Structural Basis of Vesicle Formation at the Inner Nuclear Membrane

1Intracytoplasmic transport between compartments is primarily mediated by vesicles. These vesicles are shaped by specific coat proteins that are recruited to the site of assembly and function to deform the membrane. In contrast, movement into and out of the nucleus is effected by “gated transport” via the nuclear pore complexes (NPCs). NPCs allow free diffusion of small molecules and can mediate active transport of cargo up to ∼39 nm in diameter. Larger macromolecular assemblies, however, are unable to pass through the NPC. Recently, vesicular trafficking was reported to mediate nucleo-cytoplasmic transport of ribonucleoprotein particles. This non-canonical pathway across the nuclear double membrane involves vesicle formation at the INM and fusion at the outer nuclear membrane (ONM). Another suggested role of this pathway is in nuclear recycling, i.e., transport of nuclear protein aggregates like defective NPC assembly intermediates to the cytosolic autophagy machinery for degradation.

2The ribonucleoprotein particle transport mechanism is in many respects similar to nuclear egress of herpesviruses discovered a decade earlier. In the latter, ∼125-nm-diameter icosahedral herpesvirus capsids assemble inside the nucleus and use vesicle-mediated transport across the nuclear envelope to gain access to the cytoplasm for further maturation. The combined evidence from the cellular ribonucleoprotein particle and viral capsid transport systems makes it likely that vesicle transport represents a general mechanism for translocation of large cargo from the nucleus to the cytoplasm that herpesviruses have usurped during evolution. Whereas the overall topology of the process of herpesvirus nuclear egress resembles cellular vesicle trafficking, little is known about the nanostructural details that lead to formation, scission, and fusion of INM-derived vesicles. Studied already in great biological detail, herpesvirus nuclear egress therefore represents a unique tractable model system to delineate the general structural and functional basis of nucleo-cytoplasmic vesicle transport.

3Studies of herpesvirus nuclear egress showed that, during infection, newly formed intranuclear capsids bud at the INM (Figure 1A) followed by membrane scission, resulting in enveloped capsids located in the perinuclear space. The envelope then fuses with the ONM to deliver the capsids to the cytoplasm. A large body of experimental studies has established that, throughout the Herpesviridae, two viral proteins, designated as pUL31 and pUL34 in the alphaherpesviruses herpes simplex virus 1 (HSV-1) and pseudorabies virus (PrV), form the heterodimeric nuclear egress complex (NEC; Figure 1A, green). The NEC is required and sufficient for vesicle formation, i.e., budding and scission, at the INM. The C terminus of the type II membrane protein pUL34 tethers the NEC to the INM, while pUL31 is exposed to the nucleoplasm. pUL31 then associates with the capsid surface in the lumen of the nascent perinuclear vesicle. After vesicle fusion with the ONM and release of the capsid, the NEC is exposed to the cytoplasm. NEC components are also likely to mediate cargo selection. Additionally, kinases recruited to the NEC are responsible for phosphorylation of lamins for local dissolution of the nuclear lamina to allow access of capsids to the INM and for phosphorylation of NEC components. Whereas this prototypic budding process at the INM and its components are well characterized, the fusion process with the ONM is still under debate, including a possible role of viral fusogenic glycoproteins.

4Recently, in vitro studies showed that partially truncated NEC components artificially membrane tethered to giant unilamellar vesicles formed a coat that can function as a minimal virus-encoded vesiculation machinery, not requiring additional viral or cellular factors. Furthermore, artificial membrane tethering of pUL31 alone was sufficient for induction of membrane invaginations and membrane scission in giant unilamellar vesicles. However, owing in part to the reduced complexity of the models used, these studies did not provide sufficient ultrastructural detail to elucidate the architecture and functionality of the NEC coat. Thus, we here investigated the NEC in its native location, in vesicles at the periphery of the nucleus.

5The size of the nucleus makes it a challenging target for visualization of intra-nuclear structures at molecular resolution in situ. Nevertheless, by applying an integrated multi-modal approach that enabled near-native imaging over variable scales and resolutions, we were able to characterize in detail both the extent of nuclear membrane remodeling and the architecture of the NEC at the INM. We first show in cryo-sections of herpesvirus-infected cells that the NEC forms a protein coat that lines capsid-containing perinuclear vesicles during egress. We then characterize the ultrastructure of NEC-coated vesicles in non-infected cells that co-express pUL31 and pUL34. Further, that latter experimental system provided a higher frequency of these vesicles, allowing successful cellular electron cryo-tomography of lamellae prepared by advanced focused ion beam cryo-milling (cryoFIB). Subsequent three-dimensional averaging of the NEC coat revealed that it forms an ordered lattice with two different hexameric faces. X-ray scattering data of solubilized NEC complexes, combined with integrated modeling, allowed us to determine that these two faces represent pUL34 anchored in the vesicle membrane and pUL31 forming the inner layer. The unique structure and interactions between the two protein layers result in a defined membrane curvature, ensuring that viral capsids are tightly enveloped. Our data reveal how formation of correctly sized perinuclear vesicles is achieved and establish a mechanistic basis for nucleo-cytoplasmic transport of large cargoes.

6To analyze the NEC coat formed in situ during viral infection in its most native environment, we used electron cryo-microscopy and tomography (cryoEM/T). CryoET imaging of areas deeper inside cells typically requires vitreous sections in order to provide electron transparent specimens of <500 nm thickness. In electron cryo-microscopy of vitreous sections (CEMOVIS), a method for imaging hydrated and unstained cellular ultrastructural detail, NECs were observed as electron-dense coats at the nucleoplasmic side of the INM and in the perinuclear space of HSV-1-infected Vero cells (Figures 1B–1F and Movies S1 and S2). When nuclear capsids were in close contact to the INM, a planar NEC coat of ∼100 nm diameter, i.e., about the width of the capsid, was observed (Figure 1B′, right HSV-1 capsid). The coat curved and expanded during budding of the INM into the perinuclear space (Figure 1B′, left HSV-1 capsid). Interestingly, the electron-dense NEC coat did not extend beyond the individual sites of budding (Figures 1C and 1D and Movie S1). Ultimately, the NEC formed a tightly fitting complete coat around the capsid (Figures 1D–1F and Movies S1 and S2). In HSV-1-infected Vero cells, not only DNA-filled C-capsids underwent primary envelopment (Figures 1B, 1D, and 1F), but also empty A-capsids (Figure 1C) and scaffold-containing immature B-capsids (Figure 1E, right vesicle). Intraluminal vesicles (ILVs, defined as possessing the NEC coat but lacking capsids; Figure 1E, asterisk; Movie S2) represented 34% of all observed perinuclear vesicles in HSV-1-infected Vero cells (14 of 41 vesicles, from 21 tomograms total), with a mean inner diameter of 115 nm ± 11 nm SD (n = 12). During de-envelopment, the NEC coat was left behind at the cytoplasmic face of the outer nuclear membrane (Figure 1F, arrow), and cytoplasmic capsids, now devoid of the NEC coat, subsequently underwent virion assembly (Figure 1F, right, and Movie S2, right upper-corner). This result contradicts previous conclusions drawn on the basis of interpreting densities in heavy-metal-stained, freeze-substituted, plastic-embedded samples and is in line with the absence of pUL31 and pUL34 in extracellular HSV-1 virions.

7A grainy nature of the NEC coat was readily visible in computational slices through cryoET reconstructions (Figures 1C–1E and Movies S1 and S2), suggesting a modular lattice-type architecture consisting of repetitive units. To analyze the structure and function of this coat in greater detail, a multimodal imaging approach was needed, spanning several length scales and covering from the nuclear distribution of its fully assembled form down to interactions of its single constituents.

8In HSV-1-infected cells, the number of capsid envelopment events captured at the INM was low. Therefore, we used a previously described porcine cell line that stably co-expresses pUL31 and pUL34 of PrV as a model frequently showing NEC-mediated vesicle formation. In this BK cell line, pUL34 is anchored to the INM by its authentic C-terminal transmembrane region, with the C-terminal GFP tag exposed on the vesicle outside, i.e., on the opposite membrane side of the NEC. Tagging allowed visualization of the NEC in vivo. By using three-dimensional structured illumination microscopy (3D-SIM), volumetric live-cell imaging of the nucleus at sub-diffraction resolution was achieved. This revealed clusters of fluorescent speckles of ∼160–1,500 nm diameter at multiple sites around the nuclear periphery, as well as within the nuclear interior along membranous invaginations (Figures 2A and Movie S3). These clusters represent accumulations of NEC-containing vesicles in the perinuclear space and were intensely fluorescent, suggesting high local concentrations of pUL34-GFP.

9Imaging of similar regions of BK cells at higher resolution by soft X-ray cryo-microscopy/tomography, guided by correlation with GFP fluorescence, provided detailed information about the spatial distribution of the NEC-containing target structures/vesicle clusters throughout the nucleus (Figure S1 and Movies S4 and S5). The ultrastructure of these intranuclear vesicle clusters was next characterized in 3D by CEMOVIS (Figure 2B), resulting in visualization of vesicles with a grainy inner NEC coat and a mean inner diameter of 107 ± 33 nm SD (n = 79). These vesicles were closely similar in size and structure to the capsid-less ILVs in infected cells (Figure 1E, asterisk).

10Characterization of larger volumes by serial CEMOVIS sections enabled us to localize and characterize the occurrence of repetitive NEC structures/lattices suited for sub-tomogram averaging, even in rare developmental states (Figure 1). An alternative cryo-thinning technique, cryoFIB, has recently been developed to produce 100–300 nm thick lamellae from vitreous samples. This approach does not rely on physical cutting and, thereby, avoids sectioning artifacts. CryoET data were recorded from cryoFIB-prepared lamellae of plunge-frozen BK cells. Perinuclear vesicles were typically spherical, although some exhibited a more irregular shape (Figures 3A–3C and Movie S6), possibly due to crowding. The NEC protein layer was evident as a clear lattice-like, ∼10-nm-thick coat lining the entire inside of each vesicle with periodic connections to the vesicle membrane (Figures 3A and 3B). Size measurements of vesicles from three tomograms showed a peaked distribution with a mean inner diameter of 103 ± 10 nm SD (n = 31) (Figure 3D). The thickness of the coat and the size of ILVs in BK cells measured in 3D from CEMOVIS and cryoFIB-based data were in agreement (Figures 2B and 3 and Movie S6) and were similar to CEMOVIS data from HSV-1-infected Vero cells (Figure 1E). The vesicle diameters did not show a Gaussian distribution. Instead, the distribution is heavily skewed and peaked with very light tails (Figure 3D). These properties suggest that a specific mechanism inherent to NEC coat assembly is a predominant determinant of vesicle size, with positive skewness indicating a lower limit of the measured parameter.

11The previous observations suggested a highly repetitive organization of the NEC protein layer. Taking different curvature into account, the structure of the NEC coat was therefore determined independently for each vesicle by sub-tomogram averaging from the cryoFIB/ET data (Figure 4). Each vesicle average revealed a curved hexagonal lattice composed of two tightly interconnected layers of distinct appearance (Figures 4A, 4B, and S2 and Movies S7 and S8 [resolution 3.5–4 nm]). We termed the two NEC layers the membrane-proximal (MP) and membrane-distal (MD) layers. The MD layer is a ∼3-nm-thick hexagonal lattice with a spacing between repeating unit centers of ∼11.5 nm (purple), and the MP layer is ∼7 nm thick (magenta) (Figure 4C and Movie S9). In cross-sections, the repeating unit of the NEC coat (a single hexagon) shared a characteristic “archway” motif—similar to an inverted ‘U’ (Figure 4A and Movie S7). The hexagonal unit could be further decomposed into a motif of angular appearance, one side of the archway, which appeared kinked at approximately two-thirds along its length. Thus, the MP layer consists of a conical arrangement of six independent densities that originate at the unit cell “keystone” extending toward the vesicle center to form an “arch” and connecting with the MD layer near the 2-fold axes. Densities connecting the arch/keystone of the unit cell to the vesicle membrane were already apparent in raw tomograms (Figure 3 and Movie S6) and became accentuated after sub-tomogram averaging (Figure 4A and Movie S7).

12We assigned the MP layer to the membrane-anchored pUL34 and the MD layer to pUL31 (Figure 4C and Movie S9). The schematic interpretation shown in Figure 5 is based on analysis of tangential slices revealing the characteristic arrangements of protein density in each layer (Figure 4B and Movie S8). Orthogonal cross-section slices, shown adjacent to the model (Figure 5), reveal that the local curvature of the MD layer (pUL31) is not isotropic. Between 3-fold axes (“3-2-3”) the MD layer is distinctly planar, whereas between 6-fold axes (“6-2-6”), the curvature of MD is consistent with that of the vesicle membrane.

13The cryoFIB/ET sub-tomogram averages do not readily reveal the stoichiometry of pUL31/34 heterodimers in the NEC coat. The schematic interpretation shown in Figure 5 suggests that the unit cell is composed of a hexamer of heterodimers ([pUL31/34]6). To independently validate the model, we characterized a soluble form of the PrV pUL31/34 heterodimer (Figure S3) by small-angle X-ray scattering (SAXS) (Figure 6). We determined the shape of the soluble heterodimer by ab initio modeling from the 1D SAXS scattering curve (Figures 6A and 6B). The angular shape (87 nm3 in volume, radius of gyration of 2.96 nm, and maximum dimension of 10.2 nm) was similar to that observed in the cryoEM sub-tomogram average as one side of the archway. To orient the soluble heterodimers within the cryoEM map, we carried out a fitting search using the SAXS model and sampled the full rotational range. By locally fitting multiple copies of the highest-scoring model into the cryoEM map, we were able to account for the cryoEM density as well as to reproduce its characteristic features (Figures 6C–6I and Movie S10). Together, this integrated modeling suggests that the NEC coat is composed of a ∼10-nm-thick layer of interacting hexameric cores of NEC heterodimers in lateral self-association.

14The densities of the MP layer leading to the membrane are not accounted for by the SAXS model (Figures 6D and 6E), consistent with it being the truncated sequence from pUL34 that leads to the transmembrane domain (residues 180–240). The calculated mass from these residues (∼40 kDa per hexamer) and the volume of the unaccounted density are congruent, confirming that the NEC coat consists only of pUL31 and pUL34, without direct contribution from any other viral or cellular factors, as was shown in vitro for HSV-1. Previously, it has been reported that this membrane-connecting part of pUL34 containing low-complexity/high-flexibility domains (Figure S3C) can be deleted and substituted by heterologous transmembrane-containing peptides.

15To unveil the architectural basis for its constrained curvature, we devised a simplified mathematical description of the NEC coat (Figure 7) and used this to produce a model of the coat that closely matches the measurements from the cryoFIB/ET average (Figure S4). The results confirm that the interplay of interactions within each layer and repeated heteromeric interactions between pUL31 and pUL34 define the curvature of the NEC coat.

16Observation of a hexagonal NEC coat for the two alphaherpesviruses, PrV, as reported here, and HSV-1, suggests that interactions occurring at 2-fold and 3-fold axes of the MD (pUL31) layer are likely evolutionarily conserved (lattice spacing ∼11 nm in both cases). Interestingly, artificially membrane-tethered pUL31 oligomers, i.e., in the absence of the native membrane tether pUL34, did not show any regular/hexagonal pattern. Thus, by mutually confining the position of each of the six NEC heterodimers in space, the role of the pUL34 membrane-connecting region is critically central in determining structural properties of the MP layer and thereby the curvature of the NEC coat. However, the distribution of vesicle sizes suggests that the NEC coat does not function as two rigidly imposed layers, i.e., it is not crystalline. While constrained in space by arch-forming interactions, the range of curvatures, vesicle sizes, and shapes observed (Figures 3D and 7C), starting from a planar NEC coat at initial budding sites (Figure 1B), is mediated by a high degree of flexibility in the membrane-connecting region of the coat.

17In our experiments, we confirmed that, at artificially high local concentrations, in cells under constitutive expression (or in incubated vesicles), NECs alone can spontaneously form a coat and are able to mediate a complex process that involves induction of membrane curvature, vesicle budding, and scission. However, our observations at concentrations typical of the native situation, i.e., in infected cells, suggest that, in two-thirds of the perinuclear vesicles, the initial nucleation of pUL31/34 heterodimers to form the NEC coat depended on presence of the capsid cargo (Figure 1).

18In the cellular context, the NEC coat has the ability to form uniformly sized coated vesicles independent of the capsid cargo (ILVs in HSV-1-infected cells, Figures 1E and 3). A higher variability in curvature and hence size and shape has been observed in artificial model vesicles using partially truncated pUL31/34 constructs without the genuine membrane anchor. The heteromeric combination of both pUL31 and pUL34 and their arrangement as hexamers as a result of the arch-forming interactions yield a structural environment (i.e., the inner surface of the coat) conceivably central to the task of selectively and efficiently recruiting and transporting the viral capsid. Thus, we propose that, while membrane-anchored pUL31 is able to drive budding on its own, formation of vesicles of a curvature tailored specifically to the herpesviral capsid requires pUL34.

19Finally, we found that NEC coat assembly in situ produces vesicles of a size closely approximating but being somewhat smaller than capsids (Figures 3D and 7C). This size distribution in the absence of the capsid cargo suggests that it is most likely the capsid itself that determines the minimum diameter of an enveloped capsid (Figure 7C, gray capsid symbol), as the NEC coat appears to inherently favor a slightly higher curvature and thus smaller vesicle size. Concomitantly, this ensures a very tight fit and interaction between the NEC coat and the capsid, leading to a cargo vesicle of the smallest possible size given the components involved.

20Our current functional model of capsid envelopment at the INM can be summarized as follows: sparsely distributed NEC heterodimers form a planar layer at the INM, either spontaneously or initiated by cargo/capsid docking. At this point, the lattice already shows the ∼11 nm spacing of the hexagonal MD/pUL31 layer. Driven potentially by structural changes in the pUL34 region during the concomitant formation of a second hexagonal layer (MP), budding of the INM into the perinuclear space is induced. New NEC heterodimers are recruited at the rim of the coat until it reaches its curvature limit through interaction of the MD/MP layers at a size to precisely envelope a herpesviral capsid.

21Our insights into the molecular mechanism of remodeling the nuclear envelope for viral nuclear egress provide a molecular template by which nucleo-cytoplasmic transport can occur. The precise architecture of the NEC defines vesicles with a specific size, allowing an efficient but highly controlled method of egress. This mechanism allows the transport of cargoes with minimal disruption to the INM, a feature essential for the egress of equivalent cellular cargoes. Crucially, membrane-anchored proteins mediating a divergent process would pre-assemble with a potentially modular cargo-recruitment adaptor to form heterodimeric units capable of forming lateral, self-assembling lattices. Formation of curvature by this lattice is then a pre-requisite for envelopment of egressing cargoes, features that would be evident when investigated in vitro. High-resolution structures of NEC from different herpesvirus species have now emerged that may reveal a common structural homology to cargo recruitment at the INM. Interestingly, pUL31 contains a conserved zinc-finger motif essential for vesicle formation and NEC function. Using the curved lattice structure described here as a model for fitting the atomic structure of the NEC heterodimer, we have defined exact interaction surfaces that could be used as a further constraint for structural and functional homology modeling of putative cellular counterparts.

22Recently, it has been shown that TorsinA AAA+ ATPase is activated in a complex with type II membrane protein LAP1 at the INM. As speculated in, that complex might also be involved in perinuclear vesicle formation during transport of ribonucleoprotein particles in Drosophila cells in which TorsinA has been shown to promote INM scission. This complex and the NEC might share molecular attributes like the zinc-finger motif coming from a common ancestor when (and if) herpesvirus has hijacked this pathway in evolution. However, there are many issues in determining common ancestors for protein structures, including the increased mutation rate of viral genomes. A next practical step to analyze that further might be to apply cryoEM also in vesicle-accumulating TorsinA-mutated cells described in. This imaging technique is the sole method that can elucidate the direct presence of a (protein) coat along a membrane unequivocally as it avoids artifacts by chemical fixation and heavy metal staining and can be combined with immunostaining. Finding a coat of TorsinA-LAP1 complexes, or any other players implicated in nuclear egress, might then suggest a similar mechanism of vesicle formation in a general nucleo-cytoplasmic transport pathway of large cargo, as described here for nuclear egress of herpesviral capsids.

23The described NEC coat architecture is an elegant solution for induction of membrane curvature based solely on the formation of a highly defined lattice of heterodimer interactions. This is reminiscent of virus budding at the plasma membrane, e.g., HIV. However, the NEC targets the INM, a membrane for which no other vesicle transport has yet been mechanistically fully elucidated. Furthermore, while most cellular vesicle formation processes involve a dedicated cellular scission machinery and consume energy in form of ATP or similar, the NEC (1) appears capable of autoscission by continuing assembly of NEC units on the inside of the forming vesicle and (2) requires at least under in vitro conditions no external energy input for both membrane budding and scission. Elucidating the unique features of the binary pUL31/34 vesicle formation machinery might provide the blueprint for designing vesicles of highly defined sizes or specific volumes to be used in pharmaceutical and nanobiotechnological applications. Moreover, the characterization of the nature of the viral cargo packing system at the INM opens the search for the respective cellular counterparts and molecular determinants mediating nuclear egress of cellular large cargo, including ribonucleoprotein particles.

24Sixteen hours after infection with herpes simplex virus 1 (HSV-1) strain K26GFP at a multiplicity of infection of 10, trypsinized African green monkey kidney cells (Vero cells, strain CCL-81; ATCC) or proteinase K-treated porcine epithelial-like embryonic EFN-R kidney cells stably co-expressing PrV pUL31 and pUL34, the latter fused with GFP (cell line designated as BK/EFN/UL31/34 gfp, here abbreviated to BK, catalog No. RIE 1083 of the Collection of Cell Lines in Veterinary Medicine at the FLI, Greifswald-Insel Riems, Germany), were physically fixed and analyzed. Cryo-immobilization was performed by high-pressure freezing followed by cryo-electron microscopy of vitreous sections (CEMOVIS), essentially as described in. Further details are available in the Supplemental Experimental Procedures.

25BK cells were grown on high-precision 22 × 22 mm No. 1.5H glass coverslips (Marienfeld Superior) or in μ-Dish, high glass bottom 35-mm dishes (Ibidi GmbH, Martinsried, Germany) to a confluency of ∼70%–80% in 10% (w/v) fetal bovine serum in Dulbecco’s modified Eagle medium (DMEM; GIBCO-Invitrogen). Before imaging, the medium was replaced with pre-warmed Opti-MEM (GIBCO-Invitrogen). Three-dimensional structured illumination microscopy (3D-SIM) on live-cell samples was performed using an OMX V3 Blaze system (Applied Precision, GE Healthcare) equipped with a 60×/1.42 NA PlanApo oil-immersion objective (Olympus), a 488-nm diode laser with standard filter sets, and Edge sCMOS cameras (PCO). Further details are available in the Supplemental Experimental Procedures.

26Standard 3.05 mm electron microscopy 200 mesh gold grids covered with a perforated carbon foil (R2/1; Quantifoil Micro Tools GmbH, Jena, Germany) were hydrophilised in a PDC-002 plasma cleaner (Harrick Plasma, Ithaca, NY, USA). BK cells were grown on these grids in DMEM supplemented with 10% (w/v) fetal calf serum and 1% (v/v) PSN Antibiotic Mixture (GIBCO-Invitrogen), essentially as performed for 3D-SIM and soft X-ray microscopy samples. After 2 days of incubation (37°C, 5% CO2) in plastic microscope slide growth chambers (μ-slide 2 × 9 well; Ibidi GmbH) and light microscopic screening for optimal growth, cells were cryo-immobilized by plunge freezing, as described in.

27CryoFIB was essentially performed as recently described. It is detailed in the Supplemental Experimental Procedures.

28For tomography of the cryoFIB lamellae, a Tecnai G2 Polara transmission electron microscope (FEI) equipped with a field emission gun operated at 300 kV, a GIF 2002 post-column energy filter (Gatan, Pleasanton, CA), and a 2048 × 2048 Gatan Multiscan CCD camera were used. Tomographic tilt-series acquisition under low-dose conditions (10 tilt series out of 14 lamellae, tilt range: −55° to 59°, cumulative dose: 110 electrons per Å2) was controlled by SerialEM. Tilt-series images were recorded at 3° tilt increments, with −6 μm defocus, at an object pixel size of 0.57 nm.

29For alignment of the tilt-series projections, small spherical cellular features or ice contaminants were employed as tracking markers, or patch tracking following the routine in the Etomo GUI of IMOD was applied. Tomograms were reconstructed using weighted back projections, and visualization was performed with Amira 5.2 (FEI).

30Sub-tomogram averaging was carried out using the PEET package applying constrained cross-correlation (CCC). Details of data processing, integrated analysis, and model building are available in the Supplemental Experimental Procedures.

31Details on the construction of a soluble NEC expression vector are provided in the Supplemental Experimental Procedures. Escherichia coli BL21 (DE3) transformed with pETDuet::UL34(1–179)::UL31-NLS was grown to saturation overnight at 25°C in LB medium containing ampicillin (100 μg ml−1). An aliquot of overnight culture was diluted 1/20 in medium containing ampicillin and was grown at 37°C to an OD600 of 0.6, at which time expression was induced by addition of isopropyl-D- galactoside (IPTG) to a final concentration of 1 mM. Cells were incubated for a further 4.5 hr at 25°C before being harvested by centrifugation (3,500 × g, 10 min, 4°C), and stored at −20°C.

32Cells were re-suspended in buffer A (PBS pH 7.4); supplemented with 2,500 units DNase 1, 0.2 mg ml−1 lysozyme, 5 mM MgSO4, and 1% protease inhibitor cocktail (Sigma); and lysed by sonication. The supernatant was clarified by centrifugation (45,000 × g, 30 min, 4°C).

33The complex was purified by metal affinity chromatography (co-sepharose 6-fast flow) followed by size-exclusion chromatography in buffer B (10 mM Tris-HCl [pH 7.4], 75 mM NaCl, 3 mM DTT). Purified proteins were analyzed by SDS-PAGE.

34SAXS data for soluble NEC were collected on the BM29 beamline at the ESRF synchrotron (Grenoble, France). Details are available in the Supplemental Experimental Procedures.

35All authors intellectually conceived the experiments. C.H. performed CEMOVIS, cryoXM/T, and cryoFIB; K.C.D. did cryoET sub-tomogram analysis and integrated modeling; T.Z.-B.-M. provided SAXS data and models; and M.G. and J.B.B. imaged cells by live 3D-SIM. Furthermore, C.W., B.G.K., C.A.S., F.J.B.B., J.C., S.W., P.G., S.R., K.H., J.D., and B.A. performed experiments; D.V. processed and analyzed data; C.H., K.C.D., T.Z.-B.-M., M.G., J.B.B., T.C.M., and K.G. wrote the manuscript, and all authors commented on it.