Resolving flexible protein

[leti]

Hi, cryoSPARC community,

I’m working on resolving a protein complex structure of two proteins, one an outer membrane protein associated with DDM micelle and its binding partner. Since I’m working with membrane proteins, I used the following scheme: ab initio reconstruction (two different jobs: 1 class and 3 classes) and non-uniform refinement. Using non-uniform refinement I have resolved a nice ~3ish Å resolution map of one of the proteins, but its binding partner appears to be more flexible and at a low threshold appears to have a density at the same level of noise. When the threshold is increased to remove the noise density from the binding partner also disappears (example shown below). From what I have read non-uniform refinement is best at resolving regions of flexibility, but this doesn’t seem to be the case in my situation. I’m hoping you all may have suggestions I have not tried yet. So far, I have tried particle subtraction, subtracting the outer membrane protein’s density, local refinement, and different masking strategies, but have had no luck.

Any suggestions or insight is much appreciated. Thank you, Leti

[olibclarke]

Hi Leti,

Have you tried classification without alignment in relion? This may not help if the issue is truly flexibility, but it could if it is low occupancy that is causing the issue. Also, when you say you have tried local refinement, what was your experience there, and what parameters did you use?

Cheers
Oli

[leti]

Hi Oli,

Thanks for the quick reply. I have not tried Relion. We suspect occupancy could be an issue. I’ll give Relion a try. Local refinement results are below. It’s much “better” in terms of more visible density at a low threshold but if I want to remove more noise (assuming this is the micelle belt) around the outer membrane protein by increasing the threshold then the density of the flexible protein disappears. The overall resolution of the outer membrane protein also takes a hit. The parameters I used for local refinement were defaulted, with the exception of the mask parameter. I set this to static. I supplied a mask around the complex to try to get rid of the micelles. The mask also had default parameters with the exception of the soft padding I set to 19.

[olibclarke]

Have you tried local refinement using a mask just around the flexible component? If not, I would try that, starting with a very small search range

[leti]

Hi Oli,

Yes I have and the best reconstruction I got is below (in green). There is extra density “floating around” that I could not get rid of. I don’t recall adjusting the search range. Are you referring to “shift search” and “rotation search”?

[olibclarke]

yes, those are the parameters I am referring to. perhaps if you post a snapshot of the parameters you used, I can give some suggestions?

[olibclarke]

Hi Leti, this is not local refinement (or if it is you are not showing the search parameters here)

[leti]

Hi Oli,

You are right. Sorry about that. I sent parameters for NU-refinement. Here are the parameters for local refinement.

[olibclarke]

right, so for something small like this, I’d suggest switching on “use gaussian prior”, and use priors of 3deg/1Å with search range of 9deg/3Å to start with (maybe trying with/without recentering of rotations and shifts at each iteration). I would also reduce the initial lowpass resolution to 8Å (assuming that 8Å features were present in the initial map before local refinement)

[leti]

Thanks for the suggestions Oli. I will give this a try as well.

All the best,
Leti

[marino-j]

I think the “green blobs” floating around are also in the mask you use for local refinement. If you go towards zero in the threshold, do you see those blobs in the mask ?

[Guillaume]

In the maps you showed where the bound protein is visible under the membrane protein, it is very apparent that you have conformational flexibility: these helices look stretched in one direction and squished in a perpendicular direction. Without conformational flexibility, they should look more like the helices in the membrane protein.

If you haven’t tried 3DVA yet, I recommend it. Depending on the results, you may be able to isolate a subset of particles to run NU-refinement on, and it may improve this density a lot.

[mmclean]

Hi @leti,

Thanks for the interesting post – I’d second all of the above comments/suggestions wrt. local refinement and 3DVA as potential methods to help resolve the flexibility. You may also want to check out this similar thread where tweaking masking parameters (softness) and search parameters were helpful:

• Progression after 3DVA clustering: loss of detail after local refinement VS NU-refinment

Best,
Michael

[cryo-lg]

Hi @leti,

I am working on flexible complexes, and I am a big fan of the 3DVA to disentangle conformations, followed by local refinement (the post @mmclean linked was me having problems).

A couple of pointers from my limited experience.

I started with a heterogeneous refinement from the first ab initio with 8 classes to remove really spurious stuff.

Then I pooled all remaining classes and did a NU-refinement to get a consensus structure.

In the consensus volume, I identified dynamic regions, then in chimera generated masks for all of these with the corresponding consensus-minus-local-region-masks.
Local masks have to have a decent size, otherwise the local refinement will struggle (in my case a sphere of 80 in a 512^3 box seemed to work well, but much smaller masks not as much).
I removed a bit more from the consensus-minus-mask volumes, as I will use them for particle subtraction later.
I treated all the masks with 5-10 dilation and 20-25 padding, setting an appropriate threshold (this step is critical for good results later, e.g. particle subtraction).

For each mask and consensus-minus-mask, I started with a local refinement using the consensus-minus-mask.

The particles from this alignment were then subjected to particle subtraction using the consensus-minus-mask.

The subtracted particles were downsampled in my case (3DVA crashed with my 512^3 boxsize), if your box is smaller that’s likely not a problem.

The downsampled particles were used for 3DVA with the mask corresponding to the region of interest (i.e. what remained after my particle subtraction).

3DVA display was used in clustering mode (it’s by default set to make transition movies) to generate 4-10 clusters from the 3DVA.

Here clusters with good density for my region of interest became immediately apparent.

After choosing the best cluster, I used local refinement on it, with the original blobs from the unbinned particles (from the particle subtraction job).

This procedure was repeated for each region of interest.

Then in chimera, I merged the volumes with volume maximum (chimeraX) or vop maximum (chimera or chimeraX).

I am still optimising the final map, but it looks pretty ok.

[leti]

Hi @marino-j,

That makes a lot of sense. I just checked and yes I actually do see the blobs when I go towards zero in the threshold. I’m in the process of making a better mask to exclude the blobs. Thanks for your feedback!

-Leti

[leti]

Hi @Guillaume,

I have tried using 3DVA, but have had no luck. The job seems rather straightforward but I’m having difficulties figuring out what filter resolution I should use. If I understand, setting a larger filter resolution, let’s say 15Å are best for large-scale movements and a smaller filter resolution, for example, 3Å is best for small-scale movements?

I’m also not sure which 3D variability display output mode would be best to use, simple, cluster, or intermediate. I’ve tried all three.

Leti

[leti]

Hi @mmclean,

Thanks for the reply. I have actually used the suggestions in that particular thread. From this thread, I was inspired to try using soft padding set to 19.

Leti

[marino-j]

you can do it simply by using the eraser in Chimera, and save the new map. cheers

[leti]

Hi @cryo-lg

Thanks for your comments and suggestions. I will try doing this workflow this week.

All the best,
Leti

[leti]

Got it! Thanks, @marino-j!