I am currently engaged in cryo-EM reconstruction of a protein-ligand complex with C4 symmetry, utilizing RELION and CryoSPARC software. The ligand-binding site is located on the C4 symmetry axis, which makes resolving a single ligand pose challenging.It’s important to note that I am performing all refinement and classification steps in C1 symmetry. Despite significant efforts in particle selection, focused refinements with different masks, and iterative masked 3D classification without alignment, I have encountered challenges in achieving clear, unambiguous density for the ligand and its surrounding residues. While our current progress is promising, I believe further refinement is necessary to ensure homogeneous and high-quality density representation without any ambiguity.
I am seeking insights, tips, or experiences from the community regarding strategies or techniques that have proven effective in similar scenarios. Any guidance would be greatly appreciated. Thank you.
Classify without alignments using symmetry expanded particles. Make sure to use enough classes to accommodate all possible orientations of the ligand. Some experimentation with the focus & solvent masks, and the exact classification parameters (force hard classification on/off, target resolution, highpass filter) may be required here.
As an alternative to 2, try symmetry relaxed refinement (in NU-refine) after refining in C4 - sometimes this can help resolve subtle pseudosymmetric features which can be otherwise difficult to pick up.
To add on to Oli’s suggestion #2 about 3D classification without alignment on symmetry expanded particles, it might be helpful to mask the ligand-binding site of just a single subunit. Afterwards, if you want to do some statistics and count how many particles have a given number of ligands, you can follow this guide that was also written by rposert.
I just want to confirm my understanding of the process. By refining in C4 symmetry initially, we aim to improve the overall reconstruction quality using the inherent symmetry of the protein-ligand complex. After refinement, we expand the symmetry to C1, effectively generating multiple copies of each particle in different orientations according to the C4 symmetry. This expanded dataset allows us to capture a broader range of particle orientations and improve the resolution, particularly around the ligand-binding site?
After symmetry expansion following refinement in C4 symmetry, the expanded particles represent various orientations of the protein-ligand complex according to its C4 symmetry. During classification without alignments, these particles are sorted into different classes based on their orientations, allowing us to explore the conformational variability within the dataset without imposing alignment to a reference structure. By experimenting with different classification parameters such as focus and solvent masks, as well as classification settings like forcing hard classification on or off, target resolution, and high-pass filtering, we aim to achieve optimal separation of particles into distinct classes representing different ligand orientations?
I think the issue is that the binding site is on the C4 axis - not within a subunit - but agree I would use a local mask, perhaps offset asymmetrically to include both the central portion around the sym axis and part of a single subunit.
Sorry, I misread the part where the binding site is on the C4 axis, thanks for catching that!
Could you help me understand the advantage of refining in C4 instead of C1 before doing symmetry relaxation? Does symmetry relaxation work better if the starting reference is symmetric instead of asymmetric, or is this just meant to align the reference to the correct symmetry axes?
I’m still trying to improve my understanding of different symmetry strategies, so any insights you could provide would be really appreciated.
Sort of but not exactly. The expanded dataset allows you to capture all orientations of the ligand (which are currently smeared around the axis by the C4-sym of the framework it is binding to) within a single asymmetric mask, allowing you to then (hopefully!) separate the 4 different orientations of the bound ligand without having to search over orientations. Resolution won’t necessarily improve, but hopefully the interpretability of the ligand density will!
Thanks @cbeck, I have been using a a local mask around the binding site to do my local refinement, this has helped but as @olibclarke mentioned the binding site is on the C4 axis, so I still get a ambiguous/heterogeneous ligand density which is why I have been processing the data in C1, otherwise I get the same ligand pose in all 4 directions of my protein. I can usually squeeze out a pose depending on how tight the mask is around the binding pocket , but it’s not as clear as I would like it to be.
In our experience, we get the best results with sym relaxation starting from a symmetric reference, but both approaches are valid, so long as the reference is correctly aligned to the symmetry axes. In Relion, one can also perform symmetry relaxation in local refinement, which has the advantage that one can prevent orientations diverging too much from the symmetric/psuedosymmetric equivalent orientations. This is not yet available in CS, but hopefully will be soon!
Is there really no way to force this in CS homo refinement? If one were to perform symmetric refinement followed by symmetry relaxed refinement with a sufficiently high resolution lowpass filter, like 4-6 Å, does this mean that it doesn’t actually carry over those previous alignments to reconstruct that 4-6 Å initial map? It just uses the lowpass filtered map and freshly aligns particles?
Correct, supplied orientations are not used in NU or homo refinement. Currently there is no way to apply symmetry relaxation in local refinement in CS, only in relion (I have requested this for CS: Symmetry relaxation in v4.4 - local refinement implementation?).
