A Bayesian approach to single-particle electron cryo-tomography in RELION-4.0

Curation statements for this article:
  • Curated by eLife

    eLife logo

    eLife assessment

    Single-particle tomography (SPT) is a useful method to determine the structure of proteins imaged in situ. This important work presents an easy-to-use tool for SPT that approximates the use of 2D tomographic projections using a "pseudo-subtomogram" data structure, chosen to facilitate implementation within the existing Relion codebase. The examples shown provide solid support for the claims about the efficacy of the approach.

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

We present a new approach for macromolecular structure determination from multiple particles in electron cryo-tomography (cryo-ET) data sets. Whereas existing subtomogram averaging approaches are based on 3D data models, we propose to optimise a regularised likelihood target that approximates a function of the 2D experimental images. In addition, analogous to Bayesian polishing and contrast transfer function (CTF) refinement in single-particle analysis, we describe the approaches that exploit the increased signal-to-noise ratio in the averaged structure to optimise tilt-series alignments, beam-induced motions of the particles throughout the tilt-series acquisition, defoci of the individual particles, as well as higher-order optical aberrations of the microscope. Implementation of our approaches in the open-source software package RELION aims to facilitate their general use, particularly for those researchers who are already familiar with its single-particle analysis tools. We illustrate for three applications that our approaches allow structure determination from cryo-ET data to resolutions sufficient for de novo atomic modelling.

Article activity feed

  1. eLife assessment

    Single-particle tomography (SPT) is a useful method to determine the structure of proteins imaged in situ. This important work presents an easy-to-use tool for SPT that approximates the use of 2D tomographic projections using a "pseudo-subtomogram" data structure, chosen to facilitate implementation within the existing Relion codebase. The examples shown provide solid support for the claims about the efficacy of the approach.

  2. Reviewer #1 (Public Review):

    This study presents an implementation of single-particle tomography within the Bayesian framework of the Relion software package. Similar to previously proposed strategies, the approach leverages single-particle analysis tools and tomographic geometric constraints to improve map resolution. Results on the EMPIAR-10164 benchmark dataset appear to match the performance of previous methods, but no maps were made available or deposited, and no direct comparisons with previous results are shown. Consistent with previously published strategies that use 2D projections instead of sub-volumes, the approach performs favorably in terms of resolution when compared to traditional subvolume averaging.

    Strengths

    - Use of a Bayesian framework for image refinement and reconstruction requires less parameter tweaking.
    - By making the new implementation accessible through a GUI already familiar to many SPA users, this tool will make SPT easier to use.
    - The implementation of 3D classification could be potentially beneficial to study sample heterogeneity in situ.
    - In cases where high resolution can be achieved (better than 3A), the approach has the potential to correct for higher-order optical aberrations.
    - Using two cryo-ET datasets, resolution improvements are shown over traditional subvolume averaging (as implemented in the AV3 Matlab suite of programs [Forster et al., 2007] and Dynamo suite [Castano-Diez, 2012]).

    Weaknesses

    - The approach recapitulates previously proposed strategies for SPT refinement that use raw tomographic projections instead of sub-volumes to improve resolution. Strategies that leverage the increased SNR of average structures to optimize particle pose and deformation, tilt-series alignment, and CTF refinement, were proposed and validated in earlier studies [1,24].
    - Compared to end-to-end pipelines for tomography data analysis such as EMAN2 and Dynamo, this approach only implements the subtomogram averaging step, while still relying on external tools for initial tilt-series alignment, CTF estimation, and particle picking.
    - In terms of performance, the HIV-1 Gag maps obtained from the benchmark dataset EMPIAR-10164 do not represent an improvement in resolution over previous methods.
    - No validation is provided to support the claim that the tool can correct for higher-order optical aberrations of the microscope from cryo-ET data.
    - No results are provided to validate the 3D classification routines to study heterogeneity, and no experiments are shown to support the claim that the new approach is more accurate than previous sub-volume classification strategies that compensate for the missing wedge (such as the approach implemented in the earlier version of Relion [4]).

    Overall, this implementation of SPT would be a valuable resource for the cryo-ET community.

  3. Reviewer #2 (Public Review):

    Zivanov et al. present a new approach for multi-particle averaging from cryo-electron tomography data. They propose that refining directly against 2D tilt series images instead of the traditional reconstructed 3D subtomograms would simplify and improve structure determination. This would represent the experimental data more faithfully than traditional subtomogram averaging and circumvents the need for missing wedge correction. The authors describe a data structure termed 'pseudosubtomograms' where the tilt images are represented as their Fourier transform pre-multiplied with the CTF, accompanied by an array describing how often each 3D-voxel has been observed and the sum of the squared CTF. They then present a new regularized likelihood target function for cryo-ET particle alignment which uses the pseudosubtomograms data structure. This approach is implemented within the general RELION refinement framework and allows for the use of pseudosubtomograms for 3D classification, initial model generation, and 3D refinement.

    The authors also introduce methods for refining optical and geometrical parameters in the tilt series taking advantage of the average map obtained after 3D refinement. This allows for more accurate tilt series alignment, per-particle motion tracking, and calculation of per-particle CTF. They propose that iteratively refining these parameters, extracting new pseudosubtomograms, and realigning the particles should lead to more accurate structure determination. The methods are validated using three different datasets, and the authors show that the iterative refinement within their framework increases the resolution of the 3D reconstruction and that the resulting maps are resolved to the same or better resolution than previously published methods.

    The introduction of a more direct representation of the 2D tilt series images is a novel approach to subtomogram averaging, and the authors show that it is as good or better than current approaches. Comparing the subtomogram average to the tilt series to correct for optical and geometrical parameters of the data has already been implemented in the program M. Here, the authors show that their algorithms can reach the same resolution as M for the HIV immature capsid, but discuss that M might be superior at very high resolution, as it models beam-induced rotation of particles. Nevertheless, the new approaches are implemented in a single framework - the popular open-source software package RELION - thereby greatly facilitating their accessibility to uses. This is a very welcome contribution and development in the field.