Protein production and purification

To produce soluble F13 (sF13WT and mutants), we cloned a synthetic gene, codon-optimized for expression in Escherichia coli into a pET-28a(+) vector (Novagen) with an N-terminal His- and Strep-tag followed by a thrombin site. The sequence is derived from the Western Reserve strain of VACV (Uniprot code, P04021). We removed the hydrophobic N-terminal tail (residues 2–5) and introduced five mutations in the MIR: W177A, L178A, C181A, C185A and C186A, to remove the palmitoylation sites and the hydrophobic residues around. The point mutants mentioned in the text were introduced into this vector. We transformed E. coli BL21 (DE3) cells (New England Biolabs) and induced protein expression overnight at 16 °C with 0.25 mM isopropyl β-d-1-thiogalactopyranoside. We collected cells from 3 l of culture, resuspended them in 40 ml cold resuspension buffer (Tris–HCl 10 mM pH 8, NaCl 150 mM, EDTA 1 mM) supplemented by one tablet of complete protease inhibitor (Pierce) and froze them at −20 °C. The next day, we thawed and lysed them using a sonicator. After removing the insoluble material by centrifugating at 20,000 g (30 min, 4 °C), we purified the recombinant protein using streptag-based affinity chromatography in a StrepTrapTM HP 5 ml column (Cytiva), treated it with 5 mM Tris(2-carboxyethyl)phosphine hydrochloride (Thermo Fisher Scientific) for 10 min at room temperature to reduce all exposed cysteines and purified the protein using size-exclusion chromatography with a Superdex 75 column (Cytiva) using SEC buffer (Tris–HCl 10 mM pH 8, NaCl 100 mM). The final yields obtained were: sF13WT = 3.5 mg l−1, sF13A295E = 2.5 mg l−1, sF13G277C = 2.8 mg l−1, sF134MUT = 1.4 mg l−1, sF13Δ267 = 0.6 mg l−1, sF13R291E = 1 mg l−1, sF13S292F = 1.9 mg l−1, sF13S292K = 0.2 mg l−1 and sF13L296Y = 0.4 mg l−1. All proteins were analysed by SDS–PAGE to assess their purity (Extended Data Fig. 8a).

Crystallization and structure determination

For crystallization, we digested the purification tags using 1.5 units of thrombin (Cytiva) per 0.1 mg of protein overnight at 4 °C and then treated the protein with Tris(2-carboxyethyl)phosphine hydrochloride for 10 min at room temperature. The digest was loaded to a gel filtration Superdex 75 16/60 column in 10 mM Tris–HCl pH 8.0, 100 mM NaCl, and the fractions of the main peak were pooled and concentrated to 12 mg ml−1 in the same buffer for crystallization trials. Crystallization screening trials were carried out by the vapor diffusion method using a Mosquito TM nanodispensing system (STPLabtech) following established protocols36. Monoclinic crystals of sF13WT were grown after 10 days in 20% (w/v) PEG 3350, 0.1 M HEPES pH 7.5 and 2% (v/v) Tacsimate and were cryoprotected in the same solution supplemented with 20% (v/v) glycerol. Cubic crystals of sF13WT were grown in 1 day in 1 M Na3 citrate and 0.1 M imidazole pH 8 and were cryoprotected using the crystallization solution supplemented with 33% (v/v) glycerol. To obtain the complex with tecovirimat, we soaked cubic crystals, which have a high solvent content (70%), for 5 min into a soaking solution containing 1 mM tecovirimat (BenchChem, catalogue number B611274), 10% (v/v) DMSO, 1 M Na3 citrate and 0.1 M imidazole pH 8. To obtain the complex with IMCBH, we soaked cubic crystals into a solution containing 1 mM IMCBH (BLD Pharmatech, catalogue number BL3H9998EC8C), 10% (v/v) DMSO, 1 M Na3 citrate and 0.1 M imidazole pH 8. After soaking, all crystals were cryoprotected using the soaking solution supplemented with 33% (v/v) glycerol. Similarly, cubic crystals of sF13A295E were obtained in 1 day using 1 M Na3 citrate and 0.1 M imidazole pH 8 and soaked with tecovirimat as reported above.

X-ray diffraction data were collected on beamlines PROXIMA-1 and PROXIMA-2A at the synchrotron SOLEIL (St Aubin, France) using the beamline control software MXCuBE (version 2). Diffraction images were integrated with XDS (version 10 January 2022)37, and crystallographic calculations were carried out with programs from the CCP4 program suite (version 9)38. To determine the phases, we used a model of F13 obtained using AlphaFold2 (ref. 39) as a template to perform molecular replacement in PHASER (version 2.8.3)40. To obtain the final models, we iteratively built and refined the structures using phenix.refine (Phenix version 1.19.2-4158)41 and Coot (version 0.9.8.95)42 using isotropic B factor and Translation/Libration/Screw groups as refinement strategy. We validated all the models using MolProbity (version 4.5.2)43. The crystallographic statistics are provided in Supplementary Table 1. In crystals soaked with tecovirimat or IMCBH, additional electron density appeared at the dimeric interface that was compatible with the shape and size of the drug, as shown in Fig. 2. To facilitate the modelling, all maps derived from the cubic crystals were corrected using a bulk-solvent mosaic model available in the PHENIX program (phenix.mosaic). The electron density for tecovirimat does not show clear features. We hypothesize that this is because the binding pocket is symmetric, while the molecule itself is not, allowing it to adopt two indistinguishable orientations rotated by 180°. However, it is also possible that this is a crystallographic artefact, with tecovirimat only entering in a single orientation, and the density is featureless because of the presence of a twofold symmetry axis crossing the molecule. To investigate this, we reprocessed the cubic crystals in the space group P1 and refined the tecovirimat molecule in two different ways: in a single orientation with 100% occupancy and in two orientations rotated by 180° with 50% occupancy each, mimicking what is observed in the cubic space group. When comparing the two refinements (Extended Data Fig. 8b), we observed improved R factors and reduced residual densities when using the model with two rotated molecules, supporting our original hypothesis. Coordinates and structure factors have been deposited in the Protein Data Bank. Figures showing the crystallographic models were generated with PyMol v3.0.3 (Schrödinger, LLC).

Molecular dynamics simulations

To assess the stability of the X-ray resolved F13 dimer on the membrane surface, we conducted molecular dynamics simulations. Before simulation, we made several structural modifications to the sF13 dimer. First, the unresolved N-terminal residues (residues 1–5) were modelled as unstructured and integrated into the dimer structure using the Modeller (version 10.4) tool44. Next, the structure was processed using the CHARMM-GUI (accessed on December 2023)45 server to add post-translational palmitoylation on residues C185 and C186 and neutral capping of the N- and C-terminal residues. Subsequently, the F13 dimer was positioned on a membrane (Fig. 1c) mimicking the lipid composition of the Golgi membrane (as outlined in Supplementary Table 2). For the protein force field, we used the CHARMM36m-WYF force field46,47, which includes corrections for cation–pi interactions, while lipids were described using the CHARMM36 force field48. The protein–membrane system was solvated with 52,801 water molecules using CHARMM-modified TIP3P49 water model, and the total charge of the system was neutralized by adding 82 K+ ions. The total system size is 249,131 atoms. The box dimensions were 14.34 × 14.34 × 13.56 nm in the x, y and z directions. The solvated system was energy minimized using the steepest descent algorithm to remove any steric clashes, followed by six short equilibrations ranging from 125 ps to 500 ps with restraints on either protein backbone/side chain atoms or lipid phosphate atoms.

Throughout the equilibration process, we maintained a temperature of 310 K using the Berendsen thermostat50 with a time constant (τt) of 1 ps, while pressure was maintained at 1 bar using the Berendsen semi-isotropic scheme with a time constant (τp) of 5 ps. van der Waals and electrostatic interactions were treated using the cut-off and Particle Mesh Ewald51,52 methods, respectively, with a cut-off of 1.2 nm. Covalent bonds involving hydrogen atoms were constrained using the LINCS algorithm53. For the final production run, we removed all restraints and switched to V-rescale54 and Parrinello–Rahman55,56 semi-isotropic scheme to regulate the temperature and pressure, respectively. The rest of the parameters were consistent with those used during equilibration. The production simulations were conducted for 1 µs with 5 repeats using the GROMACS (version 2021) simulation package57, using a time step of 2 fs. For analysis, we concatenated the last 300 ns from each repeat and examined monomer–monomer contacts within 5 Å. All images and plots were generated using VMD (version 1.8.1)58 and the matplotlib (version 3.10)59 library.

Although the structure of the sF13 dimer was determined at a resolution of 2.6 Å, a crystallographic 2-fold axis passes through the ligand density (tecovirimat), leading to challenges in accurately fitting the ligand. To enhance the accuracy of ligand modelling within the density, we used the recently developed RosettaEMERALD (rosetta release-362)60 protocol. This protocol integrates both RosettaGenFF and genetic algorithm optimization for robust ligand modelling within the density map. The three-dimensional structure of tecovirimat was downloaded from PubChem61 in its endo-isomeric form, which is the favoured product of the Diels–Alder reaction used to synthesize the drug. Next, the sF13 dimer–tecovirimat complex was docked to the density using the ChimeraX (v1.7.1) tool62. Following this, we used the RosettaEMERALD protocol to accurately model tecovirimat within the density. Briefly, an initial pool of 500 ligand conformations, along with protein side chains, undergo genetic algorithm optimization over 10 generations. The top 20 lowest-energy conformations obtained from genetic algorithm optimization were further refined, along with protein side chains, using a cartesian minimization in Rosetta. The protocol was executed in triplicate. Out of the total 60 ligand conformations, redundant poses were eliminated, and 15 poses were selected for binding free energy calculations. These 15 selected poses are indicated by the first number of x-axis labels of Extended Data Fig. 3. The RosettaEMERALD XML script and flags used for refining the ligand within the density are provided in the supplementary material.

ABFE calculation

The selected 15 poses were subjected to binding free energy estimation using an in-house pipeline (publication in preparation). Our ABFE protocol is similar to the one previously described by ref. 63. ABFE calculations were performed in triplicate for each pose. To optimize the ABFE calculations, the membrane was excluded from the simulation. This simplification is justified as the tecovirimat binding pocket is located at 40 Å from the membrane surface, and equilibration simulations show that the interface remains stable throughout the simulation (Extended Data Figs. 9 and 10). The AMBER-ff14sb64 parameters for F13 dimer were acquired through the use of OpenMM (version 8.0)65 and ParmEd (version 4.1.0)66 software, while the OpenFF-2.0.0 (ref. 67) parameters with AM1-BCC charges for tecovirimat were obtained via TOFF68 software v0.1.0. The TIP3P49 water model was used, along with AMBER parameters for ions. GROMACS-2022.4 (ref. 57) simulation package was used as the molecular dynamic engine.

In all cases, the simulation temperature was maintained at 298.15 K using Langevin dynamics with a collision frequency of 2 ps−1. Van der Waals and electrostatic interactions were treated using the cut-off and Particle Mesh Ewald methods51,52, respectively, with a cut-off of 1 nm. Hydrogen bonds were constrained using the LINCS algorithm53. Two different isotropic schemes were used to maintain the pressure at 1 atm: Berendsen50 with a time constant of 1 ps and Parrinello–Rahman55,56 with a time constant of 2 ps. All production simulations used the former. We used a hydrogen mass repartitioning factor of 2.5, which allowed an integration time step of 4 fs for all production simulations. Other molecular dynamics parameters are detailed in the GROMACS input files provided as Supporting Information.

Our in-house pipeline operates as follows (Supplementary Fig. 1): During the ‘Build Simulation System’ phase, the user provides configuration details for the protein–ligand complex, and topologies are generated accordingly. Two neutral solvated systems with 150 mM of NaCl are created for the ligand and protein–ligand complex in an octahedron box with 1.5 nm distance between the solute and the edges’ box with the GROMACS’ solvate module. The ‘Equilibration Setup’ and ‘Equilibration Run’ steps generate molecular dynamics parameters and conduct the corresponding equilibration simulations. For the protein–ligand complex, the process begins with minimization using the steepest-descent algorithm. This is followed by a 1 ns NVT (constant particle number, volume and temperature) phase with a 2 fs integration time step and position restraints on the heavy atoms using a force constant of 2,500 kJ mol−1 nm−2. Next, a 1.05 ns NVT phase and approximately 1 ns NPT (constant particle number, pressure and temperature) phase are conducted, both with a 3 fs integration time step and the same position restraints. The previous NPT phase uses the Berendsen scheme as detailed above. Subsequently, a 5 ns NPT phase with the Parrinello–Rahman scheme with a 4 fs integration time step is performed without restraints. This is followed by a final step of 10 ns under the same conditions. For the ligand alone, the same procedure is followed, except the initial 1 ns NVT phase with a 2 fs integration time step is omitted and the final NPT simulation is performed during 5 ns. The final 10 ns of the of the protein–ligand complex simulation was used to estimate the optimal Boresch restraints for the decoupling phase of the protein–ligand complex simulations.

Similar to the previous two steps of the workflow, ‘FEP Setup’ and ‘FEP Run’ prepare and execute the FEP simulations needed to complete the thermodynamic cycle detailed in the ‘Thermodynamic cycle for FEP’ section. Each window for both the protein–ligand complex and the ligand alone begins with minimization using the steepest-descent algorithm. This is followed by a 10 ps NVT phase with a 2 fs integration time step and position restraints on the heavy atoms using a force constant of 2,500 kJ mol−1 nm−2. Next, a 100 ps NPT phase is conducted with a 4 fs integration time step, the same position restraints and the Berendsen scheme as detailed above. Subsequently, a 500 ps NPT phase with the Parrinello–Rahman scheme and a 4 fs integration time step is performed without restraints. This is followed by a final step of 10 ns under the same conditions.

The free energy contributions of each step are computed using either the multistate Bennett acceptance ratio69 or thermodynamic integration estimators during the ‘Get Contribution’ step. The Python package alchemlyb-2.0.0 (ref. 70) was used for this purpose. Finally, all results are aggregated in the ‘Get Cycle’s ∆G’ step.

Thermodynamic cycle for FEP

The thermodynamic cycle involved decoupling the Coulomb interactions of the ligand in water over 11 λ points, followed by the decoupling of van der Waals interactions over 21 λ points with a soft-core potential activated to prevent numerical instability. Boresch restraints71, chosen from the last 10 ns of the protein–ligand complex free simulation during the equilibration phase (‘ABFE calculation’ section), had their free energy contribution analytically calculated.

Both the selection and energy contribution estimation of Boresch restraints were conducted using the software MDRestraintsGenerator (version 0.2.0)72. The selected restraints were activated for the ligand in complex with the protein, and the van der Waals interactions of the ligand were reactivated in the protein complex over 21 λ points with a soft-core potential to avoid numerical instability. Subsequently, Coulomb interactions were activated over 11 λ points to finally remove the restraints over 12 λ points. The binding free energy is calculated from the contributions of all previously mentioned steps.

Clustering and identification of the most favourable energetic pose

Frames selected by MDRestraintsGenerator and used as input structures for the FEP simulations were clustered based on the protein–ligand interaction fingerprint calculated with ProLIF (version 2.0.3)73. Each bit of the fingerprint represents a pair of atom/atom groups from the protein and ligand involved in a specific class of interaction as defined by ProLIF. This unambiguous definition allows for the separation of potential poses that may be symmetrical. The final fingerprint consists of 956 bits. The similarity among frames is calculated using the Tanimoto metric implemented in RDKit (version 2023.03.2)74 on the constructed protein–ligand interaction fingerprint, resulting in the generation of a similarity matrix.

The similarity matrix was then subjected to the hierarchical clustering algorithm in SciPy using Ward’s variance minimization algorithm75. After constructing the dendrogram, the number of clusters was determined through visual inspection.

ABFE between the most energetic favourable pose and the individual monomers

To investigate whether tecovirimat can bind to the monomer, the ABFE for each individual monomer was calculated for the identified most energetic favourable pose within the dimer. The same methodology previously described was used, with the only difference being that a single monomer was used instead of the dimer.

Calculation of average binding free energies

To estimate the average binding free energy for the dimer and monomer complexes with tecovirimat, we averaged over all independent simulations (N = 45 for the tecovirimat–dimer, N = 6 for the tecovirimat–monomer binding) as shown in equation (1), implying that we average with respect to the binding probabilities (rather than with respect to the binding free energies). Thereby, the average is dominated by the high-affinity binding poses. Here, β is the inverse temperature, and 〈·〉 denotes the average over independent simulations.

$$ _}=-^}\langle ^_},}}\rangle$$

(1)

The 95% confidence interval was estimated from 1,000 rounds of bootstrapping. In each round, N \(}}_}}}_\) samples were drawn with replacement from our N \(}}_}}}_\) values and averaged according to equation (1). After removing the largest and smallest 2.5% of the 1,000 bootstrapped averages, 95% confidence intervals were obtained from the upper and lower bounds of the remaining 950 averages.

Validation of ABFE calculation via ABFE calculations for seven additional ligands

To provide additional evidence for the tecovirimat binding pose and to validate the ABFE calculations, we carried out additional ABFE calculations with seven structurally similar ligands with available \(}}_^}}\) values alongside tecovirimat22,23. Here \(}}_^}}\) denotes the effective concentration that inhibits 50% of virus-induced cytopathic effects on VACVs. Each ligand was aligned to the tecovirimat pose reported here, and ABFE calculations were conducted using three independent replicates for each of the eight ligands, including tecovirimat. The calculated binding affinity \(\Delta _}}^\) reported for each ligand represents the mean across the three replicates, and the error was taken as the standard error of the mean (s.e.m.).

We assume that the \(}}_^}}\) value is related to the change in free energy upon two reactions, dimerization and ligand binding:

$$\begin2}\leftrightharpoons }}^}}}};\Delta _}}^\\ }}^}}}}+}\leftrightharpoons }}^}}}};\Delta _}}^\\ 2}+}\leftrightharpoons }}^}}}};\Delta _}}^+\Delta _}}^\end$$

(2)

where P denotes the protomer (a single monomer), L the ligand, \(}}^}}}}\) the homodimer observed in the crystal, and \(}}^}}}}\) is the ternary complex. \(\Delta _}}^\) denotes the free energy for dimerization of two protomers towards the dimeric crystal structure, and \(\Delta _}}^\) denotes the free energy for ligand binding to the crystallographic homodimer. For the overall reaction, the fraction of protein in ternary complex is

$$\theta =\frac_}}}_}}+_}}}$$

(3)

Here, ai denotes the activity of species i defined as \(_\equiv _/^\), where C° is the standard concentration of 1 mol l−1. The dissociation constant for equation (2) is:

$$_=\frac^}_}}_}}}_}}}$$

(4)

Equations (3) and (4) yield

Let \(^_}\) denote the ligand activity at which 50% of the protein is in complex (θ = 0.5). Here ‘*’ is used to distinguish the symbol aL, the activity of the ligand at any θ value. Thus, we have:

Assuming that \(^_}\) is proportional to \(}}_^}}\) among the eight ligands, we have \(^_}=\gamma }}_^}}/C^\circ\), where γ is an unknown constant. Thus,

$$\Delta _}}^+\Delta _}}=\,}\left(_\right)=\,}\left(\frac}}_^}}}^}\right)+\,}\left(_}}\right),$$

(7)

where R is the gas constant and T the temperature. Furthermore, we assume that the activity of the protomer aP on the cell is constant.

In our ABFE calculations, we evaluated only the second step of the two reactions of equation (2):

$$}}^}-}}+}\leftrightharpoons }}^}-}};\Delta _}}^$$

(8)

Thus, \(\Delta }_}}^\) is offset from \(}\left(\frac}}_^}}}}^}\right)\) by two constant contributions:

$$_}}=-\Delta _}}^+\,}\left(_}}\right)$$

(9)

\(_}}\) accounts for (1) the free energy cost of F13 dimerization and (2) our assumption that the intracellular ligand activity is proportional (but not equal) to extracellular ligand concentration in experiment. By comparison of our calculated AFBEs with the experimental \(}}_^}}\) values, we estimated \(_}}\approx \left\langle \Delta _}}^-\Delta _^\right\rangle\) = −15.2 ± 0.4 kcal mol−1, where 〈·〉 denotes the average over the eight ligands. The error for the quantity \(}\left(\scriptstyle\frac}}_^}}}}^}\right)+_}}\) was estimated by error propagation using the uncertainties Python library76.

Supplementary Fig. 1 correlates \(\Delta _}}^\) with \(\mathrm\left(\scriptstyle\frac}}_^}}}^}\right)\), after correcting for \(_}}\). The reasonable agreement (1) validates the ABFE protocol and (2) suggests that the crystallographic pose of tecovirimat is adopted by the other seven ligands considered in this analysis.

Mass photometry

Mass photometry experiments were done using TwoMP instrument (Refeyn) using filtered (0.22 μm) ‘protein buffer’ (10 mM Tris–HCl pH 8, 100 mM NaCl) to avoid contaminations which would increase the background signal. Contrast-to-mass calibrations were achieved by measuring the contrast of two references (bovine serum albumin (BSA) and urease, both purchased from Sigma Aldrich) diluted in protein buffer, covering mass range from 66 kDa to 272 kDa. Four contrast values were used to generate a standard calibration curve, with the following rounded average masses: 66, 132, 198 and 272 kDa. We performed the experiment using microscope coverslips (24 × 50 mm and 170 ± 5 μm thick) cleaned with isopropanol and Milli-Q water followed by drying with air. Samples were loaded into dried coverslip surface assembled into silicone gaskets. Immediately before mass photometry measurements, 2 μl of sF13 protein stocks, with increasing amounts of tecovirimat or IMCBH, was diluted in 18 μl of ‘protein buffer’ into the gasket hole and mixed twice. In all cases, the final concentration of sF13 was 25 nM. Tecovirimat/IMCBH were at different concentrations between 10 μM and 1 nM. Data acquisition was performed using AcquireMP v2.3 (Refeyn), and movies of 2,936 frames were recorded at 49 Hz framerate, adjusted to maximize camera counts while avoiding saturation. Mass photometry images were processed and analysed using DiscoverMP v2.3 (Refeyn).

Mean contrast values from the BSA and urease calibration were plotted and fitted to a linear function y = bx, where y is the contrast, x is the mass and b is the contrast-to-mass calibration factor. To extract mole fractions (percentage of each species), we converted all particle contrasts obtained from each movie to mass, applied a Gaussian fitting and calculated mole fractions as the area of each Gaussian curve. Finally, sF13 dimer percentage values were plotted against tecovirimat/IMCBH concentration using Prism Graphpad v9.0.2, and EC50 values were extracted using a nonlinear fit function (Extended Data Fig. 6).

AUC

Sedimentation velocity experiments were carried out at 20 °C in an Optima AUC analytical ultracentrifuge (Beckman Coulter) equipped with double-UV and Rayleigh interference detection. Purified sF13 proteins at 0.4 mg ml−1 in the presence or absence of tecovirimat (10 μM) were centrifuged at 42,000 r.p.m. (23,600 g) using an AN60-Ti rotor and 12 mm thick double sector centrepieces. Absorbance and interference profiles were recorded every 5 min. Buffer viscosity (η = 1.016 cP) and density (ρ = 1.0054 g ml−1) at 20 °C were estimated with SEDNTERP 1.09. Partial specific volumes at 20 °C were estimated based on amino acid sequences using SEDNTERP 1.09 software. Data were analysed with SEDFIT 16.1 (ref. 77) using a continuous size distribution c(S) model. Theoretical sedimentations of the complex were generated using hydropro 10 (ref. 78).

SAXS experiments

SAXS data were collected on the SWING beamline at Synchrotron Soleil (France) using the online HPLC system. These experiments have been performed using sF13WT digested with thrombin. sF13WT samples at 4.6 mg ml−1 were prepared in a buffer containing 10 mM Tris pH 8, 100 mM NaCl and 10 μM tecovirimat and injected into a size exclusion column (Superdex 75 increase 5/150 mm) cooled at 15 °C eluting directly into the SAXS flow-through capillary cell at a flow rate of 200 µl min−1. The data were analysed using FOXTROT and PRIMUS from ATSAS 3.2 (ref. 79), from which Guinier was generated. Scattering curves were selected for stable Rg at the apex of the elution profile, the selected curves were averaged, and buffer signal was subtracted. From these corrected scattering curves, the pair distribution function was computed using GNOM (version 5.0)80, and the normalized Kratky plot was generated. Using the structure of sF13WT (PDB 9FHS), the experimental curve was compared to theoretical curve using CRYSOL (version 2.8.3)81. Ab initio models were generated with DAMMIN (version 5.3)82, and for each model, sedimentation characteristic was calculated with hydropro (version 10)78. The SAXS statistics are provided in Supplementary Tables 6 and 7.

F13 transfection for PLA and immunofluorescence staining

To perform the PLA experiment and immunofluorescence staining of F13, HeLa cells (ATTC CCL-2) were transfected with pcDNA 3.1 plasmids coding for either F13WT or F134MUT (N267D, A288P, A290V, D294V), with an internal FLAG tag sequence (GGGDYKDDDDKGGG) inserted within residues D21 and N22. The use of an internal FLAG tag in F13 was necessary, as the N-terminus is buried into the membrane and the C-terminus is part of the dimeric interface. Thus, none of them were suitable for standard N- or C-terminal protein tagging. We selected the best region to insert the FLAG tag based on the sF13 dimer structure reported here. For this, we selected an exposed loop, away from the membrane interaction region and the dimerization interface.

For PLA and immunofluorescence, 1.2 × 104 HeLa cells per well were transfected in suspension using lipofectamine 2000 (Thermo Fisher Scientific) in a 96-well plate (μClear, Greiner Bio-One 655090) with 100 ng of DNA. In each well 50 μl of HeLa cells at 2.4 × 105 cells ml−1 were mixed with 50 μl of transfection mix and 50 μl of DMSO or DMSO/tecovirimat, resulting in tecovirimat at a final concentration of 10 μM and DMSO at 0.1%. Cells were incubated 24 h at 37 °C and 5% CO2; subsequently the cells were fixed with 4% PFA for 10 min.

PLA

PLA was performed using the Duolink PLA Fluoresence kit (DUO92008, Merck). In short, cells were permeabilized at room temperature for 3 min in PBS with Triton 0.1% and washed with PBS. About 40 μl of Duolink blocking solution was added to each well, and the plate was incubated at 37 °C for 1 h. After blocking, cells were incubated at room temperature for 45 min with primary monoclonal mouse M2 (dilution = 1:350, F3165, Sigma-Aldrich) and rabbit D6W5B (dilution = 1:500, 14793, Cell Signaling Technology) anti-FLAG antibodies (diluted in Duolink Antibody Diluent) at a final concentration of 285 ng ml−1. The wells were washed twice for 5 min at room temperature with buffer A (10 mM Tris pH = 7.4, 150 nM NaCl, 0.05% Tween). About 40 μl of PLA probe mix, containing PLA probe PLUS (anti-rabbit, dilution = 1:5, DUO92002, Merck) and PLA probe MINUS (anti-mouse, dilution = 1:5, DUO92004, Merck) was added to the wells following the manufacturer’s instructions. The plate was then incubated at 37 °C for 1 h. After incubation, the wells were washed twice for 5 min with buffer A, then 40 μl of ligase mix was added to the samples following the manufacturer’s instructions; these were incubated 30 min at 37 °C. Wells were washed twice for 5 min with buffer A at room temperature; subsequently, 40 μl of amplification mix (containing a polymerase) were added to each well, following the manufacturer’s instructions. The plate was then incubated for 100 min at 37 °C. Finally, PLA wells were washed once with buffer B (200 mM Tris, pH = 7.5, 100 mM NaCl) for 10 min at room temperature and a second time for 10 min at room temperature with buffer B supplemented with 1 μg ml−1 Hoechst 33342 nuclear staining (Invitrogen). Subsequently, a final wash was performed with 0.01× buffer B for 1 min at room temperature, and cells were then left in fresh PBS. The plates were imaged using an Opera Phenix Plus microscope (Revvity) at ×20. Forty-nine images per well, covering over 90% of the well, were acquired.

Immunofluorescence staining of F13

Immunofluorescence staining of F13 with rabbit and mouse anti-FLAG antibodies was performed in parallel with PLA, in the same plate. Cells were permeabilized at room temperature for 3 min in PBS with Triton 0.1% and washed with PBS. About 40 μl of Duolink blocking solution was added to each well, and the plate was incubated at 37 °C for 1 h. Cells were incubated at room temperature for 45 min with primary monoclonal mouse M2 (dilution = 1:350, F3165, Sigma-Aldrich) and rabbit D6W5B (dilution = 1:500, 14793, Cell Signaling Technology) anti-FLAG antibodies (diluted in Duolink Antibody Diluent) at a final concentration of 285 ng ml−1. The wells were washed twice for 5 min at room temperature with buffer A. Wells were washed twice with PBS for 5 min at room temperature, and Alexa FluorTM 488 goat anti-mouse antibody (dilution = 1:500, A-11001, Invitrogen) and Alexa FluorTM 488 goat anti-rabbit antibody (dilution = 1:500, A-11008, Invitrogen) (diluted in PBS, BSA 1%, Na Azide 0.1%) were added to the respective wells at a final concentration of 4 μg ml−1. The plate was then incubated at 37 °C for 1 h. Wells were washed once for 5 min at room temperature with PBS and once with PBS supplemented by 1 μg ml−1 Hoechst 33342 nuclear staining (Invitrogen). Finally, cells were left in fresh PBS before imaging. The plates were imaged using an Opera Phenix Plus microscope (Revvity) at ×10. Twenty-one images per well, covering over 90% of the well, were acquired.

Viral plaque assay and analysis

Six-well plates were seeded with BSC40 (ATTC CRL-2761) cells 24 h before infection. Confluent BSC40 cells were infected with wild-type VACV-WR25 (provided by J.M. (University of Birmingham)) or rVACV mutants generated from a modified VACV-WR (vNotI/tk) strain (originally provided by B. Moss to K.C., who further modified it by incorporating an mCherry reporter into the tk locus) at a 10-fold dilution in DMEM with 2.5% FBS for 1 h at 37 °C. The infection medium was then removed, and a 0.5% methylcellulose in DMEM media overlay containing 10 µM Tecovirimat was added for 3 days at 37 °C. Afterward, the overlay medium was removed, and the wells were fixed and stained with 1% crystal violet in 20% methanol for 20 min. The crystal violet was removed, wells were washed with PBS, and plates were imaged using a Cytation 7. Plaque counts and diameters were measured to determine titres (plaque-forming units (p.f.u. ml−1)) and plaque sizes (µm) using a program dev