Rotational randomness


I have a large protein with two groups of domains connected by what might be termed “spaghetti.” I have managed to solve both parts of the structure (to 3.3 and 4.3 respectively) by solving the bigger section traditionally, subtracting the particles and solving the smaller section by giving it a nice predicted volume. The question now is what is the relationship between the two domains? Is it truly spaghetti, or just rotational freedom? I have tried various 3DFlex, 3D variability and 3Dclassification jobs but the hassle is I really don’t have a mask that encompasses both domains unless I smear out one domain (I’ve just tried fat puffy domain and one I created by rotating the mask 15 degrees for 180 degrees and kept adding it together). In all cases, I just don’t think the algorithms are built for this kind of question. They are looking for smaller movements or smaller domains with moderate movements. Not 1/3 of the structure with a huge movement.

But, I have the answer in the .cs files. I used the same set of particles (albeit - one set was subtracted, in theory they are connected). Each set of particles would have 3D alignments associated with them. ALL I have to do is learn to use the python based cryosparc Tools and understand vector mathematics. Before I go that far (when I was in college I took a TrueBasic course), can anyone confirm this is probably the way to go. I appreciate that there are many variables in the 3D tools that I could continue to play with but I haven’t seen anything even remotely useful over about 7-8 different type of runs.

thanks for any feedback

My software (pyem) and Relion multibody PCA have similar ways to approach this, by looking at the differences in the alignments of subparticles from the same real particle.

First, though, I think it’s important to define the question carefully. What do you mean by spaghetti vs. rotational freedom? Could the intrinsically disordered linker really act like a rod with a limited number of rotatable bonds? Do the domains interact with each other directly?

All of the complexity in analyzing the domain motions using “multibody techniques” (inspecting the subparticle alignments) comes from the 3D rotations. Would just looking at the distribution of the distance between the domain centers, which is pretty simple, answer your question? (This would be a histogram plot of how far apart the two domains are in each particle).

Thanks for the discussion Daniel,
Related crystal structures show the two domains as a big “X”. (This is similar to a fatty acid synthase) These proteins did crystallize suggesting that the rotational flexibility isn’t infinite in some homologs. But, no, I don’t expect the two domains to interact with each other. I think of this as two pencils connected by string. They may rotate in the parallel plane distinct from each other but the distance remains pretty equal and the central C2 axis is probably where the rotational axis occurs. I will admit that I do see that one domain flops a little to the side with respect to the main axis though. It’s hard to know since the only density observed for the small domain when I solve the big domain is a small ball of complete noise. It is averaged density which is centered but not aligned. I guess I was assuming that the vector difference between alignments would show either - random orientation (where the N- and C- terminal portions are not linked in space with up to 180 degrees of motion) or limited rotation (where there really is only 10-20 degrees of freedom of one domain with respect to each other) or even clusters of rotation (where the two domains rotate according to low energy minima with respect to each other.)

The problem (other than I don’t know python or vector math) is that 3DFlex wants the threshold to relate to the expected differences - so 10-80A in motion. This just gives me balloon blobs expanding and contracting, not rotating with respect to each other. And I don’t have a mask that contains all of each molecule. If the mask holds the larger domain rigidly with a ~5A surface, the lower domain could be anywhere - and if I rotate the presumed domain around the C2 axis then I get a huge oval of very low resolution. The program doesn’t take into effect the knowledge of the 4.3A structure I obtained with subtracted particles.

I guess a histogram plot of centers would be a start. If it showed relatively small changes in center difference than we assume just rotation. There are only ~8 residues in the linker so I don’t think the two domains breathe that much. But I think your first thought it more of what I thought I wanted “differences in the alignments of subparticles from the same real particle.”


Is it possible to manually “sense” this from a movie of aligned 2D? Can you see both domains in the parallel axis from one view? Then cut out all 2D of a single view and overlay and watch the second domain. Gabe lander DFCI TPD seminar video on YouTube 19:28 mark

1 Like