Appropriate number of 3D-VA components to capture asymmetric changes in symmetric systems?


Thinking about symmetric systems (e.g. the proteasome example in the 3D-VA paper: 3D variability analysis: Resolving continuous flexibility and discrete heterogeneity from single particle cryo-EM - ScienceDirect), how can one best capture asymmetric conformational or compositional changes in such a system using 3D-VA?

In the case of an asymmetric entity binding randomly to one of N subunits in a Cn symmetric framework, it seems like we’ll need to multiply the number of required components by N (as binding to each subunit will effectively represent an independent mode).

Of course in that case (a compositional change in a single subunit) one might be able to perform symmetry expansion and local refinement first, then 3D-VA. But what about an asymmetric motion of the entire framework, e.g. a bending of the entire assembly? Again, assuming the framework has been refined with Cn symmetry, intuitively it seems like we will need at least N components to capture such a motion. Is this reasonable, or am I missing something obvious?


Hi @olibclarke ,

Interesting question! I look forward to feedback by others and the team on this topic.

If I understand well, it is not possible to observe asymmetric deformations using 3DVA within a component. We are able to understand how a ligand triggers asymmetry by integrating 3DVA with MD simulations, where MD points to a possible mechanism and 3DVA confirms that some movements are restricted, in line with biochemistry.

Since components are in decreasing order of particle variance, my hunch is that unless your variance is equally distributed across all your subunits (which I think is unlikely), you will see the main movements first and start to “loose signal”. The question is “how far” can we go in components to explain protein movements and I guess it is protein dependent. In my hands, I feel I can’t go too far as the message gets very complex and I’m not sure I can relate these movements to other biochemical/enzymology/biophysical methods to make a story.

Good luck.

Hi @vincent,

I am talking about asymmetric components - e.g. see Fig 7D of the 3D-VA paper, showing asymmetric bending of the entire (symmetric) framework of the proteasome. In this case one would expect the variance to be equally distributed in every symmetry-related direction, as the consensus refinement was performed with symmetry applied.

So my expectation is that in such a case, if you run 3D-VA with enough components, you will find seven components corresponding to asymmetric bending that are equal in magnitude, but which differ in direction.

By contrast, for symmetric motions (e.g. in the case of the proteasome example, twisting around the Cn axis), I would expect to find only one mode, as the symmetry of the motion matches the symmetry of the framework - but maybe (probably!) I am misunderstanding something.

Of course one can perform symmetry expansion first, but this will not I think change the relative prominence of symmetric vs asymmetric modes (just increase runtime by N times).


Hi @olibclarke,

Apologies for a very late reponse. FWIW, our compute team discussed this post and agree with your statement below.

So my expectation is that in such a case, if you run 3D-VA with enough components, you will find seven components corresponding to asymmetric bending that are equal in magnitude, but which differ in direction.

In general for a symmetry-breaking change represented by a particular mode, we think 3DVA would find N of these modes in a complex with global symmetry of order N.

This could be prohibitive for large N. One idea we talked about was using 3DVA (perhaps in intermediates mode) to obtain a canonical example of a symmetry-broken version of your volume (this would be one of N possible symmetry-related modes, given the above). Then, it could be possible to re-align all particles to this volume — perhaps as simple as running a global refinement again (potentially assisted via symmetry relaxation :slight_smile:), and repeating 3DVA. The hope is that after this process, 3DVA could find the same mode present only once, since all of the particles have been re-aligned to this specific symmetry-breaking feature.



Hi Michael - good idea re refining using the symmetry broken volume as a starting point, will try! Intriguing int re symmetry relaxation, look forward to trying that! :smiley: