Problem with small membrane protein

Hello Everyone.
I am working on a transporter which is a homodimer of 80 kDa. I have collected some 6000 movies and processed them in Cryosparc V4.1.

I collected these movies when i was new to cryo-EM. One movie is around 500 mb. Now i am confused if these are gain corrected or not? usually a Tiff LZW non gain corrected movie at 0.85 A pixel size is around 250 mb!
I followed standard protocol and did import movies, patch motion correction follower by patch CTF estimation.Did a micrograph curation based on CTF fit better than 5A. I used Topaz train (on 1000 micrographs) and extract on the 50k cleaned good particles resulting from one round of 2d classification from blob picker. When i measured the end to end distance of my model in pymol, it came out to be 95A. So to took both 320 as well as 256 px box size. The 256 box is currently giving a resolution of 4.21A.
I did iterative round of 2D classification and got 301,047 particles which i took for ab initio reconstruction with two classes (resolution initial 12- final 8A). Repeated second ab initio until i got rid of junk and did a Non uniform refinement and final particle number is 104,465. i tried global and local ctf refinement but it worsens the resolution. Used dynamic mask threshold of 0.4 and got the resolution of 4.21 using Non uniform refinement. Since this is a ABC transporter having c2 pseudosymmetry, after using C2 without alignment i am able to get the above resolution.
I have tried to use mask using Seager in ChimeraX where i masked the detergent micelles, but it makes the whole map even worse. Although i have used the pixel size of 0.85 i am unable to reach anywhere near 2.5-3A. I am unable to figure out where is the problem. Is there any flipping needed in gain for which there is this sort of problem of i need not use gain correction seeing the size of the file. Here are some results of final NU refinement? I would love to have suggestions from the community.

There are many things that can affect the size of compressed TIFFs - for example, they will generally be larger for lower magnifications. Even if a gain were stored in the TIFF, you would expect that to increase the size by only a few percent, not double it, so there is probably some other reason.

What you are trying in terms of masking makes sense, however, instead of doing a new global alignment search you should try doing a local refinement with this mask. The micelle is probably very helpful in aligning the particle at low resolution. Make sure to use pose/shift Gaussian priors and to allow re-centering of orientations and shifts.

It’s great you already got close to 4Å (assuming the map quality also reflects this resolution). Given that’s the case and there are still hundreds of thousands of particles, my recommendation is to try 3D classification with local searches in Relion.

1 Like

There are no obvious defect lines, which may indicate that it is gain corrected… however there is a slight ripple with opposite curvature from the hole edge, which may indicate a gain file is needed and is missing.

Also, while there are lots more things to try (Daniel mentions several) sometimes data just gets stuck when you think it would go further.

Having said that, if the map looks like 4.2Å, local refinement, classification and local refinement in RELION, optimising optics groups for better beam tilt and magnification anisotropy refinement may all yield gains.

The above micrographs are gain corrected. Yesterday i started reprocessing the data and imported the movies without gain reference and did a patch motion correction. The ripples which are visible in my previous post are gone (attached pics).
So does that mean that the movies are already gain corrected and doing a gain correction is causing ripples/artifacts on the micrographs?

Thank Denial for the advise. For a quiet long time i was trying to use the relion 3d classification and refinement. I am using Scipion too. What i did is imported the particles in scipion and then run a 3D classification using Relion on the CS NU refined particles sets. I am using Relion in default setting as i do not know how to play with the parameters of 3D refinement in Relion. So should i do a local refinement using the mask generated in cryosparc? Also when you mentioned “Make sure to use pose/shift Gaussian priors and to allow re-centering of orientations and shifts”, i would be curious how to do that?
here is the 4.2A map. the right one is contoured at step 1, 0.121 threshold.

Do local refinement in CS using your custom mask w/out micelle. Read all the job options and turn on the ones I mentioned. You can also try 3D classification in CS, with this mask as the “solvent mask” OR you give this mask as “focus mask” and a mask including entire particle + micelle as “solvent mask.”

In Relion, you can do local search classification by selecting “Local searches - Yes” and putting something like 15˚ range. I recommend you set the “strict E-step limit” to 6 Å, and the sampling to 1.8˚. Start with 4 classes and do jobs with T = 8, 16, 32, 64. It’s too high when you get a streaky map. Then you may try different numbers of classes, probably 4 - 8. If you run out of GPU memory use --pad 1.5 in the extra options. Here I would also use your custom mask w/out micelle, set the diameter to the size of your particle + 30 Å.