When collecting data I try to balance the number of particles collected per session with the expected maximum resolution of reconstruction based on the magnification used. I have relied on the simple assumption that if the local resolution of my map is not Nyquist-limited, then my data were collected at a high enough magnification.
This seems like a good forum for asking for advice in this area: Am I oversimplifying things by only considering the Nyquist limit? I have noted that the DQE for a given detector decreases as the resolution approaches Nyquist, but I’m unsure if the extent of variation is consequential in terms of a final reconstruction. I am also wondering if it is easier to align small particles with a lower magnification (and subsequently higher apparent contrast).
Any thoughts or suggested papers on the subject are appreciated.
These are good questions. My thoughts (which may be more or less accurate!) are below.
The final reconstruction resolution is limited by several factors, with the hardest limit being the Nyquist resolution chosen by setting the magnification. However, long before the Nyquist resolution the following things can limit:
- Errors in CTF estimation. CTF of an entire micrograph gets better with more particles in the field of view, but this is only if the particles really lie on a single plane. If they have variable height in the ice (as is almost always the case at high res) then more particles in a micrograph does not help or hurt.
- Inability to align. This is a catch-all but generally the size of the particle and the thickness of the ice (which modulates the SNR due to inelastic scattering and scattering from the solvent) are very important. This is also where the magnification can play an important role via DQE - as you said, most detectors have substantially higher DQE at lower sensor resolutions. If you use a higher mag, high physical resolutions at the sample (say 2A) are mapped to lower sensor resolutions (say 1/2 or 1/4 Nyquist) which will improve the SNR even on the same sample. This is the main tradeoff, as far as I am aware, of magnification vs. number of particles in the FOV.
- flexibility and disorder. These limit resolvability regardless of microscope settings. New algorithmic development (e.g. Non-uniform refinement) can help and this is an active area of progress.
- motion correction accuracy. At a certain resolution, the in-frame drift of particles will limit resolvability, as will any errors/noise in estimated motion trajectories. This can also be helped by higher DQE from higher mag, and the focusing of the beam to a tighter or wider area also have an effect since much of the motion comes from energy deposited by the beam.
Other than these things, the next main limit is just the total SNR based on the number of particles you have and the information (SNR) in each one. If you can perfectly align, have perfect CTFs, and a rigid particle, then you just need to collect enough images to reach any resolution you desire (theoretically). At that point (which pretty much only simple samples like apoferritin can reach) radiation damage to the protein during exposure is the main limitation.