Visualizing cellular processes at the molecular level by cryo-electron tomography

Cellular activities rely on the concerted actions of macromolecular complexes that function within dynamic networks. For instance, the cell cytoskeleton is remodeled within fractions of a second, thus modulating cell shape and function. Consequently, at the molecular level, cellular architecture rapidly changes during the cell cycle and throughout various biological processes. Despite the wealth of information that exists on cellular components and their dynamic properties, our current understanding of the functional interactions that lead to a given cellular process is rather limited. Developing the experimental tools necessary for the analysis of complex and variable supramolecular structures inside cells is crucial to rectify the situation. Microscopical imaging techniques, which now offer resolution at the previously unattainable nanometer scale, would thus be expected to provide novel insight into the local and global organization of functional modules and networks inside cells (Robinson et al., 2007).

Microscopy techniques have traditionally been central in driving cell biology forward. Fluorescence microscopy and confocal laser-scanning microscopy revolutionized our thinking and opened up a large set of possible strategies for investigating cellular processes (reviewed by Schliwa, 2002). In particular, the introduction of green fluorescent protein (GFP) and its analogs has allowed kinetic measurement of proteins in living cells (Tsien, 1998). Additionally, the development of ultra-high-resolution fluorescence microscopy, such as PALM (Betzig et al., 2006), STED (Hell, 2003) and structured illumination (Schermelleh et al., 2008), allowed the visualization of individual macromolecular complexes such as the nuclear pore complex (NPC) (Schermelleh et al., 2008). However, such approaches can focus on only a limited number of proteins at a time, depending on the number of fluorophores available. The molecular architecture can, however, be reconstructed in three dimensions and at high resolution by electron microscopy (EM), particularly cryo-electron tomography (cryo-ET), therefore complementing fluorescence-microscopy techniques.

In this Commentary, we focus on the principles and implementation of cryo-ET in the field of cell biology. We demonstrate the possibility of resolving three-dimensional (3D) cytoskeleton networks within intact cells by showing actin cytoskeleton networks in the native context of membranes, vesicles and other molecular complexes. The potential of using cryo-ET for following viruses as they infect cells will be shown and discussed. We also consider how the application of cryo-ET to visualize intact nuclei, in combination with 3D-averaging procedures, has yielded a 3D structure of active NPCs. Finally, we consider future prospects for cryo-ET. We focus on the application of this technique to eukaryotic cells; however, it should be noted that cryo-ET has been successfully used for the study of prokaryotes and viruses (Borgnia et al., 2008; Briggs et al., 2006; Gan et al., 2008; Grunewald et al., 2003; Kurner et al., 2005; Lieber et al., 2009; Liu et al., 2008; Morris and Jensen, 2008).

Uniquely among imaging techniques, cryo-ET can generate 3D information concerning the macromolecular architecture of cells in an unperturbed state (Cyrklaff et al., 2007; Li et al., 2007; Nickell et al., 2006). Using this technique, one can depict unique cellular states and reconstruct molecular networks. Through vitrification by rapid freezing, biological material can be physically fixed, ensuring close-to-life conditions in samples prepared for cryo-ET (Dubochet et al., 1988). Because neither chemical fixation nor staining is needed, the delicate cellular landscape is preserved during sample preparation and accurately depicts in vivo conditions.

In cryo-ET, 3D structures of specimens are retrieved from 2D micrographs. Owing to the large depth of focus, electron micrographs are essentially two-dimensional (2D) projections of a 3D object in the direction of the electron beam. Consequently, features of the sample are superimposed and cannot be separated, in contrast to the situation in confocal laser-scanning microscopy. Nevertheless, the three-dimensionality of an object can be retrieved by recording a series of projections at varying angles and synthesizing these projections into a 3D density map – that is, a tomogram (Fig. 1) (Frank, 1992). In practice, the different projection images are collected by tilting the specimen incrementally around a single axis inside the electron microscope that is perpendicular to the optical axis of the electron beam (Fig. 1A). This tilt series is then aligned to a common frame of reference, followed by calculation of the tomogram (Fig. 1A), most commonly by using a `weighted back-projection' algorithm (Radermacher, 1992).

The resolution of a tomogram is directly dependent on the angular increment between two adjacent projections and on the total number of images that are obtained (Koster et al., 1997). Therefore, the aim is to collect as many tilted projections as possible, covering the widest possible angular range, but keeping the electron dose at a sub-critical level. Thus, the cumulative electron dose in the entire tilt series must be kept within tolerable limits, typically not exceeding ∼6000 e^–/nm², to prevent radiation damage to the biological specimen. Furthermore, because of technical limitations, the tilt series cannot cover the entire spectrum of views and is limited to ±70°. In practice, a typical tilt series consists of 80-100 exposures and covers only 120°-140° of the 180° angular range. Consequently, elongation of features along the beam axis is evident because of a missing `wedge' in the 3D Fourier space (Frank et al., 2002). Overall, to minimize the exposure time and to increase the accuracy of the process, data acquisition must be fully automated, and relies on computer control (Dierksen et al., 1993; Dierksen et al., 1992).

A further limitation is that the application of cryo-ET to eukaryotic cells is restricted to relatively thin regions. When the object is thicker than the mean free path of an electron [200 nm and 350 nm for 120 keV and 300 keV (acceleration voltage of electrons in the electron microscope), respectively (Grimm et al., 1998)] multiple scattering events substantially degrade the image quality, despite the use of high to medium acceleration voltage (300 keV) and an energy filter to minimize this effect (Grimm et al., 1996). As a consequence, samples thicker than 1 μm can barely be studied in toto, and require cryosectioning before they can be subjected to tomographic analysis. Several laboratories have devoted substantial effort to establishing freeze-hydrated cryosectioning procedures. The feasibility of this approach has been shown using cryosectioned rat liver cells, mouse epidermis, and human epidermis and cardiomyocytes (Al-Amoudi et al., 2004; Al-Amoudi et al., 2007; Castano-Diez et al., 2007; Gruska et al., 2008; Hsieh et al., 2002; Salje et al., 2009), but it is still technically demanding and often gives rise to sectioning artifacts (Al-Amoudi et al., 2005).

For decades, EM of actin cytoskeletons was performed on sections of chemically fixed, detergent-extracted cells shadowed with metal (Brown et al., 1976; Svitkina et al., 1997), or by using adherent cells, which have an apical surface that has been mechanically removed, for EM analysis (Hartwig et al., 1989; Heuser and Kirschner, 1980). Although these methods provided important insight into the architecture of actin networks (Small et al., 1994), the spatial resolution of the structures revealed by these methodologies was limited, especially in the third dimension. For instance, connections between actin filaments were difficult to resolve in detail after metal decoration or replica formation. Moreover, distortion of the actin network by detergent treatment does not permit the study of anchorage of actin filaments to membranes. Most importantly, by means of total internal reflection fluorescence (TIRF) microscopy using specific probes for actin nucleation, growth and branching (e.g. Arp2/3) it was shown that, in highly motile cells, the proteins associated with cytoskeletal networks have half-lives in the order of 7-10 seconds (Bretschneider et al., 2004). Therefore, it is to be expected that actin-bundling and -crosslinking proteins, which are crucial for maintaining the architecture of the actin network, might dissociate from actin filaments and redistribute within a specimen during the course of traditional-preparation EM procedures.

In live cells, actin structures rapidly reorganize during motility, endocytosis and cytokinesis (Dalous et al., 2008; Kaksonen et al., 2004; Pantaloni et al., 2001; Pellegrin and Mellor, 2007; Walpita and Hay, 2002). Because of these dynamics, immediate arrest of the cellular processes in intact cells is a prerequisite for obtaining faithful information regarding actin networks. Electron tomography of vitrified but otherwise unaltered cells has proven to be a key technique for the 3D reconstruction of actin architecture (Medalia et al., 2002). It was demonstrated that cryo-ET of intact Dictyostelium discoideum cells could reveal the connections of the actin-filament network with the plasma membrane (Fig. 1B, upper panel), as well as massive actin filaments branching at various angles (Fig. 1B, lower panels), without the need for chemical fixation or heavy-metal decoration (Medalia et al., 2002). Similar views can be obtained when cells of higher eukaryotes are studied (Fig. 1C).

In the future, our understanding of cell motility and other cytoskeleton-dependent processes will be increased by investigating the 3D organization of the actin system in relation to specific functional states, such as during consecutive steps of particle uptake by a phagocyte or at cell-adhesion sites. The possibility of arresting cells instantly also allows the investigation of rapid cellular processes such as filopodial protrusion, which is discussed below.

Filopodia are finger-like plasma-membrane protrusions that are involved in adhesion to the extracellular matrix (ECM), sensing the environment and cell-cell signaling (Chhabra and Higgs, 2007; Mattila and Lappalainen, 2008). Filopodia also represent an excellent model system for describing the process of actin-driven membrane protrusion. These structures grow at their tips through the assembly of actin and are stabilized along their lengths by a core of bundled actin filaments (Pollard and Borisy, 2003). Given their relatively low thickness (150-400 nm), filopodia are also excellent cellular structures for study by cryo-ET. With this approach, the organization of actin filaments and membranes to which the actin network is anchored can be carefully analyzed in an unperturbed state.

We used cryo-ET to track the length and relative position of individual filaments within D. discoideum filopodia, which allowed a quantitative analysis of actin filaments. The data revealed that actin filaments in these fast-moving cells are not continuous throughout the entire protrusion (Fig. 2) (Medalia et al., 2007). Importantly, it was shown that the filopodial tip comprises many short filaments that interact with the membrane at their distal and proximal ends. These filaments, arranged at the tip in a cone shape, have been suggested to provide the driving force that pushes the membrane forward (Gerisch and Weber, 2007). Their location and length support the notion that sites of de novo actin-filament nucleation and growth are confined to the tip of the filopodia (Faix et al., 1992; Faix and Grosse, 2006; Medalia et al., 2007). It is noteworthy that no vesicles are found within filopodia that are visualized by cryo-ET, implying that the membranes needed for the formation of protrusions are supplied at a distance from the filopodial tip. The presence of Dia2, a formin with actin-nucleating activity that is important for filopodia formation and maintenance (Schirenbeck et al., 2005), at the tips of filopodia in D. discoideum implies that actin filaments undergo nucleation through filopodia protrusion. When filopodia protrude, the actin filaments grow and are then bundled while laterally connecting to the membrane along the filopodial shaft (Medalia et al., 2007). In general, we found that D. discoideum filopodia are characterized by a discontinuity of actin filaments along the filopodial axis (Fig. 2C). However, some transverse filaments connect to the membrane and the shaft filaments are found in the tip-shaft zone.

This analysis can thus be explained by the `sequential-nucleation model' (Medalia et al., 2007), which proposed that the sites of de novo nucleation and growth of actin filaments are confined to the filopodium tip. The short filaments located at the tip detach from and reattach to the cell membrane with their distal and/or proximal ends, thus enabling actin polymerization. The growing filaments are then bundled and laterally connect to the cell membrane along the filopodia shaft. Within the shaft zone, actin filaments are bundled and axially oriented (Fig. 2C). The unique organization and arrangement of D. discoideum filopodia can presumably be attributed to the fast motility of these cells. That is, the apparent discontinuity of actin filaments might be a property of filopodia in fast motile cells, which differs from filopodia in other adherent cells, which are characterized by continuous actin filaments.

Several viruses use an endocytic mechanism to enter cells prior to remodeling cortical actin (Greber, 2002; Munter et al., 2006). In addition, filopodia and other cellular protrusions are susceptible to viral docking, which eventually leads to cell infection (Clement et al., 2006). Cryo-ET can provide a unique tool for studying infected cells, supplying unprecedented information on the stages of viral assembly and maturation within cells. In a pioneering study, Maurer et al. showed snapshots of the entry of Herpes simplex virus 1 (HSV-1) into cells (Maurer et al., 2008). By means of cryo-ET, the authors showed that cytosolic capsids of HSV-1 were located between actin bundles within PtK₂ cells. They also identified capsids between individual actin filaments in the cell cortex. Furthermore, the data revealed that the virus does not induce local depolymerization of actin in its vicinity, on the basis of morphological appearance of the cell cortex, but rather might be involved in remodeling the dense cytoskeletal network as shown by Clement et al. (Clement et al., 2006) (Fig. 3).

Cryo-ET is primarily a static tool. However, by collecting a large number of datasets and correlating them with pre-existing information, one can acquire information about a dynamic process at the molecular level, as has been described above for filopodia and virus-infected cells. Thus, the remodeling and structural changes of macromolecular complexes that transpire during cellular processes can be examined. Additionally, a detailed 3D reconstruction of macromolecular complexes in situ can be achieved by combining cryo-ET with 3D-averaging procedures (Bartesaghi et al., 2008; Bostina et al., 2007; Forster et al., 2005).

Such a hybrid technique (cryo-ET and 3D-averaging approach) was applied to NPCs, which are large molecular machines that are embedded in the nuclear envelope and connect the nucleoplasm with the cytoplasm by means of an aqueous channel. NPCs, which are composed of hundreds of proteins (Alber et al., 2007) arranged in pseudo-eightfold rotational symmetry, function as a selective barrier (Rout et al., 2000) by allowing small molecules and ions to diffuse freely while mediating the passage of large molecules in an energy-dependent manner. Although work on the NPC structure using EM began in 1950 (Callan and Tomlin, 1950), only at the end of the twentieth century were the main components of the structure revealed (Akey and Radermacher, 1993; Brohawn et al., 2009; Yang et al., 1998). Despite dimensional differences between species, the basic architecture of the NPC is conserved; the consensus structure consists of a central spoke ring that is confined by a cytoplasmic and a nucleoplasmic ring. Eight cytoplasmic filaments and a nuclear basket composed of eight filamentous structures join to form a distal ring. Owing to its sheer size, the NPC presents a major challenge for structural determination. Applying cryo-ET to intact nuclei ensures that the NPC is arrested in its active form, as is evident from the preservation of a nucleocytoplasmic gradient of Ran-GTP in isolated nuclei (Becskei and Mattaj, 2003; Becskei and Mattaj, 2005; Gorlich et al., 2003). The nuclear envelope can be viewed as an ellipsoid with NPCs embedded in its surface in all possible orientations; thus, extracting these elements in silico, followed by 3D alignment and averaging, results in isotropic resolution, i.e. in all three dimensions. Using cryo-ET, we have resolved the NPC to 8-9 nm, in which some of the flexible filaments of the NPC were observed, in addition to the scaffolding features of the complex (Beck et al., 2004). An improvement in resolution (<6 nm) was later achieved when the structure of the D. discoideum NPC was resolved without imposing eightfold symmetry (Beck et al., 2007). In this study, snapshots of the trajectory of cargo transported through the NPC were visualized. Recently, work from our laboratory has shown that the application of a similar approach using nuclei from human fibroblasts (Fig. 4) yielded the first insight into the structure of the human NPC (Elad et al., 2009). Although the resolved structures of the D. discoideum NPC and the preliminary structure of the human NPC share several common features, such as the outer and inner diameters (∼120 nm and ∼50 nm, respectively), they differ in height and protein-density distribution, suggesting differences in protein positions (Elad et al., 2009). Currently, analysis of intact nuclear envelopes by means of cryo-ET is limited to a resolution of 5-8 nm, which restricts structural interpretation to the level of determining the position of subcomplexes. Increasing the resolution of tomograms and the application of specimen-thinning techniques, such as cryosectioning (see above) and focused ion beam (see below), will enable a step forward in understanding the functional organization of the nuclear envelope.

Cryo-ET is the method of choice in acquiring an insight into the molecular organization of cells and cellular components, such as filopodia and actin filaments at the cortex of the cell, and to track the entry of viruses into cells. Additionally, it allows determination of the 3D structure of large supramolecular assemblies in situ, as we demonstrated for the NPC, at a medium resolution of 4-6 nm.

A major challenge facing cryo-ET concerns the identification of macromolecular complexes within a cellular context. The different orientations of macromolecules and the current resolution of cellular tomograms prohibits unbiased identification of many molecular complexes, although some successful template-matching approaches have been introduced (Frangakis et al., 2002; Ortiz et al., 2006). These procedures aim to identify specific macromolecular complexes in vivo on the basis of their structural fingerprint, by searching for in-vitro-determined structural complexes in a tomogram (van Heel et al., 2000). An approach based on electron-dense labeling must be developed to design a clonable tag that can be genetically conjugated to proteins, which would facilitate their localization in cryo-tomograms – that is, we are in need of a GFP analog for cryo-electron microscopy. An elegant example of a clonable tag is metallothionein, a cysteine-rich protein that has been shown to bind to multiple heavy atoms and can be detected by an electron beam (Mercogliano and DeRosier, 2007). Such a labeling strategy would provide a general solution for identifying complexes whose structure is not yet determined or that intimately interact to form large assemblies. To make cryo-ET applicable not only to cellular protrusions and thin regions of the eukaryotic cell but also to thicker samples, there is a need to develop a reliable freeze-hydrated artifact-free sectioning technique that can be applied to tissues and cells to produce optimal (thinner than 500 nm) biological samples for cryo-ET. Alternative micro-dissection techniques that involve using a focused ion beam to mill frozen samples are currently being developed. In these approaches, gallium ions (Ga⁺) are directed onto the frozen cell or tissue sample at a specific angle and, by process of `sputtering', selected parts of the sample can be removed; this is known as ion-beam milling. Notably, this process does not cause major artifacts below 30 nm from the upper level of the milled surface (Marko et al., 2006; Marko et al., 2007). Realization of this technology would open a window for the entry of cryo-ET into other branches of biology and might provide, for instance, a bridge between structural and developmental biology.

Another direction of technical development lies in correlating fluorescent and cryo-electron-microscopy images; this would greatly help in identifying attractive locations within the cell for investigation by cryo-ET (Sartori et al., 2007; Schwartz et al., 2007), and would allow for the reconstruction of important cellular structures. This approach would permit the identification of specific states of cellular processes and would therefore eventually produce structural snapshots of such processes. Moreover, developing automated fast algorithms would allow for a shortening of current time-consuming procedures, making analysis more robust.

In the future, the application of other correlative approaches – such as combining cryo-ET with atomic force microscopy – would also enable correlations between physical changes in the cell, force measurements and structural information. It is expected that such hybrid methods will lead to platforms that can provide deeper insight into cellular processes. Complementary information from a variety of techniques will thus be combined to reconstruct meaningful cellular density maps (Robinson et al., 2007). With advanced instrumentation, such as advanced charge-coupled device (CCD)-camera detectors and dual-axis tilting devices, the prospects are good that higher and more isotropic resolutions of 2-3 nm can be attained. Therefore, we foresee that cell biology will increasingly rely on high-resolution 3D imaging techniques, in conjunction with other approaches.