CTFFIND5 provides improved insight into quality, tilt and thickness of TEM samples
Curation statements for this article:-
Curated by eLife
eLife assessment
This valuable work presents the latest version of CTFFIND, which is the most popular software for determination of the contrast transfer function (CTF) in cryo-electron microscopy. CTFFIND5 estimates and considers acquisition geometry and sample thickness, which leads to improved CTF determination. The paper describes convincing evidence that CTFFIND5 finds better CTF parameters than previous methods, in particular for tilted samples (e.g. for cryo-electron tomography) or where thickness is an issue (e.g. cellular samples, or electron microscopy at low voltages).
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Structural Biology and Molecular Biophysics (eLife)
- Reading list (BiophysicsColab)
Abstract
Images taken by transmission electron microscopes are usually affected by lens aberrations and image defocus, among other factors. These distortions can be modeled in reciprocal space using the contrast transfer function (CTF). Accurate estimation and correction of the CTF is essential for restoring the high-resolution signal in an image and has been one of the key aspects of the “resolution revolution” in cryogenic electron microscopy (cryoEM). Previously, we described the implementation of algorithms for this task in the cis TEM software package (Grant et al ., 2018). Here we show that taking sample characteristics, such as thickness and tilt, into account can improve CTF estimation. This is particularly important when imaging cellular samples, where measurement of sample thickness and geometry derived from accurate modeling of the Thon ring pattern helps judging the quality of the sample. This improved CTF estimation has been implemented in CTFFIND5, a new version of the cis TEM program CTFFIND. We evaluated the accuracy of these estimates using images of tilted aquaporin crystals and eukaryotic cells thinned by focused ion beam milling. We estimate that with micrographs of sufficient quality CTFFIND5 can measure sample tilt with an accuracy of 3° and sample thickness with an accuracy of 5 nm.
Article activity feed
-
eLife assessment
This valuable work presents the latest version of CTFFIND, which is the most popular software for determination of the contrast transfer function (CTF) in cryo-electron microscopy. CTFFIND5 estimates and considers acquisition geometry and sample thickness, which leads to improved CTF determination. The paper describes convincing evidence that CTFFIND5 finds better CTF parameters than previous methods, in particular for tilted samples (e.g. for cryo-electron tomography) or where thickness is an issue (e.g. cellular samples, or electron microscopy at low voltages).
-
Reviewer #1 (Public Review):
This work presents CTFFIND5, a new version of the software for determination of the Contrast Transfer Function (CTF) that models the distortions introduced by the microscope in cryoEM images. CTFFIND5 can take acquisition geometry and sample thickness into consideration to improve CTF estimation.
To estimate tilt (tilt angle and tilt axis), the input image is split into tiles and correlation coefficients are computed between their power spectra and a local CTF model that includes the defocus variation according to a tilted plane. As a final step, by applying a rescaling factor to the power spectra of the tiles, an average tilt-corrected power spectrum is obtained and used for diagnostic purposes and to estimate the goodness of fit. This global procedure and the rescaling factor resemble those used in Bsoft, …
Reviewer #1 (Public Review):
This work presents CTFFIND5, a new version of the software for determination of the Contrast Transfer Function (CTF) that models the distortions introduced by the microscope in cryoEM images. CTFFIND5 can take acquisition geometry and sample thickness into consideration to improve CTF estimation.
To estimate tilt (tilt angle and tilt axis), the input image is split into tiles and correlation coefficients are computed between their power spectra and a local CTF model that includes the defocus variation according to a tilted plane. As a final step, by applying a rescaling factor to the power spectra of the tiles, an average tilt-corrected power spectrum is obtained and used for diagnostic purposes and to estimate the goodness of fit. This global procedure and the rescaling factor resemble those used in Bsoft, Warp, etc, with determination of the tilt parameters being a feature specific of CTFFIND5 (and formerly CTFTILT). The performance of the algorithm is evaluated with tilted 2D crystals and tilt-series, demonstrating accurate tilt estimation in some cases and some limitations in others. Further analysis of CTF determination with tilt-series, particularly showing whether there is accurate or stable estimation at high tilts, might be helpful to show the robustness of CTFFIND5 in cryoET.
CTFFIND5 represents the first CTF determination tool that considers the thickness-related modulation envelope of the CTF firstly described by McMullan et al. (2015) and experimentally confirmed by Tichelaar et al. (2020). To this end, CTFFIND5 uses a new CTF model that takes the sample thickness into account. CTFFIND5 thus provides more accurate CTF estimation and, furthermore, gives an estimation of the sample thickness, which may be a valuable resource to judge the potential for high resolution. To evaluate the accuracy of thickness estimation in CTFFIND5, the authors use the Lambert-Beer law on energy-filtered data and also tomographic data, thus demonstrating that the estimates are reasonable for images with exposure around 30 e/A2. While consideration of sample thickness in CTF determination sounds ideally suited for cryoET, practical application under the standard acquisition protocols in cryoET (exposure of 3-5 e/A2 per image) is still limited. In this regard, the authors are honest in the conclusions and clearly identify the areas where thickness-aware CTF determination will be valuable at present: e.g. in situ single particle analysis and in vitro single particle cryoEM of purified samples at low voltages.
In conclusion, the manuscript introduces novel methods inside CTFFIND5 that improve CTF estimation, namely acquisition geometry and sample thickness. The evaluation demonstrates the performance of the new tool, with fairly accurate estimates of tilt axis, tilt angle and sample thickness and improved CTF estimation. The manuscript critically defines the current range of application of the new methods in cryoEM.
-
Reviewer #2 (Public Review):
Summary:
This paper describes the latest version of the most popular program for CTF estimation for cryo-EM images: CTFFIND5. New features in CTFFIND5 are the estimation of tilt geometry, including for samples, like FIB-milled lamellae, that are pre-tilted along a different axis than the tilt axis of the tomographic experiment, plus the estimation of sample thickness from the expanded CTF model described by McMullan et al (2015). The results convincingly show the added value of the program for thicker and tilted images, such as are common in modern cryo-ET experiments. The program will therefore have a considerable impact on the field.
I have only minor suggestions for improvement below:
Abstract: "[CTF estimation] has been one of the key aspects of the resolution revolution"-> This is a bit over the top. …
Reviewer #2 (Public Review):
Summary:
This paper describes the latest version of the most popular program for CTF estimation for cryo-EM images: CTFFIND5. New features in CTFFIND5 are the estimation of tilt geometry, including for samples, like FIB-milled lamellae, that are pre-tilted along a different axis than the tilt axis of the tomographic experiment, plus the estimation of sample thickness from the expanded CTF model described by McMullan et al (2015). The results convincingly show the added value of the program for thicker and tilted images, such as are common in modern cryo-ET experiments. The program will therefore have a considerable impact on the field.
I have only minor suggestions for improvement below:
Abstract: "[CTF estimation] has been one of the key aspects of the resolution revolution"-> This is a bit over the top. Not much changed in the actual algorithms for CTF estimation during the resolution revolution.
L34: "These parameters" -> Cs is typically given, only defocus (and if relevant phase shift) are estimated.
L110-116: The text is ambiguous: are rotations defined clockwise or counter-clockwise? It would be good to explicitly state what subsequent rotations, in which directions and around which axes this transformation matrix (and the input/output angles in CTFFIND5) correspond to.
L129-130: As a suggestion: it would be relatively easy, and possibly beneficial to the user, to implement a high-resolution limit that varies with the accumulated dose on the sample. One example of this exists in the tomography pipeline of RELION-5.
Substituting Eq (7) into Eq (6) yields ksi=pi, which cannot be true. If t is the sample thickness, then how can this be a function of the frequency g of the first node of the CTF function? The former is a feature of the sample, the latter is a parameter of the optical system. This needs correction. -
Reviewer #3 (Public Review):
In this manuscript, the authors detail improvements in the core CTFFIND (CTFFIND5 as implemented in cisTEM) algorithm that better estimates CTF parameters from titled micrographs and those that exhibit signal attenuation due to ice thickness. These improvements typically yield more accurate CTF values that better represent the data. Although some of the improvements result in slower calculations per micrograph, these can be easily overcome through parallelization.
There are some concerns outlined below that would benefit from further evaluation by the authors.
For the examples shown in Figure 3b, given the small differences in estimated defocus1 and 2, what type of improvements would be expected in the reconstructed tomograms? Do such improvements in estimates manifest in better tilt-series reconstruction?
Si…
Reviewer #3 (Public Review):
In this manuscript, the authors detail improvements in the core CTFFIND (CTFFIND5 as implemented in cisTEM) algorithm that better estimates CTF parameters from titled micrographs and those that exhibit signal attenuation due to ice thickness. These improvements typically yield more accurate CTF values that better represent the data. Although some of the improvements result in slower calculations per micrograph, these can be easily overcome through parallelization.
There are some concerns outlined below that would benefit from further evaluation by the authors.
For the examples shown in Figure 3b, given the small differences in estimated defocus1 and 2, what type of improvements would be expected in the reconstructed tomograms? Do such improvements in estimates manifest in better tilt-series reconstruction?
Similarly, the data shown in Figure 3C shows minimal improvements in the CTF resolution estimate (e.g., 4.3 versus 4.2 Å), but exhibited several hundred Å difference in defocus values. How do such differences impact downstream processing? Is such a difference overcame by per-particle (local) CTF refinements (like the authors mention in the discussion, see below)?
At which point does the thickness of the specimen preclude the ice thickness modulation to be included for "accurate" estimate? 500Å? 1000Å? 2000Å? Based on the data shown in Figure 3B, as high as 969 Å thick specimens benefit moderately (4.6 versus 3.4 Å fit estimate), but perhaps not significantly, from the ice thickness estimation. Considering the increased computational time for ice thickness estimation, such an estimate of when to incorporate for single-particle workflows would be beneficial.
It would seem that this statement could be evaluated herein: "the analysis of images of purified samples recorded at lower acceleration voltages, e.g., 100 keV (McMullan et al., 2023), may also benefit since thickness-dependent CTF modulations will appear at lower resolution with longer electron wavelengths". There are numerous examples of 300kV, 200kV, and 100kV EMPIAR datasets to be compared and recommendations would be welcomed.
Although logical, this statement is not supported by the data presented in this manuscript: "The improvements of CTFFIND5 will provide better starting values for this refinement, yielding better overall CTF estimation and recovery of high-resolution information during 3D reconstruction."
Moreso, the lack of single-particle data evaluation does present a concern. Naively, these improvements would benefit all cryoEM data, regardless of modality.
-
