Journal of Structural Biology 214 (2022) 107872 Contents lists available at ScienceDirect Journal of Structural Biology journal homepage: www.elsevier.com/locate/yjsbi ScipionTomo: Towards cryo-electron tomography software integration, reproducibility, and validation J. Jiménez de la Morena a, P. Conesa a, Y.C. Fonseca a, F.P. de Isidro-Gómez a, D. Herreros a, E. Fernández-Giménez a, D. Strelak a, b, E. Moebel c, T.O. Buchholz d, e, F. Jug d, f, A. Martinez- Sanchez g, h, M. Harastani n, S. Jonic n, J.J. Conesa a, A. Cuervo a, P. Losana a, I. Sánchez a, M. Iceta a, L. del Cano a, M. Gragera a, R. Melero a, G. Sharov i, D. Castaño-Díez j, A. Koster k, J.G. Piccirillo a, J.L. Vilas a, J. Otón l, R. Marabini a, m, C.O.S. Sorzano a, *, J.M. Carazo a, * a National Center of Biotechnology (CNB-CSIC), Madrid, Spain b Masaryk University, Brno, Czech Republic c Inria Rennes - Bretagne Atlantique, Rennes, France d Max Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG), Germany e Center for Systems Biology Dresden (CSBD), Germany f Fondazione Human Technopole, Milan, Italy g University of Oviedo, Department of Computer Sciences, Oviedo, Spain h Health Research Institute of Asturias (ISPA), Oviedo, Spain i Structural Studies Division, MRC Laboratory of Molecular Biology, Cambridge, United Kingdom j BioEM Lab, Biozentrum, University of Basel, Basel, Switzerland k University of Leiden, Ultrastructural and Molecular Imaging, Leiden, the Netherlands l Alba Synchrotron - CELLS (ICTS), Barcelona, Spain m Superior Polytechnic School. Univ. Autónoma of Madrid. Madrid, Spain n IMPMC-UMR 7590 CNRS, Sorbonne Université, MNHN, Paris, France A B S T R A C T Image processing in cryogenic electron tomography (cryoET) is currently at a similar state as Single Particle Analysis (SPA) in cryogenic electron microscopy (cryoEM) was a few years ago. Its data processing workflows are far from being well defined and the user experience is still not smooth. Moreover, file formats of different software packages and their associated metadata are not standardized, mainly since different packages are developed by different groups, focusing on different steps of the data processing pipeline. The Scipion framework, originally developed for SPA (de la Rosa-Trevín et al., 2016), has a generic python workflow engine that gives it the versatility to be extended to other fields, as demonstrated for model building (Martínez et al., 2020). In this article, we provide an extension of Scipion based on a set of tomography plugins (referred to as ScipionTomo hereafter), with a similar purpose: to allow users to be focused on the data processing and analysis instead of having to deal with multiple software installation issues and the inconvenience of switching from one to another, converting metadata files, managing possible incompatibilities, scripting (writing a simple program in a language that the computer must convert to machine language each time the program is run), etcetera. Additionally, having all the software available in an integrated platform allows comparing the results of different algorithms trying to solve the same problem. In this way, the com­ monalities and differences between estimated parameters shed light on which results can be more trusted than others. ScipionTomo is developed by a collaborative multidisciplinary team composed of Scipion team engineers, structural biologists, and in some cases, the developers whose software packages have been integrated. It is open to anyone in the field willing to contribute to this project. The result is a framework extension that combines the acquired knowledge of Scipion developers in close collaboration with third-party developers, and the on- demand design of functionalities requested by beta testers applying this solution to actual biological problems. 1. Introduction revolution that has pushed cryoEM and cryoET towards new paradigms with the elucidation of structures at higher and higher resolutions During the last decade, structural biology has witnessed an authentic (Smith and Rubinstein, 2014). The reasons behind this growth are the * Corresponding authors. E-mail addresses: coss@cnb.csic.es (C.O.S. Sorzano), carazo@cnb.csic.es (J.M. Carazo). https://doi.org/10.1016/j.jsb.2022.107872 Received 28 February 2022; Received in revised form 26 May 2022; Accepted 28 May 2022 Available online 2 June 2022 1047-8477/© 2022 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 recent advances in sample preparation, instrumentation (better micro­ ScipionTomo. It provides a common Graphical User Interface (GUI) for scopes and the introduction of direct electron detectors), hardware all integrated tomography packages. As it has already happened with (more computational capabilities), and software. SPA, traceability, repeatability, workflow design, agility, and all avail­ Until now, SPA has benefited from these advances given the incre­ able Scipion functionalities are provided out of the box for cryoET image ment in the number of deposited maps in EMDB (Lawson et al., 2016) processing. The main GUIs consist of a graph that connects the different with resolutions of 3 Å or better. CryoET has also taken advantage of this steps of the pipeline, forms to be filled with the parameters of each revolution, but it has been a step behind in this regard. This is caused by method, generic and specific result viewers, and finally the capture of all inherent problems of the cryoET acquisition technique that requires the logging messages written out during execution. dealing with low Signal-to-Noise-Ratio (SNR) images affected by the ScipionTomo covers the complete cryoET pipeline, starting from missing wedge problem and with a difficult correction of the CTF (Wan steps (referred to as protocols hereafter) to directly import the tilt series et al., 2016; Fernandez, 2012) complicating the image processing. movies via the mdoc files produced by SerialEM (Mastronarde, 2005), Moreover, the 3D nature of cryoET data, makes its workflow to be highly passing through subtomogram averaging (STA), to tilt series refinement computationally demanding and requires interoperability and analysis protocols. Each of the protocols provides its own help text for each of its of the intermediate results at the different steps. Fortunately, this situ­ parameters and optionally a link to a specific url describing its usage and ation is changing, and many developers, who traditionally worked on the related citations for a better understanding. There are currently SPA methods, are now focussing their efforts on the development of new more than 100 protocols specifically integrated for cryoET from 15 and better image processing methods for cryoET. plugins, and more integrations are coming (Fig. 1). In this work, the There is a broad variety of software for cryoET: Imod (Mastronarde current state and some highlights of ScipionTomo are described, and Held, 2017), Dynamo (Castaño-Díez et al., 2012), Eman (Chen et al., including: 2019), Relion (Bharat and Scheres, 2016), Xmipp (de la Rosa-Trevín et al., 2013), PySeg (Martinez-Sanchez et al., 2020), emClarity (Himes • Interoperability among all integrated software (with some limita­ and Zhang, 2018), Peet (Heumann et al., 2011), PyTom (Hrabe et al., tions explicitly mentioned in “Current status and future work” 2012) and M (Tegunov et al., 2021) to list a few. Only some of them can section) cover the whole processing pipeline. Some of them provide some • The ongoing development of protocols for alignment, picking, and export/import functionality, like M that can generate star files ready to STA to get a refined result using the integrated software packages. be used with Relion, or in some cases, there are third-party scripts to do • Modern internally developed viewers to cover specific visualization the same for other packages. These kinds of integrations are important needs in cryoET, like segmentation, oriented 3D coordinates, or tilt but limited in scale, usually interfacing only one or 2 software packages series CTF analysis. and in some cases through scripting, losing all the traceability. To use the rest, files need to be moved around the filesystem, in some cases From the users’ perspective, they will find a framework in which all creating specific metadata files manually or programmatically, and the required tools covering the whole data processing pipeline for cry­ make extensive use of the command line. Additionally, one will need to oET are integrated into the same software environment that has already deal with the installation of each of the software packages. Even if one been used for SPA and model building. Hence, they can forget about: manages to handle all this heterogeneity, for the sake of traceability and bookkeeping, all the steps done will require a strict annotation mecha­ • Dealing with metadata and image file conversions. nism to be able to report them in a future scientific article. Finally, using • Managing incompatibilities between different software packages. a single algorithm to estimate any parameter (the angular orientation of • Scripting a subtomogram, the defocus at a particular region of a tilted image, etc.) • Losing track of the completed processing attempts. does not allow determining if the estimated parameter is approximately • Using a logbook. correct or not. We may assess its correctness by comparing this esti­ mation to other estimations performed by alternative algorithms solving From the software developers’ perspective, Scipion offers a robust the same problem. These complications are the ones addressed by and maintained platform where they can integrate their software Fig. 1. Tomography software currently integrated into Scipion, organized by cryoET data processing pipeline stage (tomography reconstruction, subtomogram averaging, or other tools), and by processing step (from movie alignment to map back). 2 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 packages, making them more accessible to the community and offering a 2.2. Tilt series movies alignment better user-friendly experience. A “channel” to reach cryoET users and a versatile installation engine to free them from following, sometimes not Once the tilt series movies have been imported into Scipion, the first easy, installation instructions. step is to align the movies of each angular stack to get the tilt series. This From this point on, the Scipion plugins will be referred to as scipion- step corrects the beam-induced movement or ice doming due to the em-pluginname, for instance, the imod plugin will be denoted by sci­ interaction of the electron beam with the sample, which results in blurry pion-em-imod. images when averaging all the frames acquired for the same angle. Two plugins offer a protocol to go through this operation: “tilt series flex 2. The cryoET pipeline align”(Střelák et al. 2020) from scipion-em-xmipptomo and “tilt series motion correction” (Zheng et al., 2017) from scipion-em-motioncorr. In this section, all the steps which compose the cryoET data pro­ cessing pipeline will be briefly described. With each step, we also briefly 2.3. CTF estimation and correction present all corresponding protocols in ScipionTomo that can be used. Fig. 2 shows an illustrative overview of the steps which compose a This step attempts to model the image formation of the electron generic cryoET data processing workflow. microscope, in particular, the defocus and other aberrations that the microscope introduces in the acquired images. The CTF estimation can 2.1. Data and metadata import be carried out with ctffind4 for tilt series (Hu et al., 2019) from the scipion-em-cistem plugin, gctf (Zhang, 2016) from scipion-em-gctf, The first step is to import (registration process) data (images) and/or ctfplotter from the scipion-em-imod plugin, or with scipion-em- metadata (information on the images) in Scipion. Both are defined as a emantomo. For CTF objects, the Scipion team has developed a tool ScipionTomo object. These objects are generic representations of each of called CTF Estimation Assistant (see Fig. 3) which allows the user to the data involved in a tomography pipeline, such as “Tilt series”, determine which of the CTF estimations seems unreliable, by calculating “Subtomogram”, “CTF”, etc. Scipion offers a set of data import protocols, the deviation of the defocusU and the defocusV values for a tilt angle are at least one for each kind of object. Thus, Scipion offers a way to enter at lower than 20% (default value) respect to the mean value for the tilt different points of the whole pipeline, starting, for example, from the series. More complex metrics are planned to be implemented in the aligned tilt series if they have been calculated before. Sometimes there is future. This tool consists of a dynamic viewer which displays all the data more than one import protocol for the same object, for example for regarding the tilt series on which the CTF has been estimated, with the subtomograms. It is possible to load particles obtained with Dynamo relevant data and the advice of the assistant per tilt image, as well as the (represented with.em files and a Dynamo table), with Relion (expressed corresponding defocus, tilt angle, and resolution plot. Hence, the as.star files), or any other native file from the integrated packages. manual estimation can be carried out only for the tilt series in which the Nevertheless, the natural entry point to Scipion is considered to be automatic estimation has not been good enough, improving the effi­ importing directly the data acquired from the microscope, the so-called ciency of the data processing. The user can accept or refuse the proposed “tilt series movies” (a stack of frames acquired for each tilt angle), using estimation result, and the ‘good’ and ‘bad’ subsets are directly generated the mdoc files generated by SerialEM (The SerialEM Home Page). In the by the CTF assistant. future, other formats like the one produced by Tomo5 TFS (Thermo Fisher Scientific microscopes for electron tomography (Thermo Scien­ tific - Tomography 5 Software) could be supported. 2.4. Tilt series alignment Then, the tilt series alignment step should be carried out. This step Fig. 2. Schematic representation of a “clas­ sical” cryoET workflow. Partial tilt series movies (multi-frame images) showing only 3 different tilt angle acquisitions [a]. A beam- induced motion correction method should align all frames to get tilt series images [b]. A hypothetical 19 angles tilt series representa­ tion is presented in [c]. Tilt series alignment methods will estimate the transformation matrices to align the whole tilt series and render an aligned tilt series [d]. Tomogram reconstruction methods will generate a tomogram [e]. Tomogram segmentation and picking methods will help to identify 3D co­ ordinates of interest [e - red circles]. Tomo­ gram extraction methods will extract the 3D boxes (subtomograms) containing the parti­ cles of interest [f]. STA methods will calcu­ late a final 3D average [g]. 3 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 3. CTF Estimation assistant in Scipion. The main view can be observed in [a], in which the left panel shows the relevant data of each tilt series and its cor­ responding tilt images, with the corresponding status and the values of the quality indicators calculated, such as the defocus stability through the tilt angle values. The right panel is used to display the corresponding defocus-tilt angle-resolution graph. [b] and [c] correspond, respectively, to an ’ok’ and ’no ok’ tilt image result, obtained from images acquired at 48 deg. and 54 deg. (outlier) angles. This CTF view replaces the quality indicators when clicking on a specific tilt image. Fig. 4. Examples of scipion-em-tomoviz plugin for 3D visualization. PySeg filaments of a vesicle are shown overlapping with a slice of a cropped tomogram [a]. A membrane vesicle showing the orientations of the picked particles overlapping with a slice of a cropped tomogram [b]. The surface of 3 segmented vesicles using TomoSegMemTV rendered on a tomogram [c]. aims to correct the misalignment introduced in the images as a result of alignment”) and manual alignment, which can be done with the “Imod - tilting the sample (shifts and rotations of the sample holder) from one tilt etomo interactive” protocol where you can use patch alignment and any image to another. The plugin scipion-em-imod offers two approaches for other functionality eTomo provides. Another protocol for tilt series aligning tilt series: automatic tilt series alignment from imod (Mas­ alignment is the eman “tilt series alignment”, from the plugin scipion- tronarde and Held, 2017) (“tilt series fiducial alignment”, which re­ em-emantomo, which, in contrast to imod, does not require fiducials quires gold bead fiducials in the sample and it is composed of protocols to perform the alignment. “xcorr pre-alignment”, “generate fiducial model”, and “fiducial Optionally, the scipion-em-aretomo plugin can also be used to align 4 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 and reconstruct a tomogram or just provide the alignment parameters interest in a tomogram. Here there are two possible scenarios: the first for a later reconstruction with other software. Finally, the plugin one consists of identifying ‘isolated’ macromolecules while the second scipion-em-tomo3d has a protocol that aligns the tilt series and re­ one applies to particles that are tethered to a bigger biological entity, constructs the tomogram, called “motion compensated reconstruction” like membrane proteins. For the first scenario, ScipionTomo offers four (Fernandez et al., 2019). This protocol also attempts to correct the picking protocols, three of which are from the scipion-em-emantomo doming effect in the ice as a consequence of the radiation from one tilt plugin: manual picking, template matching-based picking, and semi- image to the next one. It should be noted that other imod protocols can automated deep-learning-based picking. The fourth protocol is Deep­ be used to pre-process a tilt series before the alignment, such as “x-ray Finder(Moebel et al. 2021), a semi-automated solution that employs a eraser”, “dose filter” or “exclude views”, all of them included in the deep neural network to learn a model for macromolecule species and plugin scipion-em-imod. their cellular environment from annotations. For the second scenario, there are two protocols integrated into ScipionTomo. The “manual 2.5. Tomogram reconstruction vectorial picking” protocol from Dynamo and the picking from PySeg, which is automatic and is based on computing spatially embedded Once the tilt series are properly aligned, they can be reconstructed to graphs with a previous segmentation and annotation of the membranes. obtain the 3D representation of a sample, i.e., a tomogram. Many For these operations, ScipionTomo also provides the TomoSegMemTV tomogram reconstruction protocols are integrated into ScipionTomo (Martinez-Sanchez et al., 2014) and membrane annotation protocols from the following plugins: respectively, both from the plugin scipion-em-tomosegmemtv and a segmentation protocol from the DeepFinder plugin. • scipion-em-emantomo: automatic tomogram reconstruction using direct Fourier inversion using fiducials or high contrast regions. • scipion-em-imod: uses imod WBP (Weighted Back-Projection) based 2.8. Picking pruning tools tomogram reconstruction. • scipion-em-novactf: uses novaCTF (Turoňová et al., 2017) a WBP Picking output can be further refined with some post-picking oper­ based tomogram reconstruction with local CTF correction. ations, such as “picking consensus”, “remove duplicates”, and “filter by • scipion-em-tomo3d: uses tomo3d to execute SIRT (Simultaneous normal” (only for oriented coordinates), which are the protocols offered Iterative Reconstruction Technique) based reconstruction or WBP by the scipion-em-tomoviz plugin. The picking consensus protocol can (Weighted Back Projection) based reconstruction (Agulleiro and be used to determine the agreement between different particle picking Fernandez, 2011; Agulleiro and Fernandez, 2015). As it was algorithms. It takes different sets of 3D coordinates obtained from mentioned before, it also provides a protocol to directly align and different softwares. These outputs can then be combined to remove reconstruct a tomogram, which uses TomoAlign algorithms (Fer­ false-positively picked particles, using multiple criteria, such as toler­ nandez et al., 2019; Fernandez et al., 2018; Fernandez and Li, 2021). ance radius. The “remove duplicates” protocol works on the coordinates • scipion-em-aretomo: it performs both tilt-series alignment and that are closer than a given distance threshold. The resulting coordinate tomogram reconstruction in one protocol (Zheng et al., 2022). The is the average of the coordinates which are closer to each other than the reconstruction can be carried out with SART (Simultaneous Alge­ provided threshold. In the case of oriented particles, like membrane braic Reconstruction Technique) or WBP. particles, the directional picking (coordinates and angles) can also be refined using the protocol “filter by normal”. This protocol takes the The user can choose the protocol that better works for the data which membrane surface and the particles and filters them by different criteria is being processed or even combine several protocols in the same project. related to the direction normal to the membrane, such as a specified tilt For example, SIRT and SART-based methods offer very good contrast for angle or the normal direction of the coordinate to the vesicle surface, picking, but the particles can be then extracted from the corresponding within a specified angular tolerance. If the user has a set of coordinates tomogram reconstructed with the local CTF correction (novaCTF) with orientation, but not the surfaces or meshes corresponding to their correction or any other reconstruction algorithm. membranes, these surfaces can be created from the oriented coordinates by using the protocol “fit vesicles” from the plugin scipion-em- 2.6. Tomogram denoising xmipptomo. Another way of pruning directionally picked particles is by applying An available optional step could be denoising reconstructed tomo­ 2D structural classification with PySeg. Firstly, 3D subvolumes are grams to enhance their visualization. This can be of special interest to flattened by radial averaging along their in-plane (locally tangent to the visualize cellular environments or to facilitate the identification of membrane surface) axis, thus obtaining 2D particle representations with particles for a later picking step. Carrying out this step or not, depends higher SNR and neglecting misalignment around in-plane angles. Sec­ mostly on the data being processed: for example, for very small particles, ondly, these 2D images are classified into structurally homogeneous it is recommended not to denoise the tomograms, because the small groups by the Affinity Propagation algorithm, this algorithm is unsu­ particles could be removed in the denoising calculations. On the other pervised as it does not require the number of input classes, nevertheless, hand, these kinds of methods may be very useful for cellular environ­ K-means and Hierarchical Clustering can also be used. Finally, the user ment analysis. But it depends on each dataset. Scipion integrates three manually selects the most promising classes based on their 2D averaged denoising algorithms, Tomoeed (Moreno et al., 2018) and Tomobflow appearance (more information in Supplementary material 2). (Fernandez, 2009), from plugin scipion-em-tomo3d, and a deep- learning-based approach named cryoCARE (Buchholz et al., 2019), from a plugin with the same name. Note that the last method requires 2.9. Subtomogram averaging (STA) and per-particle per-tilt refinement pairs of tomograms reconstructed from the tilt series generated using only the even and odd frames of the tilt series movies, respectively. This section lists the available software methods available to get a Supplementary material 1 explains how to get these pairs of even and final 3D average. There are 2 main approaches: STA and per-particle per odd tomograms with ScipionTomo. tilt refinement. STA deals with subtomogram 3D volumes as input par­ ticles whereas a per-particle per-tilt approach goes back to the tilt series 2.7. Tomogram particle picking to make a finer refinement as described in (Pyle and Zanetti, 2021). There are three plugins in ScipionTomo that offer both STA and per- The picking step is the process of identifying the complexes of particle per-tilt protocols. These are: 5 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 • scipion-em-dynamo: it offers protocols for particle extraction, sub- Finally, the scipion-em-continuousflex plugin provides a method for boxing (locating the position of subunits of interest), subtomogram synthesizing subtomograms with or without conformational heteroge­ principal component analysis (PCA), and multireference alignment neity, which can be used to synthesize data for testing tomography (MRA), which aligns each particle against a group of different model methods under development. templates. • scipion-em-emantomo: it contains another particle extraction pro­ 2.11. Post-processing tocol, an initial model generation protocol, a subtomogram average refinement protocol, and a per-particle per tilt refinement. 2.11.1. Local resolution estimation • scipion-em-reliontomo: Relion 4 (RELION) is the new version of the The plugin scipion-em-xmipptomo contains a protocol for estimating popular and powerful software package for cryoEM, which has the local resolution in tomography, called MonoTomo (Vilas et al., expanded its functionality to cryoET. The plugin includes most of the 2020). This protocol extends the SPA algorithm of MonoRes (Vilas et al., functionality but leveraging the advantages of Scipion to offer a 2018) adapted to address the problems specific to tomography, such as generalized entry point to it, using the data model and conversion the spatially variant noise or the large size of the tomograms. It requires functionalities from Scipion to allow the user to go into the scipion- two tomograms, one obtained from a tomogram reconstructed using the em-reliontomo plugin no matter the software package used to esti­ even frames of the movies that compose the tilt series and the other one mate the CTF or to align the tilt series and generating automatically obtained using its corresponding odd frames, called odd and even to­ all the required star and any other auxiliary files, such as the order mograms. Alternatively, the odd and even tomograms can be recon­ list files. Another advantage of using Relion inside Scipion is the structed with the odd and even tilt series images, but this practice is less graphical traceability. The workflow proposed by Relion is large and optimal due to the angular sampling being twice the original one. The iterative. The solution proposed by Relion consists of using a star file odd and even tomograms can be obtained with ScipionTomo as called optimisation_set which tracks the required files from one step explained in supplementary material 1. Output can be seen in Fig. 5. to another, but as the number of steps increases, it may be over­ An immediate application of local resolution estimations is the pos­ whelming to the user. In turn, graphic traceability can provide a sibility of filtering the picked subtomograms according to their local clearer perspective and comprehension of the whole procedure. The resolution average, as a pruning step. The result of the filter is a set of last advantage is related to another canonical Scipion feature, which subtomograms with a given range of local resolution, allowing the se­ is providing a set of visualization tools, hence, the user can use any of lection of the subtomograms with the highest quality. This kind of them without having to deal with data conversion, compatibility pruning is implemented in Scipion in the scipion-em-xmipptomo by the issues, etc. “filter coordinates by map” protocol. 2.10. Heterogeneity 2.11.2. Subtomogram subtraction Subtomogram subtraction is implemented in the plugin scipion-em- 2.10.1. Continuous conformational variability analysis xmipptomo as a post-processing tool. This protocol subtracts a sub­ In cryoET, continuous conformational transitions of biomolecular tomogram average from a set of subtomograms, which are internally complexes can be both a huge obstacle to high-resolution subtomogram aligned and numerically adjusted to obtain reliable results (Fernández- averaging and a unique opportunity to describe, at once, multiple bio­ Giménez et al., 2021). A mask can be provided if the user wants to molecular conformations relevant to the given environment (purified perform the subtraction within a specific region, for example, to subtract samples or cells in health and disease). HEMNMA-3D (Harastani et al., the membrane of a set of particles that are attached to it. 2021) and TomoFlow (Harastani et al., 2022) from the scipion-em- continuousflex plugin allow analyzing subtomograms in terms of 2.11.3. Map back continuous conformational variability. Both methods represent a given Once STA is complete, you may want to go back to the tomogram and set of subtomograms in a common low-dimensional space based on the place the obtained average into the cellular context (Fig. 6). This “map- motion field obtained between each subtomogram and a given refer­ back” operation can be done with the “map back” protocol from scipion- ence. HEMNMA-3D determines the motion field by displacing the em-xmipptomo. reference using normal modes, with the normal-mode amplitudes calculated by matching the subtomogram with the displaced reference. 2.12. Visualization tools In TomoFlow, the motion field is described by the optical flow between the subtomogram and the reference. For the reference, both methods can ScipionTomo offers a complete set of visualization tools, most of use an EM map such as subtomogram average (for normal mode anal­ them coming from the integrated software, such as IMOD − 3dmod for ysis, HEMNMA-3D will internally use a 3D Gaussian-based representa­ tilt series and tomograms, DeepFinder and Eman viewers for tomograms tion of this EM map) and HEMNMA-3D can optionally use an atomic and 3D coordinates, xmipp data viewer for tilt series, tomograms and reference. The obtained low-dimensional conformational space can then subtomograms to name a few. But there is also a set of internally be explored in terms of subtomogram averages obtained by interactive developed 3D viewers, part of the scipion-em-tomoviz plugin, which can grouping of subtomograms with similar conformations (closest points in be used to render in a 3D space tomograms, subtomograms, coordinates, the densest regions in this space) or in terms of animated displacement normals, vesicles, graphs and filaments. Fig. 4 illustrates some examples of the reference model along the densest regions in this space. of the plugin scipion-em-tomoviz visualization tools. Furthermore, the scipion-em-continuousflex plugin contains methods originally developed for analyzing continuous conformational 3. Workflow examples variability in single-particle cryoEM images (analysis of images with normal modes of a reference by HEMNMA (Harastani et al., 2020; Jin 3.1. Workflow 1: From tilt series to particle picking et al., 2014) and for deriving atomic models from cryoEM maps (flexible fitting of an atomic structure into an EM map using NMMD (Vuillemot Fig. 7 illustrates the versatility offered by Scipion and different et al., 2022) that combines normal modes and molecular dynamics walkthroughs that can be followed to go, in this case, from the tilt series simulation). These two single-particle cryoEM methods (HEMNMA and to the tomogram reconstruction, including the CTF estimation and NMMD) could also be used for analyzing 2D tomography data (e.g., correction. The steps of the workflow are labeled with a different color projections of subtomograms) and for deriving atomic models from for each different plugin is used. Thus, each box represents a different high-resolution subtomogram averages, respectively. protocol that belongs to a specific plugin. Although the import and the 6 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 5. Example of xmipptomo-monotomo estimation of the local resolution of a tomogram reconstructed from TS-G8-box7-001 tilt serie, from the dataset EMPIAR- 10364. Showing [a] the Z (top), X (right) and Y (bottom) slice axis and [b] the color scale ranging from 50 to 140 Å resolution. Fig. 6. HIV capsids subtomogram average from EMPIAR-10643 mapped back to their native locations in the tomogram using either Imod viewer (on the left) or represented in a 3D space using ChimeraX (on the right). alignment of the raw movie files obtained from the microscope are not plugin scipion-em-imod called “apply transformation” is used. This is included in this workflow, they are fully covered in Scipion, as explained because Scipion stores the cumulated transformations carried out over in previous sections. the data in a 3D transformation matrix (shifts and angles) to save disk The current reconstruction workflow was carried out on EMPIAR- space and avoid artifacts that might be generated due to several in­ 10453 (Iudin et al., 2016) dataset, which provides the tilt series and the terpolations over the same images. The interpolation operation of acquisition information. They were imported into Scipion using the applying the transformation matrix to the tilt series is performed inter­ plugin scipion-em-tomo. Focusing now on the fiducial alignment stage, nally in each protocol that requires it. Thus, this protocol offers the user it was mainly carried out using protocols from the plugin scipion-em- the possibility to check the quality of the alignment by applying the imod. transformation matrices. Note that although not shown in this example, Before aligning the tilt series, they were pre-processed through steps tilt series alignment could have been done with aretomo or eman instead of X-ray erasing, dose filtering, and tilt series normalization (binning). of imod. Then, the automatic tilt series alignment was carried out with “xcorr Going now to the CTF estimation, there are different protocols pre-alignment”, the “fiducial model generation”, and the “fiducial available to carry out this operation. Only the “tilt-series gctf” protocol alignment” protocols. We recommend attempting the automatic align­ was used in later reconstructions but any other could have been used. ment at first, as shown in this workflow because bad alignments can be The other estimations have been performed and included in the Fig. 7 to corrected manually later with the eTomo protocol or equivalent and illustrate the versatility and interoperability offered by Scipion. joined back to the initially well-aligned set of tilt series. Moving back to Now that the CTF has been estimated, it is possible to use this in­ the fiducial alignment, in this example, we have used two different al­ formation as an input to other algorithms to obtain a more accurate gorithms to reconstruct tomograms without any CTF estimation from reconstruction of the final tomogram. There are two main possibilities scipion-em-imod and scipion-em-tomo3d. It can be observed that before for this: (i) scipion-em-imod and (ii) scipion-em-novactf. The first one reconstructing with scipion-em-tomo3d, another protocol from the performs a 2D CTF correction of the tilt-series for a posterior 7 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 7. Example of workflow to go from the tilt series to tomograms reconstruction. A different color was used to represent the different plugins used. Different protocols to estimate the CTF (imod, ctffind4, gctf, and emantomo) and reconstruct the tomograms (imod, novaCTF and tomo3d) are shown, together with other preprocessing, auxiliary, or data managing methods, illustrating the interoperability and versatility offered by Scipion. reconstruction that can be used with any other classical reconstruction The second stage is carried out with the plugin scipion-em- algorithm offered by ScipionTomo. The second and more accurate op­ tomosegmemtv. In the first step, the membranes are segmented and tion performs a 3D CTF corrected reconstruction, which corrects the CTF then manually annotated using the Membrane Annotator tool, which is during the reconstruction process, considering the coordinates of each part of the same plugin. Finally, the segmented and annotated data are voxel, rendering results like those shown in Fig. 8. resized to their previous size for the correct picking of membrane particles. The third stage is composed of four steps of PySeg directional pick­ 3.2. Workflow 2: Directional picking ing, each performed by a different protocol of the scipion-em-pyseg plugin. The first step is the “preseg”, which is used to annotate each In this section, we show a workflow that illustrates how to pick vesicle membrane and the inner and outer surroundings. Then, the membrane proteins (in this case membrane-associated ribosomes) using graphs are calculated and refined into the filaments, which are finally segmentation, annotation, and directional picking protocols inside the used for the picking. Scipion framework (see Fig. 9). It was carried out on the dataset EMD- The fourth stage corresponds to the picking pruning. 3D coordinates 10439 (Lawson et al., 2016). Different colors have been used for are first grouped into geometrical related ellipsoids with “fit vesicles” different plugins to illustrate the interoperability, traceability, and leaving out those far from the geometry, from the scipion-em- customizability offered by Scipion. The small diagram in the center is xmipptomo plugin. Later, the “filter by normal” protocol, from the the whole workflow, which has been split into two halves for visuali­ scipion-em-tomoviz plugin, removes any oriented coordinate whose zation purposes. normal is tilted more than N degrees from the perpendicular to the We can group the protocols that compose this workflow into five membrane. stages: (i) data preparation, (ii) membrane segmentation, (iii) direc­ The final stage is the particle extraction, made with the scipion-em- tional picking, (iv) particle filtering, and (v) particle extraction. emantomo plugin, and the 2D classification carried out with the scipion- The first stage, data preparation, involves importing the tomograms em-pyseg plugin. The classification protocol requires a mask to calculate with the plugin scipion-em-tomo and binning them with the protocol a rotational average in the desired area where the particle is located. A “imod - Tomo normalization” from the plugin scipion-em-imod. The cylinder mask was created with the scipion-em-xmipp plugin to cover binning was carried out to reduce the execution time of the posterior the particle, the membrane, and a small part of the inner surroundings. denoising, segmentation, and annotation steps. At this stage, the Finally, a subset of the particles that belongs to the best class was tomogram is also denoised with the plugin scipion-em-tomo3d. 8 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 8. Results obtained for some of the steps carried out to reconstruct TS_293 of dataset EMPIAR-10453 [a] and [b] correspond to the original tilt series and pre- processed tilt series, respectively, being [b] the result of the tilt series X-ray erasing dose filtering, and downsampling to bin 4. Sub-figure [c] is the result of the automatic fiducial alignment offered by plugin scipion-em-imod, and sub-figure [d] is the result of the SIRT reconstruction performed with plugin scipion-em-tomo3d without CTF estimation in this case (it corresponds to the green box named “Imod - Tomo reconstruction (wbp)” in Fig. 7). All the subfigures represent data displayed with viewer IMOD-3dmod. selected using the “Xmipp Dataviewer”. Intermediate results can be seen the picked coordinates. The tilt series should be at bin 1, and the given in Fig. 10. coordinates are scaled according to the ratio of the sampling rate of the tomograms in which the coordinates were picked, in case they were at a different binning. To get the final model at bin 1, the tutorial proposes a 3.3. Workflow 3: Subtomogram averaging three iteration loop decreasing the binning factor from 4 to 2 and finally to 1. Each iteration step, as shown in Fig. 11, consists of generating the The dataset used for this STA workflow corresponds to the one used so-called pseudo-subtomograms, which are subtomograms generated in the STA tutorial proposed in the documentation of Relion4 (Sub­ directly from the tilt series. The refined pseudo-subtomograms are then tomogram tutorial). It goes from the tilt series and particle coordinates used to reconstruct the particle decreasing the binning and, after that, to the final model, computing iterative particle reconstructions and re­ the next iteration begins repeating those steps with the new binning finements from bin 4 to bin 1. At that point, the so-called “per particle factor. The tutorial dataset provides some centered masks at different per tilt” (pppt) methodology is applied to improve the resolution of the binning factors which can be used as a reference to estimate the Z shift model. The dataset is a subset of EMPIAR-10164, composed of the five that may be necessary to be applied to the refined pseudo- tilt series named TS_01, TS_03, TS_43, TS_45, and TS_54. The tutorial subtomograms. Scipion core offers a protocol named “edit set” that dataset also provides the TS alignment data estimated with IMOD- can be used to carry out this operation. etomo, the CTF estimation is performed with ctffind4 and the co­ Once we have obtained a model at bin 1, the so-called tomo refine­ ordinates are picked, but without specifying the picking algorithm used. ment cycle can be carried out. It consists of multiple iterations of the From Scipion, it is possible to take advantage of part of the precomputed pppt protocols, following the sequence: reconstruction at bin 1 with an data or to start from the tilt series movies and carry out all the steps FSC mask (provided) to generate an FSC reference apart from the required until the tilt series alignment, the CTF estimation, and the reconstruction, CTF refinement, make a new set of pseudo- particle picking. As can be observed in Fig. 11, the CTF estimation was subtomograms considering the refined defocus, reconstruct again to calculated with the protocol tilt series ctffind4 from the plugin scipion- get new halves and an improved FSC reference and frame alignment to em-cistem, using the same parameters as those in the defocus files generate the particles motion data. The cycle iteration finishes with provided with the dataset (this step of CTF recomputing is included just another generation of pseudo-subtomograms and another reconstruc­ to further show Scipion versatility). tion whose results will be used to feed an auto-refine protocol. The loop Once all the data is imported and/or generated into Scipion, the first can be repeated as many times as desired until the resolution stops step of the Relion4 workflow is covered by protocol “prepare data for increasing. Intermediate results are shown in Fig. 12. Relion 4”, which will generate the input to relion4 given the CTF esti­ mation, the unaligned tilt series, the alignment transformation data, and 9 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 9. Directional picking workflow to pick membrane-associated ribosomes of an in situ tomogram (EMD-10439). The workflow in the middle represents the whole workflow, while the images on both sides are zoomed-in parts of it to provide better visualization. Different colors were used to represent each plugin involved. The color legend is on the bottom left. The different stages described in the text are (i) boxes named “tomo - import tomograms” (in brown), “Imod - Tomo normalization” (in yellow), and “tomo3d - denoise tomogram” (in orange), (ii) the 3 blue boxes named “tomosegmemtv - tomogram segmentation”, “tomosegmemtv - annotate membranes” and “tomosegmemtv - resize segmentation”, (iii) the green boxes named “pyseg - preseg membranes”, “pyseg - graphs”, “pyseg - fils” and “pyseg - picking”, (iv) boxes named “tomoviz - remove duplicates” (in red), “xmipptomo - fit vesicles” (in light blue) and “tomoviz - filter by normal” (in red), and (v) boxes named “emantomo - subtomo extraction” (in dark yellow), “xmipp3 - create 3d mask” (in light red), “pyseg − 2D classification” (in green) and “class1” (in pink). 3.4. Workflow 4: The “SPA leap” join. Beta-testers are also an important part, as they are carefully pro­ cessing actual datasets and providing critical feedback for future stable Given the maturity of the existing methods of SPA and the integra­ releases. Scipion can be extended through plugins. At any time, any tion of most of them in Scipion, it is very easy to make use of SPA picking software developer can create a plugin and add more tools to this cat­ and pruning methods. In this example (Fig. 13) we have used EMPIAR- alog, and therefore more possibilities and options to get the best possible 10045: Aligned tilt series images of S. cerevisiae 80S ribosomes. Any 3D result. This is the case for Deepfinder, cryoCARE, pySeg, and Tomo­ picker will produce false positives, which have to be removed. It is here SegMenTV, where their developers’ close collaboration has been key in where we made a “SPA leap” to prune and get an initial volume. “xmipp providing the best possible integration. subtomo projection” from the scipion-em-xmipptomo plugin was used to project all the subtomograms on their Z-axis and generate the same 5. Current status and future work number of 2D particles. Then, we added a zero defocus CTF with “xmipp3 simulate CTF” as CTF parameters are requested by cryosparc To this date, we consider ScipionTomo is in Beta state. This means is “2D classification” (Punjani et al., 2017) and “ab-initio” protocols. stable enough to be carefully relesed upon request. We welcome any Particles from the “good classes” were manually selected with “Xmipp betatester volunteer and we offer direct communication to all the de­ data viewer” producing the “good ones manual” set that was used to velopers involved through our slack workspace. Although most of the prune the original set of subtomograms using the “tomo − 2D particles pipline is covered we are currently focused on per particle per tilt to subtomograms” protocol. Now, back into the 3D space, a “relion 3D methods (Relion 4, eman). One limitation we currently face is the classification” was used to get all the pruned subtomogram set aligned inability to mix per particle per tilt methods from different software. into a single class, feeding it with the initial volume generated by cry­ Providing a reference intial model from eman to relion or viceversa osparc in the SPA pruning leap. Additionally, another SPA leap can be should not be an issue, but mixing steps is still not possible. Additionally made to use SPA picking methods as explained in supplementary ma­ we need to tests the pipeline when special cases occur during re­ terial 3. constructions like “views exclusion”, “”offset reconstructions” and later on pppt approaches are used. 4. A collaborative software project Currently development efforts are the following: ScipionTomo is part of the bigger project “Scipion’’ that aims to • Streaming mode (outputting results on the fly), which would allow provide a free, open-source, high-quality smart workflow engine. Many the facilities to evaluate the results during the acquisition. developers are contributing to this project and everyone is welcome to 10 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 10. Graphical walkthrough over directional picking results: imported tomogram displayed with IMOD-3dmod [a], denoised tomogram with the plugin scipion- em-tomo3d and displayed with IMOD-3dmod [b], membrane segmentation with the plugin scipion-em-tomosegmemtv, displayed with IMOD-3dmod [c] and manual annotation of the segmented membranes, carried out and displayed with the Membrane Annotator tool from the plugin scipion-em-tomosegmemtv [d]. Sub-figures [e] to [h] show the results of the preseg, graphs, filaments, and picking, respectively, steps of the directional picking procedure with PySeg. Preseg results in [e] are displayed with xmipp-Dataviewer and the rest, for the membrane labeled as 3, with viewer tomoviz from the plugin scipion-em-tomoviz. Sub-figure [i] represents the best 2D class obtained with the 2D classification from the plugin scipion-em-pyseg, represented as a rotational average around the green axis (which is why the class representative is half of the particle box). Sub-figure [j] shows a volume resulting from averaging picked particles after some STA steps (carried out with the plugin scipion-em-reliontomo, Relion 3 version) in ChimeraX viewer (contained in the plugin scipion-em-chimera). • Aitom (Zeng and Xu, 2019) integration is being co-developed with its covered with several options for each of the steps rendering accurate developers results. • memBrain (Lamm et al. 2022) integration is being co-developed with its developers CRediT authorship contribution statement • surfaceMorphometics(Barad et al. 2022) integration is being co- developed with its developers J. Jiménez de la Morena: Conceptualization, Methodology, Soft­ • emClarity and novaSTA (turonova. turonova/novaSTA: novaSTA., ware, Writing – original draft, Visualization, Writing – review & editing. 2020) are being integrated. P. Conesa: Conceptualization, Methodology, Software, Writing – orig­ • Organizing more courses and tutorials, see supplementary material 4 inal draft, Visualization, Supervision, Writing – review & editing. Y.C. for available training material. Fonseca: Conceptualization, Methodology, Software, Writing – review & editing. F.P. de Isidro-Gómez: Software, Methodology, Formal 6. Conclusions analysis, Writing – review & editing. D. Herreros: Software, Method­ ology, Formal analysis, Visualization, Writing – review & editing. E. In a fast-growing domain with quickly evolving techniques, software Fernández-Giménez: Software, Methodology, Formal analysis, Writing engineers do not wait for standards. Thus, the software is produced in an – review & editing. D. Strelak: Software, Writing – review & editing. E. isolated manner and cryoET is no exception. In this paper, we presented Moebel: Software, Methodology, Formal analysis, Investigation, ScipionTomo, a framework that improves the usability and interopera­ Writing – review & editing. T.O. Buchholz: Software, Methodology, bility of many software packages currently available for cryoET data Formal analysis, Investigation, Writing – review & editing. F. Jug: processing. ScipionTomo is designed to be the glue to bring many of the Software, Methodology, Formal analysis, Investigation, Writing – review available software together, allowing the structural biologist to focus on & editing. A. Martinez-Sanchez: Software, Methodology, Formal a research problem and avoiding completely the need for manual data analysis, Investigation, Writing – review & editing. M. Harastani: conversion, programming, or scripting. As a consequence of the inte­ Software, Methodology, Formal analysis, Investigation, Writing – review gration, Scipion is naturally annotating all the steps done during the & editing. S. Jonic: Software, Methodology, Formal analysis, Investi­ tomography image processing, once more freeing the user from doc­ gation, Writing – review & editing. J.J. Conesa: Validation, Conceptu­ umenting all the steps done during the image processing. As an exten­ alization, Data curation, Writing – review & editing. A. Cuervo: sible framework, any future or not yet integrated software can be Validation, Conceptualization, Data curation, Writing – review & edit­ plugged in and join the ScipionTomo ecosystem. To date, ScipionTomo ing. P. Losana: Validation, Conceptualization, Data curation, Writing – is open upon request for beta testing and has not been widely released review & editing. I. Sánchez: Resources, Writing – review & editing. M. yet. Currently, all steps of the cryoET data processing workflow are Iceta: Resources, Writing – review & editing. L. del Cano: Resources, 11 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 11. STA workflow to get a model of the HIV capsid following Relion 4 tutorial (EMPIAR-10164). Different colors were used to differentiate the three stages proposed, being [a] the whole workflow structure, [b] the data generation and entry point to Relion 4 for tomography, [c] the three iterations, from binning 4 to binning 2, and finally binning 1 of the stage that takes from the tilt series and coordinates to the subtomogram volume, and [d] the first iteration of the so-called tomo refinement cycle. Fig. 12. Intermediate results of the STA workflow to get a model of the HIV capsid following Relion 4 tutorial (EMPIAR-10164), being [a] the tilt series named TS_01 displayed with IMOD-3dmod viewer, [b] the picking of some membrane particles from the same tilt series, displayed with EMAN, [c] and [d] the top and the side view of the volume obtained with the auto-refinement from Relion 4 at bin 1, displayed with ChimeraX as surfaces, and [e] and [f] the same views, but corresponding to the auto-refinement with which the first iteration of the tomo refinement cycle finishes, displayed with ChimeraX as a tilted slab. 12 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Fig. 13. “SPA pruning leap” workflow and results for EMPIAR-10045 dataset. The workflow shows the initial model and 2d classification SPA methods from cryosparc (left). 2d classes obtained during 2d classification, some resembling ribosomes others not (right). Writing – review & editing. M. Gragera: Validation, Writing – review & No. 713673. editing. R. Melero: Validation, Visualization, Writing – review & edit­ “Comunidad Autónoma de Madrid” through Grant: S2017/BMD- ing. G. Sharov: Conceptualization, Methodology, Software, Writing – 3817. review & editing. D. Castaño-Díez: Software, Methodology, Formal European Union (EU) and Horizon 2020 through grant: analysis, Investigation, Writing – review & editing. A. Koster: Valida­ HighResCells (ERC - 2018 - SyG, Proposal: 810057) tion, Writing – review & editing. J.G. Piccirillo: Validation, Writing – EOSC Life (INFRAEOSC-04-2018, Proposal: 824087) review & editing. J.L. Vilas: Formal analysis, Investigation, Validation, DCD acknowledges funding from the Swiss National Science Foun­ Writing – review & editing. J. Otón: Formal analysis, Investigation, dation (205321 179041) and the Human Frontiers of Science Program Validation, Writing – review & editing. R. Marabini: Formal analysis, (RGP0017/2020). Writing – review & editing. C.O.S. Sorzano: Supervision, Project administration, Funding acquisition, Writing – review & editing. J.M. Appendix A. Supplementary data Carazo: Supervision, Project administration, Funding acquisition, Writing – review & editing. Supplementary data to this article can be found online at https://doi. org/10.1016/j.jsb.2022.107872. Declaration of Competing Interest References The authors declare that they have no known competing financial de la Rosa-Trevín, J.M., Quintana, A., del Cano, L., Zaldívar, A., Foche, I., Gutiérrez, J., interests or personal relationships that could have appeared to influence Gómez-Blanco, J., Burguet-Castell, J., Cuenca-Alba, J., Abrishami, V., Vargas, J., the work reported in this paper. Otón, J., Sharov, G., Vilas, J.L., Navas, J., Conesa, P., Kazemi, M., Marabini, R., Sorzano, C.O.S., Carazo, J.M., 2016. Scipion: A software framework toward integration, reproducibility and validation in 3D electron microscopy. J. Struct. Biol. Acknowledgments 195 (1), 93–99. Martínez, M., Jiménez-Moreno, A., Maluenda, D., Ramírez-Aportela, E., Melero, R., Cuervo, A., Conesa, P., del Caño, L., Fonseca, Y.C., Sánchez-García, R., Strelak, D., The authors acknowledge the economical support from MICINN to Conesa, J.J., Fernández-Giménez, E., de Isidro, F., Sorzano, C.O.S., Carazo, J.M., the Instruct Image Processing Center (I2PC) as part of the Spanish Marabini, R., 2020. Integration of Cryo-EM model building software in Scipion. J. Chem. Inf. Model. 60 (5), 2533–2540. https://doi.org/10.1021/acs.jcim.9b01032. participation in Instruct-ERIC, the European Strategic Infrastructure Smith, M.T.J., Rubinstein, J.L., 2014. Structural biology. Beyond blob-ology. Science 345 Project (ESFRI) in the area of Structural Biology and Grant PID2019- (6197), 617–619. 104757RB-I00 funded by MCIN/AEI/ 10.13039/501100011033/ and Lawson, C.L., Patwardhan, A., Baker, M.L., Hryc, C., Garcia, E.S., Hudson, B.P., Lagerstedt, I., Ludtke, S.J., Pintilie, G., Sala, R., Westbrook, J.D., Berman, H.M., “ERDF A way of making Europe”, by the “European Union”, the support Kleywegt, G.J., Chiu, W., 2016. EMDataBank unified data resource for 3DEM. of the French National Research Agency - ANR (ANR-20-CE11-0020-03 Nucleic Acids Res. 44 (D1), D396–D403. and ANR-19-CE11-0008-01 to S.J.); the Sorbonne University (2019 Wan, W., Briggs, J.A.G., 2016. Chapter Thirteen - Cryo-Electron Tomography and Subtomogram Averaging. In: Crowther, R.A. (Ed.), Methods in Enzymology. “Interface pour le Vivant” PhD scholarship grant to M.H.); and the access Academic Press, pp. 329–367. to the HPC resources of CINES and IDRIS granted by GENCI Fernandez, J.-J., 2012. Computational methods for electron tomography. Micron. 43 (A0100710998, A0070710998, AP010712190, AD011012188 to S.J.). (10), 1010–1030. Mastronarde, D.N., Held, S.R., 2017. Automated tilt series alignment and tomographic The project that gave rise to these results received the support of a reconstruction in IMOD. J. Struct. Biol. 197 (2), 102–113. fellowship from the “la Caixa’’ Foundation (ID 100010434). The Castaño-Díez, D., Kudryashev, M., Arheit, M., Stahlberg, H., 2012. Dynamo: a flexible, fellowship code is LCF/BQ/DI18/11660021. This project has received user-friendly development tool for subtomogram averaging of cryo-EM data in high- funding from the European Union’s Horizon 2020 research and inno­ performance computing environments. J. Struct. Biol. 178 (2), 139–151. vation programme under the Marie Skłodowska-Curie grant agreement 13 J. Jiménez de la Morena et al. Journal of Structural Biology 214 (2022) 107872 Chen, M., Bell, J.M., Shi, X., Sun, S.Y., Wang, Z., Ludtke, S.J., 2019. A complete data Moreno, J.J., Martínez-Sánchez, A., Martínez, J.A., Garzón, E.M., Fernández, J.J., processing workflow for cryo-ET and subtomogram averaging. Nat. Methods 16 (11), Murphy, R., 2018. TomoEED: fast edge-enhancing denoising of tomographic 1161–1168. volumes. Bioinformatics 34 (21), 3776–3778. Bharat, T.A.M., Scheres, S.H.W., 2016. Resolving macromolecular structures from Fernandez, J.-J., 2009. TOMOBFLOW: feature-preserving noise filtering for electron electron cryo-tomography data using subtomogram averaging in RELION. Nat. tomography. BMC Bioinf. 10, 178. Protoc. 11 (11), 2054–2065. Buchholz, T.-O., Krull, A., Shahidi, R., Pigino, G., Jékely, G., Jug, F., 2019. Content- de la Rosa-Trevín, J.M., Otón, J., Marabini, R., Zaldívar, A., Vargas, J., Carazo, J.M., aware image restoration for electron microscopy. Methods Cell Biol. 152, 277–289. Sorzano, C.O.S., 2013. Xmipp 3.0: an improved software suite for image processing Martinez-Sanchez, A., Garcia, I., Asano, S., Lucic, V., Fernandez, J.-J., 2014. Robust in electron microscopy. J. Struct. Biol. 184 (2), 321–328. membrane detection based on tensor voting for electron tomography. J. Struct. Biol. Martinez-Sanchez, A., Kochovski, Z., Laugks, U., Meyer zum Alten Borgloh, J., 186 (1), 49–61. Chakraborty, S., Pfeffer, S., Baumeister, W., Lučić, V., 2020. Template-free detection Pyle, E., Zanetti, G., 2021. Current data processing strategies for cryo-electron and classification of membrane-bound complexes in cryo-electron tomograms. Nat. tomography and subtomogram averaging. Biochem. J. 478, 1827–1845. Methods 17 (2), 209–216. RELION — RELION documentation. Available: https://relion.readthedocs.io/en/release- Himes, B.A., Zhang, P., 2018. emClarity: software for high-resolution cryo-electron 4.0/. tomography and subtomogram averaging. Nat. Methods 15 (11), 955–961. Harastani, M., Eltsov, M., Leforestier, A., Jonic, S., 2021. HEMNMA-3D: Cryo electron Heumann, J.M., Hoenger, A., Mastronarde, D.N., 2011. Clustering and variance maps for tomography method based on normal mode analysis to study continuous cryo-electron tomography using wedge-masked differences. J. Struct. Biol. 175 (3), conformational variability of macromolecular complexes. Front Mol Biosci. 8. 288–299. Harastani, M., Eltsov, M., Leforestier, A., Jonic, S., 2022. TomoFlow: analysis of Hrabe, T., Chen, Y., Pfeffer, S., Kuhn Cuellar, L., Mangold, A.-V., Förster, F., 2012. continuous conformational variability of macromolecules in cryogenic PyTom: a python-based toolbox for localization of macromolecules in cryo-electron subtomograms based on 3D dense optical flow. J. Mol. Biol. 434 (2), 167381. tomograms and subtomogram analysis. J. Struct. Biol. 178 (2), 177–188. https://doi.org/10.1016/j.jmb.2021.167381. Tegunov, D., Xue, L., Dienemann, C., Cramer, P., Mahamid, J., 2021. Multi-particle cryo- Harastani, M., Sorzano, C.O.S., Jonić, S., 2020. Hybrid electron microscopy normal mode EM refinement with M visualizes ribosome-antibiotic complex at 3.5 Å in cells. Nat. analysis with scipion. Protein Sci. 29 (1), 223–236. Methods 18, 186–193. Jin, Q., Sorzano, C., de la Rosa-Trevín, J., Bilbao-Castro, José Román, Núñez-Ramírez, R., Mastronarde, D.N., 2005. Automated electron microscope tomography using robust Llorca, O., Tama, F., Jonić, S., 2014. Iterative elastic 3D-to-2D alignment method prediction of specimen movements. J. Struct. Biol. 152 (1), 36–51. using normal modes for studying structural dynamics of large macromolecular The SerialEM Home Page. Available: https://bio3d.colorado.edu/SerialEM/. complexes. Structure. 22 (3), 496–506. Thermo Scientific - Tomography 5 Software. Available: https://assets.thermofisher.com/ Vuillemot, R., Miyashita, O., Tama, F., Rouiller, I., Jonic, S., 2022. NMMD: efficient TFS-Assets/MSD/Datasheets/tomography-5-software-ds0362.pdf. Cryo-EM flexible fitting based on simultaneous normal mode and molecular Zheng, S.Q., Palovcak, E., Armache, J.-P., Verba, K.A., Cheng, Y., Agard, D.A., 2017. dynamics atomic displacements. J. Mol. Biol. 434 (7), 167483. https://doi.org/ MotionCor2: anisotropic correction of beam-induced motion for improved cryo- 10.1016/j.jmb.2022.167483. electron microscopy. Nat. Methods 14 (4), 331–332. Vilas, J.L., Oton, J., Messaoudi, C., Melero, R., Conesa, P., Ramirez-Aportela, E., Mota, J., Hu J, Yu Y. Polycystic Kidney Disease. CRC Press; 2019. Martinez, M., Jimenez, A., Marabini, R., Carazo, J.M., Vargas, J., Sorzano, C.O.S., Zhang, K., 2016. Gctf: Real-time CTF determination and correction. J. Struct. Biol. 193 2020. Measurement of local resolution in electron tomography. J Struct Biol X. 4, (1), 1–12. 100016. https://doi.org/10.1016/j.yjsbx.2019.100016. Fernandez, J.-J., Li, S., Agard, D.A., 2019. Consideration of sample motion in cryo- Vilas, J.L., Gómez-Blanco, J., Conesa, P., Melero, R., Miguel de la Rosa-Trevín, J., tomography based on alignment residual interpolation. J. Struct. Biol. 205 (3), 1–6. Otón, J., Cuenca, J., Marabini, R., Carazo, J.M., Vargas, J., Sorzano, C.O.S., 2018. Turoňová, B., Schur, F.K.M., Wan, W., Briggs, J.A.G., 2017. Efficient 3D-CTF correction MonoRes: Automatic and Accurate Estimation of Local Resolution for Electron for cryo-electron tomography using NovaCTF improves subtomogram averaging Microscopy Maps. Structure. 26 (2), 337–344.e4. resolution to 3.4Å. J. Struct. Biol. 199 (3), 187–195. Fernández-Giménez, E., Martínez, M., Sánchez-García, R., Marabini, R., Ramírez- Agulleiro, J.I., Fernandez, J.J., 2011. Fast tomographic reconstruction on multicore Aportela, E., Conesa, P., Carazo, J.M., Sorzano, C.O.S., 2021. Cryo-EM density maps computers. Bioinformatics 27 (4), 582–583. adjustment for subtraction, consensus and sharpening. J. Struct. Biol. 213 (4), Agulleiro, J.-I., Fernandez, J.-J., 2015. Tomo3D 2.0 – Exploitation of Advanced Vector 107780. https://doi.org/10.1016/j.jsb:2021.107780. eXtensions (AVX) for 3D reconstruction. J. Struct. Biol. 189 (2), 147–152. Iudin, A., Korir, P.K., Salavert-Torres, J., Kleywegt, G.J., Patwardhan, A., 2016. EMPIAR: Fernandez, J.-J., Li, S., Bharat, T.A.M., Agard, D.A., 2018. Cryo-tomography tilt-series a public archive for raw electron microscopy image data. Nat. Methods 13 (5), alignment with consideration of the beam-induced sample motion. J. Struct. Biol. 387–388. 202 (3), 200–209. Subtomogram tutorial — RELION documentation. [cited 25 Feb 2022]. Available: Fernandez, J.-J., Li, S., 2021. TomoAlign: A novel approach to correcting sample motion https://relion.readthedocs.io/en/release-4.0/STA_tutorial/index.html. and 3D CTF in CryoET. J. Struct. Biol. 213 (4), 107778. https://doi.org/10.1016/j. Punjani, A., Rubinstein, J.L., Fleet, D.J., Brubaker, M.A., 2017. cryoSPARC: algorithms jsb:2021.107778. for rapid unsupervised cryo-EM structure determination. Nat. Methods 14 (3), Zheng, S., Wolff, G., Greenan, G., Chen, Z., Faas, F.G.A., Bárcena, M., et al., 2022. 290–296. AreTomo: An integrated software package for automated marker-free, motion- Zeng X, Xu M. AITom: Open-source AI platform for cryo-electron tomography data corrected cryo-electron tomographic alignment and reconstruction, p. analysis. arXiv [q-bio.QM]. 2019. Available: http://arxiv.org/abs/1911.03044. 2022.02.15.480593 bioRxiv.. https://doi.org/10.1101/2022.02.15.480593. turonova. turonova/novaSTA: novaSTA. 2020. doi:10.5281/zenodo.3973623. 14