Degassed C-Flat 2/2-3C grids were glow discharged for 30 s at 20 mA. Virus or VLP solution was diluted with 10 nm colloidal gold; 2.5 μl of this mixture was applied to each grid and plunge frozen into liquid ethane using a FEI Vitrobot Mark 2. Grids were stored in liquid nitrogen until imaging.
Tomographic imaging was performed as described previously26 (link). Imaging was performed on a FEI Titan Krios at 300 keV using a Gatan Quantum 967 LS energy filter with a slit width of 20 eV and a Gatan K2xp direct detector. Tomograms were acquired from -60° to 60° with 3° steps using SerialEM27 (link) and a dose-symmetric tilt-scheme28 (link). Images were acquired in super-resolution mode.
Frames were aligned with either the K2Align software, which uses the MotionCorr algorithm29 (link), or with the frame alignment algorithm built into serialEM; frames were aligned and Fourier cropped to 4K x 4K, giving a final pixel size of 1.78 Å/pixel. Defocus for each tilt was determined by CTFFIND430 (link). Tilt images were filtered by cumulative electron dose using the exposure-dependent attenuation function and critical exposure constants described eleswhere26 (link),31 (link). Tilt images were CTF-corrected using ctfphaseflip32 (link) and tomograms were reconstructed using weighted back projection in IMOD33 (link). Tomograms with poor fiducial alignment were discarded (Extended Data Table 1). Poor fiducial alignment was defined as alignment residual above 1 pixel in 2x binned data or retaining fewer than 8 fiducial markers. CTF-corrected unbinned tomograms were binned by 2x (3.56 Å/pixel) and 4x (7.12 Å/pixel) with a Lanczos two-lobe anti-aliasing filter.
Detailed dataset parameters are given in Extended Data Table 1.