Image processing

The RELION 3.0 pipeline³⁸ was used for image processing. Raw micrographs were first motion corrected using MotionCor2³⁹, then CTF parameters were estimated using Gctf⁴⁰. Particles were selected from a subset of micrographs by manual picking and were then differentiated by 2D classification. High quality 2D classes were used as templates for autopicking the complete data set. A low picking threshold was specified for autopicking to minimize the number of ‘missed’ VLPs. Picked particles were 2 × down-sampled, then subject to 2D classification (with CTFs ignored until the first peak), leading to the identification of FAS-containing classes. 2D classes containing FAS were taken forward for initial model generation and used as templates for FAS-specific autopicking. Autopicked FAS particles were extracted without down-sampling and subjected to two rounds of 2D classification (first with CTFs ignored until the first peak, then without). Particles in high quality classes were taken forward for 3D refinement (with D3 symmetry applied) with subsequent masking and sharpening. Following this, several cycles of CTF refinement (with per-particle astigmatism correction and beamtilt estimation), Bayesian polishing and 3D refinement (with masking and use of solvent-flattened FSCs) were performed to improve the resolution of the map. After sharpening, the resolution of the final map was determined using the ‘gold standard’ Fourier shell correlation criterion (FSC = 0.143) (Supplementary Table ^S1, Supplementary Fig ^S1). RELION was used to estimate local resolution and generate a local resolution-filtered map.

To resolve the flexible ACP, focussed 3D classification (without signal subtraction) was performed as described previously^16,41–44. Briefly, a cylindrical mask was generated in SPIDER⁴⁵ and resampled onto the D3-symmetric density map of FAS, such that it was positioned over the weak ACP density. The relion_symmetry_expand tool was used to assign 6 symmetrically-redundant orientations to each particle contributing to a symmetric reconstruction, such that the ACP domain in each symmetry-related position would be included in the classification. These symmetry-expanded particles were then subject to masked 3D classification without re-alignment using a regularisation parameter (“T” number) of 40. Given that an ACP density-containing class was resolved with a “T” number of 40, greater values of “T” number were not tested due to the heavy computational resources required for focussed classification. Particles contributing to the ACP density-containing class were taken forward for asymmetric reconstruction using the relion_reconstruct tool.