I recently determined two structures of the enzymes at 3 A resolutions. The results from 3DVA may potentially be useful to shed light on mechanistic insights. I am wondering if there is the way to translate the density movements defined by 3DVA into those for the atomic model? Here I am talking about something similar to MD stimulation. Or my thought would be a bit crazy?
a low-tech way to do this: build a model at each endpoint of the motion (coot, isolde, etc.) then perform linear interpolation within chimera to visualize the transition. it would also be good for you to know (with 3DVA cluster mode and with 3D classification job) whether your 3DVA results suggest continuous motion (structures can be solved at every intermediate) or whether the protein exists in the two end conformations and these are “discrete states”.
Thanks for the idea. I will try because it is low-tech way as you indicated. Perhaps what I would do is to refine an atomic model in each state (generated by cluster mode) to fit the density and interpolate via Chimera to get some idea how resides particularly with or near catalytic site might move.
Thanks, unfortunately it won’t be available there for ~4.5 months. For some reason I couldn’t find the Biorxiv version last week, my fault I’m sure.
In the study, did you use computed (‘simple mode’) maps or reconstructed ones (‘intermediates’)? Also does phenix.varref use models for adjacent steps in the volume series to restrain refinement? In other words, is the only difference from running real-space refinement in a loop the perturbation of each input model, or is there more to it?
@DanielAsarnow , sorry for the dealy in availability I wasn’t aware of it. I guess you were finally able to locate the BioRXiv, great and thanks for posting it.
Lets me first answer your question on the refinement strategy. It actually doesn’t use just restrained refinement as it is trapped in a local minimum and is not perturbed enough to reach the next map. So (and this is Pavel Afonine’s expertise and choice) varref uses what is called “morphing” in Phenix but essentially uses Simulated annealing with slow decent to go further from that local minimum. Using 50 initial models and refinement for each of them, the best one is chosen for a given map. This later pdb is then used as a template to reach the next map via the same strategy. There is also an adaptation of the “refinement” strategy according to the resolution used for 3DVA calculation. Obviously, one cannot claim the same things at 3 or 6 Å resolution, so users need to check their data to make sure what they observe matches reality. This was exemplified on the spliseosome, where one part of the map was not well defined, varref can refine a model there, but the results cannot be used in this region to make any claim on protein movements. On other parts of the map, it was possible even at 8 Å.
About the 3DVA maps, we tend to see the same results in simple or intermediates modes for our system so we used simple mode here. But we have tested it in various scenarios and it worked well, and on different proteins (more proteins tested than on the article). If it doesn’t work for some reason, then let us know to fix it.
For the record, I have also tested on a 3DFlex map. It was on a small movement, and varref did the job very well. I haven’t gotten to trying it on larger motions, but it’s on the todo list. Maybe for these maps the strategy would need to be adapted?
I believe that a combination of 3DVA and 3D refinement provides experimental data for motions of residues - in our case at side-chain level, and the correlation between RMS of residues from 3DAV and those from MD simulation is stunningly good. So, we should explore the power of 3DVA more than simply visualizing motions of proteins.