Electron microscopy (EM) allows visualization of viruses in fixed cells with high resolution. High-pressure freezing for sample fixation in combination with freeze substitution and embedding in resin improves significantly the preservation of cellular structures and specifically of membranes. This advancement allows better visualization of human cytomegalovirus (HCMV) morphogenesis occurring at membranes. To obtain comprehensive information on viral phenotypes from ultrastructural images it is important to also quantify morphological phenotypes. This again can be much refined by three-dimensional visualization after serial sectioning. For elucidation of dynamic processes three-dimensional tomography is extremely helpful. We analysed interaction of HCMV particles with host cell membranes during final envelopment. Both wild-type virus and a viral mutant with impaired envelopment were analysed in fibroblasts, but also using in vivo relevant human endothelial cells and macrophages. The quantification of the EM data showed similar ultrastructural phenotypes regarding the envelopment efficiency in the different cell types indicating similar mechanisms in late stages of virus morphogenesis. Furthermore, thorough analysis of the viral assembly complex (AC) – a virus-induced cytosolic structure – by using three-dimensional visualization techniques combined with a quantitative analysis revealed that the events of final envelopment are equally distributed within the AC irrespective of different local membrane composition.
Human cytomegalovirus (HCMV) is a clinically highly relevant member of the Betaherpesvirinae subfamily in individuals with a compromised immune system. After primary infection HCMV establishes a lifelong latent state from which reactivation can occur (Bruggeman, 1993). Productive infection of HCMV is exclusively supported by human cells both in vivo and in vitro. This strict species specificity and the lack of an animal model are the major obstacles for the characterization of biological and pathogenic virus phenotypes. Definition of presumed viral phenotypes in humans is extremely difficult and has hardly led to any clear result. The description of virus phenotypes in cell culture is mostly based on cell tropisms. However, there is a need for defining of phenotypes by other independent methods, e.g. by electron microscopy (EM). Direct visualization of virus–host interactions is only possible using high-resolution imaging techniques such as EM. In fact, EM has significantly contributed to unravelling viral morphogenesis. Various techniques have been established to image viral particles and their interactions with host cell structures. Our investigations focus on the complex stages of HCMV late morphogenesis by using advanced EM technologies, such as three-dimensional visualization of the entire HCMV assembly compartment by serial sectioning and transmission electron microscopy (TEM) or scanning transmission electron microscope (STEM) tomography.
The composition of the HCMV virion is very complex. It consists of a large number of viral proteins but also cellular proteins (Gibson, 1996; 2008; Varnum et al., 2004). HCMV morphogenesis leads to four ultrastructurally distinct structures common to all herpesvirus particles: core, capsid, tegument and envelope. The electron-dense core contains the linear double-stranded DNA genome which is tightly packed in an icosahedral capsid shell build of 150 hexons and 12 pentons (Chen et al., 1999). The capsid is surrounded by the tegument which is enclosed in a lipid bilayer envelope of host cell-derived lipids containing virus-encoded glycoproteins (Britt and Mach, 1996). A wealth of information has been obtained from the analyses of other herpesviruses, showing that the generation of infectious virus particles is a highly complex process and involves nuclear and cytosolic maturation stages (Mettenleiter, 2002). Nevertheless, concerning HCMV the detailed mechanisms of tegumentation and envelopment, and the intracellular compartments involved are still under discussion and only partially known.
After genome packaging into preformed capsids in the nucleus, capsids exit the nucleus to complete their morphogenesis in the cytosol. The most accepted view of herpesvirus morphogenesis includes two distinct envelopment processes which are termed primary and secondary envelopment respectively (Mettenleiter, 2004). Primary envelopment of viral capsids occurs at the inner nuclear membrane when nucleocapsids exit the nucleus. Profound intranuclear and cytoplasmic changes occur in cells infected with HCMV. For instance, the virus infection induces infoldings of the inner nuclear membrane which in turn are used by nuclear capsids as primary budding site to exit the nucleus (Buser et al., 2007a).
These budding events lead to primary enveloped virus particles in the perinuclear space. The primary envelope is lost by fusion with the outer nuclear membrane. By this process the capsid is released into the cytoplasm. The different morphological features that can be postulated from this chain of events have been found by electron microscopy (reviewed in Mettenleiter, 2004; Buser et al., 2007a). The subsequent events are much less clear. The partially tegumented nucleocapsid has to complete its tegument and reach the site of secondary envelopment (Sanchez et al., 2000), which in case of HCMV is a juxtanuclear region in the cytoplasm often referred to as viral assembly complex (AC). The AC appears to be an organized structure containing early endosomes within its centre bordered by trans- and cis-Golgi membrane structures (Das et al., 2007). The formation of the AC during HCMV infection is characterized by the reorganization of the Golgi structure and endosomes and involves their relocalization, expansion and possibly the establishment of novel membrane compartments (Das and Pellett, 2011). Consistent with the view that it represents the site of secondary envelopment, many structural viral proteins as well as capsids accumulate at the AC. After secondary envelopment at a cytoplasmic vesicle the enveloped virion is located within a vesicle which can fuse with the plasma membrane to release the virion from the host cell. It is quite obvious that the detailed understanding of viral and cellular proteins and their interactions involved in HCMV morphogenesis could also help to define new targets for antiviral intervention and later on antiviral drug development.
The first crucial step in processing living samples for EM analysis is fixation, providing the basis for high-quality images. Ideally, the processed sample investigated in EM reflects the actual situation found in the original material. This imposes particular requirements on the rapid fixation and optimal preservation of cellular structures without generating artefacts. Recent advances in sample preparation provide the basis to obtain better preservation of biological specimen in a defined physiological state (reviewed in McDonald and Auer, 2006). Chemical fixation has been widely used as standard fixation method. However, there are two major drawbacks using this method. First, the chemical agents influence and change the sample on the cellular and molecular levels during the fixation process, which also increases the risk of artefacts. Second, the fixation process is relatively slow and thus not suitable to visualize rapidly occurring intracellular events (Schwarz et al., 1992). The disadvantage of the slow fixation process in case of chemical fixation can be overcome by using cryofixation methods (Vanhecke et al., 2010). Cryofixation is a very rapid physical fixation method which leads to the arrest of physiological events within milliseconds (Studer et al., 1989; McDonald and Auer, 2006). This is favourable for the examination of dynamic processes like vesicle trafficking or virus envelopment. Cryofixation by high-pressure freezing provides significant advantages over chemical fixation (Moor and Riehle, 1968; Kaneko and Walther, 1995; Shimoni and Müller, 1998) and even other cryofixation methods (Studer et al., 1995; McDonald, 1999) since it allows for processing of relatively large samples (such as infected cells or small tissue pieces) up to a thickness of about 200 μm. Given the adequate specific equipment and expertise, cryofixation by high-pressure freezing minimizes the formation of ice crystals during the fixation process and thus of artefacts in the fixed cells (Walther and Ziegler, 2002). The combination of high-pressure freezing, freeze substitution and embedding in resin greatly enhances the structural preservation of membranes and allows to resolve single leaflets of lipid membrane bilayers (Walther and Ziegler, 2002). Furthermore, the limitation to visualize structures only in two dimensions has been mostly overcome by the implementation of methods like three-dimensional visualization after serial sectioning and EM tomography.
These advances in EM appear to be ideal to study phases of highly dynamic processes during virus de-envelopment and envelopment at high resolution. Thus, we adapted this method to analyse HCMV morphogenesis with a special emphasis on secondary envelopment processes in infected cells.
Results and discussion
Quantitative measurement of HCMV secondary envelopment events
As mentioned earlier, secondary envelopment of HCMV has been found to take place exclusively in the area of the AC. The HCMV envelopment itself is characterized by a budding process of the nucleocapsids into small vesicles which can be dissected into several steps (Fig. 1A–E). The envelopment process starts when virus particles get in contact with a vesicle membrane. The contact appears to initiate the deformation of the vesicle membrane thereby wrapping around the tegumented nucleocapsid. The envelopment is finished by a membrane fusion process giving rise to the enveloped viral particle now within a vesicle. In contrast to the analysis of two-dimensional micrographs, three-dimensional imaging of envelopment processes allows to visualize the entire membrane that wraps around the capsid. For instance, STEM tomography of HCMV-infected cells helps to reveal the morphology of vesicles at every step of the envelopment process, including the formation of the fusion neck (Fig. 1F–H). The force which drives the curvature of the lipid bilayer during viral envelopment and the mechanism by which the fusion of the vesicle membrane at the end of the envelopment occurs have not been identified yet. The close contact of the virus particle with the vesicle membrane during the entire process of envelopment suggests that interactions of presumably viral tegument proteins attached to the capsid with proteins associated with the vesicle membrane contribute to the membrane wrapping process.
Human cytomegalovirus morphogenesis in different cell types
Human cytomegalovirus replicates in many different host cells in vivo and in vitro, including fibroblasts, endothelial cells, epithelial cells, hepatocytes and various leukocyte populations (Sinzger et al., 2008a). Fibroblasts represent a large population of cells in which HCMV replicates in vivo (Sinzger et al., 1995). They are also commonly used to study cytomegalovirus infection in cell culture as they are affordable and easy to handle, and replication of HCMV yielding high titres of progeny virus is restricted mainly to diploid human fibroblasts. Therefore, the current knowledge about HCMV morphogenesis is primarily limited to human fibroblasts (HFFs). However, HCMV morphogenesis might differ in other cell types, e.g. macrophages or endothelial cells. Endothelial cells play a central role in the spread of HCMV throughout the host, since they line the entire circulatory system (Adler and Sinzger, 2009), and their contribution to HCMV persistence and latency is under debate (Kahl et al., 2000; Jarvis and Nelson, 2002; Reeves et al., 2004). Macrophages are also susceptible for HCMV infection and are discussed to be crucial for the interaction between the virus and the host immune system, but also to contribute to the dissemination of HCMV in the infected host (Ibanez et al., 1991; Smith et al., 2004). Since fibroblasts, endothelial cells and macrophages differ in their susceptibility to HCMV and expression profiles of viral proteins during HCMV infection and furthermore distinct HCMV strains differ with regard to their tropism concerning these cells, we hypothesized that secondary envelopment processes might progress differently in these cell types. To determine the efficiency of secondary envelopment in a quantitative approach, we defined criteria to quantify the relative proportion of enveloped virus particles in an infected cell. HCMV-infected cells at late stages of infection (5 to 6 days post infection) contain a large number of virus particles at different morphogenetic stages. These stages can be defined as following: (i) enveloped virions, which have acquired their envelope and therefore are found within a vesicle, (ii) membrane-associated tegumented capsids, which are budding into a vesicle and (iii) non-enveloped tegumented cytoplasmic capsids, which do not seem to be membrane-associated.
Human fibroblasts, human umbilical vein endothelial cells (HUVECs) and monocyte-derived M2-polarized macrophages (MDMs) (Mantovani et al., 2007; Frascaroli et al., 2009; Romo et al., 2011) were infected with HCMV wild-type virus and the stages of HCMV secondary envelopment assessed 5 and 6 days post infection. The majority of virus particles within the ACs of all investigated cell types were fully enveloped and only a small portion was found in budding processes, while the amount of non-membrane-associated (naked) capsids was negligible (Table 1). More than 85% of all virus particles in the ACs in HFFs and MDMs were enveloped, which was a slightly higher proportion than in HUVECs with about 75% enveloped particles.
Table 1. Quantification of HCMV wild-type virus envelopment in the AC at 5 days post infection in HFF and 6 days post infection in HUVEC and MDM
a Given in percentage are the relative numbers of enveloped particles, non-enveloped particles attached to membranes designated as budding particles and non-enveloped particles termed as naked particles.
86.6 ± 14.8
13.0 ± 14.7
0.4 ± 1.7
74.5 ± 11.4
23.4 ± 10.7
2.1 ± 3.6
88.2 ± 7.5
10.9 ± 7.1
0.9 ± 2.7
These data prove that the proportion of naked, budding and enveloped HCMV capsids is similar in HFFs, HUVECs and MDMs at late times post infection.
Phenotyping of HCMV mutants by EM
The establishment of the quantification of HCMV secondary envelopment and the thorough analysis of the phenotype of HCMV wild-type virus allowed us to detect even slight changes in the morphological phenotype of recombinant viruses. Thus, we could recently identify the importance of the HCMV tegument protein pUL71 for secondary envelopment of HCMV in fibroblasts (Schauflinger et al., 2011; Meissner et al., 2012). To determine whether pUL71 plays a similar important role for secondary envelopment in HUVECs and MDMs as in HFFs, we quantified HCMV envelopment in these cells after infection with the pUL71-deficient mutant virus TBstop71 (Schauflinger et al., 2011). As shown in Fig. 2, infection with TBstop71 virus led to a dramatic impairment in secondary envelopment of virus particles in MDMs. A similar phenotype was also observed in HFFs and HUVECs, respectively, which is consistent with previously published data for this recombinant virus (Schauflinger et al., 2011). The quantitative analysis showed that only 25% of TBstop71 particles are enveloped in HFFs, while the vast majority of virus particles are found in a state of budding. Moreover, envelopment of dense bodies seems to be affected in TBstop71 virus-infected cells, since often accumulations of viral proteins are found near membranes. The influence of pUL71 on secondary envelopment is comparable if not higher in HUVECs and MDMs (Table 2).
Table 2. Quantification of HCMV TBstop71 virus envelopment in the AC at 5 days post infection in HFF and 6 days post infection in HUVEC and MDM
a Given in percentage are the relative numbers of enveloped particles, non-enveloped particles attached to membranes designated as budding particles and non-enveloped particles termed as naked particles.
24.6 ± 8.6
73.2 ± 9.7
2.3 ± 4.3
18.2 ± 6.6
74.2 ± 6.6
7.1 ± 10.8
16.0 ± 9.2
83.8 ± 9.1
0.3 ± 0.8
Summarizing these results, we could show that the phenotype of an envelopment-deficient pUL71 mutant is different from the parental wild-type virus, but clearly similar in different cell types. This supports the assumption that the role of HCMV pUL71 seems to be equally preserved in these cell types.
Analysis of the assembly complex
The nature of the vesicles used for HCMV secondary envelopment is not entirely clear yet. They are proposed to be Golgi-derived (Homman-Loudiyi et al., 2003) or early endosomal vesicles (Tooze et al., 1993) or to expose markers of both compartments (Cepeda et al., 2010). Based on the model of Das et al. (2007), which shows that the inner part of the AC contains early endosomes while the outer AC is composed of Golgi vesicles, we wondered whether we could determine an area within the AC in which envelopment of HCMV nucleocapsids preferably takes place. We hypothesized that budding events might occur preferentially in either the outer or the inner area of the AC. Secondary envelopment was analysed in micrographs containing the entire ACs of wild-type virus-infected HFFs. The area of the AC can usually be defined by an elliptic shape: it is enclosed by stacks of the Golgi, while mitochondria are usually not found within the AC. This elliptic area was further divided by another ellipse with the identical shape and same centre but only half of the main diameter leading to a viral AC which is artificially subdivided into an inner and an outer area (Fig. 3).
The different envelopment stages of viral particles were assessed separately for both areas and the relative numbers of enveloped, budding and naked particles were compared for the two regions (Table 3). No significant differences between the two compartments were found, suggesting that secondary envelopment does not occur at a preferential site within the AC but is rather evenly distributed.
Table 3. Quantification of HCMV wild-type virus envelopment in the inner and outer areas of the AC indicated in Fig. 3
a The relative numbers of enveloped particles, non-enveloped particles attached to membranes designated as budding particles and non-enveloped particles termed as naked particles per indicated area are given from a total of 1050 particles from 11 infected cells.
81.1 ± 11.6
18.9 ± 11.6
0.0 ± 0.0
80.0 ± 6.8
19.6 ± 6.2
0.5 ± 0.7
To overcome a potential bias introduced by the two-dimensional visualization of single thin sections of the AC we decided to address the same question by a three-dimensional approach. The three-dimensional visualization of virus-infected cells gains more and more interest since it might substantially contribute to a better understanding of virus morphogenesis. We have previously shown the application of STEM tomography to display virus budding events (Schauflinger et al., 2011). However, this method is best suitable to display small volumes up to a thickness of 1 μm and there are more convenient methods to get three-dimensional EM information in the range of an entire cell with a thickness of several micrometres. Serial sectioning is a method for three-dimensional visualization in light and electron microscopy. It has been established and improved for three-dimensional electron microscopic approaches since the use of transmission electron microscopy for biological samples was rendered possible in the 1950s (e.g. Birch-Andersen, 1955). The basic idea of serial sectioning is the production of consecutive ultrathin sections of a resin-embedded object, followed by the imaging of the area of interest (e.g. a particular cell) on every section. The micrographs are then aligned into an image stack, which is then computed into a three-dimensional image.
We used the serial sectioning approach on HCMV wild-type virus-infected HFFs, which were high pressure frozen at 5 days post infection, freeze substituted and embedded in Epon. One hundred-nanometre thin sections were prepared and a single cell with an AC typical for the late stage of infection (Fig. 4A) was traced and recorded on all sections. The single images are standard TEM micrographs (Fig. 4B), thus the resolution in X and Y directions is good enough to display virus–membrane interactions (Walther and Ziegler, 2002), while the resolution in Z direction depends on the thickness of the sections (Merchán-Pérez et al., 2009). After manual alignment and three-dimensional reconstruction of the AC, we could trace viral and cellular structures throughout the AC. Among the various vesicles, filling the centre of the AC where the centrioles are located, there are cellular organelles which are easy to identify on an ultrastructural level, such as Golgi stacks which can be found surrounding the AC, and multivesicular bodies which are predominately found in the periphery of the AC (Movie S1). Enveloped and non-enveloped viral particles were segmented in different colours (Fig. 4C). We found a total of 2836 viral particles in the AC out of which 80.3% were enveloped and inside a vesicle (Table 4). Additionally, we observed 1564 dense bodies (not shown). The three-dimensional representation of virus particle distribution also enables to compare the stages of envelopment between different volumes of the AC, comparable to the abovementioned determination of the location of HCMV envelopment in two dimensions (Fig. 3). We determined the number of enveloped and non-enveloped virus particles in seven arbitrarily defined distinct volumes within the AC, which comprised 8 μm3 each (Fig. 4D): one volume in the centre of the AC (# 7), three volumes in close proximity to the nucleus (# 1–3) and three volumes more distant from the nucleus (# 4–6). Seventy-six per cent to 89% of viral particles in these volumes were enveloped, and the percentage of particles undergoing secondary envelopment did not significantly differ between those volumes (Table 4). This again suggests that secondary envelopment is not preferably occurring in a certain region of the AC. While the numbers are derived only from one three-dimensional reconstruction of one AC, they are consistent with our reproduced data from the two-dimensional approach.
Table 4. Quantification of HCMV wild-type virus envelopment in the entire AC and in seven individual volumes of the AC based on the 3D reconstruction in Fig. 4
No. of particles analysed
No. of enveloped particles
% values of enveloped particles
In conclusion, serial sectioning is successfully applicable for the quantification of virus particles and their spatial distribution in the host cell. This bears the potential to reveal the phases of dynamic mechanisms of virus–membrane interaction within the cell and the proteins involved in this process. In contrast to the three-dimensional reconstruction of the AC of HCMV-infected cells by Z-stack confocal immunofluorescence microscopy (Das et al., 2007), serial sectioning TEM enables to visualize the ultrastructure of organelles and individual viral particles. For the schematic representation of particles and structures within the AC, Das et al. (2007) had to combine information gained from fluorescence microscopy and ultrastructural analyses (e.g. Severi et al., 1988).
An alternative approach to visualize infected cells three-dimensionally is the focused ion beam/scanning electron microscopy (FIB/SEM) tomography. The development of FIB technology goes back to the year 1974, when it has been successfully used for microfabrication (Seliger and Fleming, 1974). FIB/SEM tomography is also applied for TEM sample preparation (Giannuzzi and Stevie, 1999; Rigort et al., 2012), and has been increasingly used in the last 10 years for three-dimensional visualization of biological samples (e.g. Denk and Horstmann, 2004; Drobne et al., 2005; Knott et al., 2008; Hekking et al., 2009). A focused ion beam is used to very precisely mill material off a resin block which contains the resin-embedded sample. The resulting surface (block face) is recorded in a SEM. When appropriate imaging methods are used, FIB/SEM allows for three-dimensional analysis of whole cells with a resolution good enough to see the single leaflets of the lipid bilayer. We have currently shown that FIB/SEM tomography enables three-dimensional imaging of high-pressure frozen cells with TEM-like resolution (Villinger et al., 2012). We think that three-dimensional visualization of HCMV-infected cells by FIB/SEM tomography will eventually further improve our insight into HCMV morphogenesis.
In conclusion, we have shown that by combining different advanced methods of sample preparation and EM we can obtain clearer and additional information of the HCMV morphogenesis in infected cells and we strongly believe that only by combining molecular virology, cell biology, biochemistry and multiple imaging techniques will we finally be able to fully understand the secrets of HCMV morphogenesis which then may lead to new approaches for antiviral intervention.
Cells and viruses
Human foreskin fibroblasts (HFFs) were used before passage 23 and maintained in minimal essential medium (Invitrogen, Darmstadt, Germany) supplemented with 10% fetal calf serum (Invitrogen), 1× nonessential amino acids (Biochrom AG, Berlin, Germany), 2 mM l-glutamine (Biochrom AG), 100 U of penicillin and 100 μg of streptomycin (both Invitrogen). Human umbilical cord endothelial cells (HUVECs) were purchased from Clonetics (BioWhittaker, Walkersville, MA, USA) and cultured in endothelial cell basal medium supplemented with 5% fetal calf serum and growth factors (EGM-MV SingleQuots; BioWhittaker). Monocyte-derived M2 macrophages (MDMs) were produced as described earlier (Frascaroli et al., 2009). Briefly, monocytes were isolated from buffy coats of HCMV-seronegative blood donors by negative immunoselection (monocyte isolation kit II; Miltenyi Biotec, Bergisch Gladbach, Germany) and differentiated into macrophages by cultivation for 7 days in RPMI 1640 supplemented with 10% fetal calf serum, 2 mM glutamine, antibiotics and 100 ng/ml macrophage colony-stimulating factor (R&D Systems, Minneapolis, MN, USA) on hydrophobic Lumox dishes (Greiner Bio-One GmbH, Frickenhausen, Germany). The wild-type virus was reconstituted from the HCMV bacterial artificial chromosome (BAC) clone TB40-BAC4 (Sinzger et al., 2008b). The generation and propagation of the UL71 stop mutant virus, TBstop71, were described previously (Schauflinger et al., 2011).
For electron microscopy, cells were grown on carbon-coated sapphire discs (3 mm in diameter; Engineering Office M. Wohlwend GmbH, Sennwald, Switzerland), infected at an MOI of 0.5 and cryofixed by high-pressure freezing at either 5 days post infection for HFFs or 6 days post infection for HUVECs and MDMs. For the fixation, the 50 μm thick sapphire disc with the infected cells grown on it was clamped in between two aluminum planchettes and covered with 1-hexadecen so that the cells were protected in a 100 μm cavity by one of the planchettes as described earlier (Buser et al., 2007b). These sandwiches were frozen using a Wohlwend HPF Compact 01 high-pressure freezer (Engineering Office M. Wohlwend GmbH). Freeze substitution and Epon embedding was performed as described by Walther and Ziegler (2002) with some modifications. The substitution medium consisted of acetone with 0.2% osmium tetroxide, 0.1% uranyl acetate and 5% of water for good contrast of the membranes. During 17 h, the temperature was exponentially raised from −90°C to 0°C. After substitution, the samples were kept at room temperature for 1 h and then washed twice with acetone and stepwise embedded in Epon (polymerization at 60°C within 72 h). The sapphire disc was then removed, so that the cells remained at the surface of the Epon block. Thin sections with a nominal thickness of 70 nm were cut with a microtome (Leica Ultracut UCT ultramicrotome) using a diamond knife (Diatome, Biel, Switzerland), and mounted on formvar-coated single slot copper grids for transmission electron microscopy. Samples were imaged with the JEOL JEM-1400 TEM equipped with a CCD camera at an acceleration voltage of 120 kV. To quantify the morphogenesis stages in two-dimensional micrographs, fully enveloped, budding and non-enveloped virus particles, respectively, were counted in micrographs from randomly selected virus-infected cells from at least three independent experiments for each cell type and virus respectively. All micrographs used for this analysis were taken in randomly selected regions of the viral AC. For serial sectioning TEM, consecutive sections with a nominal thickness of 100 nm were processed and imaged as described above. Stacking and alignment of serial images from these sections were carried out using Avizo 6.3 (Visualization Sciences Group, Burlington, MA, USA). For STEM tomography, 500 nm-thick sections were imaged in a Titan (FEI, Eindhoven, the Netherlands) 300 kV field emission transmission electron microscope in scanning transmission mode using a high-angle annular dark-field detector (Fischione Instruments, Horley, UK); images were recorded at tilt angles from −70° to +70°; the volume was reconstructed by weighted back projection using the imod software (Kremer et al., 1996), with the help of 15 nm colloidal gold as fiducial markers for alignment. Three-dimensional modelling of cellular and viral structures in stacks of serial images as well as in STEM tomograms was carried out manually using Avizo 6.3 (Visualization Sciences Group).
We thank Eberhard Schmid for expert technical assistance. This work was supported by the DFG priority research program SPP1175.