variability analysis (3DVA) have emerged as powerful approaches to better distinguish conformational variability and structural heterogeneity within particle populations, thus markedly improving the reliability of selected particles [43,44]. Furthermore, the effectiveness of particle selection critically depends on precise contrast transfer function (CTF) estimation and correction. Innovations in real-time CTF estimation and advanced motion correction algorithms have greatly enhanced image quality and structural resolution, minimizing systematic errors during reconstruction [26,27].

This multi-steps curation protocol reflects current best practices in SPA, where aggressive classification and particle selection strategies are often needed to mitigate conformational and compositional heterogeneity [45]. Recent studies emphasize that retaining only a subset of well-aligned, high-SNR particles not only improves map resolution but also enhances interpretability by suppressing structural noise introduced by flexible or heterogeneous states [36]. Such iterative curation strategies, involving ab initio modeling followed by focused classification, have become essential in resolving conformational ensembles of dynamic complexes [11]. Moreover, in systems with modular or symmetry-mismatched components, refined particle pruning helps decouple overlapping features, enabling focused refinements that yield more biologically relevant models [23]. By benchmarking the effect of particle downsampling, this study offers a practical route for resource-limited laboratories to assess the trade-off between resolution and computation efficiency [14]. Especially for cryo-EM facilities lacking access to high-performance computing clusters, such rational subsampling combined with optimized symmetry constraints can guide efficient allocation of computing resources without substantial loss in structural quality [46].

When conducting cryo-EM SPA, one of the parameters that is input by the user and is able to influence the results, is the symmetry [47]. The initial symmetry assignments were not arbitrarily chosen but were guided by the alignment characteristics of 2D class averages and existing structural models. In particular, the E2 core of the pyruvate dehydrogenase complex is known to form a dodecahedral assembly (PDB: 1B5S), suggesting high-order symmetry. Based on these biological insights, we initially hypothesized I symmetry. To ensure this assumption was valid, we conducted directional FSC analysis and examined and verified structural consistency across symmetry groups. The influence of symmetry on the 3D reconstruction was therefore assessed by reconstructing particle sets 1 and 2 using values applicable to spherical samples, including C1, C4, D4, D7, D8, O, and I [31,48]. The resolution of the maps that were reconstructed using each symmetry parameter is presented in

Table 1. Calculated resolution with various symmetry parameters.

Particle Set 1
Symmetry	C1	C4	D4	D7	D8	O	I
Resolution (Å)	5.03	5.34	5.14	4.72	6.21	4.69	4.20
Particle Set 2
Symmetry	C1	C4	D4	D7	D8	O	I
Resolution (Å)	6.16	6.73	6.39	4.77	5.96	4.81	4.26

, and Figure 3 shows the results of the relatively high-resolution maps for particle set 1. The FSC resolution of the maps, excluding the C1 and I symmetry results from particle sets 1 and 2, were calculated to be sufficient to distinguish the secondary structure of the protein. However, their 3D structures were not well-aligned to clearly resolve structural features [49].

In Figure 3, a comparison of C1 and I reveals noticeable differences in the structures represented by C4, D4, D7, D8, and O in Figure 3C. This indicates that misunderstanding the symmetry of the target with the consequent use of inappropriate options may lead to the generation of artifacts and disadvantaged results compared to conducting the calculation without using symmetry such as when using C1. This could potentially disrupt subsequent analysis.

Such outcomes are consistent with prior observations that an incorrect imposition of symmetry can generate spurious structural features in final maps, especially when the biological assembly deviates from idealized symmetry [50]. Even when global FSC values suggest acceptable resolution, improper symmetry application often leads to directionally biased local resolution, anisotropic features, or masked heterogeneity that compromise interpretability at high resolution. Nakane et al. demonstrated that imposing Cn or Dn symmetry on complexes with only pseudo-symmetry can obscure biologically relevant domain arrangements, resulting in misleading atomic models [11]. In complexes exhibiting partial symmetry or local heterogeneity, symmetry enforcement during refinement may mask functionally relevant conformations, thereby obscuring critical mechanistic insights [5].

Figure 4 and Figure 5 show the gold standard Fourier shell correlation (GSFSC), one of the quantitative indicators of well-aligned and misaligned maps. Despite the resolution being calculated without alignment issues, structural misalignments were observed in the 3D maps, appearing as artifacts. This indicates that even if the reconstruction proceeds normally under certain symmetry options, the resulting 3D maps may still be structurally inaccurate, comparison with asymmetrical reconstructions, and assessment of structural features such as continuity, resolution anisotropy, or biological plausibility can provide critical clues for detecting improper symmetry assignment [30].

Thus, interpretation of FSC values must consider not only global resolution metrics but also local-resolution variation, consistency with atomic models, and structural coherence, particularly in datasets with low SNR or conformational diversity [51]. Composite validation strategies such as directional FSC, local-resolution heatmaps, and density variance analysis are increasingly used to supplement GSFSC and validate reconstruction integrity [52].

2.2. Comparative Analysis of 3D Reconstruction by Varying the Particle Quantity

The differences in the 3D reconstruction results based on the number of particles were assessed by conducting a comparative analysis using the appropriately reconstructed maps of particle sets 1 and 2 with the symmetry parameters C1 and I in Figure 6. The calculated resolution values for set 1 were C1 = 5.03 Å and I = 4.20 Å, while for set 2, C1 = 6.16 Å and I = 4.26 Å. Notably, for C1, visual inspection of the reconstructed maps did not reveal significant structural differences between sets 1 and 2. However, meaningful differences between the reconstructions, suggesting that the reduced particle count in set 2 negatively impacted the overall map (Figure 4 and Figure 5). In contrast, for I symmetry, the results from both sets 1 and 2 were nearly identical, with only slight differences in resolution that did not affect the ability to resolve key structural features, such as secondary structure elements and inter-domain interfaces.

These observations align with recent theoretical and empirical studies demonstrating the diminishing returns of excessively increasing particle numbers, especially when accurate symmetry parameters are employed. Studies by Rosenthal and Henderson [53] and Grant and Grigorieff [54] revealed that the improvement in resolution plateaus beyond a certain number of particles, and that computational efficiency can be significantly enhanced by selecting an optimal particle count without a substantial loss of structural accuracy.

Therefore, although the use of a large number of particles generally improves map quality, especially in reconstructions without symmetry constraints (as observed in C1), the results suggest that having accurate symmetry information for the target significantly compensates for the reduced particle count.

In the case of I symmetry, the maps produced with approximately 70% of the particle data (set 2) were highly comparable to those reconstructed using the full particle set (set 1), thereby confirming that symmetry constraints can enhance the efficiency of SPA by minimizing the computational burden without sacrificing resolution. Further analysis, shown in Figure 6C, highlights the challenge in discerning structural differences between the refined pseudo-atomic model, based on the I symmetry map of particle set 1, and the corresponding I maps from both set 1 and set 2. Visual comparison of the pseudo-atomic models fitted to the maps reveals minimal discrepancies in overall structural alignment, indicating that even with fewer particles, I symmetry maintains sufficient accuracy for model building and refinement.

This finding is particularly relevant for cases where computational resources are limited, as it demonstrates that an efficient workflow can still produce high-quality results with fewer particles, provided that symmetry information is well-understood and applied correctly. Furthermore, under the same conditions, we confirmed that the results of I symmetry constraint for set 1, comprising 43,249 particles, and set 2, comprising 29,376 particles, did not significantly differ from those obtained with the full dataset of 69,946 particles. The resolution of 4.20 Å achieved using the full particle set with I symmetry closely matches that of the smaller sets, demonstrating the robustness of the symmetry constraint in handling variations in particle count. These findings suggest that the introduction of symmetry constraints can mitigate the influence of particle reduction on the final resolution, a critical factor for laboratories with limited computational resources.

Consequently, these results suggest that users need to adopt more efficient image processing strategies that take into account the computational power and time cost associated with processing massive amounts of raw data, such as an excessive number of particles. Strategic reduction in particle count, combined with the appropriate use of symmetry, can provide a balance between computational efficiency and structural accuracy, particularly in large-scale cryo-EM research where processing speed and data management are critical concerns.

In summary, our results offer practical guidance informed by recent empirical studies, advocating for a balanced approach to particle quantity and symmetry application, ensuring optimal resource utilization while preserving structural fidelity.

3. Discussion

Cryo-EM has revolutionized structural biology by overcoming the limitations of traditional methods such as X-Ray crystallography and NMR [1]. This advancement has been particularly transformative in analyzing the 3D structure of macromolecular complexes through TEM, leading to the development of several reconstruction methods such as ET and RCT reconstruction. However, SPA has emerged as the dominant method for high-resolution structural determination, largely due to the inherent limitations of alternative methods. ET is a powerful approach for visualizing macromolecular complexes, providing critical insights into their biological context. However, ET has several limitations that hinder its application in high-resolution protein structure determination [55,56]. One major drawback is sample thickness or size, which can lead to multiple scattering effects, reducing image contrast and limiting achievable resolution. Additionally, ET suffers from the missing wedge problem, which is caused by the restricted tilt range during data collection, leading to anisotropic resolution [57]. In contrast, cryo-EM with SPA provides a more broadly applicable, efficient, and powerful approach for the 3D reconstruction of macromolecular complexes. However, the complexity of cryo-EM data processing requires the careful consideration of various factors, among which symmetry is crucial [48]. Inappropriate symmetry applications can distort the reconstruction, leading to erroneous interpretations of the biological structure under study. In addition, when numerical resolution is calculated without issues, detailed visual inspection remains essential for assessing the accuracy of 3D reconstructions. In some cases, artifacts were observed despite achieving relatively high-resolution values, suggesting that quantitative metrics alone may not fully capture structural integrity. Our results demonstrated that careful attention to symmetry parameters is essential for accurate 3D reconstruction, emphasizing that users must be well-informed about the underlying principles governing their data processing software. Our study highlights the critical role that symmetry parameters play in the SPA workflow for cryo-EM. We observed that even with high-quality raw data, an insufficient understanding or misapplication of symmetry can lead to unsuccessful 3D reconstructions and the generation of artifacts. This observation aligns with previous studies demonstrating that incorrect symmetry assumptions can introduce artifacts and hinder accurate model interpretation.

For example, studies on symmetry-mismatched complexes, such as tailed bacteriophages, have shown that enforcing incorrect symmetry constraints can obscure key structural details and produce artifacts in reconstructed maps [58]. Specifically, in the case of the bacteriophage ϕ29, where a sixfold-symmetric tail assembly is attached to a fivefold-symmetric prolate head, inappropriate symmetry application led to distorted reconstructions. By correctly adapting for symmetry parameters, an improved structural model of the viral particle was obtained, emphasizing the importance of considering symmetry parameters. This finding is particularly significant given the increasing availability and use of user-friendly software packages in this field. Our findings also reaffirm the computational implications of particle count optimization. While collecting more particles intuitively improves signal averaging and statistical robustness, the gain in resolution plateaus beyond a threshold, as established in both experimental and simulation-based studies [53,54]. This plateau effect highlights the importance of selectively curating particles using classification strategies that maximize structural homogeneity rather than brute-force volume.

In cases where the target structure is not yet known, the results of negative staining or 2D classification become increasingly important. Low-resolution 3D models reconstructed by negative staining can provide a rough understanding of the overall architecture and possible symmetry information. Alternatively, after data collection, careful consideration should be given during the 2D classification step, as the resulting class averages can reveal structural features. In addition to symmetry considerations, our study also examined the influence of particle quantity on the efficiency and accuracy of 3D reconstructions. The advancements in cryo-EM hardware, including more sophisticated microscopes and detectors, have enabled the collection of vast amounts of data. However, processing this data efficiently remains a challenge, especially for researchers with limited computational resources. Our findings indicate that a moderate number of particles can suffice for high-quality structural analysis, provided that the symmetry parameters are correctly applied. Although only 113,560 particles were extracted from 5,250 movies, this reflects the type of suboptimal experimental condition that the current study aims to address. The presented workflow demonstrates that even in such limited settings, optimizing symmetry selection and particle quantity can still yield high-quality reconstructions efficiently. Our findings indicate that a moderate number of particles can suffice for high-quality structural analysis, provided that the symmetry parameters are correctly applied. This has significant implications for the design of cryo-EM experiments, suggesting that excessive data collection might not always translate to better results and can instead lead to increased computation costs and time. These results are consistent with previous studies demonstrating a plateau in resolution improvement beyond a certain particle threshold. Rosenthal and Henderson [53] established that resolution gain follows a square-root dependence on particle number, which gradually diminishes after a few tens of thousands of particles. Grant and Grigorieff [54] further quantified that, under ideal conditions, particle counts exceeding ~40,000 often yield only marginal improvements in resolution. In our dataset, reconstructions using ~43,000 particles (set 1) and ~29,000 particles (set 2) yielded comparable map quality, indicating that a plateau effect likely begins near 40,000 particles for this target. Thus, beyond this point, increasing particle count offers diminishing returns relative to the computational cost, supporting strategic particle selection rather than indiscriminate accumulation. Notably, symmetry-aware reconstruction strategies, such as adaptive particle selection, focused refinement, and sub-volume masking offer viable solutions for improving resolution without disproportionately increasing dataset size [44]. These methods are especially useful in large complexes exhibiting mixed symmetry or modular domain organization.

The comparative analysis between different particle sets revealed that while larger datasets generally offer better resolution, the improvement plateaus beyond a certain point. For instance, in our study, set 1 with 43,249 particles and set 2 with 29,376 particles produced comparable results when appropriate symmetry was applied. This suggests that researchers can achieve substantial computational efficiency without compromising on the accuracy of the structural data by optimizing the number of particles used in the analysis. The relationship between particle number and achievable resolution has been extensively characterized, with prior studies by Liao and Frank [59] providing a detailed explanation of this dependency. Moreover, our study underscores the need for strategic planning in cryo-EM projects, particularly in the context of limited resources. Efficient image processing strategies, informed by an understanding of the key variables such as symmetry and particle quantity, can significantly reduce the time and computational power required for high-resolution 3D reconstructions. This approach is not only cost-effective but also allows researchers to allocate their resources more effectively, potentially accelerating the pace of discoveries in structural biology.

In addition to symmetry and particle number, other important parameters significantly affect reconstruction efficiency and computational resource demands. Specifically, the choice of pixel size during image acquisition and the box size during particle extraction can impact both map resolution and data processing requirements [60,61]. Larger pixel sizes reduce file sizes and speed up computation but may compromise high-resolution details, whereas smaller pixel sizes increase computational load without proportionate resolution gain unless the sample supports it [62,63]. Similarly, excessively large box sizes can capture more peripheral noise, increasing memory and storage usage during processing. Therefore, selecting a pixel size that matches the Nyquist frequency relative to the target resolution and optimizing the box size to include the particle plus minimal surrounding margin are critical for balancing resolution and resource use. Incorporating such considerations during experimental planning can significantly enhance workflow efficiency in settings with limited computing infrastructure. To aid in practical implementation, we summarize key guidelines for symmetry determination in SPA workflows. Prior to 3D reconstruction, negative-staining EM or initial 2D classification can reveal symmetry-related features. Class averages showing distinct rotational views often indicate cyclic (Cn), dihedral (Dn), or icosahedral (I) symmetry. Subsequently, local-resolution mapping and directional FSC analysis should be used to validate the chosen symmetry. When discrepancies arise between visual features and numerical metrics, lower-symmetry models (e.g., C1) should be considered to avoid over-symmetrization. Together, these strategies allow for a rational and biologically relevant selection of symmetry parameters.

4. Materials and Methods

4.1. Protein Sample Preparation

The E2 complex was purified from Escherichia coli (E. coli) BL 21 star (DE3) cells (Thermo fisher Scientific, Waltham, MA, USA) [64], using E2 encoded cDNA of Bacillus stearothermophilus (B. stearothermophilus) (UniProt: P11961) as templates. The expression of recombinant E2 was induced by isopropyl-ß-D-thiogalactoside (IPTG; Sigma-Aldrich, St. Louis, MS, USA) at 18 °C for 16 h [65]. Harvested cells were prepared by centrifugation (2808× g at 4 °C for 15 min, Hanil, Kimpo, Republic of Korea), washed with a buffer (50 mM Tris pH 7.5, 500 mM NaCl, 5% glycerol and 1 mM phenyl-methylsulfonyl fluoride), and lysed by ultra-sonication (Sonics, Newtown, CT, USA). After a second centrifugation (30,438× g at 4 °C for 15 min, Hanil, Kimpo, Republic of Korea), the supernatant was purified by heat precipitations and a final centrifugation (30,438× g at 4 °C for 25 min, Hanil, Kimpo, Republic of Korea) before use [33,66].

4.2. Cryo-EM Sample Preparation

The purified protein sample was prepared on glow discharged Quantifoil R2/2 holey carbon EM grids (EMS, Hatfield, PA, USA) using Vitrobot Mark IV (Thermo Fisher Scientific, Waltham, MA, USA; 5 s blotting time with force −1 and 100% humidity at 4 °C). A sample (4 μL, 100 nM) was applied to the grid, to which a current of 15 mA had been applied for 1 min using easiGlow (Ted-Pella, Redding, CA, USA) to ensure it was negatively charged before use, before being quickly plunge-frozen in liquid ethane cooled by liquid nitrogen. (All the instrumentation is installed at the Korea Basic Science Institute).

4.3. Data Collection and Image Processing of E2 Complex

Data collection was automatically performed using EPU software v2.12 (Thermo Fisher Scientific, Waltham, MA, USA) on a Talos Artica G2 transmission electron microscope (Thermo Fisher Scientific, Waltham, MA, USA) operating at 200 kV and equipped with a K3 detector (the instrumentation is installed at the Korea Basic Science Institute). Each movie was recorded with a total dose of 40 e⁻/Å⁻² per movie and a defocus range of −1.8 to −2.6 μm at a 1.1 Å. After estimating the contrast transfer function and motion correction, 5250 movies were acquired. After estimating the 5250 micrographs, the particles were picked from micrographs without templates (Blob picking). And performing 3 rounds of 2D classification, the best-looking 69,946 particles and 8 class averages were selected, as judged by visual inspection. To classify and visualize particles in three-dimensional space, we produced five models using ab initio reconstruction. From the five ab initio reconstruction models, approximately 75% of the particles (53,702 particles), and roughly 50%, totaling 35,911 particles, were selected. These particles were subsequently classified using 3D classification, resulting in a final classification of 43,249 particles (set 1) and 29,376 particles (set 2). The influence of symmetry in the reconstruction was assessed by subjecting each final particle set to 3D reconstruction under symmetry restraints of C1, C4, D4, D7, D8, O, and I, respectively. The well-reconstructed final model obtained by applying restraints C1 and I, was used to refine the symmetric parameter and number of particles with a soft-edge mask and sharpened. All image processing and validation procedures were carried out using CryoSPARC v4.0 [24]. The final resolution of each cryo-EM model was estimated by GSFSC (0.143 cut-off). The data collection and refinement statistics are summarized in

Table 2. Data collection and refinement statistics.

EM Data Collection	Set 1 EMD-38337	Set 1 EMD-38338	Set 2 EMD-38339	Set 2 EMD-38341
Particles	43,249	43,249	29,376	29,376
Pixel size (Å)	1.1	1.1	1.1	1.1
Voltage (keV)	200	200	200	200
Defocus range (μm)	−1.5 to −2.6	−1.5 to −2.6	−1.5 to −2.6	−1.5 to −2.6
Total electron dose	40 e−/Å−2	40 e−/Å−2	40 e−/Å−2	40 e−/Å−2
Map symmetry	I	C1	I	C1
Resolution (Å)	4.20	5.03	4.26	6.16
Map sharpening B-factor (Å)	−208.1	−101.7	−193.5	−245.1
Refinement and validation
Initial model used (PDB)	1B5S
Correlation model versus data
CC (mask, volume)	0.78, 0.79
R.m.s deviations
Bond length (Å)	0.002
Bond angle (°)	0.528
MolProbity score	1.50
Clashscore	0.79
Romater outliers (%)	1.31
Ramachandran plot
Favored (%)	79.04
Allowed (%)	19.65
Disallowed (%)	0
PDB deposition ID	9ukz

. The final models were deposited in the Electron Microscopy Data bank (EMDB) (EMD-38338, Set 1, C1; EMD-38337, Set 1, I; EMD-38341, Set 2, C1; EMD-38339, Set 2, I).

4.4. Atomic Model Building

For model building and refinement of the E2 complex were carried out using Coot v1.1 [67] and Phenix v1.20 [68], starting from deposited structure (PDB 1B5S) [32]. Atomic coordinates and temperature factors were refined using the real-space refinement tool in Phenix. Model validation and refinement statistics ins provided in Table 2. Final model was deposited as PDB: 9ukz. Molecular graphics were generated using UCSF chimera v1.16 [69].

5. Conclusions

Our results indicated that an insufficient understanding of symmetry, which is a key variable in the image processing workflow of SPA for the 3D reconstruction of cryo-EM data, can lead to unsuccessful reconstruction and the production of artifacts, even when starting with high-quality raw data. This conveys an important message for users who employ increasingly user-friendly software packages developed for this research field. Additionally, despite the continuous efforts of developers, advancements in hardware, including those in microscopes and detectors, generate vast amounts of data for SPA. As demonstrated in this study, obtaining results with an appropriate quantity of particles emphasizes that researchers can achieve sufficient structural analysis with limited computing power and time cost.