Hi cryosparc community,
I have a dataset for a transmembrane protein with both the n-terminus and c-terminus on the same side of the membrane. We have a two tag system that ideally snap to each other on one side of the membrane. i was able to work out what I believe to be a very decent 2D classes (below) but could not get it to a decent 3D reconstruction. Part of it is potential heterogeneity in how it insert in lipids and part because the two tag system sometimes do not bind each other. In 3D, masking, or particle subtraction did not do much. the only semi decent reconstruction was from a variation of HR-HAIR. But again seems like high resolution noise with regard to the lipid and poor/incomplete helices for the protein. Any thoughts on how to tease out heterogenity for such a small molecule would be appreciated.
@rwaldo @olibclarke
Thanks and best regards,
Bassem
Those are beautiful 2Ds! This should definitely be a solvable problem, especially given that the helices are clearly visible in the micelle. Could you provide a bit more data on your processing workflow? How many particles do you currently have? Can you try running the recently published GPCR workflow found here?
2 Likes
Ab initio with a lot of classes (18+) starting at, maybe 12Å and targeting 6Å should net you something usable.
Then heterogeneous refinement starting at 10Å or so.
Caution, though, if you’re not reaching <4Å from a starting point of 8-12Å, I would not consider it trustworthy.
175K particles.
I tried different picking methods, remove duplicates and run extensive 2D classification for small particles. When I get to ab initio if i leave all particles in, I end up with ab-inito maps that donot capture the snap tags i have on one side of the nanodisc. Variations of ab-initio with selected few particles give a volume that recapitulate what we put in, aka with the two tag parts snapped to each other. If I try to focus refine the tag then look in the nanodisc, I get very low resolution reconstruction. If I use the tag partial or full to orient the particles such that the tag on one side, subtract the tag and try NU refinement or local refinement, it end up in very glossy maps and I lose all details.
Tried that but not with 18+ classes. I have like 175K particles. I can try a variation of this and see how it works. I get my best resolved volumes in ab-inito so far. Did heterogenous refinement too. It seems i get slightly better resolution when i subtract some of the nanodisc but since I suspect the protein to embed at angle and not straight all the time, i think it is unreliable till I can see better details of the protein. Will try the high number classes ab-initio job and update you.
1 Like
That’s a fairly small number of particles. How many micrographs are you working from? If you’re being too aggressive w/ 2D curation that could be adding to your issue.
That’s one of the reasons the automated workflow is nice, it relies much more heavily on binning techniques such as multi-class ab initio and het. refine w/ decoy volumes to curate particles.
EDIT: Just for reference, most of the particle stacks from the pre-print describing the automated GPCR workflow are well above 250k particles.
We typically sort particle sets down to 30-40k particles to get to around 3 Ans. We often start with 200-400k particles. So it’s definitely possible to get high resolution reconstructions with a very limited number of particles with membrane proteins.
This to me also looks solvable but will be very hard. We have a similar situation with a very heterogeneous extra membrane domains which took 2-3k jobs to solve. It’s difficult to say exactly what should be done but I do agree that working more in 3D is often the best, especially once you can resolve TMHs. Good luck.
1 Like