Extended Data Fig. 6: Cryo-EM image processing workflow and strategy.

Initial data collection by Arctica and pre-processing indicated quality particles with density for the N-terminal module. An ab-initio volume and 2D class averages were used to bootstrap 3D classification and particle picking, respectively. A topaz³⁹ model was trained using the top 100 micrographs ranked by number of quality particles. This was subsequently used to pick roughly 0.9 million particles. Template matching from re-projected templates of the ab-initio volume was performed yielding 6.5 million particles. Multiple rounds of 2D classification of this combined set yielded 0.9 million particles with good secondary structure. Two rounds of 3D classification with incrementally higher angular sampling was performed, and a single class was selected for refinement in RELION³⁴ with SIDESPLITTER⁴⁰. Local particle motion and CTF were further refined by Bayesian polishing and CTF refinement. Two modules of P-Rex1 showed notable flexibility and were therefore separated for local refinements. The final maps were sharped using DeepEMhancer⁴² (1.0) to suppress effects of anisotropy and visualise high-resolution features.