In this work, we developed Serial Section Array-Scanning Electron Microscopy (SSA-SEM), a novel three-dimensional (3D) imaging and reconstruction strategy, and applied the technique to the analysis of VZV-infected cells. Using SSA-SEM and EM tomography, we have reconstructed the nuclei of host cells infected with this representative herpesvirus and, for the first time, revealed the numbers and precise location of thousands of VZV nucleocapsids, visualized the 3D shape and ultrastructure of nuclear PML cages that entrap nucleocapsids, and provided quantitative estimates of the volume, sequestration efficiency and sequestration capacity of these PML cages. The large volume reconstruction of nuclei in VZV-infected cells also provided basic information on how VZV infection affects the shape of the host cell nuclei and how subnuclear domains like electron dense heterochromatin or PML cages and nucleocapsids are spatially related. Of interest, our 3D analysis revealed that PML cages with entrapped capsids were consistently located at the periphery of the nucleus and associated with domains of electron dense heterochromatin, suggesting that the formation of PML cages and VZV capsid sequestration are initiated adjacent to these domains.
Our experimental challenge, which has many similarities to obstacles encountered in addressing other virology and cell biology questions, consisted in how to combine an efficient approach for the large volume 3D reconstruction of infected cell nuclei and complete PML cages with the high ultrastructural resolution necessary to localize VZV nucleocapsids and differentiate mature from immature capsids. Infected cell nuclei have diameters of about 5–10 µm and PML cages are about 0.5–5 µm 
. These structures are about one order of magnitude too large to be readily reconstructed by conventional electron tomography approaches that usually use 100–300 nm sections. Recent technical and computational improvements have enabled some specialized laboratories to apply serial-sectioning tomography for the reconstruction of large organelles and even complete cells by merging individual tomograms from consecutive sections into a single large volume reconstruction 
. However, this approach is very labor-intensive so that only a few 3D reconstructions can be generated and this limitation may raise questions about whether these models are fully representative of the structures of interest.
SSA-SEM combines a sample preparation strategy (serial section arrays) similar to the method used in immunofluorescence (IF) array tomography with imaging and detection principles (high resolution SEM with back scattered electron detection) that have been used in serial block face (SBF)-SEM or focus ion beam (FIB)-SEM 
. In principle, the latter two methods could also be used to analyze herpesvirus-infected cell nuclei or PML cages. In fact, Feierbach et al
. used SBF-SEM to locate structures reminiscent of actin filaments and nucleocapsids in cells infected with HSV-1 and PRV, which have caspids that are similar to VZV capsids in size and shape 
. Bennett et al
. used FIB-SEM to locate human immunodeficiency virus (HIV) particles in surface-connected tubular conduits in HIV-infected macrophages 
. However, these approaches require highly specialized equipment that may not be readily accessible. Most importantly, these techniques are destructive imaging methods that destroy the sample block during image stack acquisition by step wise FIB-milling or cutting the sample surface to allow successive surface imaging at different sample levels, while discarding the serially-cut sections. SBF-SEM and FIB-SEM may therefore not be ideal for valuable samples that are difficult to obtain or to prepare. In SSA-SEM, serial sections are secured on a glass slide, creating stable arrays that can be stored and imaged repeatedly, allowing the acquisition of several image series of the same sample at different magnification, resolutions, contrast modes or with different equipment. A major advantage of SSA-SEM is that the interior of cells and tissue become exposed at the section surface, enabling the use of immuno-histochemistry protocols to localize proteins or nuclei acids within the context of the 3D ultrastructure of cells or tissues.
Our 3D reconstructions confirmed that most PML nuclear bodies in VZV infected cells expressing endogenous PML are disassembled efficiently during the course of infection. This process involves the interaction of SUMO-interacting domains (SIM) of the VZV ORF61 protein with sumoylated PML 
. As a result, most of the several thousand VZV nucleocapsids that were produced in VZV-infected cells appeared randomly distributed in the reconstructed nuclear volume when examined by SSA-SM. As noted, other alphaherpesviruses disrupt PML nuclear bodies and in most cases, also eliminate PML protein by rapid ICP0-mediated degradation 
However, when PML disassembly is incomplete, as it is in VZV-infected cells in skin and neural cells in vivo
, nucleocapsids become sequestered in PML cages. Systematic random sampling analysis of hundreds of ultrathin sections through different PML cages suggested that >95% of all types of VZV nucleocapsids (A, B and C-type) were efficiently sequestered in PML cages 
. Nevertheless, random ultrathin sections do not reveal the 3D shape and volume of single PML cages because these sections (50–100 nm) may encompass only 1–10% of the diameter of PML cages. Therefore, techniques used in the earlier study did not allow an assessment of the size, volume and shape of PML cages or how many VZV nucleocapsids may be sequestered within individual PML cages. Furthermore, since ultrathin cross-sections through a nucleus encompass only a very small fraction of the nuclear volume, the sequestration efficiency of PML cages could not be determined for single nuclei. These limitations were addressed by using SSA-SEM to reconstruct the shape and volume of individual PML cages, which demonstrated that up to several thousand (2,780) nucleocapsids can be sequestered by single PML cages. Furthermore, quantitative analysis of several thousand nucleocapsids in reconstructed volumes of single nuclei showed that more than 98% of all capsids could become entrapped in PML cages, proving their very high sequestration capacity and explaining the antiviral activity of PML IV 
. Our method to estimate the sequestration capacity and efficiency of PML cages made it possible to provide information beyond just a morphological description and demonstrates that SSA-SEM can be used in quantitative analyses of virus interactions with nuclear structures.
Given the high sequestration capacity of PML cages, now established by single 3D nuclear analysis and by quantitative random sampling analysis of hundreds of ultrathin cross-sections, it is somewhat surprising that infectious VZV titers were reduced only by about 50% in cell lines expressing PML IV 
. These results indicate that only very few VZV infectious particles are needed to successfully enter and replicate in adjacent cells. This explanation is consistent with the observation that only very few PRV genomes are required to establish nuclear replication compartments and initiate productive replication, as shown using recombinant PRV, which is also an alphaherpesvirus, carrying a Brainbow cassette 
. VZV does not release virus particles into the supernatant in cell culture and spreads only from cell to cell by a mechanism that may be facilitated by extensive syncytia formation 
; therefore, even the few nucleocapsids that may escape sequestration in PML cages should be sufficient to infect adjacent cells in vitro
. In contrast, in the human host, VZV must infect complex tissues and overcome the barriers of intrinsic and adaptive immunity, which is likely to depend on production of larger numbers of infectious virus particles. Therefore the PML-mediated nuclear sequestration of many VZV capsids observed in human skin or DRG may be expected to have a more substantial antiviral effect 
. The quantitative analysis of the different types of capsids present within infected cell nuclei revealed that the majority (70–90%) were immature (A and B-type capsids) while only a minority was in a mature stage (C-type, 10–30%). We speculate that large numbers of immature nucleocapsids help to outcompete the limited sequestration capacity of PML cages, giving mature capsids a better chance to egress from the nucleus. These observations also suggest that the relatively few mature virions observed in VZV infected cells in vitro
is not just a tissue culture phenomenon.
Using conventional EM tomography, we obtained the first insights about the 3D ultrastructure of PML cages, suggesting how VZV nucleocapsids may be kept entrapped in these nuclear domains. Tomographic 3D reconstructions revealed the presence of an electron dense meshwork surrounding sequestered nucleocapsids and fiber-like like structures, that often cross-linked adjacent nucleocapsids, suggesting that capsids were entrapped by restricting their mobility and ‘gluing’ them together. The 3D analysis of PML-labeled sections by serial section immunoTEM showed that PML protein was present both in the periphery of the cage (the ‘shell’) and associated with the capsids entrapped in the center of PML cages. PML protein which is the main structural component of PML nuclear bodies, forms homo-and heterooligomers 
; therefore at least part of the electron dense meshwork is likely to consist of PML-oligomers that crosslink and embed capsids in a protein meshwork. The PML-positive meshwork and fibers were in general directly associated with the edges of VZV capsids, which is consistent with our previous biochemical data that demonstrated an interaction of PML with the small outer capsid protein ORF23 
. Many other proteins resident in PML-nuclear bodies, e.g. hDaxx or Sp100, may be part of this meshwork 
Cryo-tomography is an alternative that would enable a 3D reconstruction of PML cages at even higher resolution and more native conditions (avoiding resin embedding and heavy metal staining) but this approach can be predicted to encounter major experimental challenges. Frozen-hydrated sections through cell nuclei will be required, the identification of PML cages will be very demanding as these structures are not abundant, and identification would need to occur at a very low electron dose that creates noisy imaging conditions. Molecular docking of known crystal structures to electron densities will also be difficult, because of the many other proteins present in PML nuclear domain and currently only the PML RING domain has been crystallized 
, whereas the PML IV C-terminal domain is critical for nucleocapsid sequestration 
In summary, we were able to create complete reconstructions of herpesvirus-infected cell nuclei and PML nuclear domains in three dimensions for the first time using 3D SSA-SEM and EM tomography. This study supports and extends our recent discovery and characterization of PML cages that efficiently sequester VZV nucleocapsids in cell culture and in differentiated human skin and neural cells infected in vivo
and represents a novel antiviral mechanism, distinct from the established role of PML in controlling several alphaherpesviruses shortly after virus entry by limiting early viral gene transcription. Visualization of the shape and measurements of the volumes of host cell nuclei and PML cages together with the 3D localization of VZV nucleocapsids with ultrastructural precision enabled us to determine the sequestration efficiency and capacity of PML nuclear cages. This work contributes not only to a more comprehensive understanding of the antiviral activity of PML cages against VZV, a pathogenic human herpesvirus, but also provides a novel method to undertake the 3D reconstruction and quantitative investigation of nuclear PML domains that have also been found to be associated with capsids of papillomaviruses and polyomaviruses 
. The method has broad relevance for addressing other questions in virology and cell biology where large volume 3D reconstruction with high precision imaging of intracellular structures is needed.