Extracting xyz coordinates of subunits from particles

Hi all, I have a set of particles that is in apparent pseudo-C3 symmetry, and they are quite heterogenous in term of size. I was able to solve a ~3Å high res map for one of the subunits using Cryosparc. From 2DClass, there are classes where that particular subunit produces a sharp signal with a blurry background with different diameters, and for remaining classes the particle looks triangular and with blurry densities for the subunits. My current goal is to determine how the whole particle looks like.

From hetero refinement, I can only get low res maps coming with different diameters, but with contours of 3 subunits when try to refine the particles as whole. It seems that heterogenous refinement works best (but not satisfactory because I cannot fit the high res model into “subunit” densities confidently) to separate the whole complex in population with different sizes. 3DVA, cryodrgn and relion Class3D do not give good results, probably because a single consensus map cannot be generated or represent most particles. Is there any other way I can analyze the particle set to get quaternary structure map?

Second idea in my mind is to use the current map as input of Topaz train and extract to locate the location of each subunit, then extract their coordinates after remove duplicates and NU refinement. I wonder if I can simply use micrograph coordinate x,y as x,y of a particular particle, and using mean of defocusU and defocusV as z coordinate? Using these coordinates, I may try to plot relative positions as triangles and analysis the heterogeneity of the pool.

Another idea that we can think of is to do particle subtraction on the particle stack using the high res map, then iteratively run NU refinement using high res map as initial model and perform particle subtraction again. It feels like manual symmetry expansion to locate the coordinates of all usable subunits and plot the relative positions like in the second idea. Do you think this workflow adequate/work?

Apart from the above ideas, do you have other better ways to represent quaternary structure of complex with non-point group symmetry but with multiple subunit copies? Thank you for your help in advance!

Is it a homotrimer that is flexible or a heterotrimer? If it is a homotrimer, what do you get with C3 imposed? Try also C1 in Relion, which will less likely to over refine and might give you a better starting point. One could try symmetry expansion and local refinement with a mask around one subunit. If it is a heterotrimer have you tried multibody refinement in relion(?)

Depending on how that looks, you could also do signal subtraction (erase the good part) followed by 3D classification without realignment (both steps in relion) with a mask around the remaining two subunits. You might be able to pull out a class where all three (or two) subunits align better.

Peter