Hey everyone,
I’m having troubles processing a one recent dataset. It’s a oligomeric membrane protein in detergent, collected at Krios with K3 detector in counting super-resolution mode.
I had this sample collected and processed earlier with slightly different buffer composition and got a sub-3A reconstruction. But now I’m suddently struggling a lot. I can see a lot of particles (around a million after few rounds of cleaning), 2D classes look ok, I even see the proper oligomeric assembly within the classes. But after refinement, the best I get is ~8A without oligomeric symmetry / 5A with it, even though I was under 3A for the other sample.
I tried multiple classes during ab initio reconstruction – didn’t help much. Refinement of CTF parameters also didn’t work out for me.
What can be the problem here?
If it was crystallography, I’d check for radiation damage – but I have no idea how to do that in case of cryoEM. Does the radiation damage indeed look like that? Can I e.g. process the first half of the movies and see if the resolution is increasing?
Radiation damage in single-particle EM is accounted for by dose fractionation; high frequency contributions are downweighted in the later movie frames, where they have been corrupted by radiation. Regardless of dose weighting, radiation damage occurs randomly over different particles and becomes another B-factor that can be overcome by averaging. Very labile features, such as disulfides, may only be visible with dose weighting or by reconstructing (not refining) with a restricted number of frames.
Unfortunately the most likely explanation is variation in sample quality for some reason. Are the micrograph CTF fits all good? Are you doing 3D classification aside from ab initio?
CTF fits are good (not worse than in another dataset).
I’m running 3D classification now, with 20 classes, but it seems that it prefers to put all the particles into one single class for now, after >50% of iterations.
Also, I noticed that there’s a preferred orientation issue, but things like “rebalance 2D classes” don’t seem to help – after removing 70% of preferred orientation particles, it still reconstructs to ~6A with 100k particles.
I meant regular 3d classification, with search. “Heterogeneous refinement” in cryoSPARC. If there is a true preferred orientation - it sits one way in the ice - that wasn’t there before then that is almost certainly a sample prep issue. If there’s just an orientation bias then it’s usually not that big of a problem (depends on the shape of the particle). Since you already had a high resolution structure, you can try using a high resolution initial reference (or sometimes another software will do better with a given particle and orientation distribution, because of different filtering).
In my experience this kind of variation usually comes down to sample quality - prep less monodisperse than usual, ice too thick, change in ligand or stabilizing buffer, etc.