Chikungunya virus assembly and budding visualized in situ using cryogenic electron tomography

Chikungunya virus (CHIKV) is a representative alphavirus causing debilitating arthritogenic disease in humans. Alphavirus particles assemble into two icosahedral layers: the glycoprotein spike shell embedded in a lipid envelope and the inner nucleocapsid (NC) core. In contrast to matrix-driven assembly of some enveloped viruses, the assembly/budding process of two-layered icosahedral particles remains poorly understood. Here we used cryogenic electron tomography (cryo-ET) to capture snapshots of the CHIKV assembly in infected human cells. Subvolume classification of the snapshots revealed 12 intermediates representing different stages of assembly at the plasma membrane. Further subtomogram average structures ranging from subnanometre to nanometre resolutions show that immature non-icosahedral NCs function as rough scaffolds to trigger icosahedral assembly of the spike lattice, which in turn progressively transforms the underlying NCs into icosahedral cores during budding. Further, analysis of CHIKV-infected cells treated with budding-inhibiting antibodies revealed wider spaces between spikes than in icosahedral spike lattice, suggesting that spacing spikes apart to prevent their lateral interactions prevents the plasma membrane from bending around the NC, thus blocking virus budding. These findings provide the molecular mechanisms for alphavirus assembly and antibody-mediated budding inhibition that provide valuable insights for the development of broad therapeutics targeting the assembly of icosahedral enveloped viruses. Cryogenic electron tomography analysis of Chikungunya virus particle assembly reveals 12 intermediate structural stages during virus assembly/budding at the plasma membrane and shows that non-icosahedral nucleocapsid proteins serve as scaffold to induce icosahedral assembly of the glycoprotein spike lattice. Structural analysis also shows that budding-inhibiting antibodies act by interfering with lateral spike interactions.

nveloped virus assembly is a highly coordinated process that requires budding the cell membrane and incorporating all necessary components into the viral particle for subsequent cell infection. This process is relatively well understood for viruses whose assembly and budding are driven solely by capsid or matrix proteins via recruiting host endosomal sorting complex required for transport machinery, such as retroviruses and filoviruses 1 . In contrast, assembly and budding of alphaviruses, a representative genus of enveloped spherical viruses containing two icosahedral structural protein layers, is less understood, with multiple contradictory models proposed in previous studies 2 . Chikungunya virus (CHIKV) is the most common alphavirus infecting humans and has caused global epidemics of debilitating and often chronic polyarthritis 3 . No licensed vaccine or antivirals are available for treating any alphavirus infection. All alphaviruses share a spherical virion architecture ~70 nm in diameter, with concentric spike and nucleocapsid (NC) layers separated by a lipid envelope. The outer glycoprotein (GP) shell of 240 copies of E1-E2·(E3) heterodimers are arranged as 80 prominent trimeric surface spikes embedded in the viral envelope, while an inner-NC core of 240 capsid proteins (Cps) encapsulates the 11.5 kb plus-sense (+) genomic RNA 4 . At the late stage of virus infection, NC cores assembled in the cytosol and trimeric E1-E2·(E3) spikes, synthesized and processed through the membrane secretory system, co-assemble into virus particles at the plasma membrane. Vertical links across the lipid envelope between spikes and NC are mediated by insertion of the intracellular tail of E2 into the hydrophobic pocket of Cp [5][6][7] . Contradictory mechanisms were proposed for the assembly of two-layered icosahedral alphavirus particles, largely centred around whether Cps pre-assemble into icosahedral NCs 5,6 that serve as structural templates or whether spikes drive icosahedral co-assembly of outer-spike and inner-NC lattices 2 . Such assembly of spherical icosahedral shells during assembly/budding is relevant to other enveloped viruses, including the notable flaviviruses Dengue virus and Zika virus. Importantly, we previously demonstrated that the assembly of the two-layered icosahedral CHIKV particles can be blocked by antiviral antibodies [8][9][10] . Without direct visualization at molecular-resolution of both the assembly/budding process and spikes in a budding-blocked conformation in virus-infected cells, the mechanisms of icosahedral particle assembly and antibody-induced budding inhibition remain poorly defined.
Here we imaged vitrified CHIKV-infected human cells and determined structures of viral assembly intermediates in situ using cryogenic electron tomography (cryo-ET). By analysing both the full progression of CHIKV assembly in the cell and budding blocking by antibodies, we revealed the structural organization of spikes and NCs before budding, during their orchestrated co-assembly and in a budding-blocked conformation. This resolved the structural mechanism of two-layered icosahedral particle assembly and importantly, that of budding-blocking molecules that can serve as anti-alphavirus therapeutics.

Cell tomography captures CHIKV budding intermediates.
To capture the dynamic CHIKV assembly/budding process in the near-native state, we imaged U2OS cells, a human bone osteosarcoma cell line, that were infected with CHIKV-181 vaccine strain on a cryogenic electron microscopy (cryo-EM) grid and embedded Chikungunya virus assembly and budding visualized in situ using cryogenic electron tomography David Chmielewski 1 , Michael F. Schmid 2 , Graham Simmons 3,4 , Jing Jin 3,4,6 ✉ and Wah Chiu 1,2,5,6 ✉ Chikungunya virus (CHIKV) is a representative alphavirus causing debilitating arthritogenic disease in humans. Alphavirus particles assemble into two icosahedral layers: the glycoprotein spike shell embedded in a lipid envelope and the inner nucleocapsid (NC) core. In contrast to matrix-driven assembly of some enveloped viruses, the assembly/budding process of two-layered icosahedral particles remains poorly understood. Here we used cryogenic electron tomography (cryo-ET) to capture snapshots of the CHIKV assembly in infected human cells. Subvolume classification of the snapshots revealed 12 intermediates representing different stages of assembly at the plasma membrane. Further subtomogram average structures ranging from subnanometre to nanometre resolutions show that immature non-icosahedral NCs function as rough scaffolds to trigger icosahedral assembly of the spike lattice, which in turn progressively transforms the underlying NCs into icosahedral cores during budding. Further, analysis of CHIKV-infected cells treated with budding-inhibiting antibodies revealed wider spaces between spikes than in icosahedral spike lattice, suggesting that spacing spikes apart to prevent their lateral interactions prevents the plasma membrane from bending around the NC, thus blocking virus budding. These findings provide the molecular mechanisms for alphavirus assembly and antibody-mediated budding inhibition that provide valuable insights for the development of broad therapeutics targeting the assembly of icosahedral enveloped viruses.
in vitreous ice. Tomographic tilt series were collected at the infected-cell peripheries at 8 h post infection (hpi) and revealed a variety of CHIKV assembly phenotypes (Supplementary Movie 1): budding at the PM of the cell body ( Fig. 1a-d), budding on long intercellular extensions (>10 µm) and short extensions (typically 2-10 µm in length), and thin branching extensions composed solely of incomplete viral particles (Fig. 1a,b). Particles were observed at the PM at various stages of budding and as fully assembled virions released into the extracellular space (Fig. 1e,f), thus capturing snapshots of the entire CHIKV assembly/budding process. Interestingly, CHIKV replication spherules, where viral RNAs are synthesized, were occasionally observed near cytosolic nucleocapsid-like particles (NLPs) and budding viruses (Fig. 1c,d and Extended Data Fig. 1). This is consistent with previous reports of CHIKV replication spherules located predominantly on the cell surface 11 . It is conceivable that viral RNAs can be synthesized and immediately packaged into NLPs that bud into virions, all near the PM, which is different from other old-world alphaviruses that have RNA replication and virus budding spatially separated 12 . We also identified thin extensions from the cell periphery, formed by incomplete particles often the width of a single virion (~70 nm diameter, <5 µm length) and linked by a continuous membrane (Extended Data Fig. 2). These extensions displayed differences in the levels of particle completion and structural conformations (Extended Data Fig. 2), but due to the lack of sufficient spikes to finish enwrapping the NC as an icosahedron, the linked particles are unlikely to complete the assembly of full virions. Therefore, such particles were excluded from analysis of two-layered icosahedral particle assembly.
Next, we studied how icosahedral alphavirus particles form at the PM by analysing snapshot images of 1,918 individual budding states picked from the three-dimensional (3D) tomograms. These snapshots were subjected to subtomogram alignment and classification through an unbiased and iterative multi-reference refinement protocol based on structural similarity (Extended Data Fig. 3a and Methods). This resulted in 12 class average maps with distinct stages of the virus assembly ( Fig. 2)  budding shell (Fig. 2d). The 12 class average maps were ranked by increasing budding level and the particle numbers of individual states within each class were determined (Fig. 2d). This classification scheme resulted in large numbers of particles (868 particles, 63% of total) grouped into classes (9)(10)(11)(12) at the very late stages of assembly/budding with 94-100% complete glycoprotein shells. In this analysis, the 100% assembled particles are still tethered to the PM. The proportion of particles within early budding classes (1-3, 12-50% particle completion) is less (288 particles, 21% of total) and even fewer particles (191, 14% of total) were grouped in the remaining classes (4-8, 50-88% completion). In our analysis, we assumed that a snapshot of any intermediate state can be captured and that at the time we imaged the cells (8 hpi), virus assembly and supply of Cp and spikes were at a steady state across cells on the grid. The discrete classes with a larger number of particles could be interpreted as the states having a low free energy and would thus transition to the next assembly state at a slower rate. On the contrary, the classes with fewer particles could have a faster rate of transition to the next assembly state. Our results suggest that the kinetic progression of CHIKV budding is non-uniform, with formation of the first half of the glycoprotein shell being a minor rate-limiting step, and final pinch-off of the complete virions being the major rate-limiting step. Further work is warranted to support this proposed model of assembly kinetics, while alternative interpretations related to bursts of virus assembly in the cells cannot be excluded from the current study.
In light of previous studies that described notable inter-and intra-particle heterogeneity in alphavirus particles following in vitro purification, we then evaluated images of released virus particles in the 3D tomograms 6,13 . Concerns about the fragility of enveloped viruses to purification have called into question the relevance of observed structural heterogeneity in single particle cryo-EM analysis of biochemically purified virus to alphavirus assembly in situ 13,14 . Therefore, our direct imaging of CHIKV-infected cells eliminates potential structural perturbations to biochemically purified virus particles.
Using a Volta phase plate to collect images with increased contrast, released virions typically displayed relatively weak or absent density at one side of the particle between spike and NC core layers (Extended Data Fig. 4). In some rare extreme events, multiple NC cores bud into one released particle and similarly at the un-enveloped side of each core an absent density was observed (Extended Data Fig. 4e). Interestingly, the trailing end of late-stage budding particles still tethered to the PM also displayed relatively absent density with similar non-icosahedral geometry of the viral envelope (Extended Data Fig. 4b). In some tomograms, unidentified molecular complexes were observed at the base of the viral envelope in the relatively absent density region (Extended Data Fig. 4d). Therefore, released CHIKV particles possess imperfect icosahedral symmetry that probably results from the final membrane scission step of virus assembly, reminiscent of imperfect symmetry of flaviviruses 15 .
Detailed CHIKV budding structures. Two-layered icosahedral (5-3-2) symmetry is a feature of purified and mature CHIKV particles 4 , but the role of spike and NC layers in assembly/budding has not been established. We next explored how the glycoprotein spike and NC layers were organized at different stages of budding to decipher the mechanism of alphavirus assembly. We performed further subtomogram refinement of released virions and four budding intermediate classes that displayed weak icosahedral features during previous 3D classification (Fig. 2), applying five-fold symmetry (Methods). Among the resulting five maps, the earliest budding structure displays a 5-fold pentagon of spikes only at the leading end of budding, while the other four structures displayed excellent pentagon and hexagon assemblies with 5-fold, 3-fold and 2-fold symmetry axes (Fig. 3b). The maps ranged in resolution from 8.3 Å (released) to ~44 Å (docking) (0.143 Fourier Shell Correlation (FSC) criterion) ( Fig. 3b and Extended Data Fig. 3b). Subnanometer resolution in the released virion average is validated by the visualization of E1/E2 transmembrane helices spanning the viral envelope (Extended Data Fig. 3c,d). The resolution progressively decreases in maps of less-complete particles and is probably caused by increased conformational flexibility due to lack of icosahedral symmetry constraints. A striking correlation between icosahedrally symmetric regions of the spike budding shells and Cps of the underlying NC core was observed at the 'leading end' of budding (Fig. 3b). In contrast, there is a lack of detectable icosahedral symmetry in the 'trailing end' of each intermediate NC's structure, where no spikes are present ( Fig. 3b and Extended Data Fig. 4a-c). The NC of the earliest budding map (docking) only shows five-fold symmetry at the 'leading end' with a small cluster of spikes above and is notably smaller (long axis: 37 nm, short axis: 31 nm) than the icosahedral NC of a released virion (41 nm). Overall, as the growing icosahedral spike shell enwraps the NC, it reorganizes Cps from asymmetry into matching icosahedral symmetry. This result explains the origin of two-layered icosahedral symmetry evident in mature alphavirus virions 7 .
To assemble a two-layered icosahedral particle, the previously proposed NC-centric model of alphavirus budding 16 suggested that preformed icosahedral NCs provide a symmetric template for spike incorporation. To test this model and our previous interpretation of spike-driven reorganization of NCs during budding, we performed subtomogram classification and refinement of apparently cytosolic NLPs present in the 3D tomograms. Apparently cytosolic NLP subvolumes (545) were picked from the 3D tomograms on condition that there was no clear budding or attachment to spikes at the PM ( Fig. 3a and Extended Data Fig. 5a). Due to the orientation of the membrane in the tomograms, clear budding structures at the plasma membrane were only visible in the side view, while the top and bottom membrane surfaces were invisible. Subtomogram classification of the apparently cytosolic NLP particles was performed with five-fold symmetry applied and resulted in two classes with NC-like structures (classes I and II). Further subtomogram refinement of those two sets of particles, again with applied five-fold symmetry, produced low-resolution structures shown in Fig. 3c at resolutions of 47.6 Å (class I) and 43.5 Å (class II). However, both maps lacked icosahedral symmetry, while only class II displayed local five-fold symmetry features at one pole (Extended Data Fig. 4a). The diameters of the two maps were both ~37 nm, with one structure (class I) largely spherical and the other (class II) an oblate spheroid with a short axis of 33 nm ( Fig. 3c and Extended Data Fig. 4a). Due to overall similarity between the class II map with the NC of the earliest budding intermediate, we next asked whether the local five-fold symmetry at one pole of the class II map was a result of interactions with spikes at the PM in orientations that were not clear in the 3D tomograms.
We proceeded by utilizing the refined subvolume orientations of each particle in class II to map the particles back into the original 3D tomograms for cellular context. Most particles within class II were positioned in the cell with the five-fold symmetric pole oriented towards the PM. This indicates that the five-fold structure arises from interactions with membrane-embedded spikes (Extended Data Fig. 5), and different from cytosolic NLPs, these particles are early-stage budding NCs docking at the plasma membrane. Patches of spikes were observed in tomogram slices above or below many of these particles, adding further support to this conclusion (Extended Data Fig. 5). At the current resolution, the lack of symmetry in class I cytosolic NLP particles cannot rule out that the Cp-RNA mixture is organized into assemblies with alternate non-icosahedral and non-five-fold symmetry. Taken together, the results indicate that the spike lattice must reorganize cytosolic NLPs from asymmetric assemblies to icosahedral viral cores during assembly/budding.
Assembly of spike lattices. Our structures of budding-intermediate states (Fig. 2) revealed a progressive spike-driven NC morphogenesis, demonstrating the mechanistic role of the spike lattice in alphavirus assembly. However, how the spike layer acquires the proper icosahedral symmetry is not well understood. Detailed spike organization at the PM has never been observed in virus-infected cells due to technical challenges in resolving small macromolecules in different orientations without chemical stains. Here we applied contrast-enhancing Volta phase plate cryo-ET imaging to investigate how spikes organize outside virus budding sites. From 3D tomograms of infected cells, we observed rare non-icosahedral spike assembly products that provided unexpected insights into icosahedral spike assembly. First, we identified rare instances of near-planar sheets of hexagonal spike lattices without underlying NCs (Extended Data Fig. 6). Secondly, we observed highly curved tubular spike lattices in filopodia-like structures: on short extensions with budding particles at the tip (Fig. 4a), on long extensions entirely devoid of NCs ( Fig. 4c and Extended Data Fig. 7) and on thin extensions loaded with budding particles (Fig. 4e and Extended Data Fig. 6). We determined subvolume averages of three different tubular spike lattices by applying helical rotations to compensate for the tomographic missing wedge. The results revealed helical organization of trimeric spikes arranged as hexagons with uniform spacing and conformation, but no internal NCs or helically arranged Cps were found within any of the membrane tubes ( Fig. 4). Interestingly, the average diameter of these spike coated membrane tubes varies from 55 to 65 nm, close to the diameter of the icosahedral enveloped CHIKV virion (~70 nm) (Supplementary Table 2). Of the different spike assemblies we observed, pentagons of trimeric spikes were only observed on the surface of budding intermediates and released particles ( Fig. 3 and Extended Data Fig. 6). Self-assembled hexagonal spike lattices can cover flat and highly curved membrane surfaces (Fig. 4, and Extended Data Figs. 6 and 7) but do not form spherical icosahedrons. Taken together with the heterogeneous non-icosahedral feature of cytosolic NLPs (Fig. 3), these data suggest that NCs serve as rough spherical scaffolds that initiate assembly of the proper icosahedral spike lattice that then drives assembly/budding.
Besides the hexagonal spike lattices that were observed on thin membrane extensions of CHIKV-infected cells, no organized spike assemblies were observed at the PM near virus budding sites. Some spike-like proteins were often identified around the base of the budding particles but were not organized into detectable higher-order assemblies (Extended Data Fig. 8), in marked contrast to organized spikes in the hexagonal and icosahedral lattices (Figs. 3 and 4, and Extended Data Figs. 6 and 7). These results suggest that the icosahedral spike lattice on budding particles is unlikely to be reorganized from any pre-assembled higher-order spike assembly on the plasma membrane.

Disrupting lateral spike interactions blocks CHIKV budding.
Building on our results showing that icosahedral spike lattice formation is essential to CHIKV assembly/budding, we investigated the mechanism for budding inhibition by anti-CHIKV antibodies that we previously reported 8,10 . We utilized anti-CHIKV NAb C9 that bivalently binds to spikes at the PM and blocks virus budding, to investigate the organization of spikes in a buddingblocked state 8,10 . CHIKV-181-infected U2OS cells were treated with C9 at 2 hpi and tomographic tilt series of the cells were collected 6 h later using Volta phase plate. 3D cryo-ET tomograms revealed that large numbers of cytosolic NLPs docked to the inner leaflet of the PM without virus budding ( Fig. 5 and Supplementary Movie 2), consistent with our previous report 10 . Computational slices of budding-arrested NLPs clearly resolved a flattened docking end below the near-planar PM. 3D classification of 1,727 budding-arrested NLPs revealed six conformations, two of which contained membrane density, and four without any membrane feature that resembled the cytosolic NLP class I described above (Fig. 3c and Extended Data Fig. 3a). Budding-arrested NLPs were predominantly docked to the plasma membrane (~82% of particles) with flattened ends that directly interact with the cytoplasmic tail of spikes (Fig. 5b,k). The lack of icosahedral or five-fold structure in cytosolic budding-arrested NLPs and a transformation in structure upon docking to the PM further support the model of spike-driven NC reorganization. Tomogram slices displaying side-views of the PM revealed three characteristic layers: the PM, spike ectodomains above the PM and a second protein layer above the spikes, approximately 150-250 Å from the inner leaflet of the PM (Fig. 5a,b). The dense protein layer above spike ectodomains, not seen in regular CHIKV-infected cells (Fig. 1f), indicated binding of C9 IgGs to the previously reported epitope at the crest of spikes 8 . Computational slices corresponding to top and bottom views of the cells revealed clear trimeric spikes in large patches at micrometre scale (Fig. 5e-h

and Supplementary
Movie 2). Densities were observed bridging across spikes with Y-shaped features, consistent with IgG structure (Fig. 5e-h). Taken together, these observations indicated that spikes were bridged by C9 and blocked from the structural rearrangements necessary for icosahedral assembly. The flat feature of spike patches and the docking end of underlying NLPs further supports our model of outside-in, spike-driven reorganization of Cps in those interacting NLPs.
After confirming the lack of virus budding and NAbs crosslinking spikes in the cell tomograms, we asked how the NAbs blocked budding at the molecular level. We first determined the precise location of individual spikes in the 3D tomograms. Trimer spikes (7,678) were manually picked, followed by subtomogram refinement that yielded a density map at ~24 Å resolution (0.143 FSC criterion). The map approximately matches the atomic model of the CHIKV trimeric spike (Fig. 5i). Using the refined orientations of each spike in a tomogram, we computed the distance between a spike and its nearest-neighbour spike at the PM. The median distance between centres of nearest-neighbour spikes in the C9-coalesced spike patches was 185.2 Å (98 Å-309 Å) (Fig. 5i). This distance is notably greater than that between neighbouring spikes in pentagons and hexagons of icosahedral CHIKV particles (117 Å and 120 Å, respectively) (Fig. 5j). This data revealed the molecular mechanism of budding inhibition by anti-alphavirus antibodies: crosslinking spikes to space them apart, thereby preventing lateral spike-spike interactions required to drive the icosahedral shell assembly. This further supports the spike-driven alphavirus assembly/budding model. The absence of ordered spike assemblies in the spike-C9 coalescence also suggests that spikes are again unlikely to be delivered to the PM as pre-assembled lattices, arguing against what was proposed from the observation of hexagonal spike lattice tubes inside cytopathic vacuole type-II in Semliki Forest virus-infected cells 17 . Subsequent virus budding is predicated on assembly of the icosahedral spike lattice that enwraps NLPs and reorganizes them into icosahedral NCs through sequential spike:Cp interactions. Rate-limiting steps to particle formation probably occur at early and late stages associated with assembly of the first half of the virions and membrane scission following completion of full virions, respectively (upside-down red triangles). Released virions contain near-icosahedral spike and NC layers, with local disruptions in the lattices probably related to membrane scission and virus release from the PM. Binding of mAbs to exposed spike surfaces at the PM (boxed) inhibits virus biogenesis by preventing formation of the curved, icosahedral spike shell. b, Spikes can self-assemble into non-icosahedral structures, giving rise to rare alternative assembly products, including helical tubes formed by spike hexagons (i,iii), thin extensions of linked incomplete particles (ii) and planar hexagonal sheets of spikes (iv).

Discussion
Cryo-EM has been a revolutionary tool capable of resolving near-atomic resolution structures of purified macromolecules and macromolecular assemblies both with and without global symmetry. Advances in image reconstruction have further allowed the possibility of resolving multiple structures from compositionally or conformationally mixed assemblies within a single biochemical preparation 18 . This work presents the technical feasibility of imaging an entire virus assembly/budding process in situ and computationally classifying the snapshots into discrete intermediate-state structures at subnanometre to nanometre resolution. Importantly, such an approach is scalable with large datasets, where more particles per class result in more interpretable intermediate structures that give further insights into the assembly path.
Here we captured cytosolic NLPs before budding and the entire progressive co-assembly of two-layered icosahedral shells upon NC interaction with spikes in CHIKV-infected human cells. These structures revealed the molecular mechanism of alphavirus assembly/budding as illustrated by our model (Fig. 6). Briefly, cytosolic NLPs lack observable symmetry and serve as a rough scaffold to trigger icosahedral assembly of the envelope spike lattice. The spike lattice probably grows through lateral self-interactions with additional spikes and progressively reorganizes the underlying, initially asymmetric NCs into expanded icosahedral viral cores through direct spike:Cp contacts. Without underlying NCs, spikes can self-assemble into rare hexagonal lattices covering flat sheets or thin membrane tubes. These spike lattices can potentially shed from the cell as capsid-less subviral particles such as reported for many virus families [19][20][21][22][23][24][25] , or form long extensions to promote virus cell-to-cell transmission although different from other reported virus-induced cytoskeleton-containing cell extensions [26][27][28][29] . Our spike-driven assembly/budding model for alphaviruses can also be applicable to other two-layer icosahedral enveloped viruses such as flaviviruses, which similarly assemble concentric icosahedral spike and NC lattices but apparently without utilizing a preformed NC. In the cell, we show that asymmetric preformed NCs are not required for assembly of highly curved tubular spike lattices but actually initiate formation of a less-curved icosahedral spike lattice of T = 4 symmetry around them. This mechanistic function of NCs in serving as rough scaffolds to promote the less-curved icosahedral architecture of the infectious particle is consistent with the role of flavivirus capsid protein in promoting assembly of T = 3 virions rather than T = 1 subviral particles formed by spikes alone. Further, this NC scaffold function in RNA viruses is reminiscent of double-stranded DNA bacteriophage scaffolding proteins that control the hexamer:pentamer ratio during icosahedral capsid assembly and lead to production of infectious, larger icosahedral particles 30 . Our work reconciles previous contradictory NC-or spike-centric budding models derived from structures of purified fully assembled virions 7 , in vitro-assembled nucleocapsids 31,32 and various mutations disrupting virus assembly and budding [33][34][35] . Similar to this study, we expect that further mechanistic discoveries will be made using this experimental framework to image assembly processes and functional cycles of other biological assemblies in the native cellular context.
Our data also resolved the structural organization of NAbcrosslinked spikes on the CHIKV-infected cell surface and revealed the molecular mechanism for antibody-mediated budding inhibition of alphaviruses 8,9,36 . Crosslinking antibodies serve as molecular wedges between neighbouring spikes and prevent spike-spike lateral interactions required for spike lattice assembly that is shown to drive budding through bending of the PM around NCs. Based on this mechanism, targeting conserved regions of spikes with antibodies 36,37 or other crosslinking molecules that multivalently bind spikes can serve as pan-alphavirus antivirals without the need to neutralize virus entry. Our study thus provides a mechanistic basis for the development of novel therapeutics against alphaviruses and other enveloped viruses with similar spike-driven assembly mechanisms. Acquisition and processing of cryo-ET tilt series. Grids of vitrified virus-infected cells were imaged on two instruments: (1) a Titan Krios microscope (Thermo Fisher) operated at 300 kV with post-column energy filter (20 eV) and K2 Summit detector (Gatan) with a calibrated pixel size of 2.72 Å and (2) a Talos Arctica (Thermo Fisher) operated at 200 kV with post-column energy filter (20 eV) and K2 Summit detector with calibrated pixel size of 3.54 Å. Single-axis bi-directional tilt series were collected using Tomography Software 4.0 (Thermo Fisher, https://www. thermofisher.com/us/en/home/electron-microscopy/products/software-em-3d-vis/ tomography-software.html) or SerialEM software v3.7 39,40 (http://bio3d.colorado. edu/SerialEM/) with low-dose settings and defocus range of −3 to −5.5 µm. For data of CHIKV-181-infected cells collected with the Titan Krios, a total cumulative dose of 110e − A −2 was applied to the specimen, while for data collected with Talos Arctica, the total average dose at the specimen was 90e − A −2 . In both cases, the electron dose was distributed over 51 tilt images, covering an angular range of −50° to +50°, with an angular increment of 2°. Additional data collection on both electron microscopes was conducted using a Volta phase plate, whereby the objective aperture was removed, phase plate inserted and activated, and tilt series collected under the above conditions. The activated Volta phase plate was operated at phase shift 0.3-0.6 radians as measured by AutoCTF software (Thermo Fisher). The motion between frames of each tilt image in the tilt series was corrected using MotionCor2 software v1.3.0 41 (https://emcore.ucsf.edu/ucsf-software). Tilt images were compiled, fiducial-based aligned and reconstructed using IMOD software v4.7 42 (https://bio3d.colorado.edu/imod/), or automatically aligned and reconstructed using EMAN2 software v2.3 43 (https://blake.bcm.edu/emanwiki/ EMAN2). In total, 144 tomograms were judged as sufficient for further analysis from the Titan Krios data collections and 20 tomograms from the Talos Arctica data collections. A summary of the Cryo-ET data collection can be found in Supplementary Table 1.

Cells, virus and antibody. Human bone epithelial cell line U2OS cells (HTB
For analysis of CHIKV-181-infected cells treated with NAb C9, 61 single-axis bi-directional tilt series were collected on the Titan Krios microscope operated at 300 kV with post-column energy filter and K2 Summit detector, and calibrated pixel size of 2.72 Å. Data were acquired using SerialEM software with low-dose settings and defocus range of −3 to −5.5 µm. Tilt series were collected with a total cumulative electron dose of 120e − A −2 distributed over 51 tilt images, again covering an angular range of −50° to +50°, with an angular increment of 2°. Data were exclusively collected using an activated Volta phase plate, with phase shift targeted in the range 0.3-0.6 radians. A total of 51 tomograms were judged as sufficient for further analysis on the basis of achieved phase shift and tomogram reconstruction quality, and were used for subvolume analysis.

Statistics and reproducibility.
Of the 144 reconstructed tomograms from CHIKV-181-infected cells imaged with Talos Artica and selected for further analysis on the basis of proper alignment, 117 contained features of the cell periphery of short extensions with virus budding events (Figs. 1a,b and 2a,b). Twenty tomograms contained strands of incomplete particles extending from the cell surface (Extended Data Fig. 2). Seven tomograms contained cell extensions with self-assembled spike lattice without nearby virus budding or NCs (Extended Data Fig. 7).
Budding intermediate subvolume classification. Subvolume analysis steps were performed using the previously published EMAN2 Tomo pipeline version 2.3 43 . The contrast transfer function (CTF) estimation for each tilt image was performed using the EMAN2 programme 'e2spt_tomoctf.py' . Budding intermediate particles (1,918) were manually picked using the EMAN2 3D slice picker and extracted into subvolumes with ×4,×2,×1 binning. High signal-to-noise ratio (SNR) particles (50) (×4 binning) were picked from the dataset for each of three rough stages of budding (early, mid and late) for initial model generation. The initial model for each budding class was produced using the EMAN2 initial model generation programme 'e2spt_sgd.py' , first imposing c1 symmetry and running 5 iterations. After aligning the C1 initial models to the symmetry axes, 5 additional iterations were run with C5 symmetry imposed for each. These 3 maps were then used as initial models for subtomogram multi-reference refinement (e2spt_refinemulti.py).
The full dataset of 1,918 budding-intermediate subvolumes (×4 binning) was input into EMAN2 multi-reference refinement with 10 initial models (3 early-, 3 mid-, 4 late-budding) and run for 12 iterations, imposing C5 symmetry and limiting resolution to 40 Å for alignments. Due to poor convergence of the earliest-budding classes, all budding particles in the tomograms were re-picked with two points defining an initial budding orientation: one at the centre of NC and the other at the apex of the budding shell. Multi-reference refinement of the pre-oriented subvolumes was repeated as described above, with a refinement angular difference constraint of 30° to prevent particle 'flipping' from the initial rough budding orientation. If a resulting class displayed budding virus structural features with sufficient particle count, those particles were subjected to further classification with either 2 or 3 low-passed versions of the class average as initial references. In this way, particles within 5 of the 10 3D classes were subjected to a second round of multi-reference refinement for further identification of budding conformations, with refinement parameters described above (Extended Data Fig. 3). Between the 2 rounds of classification, 12 different 3D budding structures were determined in total. Subvolume particles within 'junk' class averages lacking interpretable structure were viewed in the original 3D tomograms, revealing that these particles covered a wide range of budding levels and were typically located near high-density gold fiducials that biased the alignment.

Subtomogram averaging of budding intermediates, released virions and NLPs.
For each of those budding intermediate 3D classes (5) displaying low-resolution icosahedral features, particles were re-extracted (×4,×2,×1 binning) for subtomogram refinement (e2spt_refine.py). For each class, 4-6 iterations of refinement were performed for each binned (×4,×2,×1) particle set, imposing C5 symmetry at each step and following gold-standard protocol: all particles were split into 2 independent subsets and resolution measured by Fourier shell correlation (0.143 FSC criterion) of the 2 density maps. Following subtomogram refinement of the least-binned particle set for each class, 2 iterations of sub-tilt refinement (e2spt_tiltrefine.py) with imposed C5 symmetry were performed to produce final budding-intermediate subvolume averages. A summary of the Cryo-ET data collection and subtomogram analysis of viral intermediates can be found in Supplementary Table 1.
Subtomogram averaging of released virions was performed by manually picking and extracting 521 released particles (×4,×2,×1 binning) into subvolumes, followed by EMAN2 3D refinement and sub-tilt refinement. An initial reference for 3D refinement was generated from 50 high-SNR particles with different defocus using EMAN2, with C5 symmetry imposed as described previously. 3D refinement was performed with C5 symmetry imposed, working from ×4 to ×2 to ×1 binned subvolumes as resolution improved. After visual observation of icosahedral structure in the map, icos. symmetry was applied during final sub-tilt refinement of ×1 binned subvolumes. This resulted in a converged map with pixel size of 2.72 Å per pixel and resolution (0.143 FSC criterion) of 8.2 Å.
For subtomogram averaging of NLPs, 545 NLPs apparently within the cytosol of virus-infected cell tomograms were manually picked using the EMAN2 3D slice picker and extracted (×4 binning) into subvolumes. High-SNR particles (50) with varying defocus were used to generate an initial reference with C5 symmetry as described above. Multi-reference refinement of the 545 NLPs (×4 binning) was performed with three classes and similar refinement parameters described above for budding intermediate classification, without the angular difference constraints. This resulted in two cytosolic NLP 3D classes (classes I and II) with interpretable structure (Extended Data Fig. 3). Additional 3D refinements of particles within these two respective classes, with imposed C5 symmetry, resulted in maps with resolutions of 47.6 Å (class I) and 43.5 Å (class II) (gold-standard, 0.143 FSC criterion). The refined orientations of cytosolic NLPs within one class displaying local five-fold symmetry (class II, Fig. 3 and Extended Data Fig. 4) were mapped back in 3D to originating tomogram reconstructions using EMAN2 programme 'e2spt_mapptclstotomo.py' .

Subvolume analysis of NAb-crosslinked spikes and budding-arrested NLPs.
For analysis of the C9-treated CHIKV-181-infected cells, 7,678 individual spikes were automatically picked from tomograms on the basis of a low-resolution reference and judged individually for false positives. Any additional spikes in the tomogram were picked manually. This extensive manual picking protocol was meant to ensure that all spikes were properly extracted for nearest-neighbour distance analysis. 3D subvolumes (×4,×2 binning) of each spike were then extracted and a C3-symmetric initial model was built from a subset of 500 (×4 binning) high-SNR particles using the reference-free initial model programme in EMAN2 (e2spt_sgd.py). The full set of 7,678 (×4,×2 binned) spike particles was then subjected to iterative 3D subtomogram refinement (e2spt_refine.py) with C3 symmetry imposed until no improvement in refined orientations was achieved. The final converged average map had a resolution of 24.4 Å (gold-standard, 0.143 FSC criterion) and pixel size 5.44 Å per pixel. The Euclidean distance between each refined spike and its nearest neighbour in the dataset was determined using the refined centre-of-mass orientations of spike subvolumes in each tomogram.
From the same tomograms, 1,727 budding-arrested NLPs were manually picked, extracted into subvolumes (×4 binned) and an initial model was generated from 50 high-contrast particles in the dataset as described above with C5 symmetry imposed. The initial model was used as input for 3D classification of the 1,727 NLP subvolumes, imposing C5 symmetry, performed using 'e2spt_refinemulti.py' . The multi-reference refinement resulted in six class average structures ranging from ~37 Å to 63 Å resolution (gold-standard, 0.143 FSC criterion) (Extended Data Fig. 3a).
Visualization, figure generation and model docking were performed in UCSF Chimera v1.16 44

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection Serial EM software v3.7 (http://bio3d.colorado.edu/SerialEM/) and Tomography software 4.0 (ThermoFisher Scientific) were used to collect electron cryotomography tilt series data.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy Cryo-EM maps reported in this study have been deposited in the Electron Microscopy Data Bank (EMDB) under the following accession codes: EMDB-26446 (released virion), EMDB-26447, -26448, -26449, -26450 (budding intermediates), and EMDB-26451, -26452 (cytosolic NLPs). The publicly deposited atomic model of

March 2021
VEEV TC-83 (PDB:3J0C) was used for comparison to the subtomogram average structure of the CHIKV spike trimer determined in this study.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
Sample size was not pre-determined in our study. Data was collected such that sample size was sufficient to resolve interpretable biological structures during averaging.
Data exclusions Subvolume classification of virus budding intermediate snapshots resulted in 3D classes with interpretable virus structure and others without clear structure. Particles within those classes that did not align were discarded from further analysis.

Replication
The data here was collected over 10+ independent sessions following independently reproducible replications of the experiments.
Randomization We investigated CHIKV budding intermediates in 100+ cells from 10+ independent sessions and each virus-infected cell produces virus particle at different rates, so the sample itself is naturally randomized. Randomization in sampling is not relevant to our study.

Blinding
Particles were picked up from virus-infected cells based on their features and then subjected to subvolume averaging and classification with EMAN2 software based on their 3D features. Blinding in investigation is not relevant to our study.

Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Authentication
All the cell line were ordered from ATCC and authenticated using ATCC Cell Line Authentication Service with STR profiling.

Mycoplasma contamination
All the cell line used in the study were confirmed mycoplasma negative with Myco-sniff-valid mycoplasma PCR detection kit (MP bio, SKU093050301).