Structure of hexameric S-layer protein from Haloferax volcanii archaea
ELECTRON MICROSCOPY
Sample
Structure of hexameric S-layer protein csg
Specimen Preparation
Sample Aggregation State
2D ARRAY
Vitrification Instrument
FEI VITROBOT MARK IV
Cryogen Name
ETHANE
Sample Vitrification Details
Vitrobot options:
Blot time 4.5 seconds,
Blot force -10,1,
Wait time 10 seconds,
Drain time 0.5 seconds
3D Reconstruction
Reconstruction Method
SINGLE PARTICLE
Number of Particles
1087798
Reported Resolution (Å)
3.46
Resolution Method
FSC 0.143 CUT-OFF
Other Details
Particles from classes with the same curvature were combined, re-extracted in 400 x 400 boxes and subjected to a focused 3D auto refinement on the cen ...
Particles from classes with the same curvature were combined, re-extracted in 400 x 400 boxes and subjected to a focused 3D auto refinement on the central 6 subunits using the scaled and lowpass filtered output from the 3D classification as a starting model. Per-particle defocus, anisotropy magnification and higher-order aberrations were refined inside RELION3.1, followed by another round of focused 3D auto refinement and Bayesian particle polishing (Zivanov et al., 2020).
Refinement Type
Symmetry Type
POINT
Point Symmetry
C6
Map-Model Fitting and Refinement
Id
1
Refinement Space
REAL
Refinement Protocol
AB INITIO MODEL
Refinement Target
Best Fit
Overall B Value
143.26
Fitting Procedure
Details
The boundaries of the six Ig-like domains, D1-D6, were predicted using HHpred (Steinegger et al., 2019) in default settings within the MPI Bioinformat ...
The boundaries of the six Ig-like domains, D1-D6, were predicted using HHpred (Steinegger et al., 2019) in default settings within the MPI Bioinformatics Toolkit (Zimmermann et al., 2018). Subsequently, structural models for these domains were built using the Robetta structure prediction server, employing the deep learning-based modelling method TrRosetta (Yang et al., 2020). The obtained structural models of domains D3-D6 resulted in an overall fit into the hexameric cryo-EM map of csg from the reconstituted sheets. D1-D2 deviated significantly from any obtained homology models, and for those domains, the carbon backbone of the csg protein was manually traced through a single subunit of the hexameric cryo-EM density using Coot (Emsley and Cowtan, 2004). Due to the edge effect of the box used in the refinement of the 3.5 angstrom map, parts of D6 displayed edge artefacts. These artefacts were removed using single-particle cryo-EM refinement in a larger box, which led to an overall slightly lower resolution (3.8 angstrom) but allowed fitting of the D6 homology model unambiguously. Following initial manual building (for D1-D2) or fitting in of structural models (for D3-D6), side chains were assigned in regions with density corresponding to characteristic aromatic residues allowing us to deduce the register of the amino acid sequence in the map. Another important check of the model building was the position of known glycan positions, which were readily assigned based on large unexplained densities on characteristic asparagine residues. The atomic model was then placed into the hexameric map in six copies and subjected to several rounds of refinement using refmac5 (Murshudov et al., 2011) inside the CCP-EM software suite (Burnley et al., 2017) and PHENIX (Liebschner et al., 2019), followed by manually rebuilding in Coot (Emsley and Cowtan, 2004). Model validation was performed in PHENIX and CCP-EM.
RELION refinement with in-built CTF correction. The function is similar to a Wiener filter, so amplitude correction included.
10558369
Top and tilted views were manually picked at the central hexameric axis. Manually picked particles were extracted in 4x downsampled 100 x 100 boxes and classified using reference-free 2D classification inside RELION3.1 (Zivanov et al., 2020). Class averages centered at a hexameric axis were used to automatically pick particles inside RELION3.1. Automatically picked particles were extracted in 4x downsampled 100x100 pixel boxes and classified using reference-free 2D classification. Particle coordinates belonging to class averages centered at the hexameric axis were used to train TOPAZ (Bepler et al., 2019) in 5x downsampled micrographs with the neural network architecture ResNet8. For the final reconstruction, particles were picked using TOPAZ and the previously trained neural network above. Additionally, top and bottom views were picked using the reference-based autopicker inside RELION3.1, which were not readily identified by TOPAZ. Particles were extracted in 4x downsampled 100 x 100 boxes and classified using reference-free 2D classification inside RELION3.1. Particles belonging to class averages centered at the hexameric axis were combined, and particles within 100 angstrom were removed to prevent duplication after alignment.
