METHOD An image processing pipeline for electron cryo-tomography in RELION-5 Alister Burt1,2 , Bogdan Toader1 , Rangana Warshamanage3,4 , Andriko von Kügelgen1 , Euan Pyle5,6,* , Jasenko Zivanov1 , Dari Kimanius1,† , Tanmay A. M. Bharat1 and Sjors H. W. Scheres1 1 MRC Laboratory of Molecular Biology, Cambridge Biomedical Campus, Cambridge, UK 2 Department of Structural Biology, Genentech, South San Francisco, CA, USA 3 CCP-EM, Scientific Computing Department, UKRI Science and Technology Facilities Council, Harwell Campus, Didcot, UK 4 Department of Psychiatry, University of Pittsburgh, Pittsburgh, PA, USA 5 Institute of Structural and Molecular Biology, Birkbeck College, London, UK 6 The Francis Crick Institute, London, UK Keywords electron microscopy; electron tomography; image processing; software Correspondence S. H. W. Scheres, MRC Laboratory of Molecular Biology, Cambridge Biomedical Campus, Cambridge, UK E-mail: scheres@mrc-lmb.cam.ac.uk Present address *European Molecular Biology Laboratory, Heidelberg, Germany †CZ Imaging Institute, Redwood City, CA, USA Alister Burt and Bogdan Toader contributed equally to this article. (Received 30 April 2024, revised 20 June 2024, accepted 16 July 2024) doi:10.1002/2211-5463.13873 Edited by Cláudio Soares Electron tomography of frozen, hydrated samples allows structure determi- nation of macromolecular complexes that are embedded in complex envi- ronments. Provided that the target complexes may be localised in noisy, three-dimensional tomographic reconstructions, averaging images of multi- ple instances of these molecules can lead to structures with sufficient reso- lution for de novo atomic modelling. Although many research groups have contributed image processing tools for these tasks, a lack of standardisa- tion and interoperability represents a barrier for newcomers to the field. Here, we present an image processing pipeline for electron tomography data in RELION-5, with functionality ranging from the import of unpro- cessed movies to the automated building of atomic models in the final maps. Our explicit definition of metadata items that describe the steps of our pipeline has been designed for interoperability with other software tools and provides a framework for further standardisation. In the electron cryo-tomography (cryo-ET) approach, a three-dimensional (3D) reconstruction called a tomo- gram is calculated from a series of images that are taken as a sample is rotated around a tilt axis in the electron microscope. Because radiation damage limits the dose that can be applied to the specimen, tomograms typically suffer from large amounts of noise. Moreover, samples are typically thin, slab-like sections, which dictates that some views of the sample cannot be acquired, leading to artefacts in the recon- structed tomogram. Provided that the structures of interest, or particles, can be localised in the noisy Abbreviations CTF, Contrast Transfer function; ET, Electron tomography; FSC, Fourier Shell Correlation; GUI, Graphical User Interface; MPI, Message Passing Interface; MRC, Medical Research Council; 2D, Two-dimensional; 3D, Three-dimensional. 1FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. https://orcid.org/0000-0002-9341-2295 https://orcid.org/0000-0002-9341-2295 https://orcid.org/0000-0002-9341-2295 https://orcid.org/0000-0001-5444-2179 https://orcid.org/0000-0001-5444-2179 https://orcid.org/0000-0001-5444-2179 https://orcid.org/0000-0001-9756-0001 https://orcid.org/0000-0001-9756-0001 https://orcid.org/0000-0001-9756-0001 https://orcid.org/0000-0002-0017-2414 https://orcid.org/0000-0002-0017-2414 https://orcid.org/0000-0002-0017-2414 https://orcid.org/0000-0002-4633-4917 https://orcid.org/0000-0002-4633-4917 https://orcid.org/0000-0002-4633-4917 https://orcid.org/0000-0001-8407-0759 https://orcid.org/0000-0001-8407-0759 https://orcid.org/0000-0001-8407-0759 https://orcid.org/0000-0002-2662-6373 https://orcid.org/0000-0002-2662-6373 https://orcid.org/0000-0002-2662-6373 https://orcid.org/0000-0002-0168-0277 https://orcid.org/0000-0002-0168-0277 https://orcid.org/0000-0002-0168-0277 https://orcid.org/0000-0002-0462-6540 https://orcid.org/0000-0002-0462-6540 https://orcid.org/0000-0002-0462-6540 mailto:scheres@mrc-lmb.cam.ac.uk http://creativecommons.org/licenses/by/4.0/ tomograms and that they can be brought into register, many instances of a particle can be combined into a single 3D reconstruction with an increased signal-to- noise ratio. This approach, commonly referred to as subtomogram averaging, has been used to calculate structures of macromolecular complexes to resolutions sufficient for de novo atomic modelling for macromole- cules inside cells.1–3 Many tools exist for acquiring and analysing cryo-ET data. SerialEM,4,5 one of the most popular softwares for tilt series acquisition, saves acquisition-related meta- data in text files with the mdoc extension. Tomo5, a commercial software provided by Thermo Fisher, also writes metadata in the mdoc format. Accurate 3D reconstruction from a tilt series requires determining the parameters of a projection model describing the posi- tion, orientation and possibly also deformation of the sample in the electron microscope. The IMOD software package6,7 provides some of the most popular tools for this task, which is commonly referred to as tilt series alignment. Recently, AreTomo8 has also gained popu- larity by providing robust and automated tools for tilt series alignment, particularly for in situ data without fiducial markers. Once the parameters of a projection model are determined, a tomogram may be recon- structed from tilt images. Again, many approaches exist, including weighted back-projection,9 simultaneous itera- tive reconstruction technique,10–12 algebraic reconstruc- tion technique13–15 and compressed sensing.16 The reconstruction method used depends on what the tomo- gram, or subvolumes thereof, will be used for. For example, tomograms with high contrast at lower spatial frequencies are often used for direct visual interpretation and segmentation, while for subtomogram averaging, high-resolution features need to be preserved in the reconstruction and lower resolution features may be less pronounced. Direct reconstruction of local regions of interest in tomograms1,17–20 has obviated the need for calculating large, high-resolution tomograms from which subvolumes are cropped. To assist in feature identifica- tion, deep-learning approaches have gained traction for denoising21–24 and template matching.25,26 Despite the advent of near-complete cryo-ET image processing workflows, like those implemented in IMOD7, PEET,27 EMAN2,28 Dynamo,29 STOPGAP,30 TomoBEAR,31 SCIPION32,33 or NextPyp,19 a lack of standardisation represents a hurdle to newcomers and makes it difficult to effectively use the best parts of each package. Here, we present a cryo-ET image processing pipeline inside the free, open-source software RELION-5. Closely mirroring similar procedures in single-particle analysis, our cryo-ET pipeline starts from unprocessed movies and acquisition metadata in mdoc files and ends with high-resolution 3D reconstructions and the auto- mated building of atomic models.34 The pipeline pro- vides wrappers to CTFFIND435 for estimation of contrast transfer function (CTF) parameters; to IMOD7 and AreTomo8 for tilt series alignment; and to cryoCARE22 for tomogram denoising. We present graphical tools for the manual curation of images in tilt series and for the picking of particles in reconstructed tomograms. New to RELION-5 is the option to write out CTF-premultiplied 2D stacks of subimages that are cropped from the tilt series at the corresponding posi- tions for each particle. 2D particle stacks can then be used in the same alignment and classification procedures that were introduced in RELION-4.18 Refinement of 2D particle stacks provides major computational advan- tages, both in disc space requirements and in processing speed, compared with methods that reconstruct 3D sub- volumes, such as the pseudo-subtomogram approach introduced in RELION-4.18 We also provide an explicit definition of the metadata required for and generated in our pipeline, providing a framework for further standar- disation that will also be adopted in the CCP-EM software36,37 (Tom Burnley, personal communication). Materials and methods All new developments in the RELION-5 tomography pipe- line are accessible from the graphical user interface (GUI; Fig. 1), which can be launched from the command line using the command ‘relion --tomo’. The side bar in the top left of the GUI shows the names of all job types. Below, we describe all jobs that have new developments for cryo-ET in RELION-5. Importing into RELION’s data model Starting from directories with the unprocessed movie files and their corresponding mdoc files, the ‘Import’ job in the cryo-ET pipeline of RELION-5 writes metadata files that describe the acquired tilt series in the STAR format.38 A primary tilt_series.star file (Fig. 2; Table 1) con- tains a global table with one line per tilt series. The global table provides general information, including the accelera- tion voltage, pixel size, amplitude contrast and spherical aberration, plus a reference to a STAR file that contains information about individual images in each tilt series (Fig. 3; Table 2). Tilt series STAR files are stored in the tilt_series/ subdirectory of the Import job directory. They contain a table with one line for each image of the tilt series that stores the name of the unprocessed movie, the nominal tilt angle of the stage, the nominal orientation of the tilt axis, the nominal defocus and the accumulated elec- tron dose (Fig. 3; black). 2 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. One option that the user needs to decide when importing tilt series is the defocus handedness (rlnTomoHand in Fig. 2; Table 1), which defines the direction in which defocus changes while moving across the tilt axis in tilted images. In particular, when rlnTomoHand is set to 1, a particle with positive X coordinate (in the 3D space) will move closer to focus and a particle with negative X coordinate will move away from focus as the tilt angle of the specimen is increased, while the opposite effect is achieved when rlnTo- moHand is set to �1. Although the defocus handedness could be estimated automatically during CTF estimation,17,39,40 for now, it needs to be determined by the user using trial and error. For all Thermo Fisher Krios and Glacios microscopes that we have tested so far, the rlnTomoHand was always Fig. 1. The RELION-5 tomography GUI. The sidebar on the top left provides access to all job types. The top right panel provides multiple tabs with input parameters for each job type. The bottom half of the GUI provides an overview of finished, running and scheduled jobs; how different jobs relate to each other through their input and output; and the output from the currently selected job. 3FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 �1, which is the default option on the GUI (Yes, to invert defocus handedness). Motion correction The tilt_series.star file from the ‘Import’ job can be used as input for a ‘Motion correction’ job. Internally, all images of all tilt series are combined into a single list of images that can then be processed in parallel using a combination of message passing interface (MPI) and threads. Either UCSF MotionCor241 or RELION’s own implementation of that algorithm42 may be used for movie frame alignment. An option exists to write out summed micrographs that are calcu- lated only from the even or the odd movie frames, which may be used for tomogram denoising as described below. Although an option exists to bin images at this stage, this option will limit the achievable resolution of all downstream steps. Because there is no option for un-binning (i.e. reverting to the original pixel size) at a later step, binning is only recommended if the images were recorded in super-resolution mode and the resolution of the average is not expected to reach the physical Nyquist frequency. Fig. 2. Tomograms metadata description. Metadata to describe a set of tomograms is stored in the STAR format. Labels and data columns shown in black are added during import; labels and columns in violet are added during tomogram reconstruction. Corresponding definitions of metadata labels are given in Table 1. Table 1. Metadata items for tomograms. Metadata labels, types and definitions for the description of a set of tomograms. Entries in bold are compulsory for sub-tomogram averaging; entries in regular are not necessary for sub-tomogram averaging. Label Type Definition _rlnTomoName String Name of a tomogram _rlnVoltage Float Acceleration voltage (in kV) _rlnSphericalAberration Float Spherical aberration (in mm) _rlnAmplitudeContrast Float Amplitude contrast ratio _rlnMicrographOriginalPixelSize Float Pixel size of the original movie, before binning in motion correction (in Å) _rlnTomoHand Integer Tomogram handedness, i.e. sign of the slope of defocus over the Z-coordinate (�1 or 1) _rlnOpticsGroupName String Name of an optics group _rlnTomoTiltSeriesPixelSize Float Pixel size of the tilt series, after binning in motion correction (in Å) _rlnTomoTiltSeriesStarFile String Path of the STAR file with metadata about the individual tilt series images _rlnEtomoDirectiveFile String Path of the ETomo directive file from tilt series alignment _rlnTomoTomogramBinning Float Binning of the tomogram reconstruction _rlnTomoSizeX Integer Width of the tomogram in unbinned voxels (i.e. of size _rlnTomoTiltSeriesPixelSize) _rlnTomoSizeY Integer Length of the tomogram in unbinned voxels (i.e. of size _rlnTomoTiltSeriesPixelSize) _rlnTomoSizeZ Integer Height of the tomogram in unbinned voxels (i.e. of size _rlnTomoTiltSeriesPixelSize) _rlnTomoReconstructedTomogramHalf1 String Path of a tomogram reconstructed from even movie frames only _rlnTomoReconstructedTomogramHalf2 String Path of a tomogram reconstructed from odd movie frames only 4 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. The motion correction job adds several columns to the tables in the STAR files of the individual tilt series, includ- ing the name of the motion-corrected micrograph and sta- tistics about the estimated movements in each of the tilt series images (Fig. 3; red). CTF estimation The tilt_series.star file from the ‘Motion correction’ job can be used as input for a ‘CTF estimation’ job, where individual images of all tilt series are again processed in parallel using MPI. CTF estimation is performed using a wrapper to CTFFIND4,35 with two modifications from the equivalent approach for single-particle analysis that increase robustness for images at high tilt angles, where the signal to noise ratio is low. First, whereas the same minimum and maxi- mum defocus values are used for all micrographs in single-particle analysis, in the tomography pipeline the user may provide a search range that will be used around the nomi- nal defocus value for each image. Second, the maximum reso- lution used for CTF estimation may be varied with the accumulated electron dose for each image in the tilt series. Fig. 3. Tilt series images metadata description. Metadata to describe a set of images in a tilt series is stored in the STAR format. Labels and data columns shown in black are added during import; labels and columns in red are added during motion correction; labels and columns in green are added during CTF estimation; labels and columns in blue are added during tilt series alignment; labels and columns in violet are added during tomogram reconstruction. Corresponding definitions of metadata labels are given in Table 2. 5FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 This job again adds information to the STAR files of the individual tilt series, including the estimated defocus values, a figure-of-merit for the CTF fit, and the maxi- mum resolution to which a good fit was obtained for each image of the tilt series (Fig. 3; green). The values for rlnCtfIceRingDensity are calculated as the power of the tilt series images between 0.25 and 0.28 Å�1.43 Selection of tilt series images Given a tilt_series.star file, the ‘Exclude tilt-images’ job launches a Napari-based44,45 viewer that displays the images in a tilt series and allows the user to de-select unsuitable images, for example images in which the field of view has shifted or is obstructed by the grid bars, or if there is significant radiation induced move- ment detected. A screenshot of the viewer is shown in Fig. 4A. The output selected_tilt_series.star file will point towards individual tilt series STAR files that no lon- ger contain lines with de-selected images. Tilt series alignment The ‘Align tilt series’ job implements wrappers for perform- ing tilt series alignment in IMOD7 and AreTomo.8 In IMOD, fiducial markers or local image patches are tracked through a tilt series yielding observed 2D positions of a 3D object, and parameters of a projection model are fit to these observations. In AreTomo, a projection matching routine is used to iteratively improve projection model parameters. Each wrapper takes the information from an input tilt_series.star file and prepares the appropri- ate commands to align tilt series. This wrapper has not been parallelised. The RELION-5 projection model at this stage is defined as 5 parameters per tilt image. rlnTomoXTilt, rlnTo- moYTilt and rlnTomoZRot constitute a set of extrinsic Euler angles (in °) which rotate the specimen within a fixed microscope coordinate system. In this coordinate system, the optical axis is aligned with the Z-axis and the stage tilt axis is aligned with the Y-axis. The centre of rotation is the centre of the tomogram. The first rotation around the X-axis (rlnTomoXTilt) accounts for nonperpendicularity of the stage tilt axis to the optical (Z) axis. In most cases, Table 2. Metadata items for tilt series images. Metadata labels, types and definitions for the description of a set of images in a tilt series. Entries in bold are compulsory for sub-tomogram averaging; entries in regular are not necessary for sub-tomogram averaging. Label Type Definition _rlnMicrographMovieName String Path of the movie stack for an individual tilt series image _rlnTomoTiltMovieFrameCount Integer Number of frames in the movie stack _rlnTomoNominalStageTiltAngle Float Nominal value for the stage tilt angle (in °) _rlnTomoNominalTiltAxisAngle Float Nominal value for the angle of the tilt axis with the Y-axis (in °) _rlnMicrographPreExposure Float Pre-exposure dose (in electrons/Å2) _rlnTomoNominalDefocus Float Nominal value for the defocus (in μm, negative values for underfocus) _rlnCtfPowerSpectrum String Path of the power spectrum image for CTF estimation _rlnMicrographNameEven String Path of the summed micrograph image from even movie frames only _rlnMicrographNameOdd String Path of the summed micrograph image from odd movie frames only _rlnMicrographName String Path of the summed micrograph image from all movie frames _rlnMicrographMetadata String Path of a STAR file with metadata from motion correction _rlnAccumMotionTotal Float Accumulated global motion during the entire movie (in Å) _rlnAccumMotionEarly Float Accumulated global motion during early frames of the movie (in Å) _rlnAccumMotionLate Float Accumulated global motion during late frames of the movie (in Å) _rlnCtfImage String Path of the Thon-ring image from CTF estimation _rlnDefocusU Float Defocus in the U-direction (in Å, positive values for underfocus) _rlnDefocusV Float Defocus in the V-direction (in Å, positive values for underfocus) _rlnCtfAstigmatism Float Absolute value of the difference between defocus in U- and V-direction (in Å) _rlnDefocusAngle Float Angle between X and defocus U direction (in °) _rlnCtfFigureOfMerit Float Figure of merit from CTF estimation _rlnCtfMaxResolution Float Maximum resolution (in Å) of fitted Thon rings from CTF estimation _rlnCtfIceRingDensity Float Accumulated power of the image in the frequency range (0.25–0.28 Å�1) _rlnTomoXTilt Float Angle for rotation of the tomogram around the X-axis (in °) _rlnTomoYTilt Float Angle for rotation of the tomogram around the Y-axis (in °) _rlnTomoZRot Float Angle for rotation of the tomogram around the Z-axis (in °) _rlnTomoXShiftAngst Float Shift in X-direction (in Å) to align the projection of a tomogram with the tilt series image _rlnTomoYShiftAngst Float Shift in Y-direction (in Å) to align the projection of a tomogram with the tilt series image _rlnCtfScalefactor Float Linear scale-factor to be applied to the CTF values 6 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. this angle will be close to 0°. The second rotation around the Y-axis (rlnTomoYTilt) is the stage tilt angle. The third rotation around the Z-axis (rlnTomoZRot) aligns the Y axis to the tilt axis in the projection image. rlnTomoX- ShiftAngst and rlnTomoYShiftAngst are shifts (in Å from the centre of rotation) applied after rotating the speci- men to align its projection to the experimental image data. The five tilt alignment parameters describe the transfor- mation required to go from 3D coordinates in the tomo- gram to 2D coordinates in the tilt series images. More specifically, denoting rlnTomoXTilt, rlnTomoYTilt and rlnTomoZRot as θx, θy and θz, and rlnTomoXShif- tAngst and rlnTomoYShiftAngst as Δx and Δy, respec- tively, the transformation matrix R that converts (centred) coordinates in the 3D tomogram to (centred) coordinates in the 2D tilt series images is thereby defined as: R=Rt � Rz � Ry � Rx (1) with Rx = 1 0 0 0 0 cos θxð Þ � sin θxð Þ 0 0 0 sin θxð Þ 0 cos θxð Þ 0 0 1 0 BBBBB@ 1 CCCCCA ; Ry = cos θy � � 0 sin θy � � 0 0 1 0 0 � sin θy � � 0 0 0 cos θy � � 0 0 1 0 BBBBB@ 1 CCCCCA ; Rz = cos θzð Þ � sin θzð Þ 0 0 sin θzð Þ cos θzð Þ 0 0 0 0 0 0 1 0 0 1 0 BBBBB@ 1 CCCCCA ; Rt = 1 0 0 Δx 0 1 0 Δy 0 0 0 0 1 0 0 1 0 BBBBB@ 1 CCCCCA Alignment parameters are extracted from output files of IMOD or AreTomo, and written in an output aligned_ tilt_series.star file (Fig. 3; blue). Python application programming interfaces for tilt series alignment procedures in RELION are available as stand-alone python packages ‘yet-another-imod-wrapper’ (https://github.com/teamtomo/yet-another-imod-wrapper) and ‘lil-aretomo’ (https://github.com/teamtomo/lil-aretomo). Tomogram reconstruction The ‘Reconstruct tomograms’ job takes an aligned_- tilt_series.star file as input and uses real-space weighted back-projection, with premultiplication of the CTF using a single defocus value for each tilt image, to reconstruct the corresponding tomograms. As such, these tomograms will have small errors in defocus, which become worse as one moves away from the centre of the recon- struction. In the presented workflow, the reconstructed tomograms are only used for picking particles and subse- quent averaging approaches use cropped 2D stacks as indi- vidual particles. Tomograms are therefore typically reconstructed with relatively large pixel sizes (e.g. 10 Å). If even/odd tilt series were calculated in the ‘Motion correc- tion’ job, then even/odd tomograms can also be calculated for denoising in the ‘Denoise tomograms’ job below. In addition, an option exists to apply an overall tilt angle offset to the tilt series, which may be useful, for example, when tilt series were collected on lamellae that were milled from a thick specimen at a preselected angle away from the direction of the electron beam using a focussed ion beam (FIB). While tilt alignment procedures in software like IMOD and AreTomo can solve for the tilt angle offset, the wrappers in RELION do not use this feature, and the tilt angle offset can only be provided at the tomogram reconstruction stage if known, for exam- ple, from the geometry of milled lamellae. Using a tilt angle offset to minimise the Z-height of tomograms and thereby the required disc space is less important in the RELION pipeline than it is in subtomogram averaging pipelines where particles are extracted as sub-volumes from large tomograms with small pixels. For the same reason, no X-axis offset is estimated or corrected for in addition to the tilt image-specific values of the rlnTo- moXTilt parameter that is estimated during tilt series alignment. The output tomograms.star file, besides containing all the information of the input tilt series file, contains links to the names of the reconstructed tomograms, which are saved in MRC format in a sub-directory of the job called tomograms/. This job also adds a column with a CTF scale-factor, calculated as the cosine of the tilt angle, to the STAR files of the individual tilt series images (Fig. 3; violet). Denoising tomograms The ‘Denoise tomograms’ job takes a tomograms.star file as input and implements a wrapper to the noise2noise denoising program cryo-CARE.46 In the first part of the job, a neural network is trained on a selected subset of rep- resentative even/odd tomograms that were constructed in the ‘Reconstruct tomograms’ job. In the second part of the job, the trained network is applied to the complete set of tomograms that were calculated from all movie frames. The output tomograms.star file from this job points to the denoised tomograms, which are again stored in a sub-directory of the job called tomograms/. 7FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/yet-another-imod-wrapper https://github.com/teamtomo/lil-aretomo https://github.com/teamtomo/lil-aretomo https://github.com/teamtomo/lil-aretomo 8 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. Picking particles A tomograms.star file is used as input for the ‘Pick tomograms’ job, with Napari44 as image viewer and napari-threedee47 to provide tools for the interactive 3D annotation of isolated particles, filaments or spheres on tomogram slices, similar to dtmslice from Dynamo.48 Parti- cles can be taken directly from particle annotations, or sampled from filaments or spheres. Picking on spheres (Fig. 4B) is useful for particles that are arranged, for exam- ple, on spherical virus capsids, vesicles or cells. Filaments are picked as 1D curves. An option to sample particles from nonspherical 2D surfaces was planned, but has not yet been implemented. The napari viewer can also be used to visualise sets of particle coordinates after alignment and/or classification in the latter parts of the pipeline (Fig. 4C). Sampling from different geometries is implemen- ted in a stand-alone python package ‘morphosamplers’ (https://github.com/morphometrics/morphosamplers). The coordinates of manually picked particles, spheres or filaments of individual tomograms are saved in STAR files in a subdirectory called annotations/. When using spheres or filaments, particles are sampled from the spheres or filaments according to a user-specified inter-particle dis- tance. The output particles.star file contains the 3D coordinates of all picked particles in all tomograms, with coordinates defined in Ångstroms relative to the centre of the tomogram (Fig. 5; Table 3). When particles are sam- pled from spheres or filaments, the output STAR file will also contain Euler angles that define a rotation that orients the Z-axis of the particle tangential to the surface of the sphere or perpendicular to the long filament axis. Thereby, in subsequent particle alignment and classification, a 90° prior on the tilt angle will orient particles with their Z-axis normal to the sphere surface or parallel to the long fila- ment axis. Having non-zero tilt angles prevents gimbal locks in the Euler angles during refinement in RELION. Pseudo-subtomograms and 2D particle stacks The particles.star file from the ‘Pick tomograms’ job, together with the tomograms.star file from the ‘Recon- struct tomograms’ job, form the input to the ‘Extract sub- tomos’ job. Because these two files together define the input, they are bundled in a file called optimisation_ set.star, which is also written out by the ‘Pick tomo- grams’ job. The optimisation_set.star contains links to the filenames of both individual STAR files. One can either use the optimisation_set.star file or the two separate STAR files as input for the ‘Extract subtomos’ job. This job will combine the 3D positions of the particles with the alignment parameters of the tomograms either to construct 3D pseudo-subtomograms (as introduced in RELION-418), or to crop the relevant regions around indi- vidual particles in the tilt series images and save these as 2D particle stacks. Both the pseudo-subtomograms and the 2D stacks will be premultiplied with the CTF in images with a user-defined box size, and then possibly cropped to a smaller box size after the signal delocalisation has been compensated. These calculations are parallelised using both MPI and threads. Because there are typically fewer images in the tilt series (Ntilt) than the cropped boxed size (B), 2D stacks (with Ntilt * B2 pixels) often occupy less disc space than 3D pseudo-subtomograms (with B3 voxels). Options exist to only use tilt series images below a user-defined maximum dose and to only output particles that are visible on a user-specified minimum number of tilt series images. Writing particles in float16 saves a factor of 2 in disc space compared to writing in the default float32 MRC format, although not all third-party programs may be able to read such images. The particle images are saved in a subdirectory of the job called Subtomograms/. The job also saves a particles.- star file with the names of the particle images, and (for 2D stacks only) an array with names of images of the tilt series saved in the stacks. The job also writes out a new opti- misation_set.star with the filenames of the output par- ticles.star file and the input tomograms.star file. Subsequent averaging approaches The optimisation_set.star from the ‘Extract subto- mos’ job can be used as input to any of the existing ‘3D initial reference’, ‘3D classification’, ‘3D auto-refine’ or ‘3D multi-body’ jobs in the standard RELION pipeline. These methods already existed in single-particle and tomography pipelines of RELION18,42 and are not described in detail here. Two new features are an option to resize input 3D reference maps and mask if they have a different pixel and/or box size than the input images (the same also works for standard single-particle analysis), and an option to impose a prior on the tilt angle, which is useful if particles were sampled from spheres or filaments. Each of these jobs will also output optimisation_set.star files with links to the STAR files with the tomograms and the refined particle coordinates. Fig. 4. Screenshots of Napari-based picking tools. (A) The tilt-series selection program opened by the “Exclude tilt-images” job allows the user to de-select tilt images for each tilt series. (B) The picker plug-in opened by the “Pick tomograms” job allows one to annotate the reconstructed tomograms with spheres (shown here) or 1D-curves (not shown), which will then be used to randomly sample particles with priors that orient the normal to the sphere or along the curves with the Z-axis. (C) Individual particles can also be annotated manually, or read from a STAR file for visualisation. 9FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 https://github.com/morphometrics/morphosamplers Bayesian polishing and CTF refinement The methods for Bayesian polishing, which performs tilt series realignment and estimation of per-particle motion throughout the tilt series (but does not follow motion of individual particles through individual movie frames), and CTF refinement, which refines the defocus estimates for each tilt series image, have been described previously18 and will not be discussed here. Besides an input optimisa- tion_set.star file, these methods also need reference half-maps, a reference mask and a postprocess.star file from a ‘Post-processing’ job as input. Because these cal- culations are sensitive to the size and grey-scale of the ref- erence, the reference maps need to be calculated by the ‘Reconstruct particle’ job, and a binning factor of 1 must be used. A larger binning factor may be used in the preced- ing ‘3D classification’ or ‘3D auto-refine’ jobs. The ‘CTF refinement’ job will update the defocus values for all tilt series images and store links to updated STAR Fig. 5. Particle metadata description after picking. Metadata to describe a set of particles after picking in the Napari GUI. A single STAR file contains particle coordinates (in Ångstroms relative to the centre of the tomogram) for all tomograms. The labels and data columns in violet arise from the sampling of the particles on manually picked spheres, and express information that aligns their Z-axis with the normal to the surface sphere. Corresponding definitions of metadata labels are given in Table 3. Table 3. Metadata items for particle picks. Metadata labels, types and definitions for the description of a set of picked particles. Entries in bold are always present; entries in regular are only present when particles are sampled from spheres or filaments. Label Type Definition _rlnTomoName String Name of a tomogram (same as in Table 1) _rlnCenteredCoordinateXAngst Float X-position in a tomogram, in Ångstroms from the centre _rlnCenteredCoordinateYAngst Float Y-position in a tomogram, in Ångstroms from the centre _rlnCenteredCoordinateZAngst Float Z-position in a tomogram, in Ångstroms from the centre _rlnTomoSubtomogramRot Float First Euler angle (in °) that pre-orients subtomograms in the tomogram coordinate system _rlnTomoSubtomogramTilt Float Second Euler angle (in °) that pre-orients subtomograms in the tomogram coordinate system _rlnTomoSubtomogramPsi Float Third Euler angle (in °) that pre-orients subtomograms in the tomogram coordinate system _rlnAngleRot Float First Euler angle for particle alignment (rot, in °) as defined in49 _rlnAngleTilt Float Second Euler angle for particle alignment (tilt, in °) as defined in49 _rlnAnglePsi Float Third Euler angle for particle alignment (psi, in °) as defined in49 _rlnAngleTiltPrior Float Centre of the prior on the second Euler angle for particle alignment (in °) _rlnAnglePsiPrior Float Centre of the prior on the third Euler angle for particle alignment (in °) 10 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. files for the individual tilt series in an updated tomo- grams.star file. Likewise, the ‘Bayesian polishing’ job will update the tilt series alignment parameters in a new tomograms.star file. If per-particle motion estimation is performed, then the ‘Bayesian polishing’ job will also out- put a motion.star file with the estimated per-particle motion tracks for all particles. A link to these files will also be added to the output optimisation_set.star file. If one wishes to perform further refinements or classifica- tions after a ‘CTF refinement’ and/or a ‘Bayesian polishing’ job, new particles should be extracted using the ‘Extract subtomos’ job with the updated CTF parameters and/or tilt series alignment parameters and particle positions. If no further refinement or classification is deemed necessary, then calculating a new reconstruction with the ‘Reconstruct particle’ job will suffice. Results The RELION 5 pipeline has already proven useful for the processing of two cryo-ET data sets.50,51 Here, we also illustrate its capability on a benchmark data set consisting of five tomograms of virus-like particles (VLPs) of the capsid and spacer peptide 1 (CA-SP1) region of the Gag polyprotein in the immature human immunodeficiency virus 1 (HIV-1).52 The same subset of five tomograms has previously been used to bench- mark various cryo-ET tools. NovaCTF introduced 3D CTF correction, leading to 3.9 Å resolution53; optimis- ing frame alignment and CTF parameters in Warp led to 3.8 Å40; a combination of Warp-RELION-M and Dynamo yielded 3.4 Å54; and RELION-4 gave 3.2 Å.18 Starting from the import of raw movie frames and SerialEM mdoc files, all processing steps were per- formed using the pipeline described in the Methods section. Movie frames were aligned using RELION’s own motion-correction program; tilt series images without signals (dark images) were manually excluded from the first two tilt series; CTF parameters were esti- mated using CTFFIND435; and tilt alignment was per- formed using the fiducial-based procedure in IMOD.7 Next, the five tomograms were reconstructed with a binned voxel size of 10 Å and the napari-based 3D picker was used to manually annotate 47 spheres coin- ciding with the outer surfaces of the VLPs observed in the tomograms. Using an interparticle distance of 60 Å, 30597 particles were extracted by randomly sam- pling the spheres. Priors on the orientations of the par- ticles, which position the normal to the sphere parallel to the Z-axis for each particle at a tilt angle of 90°, allowed us to obtain an initial 3D reference map by simply running a ‘Reconstruct particle’ job with six- fold rotational symmetry. The initial model was subjected to a first ‘3D auto-refine’ job at a binning factor of 6, followed by a second ‘3D auto-refine’ job at a binning factor of 2. After removing 8937 dupli- cated particles (using the corresponding option in a ‘Subset selection’ job), we then ran a ‘3D classification’ job with nine classes, local alignment and regularisa- tion parameter T= 1. Manual selection of the best two classes resulted in a subset of 9053 particles. These particles were used for a 3D auto-refinement at a bin- ning factor of 1, which gave a map at 4.0 Å resolution. Five iterations of CTF refinement, Bayesian polishing and 3D auto-refinement at a binning factor of 1 yielded a final map at 3.3 Å. Automated model build- ing with ModelAngelo34 led to an atomic model com- prising 211 (94%) of the 224 residues that were built in the original study.52 Fourier shell correlation (FSC) curves of some of the intermediate steps and the final reconstruction are shown in Fig. 6A and the final map is shown in Fig. 6B. In Fig. 6C, we show representative regions from the reconstructed map after the first 3D refine- ment step at binning factor 1 (first row), after the first iteration of CTF refinement, Bayesian polishing and 3D auto-refinement (second row), and in the final map (third row), as well as the final map with the atomic model as output by the ModelAngelo job, with no additional model refinements. The computational costs of the individual steps in the pipeline are summarised in Table 4. For all 3D auto-refinements and the 3D classification leading to the final map, individual parti- cles were written as 2D image stacks in the corre- sponding ‘Extract subtomos’ jobs. For comparison with RELION-4, we also extracted particles at a bin- ning factor of 1 as 3D pseudo-subtomograms. This required 20 times more disc space and resulted in an eightfold increase in wall-clock compute time for 3D autorefinement, compared to extracting particles as 2D image stacks. The results described above have been made avail- able through Zenodo (DOI: 10.5281/zenodo.11068319) and form the basis for a detailed tutorial of the RELION-5 tomography pipeline, which can be found at http://relion.readthedocs.io. The final map is also available from the EMDB (EMD-50672). Discussion Gaining structural insights from cryo-ET data is com- plex and remains an active area of research, with ongoing developments in CTF estimation,39 tilt series alignment,8 tomogram denoising,21,23,24,46,55 automated localisation of structures of interest30,56–65 and approaches to deal with structural heterogeneity 11FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 https://doi.org/10.5281/zenodo.11068319 http://relion.readthedocs.io among extracted particles,66–69 among many others. As a result, cryo-ET image processing pipelines, including the one described here, will need to be continuously updated in the coming years. Probably, the most urgent extensions to the pipeline in RELION-5 would include a more robust handling of reconstruction and defocus handedness from the provided metadata, more robust tools for automated tilt series alignment, and Fig. 6. Subtomogram averaging of the CA-SP1 hexamer of the immature HIV1 capsid. (A) FSC curves at intermediate and final refinement steps in the RELION-5 pipeline. (B) Final reconstructed map, side view (top), top (bottom left) and bottom views (bottom right). (C) Representative regions in the reconstructed map after the first 3D auto-refinement at binning factor of 1 (blue, top row), after the first iteration of CTF refinement, Bayesian polishing and 3D auto-refinement (orange, second row), after the fifth iteration (magenta, third row), and the final map with the atomic model generated by ModelAngelo (green, bottom row). 12 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. the incorporation of tools for automated tomogram segmentation and particle picking. Given the popular- ity of the combination of Warp/M1,40 with subtomo- gram averaging in RELION-3,70 updated interoperability between Warp/M and RELION-5 is also desirable. Moreover, the use of napari as a plat- form for the development of these picking tools intro- duces complexity, especially for viewing data remotely. To avoid the maintenance burden of this complexity in RELION, the napari-based tools are intended to become standalone tools in the near future. However, different research groups will require dif- ferent tools to fulfil their specific image-processing needs, and it will remain difficult to include all of these in a single software pipeline. Recognising this, much of the work presented here was implemented as stand- alone python packages designed to maximise their reusability outside of the RELION framework. It would be helpful if the different groups that use and write cryo-ET software were to agree on a standar- dised description that allows passing metadata between different programs. In single-particle analysis, the exchange of metadata is often performed through files in the STAR format,38 with geometric definitions origi- nally defined by Heymann et al49 and then implemented in RELION.71 The explicit definition of metadata structures for tilt series, tomograms and extracted particles presented in this paper could fulfil a similar role for the cryo-ET field. A first step on this journey may be the adoption of RELION-5’s STAR files for cryo-ET data (Figs 2, 3 & 5 and Tables 1–3) in the upcoming cryo-ET pipeline of the Collaborative Computing Project for Electron cryo-Microscopy (CCP-EM) software.37 This new software will provide mechanisms for pipelining a wider range of cryo-ET programs than those available within RELION-5 (Tom Burnley, personal communication), thus further improving the accessibility for newcomers to this rap- idly developing field. Conclusions We present a pipeline for the analysis of cryo-ET data in RELION-5 that ranges from the import of unpro- cessed movies to automated atomic modelling in high- resolution subtomogram averaging maps. The explicit metadata definitions of tilt series, tomograms and extracted particles in RELION-5 may also serve wider efforts at standardisation and software inter- operability. Table 4. Computational costs. Each row represents a job described in the Results section, reporting its wall-clock time required for execution; how many MPI processes, threads and/or GPUs were used; and how much disc space the job required. All computations were performed on an (Ubuntu 22.04.3 LTS) Linux workstation with an AMD Ryzen Threadripper PRO 5975WX CPU with 32 cores and 64 threads, 512 GB RAM and two NVIDIA GeForce RTX 4090 GPUs, each with 24GB memory. Jobs Time Disk space MPI processes, threads, GPUs Other information Motion correction 6min 17 GB 32 MPI, 4 threads CTF estimation <1min 212MB 32 MPI CTFFIND-4.1 Align tilt-series 6min 11 GB IMOD Reconstruct tomograms 3min 3 GB 5 MPI, 12 threads Binned pixel size 10 Å Denoise tomograms/train 1 h 22min 20MB 1 GPU Denoise tomograms/predict 1min 1.5 GB Extract subtomos/bin6 17min 8.6 GB 5 MPI, 12 threads 30 597 particles; Binning factor 6; Box size 96 3D auto-refine/bin6 3 h 31min 600MB 5 MPI, 6 threads, 2 GPUs 26 iterations Extract subtomos/bin2 3min 16 GB 5 MPI, 12 threads 30 597 particles; Binning factor 2; Box size 128Reconstruct particle/bin2 5min 600MB 3D auto-refine/bin2 7 h 35min 754MB 5 MPI, 6 threads, 2 GPUs 24 iterations 3D classification 1 h 55min 2.1 GB 21 660 particles; 25 iterations Extract subtomos/bin1 3min 11 GB 5 MPI, 12 threads 9053 particles; Binning factor 1; Box size 192Reconstruct particle/bin1 4min 4.6 GB 3D auto-refine/bin1 1 h 41min 1.6 GB 5 MPI, 6 threads, 2 GPUs 13 iterations CTF refinement 6min <1MB 5 MPI, 12 threads Bayesian polishing 1 h 32min 270MB Extract subtomos/bin1 (3D pseudo- subtomos) 24min 240 GB 5 MPI, 12 threads 9053 particles; Binning factor 1; Box size 192 3D auto-refine/bin1 (3D pseudo- subtomos) 13 h 57min 1.5 GB 5 MPI, 6 threads, 2 GPUs 12 iterations 13FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 Acknowledgements We are grateful to Jake Grimmett, Toby Darling and Ivan Clayson for help with high-performance computing, and to Takanori Nakane for helpful discussions and RELION support. AB was a member of the group of David Barford (DB). This work was supported by the Medical Research Council as part of the United Kingdom Research and Innovation (MC_UP_A025_1013 to SHWS; MC_UP_1201/31 to TAMB and MC_UP_1201/6 to DB). AvK is supported by a Philip Leverhulme Prize to TAMB. RW is supported by The Pittsburgh Founda- tion & The Commonwealth of Pennsylvania Formula Fund to Zachary Freyberg; EP was supported by grants from the European Research Council (ERC-StG-2019 grant 852915) and the BBSRC (grant BB/T002670/1) to Giulia Zanetti. For the purpose of open access, the MRC Laboratory of Molecular Biology has applied a CC BY public copyright licence to any Author Accepted Manu- script version arising. Conflict of interest The authors declare no conflicts of interest. Peer review The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer- review/10.1002/2211-5463.13873. Data accessibility RELION is distributed under a GPLv2 open-source soft- ware licence and can be downloaded for free from http:// www.github.com/3dem/relion. All stand-alone python packages are distributed under a BSD-3 open-source soft- ware licence. The data set described in the Results section can be downloaded from EMPIAR under accession num- ber 10164, whereas the results themselves can be down- loaded from Zenodo (DOI: 10.5281/zenodo.11068319). Author contributions AB, BT, RW, EP, JZ, DK and SHWS wrote computer code; AB, BT, RW and SHWS performed experiments and analysed results. SHWS supervised the project. All authors contributed to writing the manuscript. References 1 Tegunov D, Xue L, Dienemann C, Cramer P and Mahamid J (2021) Multi-particle cryo-EM refinement with M visualizes ribosome-antibiotic complex at 3.5 Å in cells. Nat Methods 18, 186–193. 2 Xing H, Taniguchi R, Khusainov I, Kreysing JP, Welsch S, Turoňová B and Beck M (2023) Translation dynamics in human cells visualized at high resolution reveal cancer drug action. Science 381, 70–75. 3 Xue L, Lenz S, Zimmermann-Kogadeeva M, Tegunov D, Cramer P, Bork P, Rappsilber J and Mahamid J (2022) Visualizing translation dynamics at atomic detail inside a bacterial cell. Nature 610, 205–211. 4 Mastronarde DN (2005) Automated electron microscope tomography using robust prediction of specimen movements. J Struct Biol 152, 36–51. 5 Schorb M, Haberbosch I, Hagen WJH, Schwab Y and Mastronarde DN (2019) Software tools for automated transmission electron microscopy. Nat Methods 16, 471– 477. 6 Kremer JR, Mastronarde DN and McIntosh JR (1996) Computer visualization of three-dimensional image data using IMOD. J Struct Biol 116, 71–76. 7 Mastronarde DN and Held SR (2017) Automated tilt series alignment and tomographic reconstruction in IMOD. J Struct Biol 197, 102–113. 8 Zheng S, Wolff G, Greenan G, Chen Z, Faas FGA, Bárcena M, Koster AJ, Cheng Y and Agard DA (2022) AreTomo: an integrated software package for automated marker-free, motion-corrected cryo-electron tomographic alignment and reconstruction. J Struct Biol X 6, 100068. 9 Radermacher M (1988) Three-dimensional reconstruction of single particles from random and nonrandom tilt series. J Electron Microsc Tech 9, 359– 394. 10 Gilbert P (1972) Iterative methods for the three- dimensional reconstruction of an object from projections. J Theor Biol 36, 105–117. 11 Agulleiro JI and Fernandez JJ (2011) Fast tomographic reconstruction on multicore computers. Bioinformatics 27, 582–583. 12 Fernandez J-J, Li S, Bharat TAM and Agard DA (2018) Cryo-tomography tilt-series alignment with consideration of the beam-induced sample motion. J Struct Biol 202, 200–209. 13 Gordon R, Bender R and Herman GT (1970) Algebraic reconstruction techniques (ART) for three-dimensional electron microscopy and x-ray photography. J Theor Biol 29, 471–481. 14 Marabini R, Herman GT and Carazo JM (1998) 3D reconstruction in electron microscopy using ART with smooth spherically symmetric volume elements (blobs). Ultramicroscopy 72, 53–65. 15 Kunz M and Frangakis AS (2014) Super-sampling SART with ordered subsets. J Struct Biol 188, 107–115. 16 Böhning J, Bharat TAM and Collins SM (2022) Compressed sensing for electron cryotomography and 14 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2211-5463.13873 http://www.github.com/3dem/relion http://www.github.com/3dem/relion https://doi.org/10.5281/zenodo.11068319 high-resolution subtomogram averaging of biological specimens. Structure 30, 408–417.e4. 17 Himes BA and Zhang P (2018) emClarity: software for high-resolution cryo-electron tomography and subtomogram averaging. Nat Methods 15, 955–961. 18 Zivanov J, Otón J, Ke Z, von Kügelgen A, Pyle E, Qu K, Morado D, Castaño-Dı́ez D, Zanetti G, Bharat TAM et al. (2022) A Bayesian approach to single- particle electron cryo-tomography in RELION-4.0. eLife 11, e83724. 19 Liu H-F, Zhou Y, Huang Q, Piland J, Jin W, Mandel J, Du X, Martin J and Bartesaghi A (2023) nextPYP: a comprehensive and scalable platform for characterizing protein variability in situ using single-particle cryo- electron tomography. Nat Methods 20, 1909–1919. 20 Bartesaghi A, Lecumberry F, Sapiro G and Subramaniam S (2012) Protein secondary structure determination by constrained single-particle cryo- electron tomography. Structure 20, 2003–2013. 21 Liu Y-T, Zhang H, Wang H, Tao C-L, Bi G-Q and Zhou ZH (2022) Isotropic reconstruction for electron tomography with deep learning. Nat Commun 13, 6482. 22 Buchholz T-O, Krull A, Shahidi R, Pigino G, Jékely G and Jug F (2019) Content-aware image restoration for electron microscopy. Methods Cell Biol 152, 277–289. 23 Bepler T, Kelley K, Noble AJ and Berger B (2020) Topaz-denoise: general deep denoising models for cryoEM and cryoET. Nat Commun 11, 5208. 24 Zhang H, Li Y, Liu Y, Li D, Wang L, Song K, Bao K and Zhu P (2023) A method for restoring signals and revealing individual macromolecule states in cryo-ET, REST. Nat Commun 14, 2937. 25 de Teresa-Trueba I, Goetz SK, Mattausch A, Stojanovska F, Zimmerli CE, Toro-Nahuelpan M, Cheng DWC, Tollervey F, Pape C, Beck M et al. (2023) Convolutional networks for supervised mining of molecular patterns within cellular context. Nat Methods 20, 284–294. 26 Moebel E, Martinez-Sanchez A, Lamm L, Righetto RD, Wietrzynski W, Albert S, Larivière D, Fourmentin E, Pfeffer S, Ortiz J et al. (2021) Deep learning improves macromolecule identification in 3D cellular cryo-electron tomograms. Nat Methods 18, 1386–1394. 27 Heumann JM, Hoenger A and Mastronarde DN (2011) Clustering and variance maps for cryo-electron tomography using wedge-masked differences. J Struct Biol 175, 288–299. 28 Galaz-Montoya JG, Flanagan J, Schmid MF and Ludtke SJ (2015) Single particle tomography in EMAN2. J Struct Biol 190, 279–290. 29 Castaño-Dı́ez D, Kudryashev M, Arheit M and Stahlberg H (2012) Dynamo: a flexible, user-friendly development tool for subtomogram averaging of cryo- EM data in high-performance computing environments. J Struct Biol 178, 139–151. 30 Wan W, Khavnekar S and Wagner J (2024) STOPGAP: an open-source package for template matching, subtomogram alignment and classification. Acta Cryst, Sect D: Struct Biol 80. doi: 10.1107/S205979832400295X 31 Balyschew N, Yushkevich A, Mikirtumov V, Sanchez RM, Sprink T and Kudryashev M (2023) Streamlined structure determination by cryo-electron tomography and subtomogram averaging using TomoBEAR. Nat Commun 14, 6543. 32 de la Rosa-Trevı́n JM, Quintana A, Del Cano L, Zaldı́var A, Foche I, Gutiérrez J, Gómez-Blanco J, Burguet-Castell J, Cuenca-Alba J, Abrishami V et al. (2016) Scipion: a software framework toward integration, reproducibility and validation in 3D electron microscopy. J Struct Biol 195, 93–99. 33 Jiménez de la Morena J, Conesa P, Fonseca YC, de Isidro-Gómez FP, Herreros D, Fernández-Giménez E, Strelak D, Moebel E, Buchholz TO, Jug F et al. (2022) ScipionTomo: towards cryo-electron tomography software integration, reproducibility, and validation. J Struct Biol 214, 107872. 34 Jamali K, Käll L, Zhang R, Brown A, Kimanius D and Scheres SHW (2024) Automated model building and protein identification in cryo-EM maps. Nature 628, 450–457. 35 Rohou A and Grigorieff N (2015) CTFFIND4: fast and accurate defocus estimation from electron micrographs. J Struct Biol 192, 216–221. 36 Wood C, Burnley T, Patwardhan A, Scheres S, Topf M, Roseman A and Winn M (2015) Collaborative computational project for electron cryo-microscopy. Acta Crystallogr D Biol Crystallogr 71, 123–126. 37 Burnley T, Palmer CM and Winn M (2017) Recent developments in the CCP-EM software suite. Acta Crystallogr D Struct Biol 73, 469–477. 38 Hall SR (1991) The STAR file: a new format for electronic data transfer and archiving. J Chem Inf Comput Sci 31, 326–333. 39 Elferich J, Kong L, Zottig X and Grigorieff N (2024) CTFFIND5 provides improved insight into quality, tilt and thickness of TEM samples. bioRxiv 2024-02. doi: 10.7554/eLife.97227.1 40 Tegunov D and Cramer P (2019) Real-time cryo- electron microscopy data preprocessing with Warp. Nat Methods 16, 1146–1152. 41 Zheng SQ, Palovcak E, Armache J-P, Verba KA, Cheng Y and Agard DA (2017) MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat Methods 14, 331–332. 42 Zivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E and Scheres SH (2018) New tools for automated high-resolution cryo-EM structure determination in RELION-3. eLife 7, e42166. 15FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 https://doi.org/10.1107/S205979832400295X https://doi.org/10.7554/eLife.97227.1 43 González-Rodrı́guez N, Areán-Ulloa E and Fernández- Leiro R (2024) A web-based dashboard for RELION metadata visualization. Acta Cryst D 80, 93–100. 44 Ahlers J, Althviz Moré D, Amsalem O, Anderson A, Bokota G, Boone P, Bragantini J, Buckley G, Burt A, Bussonnier M et al. (2023) Napari: a multi-dimensional image viewer for python. Version v0.4.18 (Zenodo). doi: 10.5281/zenodo.8115575 45 Gaifas L, Kirchner MA, Timmins J and Gutsche I (2024) blik: an extensible napari plugin for cryo-ET data visualisation, annotation and analysis. bioRxiv 2023-12.doi: 10.1371/journal.pbio.3002447 46 Buchholz T-O, Jordan M, Pigino G and Jug F (2018) Cryo-CARE: content-aware image restoration for Cryo- transmission electron microscopy data. arXiv [Preprint]. doi: 10.48550/arXiv.1810.05420 47 Yamauchi KA and Burt A (2023) Napari-threedee: a toolkit for human-in-the-loop 3D image analysis in napari. bioRxiv [Preprint]. doi: 10.1101/2023.07.28. 550950 48 Castaño-Dı́ez D, Kudryashev M and Stahlberg H (2017) Dynamo catalogue: geometrical tools and data management for particle picking in subtomogram averaging of cryo-electron tomograms. J Struct Biol 197, 135–144. 49 Heymann JB, Chagoyen M and Belnap DM (2005) Common conventions for interchange and archiving of three-dimensional electron microscopy information in structural biology. J Struct Biol 151, 196–207. 50 Ochner H, Böhning J, Wang Z, Tarafder AK, Caspy I and Bharat TAM (2024) Structure of the Pseudomonas aeruginosa PAO1 type IV pilus. bioRxiv [Preprint]. doi: 10.1101/2024.04.09.588664 51 Pyle E and Zanetti G (2024) Cryo-electron tomography reveals how COPII assembles on cargo-containing membranes. bioRxiv [Preprint]. doi: 10.1101/2024.01.17. 576008 52 Schur FKM, Obr M, Hagen WJH, Wan W, Jakobi AJ, Kirkpatrick JM, Sachse C, Kräusslich H-G and Briggs JAG (2016) An atomic model of HIV-1 capsid-SP1 reveals structures regulating assembly and maturation. Science 353, 506–508. 53 Turoňová B, Schur FKM, Wan W and Briggs JAG (2017) Efficient 3D-CTF correction for cryo-electron tomography using NovaCTF improves subtomogram averaging resolution to 3.4 Å. J Struct Biol 199, 187– 195. 54 Burt A, Gaifas L, Dendooven T and Gutsche I (2021) A flexible framework for multi-particle refinement in cryo-electron tomography. PLoS Biol 19, e3001319. 55 Yang Z, Zang D, Li H, Zhang Z, Zhang F and Han R (2024) Self-supervised noise modeling and sparsity guided electron tomography volumetric image denoising. Ultramicroscopy 255, 113860. 56 Heebner JE, Purnell C, Hylton RK, Marsh M, Grillo MA and Swulius MT (2022) Deep learning-based segmentation of Cryo-electron tomograms. J Vis Exp doi: 10.3791/64435 57 Zhou B, Yu H, Zeng X, Yang X, Zhang J and Xu M (2020) One-shot learning with attention-guided segmentation in Cryo-electron tomography. Front Mol Biosci 7, 613347. 58 Lamm L, Righetto RD, Wietrzynski W, Pöge M, Martinez-Sanchez A, Peng T and Engel BD (2022) MemBrain: a deep learning-aided pipeline for detection of membrane proteins in Cryo-electron tomograms. Comput Methods Prog Biomed 224, 106990. 59 Siggel M, Jensen RK, Mahamid J and Kosinski J (2024) ColabSeg: an interactive tool for editing, processing, and visualizing membrane segmentations from cryo-ET data. J Struct Biol 216, 108067. doi: 10.1016/j.jsb.2024.108067 60 Sazzed S, Scheible P, He J and Wriggers W (2022) Spaghetti tracer: a framework for tracing semiregular filamentous densities in 3D tomograms. Biomol Ther 12, 1022. 61 Dimchev G, Amiri B, Fäßler F, Falcke M and Schur FK (2021) Computational toolbox for ultrastructural quantitative analysis of filament networks in cryo-ET data. J Struct Biol 213, 107808. doi: 10.1016/j.jsb.2021. 107808 62 Pyle E, Hutchings J and Zanetti G (2022) Strategies for picking membrane-associated particles within subtomogram averaging workflows. Faraday Discuss 240, 101–113. 63 Rice G, Wagner T, Stabrin M, Sitsel O, Prumbaum D and Raunser S (2023) TomoTwin: generalized 3D localization of macromolecules in cryo-electron tomograms with structural data mining. Nat Methods 20, 871–880. 64 Genthe E, Miletic S, Tekkali I, Hennell James R, Marlovits TC and Heuser P (2023) PickYOLO: fast deep learning particle detector for annotation of cryo electron tomograms. J Struct Biol 215, 107990. 65 Chaillet ML, van der Schot G, Gubins I, Roet S, Veltkamp RC and Förster F (2023) Extensive angular sampling enables the sensitive localization of macromolecules in electron tomograms. Int J Mol Sci 24, 13375. 66 Powell BM and Davis JH (2024) Learning structural heterogeneity from cryo-electron sub-tomograms with tomoDRGN. Nat Methods 1–2. doi: 10.1038/s41592- 024-02210-z 67 Harastani M, Eltsov M, Leforestier A and Jonic S (2021) HEMNMA-3D: Cryo electron tomography method based on Normal mode analysis to study continuous conformational variability of macromolecular complexes. Front Mol Biosci 8, 663121. 16 FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A tomography pipeline in RELION-5 A. Burt et al. https://doi.org/10.5281/zenodo.8115575 https://doi.org/10.1371/journal.pbio.3002447 https://doi.org/10.48550/arXiv.1810.05420 https://doi.org/10.1101/2023.07.28.550950 https://doi.org/10.1101/2023.07.28.550950 https://doi.org/10.1101/2024.04.09.588664 https://doi.org/10.1101/2024.01.17.576008 https://doi.org/10.1101/2024.01.17.576008 https://doi.org/10.3791/64435 https://doi.org/10.1016/j.jsb.2024.108067 https://doi.org/10.1016/j.jsb.2021.107808 https://doi.org/10.1016/j.jsb.2021.107808 https://doi.org/10.1038/s41592-024-02210-z https://doi.org/10.1038/s41592-024-02210-z 68 Barchet C, Fréchin L, Holvec S, Hazemann I, von Loeffelholz O and Klaholz BP (2023) Focused classifications and refinements in high-resolution single particle cryo-EM analysis. J Struct Biol 215, 108015. 69 Sibert BS, Kim JY, Yang JE, Hannon-Hatfield A, Ke Z, Garfinkel DJ and Wright ER (2023) Workflow for high-resolution sub-volume averaging from heterogenous viral and virus-like assemblies. Microsc Microanal 29, 943–944. 70 Bharat TAM and Scheres SHW (2016) Resolving macromolecular structures from electron cryo- tomography data using subtomogram averaging in RELION. Nat Protoc 11, 2054–2065. 71 Scheres SHW (2012) RELION: implementation of a Bayesian approach to cryo-EM structure determination. J Struct Biol 180, 519–530. 17FEBS Open Bio (2024) ª 2024 The Author(s). FEBS Open Bio published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. A. Burt et al. A tomography pipeline in RELION-5 Outline placeholder feb413873-aff-0001 feb413873-aff-0002 feb413873-aff-0003 feb413873-aff-0004 feb413873-aff-0005 feb413873-aff-0006 Materials and methods Importing into RELION's data�model feb413873-fig-0001 Motion correction feb413873-fig-0002 feb413873-tbl-0001 CTF estimation feb413873-fig-0003 Selection of tilt series images Tilt series alignment feb413873-tbl-0002 Tomogram reconstruction Denoising tomograms Picking particles �Pseudo-�subtomograms� and 2D particle stacks Subsequent averaging approaches feb413873-fig-0004 Bayesian polishing and CTF refinement feb413873-fig-0005 feb413873-tbl-0003 Results Discussion feb413873-fig-0006 Conclusions feb413873-tbl-0004 Acknowledgements Conflict of interest Peer review Data accessibility Author contributions feb413873-bib-0001 feb413873-bib-0002 feb413873-bib-0003 feb413873-bib-0004 feb413873-bib-0005 feb413873-bib-0006 feb413873-bib-0007 feb413873-bib-0008 feb413873-bib-0009 feb413873-bib-0010 feb413873-bib-0011 feb413873-bib-0012 feb413873-bib-0013 feb413873-bib-0014 feb413873-bib-0015 feb413873-bib-0016 feb413873-bib-0017 feb413873-bib-0018 feb413873-bib-0019 feb413873-bib-0020 feb413873-bib-0021 feb413873-bib-0022 feb413873-bib-0023 feb413873-bib-0024 feb413873-bib-0025 feb413873-bib-0026 feb413873-bib-0027 feb413873-bib-0028 feb413873-bib-0029 feb413873-bib-0030 feb413873-bib-0031 feb413873-bib-0032 feb413873-bib-0033 feb413873-bib-0034 feb413873-bib-0035 feb413873-bib-0036 feb413873-bib-0037 feb413873-bib-0038 feb413873-bib-0039 feb413873-bib-0040 feb413873-bib-0041 feb413873-bib-0042 feb413873-bib-0043 feb413873-bib-0044 feb413873-bib-0045 feb413873-bib-0046 feb413873-bib-0047 feb413873-bib-0048 feb413873-bib-0049 feb413873-bib-0050 feb413873-bib-0051 feb413873-bib-0052 feb413873-bib-0053 feb413873-bib-0054 feb413873-bib-0055 feb413873-bib-0056 feb413873-bib-0057 feb413873-bib-0058 feb413873-bib-0059 feb413873-bib-0060 feb413873-bib-0061 feb413873-bib-0062 feb413873-bib-0063 feb413873-bib-0064 feb413873-bib-0065 feb413873-bib-0066 feb413873-bib-0067 feb413873-bib-0068 feb413873-bib-0069 feb413873-bib-0070 feb413873-bib-0071