Determining tomogram handedness, and CTF correcting

[Colin Gauvin]

Hi,

I am trying to reconstruct and CTF correct an entire tomogram. I was hoping to check my understanding of some of the concepts, to ensure I'm doing things properly.

My understanding is that for CTF correction to be accurate, the handedness of the tomogram must be known, because the distance of the tile from the tilt-axis is used in estimating the CTF. Distance from the tilt axis in one direction is +Δz, and distance in the other direction is -Δz. (How exactly those directions are defined depends on whether the tilt angle is + or -) Is that correct?

I think I know my tilt axis, SerialEM says it's very close to zero, estimated at 0.9°. However, because I don't necessarily know the rotation of the image in the microscope, my understanding is that it's possible it's actually -179.1°  Is that correct? 

And if it were -179.1°, and I reconstructed a tomogram at 0.9°, the +Δz and -Δz would be flipped, and I would have a tomogram that was fit to an inverted defocus gradient. i.e. EMAN2 would be fitting less defocus when it should be fitting greater defocus, and vice versa, because I would have the side of the tomogram that's tilting up vs the side that's tilting down, incorrect. Hopefully I'm still understanding here?

One further question is - what happens if I don't CTF correct the tomogram, and instead CTF correct the micrograph (basically just phase flip each tilt series image)? My understanding is that the power spectrum of a tilted image will show the average defocus * some weight of what's actually scattering electrons (i.e. sample, if the sample is located in the middle maybe this average would be more accurate than if the sample is located throughout the entire image). This average defocus would have both the + and - Δz portions of the image aliased together, and in areas where they are completely out of phase, actually canceling each other out, and so that's why it's better to use the method where the image is chunked up, and the distance from the tilt axis is used to help model?

Finally, when I take a reconstructed tomogram, with a tilt axis that I set (0.9°), and run it through EMAN2's CTF estimation program, I get very ambiguous results. It scores well (I think the fits are real), but says that in about 65% of my tilts, the flipped hand is better. But then, if I reconstruct at -179.1°, and run that through the CTF estimation software with checkhand, it says that the flipped hand (0.9°) is better in about 55% of tilts. Is this because my tilt axis is so close to zero? Is my only real option here to solve a structure of known handedness?

Thanks for any and all input/help.

Colin Gauvin

Montana State University

PhD Candidate/Assistant Facility Manager

[Ludtke, Steven J.]

On Nov 4, 2022, at 2:41 PM, Colin Gauvin <ccgau...@gmail.com> wrote:

Hi,

I am trying to reconstruct and CTF correct an entire tomogram. I was hoping to check my understanding of some of the concepts, to ensure I'm doing things properly.

My understanding is that for CTF correction to be accurate, the handedness of the tomogram must be known, because the distance of the tile from the tilt-axis is used in estimating the CTF. Distance from the tilt axis in one direction is +Δz, and distance in the other direction is -Δz. (How exactly those directions are defined depends on whether the tilt angle is + or -) Is that correct?

I think I know my tilt axis, SerialEM says it's very close to zero, estimated at 0.9°. However, because I don't necessarily know the rotation of the image in the microscope, my understanding is that it's possible it's actually -179.1°  Is that correct? 

And if it were -179.1°, and I reconstructed a tomogram at 0.9°, the +Δz and -Δz would be flipped, and I would have a tomogram that was fit to an inverted defocus gradient. i.e. EMAN2 would be fitting less defocus when it should be fitting greater defocus, and vice versa, because I would have the side of the tomogram that's tilting up vs the side that's tilting down, incorrect. Hopefully I'm still understanding here?

Yes, your understanding of all this sounds correct. The --checkhand option is used (by checking whether a positive or negative CTF gradient fits the micrographs better) to resolve the 0.9 vs -179.1 ambiguity. This is (or was recently) a unique capability of EMAN2. 

While this does guarantee you get the correct CTF correction, technically it is still possible for the handedness of your final reconstructions to be inverted if the CCD data collection software inverts the image as part of data saving (due to a misunderstanding of the MRC format on the side of the Camera maker). 

One further question is - what happens if I don't CTF correct the tomogram, and instead CTF correct the micrograph (basically just phase flip each tilt series image)? My understanding is that the power spectrum of a tilted image will show the average defocus * some weight of what's actually scattering electrons (i.e. sample, if the sample is located in the middle maybe this average would be more accurate than if the sample is located throughout the entire image). This average defocus would have both the + and - Δz portions of the image aliased together, and in areas where they are completely out of phase, actually canceling each other out, and so that's why it's better to use the method where the image is chunked up, and the distance from the tilt axis is used to help model?

To be clear, while it is possible to generate strip phase-flipped tomograms in EMAN2, this does not happen in the normal workflow, and even if produced, these "ctf corrected" tomograms are not used as part of the subtomogram averaging process. If you follow the normal pipeline:

- you reconstruct a few tomograms

- then run the handedness check to get the tilt axis correct

- then reconstruct the full set of tomograms with the correct tilt axis

- THEN do CTF determination on the full set of tilt-series (which requires the reconstruction geometry to work)

so clearly the tomograms are not normally CTF corrected, since you don't have the CTF parameters yet when you do them. If you want CTF corrected tomograms for some purpose of your own, you have to rerun the reconstructions again after determining the CTF.

Regardless, the current subtomogram averaging process does not use the reconstructed tomograms in any way. The tomograms are really only used for particle picking. Once the locations of the particles are known, subtilt series are extracted from the original tilt series, one for each particle. These subtilt series come with CTF information based on where each particle is located in 3-D, independent for each tilt in the subtilt series. This information is used for CTF correction during "subtomogram averaging" which is really "subtilt averaging".

Finally, when I take a reconstructed tomogram, with a tilt axis that I set (0.9°), and run it through EMAN2's CTF estimation program, I get very ambiguous results. It scores well (I think the fits are real), but says that in about 65% of my tilts, the flipped hand is better. But then, if I reconstruct at -179.1°, and run that through the CTF estimation software with checkhand, it says that the flipped hand (0.9°) is better in about 55% of tilts. Is this because my tilt axis is so close to zero? Is my only real option here to solve a structure of known handedness?

Again, given the possible error in data collection software, handedness of the final result isn't an absolute measure of correctness here. The tilt axis determination process should only need to be done once for a single microscope/detector/magnification. If the contrast in the tilt series isn't great (few particles, or other issues) or if the specimen is particularly thick(?) or maybe if there are a lot of alignment issues (?), you may get some ambiguity in the results. You should get self consistent results from all of the tilt series in a data set. If a few are ambiguous or off, that's fine as long as most agree, use the consensus value. 

Thanks for any and all input/help.

Colin Gauvin

Montana State University

PhD Candidate/Assistant Facility Manager

--
--
----------------------------------------------------------------------------------------------
You received this message because you are subscribed to the Google
Groups "EMAN2" group.
To post to this group, send email to em...@googlegroups.com
To unsubscribe from this group, send email to eman2+un...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/eman2

---
You received this message because you are subscribed to the Google Groups "EMAN2" group.
To unsubscribe from this group and stop receiving emails from it, send an email to eman2+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/eman2/99d64e12-6cdb-4ac5-927a-de90dd339336n%40googlegroups.com.

[Colin Gauvin]

>While this does guarantee you get the correct CTF correction, technically it is still possible for the handedness of your final reconstructions to be inverted if the CCD data collection software inverts the image as part of data saving (due to a misunderstanding of the MRC format on the side of the Camera maker).

OK, this makes sense. Hopefully I will be able to determine that by getting a high resolution structure via tomography at some point.

>so clearly the tomograms are not normally CTF corrected, since you don't have the CTF parameters yet when you do them. If you want CTF corrected tomograms for some purpose of your own, you have to rerun the reconstructions again after determining the CTF.

This is my current plan, as I don't think the STA pipeline in EMAN2 is suitable for this project, but the fiducialess alignment worked very well, so I want to construct the tomogram with EMAN2. It seems that doing the CTF estimation, and running the construction again with --load will write out a CTF corrected tomogram. In this case, these are large viruses. I wonder if it would make sense to place a very large particle and have EMAN2 extract the whole virus as a particle...

>The tilt axis determination process should only need to be done once for a single microscope/detector/magnification

This is good to know, thanks.

>You should get self consistent results from all of the tilt series in a data set. If a few are ambiguous or off, that's fine as long as most agree, use the consensus value.

Awesome, I have a number of tomograms and they seem to agree, so I will go with the consensus here.

I think this cleared up all my questions! Thank you for the help.