Protein expression and purification
VACV D13 was prepared as previously described by Hyun et al.9. Briefly, the D13 gene in the pPROEX-Hta vector was expressed in BL21(DE3) cells and purified by immobilized metal affinity chromatography (IMAC) using HisTrapHP (Cytiva). The N-terminal His6-tag was removed by TEV protease digestion and products were further purified by IMAC to remove remaining tagged protein. TEV protease digestion was omitted for production of His6-tagged protein used in D13 self-assembly experiments. The protein was further purified by size exclusion chromatography in a buffer containing 50 mM Tris-HCl (pH 8.0), 600 mM NaCl, 50 mM L-arginine, 50 mM L-glutamic acid and 2 mM β-mercaptoethanol.
Mutagenesis
The N-terminal truncation mutant (D1318-548) was generated by PCR using a forward primer lacking the leading 51 bases of the D13 DNA sequence. Point mutations R353A and Y62A were introduced to wtD13 and D1318-548 using a Q5 site-directed mutagenesis kit (New England Biolabs Inc.). Expression and purification procedures for the mutant were identical for those of wtD13. Primers used for mutagenesis are listed in Supplementary Information.
D13 self-assembly
A 100-µL aliquot of purified protein at approximately 2 mg/mL was dialyzed in buffer containing 10 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 2 mM β-mercaptoethanol, at 4°C for at least 12 h. The resulting turbid solution was centrifuged at 15,000 rpm. Then 80 µL supernatant were removed and the pellet was resuspended using the remaining solution. The pellet sample was negatively stained with 2% uranyl acetate and examined using TEM to confirm the assembly. The remaining supernatant was checked by TEM to confirm presence of homogeneous trimers. This preparation was used for cryo-EM of singlet trimer particles.
Negative staining and transmission electron microscopy
Routine TEM examinations were performed using negatively stained samples. 5 µL of protein at approximately 0.01 mg/mL were loaded onto freshly glow-discharged EM grids with continuous carbon support films. 90 sec were allowed for sample adsorption and then grids were washed with three droplets of distilled water. 5 µL of 2% uranyl acetate solution were loaded onto grids, followed by 60-sec incubation. A piece of filter paper was used to blot excess stain solution and grids were air-dried. Specimen grids were examined using a Talos L120C TEM equipped with Ceta CMOS detector (Thermo Fisher Scientific (TFS)), operating at 120 kV acceleration voltage.
Cryo-electron microscopy
For wtD13 trimers, His6-tagged D13 trimers and D1318-548 trimers, 3 µL of protein at 0.12 mg/ml were loaded onto holey EM grids (Quantifoil R1.2/1.3 Cu300, Quantifoil Micro Tools GmbH) prior treated with graphene oxide film flakes (Sigma Aldrich). The sample was vitrified on a Vitrobot Mark IV (TFS) operating at 4°C and >90% relative humidity. wtD13 image data were acquired with a Titan Krios TEM (TFS) operating at 300 kV at a nominal magnification of 155,000x, corresponding to 0.518 Å/pixel at the specimen level, with the defocus ranging between 0.5 µm and 1.5 µm. Spot size 8, C2 aperture 70 µm, OL aperture 100 µm and 1.8 µm illumination area were used. 50 movie fractions were collected on a Falcon 3EC direct electron detector in electron counting mode using EPU software (TFS), with a total electron dose of 50 e-/Å2, and dose rate of 1.6 e-/Å2/sec. His6-tagged D13 image data were collected with a Titan Krios TEM (TFS) operating at 300 kV at a nominal magnification of 155,000x, corresponding to 0.518 Å/pixel at the specimen level, with the defocus ranging between 0.5 µm and 1.5 µm. Spot size 8, C2 aperture 70 µm, OL aperture 100 µm and 1.8 µm illumination area were used. 50 movie fractions were collected on a Falcon 3EC direct electron detector in electron-counting mode using EPU software (TFS), with a total electron dose of 50 e-/Å2, and dose rate of 1.6 e-/Å2/sec. D1318-548 data were collected on a Talos Arctica TEM (TFS) operating at 200kV at a nominal magnification of 92,000x, which corresponds to 1.12 Å/pixel at the specimen level, with defocus ranging between 0.6 µm and 1.2 µm. The spot size was 9. A 70-µm C2 aperture and 100-µm OL aperture were used. Exposures were acquired as movies of 50 dose fractions on a Falcon 3 direct electron detector in EC mode using EPU software (TFS), with a total electron dose of 50 e-/Å2 at a dose rate of 0.6 e-/Å2/sec.
For D13 doublets, multiple datasets were collected in order to enrich the angular orientation of particle images. The first two datasets were collected from a grid prepared as for the D13 trimer singlet, using holey carbon grids with additional graphene oxide film. For the third and fourth datasets, samples were prepared on gold grids (UltrAuFoil R1.2/1.3 300, Quantifoil Micro Tools GmbH) without a graphene oxide support film. Vitrification methods were the same for all grid preparations. For the third and fourth datasets, in which D13 trimers were preferentially oriented and partially assembled, the microscope stage was tilted to 30° and 45°, respectively. Image data were collected on a Talos Arctica TEM (TFS) operating at 200kV at a nominal magnification of 92,000x, which corresponds to 1.12 Å/pixel at the specimen level, with defocus ranging between 0.5 µm and 5.0 µm. The spot size was 9 or 10. A 50- or 70-µm C2 aperture and 100-µm OL aperture were used. Exposures were acquired as movies of 50 dose fractions on a Falcon 3 direct electron detector in electron-counting mode using EPU software (TFS), with a total electron dose of 50 e-/Å2 at a dose rate between 0.6-0.9 e-/Å2/sec.
For the tubular D13 assembly, 3 µL of resuspended pellet from the assembly solution were loaded onto a holey EM grid (Quantifoil R2/2 Cu300). Vitrification methods were the same as above. Data were acquired on a Titan Krios TEM (TFS) operating at 300kV in EFTEM mode (Gatan Quantum 968) at a nominal magnification of 105,000x corresponding to 1.4 Å/pixel at the specimen level, with defocus ranging between 0.5 and 2.5 µm. Spot size 7, a 70-µm C2 aperture and 100-µm OL aperture were used. 50 dose fractions/movie were recorded with a K2 Summit direct electron detector (Gatan) and EPU software (TFS), with a total electron dose of 50 e-/Å2 at a dose rate of 5 e-/Å2/sec.
For cryo-ET of spherical IV-like particles, assembly solutions that resulted from introducing His6-tagged D13 into low-salt buffer were mixed with 10-nm gold fiducial marker (AURION) in a 4:1 (v:v) ratio. 3 µL of the mixture were loaded onto a non-glow-discharged holey EM grid (Quantifoil R2/2 Cu300) and vitrified on a Vitrobot Mark IV (TFS) operating at 4°C and >90% relative humidity. Tomographic data were acquired on a Titan Krios TEM (TFS) operating at 300kV at a nominal magnification of 45,000x corresponding to 2.26 Å/pixel at the specimen level, defocused between 5.0 and 6.0 µm. The spot size was 9, and a 70-µm CL aperture, with a 100-µm OL aperture and a 1.8-µm beam diameter area were used. A dose-symmetric tomography acquisition scheme was applied on a Falcon 3 direct electron detector in EC mode using Tomo 4 software (TFS) over +/- 63° tilt with 3° intervals. The total dose applied to each tomogram was approximately 120 e-/Å2, and 5 movie fractions were collected for each tilt image at a dose rate 0.14 e-/Å2/sec.
Cryo-EM Image processing
All single-particle datasets were processed with Relion 3.1 software, unless stated otherwise30-33. Motion correction and contrast transfer function (CTF) estimation were performed using MotionCor234 and CTFFIND435, respectively. Falcon 4 movie data in EER format from the Krios G4 were processed with Relion 3.1.2, Statistics of individual datasets can be found in Supplementary Table 1. Cryo-ET data were processed using IMOD. Visual examination of maps and figure preparations were done with UCSF Chimera36, Chimera X37 and PyMol (The PyMol Molecular Graphic System, Version 2.4.0, Schrödinger, LLC).
For wtD13 trimers, 2,168 dose-weighed micrographs were manually inspected. Images without graphene oxide support or poor estimated maximum resolution or strong astigmatism from CTFFIND calculation were discarded. From 1,862 assorted micrographs, particles were automatically picked using 2D class averages of manually picked particles as a template. 668,437 particles were down-sampled by 4 and extracted into 128 pixel squared boxes. The particles were then subjected to two rounds of 2D classification to eliminate poorly aligned particle images (Supplementary Fig. 2a). The resulting 204,563 particles were re-extracted from micrographs with a down-sampling factor of 2, into 256 pixel squares boxes. A consensus reconstruction was generated by refinement without imposition of symmetry. Then, 3D classification was performed without particle alignment, from which 130,384 particles that belong in the class exhibiting the best structural details were selected for the final round of refinement (Supplementary Fig. 2b). Further refinements on the assorted particles were performed with C3 symmetry imposition. CTF refinement for magnification anisotropy, optical aberrations and per-particle defocus were performed, followed by Bayesian particle polishing. Final 3D refinement and map sharpening resulted in a reconstruction at 2.25 Å spatial resolution based on 0.143 Fourier shell correlation (FSC) criterion of independently refined halfset reconstructions (Supplementary Fig. 2c-e). The particles that had been selected from the 3D classification were also subjected to 3D refinement without symmetry imposition, CTF refinement and Bayesian polishing. The final resolution of the unsymmetrized 3D reconstruction was estimated at 2.63 Å resolution (FSC=0.143).
For His6-tagged D13 trimers, 1,328 dose-weighed micrographs were selected from 1,418 total micrographs after removing micrographs with poor estimated maximum resolution or strong astigmatism. 284,340 particles were semi-automatically picked, down-sampled by 4, and boxed in 128-pixel squared boxes. Two rounds of 2D classifications were performed to remove poorly aligned images. The resulting 216,590 particles were re-extracted from micrographs with a down-sampling factor of 2 into 256-pixel squared boxes, resulting in a sampling interval of 1.036 Å/pixel. 3D classification was performed without symmetry imposition and only particles that belong in classes with detailed structural features were selected. The resulting 173,354 particles were subjected to 3 rounds of 3D refinement with C3 symmetry. CTF refinement and particle polishing were performed between refinements. The final resolution of the 3D reconstruction was estimated at 2.63 Å (FSC=0.143).
For D1318-548 trimers, 1,112 dose-weighed micrographs out of 1,227 total micrographs were selected after removing micrographs with poor estimated maximum resolution or strong astigmatism. 743,834 particles were semi-automatically picked, binned by 2, and boxed in 128-pixel squared boxes. A 2D classification was performed to remove poorly aligned images, and 475,652 good particles were re-extracted from unbinned micrographs (1.12 Å/pixel), into 256-pixel squares boxes. 3D classification was performed without symmetry imposition and 156,813 particles that belong in classes with detailed structural features were selected. 3 rounds of 3D refinement with C3 symmetry were performed, with CTF refinement and particle polishing between. The final resolution of the 3D reconstruction was estimated at 4.10 Å resolution (FSC=0.143).
For D13 trimer doublets, four datasets were used (550, 280, 676 and 189 movies each), in which each movie dataset was independently processed for beam-induced motion correction, CTF estimation, and particle auto-picking. Particle auto-picking was performed using a template 3D map produced by artificially joining two copies of D13 trimer maps. To minimize the effect of reference bias on the high-resolution signal imposed by particle picking, the template was low-pass filtered to 20 Å. 835,797 particles from combined 1,695 micrographs were down-sampled by a factor of 2 and extracted into 128-pixel squared boxes. Particle images were subjected to 2D class averaging, from which images that belong in classes with poor structural features were eliminated (Supplementary Fig. 3a). The resulting 746,600 particles were subjected to 3D classification without symmetry imposition, using an initial 3D reference generated from image data. The majority of particles partitioned into 3D classes with a pronounced trimer singlet featuring an adjacent ghost-like density of an extra trimer (Supplementary Fig. 3b). Only 164,259 particles from a 3D class that clearly exhibited the trimer doublet were selected for further processing. Particle images were re-extracted from micrographs into 256-pixel squared boxes without down-sampling. Because of clear 2-fold symmetry of the 3D class, C2 symmetry was imposed in subsequent steps. The initial 3D refinement resulted in a 3D reconstruction at 4.78 Å at 0.143 FSC. Then, CTF refinement for magnification anisotropy, optical aberrations, and per-particle defocus was performed on each optical group, followed by Bayesian particle polishing. These procedures were repeated three times. Next, the aligned particles were 3D-classified without further image alignment, from which 42,854 particles that belong in classes with detailed structural features were selected (Supplementary Fig. 3b). These particles were re-extracted into 512-pixel squared boxes to include the delocalized signal from highly defocused images (up to 5 mm underfocus). 3D refinement, CTF refinement, and particle polishing were repeated twice. The resolution of the final 3D reconstruction was 3.87 Å at 0.143 FSC (Supplementary Fig. 3d). 3D resolution anisotropy of the reconstruction was calculated using the 3DFSC server (https://3dfsc.salk.edu/)38. The resulting estimate indicates global resolution at 4.55 Å at 0.143 FSC and sphericity of 0.769 (Supplementary Fig. 3e).
For the D13 tubular assembly, 7,621 movies were motion-corrected followed by CTF estimation. Micrographs with poor estimated maximum resolution and severe astigmatism were discarded, based on the CTFFIND4 calculation. From 7,529 micrographs, start and end coordinates of the tubes were manually selected, and helical segments with a 167.5-Å inter-box distance (5 ´ the estimated helical rise of 33.5 Å) were down-sampled by 4 and extracted into 256-pixel squared boxes. 194,960 particle images were subjected to 2D classification, from which 129,652 images belonging to class averages with good structural details were selected. Helical parameters were estimated by manual analysis of the helical layer-line pattern from the 2D class averages (Supplementary Fig. 4d). The images were classified in 3D, using an artificial canonical helix reference consisting of spheres generated from helical parameters (77° twist, 33.5 Å rise with the command relion_helix_toolbox --simulate helix --o ref.mrc --subunit_diameter 140 --cyl_outer_diameter 900 --angpix 5.6 --rise 33.5 --twist 77 --boxdim 256). 75,070 particle images belonging to the 3D class with the best structural features were re-extracted into 512-pixel squared boxes after down-sampling by a factor of 2 (Supplementary Fig. 4b). After 3D refinement, CTF refinement and Bayesian particle polishing were performed, followed by reconstruction with Ewald sphere correction31,39. The final map resolution was 7.3 Å at 0.143 FSC (Supplementary Fig. 4c). Particles were re-extracted into 1024-pixel squared boxes and a 3D reconstruction was generated prior to symmetry expansion and signal subtraction. From the helical reconstruction, 5 asymmetric units were symmetry-expanded based on the final estimated helical operator (76.98° twist, 33.86 Å rise). For signal subtraction, a volume segment that corresponds to a concentric ring of six trimers in the tube were generated from the reconstruction using UCSF Chimera. The map was then used to create a binary mask embracing six trimers, which was applied to the symmetry-expanded stack of particle images. Signal-subtracted sextet images were subjected to 2D class averaging using cisTEM40, and 303,052 particles from good class averages were selected for auto-refinement in 3 classes (Supplementary Fig. 4e-f). 247,311 particles from 3D classes with detailed structural features were selected for final refinement in cisTEM. The resulting half-maps were examined and post-processed using Relion 3.1. Final resolution of the map was 3.9 Å at 0.143 FSC (Supplementary Fig. 4g-h).
For Cryo-ET data, movie frames were aligned and summed using MotionCor2. Images were phase-flipped using ctfphaseflip41 and the tilt series was aligned based on gold fiducial markers. Tomograms were reconstructed from images that were binned by 2, using simultaneous iterative reconstruction technique (SIRT) in IMOD42.
Protein structure modeling
3D reconstructions were subjected to density modification with the ResolveCryoEM tool43 in the Phenix software suite prior to model refinement. All coordinate refinements were performed using the real-space refinement routine in Phenix44. The X-ray crystal structure of D13 (PDB ID 6BEI) was manually fitted into the cryo-EM map of our D13 trimer in UCSF Chimera, and used as a reference model for atomic coordinate refinement in Phenix, while enforcing a non-crystallographic symmetry constraint. The resulting refined trimer model was then used as a reference to refine models of the trimer doublet and sextet. These refined models were manually inspected and adjusted in Coot45, followed by final refinement with Phenix. The quality of the final models and map-to-model correlations were calculated using Phenix' Cryo-EM validation and Mtriage tools46.