Origins of Enterovirus Replication Organelles Established by Whole-Cell Electron Microscopy

To assess the resolving power of SBF-SEM on enterovirus-infected cell ultrastructure, Vero E6 cells were infected with coxsackie B virus 3 (CVB3) and fixed at 6 h postinfection (hpi). Our previous studies with electron tomography showed that enterovirus ROs with different morphologies (i.e., single-membrane tubules, double-membrane vesicles, and multilamellar vesicles) are present at this time point. After fixation, cells were prepared and imaged by SBF-SEM along with mock-infected cells for comparison. These data revealed expansive fields of ROs in infected cells, which dominated a large volume of the cell cytoplasm (Fig. 1A; see also Movie S1 in the supplemental material). Although the individual membranes of single- and double-membrane ROs were not resolvable by SBF-SEM under the imaging conditions used, single-membrane ROs (white arrowheads) were distinguishable from double-membrane (hatched arrowheads) or multilamellar (black arrowheads) ROs by the apparent thickness of the stained membrane (Fig. 1A, left). To support this interpretation, SBF-SEM images were compared with higher-resolution TEM images collected from the same type of samples (Fig. S1A and B), confirming that enterovirus ROs were discernible in the SBF-SEM data. Cellular organelles like the ER, lipid droplets, cis-Golgi cisternae, and trans-Golgi network, could be unambiguously identified in SBF-SEM reconstructions of mock-infected cells through comparisons with TEM images (Fig. S1C). In infected cells, prominent cellular organelles, like the nucleus, ER, and mitochondria, were also identified (highlighted in Fig. 1A, right). While Golgi cisternae were absent throughout these infected cell volumes, Golgi cisternae were clearly visible in all mock cell volumes assessed (n = 13). These whole-cell visualizations confirm that Golgi apparatus cisternae disintegrate during enterovirus infection, as was previously inferred from live-cell imaging studies monitoring Golgi markers across infection, and two-dimensional (2D) ultrastructural data at early and late infection time points (12, 17, 19, 22).

In order to obtain quantitative insights into the abundance and volumes of enterovirus ROs at this stage of infection, the three-dimensional (3D) distribution of the viral ROs was highlighted by manual segmentation of the RO regions, nucleus, and the cytoplasm in two reconstructed cells, A and B (Fig. 1B). CVB3-induced RO clusters accounted for large parts of the cytoplasmic volume, at 16% of in cell A (104 μm³) and 6% in cell B (37 μm³). The difference likely reflects a more advanced infection stage in cell A, which is supported by the appearance of extensive ER dilation (Fig. 1A) characteristic of later stages of enterovirus infection (12, 23). The nanometer resolution achievable by SBF-SEM allowed RO clusters in close proximity to be segmented apart to analyze their size distribution (Fig. 1C). For both cells, most ROs were not homogeneously distributed across the cytosol but instead accumulated in a single large cluster (bright green segmented volumes, 98 μm³ in cell A and 27 μm³ in cell B) proximal to the nucleus. Most of the remaining ROs could be found in relatively small clusters (colored), with more than 90% of these clusters smaller than 0.1 μm³ in both cells (Fig. 1C).

Our previous live-cell imaging data demonstrated that the onset of Golgi apparatus disintegration in CVB3 infection is concomitant with the accumulation of viral protein (e.g., 3A) in the Golgi region (19, 21). Intriguingly, the first foci of 3A protein arise in the cell periphery, often ahead of the dramatic 3A accumulation in the Golgi region that expands to dominate the cell cytoplasm as infection progresses (Movie S2, white and black arrowheads, respectively). It is unknown whether both the peripheral and Golgi-proximal 3A foci that emerge during this phase represent developing ROs. One possibility is that the small peripheral 3A foci that form early in infection are replication independent and that these 3A accumulations are merely the result of translation of the incoming vRNA delivered by uncoating of virus particles. To test this, we inhibited vRNA synthesis by treating infected cells with guanidine hydrochloride (GuHCl) and monitored these cells by live-cell imaging. The CVB3 3A was visualized by utilizing a split-GFP system (21). Neither peripheral nor any other 3A-GFP signal emerged under GuHCl inhibition, demonstrating that their formation is not a product of translation of only incoming vRNA but depends upon vRNA replication (Fig. S2).

Next, we investigated whether virus-induced membrane structures could be detected at both peripheral and Golgi-associated 3A foci. To do so, we imaged living cells expressing mCherry-GM130, a cis-Golgi marker, and infected them with CVB3 encoding split-GFP-tagged 3A (21). Confocal z-stacks were collected of cells with emerging 3A-GFP signal but at different stages of Golgi disassembly (∼5 hpi), which were then fixed and prepared for SBF-SEM. Acquired SBF-SEM volumes were manually segmented to highlight Golgi cisternae and RO clusters based solely on morphological features, which could then be compared with 3D reconstructions of corresponding confocal data (Fig. 2A, workflow illustrated in Movie S3).

A striking correspondence was found between 3A-GFP foci in the confocal data and the location of ROs highlighted by the SBF-SEM segmentation, although in some cases, a shift in the relative positions of 3A-GFP foci and RO clusters was apparent (e.g., hatched arrowheads, Fig. 2A). This may be explained by migration of RO clusters in the delay between confocal z-stack acquisition and fixation (ca. 10 min for CLEM data sets), as peripheral 3A-GFP foci are often dynamic (Movie S2, hatched arrowheads).

The cell presented in Fig. 2 represents an early stage of infection encompassing the emergence of the first peripheral 3A-GFP foci. These foci are largely distal to the Golgi region, where minimal 3A-GFP signal is apparent. The Golgi apparatus appears relatively intact at this stage of infection, as recognizable Golgi cisternae could be detected. All RO clusters at this stage (Fig. 2B, white arrowheads) were observed in close contact with the ER (black arrowheads), with apparent membrane continuities between them (hatched arrowheads). The existence of membrane connections between peripheral RO clusters and ER was confirmed in higher-resolution TEM images of parallel samples (Fig. 2C). These data suggest that 3A foci that emerge in the cell periphery correspond with ROs that are derived from ER membranes.

Following the emergence of peripheral 3A signal, 3A-GFP accumulates in the Golgi region (Movie S2, white and black arrowheads, respectively). An example of a cell fixed at this stage of infection is shown in Fig. 3A. The 3D confocal model reveals 3A-GFP signal in close proximity to mCherry-GM130 signal, which corresponded to ROs and Golgi cisternae, respectively, in SBF-SEM data (Fig. 3A).

ROs in the Golgi region (Fig. 3B, boxed region in 3A, white arrowhead) were connected to membranous regions that bridged the trans-Golgi network (hatched arrowhead) and RO clusters (black arrowhead). Peripheral RO clusters (Fig. 3A, asterisks) were found to be associated with the ER (Fig. S3), demonstrating that ER- and Golgi-associated ROs can coexist in infected cells and suggesting that ER-RO interactions persist at later stages of infection. To examine Golgi-associated ROs at higher resolution, samples prepared for SBF-SEM were imaged in parallel by TEM (Fig. 3C). In addition to Golgi cisternae (Fig. 3C, black arrowhead) and ROs (white arrowheads), TEM images revealed continuities between trans-Golgi network tubules and single-membrane ROs (hatched arrowhead). Altogether, these data suggest that early RO biogenesis occurs at seed points distributed across the ER, followed by RO formation at trans-Golgi membranes.

Beyond the ER- and Golgi-associated RO foci highlighted by the 3A-GFP signal, significant spatial relationships between ROs and other cellular organelles, including endosomes, lysosomes, autophagosomes, and mitochondria, were not clearly apparent in any cells analyzed. However, extensive contact regions could be found between RO clusters and lipid droplets (LDs), which in some cases appeared to be surrounded by ROs (Fig. 4, arrowheads). Interestingly, a qualitative analysis of the number of LDs in SBF-SEM cell reconstructions of cells fixed at late infection stages (n = 6) hinted at a possible inverse correlation between the infection stage (assessed by RO abundance) and the LD content. A quantitative analysis of LDs, visualized with Oil Red O stain by light microscopy, confirmed that LDs were depleted across CVB3 infection in our experimental setup and showed a particularly rapid drop early in infection (between 0 and 4 hpi) both in LD counts and the total LD area per cell (Fig. S4), which aligns with previous observations in enterovirus-infected cells (9, 24). Together, our results suggest that the extensive contact regions between ROs and LDs could represent sites for the transfer of critical lipids to ROs, supporting their formation and proliferation.

Vero E6 or BGM cells were cultured using Dulbecco’s modified Eagle’s medium (Gibco) supplemented with 10% fetal calf serum, penicillin, and streptomycin, and maintained at 37°C in 5% CO₂. BGM(S1–10) cells, which express GFP(S1–10) and the puromycin resistance gene (Pac), have been described in reference 21. BGM(S1–10) cell culture medium was supplemented with an additional 30 μg/ml puromycin as a selection agent.

CVB3 3A(S11aa2), which encodes a 3A with the S11 region of GFP inserted after amino acid position 2, has been described in (21). CVB3 wild type (wt; strain Nancy) was generated as described in reference 19. For infection, cells were inoculated for 1 h with CVB3 wt or CVB3 3A(S11) at MOIs of 50 and 5, respectively, after which the inoculum was removed and fresh medium was added. At specified time points after infection, cells were imaged by light microscopy and/or prepared for electron microscopy by chemical fixation.

BGM(S1–10) cells were grown to ∼35% confluence in glass-bottom 4-chamber 35-mm dishes (CELLview) and transduced with murine leukemia virus (MLV) mCherry-GM130 particles (described in reference 21). Transduced cells were infected with CVB3 3A(S11) 18 to 24 h later. Cells were washed with FluoroBrite medium (Thermo Fisher Scientific) supplemented with 8% fetal calf serum (FCS) and 25 mM HEPES just prior to imaging. LM data were collected using a Leica SP5 confocal microscope equipped with a HyD detector and a 63× (1.4 numerical aperture [NA]) oil immersion objective, with the confocal pinhole adjusted to 1 airy unit for GFP emission (95.56-μm pinhole). For time-series imaging, positions of interest (xyz) were marked and imaged sequentially at 5-min intervals. For the duration of all imaging, cells were maintained at 37°C and 5% CO₂ in a live-cell chamber. For correlative light and electron microscopy, cells were grown in glass-bottom MatTek dishes with an etched alphanumeric coordinate system (MatTek Corporation) to facilitate the relocation of the regions identified during light microscopy. Cells were monitored until the point of interest, at which time a confocal z-stack was acquired, after which samples were immediately fixed for electron microscopy. To minimize the time between the first image acquisition of each z-stack and fixation (ca. 10 min total), one field of view was imaged per sample.

Vero E6 cells grown on coverslips were infected with CVB3 wt (MOI 100) and fixed with 3% paraformaldehyde for 1 h at different infection times. Mock-infected cells were also fixed as a control. After washing with PBS, the coverslips were incubated for 1 h at room temperature with a freshly prepared mixture of Oil Red O (Sigma) (0.5% [wt/vol] in isopropanol) and double-distilled water (6:10) to stain LDs, washed with PBS, and incubated with the nuclear stain Hoechst. The samples were then imaged in a Leica DM6 wide-field microscope with a 40× oil immersion objective using both differential interference contrast (DIC) and fluorescence imaging modes. LDs appeared as darkly stained circular profiles in the DIC images that were absent in unstained samples and could be also detected in fluorescent images using a Texas Red excitation filter (540 to 580 nm) (31). A total of 25 cells per condition were randomly selected from the DIC images, and the LDs contained in those cells were counted and manually segmented from the fluorescent images using Amira 6.0.1 (Thermo Fisher), which was further used to compute the LD area per cell.

(i) Preparation of chemically fixed samples. Sample fixation, staining, and embedding procedures were adapted from reference 32. Cells were fixed with 2.5% glutaraldehyde and 4% formaldehyde in 0.15 M cacodylate buffer for 60 min at room temperature. Initial postfixation was carried out using a solution of 2% osmium tetroxide and 1.5% potassium ferricyanide in 0.15 M cacodylate for 60 min on ice. Samples were then treated with 1% aqueous thiocarbohydrazide for 20 min, followed by 2% aqueous osmium tetroxide for 30 min, and were finally incubated overnight at 4°C in 1% aqueous uranyl acetate. Samples were then stained using Walton’s lead aspartate at 60°C for 30 min ahead of stepwise dehydration in ethanol. Samples were infiltrated with resin initially using a 50/50 mixture of propylene oxide and Durcupan ACM resin (Ladd Research) for 60 min. Where MatTek dishes were used, coverslips were excised from the surrounding culture dish using a razor blade and removed to a container resistant to propylene oxide ahead of this step. Samples were then infiltrated with fresh undiluted Durcupan for 90 min before covering regions of interest with inverted resin-filled BEEM capsules (Ted Pella) and polymerization for 48 h at 60°C. For TEM, sections of 70 nm were cut from the sample block for imaging and poststained with uranyl acetate and lead citrate. For SEM, sample blocks were prepared as described in reference 33. Briefly, small portions of the cell monolayers were mounted on pins using conductive epoxy resin (CircuitWorks CW2400), trimmed to form a pillar approximately 400 by 400 by 150 μm, ensuring that regions of interest were retained in the block-face for correlative samples, and coated with a 2-nm layer of platinum ahead of imaging.

(ii) Electron microscopy imaging. TEM images were collected in an FEI Tecnai 12 BioTWIN microscope using an Eagle 4K slow-scan charge-coupled-device (CCD) camera (FEI) or an FEI Twin at 120 kV with OneView 4K high-frame-rate camera (Gatan), both in binning mode 2. Serial block-face scanning electron microscopy (SBF-SEM) data were collected using a 3View 2XP system (Gatan, Pleasanton, CA) attached to a Sigma VP SEM (Zeiss, Cambridge, United Kingdom). Back-scattered electron images were acquired using the 3VBSE detector. The SEM was operated at a chamber pressure of 5 Pa, with high-current mode inactive. Variable pressure imaging conditions, which reduce the final resolution, aided in the suppression of charging resulting from the large regions of bare resin surrounding single cells. For CLEM data, the 30-μm aperture was used with an accelerating voltage of 2 kV, a dwell time of 3 μs (10-nm reported pixel size; horizontal frame width, 81.9 μm), and 40-nm slice thickness. Two collection setups were used for samples fixed at 6 hpi, using either the 30- or 20-μm aperture and an accelerating voltage of 2 or 3 kV, respectively. The imaging dwell time was 3 μs (5-nm reported pixel size; horizontal frame width, 20.5 μm) or 10 μs (6-nm reported pixel size; horizontal frame width, 24.6 μm), with a slice thickness of 50 or 30 nm.

(iii) SBF-SEM image processing and segmentation. SBF-SEM image stacks were acquired as dm3 files and converted to 32-bit tagged image files (TIFs) for initial batch processing. Images were Gaussian filtered (1 pixel), followed by two rounds of unsharp mask and gray-level normalization (Adobe Photoshop). Image stacks were then converted to 8-bit TIFs and aligned using the virtual align slices plugin (Fiji). Segmentation of features within SBF-SEM data was carried out in a semiautomatic threshold-based way for the cell cytoplasm and manually for the viral ROs, nuclei and LDs using either Segmentation Editor (Fiji) or Amira 6.0.1 (Thermo Fisher), which was then also employed to extract volume information on the different RO clusters.