Helical polymers are found throughout biology and account for a substantial fraction of the protein in a cell. These filaments are very attractive for three-dimensional reconstruction from electron micrographs due to the fact that projections of these filaments show many different views of identical subunits in identical environments. However, ambiguities exist in defining the symmetry of a helical filament when one has limited resolution, and mistakes can be made. Until one reaches a near-atomic level of resolution, there are not necessarily reality-checks that can distinguish between correct and incorrect solutions. A recent paper in eLife (Xu et al., 2014) almost certainly imposed an incorrect helical symmetry and this can be seen using filament images posted by Xu et al. A comparison between the atomic model proposed and the published three-dimensional reconstruction should have suggested that an incorrect solution was found.https://doi.org/10.7554/eLife.04969.001
Helical polymers are ubiquitous in biology, and are found extensively in viruses (Ge and Zhou, 2011), bacteria (Polka et al., 2009; Ozyamak et al., 2013), eukaryotes (Galkin et al., 2010) and archaea (Yu et al., 2012). Because of the symmetry inherent in helical filaments (where identical subunits are rotated and translated along a filament axis) projections of helical filaments can provide all of the information needed to generate a three-dimensional reconstruction from a limited number of images, and this property was exploited in the first electron microscopic (EM) three dimensional reconstruction, which was made from a helical bacteriophage tail (DeRosier and Klug, 1968). New advances in both computational approaches (Egelman, 2000) and cryo-imaging with direct electron detectors (Bai et al., 2013) have meant that the determination of structures of helical polymers at near-atomic resolution (Lu et al., 2014; Wu et al., 2014) is becoming much more common. However, the assignment of the correct symmetry to a helical polymer can still be problematic (Egelman, 2010; Desfosses et al., 2014), and examples exist where the wrong symmetry has been mistakenly imposed (Okorokov et al., 2010; Sen et al., 2010; Yu and Egelman, 2010; Heymann et al., 2013). Insufficient understanding exists about how solutions (three-dimensional structures) may be obtained that are stable under iterative refinement, consistent with the images, but simply wrong. This arises from imposing a helical symmetry that is incorrect, but the available resolution does not allow one to distinguish between wrong symmetries and the correct one. This short communication seeks to reconcile the very different reconstructions of the MAVS filament by two groups (Wu et al., 2014; Xu et al., 2014) and in the process raise some general issues about helical reconstruction. While the points may appear to be quite technical to the non-specialist, they raise important questions about the expertise needed to evaluate some cryo-EM results.
The reconstruction of the MAVS filament presented in Wu et al. had a stated resolution of 3.6 Å and had a helical symmetry of a rotation of −101.1° (the negative rotation corresponding to a left-handed helix) and a rise of 5.1 Å per subunit. In contrast, the reconstruction of Xu et al. had a stated resolution of 9.6 Å, a C3 point group rotational symmetry, with a rotation of −53.6° and a rise of 16.8 Å for every ring of three subunits. One of the differences between the two reconstructions is that in Wu et al. not only are α-helices clearly resolved, but bulky aromatic side-chains can be seen that are consistent with not only the sequence but a crystal structure of the subunit which can be fit quite well as a rigid body into the reconstruction. In the Comment appended to their paper (Jiang, 2014), Xu et al. do not dispute the validity of the reconstruction in Wu et al., but argue that the filament being reconstructed by Wu et al. is an artifact of harsh preparative procedures, and that their filaments have a different symmetry and represent a more native conformation. They advance three arguments for why their filaments have a different symmetry. Given that the authors of Xu et al. have deposited the images that they used in the EMPIAR database, these arguments can be directly tested.
Their first argument for a different helical symmetry is that a layer line at ∼1/(17 Å) in their power spectrum has intensity on the meridian, while the corresponding layer line in Wu et al. does not and is clearly arising from an n = −1 Bessel order. Figure 1 shows how the power spectrum of an image is the central section of the three-dimensional power spectrum of the object being projected onto the image. As such, the projection is sensitive to any tilt of the axis of the object out of the plane of projection. An untilted central section (Figure 1B) would generate the two separated peaks on both sides of the meridian seen in Figure 1D (arrow), while a sufficiently tilted central section (Figure 1C) would generate a single intensity on the meridian (Figure 1E). For negatively stained polymers, where filaments are adsorbed to a carbon film, out-of-plane tilt can be largely ignored. In cryo-EM, where filaments are imaged within an ice film of finite thickness, such tilt cannot be ignored, particularly as one becomes more sensitive to it the higher the resolution. The first question I asked is whether there was any substantial out-of-plane tilt for the filaments imaged by Xu et al. I used an approach of providing references with different tilt angles to generate the histogram shown in Figure 2A. The large peaks at −20° and +20° correspond to segments with tilt angles greater than 20°, showing the unexpectedly large degree of tilt for these filaments. If one only takes the segments in the three central bins (−4°, 0°, +4°) a power spectrum is generated (Figure 2B) that clearly shows (arrow) an n = −1 layer line with no intensity on the meridian. This power spectrum looks to be the same as that shown by Wu et al., undercutting the argument that the filaments being used by the two groups are different. One can calculate the tilt that would be needed to make the two separated peaks in Figure 2B appear as a single peak on the meridian, and it is ∼9°, entirely consistent with the histogram in Figure 2A.
Is it possible that the sorting I have done for out-of-plane tilt is completely wrong, since it has involved the application of a helical symmetry, and the power spectrum in Figure 2B is actually from a highly tilted subset, while the power spectrum in Xu et al. is from an untilted subset? The answer is no, since we can see from Figure 1 how the 2D power spectrum will change with tilt. The two separated peaks seen in Figure 2B would also need to be present as secondary maxima in the power spectrum shown in Xu et al. to appear as a result of tilt, and they simply are not present there. On the other hand, the power spectrum in Figure 2B, if assumed to be representative of filaments with little tilt, can explain the power spectrum of Xu et al. when that is assumed to represent a large tilt.
The second argument advanced by Xu et al. in their Comment is that the symmetry used by Wu et al. is unstable in the Iterative Helical Real Space Reconstruction (IHRSR) approach when applied to their filaments, and therefore their filaments must have a different symmetry. Since I developed the IHRSR method (Egelman, 2000), I have some experience with the application of the algorithm to many helical systems. I have argued (Egelman, 2007, 2010) that a stable solution in IHRSR is a necessary, but not sufficient, requirement for the solution to be correct, since it has been clear that many wrong solutions can also be stable. However, it has become apparent in cryo-EM, when out-of-plane tilt cannot be ignored, that correct solutions can be unstable when this tilt is ignored. Figure 3 shows the symmetry parameters (rotation and axial rise per subunit) for the filaments of Xu et al. when references are included with out-of-plane tilt. It can be seen that the symmetry determined by Wu et al. can be applied to the filaments of Xu et al. and that this is quite stable in IHRSR cycles, again undercutting their argument that the two sets of filaments are different.
The third and last argument advanced by Xu et al. in their Comment is that their reconstruction has a hole in the center, while the model of Wu et al. that has been built into the 3.6 Å resolution reconstruction would have no such hole. They correctly state that such a hole can be seen in their filaments independently of the symmetry simply by cylindrically averaging their 3D density map. If one takes the cryo-EM density map deposited by Wu et al. (EMDB-5922) and filters it to 12 Å resolution a hole is seen in the center (Figure 4B, arrow). The reason that this is consistent with the deposited model (3j6j.PDB) is simply that the subunits are helically arranged and do not pack to form a solid core. Since the packing of protein is less dense in the center of the filament than it is at higher radius, a cylindrically averaged density distribution will show a continuous hole in the center of the filament. The reason that I have chosen 12 Å for this comparison is that this is the actual resolution of the reconstruction (Figure 4A,C) that I have been able to generate from the images of Xu et al. Why is the resolution so poor? The most obvious differences are that the Xu et al. images appear to be from thick ice, there is a very noisy background, and they have applied a carbon film which degrades the signal-to-noise ratio. Since they are sampling at 2.3 Å/px, the best one might hope to achieve would be ∼7 Å resolution (2/3 Nyquist) but given the problems with the images one would never get even close to this.
I have thus shown that the three arguments for why their filaments are different from those of Wu et al., advanced by Xu et al. in their Comment, are not true. This does not establish, however, that the filaments are the same. Establishing that would require a high resolution reconstruction from the filaments of Xu et al. which I fear would be impossible. But we can look at their published reconstruction, which gets to the most important point: given all of the potential problems and ambiguities in helical reconstruction, how does one know that a structure is correct? The great success of protein x-ray crystallography lies in the fact that structures are being solved that all have a known stereochemistry. One has prior knowledge about what an α-helix and a β-sheet look like, as well as knowing the amino acid sequence that must be built into a model. At lower resolution by EM most of these reality-checks are simply absent, and one needs to generate methods, such as the tilt-pair validation used for single-particle cryo-EM (Rosenthal and Henderson, 2003) that can distinguish between correct and incorrect reconstructions. As the field of cryo-EM is now rapidly moving to near-atomic resolution, seeing recognizable secondary structure provides a necessary and sufficient validation of a reconstruction. In the case of Wu et al. having a map with near-atomic detail, showing right-handed α-helices, confirmed not only the symmetry but the absolute hand of the reconstruction, whereas in the absence of this level of detail one must use means such as metal-shadowing or tomography to resolve the enantiomorphic ambiguity present. But at lower resolution one can still ask how well a reconstruction agrees with a model. Unfortunately, there is vanishingly little agreement between their deposited map (EMDB-5890) and their deposited model (PDB 3j6c).
I show a comparison between their actual reconstruction (Figure 5A,C), and their model filtered to 9.6 Å (Figure 5B,D), the stated resolution of their reconstruction. One sees at these higher thresholds than used in Xu et al. that there is actually no match between the features in one and in the other. A reasonable question is whether such comparisons are fair, since the atomic models are free from any noise, while the reconstruction may contain noise at the resolution limit. Several published examples suggest that this method is reasonable, and might be the best gauge of the actual reconstruction. In a 9 Å resolution reconstruction of an actin-cofilin filament a comparison with an atomic model filtered to this resolution (Supplementary figure 1D, Galkin et al., 2011) shows an excellent match. Similarly, a 7.5 Å resolution reconstruction of a naked actin filament shows an excellent match with an atomic model filtered to this resolution (Supplementary figure 2, Fujii et al., 2010).
This comparison between a map and a model can be made quantitatively, and Figure 5E shows a Fourier Shell Correlation (FSC) plot between the map and model of Xu et al., with a resolution estimate of ∼22 Å. Such a resolution suggests that the only agreement between the map and the model is for compact ‘blobs’ (which is what this small subunit looks like at a resolution of 22 Å) related by the same imposed helical symmetry. This type of comparison, had it been done by the authors, should have raised many questions. One is whether the helical symmetry imposed was actually correct. Another is whether the FSC, which is only a measure of internal consistency between two halves of the image set and not a measure of actual resolution, has any meaning in this instance since it yields such a different result between two reconstructions (each generated from half of the image set) and between the overall reconstruction and the model.
In conclusion, the arguments of Xu et al. for a different symmetry in their filaments than that in Wu et al. collapse completely when one actually examines the publicly deposited data.
The micrographs in set04 within the EMPIAR-10014 deposition (http://pdbe.org/empiar) from Xu et al. (2014) were used for analysis. Of the 581 images, the first 160 were used for filament extraction. The Contrast Transfer Function was measured using CTFFIND3 (Mindell and Grigorieff, 2003). For the 160 images, the mean defocus was 3.2 μ and the range was from 1.7 μ to 5.4 μ. The images were then multiplied by the theoretical CTF function to correct the phases and improve the SNR. From these 160 images, 1066 long filament boxes were cut using the e2helixboxer routine within EMAN2 (Tang et al., 2007). The SPIDER software package (Frank et al., 1996) was used for all further steps. Overlapping boxes 192 px long (2.3 Å/px) were cut from the long filament boxes using a shift of 6 px (97% overlap), generating 64,980 boxes. These were used for IHRSR (Egelman, 2000) which converged to a symmetry of −101.2° rotation with an axial rise of 5.3 Å, starting from a solid cylinder as an initial reference. Due to the very poor resolution (∼17 Å), an attempt was made to improve this by only using the 58 images with a defocus less than 3.0 μ, which yielded 15,624 overlapping boxes. Because segments with a very large out-of-plane tilt were excluded, the final reconstruction was generated from 8600 segments. The modulation of the amplitudes of the final reconstruction by the CTF (the phase reversals of the CTF were previously corrected when the images were multiplied by the CTF) was corrected by dividing the Fourier coefficients in the reconstruction by the sum of the squared CTFs, imposing a negative B-factor of 2000 to correct for the decay of the high frequencies, and filtering to 12 Å.
SPRING - an image processing package for single-particle based helical reconstruction from electron cryomicrographsJournal of Structural Biology 185:15–26.https://doi.org/10.1016/j.jsb.2013.11.003
The iterative helical real space reconstruction method: Surmounting the problems posed by real polymersJournal of Structural Biology 157:83–94.https://doi.org/10.1016/j.jsb.2006.05.015
SPIDER and WEB: processing and visualization of images in 3D electron microscopy and related fieldsJournal of Structural Biology 116:190–199.https://doi.org/10.1006/jsbi.1996.0030
Remodeling of actin filaments by ADF/cofilin proteinsProceedings of the National Academy of Sciences of USA 108:20568–20572.https://doi.org/10.1073/pnas.1110109108
Hydrogen-bonding networks and RNA bases revealed by cryo electron microscopy suggest a triggering mechanism for calcium switchesProceedings of the National Academy of Sciences of USA 108:9637–9642.https://doi.org/10.1073/pnas.1018104108
Three-dimensional structure of the toxin-delivery particle antifeeding prophage of Serratia entomophilaThe Journal of Biological Chemistry 288:25276–25284.https://doi.org/10.1074/jbc.M113.456145
Accurate determination of local defocus and specimen tilt in electron microscopyJournal of Structural Biology 142:334–347.https://doi.org/10.1016/S1047-8477(03)00069-8
The bacterial actin MamK: in vitro assembly behavior and filament architectureThe Journal of Biological Chemistry 288:4265–4277.https://doi.org/10.1074/jbc.M112.417030
The structure and assembly dynamics of plasmid actin AlfA imply a novel mechanism of DNA segregationJournal of Bacteriology 191:6219–6230.https://doi.org/10.1128/JB.00676-09
EMAN2: an extensible image processing suite for electron microscopyJournal of Structural Biology 157:38–46.https://doi.org/10.1016/j.jsb.2006.05.009
Helical filaments of human Dmc1 protein on single-stranded DNA: a cautionary taleJournal of Molecular Biology 401:544–551.https://doi.org/10.1016/j.jmb.2010.06.049
Filaments from Ignicoccus hospitalis show diversity of packing in proteins containing n-Terminal type IV pilin helicesJournal of Molecular Biology 422:274–281.https://doi.org/10.1016/j.jmb.2012.05.031
Wesley I SundquistReviewing Editor; University of Utah, United States
eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.
Thank you for sending your work entitled “Ambiguities in Helical Reconstruction” for consideration at eLife. Your article has been favorably evaluated by John Kuriyan (Senior editor), Wes Sundquist (Reviewing editor), and three reviewers, of whom Niko Grigorieff and Carsten Sachse have agreed to reveal their identity. A third reviewer remains anonymous.
The Reviewing editor and the three reviewers discussed their comments before we reached this decision, and the Reviewing editor has assembled the following comments to help you prepare a revised submission.
Dr Egelman analyzes the validity of a cryoEM helical reconstruction of the prion-like filament formed by the CARD domain of the RIG-I adaptor molecule MAVS, as reported by Xu et al. in eLife (Xu et al. eLife, 2014, 3:e01489). The study by Xu et al. was published before a related study was published by Wu et al. in Molecular Cell (Wu et al. Molecular Cell 2014, 55:511-524). The situation merits follow-up because the two reported MAVS structures are entirely different. There is no dispute that the Wu et al. study contained high quality EM data, that their reconstruction was technically correct, and that they successfully generated a near-atomic resolution structure of their MAVS filaments. The central question is whether differences in sample preparation generated distinct MAVS filaments with different physical structures, or whether the differences in the reported structures instead reflect methodological problems in the Xu et al study. Dr Jiang (one of the corresponding authors on the Xu et al. paper) has posted a Comment that accompanies the original eLife manuscript in which he argues that the reported differences in the filament helical symmetries and structures reflect genuine physical differences in the MAVS filaments generated by the two groups. Here, Egelman analyzes the central claims of that Comment and also discusses the validity of the original Xu et al. reconstruction.
Egelman downloaded part of the original data deposited by Xu et al. in the EMPIAR data base and performed image analyses to test the arguments put forth by Jiang that the filaments used in the two studies were indeed different. He shows convincingly that the three central pieces of evidence presented in the Jiang Comment are invalid. He makes several important points, including that Xu et al. did not appreciate that their filaments had significant out-of-plane tilt and that convergence stability in a symmetry search is not sufficient to establish the validity of that symmetry. Importantly, he also shows that the Xu et al. structure does not agree with its own PDB model beyond a resolution of ∼20 Å, suggesting that the computed EM density is not correct. These arguments convincingly rebut the points raised in Jiang's Comment, reveal potential inconsistencies in the processing procedure used by Xu et al., and raise substantial doubts about the validity of the Xu et al. reconstruction. These analyses are therefore an important contribution.
The Egelman analysis could be strengthened even further by providing additional evidence that Xu et al. must have imposed the wrong helical symmetry. The author could do this by computing a structure with the symmetry parameters from Wu et al. imposed on the EMPIAR data set from Xu et al. (which he has already been analyzing). Such a structure should contain clearer densities and recognizable secondary structure features.https://doi.org/10.7554/eLife.04969.007
[…] The Egelman analysis could be strengthened even further by providing additional evidence that Xu et al. must have imposed the wrong helical symmetry. The author could do this by computing a structure with the symmetry parameters from Wu et al. imposed on the EMPIAR data set from Xu et al. (which he has already been analyzing). Such a structure should contain clearer densities and recognizable secondary structure features.
I have been able to improve the resolution that I initially obtained with their images by excluding those with a defocus greater than 3.0μ. This reduced the data set from ∼64k segments to ∼15k segments, but the reconstruction is clearly improved. I have expanded Figure 4 to show a better comparison of the reconstruction from Xu et al. with the correct symmetry with the one from Wu et al. filtered to 12 Å resolution, and shown the atomic model of Wu et al. fit into both volumes.https://doi.org/10.7554/eLife.04969.008
- Edward H Egelman
The funder had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
This work has been supported by NIH EB001567. I thank Albina Orlova for assistance with particle picking.
- Wesley I Sundquist, University of Utah, United States
- Received: September 30, 2014
- Accepted: November 17, 2014
- Version of Record published: December 8, 2014 (version 1)
© 2014, Egelman
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
The chloride-proton exchanger CLC-7 plays critical roles in lysosomal homeostasis and bone regeneration and its mutation can lead to osteopetrosis, lysosomal storage disease and neurological disorders. In lysosomes and the ruffled border of osteoclasts, CLC-7 requires a b-subunit, OSTM1, for stability and activity. Here we present electron cryomicroscopy structures of CLC-7 in occluded states by itself and in complex with OSTM1, determined at resolutions up to 2.8 Å. In the complex, the luminal surface of CLC-7 is entirely covered by a dimer of the heavily glycosylated and disulfide-bonded OSTM1, which serves to protect CLC-7 from the degradative environment of the lysosomal lumen. OSTM1 binding does not induce large-scale rearrangements of CLC-7, but does have minor effects the conformation of the ion-conduction pathway, potentially contributing to its regulatory role. These studies provide insights into the role of OSTM1 and serve as a foundation for understanding the mechanisms of CLC-7 regulation.