Towards atomic resolution structural determination by single-particle cryo-electron microscopy

https://doi.org/10.1016/j.sbi.2008.03.004 Get rights and content

Recent advances in cryo-electron microscopy and single-particle reconstruction (collectively referred to as ‘cryoEM’) have made it possible to determine the three-dimensional (3D) structures of several macromolecular complexes at near-atomic resolution (∼3.8–4.5 Å). These achievements were accomplished by overcoming the challenges in sample handling, instrumentation, image processing, and model building. At near-atomic resolution, many detailed structural features can be resolved, such as the turns and deep grooves of helices, strand separation in β sheets, and densities for loops and bulky amino acid side chains. Such structural data of the cytoplasmic polyhedrosis virus (CPV), the Epsilon 15 bacteriophage and the GroEL complex have provided valuable constraints for atomic model building using integrative tools, thus significantly enhancing the value of the cryoEM structures. The CPV structure revealed a drastic conformational change from a helix to a β hairpin associated with RNA packaging and replication, coupling of RNA processing and release, and the long sought-after polyhedrin-binding domain. These latest advances in single-particle cryoEM provide exciting opportunities for the 3D structural determination of viruses and macromolecular complexes that are either too large or too heterogeneous to be investigated by conventional X-ray crystallography or nuclear magnetic resonance (NMR) methods.

Introduction

Electron imaging is rapidly emerging as an indispensable tool in biology that offers great promises for three-dimensional (3D) structural studies of macromolecular complexes or biological nanomachines. Electron diffraction data have long been successfully used to derive 3D structures of two-dimensional crystalline samples, up to a resolution of 1.9 Å [1, 2, 3, 4]. For macromolecule complexes in their noncrystalline ‘native’ states, single-particle cryo-electron microscopy (commonly referred to as ‘cryoEM’) has played an increasingly important role in determining 3D structures up to subnanometer resolution (e.g. [5•, 6•, 7, 8, 9, 10]). Such large assemblies often are either too large or too heterogeneous to be studied by conventional methods, such as X-ray crystallography and nuclear magnetic resonance (NMR) (e.g. [11, 12, 13, 14, 15, 16]).

Four recent studies have further pushed the limit of cryoEM single-particle reconstruction to near-atomic resolution (∼3.8–4.5 Å) [17••, 18••, 19••, 20••] (Table 1). At this resolution, many detailed structural features such as turns and deep grooves of helices, strand separation in β sheets, densities for loops and bulky amino acid side chains can be resolved. Such structure features revealed in the cytoplasmic polyhedrosis virus (CPV) [18••], the GroEL [20••], and the Epsilon 15 bacteriophage [21] were used to trace the amino-acid backbones using integrative modeling tools. In the case of CPV and Epsilon 15 phage where no prior atomic structures were available, model building increased the information content of cryoEM-derived structures and significantly enhanced their overall values to address biological questions. In the cases of the rotavirus [19••] and the GroEL [20••], where X-ray models were available for the same complexes, an excellent agreement between the cryoEM structure and X-ray structure were demonstrated, up to the level of amino acid side chains, thus firmly establishing the validity of near-atomic resolution structures from cryoEM. Crucial to these latest developments of single-particle cryoEM are careful and systematic considerations of sample preparation and cooling strategies, instrumentation and imaging conditions, and computational methods for processing noisy cryoEM images and atomic model building. Here, I will first provide a discussion about these issues by comparing the approaches used to obtain the latest results (Table 1). Then, I will illustrate how a near-atomic resolution structure can address significant biological questions using the 3.8-Å CPV structure as an example [18••].

The goal of cryoEM sample preparation is to obtain frozen hydrated specimen on grids from freshly prepared samples to minimize possible deformation or damage to the macromolecular complexes being studied. In high-resolution cryoEM, structural homogeneity or integrity is more important than purity, as opposed to X-ray crystallography and NMR, in which sample purity is essential. In cryoEM images, obvious impurities or contaminations can be distinguished from particles of interest either visually or by computational classification, and thus will not severely impact or prevent subsequent data processing. However, sorting out structural deformation or flexibility computationally is much more challenging and is often a limiting factor toward obtaining a high-resolution reconstruction. In this regard, it is desirable to avoid possible structural damage caused by density gradient centrifugation. For example, a simple protocol with single-step purification was used for the structurally robust or stable CPV particles [18••]. In the case of the rotavirus, double-layered particle reconstruction [19••], structural heterogeneity was not as large a concern because highly ordered crystals were grown from the same sample, indicating that it is likely to be structurally homogenous. In most cryoEM studies, structural heterogeneity could very well be the first limiting factor toward atomic resolution, even if the sample is chemically pure (e.g. [22]). Therefore, it is desirable to use a freshly prepared, minimally treated or disturbed sample that has never been frozen and thawed in order to increase the chance of obtaining high-resolution cryoEM reconstructions.

The choice of using liquid helium versus liquid nitrogen for sample cooling in cryoEM imaging has been a hotly debated issue. In fact, among the four near-atomic resolution structures reported to date, the 3.8-Å resolution structures of the CPV [18••] and rotavirus [19••] were obtained from liquid nitrogen-cooled samples, but the 4.2-Å GroEL structure [20••] and the 4.5-Å Epsilon 15 bacteriophage structure [17••] were from liquid helium-cooled samples (Table 1). An increase (1.4–2.5 times) of dose tolerance in liquid helium was estimated based on the measurement of electron diffraction spot intensity [23], but most of those earlier studies were largely inconclusive, as comparisons were made using data from different instruments [24]. Recent direct comparison between imaging the same samples at liquid nitrogen and liquid helium temperatures showed that, surprisingly, liquid helium is actually a worse cryogen for cryo-electron tomography applications due to decreased low-resolution contrast between ice and biological samples at liquid helium temperature [25, 26, 27]. To address the question of which cryogen is better suited for high-resolution single-particle cryoEM imaging, we used a FEI Polara G2 dual temperature cryo-electron microscope to compare the data quality between images obtained from the same sample cooled at liquid nitrogen and liquid helium temperatures, respectively (I Atanasov, ZH Zhou, unpublished data). A systematic comparative analysis of specimen charging, image contrast, dose tolerance, and potential resolution limit has led us to conclude that there is no overall benefit to use liquid helium other than liquid nitrogen for sample cooling for high-resolution imaging. When liquid helium is used, the benefit of the slightly better dose tolerance was negated by the lower image contrast because of increased ice density, the more stringent treatment requirement of support film to prevent charging, as well as the significantly higher cost of helium.

When pushing toward high resolution for very large particles, the depth-of-field problem emerges as a limiting factor [28, 29, 30•, 31, 32•]. The current formulation based on the Central Projection Theorem [33, 34], as implemented in most current 3D reconstruction programs, may no longer hold under this situation. Instead, the Fourier transform of a projected image corresponds to neither a slice nor an Ewald sphere in the Fourier space, but to the sum of the values on two quadratic surfaces Ewald spheres in 3D Fourier space [32]. Higher voltage (thus shorter electron wavelength) alleviates this problem. But for an 800 Å or larger particle, a full correction of the transfer function is necessary for pushing resolution beyond 4 Å using 300 keV electrons [28, 29, 32•]. Indeed, 300 keV electrons were used in all four near-atomic resolution structures (Table 1).

Except for the still limited field of view, CCD cameras offer a number of significant advantages over film as a recording media [35, 36, 37, 38] and have been rigorously demonstrated to be a preferred recording device for single-particle cryoEM imaging up to subnanometer resolution [39]. First, the broad dynamic range, linearity, and low level of noise make these devices ideal for recording diffraction intensities of 2D crystals [40]. Second, a CCD camera gives instant feedback about image quality, thus allowing for the efficient optimization of sample preparation and microscope alignment. Third, without the step of loading and unloading of photographic films in the microscope vacuum, the usable time of a cryoEM grid can be significantly extended from the typical one day for a film session to days or weeks for a CCD session. Fourth, although high-resolution data is more severely dampened due to worse point-spread function of CCD cameras, CCD images have better signal-to-noise ratio at low-resolution region, thus providing better image ‘contrast’ crucial for aligning single-particle images [39]. Finally, compared to conventional photographic films, imaging on a CCD exposes a much smaller sample area. Thus, it minimizes possible beam-induced specimen movement or charging similar to the ‘spot-scan’ approach [41] and reduces the effect of defocus variation within each image. For near-atomic-resolution cryoEM reconstruction, the raw cryoEM images should be rigorously screened by selecting only those with visible contrast transfer function (CTF) rings up to 5 Å in their power spectra [42]. These images should also have no visible specimen drift, charging, and astigmatism in their power spectra to ensure that only high-quality images are used in the final 3D reconstruction.

Data processing for single-particle reconstruction consists of two essential steps: orientation-center parameter determination and 3D reconstruction. Structure refinement is carried out as an iterative procedure of these two steps by gradually pushing toward higher resolution. Various software packages, including EMAN [43], FREALIGN [44], and IMIRS [45, 46], were successfully used for the near-atomic resolution structure reconstructions (Table 1). The principles for orientation-center parameter determination used in these packages are similar, including searching for common lines [33] and matching computed projections [47]. For the 3D reconstruction step, EMAN and FREALIGN use direct Fourier inversion method, which is the most computationally efficient, but may have large memory requirement (e.g. ∼30 GB for an 800 Å particle) and is generally more sensitive to noise. By contrast, the less computation efficient Fourier–Bessel synthesis method [33] and spherical harmonics synthesis method [48] implemented in IMIRS are less sensitive to image noise and have less memory requirement. Overall, the total computing times used in these studies differ by several orders of magnitude (Table 1) and are probably related to differences in implementation details and user procedures.

The effective resolution (or better yet, the resolving power) of a single-particle cryoEM reconstruction can be estimated both by critically evaluating the structural features or hallmarks resolved in the density map and by using statistical analysis of independent reconstructed maps. Figure 1 illustrates the structure feature-based method. Simulated densities are shown for the three types of secondary structure element – α helix, β sheet, and connecting loop – at different resolutions by filtering an atomic-resolution model using a Gaussian filter implemented in EMAN [20••]. At ∼3.8–4.0 Å resolution, an ideal density map of a helix should reveal deep grooves and clear pitches. Also resolved are strands in β sheets with an interstrand distance of ∼4.4 Å (except for regions of hydrogen-bond densities) and the zigzagging pattern of Cα atoms separated by ∼3.8 Å. Densities for bulky side-chains should also begin to appear. In the CPV cryoEM density map, these features are all present, including the 3.8-Å distance between adjacent Cα atoms and convincing densities for many side-chains (Figure 2). Commonly used statistical criteria for assessing effective resolution include Fourier shell correlation (FSC) [49] and spectral signal-to-noise ratio (SSNR) [50], which were shown to be equivalent [51].

It cannot be over emphasized that it is critical to evaluate map quality by examining the resolvable features in the maps, such as helix grooves, bulky side-chain densities, and well-separated strands in β sheets. Additionally, because self-correlating systematic error may exist in reconstruction methods, one needs to be cautious in using the FSC curves to assess the effective resolution of a cryoEM map, particularly when the resolution is approaching near-atomic scale. It has been recognized that the possible noise correlation may give rise to misleading ‘high-resolution’ FSC assessment, due to model bias in a template-matching-based refinement using cross correlation evaluation [52, 53]. For this reason, it is necessary to use structural features or stereochemistry as an internal control to judge whether such model bias has occurred during refinement. When the resolution reaches near-atomic scale, the aforementioned structural features can be used to judge whether the FSC assessment has been influenced by model bias. This model bias problem in FSC evaluation is more severe for image data with low signal-to-noise ratio, such as those at close-to-focus conditions, or for asymmetric or low-symmetric objects, and small particles [52]. Another way to monitor possible model bias is to refine the full image data set against two different initial models and monitor the convergence of the two reconstructions [52]. Finally, model bias can be eliminated by using the focal-pair approach [45, 46, 54] with model-free common lines method for orientation-center estimation [33]. Subsequent model-based orientation-center refinement can be performed as a ‘local’ search by limiting the range of orientation or center changes [46], thus reducing the risk of model bias in resolution estimation.

The resolution of cryoEM structures can also be judged empirically by direct comparison with published structures obtained through other well-established structural determination methods. For example, it is obvious that features revealed in an experimentally determined X-ray structure at 3.7-Å resolution [55] are not nearly as detailed as the simulated ‘perfect’ structure with a Gaussian filter to only 4 Å (cf. Figure 1, Figure 3). One can see that the structural features revealed in the 3.8-Å CPV map (Figure 2) are very much comparable to those revealed in the 3.7-Å X-ray map (Figure 3). Likewise, some regions of the CPV cryoEM map, such as the small protrusion domain of CSP, have lower resolutions than other regions, probably because of local intrinsic flexibility. This resolution variance in different regions of the same map is also common in X-ray structures (Figure 3a and b). It can easily be seen that the structure of Dsbb (Figure 3b) is less resolved.

Surface rendering of large volume data is very computationally demanding and can be a rate-limiting step in structural interpretation. Therefore it is desirable to dissect or segment individual structural components from the entire complex so that detailed features can be conveniently identified and compared. For icosahedral viruses, only structural components within an asymmetric unit are structurally unique and warrant examination. Segmentation of individual components also allow nonicosahedral averaging of structurally similar components to enhance the S/N ratio and further improve the resolution of the averaged subunits [19••, 56]. For example, nonicosahedral averaging of the 13 subunits of VP6 molecules in an asymmetric unit of a rotavirus reconstructed from 8400 particles significantly improved the resolution of the map (Figure 4) [19••]. The noise in the VP6 trimer was reduced by averaging, resulting in better resolved side-chain densities (Figure 4c and d).

Choices are limited for model building tools suitable for interpreting a near-atomic resolution cryoEM structure. Further complicating the issue is the fact that cryoEM maps often contain many subunits with over 5000 amino acids, yet less than 10–20% of those have identifiable, bulky side-chains. This is problematic because widely used atomic model-building tools, most notably O [57] and Coot [58], employ a bottom-up approach — that is, from side chains, to amino acids, to atomic model. Such bottom-up approaches have stringent requirements for map quality and resolution and are not readily applicable to near-atomic resolution cryoEM maps. Novel integrative tools, such as skeletonization and graph optimization, can be used together with other modeling tools for building rough Cα models [20••, 59] (Table 1).

Homologous structures can be of great help in model building based on near-atomic resolution data. For example, the CPV CSP and orthoreovirus λ1 protein sequences (and their structural homologs in BTV and RDV) have sequence identities of less than 10%, which is about the level expected between random sequences. Therefore, these proteins have no detectable sequence homology using current sequence alignment programs. Nevertheless, the cryoEM map clearly reveals that CSP shares a similar overall topology (or fold) with λ1, though there are significant insertions and deletions of helices. Therefore, one could use the known orthoreovirus λ1 structure as a topological reference, when building models for CSP-A and CSP-B.

Uncertainties in amino acid residue registration exist in interpreting maps around 3.8–4.0 Å resolution. In order to avoid over-interpretation of less resolved regions, one should be cautious not to go into unwarranted detail such as side-chain orientation, when describing these structures and their implications. It is important to remember that at near-atomic resolution, accurate registration of residues remains difficult. For this reason, model building from cryoEM reconstructions in the absence of existing atomic models of proteins with recognizable sequence homology is a significant, yet a very challenging undertaking. The models presented by Yu et al. [18••] are a combination of full-atom residue models (for helix-rich regions) and Cα models (for regions of less-defined densities). Because there is no refinement process for improving the model derived from cryoEM maps, in contrast to well-established model refinement schemes routinely performed in X-ray crystallography, these models are essentially Cα models. As the technology of single-particle cryoEM advances further to true ‘atomic resolution,’ it is expected that these uncertainties will be resolved.

Existing methods well established in crystallography can be readily adapted for verifying models derived from cryoEM maps. For example, models can be submitted to the Dali server to check for similarity to any existing fold. In the case of CPV, for example, when the CSP-A model was submitted to the Dali server, the search identified Bluetongue virus (BTV) capsid protein VP3 as a top hit, consistent with their fold similarity. However, when searching the structure database for the GTase domain of CPV, Dali did not identify any known GTases, suggesting CPV GTase does not share a significant fold homology with other known GTases. CPV GTase is more like the GTase of the unrelated Paramecium bursaria Chlorella virus 1 in the sense that it clearly consists of two domains, one large and one small, between which there exists a deep and narrow cleft [18••]. Even in the absence of a Cα model, a set of helices identified in a cryoEM structure can also be used as elements for probing structure databases for similar folds [8]. To this end, DejaVu and COSEC [60, 61, 62] can be used for such ‘spatial fold recognition’. A successful match suggests that the intermediate resolution structure under investigation has a possible homolog within the fold space. Other considerations include directions of side-chain densities, which should normally be pointing toward the N-terminus and the densities of helices. However, at the limited resolution of 3.8–4.0 Å, there could be occasional ambiguities in which side-chain densities appear sideways or horizontal, instead of pointing clearly toward the N-terminus.

Section snippets

Addressing biological questions at near-atomic resolution

Because no crystal structures are available for CPV and Epsilon 15 bacteriophage, atomic models derived from their cryoEM structures offered clues for functional implications [17••, 18••]. The 3.88-Å structure of CPV illustrates how a near-atomic resolution cryoEM structure can address interesting biological questions (Figure 5) [18••]. CPV is unique within Reoviridae in that it has only a single-layered capsid contained within polyhedrin inclusion bodies, yet is still fully capable of cell

Conclusions

Electron diffraction patterns from two-dimensional crystals were successfully used to derive atomic structures by crystallography [75, 76, 77]. High-resolution single-particle electron imaging has proven to be more challenging than electron crystallography due to many more limiting factors encountered in electron optics, imaging, image correction, and structure interpretation. The milestone accomplishments in breaking the nanometer barrier occurred about 10 years ago by using the structural

References and recommended reading

Papers of particular interest, published within the annual period of review, have been highlighted as:

  • • of special interest

  • •• of outstanding interest

Acknowledgements

This research is supported in part by grants from NIH and the Welch Foundation. I am grateful to Dr Wah Chiu for his advice and encouragement. I thank Xuekui Yu, Lei Jin for stimulating discussions and contributions to Figure 2, Figure 3, Figure 5; Yuyao Liang, Joanita Jakana, Matthew Baker, and Wah Chiu for their participation at the preliminary stage of the our efforts; Prof Jing-Qiang Zhang for providing the CPV-containing polyhedra sample and many years of fruitful collaboration; Ivo

References (82)

  • R.A. Crowther et al.

    Three dimensional reconstructions of spherical viruses by Fourier synthesis from electron micrographs

    Nature

    (1970)
  • M.B. Sherman et al.

    Reliability of phases retrieved from 400-kV spot-scan images of purple membranes acquired on a slow-scan CCD camera

    J Microsc

    (1997)
  • P.L. Stewart et al.

    Digitally collected cryo-electron micrographs for single particle reconstruction

    Microsc Res Technol

    (2000)
  • P. Zhang et al.

    Automated image acquisition and processing using a new generation of 4 K × 4 K CCD cameras for cryo electron microscopic studies of macromolecular assemblies

    J Struct Biol

    (2003)
  • Y. Liang et al.

    IMIRS: a high-resolution 3D reconstruction package integrated with a relational image database

    J Struct Biol

    (2002)
  • Z.H. Zhou et al.

    Determination of icosahedral virus structures by electron cryomicroscopy at subnanometer resolution

    Adv Protein Chem

    (2003)
  • T.S. Baker et al.

    A model-based approach for determining orientations of biological macromolecules imaged by cryoelectron microscopy

    J Struct Biol

    (1996)
  • M. Unser et al.

    A new resolution criterion based on spectral signal-to-noise ratios

    Ultramicroscopy

    (1987)
  • P.A. Penczek

    Three-dimensional spectral signal-to-noise ratio for a class of reconstruction algorithms

    J Struct Biol

    (2002)
  • S. Yang et al.

    Issues of resolution and polymorphism in single-particle reconstruction

    J Struct Biol

    (2003)
  • A. Stewart et al.

    Noise bias in the refinement of structures derived from single particles

    Ultramicroscopy

    (2004)
  • S.D. Fuller et al.

    Is Sindbis a simple picornavirus with an envelope?

    EMBO J

    (1987)
  • K. Inaba et al.

    Crystal structure of the DsbB–DsbA complex reveals a mechanism of disulfide bond generation

    Cell

    (2006)
  • J. He et al.

    Finding and using local symmetry in identifying lower domain movements in hexon subunits of the herpes simplex virus type 1 b capsid

    J Mol Biol

    (2001)
  • T.A. Jones et al.

    Improved methods for building protein models in electron density maps and the location of errors in these models

    Acta Crystallogr A

    (1991)
  • M.L. Baker et al.

    Identification of secondary structure elements in intermediate-resolution density maps

    Structure

    (2007)
  • K. Kinoshita et al.

    Diversity of functions of proteins with internal symmetry in spatial arrangement of secondary structural elements

    Protein Sci

    (1999)
  • G.J. Kleywegt et al.
  • K. Mizuguchi et al.

    Comparison of spatial arrangements of secondary structural elements in proteins

    Protein Eng

    (1995)
  • Z.H. Zhou

    Cypovirus

  • C.L. Hill et al.

    The structure of a cypovirus and the functional organization of dsRNA viruses

    Nat Struct Biol

    (1999)
  • K.M. Reinisch et al.

    Structure of the reovirus core at 3.6 Å resolution

    Nature

    (2000)
  • E. Nogales et al.

    Structure of the alpha beta tubulin dimer by electron crystallography

    Nature

    (1998)
  • W.R. Wikoff et al.

    Topologically linked protein rings in the bacteriophage HK97 capsid

    Science

    (2000)
  • S. Subramaniam et al.

    Molecular mechanism of vectorial proton translocation by bacteriorhodopsin

    Nature

    (2000)
  • K. Murata et al.

    Structural determinants of water permeation through aquaporin-1

    Nature

    (2000)
  • T. Gonen et al.

    Aquaporin-0 membrane junctions reveal the structure of a closed water pore

    Nature

    (2004)
  • B. Böttcher et al.

    Determination of the fold of the core protein of hepatitis B virus by electron cryomicroscopy

    Nature

    (1997)
  • J.F. Conway et al.

    Visualization of a 4-helix bundle in the hepatitis B virus capsid by cryo-electron microscopy

    Nature

    (1997)
  • R. Matadeen et al.

    The Escherichia coli large ribosomal subunit at 7.5 Å resolution

    Struct Fold Des

    (1999)
  • Z.H. Zhou et al.

    Electron cryomicroscopy and bioinformatics suggest protein fold models for rice dwarf virus

    Nat Struct Biol

    (2001)
  • Cited by (141)

    View all citing articles on Scopus
    View full text