Can multiple conformations appear as a single density in cryo-EM maps?

Model Building
1

Hi everyone,

I’m currently working with a relatively low-resolution cryo-EM map and encountered an issue during model building.

While interpreting the density, it seems that two different models/conformations could plausibly fit the same region. In crystallography, I understand that multiple conformations can sometimes coexist and appear simultaneously in the electron density (for example through alternative conformations or partial occupancy). I was wondering whether a similar situation can occur in cryo-EM maps.

In my case, I suspected that structural heterogeneity might be contributing to this ambiguity. To investigate this, I tried several approaches including ab-initio reconstruction and 3D classification, hoping to separate the two possible states. However, these attempts did not help much, possibly because the overall resolution of the data is relatively limited.

So I wanted to ask:

Can multiple conformations/states appear merged into a single cryo-EM density map in a way that makes two different models appear plausible?
Have others experienced similar situations during model building?
If classification does not separate them because of limited resolution, are there alternative strategies to investigate whether multiple states are actually present?

Any suggestions or experiences would be greatly appreciated.

Thanks!

2

absolutely this is always happening - and can be easily exemplified in the thought experiment: take any two sets of particles from different structures and use them as input to generate a single refinement volume. The goal of data processing is to limit the compositional and conformational heterogeneity in the sample by classifying the different forms into different particle sets, but this is very inefficient and requires close attention and iteration.

If 3D classification with a low filter resolution (10, 12, 14) and a high number of classes (10-20 keeping 20k particles per class or more) doesn’t automatically separate the motion, then you will need to try 3DVA, masked classification, multi-model ab initio, or multi-model het refinement preferably using volumes which are pure for the two states (possibly by molmap the built model in each of the two states). You can also potentially get lucky with 2D orientation and see the difference in 2D classes - same view but one region of interest different - and use this to bias your initial model development for each state in addition to ‘proof’ that multiple states exist. but you have to know your particle 2D representations pretty well.