Journal of Structural Biology: X

We describe how to use several machine learning techniques organized in a learning pipeline to segment and identify subcellular structures from cryo electron tomograms. These tomograms are difficult to analyze with traditional segmentation tools. The learning pipeline in our approach starts from supervised learning via a special convolutional neural network trained with simulated data. It continues with semi-supervised reinforcement learning and/or a region merging techniques that try to piece together disconnected components that should belong to the same subcellular structure. A parametric or non-parametric fitting procedure is then used to enhance the segmentation results and quantify uncertainties in the fitting. Domain knowledge is used in generating the training data for the neural network and in guiding the fitting procedure through the use of appropriately chosen priors and constraints. We demonstrate that the approach proposed here work well for extracting membrane surfaces of protein reconstituted liposomes in a cellular environment that contains other artifacts.

Cryo-electron tomography (CET) combined with sub-volume averaging (SVA), is currently the only imaging technique capable of determining the structure of proteins imaged inside cells at molecular resolution. To obtain high-resolution reconstructions, sub-volumes containing randomly distributed copies of the protein of interest need be identified, extracted and subjected to SVA, making accurate particle detection a critical step in the CET processing pipeline. Classical template-based methods have high false-positive rates due to the very low signal-to-noise ratios (SNR) typical of CET volumes, while more recent neural-network based detection algorithms require extensive labeling, are very slow to train and can take days to run. To address these issues, we propose a novel particle detection framework that uses positive-unlabeled learning and exploits the unique properties of 3D tomograms to improve detection performance. Our end-to-end framework is able to identify particles within minutes when trained using a single partially labeled tomogram. We conducted extensive validation experiments on two challenging CET datasets representing different experimental conditions, and observed more than 10% improvement in mAP and F1 scores compared to existing particle picking methods used in CET. Ultimately, the proposed framework will facilitate the structural analysis of challenging biomedical targets imaged within the native environment of cells.

Cryo-electron microscopy (cryo-EM) has become an indispensable technique for determining three-dimensional structures of biological macromolecules. A critical aspect of achieving high-resolution cryo-EM reconstructions is accurately determining and correcting for the microscopes contrast transfer function (CTF). The CTF introduces defocus-dependent distortions during imaging; if not properly accounted for, the CTF can distort features in and limit the resolution of 3D reconstructions. For tilt-series data used in cryo-electron tomography (cryo-ET), CTF estimation becomes even more challenging due to the tilt of the specimen, which introduces a defocus gradient across the field of view, as well as the low dose and signal in individual tilt images. Here, we describe a simple algorithm to improve the accuracy of CTF estimation of tilted images by leveraging the tilt-series alignment parameters determined for tomographic reconstruction to explicitly account for the tilted specimen geometry. In brief, each tilt image is divided into patches, each of which are then stretched according to their defocus shift. These are then summed to provide a coherent power spectra at the tilt axis, which can then be used in standard CTF estimation algorithms. This uses all the data in each image to enhance the visibility of Thon rings, thereby improving high-resolution CTF estimation and subsequent enhancements in the resolution of subtomogram averages.

The enzyme nitrogenase catalyzes the reduction of dinitrogen to ammonia during biological nitrogen fixation through a mechanism involving the ATP dependent interaction of two component proteins adopting multiple conformational states. To date, high resolution structural information has been provided by X-ray crystallography, which restricts the states that can be accessed to those that can be crystallized. Cryo-electron microscopy (cryoEM) presents a new opportunity for structural characterization of nitrogenase solution structures, and may yield new information on the mechanism of nitrogenase by revealing structures of transient or heterogeneous states. In this study, we present single particle cryoEM structures of the MoFe-nitrogenase endogenously isolated from Azotobacter vinelandii. To maintain the fully reduced cluster states of this oxygen sensitive protein, we prepared samples within an anaerobic chamber and employed specialized conditions to minimize partial disordering of the -subunit at the air-water interface during freezing. Under these conditions, cryoEM structures of the as-isolated MoFe-protein and stabilized MoFe-protein-Fe-protein ADP-AlF4-complex were generally found to closely resemble their corresponding X-ray crystallographic structures. The cryoEM structures did reveal disordering in regions of the MoFe-protein -subunit reminiscent of that observed previously for the {Delta}nifB MoFe-protein lacking the FeMo-cofactor, suggesting that this disorder may reflect functionally relevant dynamics, as well as the possibility of asymmetric binding of the Fe-protein to the MoFe-protein in solution. The methods presented here pave the way toward the capture and interrogation of turnover-relevant nitrogenase states by cryoEM.

Many tasks in single-particle cryo-electron microscopy (cryo-EM), such as 2D/3D classification and homo/heterogeneous reconstruction, require optimizing model parameters to minimize the discrepancy between observed data and a forward model. The standard Mean Squared Error (MSE) loss function is computationally efficient but suffers from a non-convex rugged loss landscape, particularly for high-resolution heterogeneity inference. In this work, we investigate the practical utility of Sliced Wasserstein (SW) distances. We implement exact W2 estimators (inverse-CDF and greedy matching) of projections alongside a computationally efficient proxy based on the L2 norm of CDFs, a formulation akin to the sliced Cramer-von Mises distance. We establish the latter as a robust, fully differentiable workhorse for the cryo-EM forward model. We evaluate its performance against the MSE in joint inference tasks recovering pose, CTF parameters, and conformational heterogeneity. Our results demonstrate that SW significantly broadens the basin of attraction, enabling robust gradient-based optimization from distant initializations where MSE fails. Using a helical spiral toy model, we highlight how SW losses are sensitive to per-particle contrast, where background noise level miscalibration can induce geometric bias in the inferred structure. We show that this bias is manageable through a joint optimization strategy that treats background contrast as a learnable parameter. Finally, we validate the approach on a synthetic dataset using the Zernike3D framework, showing that the SW loss works and yields an accurate landscape representations, comparable with MSE. These findings establish SW as a powerful tool for navigating the rugged landscapes of cryo-EM forward model parameters. SynopsisThe Sliced Wasserstein loss provides a smoother optimization landscapes than mean squared error for single particle cryo-EM joint inference of pose, CTF defocus and conformational heterogeneity. Estimating background contrast is essential to avoid biasing other parameters.

Orientation estimation of macromolecules in cryo-electron tomography (cryo-ET) images is one of the fundamental steps in applying subtomogram averaging. The standard method in particle picking and orientation estimation is template matching (TM), which is computationally very expensive, with its performance depending linearly on the number of template orientations. In addition to conventional image processing methods like TM, the investigation of crowded cell environments using cryo-ET has also been attempted with deep learning (DL) methods. These attempts were restricted to macromolecule localization and identification while orientation estimation was not addressed due to a lack of a large enough dataset of ground truth annotations suitable for DL. To this end, we first generate a large-scale synthetic dataset of 450 tomograms containing almost 200K samples of two macromolecular structures using the PolNet simulator. Utilizing this synthetic dataset, we address the problem of particle orientation estimation as a regression problem by proposing a DL-based model based on multi-layer perceptron networks and a six-degree-of-freedom orientation representation. The iso-surface visualizations of the averaged subtomograms show that the predicted results by the network are overly similar to that of ground truth. Our work shows that orientation estimation of particles using DL methods is in principle possible provided that ground truth data is available. What remains to be solved is the gap between synthetic and experimental data. The source code is available at https://github.com/noushinha/DeepOrientation.

The structural analysis of protein complexes by cryo-electron microscopy (cryo-EM) single particle analysis (SPA) has had great impact as a biophysical method in recent years. Many results of cryo-EM SPA are based on state-of-the-art cryo-electron microscopes customized for SPA. These are currently only available in limited locations around the world, where securing machine time is highly competitive. One potential solution for this time-competitive situation is to reuse existing multi-purpose equipment. Here, we used a multi-purpose TEM with a side entry cryo-holder at our facility to evaluate the potential of high-resolution SPA. We report a 3 [A] resolution map of apoferritin with local resolution extending to 2.6 [A]. The map clearly showed two positions of an aromatic side chain. We also verified the optimal imaging conditions depending on different electron microscope and camera combinations. This study demonstrates the possibilities of more widely available and established electron microscopes, and their applications for cryo-EM SPA.

Recently it has been demonstrated that single-particle cryogenic electron microscopy (cryo-EM) at 200 keV is capable of determining protein structures, including those smaller than 100 kDa, at sub-3.0 [A] resolutions, without using significant defocus or a phase plate. However, the majority of near-atomic resolution cryo-EM structures has been determined using 300 keV. Consequently, many typical parameter settings for the cryo-EM computational image processing steps, especially those associated with the contrast transfer function, are based on the accumulated experience of 300 kV cryo-EM. We have therefore revised these parameters, established theoretical bases for criteria to find an optimal mask diameter and box size for a given dataset irrespective of acceleration voltage or protein size, and proposed a protocol. Considering the defocus distributions of the datasets, merely optimizing the mask diameters and box sizes yielded meaningful resolution improvements for the reconstruction of < 200 kDa proteins using 200 kV cryo-EM.

SummaryCryoEM multi-grid screening is often a tedious process that demands hours of attention. Here, this protocol shows how to set up standard Leginon collection and Smart Leginon Autoscreen to automate this process. This protocol can be applied to the majority of cryoEM holey foil grids. Advancements in cryo-electron microscopy (cryoEM) techniques over the past decade have allowed structural biologists to routinely resolve macromolecular protein complexes to near-atomic resolution. The general workflow of the entire cryoEM pipeline involves iterating between sample preparation, cryoEM grid preparation, and sample/grid screening before moving on to high-resolution data collection. Iterating between sample/grid preparation and screening is typically a major bottleneck for researchers, as every iterative experiment must optimize for sample concentration, buffer conditions, grid material, grid hole size, ice thickness, and protein particle behavior in the ice, amongst other variables. Furthermore, once these variables are satisfactorily determined, grids prepared under identical conditions vary widely in whether they are ready for data collection, so additional screening sessions prior to selecting optimal grids for high-resolution data collection are recommended. This sample/grid preparation and screening process often consumes several dozen grids and days of operator time at the microscope. Furthermore, the screening process is limited to operator/microscope availability and microscope accessibility. Here, we demonstrate how to use Leginon and Smart Leginon Autoscreen to automate the majority of cryoEM grid screening. Autoscreen combines machine learning, computer vision algorithms, and microscope-handling algorithms to remove the need for constant manual operator input. Autoscreen can autonomously load and image grids with multi-scale imaging using an automated specimen-exchange cassette system, resulting in unattended grid screening for an entire cassette. As a result, operator time for screening 12 grids may be reduced to [~]10 minutes with Autoscreen compared to [~]6 hours using previous methods which are hampered by their inability to account for high variability between grids. This protocol and video tutorial first introduces basic Leginon setup and functionality, then demonstrates Autoscreen functionality step-by-step from the creation of a template session to the end of a 12 grid automated screening session.

Cryo-electron tomography (cryoET) has emerged as a powerful structural biology tool for understanding protein complexes in their native cellular environments. Presently, 3D volumes of cellular environments can be acquired in the thousands in a few days where each volume provides a rich and complex cellular landscape. Despite numerous innovations, localizing and identifying the vast majority of protein species in these volumes remains prohibitively difficult. Machine learning based methods provide an opportunity to automate the process of labeling and annotating cryoET volumes. Due to current bottlenecks in the annotation process, and a lack of large standardized datasets, training datasets for machine learning algorithms have been scarce. Here, we present a defined "phantom" sample, along with "ground truth" annotations, that will be the basis of a machine learning challenge to bring cryoET and ML experts together and spur creativity to address this annotation problem. We have also set up a cryoET data portal that provides additional diverse sets of annotated 3D volumes from cryoET experts across the world for the machine learning challenge.

Sample thickness is a known key parameter in cryo-electron microscopy (cryo-EM) and can affect the amount of high-resolution information retained in the image. Yet, common data acquisition approaches in single particle cryo-EM do not take it into account. Here, we demonstrate how the sample thickness can be determined before data acquisition, allowing to identify optimal regions and restrict automated data collection to images with preserved high-resolution details. This quality over quantity approach, almost entirely eliminates the time- and storage-consuming collection of suboptimal images, which are discarded after a recorded session or during early image processing due to lack of high-resolution information. It maximizes data collection efficiency and lowers the electron microscopy time required per dataset. This strategy is especially useful, if the speed of data collection is restricted by the microscope hardware and software, or if microscope access time, data transfer, data storage and computational power are a bottleneck. SynopsisSample thickness is a key parameter in single particle cryo-electron microscopy. Determining sample thickness before data acquisition allows to target optimal areas and maximize data output quality of single particle cryo-electron microscopy sessions. Scripts and optimized workflows for EPU and SerialEM are presented and available as open-source.

Advances in single-particle cryogenic electron microscopy (cryoEM) now allow for routine structure determination of well-behaved biological specimens to high-resolution. Despite advances in the electron microscope, direct electron detectors, and data processing software, the preparation of high-quality grids with thin layers of vitreous ice containing the specimen of interest in random orientations remains a critical bottleneck for many projects. Although numerous efforts have been dedicated to overcoming hurdles frequently encountered during specimen vitrification using traditional blot-and-plunge specimen preparation techniques, the development of blot-free grid preparation devices provide a unique opportunity to carefully tune ice thickness, particle density, and specimen behavior during the vitrification process for improvements in image quality. Here, we describe critical steps of high-quality grid preparation using a SPT Labtech chameleon, evaluation of grid quality/ice thickness using the chameleon software, high-throughput imaging in the electron microscope, and recommend steps for troubleshooting grid preparation when standard parameters fail to yield suitable specimen. Video LinkContents of this manuscript are available as a video tutorial. This video can be found here

Cryogenic electron tomography (CryoET) provides 3D views of vitrified cellular samples, and protein structures can be determined from the tomograms by averaging many copies of the same protein computationally. However, the resolution of these averaged structures, particularly for smaller proteins, is often constrained by the precision of tilt-series alignment. In this study, we introduce a gradient descent-based approach to refine alignment parameters, enhancing the contrast in tomograms of the sample regions. This refinement not only improves contrast but also yields higher-resolution protein structures derived from the same particle populations.

Super resolution detector acquisition for cryo-EM has been used to improve the clarity of cryo-EM reconstructions. Recent reports have demonstrated achieving resolutions beyond the physical Nyquist limit using super resolution acquisition. Here, we demonstrate exceeding the physical Nyquist limitation by pre-processing the raw micrograph movies from "counting mode" data which has already reached physical Nyquist reconstruction resolution. To demonstrate functionality, micrograph movies of five datasets were pre-processed and demonstrate that it is possible to exceed the physical Nyquist limit via pixel doubling before motion correction. We call this "post-acquisition super resolution", or PASR. While this was originally developed for processing of giant virus datasets, where acquiring at high magnification is not always possible or desirable, it is also shown to work for smaller objects such as adeno-associated virus (AAV) and apoferritin, both of which are still high symmetry, and jack bean urease, with lower symmetry. PASR could reduce the magnification required to achieve desired resolutions, which may increase collection efficiency. PASR can also be of use for in vivo tomography and facilities with high storage demands. However, this method should only be used for data which is able to achieve the Nyquist limit without PASR pre-processing. It will not improve attained resolutions of data which does not already reach the Nyquist limit.

Single particle cryogenic electron microscopy (cryo-EM) as a structural biology methodology has become increasingly attractive and accessible to investigators in both academia and industry as this ever-advancing technology enables successful structural determination of a wide range of protein and nucleic acid targets. Although data for many high resolution cryo-EM structures are still obtained using a 300 kV cryogenic transmission electron microscope (cryo-TEM), a modern 200 kV cryo-TEM equipped with an advanced direct electron detector and energy filter is a cost-effective choice for most single particle applications, routinely achieving sub 3 angstrom ([A]) resolution. Here, we systematically evaluate performance of one such high-end configuration - a 200 kV Glacios microscope coupled with a Falcon 4 direct electron detector and Selectris energy filter (Glacios-F4-S). First, we evaluated data quality on the standard benchmarking sample, rabbit muscle aldolase, using three of the most frequently used cryo-EM data collection software: SerialEM, Leginon and EPU, and found that - despite sample heterogeneity - all final reconstructions yield same overall resolutions of 2.6 [A] and map quality when using either of the three software. Furthermore, comparison between Glacios-F4-S and a 300 kV cryo-TEM (Titan Krios with Falcon 4) revealed nominal resolution differences in overall reconstructions of a reconstituted human nucleosome core particle, achieving 2.8 and 2.5 [A], respectively. Finally, we performed comparative data analysis on the human RAD51 paralog complex, BCDX2, a four-protein complex of approximately 150 kilodaltons, and found that a small dataset ([&le;]1,000 micrographs) was sufficient to generate a 3.3 [A] reconstruction, with sufficient detail to resolve co-bound ligands, AMP-PNP and Mg+2. In summary, this study provides evidence that the Glacios-F4-S operates equally well with all standard data collection software, and is sufficient to obtain high resolution structural information of novel macromolecular complexes, readily acquiring single particle data rivaling that of 300 kV cryo-TEMs.

Rapid structural analysis of purified proteins and their complexes has become increasingly common thanks to key methodological advances in cryo-electron microscopy (cryo-EM) and associated data processing software packages. In contrast, analogous structural analysis in cells via cryo-electron tomography (cryo-ET) remains challenging due to critical technical bottlenecks, including low-throughput sample preparation and imaging, and laborious data processing methods. Here, we describe the development of a rapid in situ cryo-ET sample preparation and data analysis workflow that results in the routine determination of sub-nm resolution ribosomal structures. We apply this workflow to E. coli, producing a 5.8 [A] structure of the 70S ribosome from cells in less than 10 days, and we expect this workflow will be widely applicable to related bacterial samples.

Resolution estimation by Fourier shell correlation (FSC) using half data sets is the standard method for map quality assessment in cryo-EM. Currently, the FSC method is largely used for refined cryo-EM maps in the context of single particle cryo-EM or subtomogram averaging. Here, we extended resolution estimation to assess the quality of electron micrographs, tilt-series and tomograms. We developed a robust statistics-based framework, capable of determining local quality estimates in the above cryo-EM data types. We show that the determined quality values on a micrograph and tomogram level can be used as a particle quality criterion to improve averaged 3D reconstructions. Using local quality assessments of tomograms, we were able to characterize tomogram quality dependence on distance inferred by radiation damage of FIB-milled lamella. This robust resolution-based quality assessment approach suitable for multiple cryo-EM data types opens new possibilities for automated quality control and method development in cryo-EM maps as well as tomograms and micrographs.

Continued exponential growth in the number of structures resolved by single particle cryoEM, as seen in the last decade, requires ever more effective data analysis workflows. Datasets are rarely homogeneous, demanding a multistep procedure for discarding outliers. Since individual particles are very noisy, either 2D or 3D averages are normally used for discrimination. This becomes challenging when the 2D classes themselves are heterogeneous, leading to selection of contaminants or discarding useful rare views/poses. The 3D model-based discrimination requires trustworthy 3D maps and a correct assignment of Euler angles, which in turn depends on the quality of the initial data and might not be available at the very early stages of the analysis. We propose a novel deep-learning approach for improving quality of single particle datasets. The two-stage procedure consists of denoising single particle images using Variational AutoEncoder framework followed by particle quality filtering based on the score inferred for every particle by Domain Adaptation Neural Network trained on a large data set of categorised 2D averages. This approach allows an automated scoring of noisy raw images using data patterns learned from the high signal-to-noise ratio, externally derived 2D classes. Consequently, a higher quality data set enters computationally expensive steps of the data analysis, reducing the need for protracted and expensive calculations. Importantly, our method does not require any prior knowledge about the data or existence of a 3D model, making it universally applicable. Tests on publicly available datasets demonstrated that our approach largely outperformed 2D class-based particle discrimination. Smaller subsets of the top-scoring particles selected with our method were required to obtain the author-reported 3D model resolution. When applied to the user data in the automated on-the-fly data processing pipeline, the method rescued 30% of cases, which otherwise would not reach confidence threshold required for making decision to proceed to the 3D model refinement. It also led to general improvements in the quality of the 3D models for many datasets which were selected for the high-resolution processing.

Cryo-electron tomography (cryo-ET) enables high-resolution, three-dimensional imaging of cellular structures in their native, frozen state. However, image quality is limited by a trade-off between angular sampling and radiation damage, making the choice of angular increment during data collection a critical parameter, affecting tomogram quality and downstream analyses. Optimising this increment is challenging due to the high demands on microscope time, storage, and computation. In this study, we systematically evaluated tilt increments of 1{degrees}, 2{degrees}, 3{degrees}, 5{degrees}, and 10{degrees} using lamellae from Dictyostelium discoideum cells. Keeping total electron dose constant, we found that finer tilt increments (1-3{degrees}) produced better-aligned tomograms with higher signal-to-noise ratios and improved outcomes in template matching and subtomogram averaging. A 3{degrees} increment emerged as the optimal balance between data quality, alignment accuracy, dose per image, and processing efficiency. This practical recommendation supports both high-throughput and high-resolution structural studies and can guide future cryo-ET data acquisition strategies.

Cryo-EM is a powerful tool in structural biology, providing insights through techniques like single-particle analysis (SPA) and cryogenic electron tomography (cryo-ET). In thick specimens, challenges arise as an exponentially larger fraction of the transmitted electrons lose energy from inelastic scattering and can no longer be properly focused as a result of chromatic aberrations in the post-specimen optics. Rather than filtering out the inelastic scattering at the price of reducing potential signal, as is done in energy-filtered transmission electron microscopy (EFTEM), we show how a dose-efficient and unfiltered image can be rapidly obtained using tilt-corrected bright-field scanning-TEM (tcBF-STEM) data collected on a pixelated detector. Enhanced contrast and a 3-5x improvement in collection efficiency are observed for 2D images of intact bacterial cells and large organelles using tcBF-STEM compared to EFTEM for thicknesses beyond 500 nm. As a proof of concept for the techniques performance in structural determination, we present an SPA map at subnanometer resolution for a highly symmetric virus-like particle (VLP) with 789 particles. These findings suggest applications for tcBF-STEM in cryo-EM of thicker cellular volumes where current approaches struggle.