Computational protocol: Long-range allosteric regulation of the human 26S proteasome by 20S proteasome-targeting cancer drugs

Similar protocols

Protocol publication

[…] Individual image frames were aligned and weighted according to electron dose using the software unblur to reduce the effects of drift and charging. The CTF of the remaining micrographs was determined using Gctf. Particles were selected in a template-free manner, using image statistical properties in combination with mass centring. Individual particle coordinates were additionally refined by alignment against twelve low resolution reference images representing different views of the proteasome ().Subsequently, we performed several image sorting steps to remove contaminations, blurred images and bad particles. In a first step, power spectra were calculated, for each particle and classified using a hierarchical clustering scheme. The resulting class averages were visually inspected for Thon ring appearance and particles belonging to strongly charged or blurred classes were discarded. Second, several rounds of multi-reference alignment and two-dimensional classification were performed. Particles belonging to classes that did not show clear molecule views were discarded.After having applied these image sorting procedures, the best class averages were used to generate an initial 3D model using simple PRIME. This 3D model was used as an initial reference in a 3D classification in RELION, which we used to classify the particles according to their two main conformational states. To ensure correct class assignment, all particles were aligned competitively against averaged maps obtained for the two main states. The flexible protein Rpn1 interferes with the alignments and therefore its density was masked out.Particle images belonging to the non-rotated state of the proteasome were refined by RELION auto-refine. Subsequent hierarchical sorting discriminated further sub classes of various RP conformations. Specifically, a series of 3D classification steps without alignment using increasingly smaller masks was performed in RELION. In the first classification step, we used a mask for the whole proteasome holoenzyme excluding Rpn1, in the second iteration we used a mask for the whole RP (19S) subcomplex, in the third iteration a mask for the whole lid and finally a mask for Rpn2 only. The remaining particles (233,000) were refined to a final resolution of 3.8 Å and B-factor corrected in RELION. To further improve the map, particle polishing was performed on the final particle stack in RELION.A local resolution map was calculated in ResMap by calculating local FSC values in a sphere with a diameter of 13 voxels moving over the entire 3D volume. In addition, the signal of CP was subtracted from the raw particles. These subtracted particles were centred and again refined in RELION. Masks for the Rpt2/6, Rpn9/10 and Rpn5/6 regions were created in Chimera and focussed 3D classification without alignment was performed on the computationally generated 19S particles. Resulting 3D classes were refined. [...] The initial atomic coordinate model for the 20S particle was taken from a crystal structure of the Oprozomib-inhibited human 20S complex (PDB 5LEY). Models of each RP (19S) subunit were generated with Robetta and docked as rigid bodies into the EM density map with UCSF Chimera. The six nucleotides of the ATPase subunits were placed by fitting the crystal structure of PAN (PDB 3H4M) into our density. Additional aid for regions, which had to be modelled at least partly de novo, was obtained using the secondary structure prediction server psipred.An initial rigid body refinement was performed using real space refinement in Phenix and subsequent manual modelling in coot. Next, secondary structure restraints were generated using phenix.ksdssp. All secondary structure restraints were visually inspected and additional restraints were added manually. Several iterative rounds of real space refinement in Phenix and manual modelling in coot followed, where the last Phenix refinements included ADP refinement to calculate B-factors.The present map quality does not allow to distinguish clearly between ATP and ADP in the ATPase and hence we modelled all nucleotides as ADP. In addition, to account for local resolution differences in the EM density map, we used calculated model B-factor distributions as a guideline to define the level of structural details interpreted in the final model (). Accordingly, we analysed B factors in segments of five amino acids. Side chains were only modelled if the mean atomic B-factor per segment was smaller than 110 Å2, segments with mean B-factors between 110 and 150 Å2 were truncated to poly-alanine. Residues with mean B-factors higher than 150 Å2 were not included in the final deposited PDB model. […]

Pipeline specifications