## Main

The Poxviridae family of viruses—which includes the causative agent of smallpox (variola), monkeypox and the smallpox vaccine (vaccinia virus)—has genomes that encode 4 binding proteins and 11 proteins that are required for fusion4. The binding proteins (D8, H3, A26 and A27) are not individually essential6,7,8,9. However, genetic repression of any of the 11 proteins required for fusion, collectively termed the entry fusion complex (EFC)5, results in the formation of morphologically normal virions that are incompetent for hemifusion (A16, A21, F9, G3, G9, H2, J5 and O3) or full fusion (A28, L1 and L5)5. All of the EFC components are transmembrane proteins and nine (A16, A21, A28, G3, G9, H2, J5, L5 and O3) form a stable core complex with which L1 and F9 associate5.

EFCs within mature virions (MVs) are required for fusion of both infectious forms of Poxviridae10—single-membrane MVs and double-membrane extracellular enveloped virions (EEVs), which shed their outermost membrane to allow for fusion of the underlying MV-like particle11,12,13. Here we used a combination of electron microscopy (EM), super-resolution microscopy, single-particle analysis and a large collection of virus mutants to investigate virion binding and fusion orientation, the spatial distribution of binding and fusion proteins on individual vaccinia virus (VACV) particles, and how fusion-protein distribution correlates with fusion activity.

Noting an orientation bias in previous poxvirus EM studies (Supplementary Table 1), we analysed the orientation of VACV MV binding and plasma membrane (PM) fusion using scanning EM (SEM) and transmission EM (TEM), respectively. Quantification of binding orientation showed that more than 99% of MVs bound to the cell surface at the sides of the virions (Fig. 1a). When fusion was forced by lowering the pH13,14,15, fusion of virion and cell membranes occurred at MV tips in 96% of cases (Fig. 1b). Our results were consistent with the literature, in which 98% of MV binding events were shown to occur at the sides of the virions and 100% of fusion events occurred at the tips (Supplementary Table 1).

These results suggest that viral membrane proteins may be organized into functional domains. To investigate this, we applied structured illumination microscopy (SIM)16 and single-particle averaging to generate models of the distribution of binding and EFC proteins in MVs17,18. For this, mCherry-tagged core protein A4 was used to identify virion position and orientation (Fig. 1c), and EGFP-tagged A1319 was used as a viral membrane marker (Fig. 1c). VACV binding (A27 and D8) and EFC proteins (A21, A28, F9, J5, H2 and L1) were visualized using immunofluorescence. The localization models showed that MV binding proteins reside at the sides of the virions and EFC components localize to virion tips, independent of virion orientation (Fig. 1c, Supplementary Fig. 1). Polarity factor quantification indicated that binding proteins were enriched 1.2-fold at the sides, and EFC proteins were enriched 1.7-fold at the tips of virions (Fig. 1d, Supplementary Fig. 2).

The EFC is held together by a complex network of interactions20,21. Repression of any core component results in disruption of the EFC into subcomplexes without compromising the expression or virion incorporation of other EFC proteins5. This allowed us to investigate whether the polarized distribution of binding and EFC components is dependent on the intactness of EFCs. MVs that lack EFC core components A28, G9 or O3 were immunolabelled for binding-protein D8 and EFC protein L1. Localization models showed that D8 distribution was unaffected by the loss of EFC components, whereas L1 was redistributed evenly around the MV membrane (Fig. 1e). Models of A21, A28, F9, H2 and J5 on A28(−), G9(−) and O3(−) MVs showed that all of the EFC components were depolarized in these mutants (Supplementary Fig. 3a). No redistribution of D8 or EFC proteins was seen on virions that lacked the binding-protein H3 (Fig. 1e, Supplementary Fig. 3a,b). Polarity factor quantification confirmed that the deletion of EFC core components does not alter D8 localization, whereas EFC distribution was shifted from polarized to isotropic (Fig. 1f, Supplementary Fig. 3b). Collectively, these results show that VACV binding and fusion machineries are organized as distinct functional domains within the viral membrane (Fig. 1g), and that the polarized distribution of EFCs relies on their intactness.

Using stochastic optical reconstruction microscopy (STORM)22, we extended our investigation to single virions. Immunolabelling of D8 and L1 was performed on wild-type (WT), EFC mutant (A28(−), G9(−) and O3(−)) and binding-protein (ΔH3) mutant MVs. As expected, D8 was distributed to the sides of all virions, whereas L1 was polarized to the tips of WT and ΔH3 MVs and distributed throughout the membrane in A28(−), G9(−) and O3(−) virions (Fig. 2a, Supplementary Fig. 4a). STORM imaging revealed that D8 and L1 were localized to distinct clusters in WT and ΔH3 MVs (Fig. 2a). D8 clusters appeared to be unaffected in the EFC mutants, whereas L1 clustering was largely disrupted (Fig. 2a). Similar D8 and L1 distributions were observed in WT MVs that were stained with fluorescently conjugated primary antibodies or antigen-binding fragments (Fab), indicating that clustering was not induced by signal amplification through secondary antibodies23 (Supplementary Fig. 4b,c). Furthermore, STORM imaging showed that both D8 distribution and L1 polarization is maintained on the MV-like particles within EEVs (Supplementary Fig. 5).

To analyse D8 and L1 clustering on MVs, we applied Voronoi tessellation (SR-Tesseler software24). A13–EGFP was labelled with fluorescently conjugated anti-EGFP nanobodies to determine a baseline for clustering (Supplementary Fig. 4d). Images were subjected to localization, segmentation and cluster identification using SR-Tesseler (Supplementary Fig. 6a). The clustering threshold was set at three times the average density for each particle (Supplementary Fig. 6b). We used this analysis to quantify D8 and L1 clustering in WT, EFC mutants (A28(−), G9(−) and O3(−)) and binding-protein (ΔH3) mutants (Fig. 2b). The percentage of D8 localizations in clusters, and the number of D8 clusters, did not differ between the WT and EFC or binding-protein mutants (Fig. 2c,d). Conversely, L1 localizations in clusters were reduced from 44% to 3%, and L1 clusters were reduced from 8 to ≤4 on EFC mutants compared with the WT (Fig. 2e,f, Supplementary Fig. 6c,d). The total cluster area of L1 on A28(−), G9(−) and O3(−) virions was reduced from 4,400 nm2 per virion on the WT and ΔH3 mutant MVs to below the imaging resolution (Supplementary Fig. 6e). The loss of fusion machinery polarization and clustering observed in EFC mutants correlates with the fusion defects exhibited by these viruses5,25.

It was reported previously that—unlike other EFC mutants—A28(−) MVs are able to direct hemifusion25. EFC polarization and clustering were equally disrupted in A28(−) and other EFC mutants, suggesting that polarization and clustering may be more important for full fusion than hemifusion. To test this, we needed a viral protein that is not a component of the EFC, and the loss of which affects MV fusion. The VACV protein A27 has multiple roles in the virus lifecycle, including virus binding, fusion, and the intracellular transport and wrapping of MVs9,26,27,28,29,30. Direct evidence of the involvement of A27 in fusion was provided by Vazquez and Esteban, who showed that A27(−) virions cannot mediate acid-induced cell–cell fusion26. This phenotype was confounded by the fact that the A27(−) virus had no defect in MV production. Using 24 h yield and cell–cell fusion experiments, we confirmed that A27(−) MVs display no defect in MV production (Fig. 3a), but are eightfold less capable of mediating cell–cell fusion than A27(+) virions (Fig. 3b,c).

Lacking a transmembrane domain, A27 is tethered to the MV surface by the membrane protein A1727. STORM imaging showed that A27 and A17 are homogeneously distributed on MVs, suggesting that there is no exclusive interaction with EFCs (Fig. 3d, Supplementary Fig. 4e). However, localization models of D8 and EFCs in A27(−) MVs indicated that their distribution on these virions was shifted (Fig. 3e, Supplementary Fig. 7a). Interestingly, A27(−) MVs phenocopied EFC mutants, displaying a complete loss of fusion protein polarization (Fig. 3f, Supplementary Fig. 7b). STORM imaging showed that D8 clusters appeared redistributed and that L1 clusters were reduced and redistributed on A27(−) MVs (Fig. 3g,h). D8 clusters were reduced from nine to seven with no difference in the cluster area covered by D8 (Supplementary Fig. 8a,b). L1 clusters were reduced from nine to six, and the total L1 cluster area was reduced from 3,015 nm2 on A27(+) virions to 1,411 nm2 (Supplementary Fig. 8c,d). Applying Ripley’s H-function analysis to L1 in WT, A27(−) and G9(−) MVs validated and confirmed31 (Supplementary Fig. 8e) that A27 is required for polarized clustering of EFCs on MV tips (Fig. 3i). These results show that EFC polarization requires both EFCs being intact and A27. As A27 is not a component of the EFC, these results strongly suggest that the loss of EFC polarization in A27(−) virions underlies their inability to mediate the full-fusion reaction required for cell–cell fusion26.

We therefore investigated whether A27(−) MVs can undergo acid-induced hemifusion and full fusion with cells25,32 using the assay that is illustrated in Fig. 4a. MVs containing an enhanced green fluorescent protein (EGFP) core were labelled with the self-quenching membrane dye R18. The MVs were bound to cells in medium at 4 °C and pH 7.4, and medium (37 °C and pH 5.0) was added to induce MV fusion with the PM. Proton influx into virions quenches the pH-sensitive EGFP core fluorescence, followed by R18 dequenching as a result of lipid mixing during hemifusion32. On full fusion, cores were exposed to cytosolic pH enabling the recovery of EGFP core fluorescence. Measurement of the R18 and EGFP fluorescence over time allowed for the quantitative assessment of MV hemifusion and full fusion25,32. Comparison of R18 dequenching rates indicated that A27(+) and A27(−) MV hemifusion was comparable (Fig. 4b). Although core EGFP quenching occurred in both A27(+) and A27(−) MVs, EGFP recovery was observed only in A27(+) MVs (Fig. 4c). These results indicate that A27(−) MVs can undergo hemifusion but their ability to mediate full fusion is impaired.

To assess how A27(−) MVs produce equivalent numbers of infectious virus as A27(+) MVs despite being impaired for full fusion, cells infected with A27(+) or A27(−) MVs that expressed EGFP under the control of an early viral promoter were monitored from 2 h until 8 h after infection (Fig. 4d, Supplementary Fig. 9). This assay—which we used as a surrogate to assess delayed VACV fusion25,32—showed detectable early gene expression in 61% of A27(+)-infected cells and 38% of A27(−)-infected cells at 2 h after infection. The percentage of A27(+)-infected cells expressing early genes increased only slightly over 8 h, whereas the percentage of A27(−)-infected cells expressing early genes increased steadily, reaching the levels of A27(+) within 8 h. The ability of A27(−) MVs to overcome delayed full-fusion kinetics (Fig. 4c) is consistent with the production of equivalent numbers of MVs in A27(+) and A27(−) infections over a 24 h period (Fig. 3a).

The experimental data indicate that clustered polarization of the fusion machinery is important for VACV full-fusion efficiency. To further explore the impact of EFC clustering on fusion, we implemented a simple kinetic model33. The contact area between the virus and cell was modelled to contain a variable density of fusion complexes that stochastically activate to drive fast hemifusion followed by rate-limiting full fusion. By varying the density of the fusion complexes within the virus–cell contact area, we could assess how fusion-complex clustering effects the rate of full fusion. We tested this model at two fusion thresholds33—low, which requires three activated fusion complexes, and high, which requires five activated fusion complexes. Simulations of 1,000 viruses indicated that, regardless of the fusion threshold, the rate of full-fusion kinetics decreases when fusion complex density is reduced from high to low (Fig. 4e). Notably, at the high threshold, the rate of fusion was more dependent on complex density, indicating that full fusion is highly dependent on complex clustering.

On the basis of these collective findings, we propose that the organization of poxvirus membrane proteins into functional domains is important for virus entry, and that EFC polarization and clustering is critical for efficient MV and EEV virus–cell fusion. The only other documented example of polarized virus fusion machinery is in HIV-134. The authors proposed that Env clustering is important owing to the low number of Env trimers present in the viral membrane34. Formation of a single Env cluster on mature HIV virions is required for efficient CD4 engagement and fusion at the PM. Although a minority of VACV MVs and EEVs enter by PM fusion, the majority enter by acid-mediated endocytosis13,15,35,36,37,38,39. This—together with our findings—invokes a model in which VACV MVs bind to the PM in a side-on orientation (see Fig. 4f). This is consistent with MV binding data, the position of the binding machinery on the virus and accounts for the low number of direct PM fusion events that were observed after low-pH treatment.

During endocytosis and, in the case of EEVs, low-pH-mediated rupture of the outer membrane32, both MVs and MV-like particles are completely enveloped within cellular membrane compartments. This allows for virus orientation-independent contact between virion tips and the limiting membrane of endosomes to drive EFC-mediated fusion and core deposition. Consistent with this model, MV and EEV fusion from endosomes has only been observed to occur at virion tips15,40,41.

In summary, by revealing the nanoscale organization of the poxvirus membrane and showing the consequences of its disruption, we demonstrate that virion protein architecture is critical to virus function. We suggest that the organization of the VACV membrane into functionally distinct domains has evolved as a mechanism to maximize virion binding and fusion efficiency for productive infection.

## Methods

### Cells and viruses

African green monkey kidney (BSC-40) cells and human HeLa cells were cultivated in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS), 2 mM GlutaMAX, 100 units ml−1 penicillin and 100 μg ml−1 streptomycin, 100 μM non-essential amino acids and 1 mM sodium pyruvate. Routine mycoplasma tests of the cell culture medium were negative. Recombinant VACV strains were based on the VACV strain Western Reserve (WR). WR mCherry–A4 and WR L4–mCherry/EGFP–F17 were described previously as WR mCherry–A5 (ref. 13) and WR EGFP–F17 VP8–mCherry, respectively42. WR mCherry–A4 F13–EGFP was described previously as WR mCherry–A5 F13–EGFP (ref. 13). ∆H3, A28(+/−), G9(+/−) and O3(+/−) were described previously as vH3∆43, vA28–HAi10, vG9i44 and vO3–Hai45, respectively. WRA27(+/−) was previously described as VVIndA27L (ref. 30). WR A4–mCherry/A13–EGFP, WRA27(+/−) EGFP–A4 and WRA27(+/−) early/late (E/L) EGFP were constructed as previously described42. For WR A4–mCherry/A13–EGFP, A13 was replaced with A13–EGFP at its endogenous locus within the WR A4–mCherry virus. First, the primers GCGCCTCGAGATCTCGACATTGTTGAATCATTATTAC and GCGCGGTACCCACCAGAAGTATTTTTGGAGCC were used to amplify the A13 region of the viral genome with XhoI and KpnI sites, which was inserted into the pBluescript II KS (+) backbone. An EGFP tag was then inserted using Gibson assembly. To generate WRA27(+/−) EGFP–A4, A4 was replaced with EGFP–A4 at its endogenous locus within WRA27(+/−). First, the primers CCATCGATGATGACTATAGGACAAGAACCCTCCTC and CGGAATTCCGCTTGAACAGCATTGC were used to amplify the A4 region of the viral genome with EcoRI and ClaI sites, which was then inserted into the pBluescript II KS backbone (+). An EGFP tag was then inserted using Gibson assembly. For WRA27(+/−) E/L EGFP, EGFP under the control of an early/late viral promoter was inserted between the H2 and H3 loci of the WRA27(+/−) virus. First, the primers GCGCGGTACCCTAGCCGCTGGTAAGGATGA and GCGCGGTACCGCAGATACTGGATAATGCCG were used to amplify the H2–H3 region from the viral genome with KpnI sites, which was then inserted into the pUC/neo backbone. Then, E/L EGFP was introduced into the H2–H3 locus using the primers GTACAAGTAAGAATTCTGTTAGATAAATGCGGTAACGAAT and TTTAGTAATATGGAATAGAAGCTTAAAAATTGAAATTTT to introduce EcoRI and HindIII sites into the H2–H3 region, and the primers AAGCTTAAAAATTGAAATTTTATTTTTTTTT and GAATTCTTACTTGTACAGCTCGTCC were used to introduce the same sites into E/L EGFP. In brief, BSC-40 cells were infected with the parental viruses and subsequently transfected with linearized plasmid that contained the region of genome to be replaced and was flanked at its 5′ and 3′ ends by 300 bp of genomic sequence for targeted homologous recombination. Recombinant viruses were selected by fluorescence through four rounds of plaque purification. All viruses were produced in BSC-40 cells, and MVs were purified from cytoplasmic lysates through bands on sucrose gradients as previously described35. For WRA27(+/−) and its derivatives, virus stocks were generated in the presence or absence of 2 mM isopropyl β-d-thiogalactopyranoside (IPTG, Sigma) to obtain A27+ or A27− MVs. WRA28(+), WRG9(+) and WRO3(+) stocks were produced using 100 μM, 50 μM and 20 μM of IPTG, respectively. EEVs were prepared as previously described13. In brief, RK13 cells were infected with WR F13–EGFP mCherry–A4 with an MOI of 1 and EEVs were purified from the supernatant at 24 h after infection. Debris was removed by centrifugation at 1,000g and EEVs were concentrated by centrifugation at 38,000g before being resuspended in 1 mM Tris pH 9.

### Antibodies

Anti-L1 mouse monoclonal antibodies (clone 7D11) were purified from a hybridoma cell line that was provided by B. Moss (National Institutes of Health) with permission of A. Schmaljohn (University of Maryland). Anti-D8 rabbit polyclonal antibodies were made by immunizing a rabbit with purified D8 protein and adjuvant. Anti-A17 rabbit polyclonal antibodies were a gift from J. Krijnse-Locker. Antibodies against viral proteins A21 (R206), A28 (R204), F9 (R192), H2 (R202) and J5 (R264) were produced by G.H.C. using purified recombinant baculovirus-expressed proteins as previously described46.

### Structured illumination imaging

High-performance coverslips (18 × 18 mm, 1.5H, Zeiss) were washed as previously described17. Purified virus was diluted in 1 mM Tris pH 9, bound to the ultra-clean coverslips for 30 min and fixed using 4% formaldehyde. Samples were washed three times with PBS before mounting for imaging. To visualize membrane proteins, virus was blocked after fixation using 5% bovine serum albumen (BSA, Sigma) in PBS for 30 min, incubated in 1% BSA in PBS with primary antibody, followed by Alexa Fluor 488-conjugated goat anti-mouse or anti-rabbit secondary antibody for 1 h each. Samples were washed three times with PBS after each staining step. The coverslips were mounted in VECTASHIELD (Vector Laboratories) and sealed with nail polish.

SIM imaging was performed using a Plan-Apochromat ×63/1.4 NA oil differential interference contrast (DIC) M27 objective and an ELYRA PS.1 microscope (Zeiss). Images were acquired using five phase shifts and three grid rotations, the 561 nm (32 μm grating period) and the 488 nm (32 μm grating period) lasers, and filter set 3 (1850–553, Zeiss). Two-dimensional images were acquired using a sCMOS camera and processed using the ZEN software (2012, v.11.0.3.190, Zeiss). For channel alignment, TetraSpeck beads (ThermoFisher) were mounted on a slide, imaged using the same image acquisition settings and used for the alignment of the different channels.

### Single-particle analysis

Individual viral particles were extracted from the SIM images. Seed images were generated using VirusMapper v1.0 as described previously17. VirusMapper models were then created by registration of the entire set of particles according to cross-correlation with the seeds and calculation of a weighted average of a subset of particles. These values of (n) are shown in Supplementary Figs. 3 and 7. Models are normalized and therefore intensity does not reflect protein abundance.

### Polarity factor

Polarity factors were calculated directly from the sets of particles that were used to generate the models (Supplementary Fig. 2). Following the cross-validation method of Szymborska et al.47, particles were randomly divided into subsets of 50 particles and separately averaged. Radial profiles were generated from these images by transforming from xy coordinates to rθ. The radial profiles were divided into four regions according to the parameter φ and the mean intensity within the viral membrane in these regions was evaluated. The four regions were defined in θ by:

$$I_1:360^\circ - \varphi < \theta \,{\mathrm{or}}\,0 < \theta < \varphi$$
$$I_2:\varphi < \theta < 180^\circ - \varphi$$
$$I_3:180^\circ - \varphi < \theta < 180^\circ + \varphi$$
$$I_4:180^\circ + \varphi < \theta < 360^\circ - \varphi$$

where $$0 < \theta \le 360^\circ$$. The images are therefore divided by two lines through the centre of the image that intersect at angle φ and divide the image symmetrically about the horizontal and vertical axes. The polarity factor was then calculated as

$$p = \frac{{I_2 + I_4}}{{I_1 + I_3}}$$

A value for φ was used that resulted in a mean polarity factor of 1 for the A4 core protein (48.5°). The values quoted in the text were calculated by averaging the mean polarity factors for the sets of binding and fusion proteins.

### Labelling of primary antibodies

Mouse monoclonal antibodies (IgG) against L1 from hybridoma supernatant and rabbit polyclonal antibody against D8 from whole serum were purified using a NAb Protein A/G Spin Kit (Thermo Scientific) according to the manufacturer’s instructions. Fab fragments were generated from primary antibodies with a Pierce Fab Preparation Kit (Thermo Scientific). Antibodies and Fab fragments were buffer exchanged into 0.1 M NaHCO3 pH 8.2 and concentrated to greater than 1 mg ml−1 using Amicon Ultra centrifugal filters 10 kDa molecular weight cut-off (MWCO; Merck). Antibodies and Fab fragments were custom-labelled with a 50-fold molar excess with Alexa Fluor 647-NHS (Invitrogen) for 1 h at room temperature. The reaction was quenched by the addition of 2 µl of 1 M Tris pH 9. Unreacted dye was removed by three passes through Zeba Spin columns 3.5K MWCO (Thermo Scientific).

### STORM imaging

Single-molecule localization microscopy was performed by direct STORM48, a method developed on the basis of STORM22. High-performance coverslips (18 mm, 1.5H, Zeiss) were washed with ultrapure ethanol (Sigma) and deionized water to clean them and make their surface hydrophobic. Purified MVs or EEVs were diluted in 20 µl of 1 mM Tris pH 9, placed in the centre of the clean coverslips for 30 min and the bound virus was fixed using 4% EM-grade formaldehyde (EMS). The viruses were blocked after fixation using 5% BSA (Sigma), 1% FCS and 0.2–1% Triton X-100 in PBS for 30 min. The viruses were immunostained in 5% BSA in PBS with primary antibodies at 4 °C overnight and Alexa Fluor 647-conjugated secondary antibodies (Invitrogen) for 1–2 h at room temperature. The samples were washed three times with PBS after each staining step. Optionally, a step was included after fixation using 4% PFA. Autofluorescence was quenched by brief incubation in 0.25% (w/v) NH4Cl in PBS. Coverslips were mounted on a Secure-Seal incubation chamber (EMS) in BME buffer (1% (v/v), β-mercaptoethanol (Sigma), 150 mM Tris, 1% glucose, 1% glycerol and 10 mM NaCl, at pH 8) or in MEA-buffer49 (50 mM cysteamine (Sigma), 3% (v/v) OxyFluor (Oxyrase Inc), 20% sodium lactate (Sigma) and PBS, at pH 8).

Imaging was performed on an Elyra PS.1 inverted microscope (Zeiss) using an alpha Plan-Apochromat ×100/1.46 NA oil DIC M27 objective with a ×1.6 tube lense and an iXon 897 EMCCD camera (Andor). Images were acquired using a 25–30 ms exposure time with 642 nm excitation at 100% laser power and a 655 nm longpass filter. Fluorophore activation was dynamically controlled using a 405 nm laser at 0–2% laser power. Images were processed in Fiji (ImageJ v.15250 using ThunderSTORM51. Localizations were fitted with a maximum-likelihood estimator, lateral drift was corrected by cross-correlation, localizations that were less than 20 nm apart with ≤1 frames in between were merged, and images were rendered using a Gaussian profile with the NanoJ-Orange LUT (NanoJ v1.1). Lateral resolution was 25 nm, as determined by Fourier ring correlation. Dual-colour STORM and SIM images (Supplementary Figs. 4 and 5) were registered using NanoJ52.

### SR-Tesseler analysis

Cluster analysis was performed using SR-Tesseler24. Localization tables from ThunderSTORM were imported and Voronoi diagrams were created. Individual virions were selected as regions of interest and segmented as single objects with a density factor δ of 0.1–0.5. Within individual objects, clusters were identified with δ = 3 that yielded less than 2% clustering in the non-clustered reference probe A13–EGFP (Supplementary Fig. 6). Statistical analysis was performed using GraphPad Prism 7 (Prism Software). Significance was calculated using unpaired t-tests: *P < 0.05; **P < 0.01; ***P < 0.001.

### Ripley’s H-function analysis

Additional cluster analysis was performed as previously described31. Ten representative virus particles were selected from STORM images of L1 on WT, A27(+/−), G9(−) VACV (Supplementary Fig. 8e). Localizations within the selected regions of interest were separately used to calculate the Ripley’s H-function as a function of increasing radius H(r) according to:

$$K(r) = \frac{A}{{n({n - 1})}}\mathop {\sum }\limits_j^n \mathop {\sum }\limits_i^n \delta _{ij}(r)$$

where A is the approximate area of the particle, n is the number of localizations and

ij (r) = 1 if |xi – xj| < r; 0 otherwise

where xi is the spatial location of localization i. Then

$$L\left( r \right) = \frac{1}{\pi }\sqrt {K\left( r \right)}$$

and

$$H\left( r \right) = L\left( r \right) - r$$

so that the expected value of H(r) is 0 for a random uniform distribution, and values of H(r) above 0 indicate clustering at a range of approximately r. Custom Python scripts were used to calculate the H-function from STORM localization tables. The mean of H(r) and the 95% confidence interval were evaluated for each case.

### The 24 h yields

Confluent BSC-40 cells were infected with A27(+/−) virus in DMEM with an MOI of 1. After 60 min, the inoculum was removed and DMEM with serum and 2 mM IPTG was added for 24 h. Cells were collected, resuspended in 100 μl of 1 mM Tris pH 9 and freeze–thawed three times in liquid nitrogen. The virus concentration was then measured by plaque assay.

### Fusion-from-without experiments

WRA27(+/−) was bound to HeLa cells that were grown on coverslips on ice with an MOI of 10 for 1 h. The cells were washed and incubated at 37 °C with 20 mM 2-(N-morpholino)ethanesulfonic acid (MES) pH 5 or pH 7.4 for 5 min, then washed and incubated in full medium for 2 h. Cells were fixed with 4% formaldehyde in PBS for 20 min, blocked and permeabilized using 0.1% Triton X-100 and 5% BSA in PBS, stained with Alexa Fluor 488 Phalloidin and Hoechst 33258 (both Invitrogen) in PBS for 1 h. Samples were then mounted on slides with ProLong Gold Antifade Mountant (Thermo Fisher). Samples were imaged on a Leica TCS SP8 STED 3X microscope in confocal mode running LAS X v.2.01 acquisition software. Alexa Fluor 488 fluorescence was excited using the 488 nm line and Hoechst was excited with the 405 nm LED. Subsequently, z-stacks were acquired at a scan speed of 400z Hz in bidirectional scan mode using a Hybrid Detector (HyD, standard mode) and an Acousto-Optical Beam Splitter for filtering and time-gating of 0.5–8 ns. The fusion index was calculated from maximum intensity projections of the z-stacks as described previously53.

### Bulk fusion measurements

MVs were labelled by incubating with 22.5 μM R18 (ThermoFisher) in 1 mM Tris pH 9 at room temperature for 2 h. Labelled viruses were pelleted by centrifugation at 16,000g for 10 min at 4 °C and then resuspended in 1 mM Tris pH 9 twice to remove excess R18. Labelled viruses were bound to 7 × 105 HeLa cells with an MOI of 30 in DMEM on ice for 1 h. Cells were sedimented by centrifugation at 300g for 5 min at 4 °C then resuspended in 100 μl PBS. The cell suspension with bound virions was added to 630 μl of prewarmed PBS in a quartz cuvette. After 2 min, the pH in the cuvette was lowered by the addition of 100 μl of 100 mM MES, resulting in a pH of 5.0. After the acquisition, all R18 was dequenched by the addition of 83 μl 10% Triton X-100 in PBS. R18 fluorescence was normalized to the signal intensity after the addition of Triton X-100. R18 fluorescence was measured using a Horiba FluoroMax-4 (Horiba Jobin Yvon) spectrofluorometer with an excitation wavelength of 560 ± 5 nm and an emission wavelength of 590 ± 5 nm. To measure EGFP fluorescence, unlabelled viruses were bound to cells and the pH was lowered as above, and fluorescence was recorded using an excitation wavelength of 488 ± 5 nm and an emission wavelength of 509 ± 10 nm.

### Flow cytometry

HeLa cells were infected with WR A27(+/−) E/L EGFP with an MOI of 4 in DMEM and full medium containing 10 μM cytosine β-d-arabinofuranoside (AraC; Sigma) was added after 1 h. Cells were washed with PBS, trypsinized and resuspended in PBS and fixed with 4% formaldehyde in PBS at the indicated times. Cells were then sedimented by centrifugation at 300g for 5 min and resuspended in PBS with 2% FBS and 5 mM EDTA. Flow cytometry was performed using a Guava easyCyte HT flow cytometer, recording the EGFP fluorescence with the 488 nm laser. Flow cytometry data were analysed using FlowJo v10.

### Fusion-kinetics mathematical model

We modelled the process of fusion undergone by a single fusion complex as a fast irreversible hemifusion step followed by a rate-limiting, irreversible full-fusion step.

$$I\mathop{\longrightarrow}\limits^{{k_{\mathrm{hf}}}}{\mathrm{HF}}\mathop{\longrightarrow}\limits^{{k_{\mathrm{fuse}}}}F$$

Where I, HF and F are initial, hemifused and fused states, respectively. Following similar work on influenza33, for each virus, we required that Tf (fusion threshold) fusion complexes transition to state F before the virus is considered fused. We simulated the contact area with the cell as containing an initial density $$\rho _0$$ (high density), $$\frac{{\rho _0}}{3}$$ (medium density) and $$\frac{{\rho _0}}{{10}}$$ (low density) of fusion complexes. We then modelled the effects of fusion-complex density on fusion kinetics at two different values of Tf—low (Tf = 3) and high (Tf = 5). The true value of Tf is unknown for vaccinia but for influenza is thought to be in this range.

### Electron microscopy

For binding-orientation experiments, VACV was bound to HeLa cells for 15 min on ice with an MOI of 50. The unbound virus was removed, and the samples were fixed for 20 min using fixation buffer (final concentration of 1.5% formaldehyde, 0.1% glutaraldehyde, both EM grade). The fixation solution was replaced with 2% glutaraldehyde in 0.1 M sodium cacodylate for a further 20 min. The samples were washed in 0.1 M sodium cacodylate and incubated for 1 h at 4 °C in reduced osmium tetroxide. After further washes, samples were dehydrated through a series of increased ethanol and critical point dried (Leica Auto CPD). Samples were gold and palladium coated and imaged using a FEI Quanta 200 FEG ESEM (Thermo Fisher Scientific) operated at 5 kV.

For fusion-orientation experiments, VACV was bound to HeLa cells for 30 min on ice with an MOI of 100. Unbound virus was removed, and the samples were incubated for 7.5 min at 37 °C in DMEM with 100 mM MES adjusted to pH 5. Samples were fixed with 1.5% glutaraldehyde, 2% paraformaldehyde (EM-grade) in 0.1 M sodium cacodylate for 45 min at room temperature and prepared for TEM imaging. Transmission electron micrographs were obtained using a Tecnai T12 FEI equipped with a charge-coupled device camera (SIS Morada; Olympus).

### Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.