Optimod--an automated approach for constructing and optimizing initial models for single-particle electron microscopy

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy

Single particle cryo-electron microscopy (cryo-EM) has emerged as the prevailing method for near-atomic structure determination, shedding light on the important molecular mechanisms of biological macromolecules. However, the inherent dynamics and structural variability of biological complexes coupled with the large number of experimental images generated by a cryo-EM experiment make data processing nontrivial. In particular, ab initio reconstruction and atomic model building remain major bottlenecks that demand substantial computational resources and manual intervention. Approaches utilizing recent innovations in artificial intelligence (AI) technology, particularly deep learning, have the potential to overcome the limitations that cannot be adequately addressed by traditional image processing approaches. Here, we review newly proposed AI-based methods for ab initio volume generation, heterogeneous 3D reconstruction, and atomic model building. We highlight the advancements made by the implementation of AI methods, as well as discuss remaining limitations and areas for future development.

Emerging Themes in CryoEM─Single Particle Analysis Image Processing

Cryo-electron microscopy (CryoEM) has become a vital technique in structural biology. It is an interdisciplinary field that takes advantage of advances in biochemistry, physics, and image processing, among other disciplines. Innovations in these three basic pillars have contributed to the boosting of CryoEM in the past decade. This work reviews the main contributions in image processing to the current reconstruction workflow of single particle analysis (SPA) by CryoEM. Our review emphasizes the time evolution of the algorithms across the different steps of the workflow differentiating between two groups of approaches: analytical methods and deep learning algorithms. We present an analysis of the current state of the art. Finally, we discuss the emerging problems and challenges still to be addressed in the evolution of CryoEM image processing methods in SPA.

Bayesian Random Tomography of Particle Systems

Random tomography is a common problem in imaging science and refers to the task of reconstructing a three-dimensional volume from two-dimensional projection images acquired in unknown random directions. We present a Bayesian approach to random tomography. At the center of our approach is a meshless representation of the unknown volume as a mixture of spherical Gaussians. Each Gaussian can be interpreted as a particle such that the unknown volume is represented by a particle cloud. The particle representation allows us to speed up the computation of projection images and to represent a large variety of structures accurately and efficiently. We develop Markov chain Monte Carlo algorithms to infer the particle positions as well as the unknown orientations. Posterior sampling is challenging due to the high dimensionality and multimodality of the posterior distribution. We tackle these challenges by using Hamiltonian Monte Carlo and a global rotational sampling strategy. We test the approach on various simulated and real datasets.

State-dependent sequential allostery exhibited by chaperonin TRiC/CCT revealed by network analysis of Cryo-EM maps

The eukaryotic chaperonin TRiC/CCT plays a major role in assisting the folding of many proteins through an ATP-driven allosteric cycle. Recent structures elucidated by cryo-electron microscopy provide a broad view of the conformations visited at various stages of the chaperonin cycle, including a sequential activation of its subunits in response to nucleotide binding. But we lack a thorough mechanistic understanding of the structure-based dynamics and communication properties that underlie the TRiC/CCT machinery. In this study, we present a computational methodology based on elastic network models adapted to cryo-EM density maps to gain a deeper understanding of the structure-encoded allosteric dynamics of this hexadecameric machine. We have analysed several structures of the chaperonin resolved in different states toward mapping its conformational landscape. Our study indicates that the overall architecture intrinsically favours cooperative movements that comply with the structural variabilities observed in experiments. Furthermore, the individual subunits CCT1-CCT8 exhibit state-dependent sequential events at different states of the allosteric cycle. For example, in the ATP-bound state, subunits CCT5 and CCT4 selectively initiate the lid closure motions favoured by the overall architecture; whereas in the apo form of the heteromer, the subunit CCT7 exhibits the highest predisposition to structural change. The changes then propagate through parallel fluxes of allosteric signals to neighbours on both rings. The predicted state-dependent mechanisms of sequential activation provide new insights into TRiC/CCT intra- and inter-ring signal transduction events.

A robust approach to ab initio cryo-electron microscopy initial volume determination

Structural information from macromolecules provides key insights into the way complexes perform their biological functions. The reconstruction process leading to the final three-dimensional (3D) map is iterative and requires an initial volume to prime the refinement procedure. Particle images are aligned to this first reference and subsequently a new map is calculated from these particles. The accurate determination of an ab initio initial volume is still a challenging and open problem in cryo-electron microscopy (cryo-EM). Different algorithms are available to estimate an initial volume from the dataset. Some of these methods provide multiple candidate initial maps and users looking for robustness typically run different approaches. In this case, users arbitrarily evaluate the different obtained candidate maps, as we lack robust methods to objectively assess the accuracy of initial references. This workflow is subjective and error-prone preventing implementation of high-throughput data processing procedures. In this work, we present a robust method to determine the best initial map or maps from a set of ab initio initial volumes obtained from one or multiple different approaches. The method is based on evaluating multiple small subsets of candidate initial volumes and particle images through reference-based 3D classifications. Obtained 3D classes of accurate initial maps will result majoritarian and the respective attracted particles will be aligned with high angular accuracies. We have tested the proposed approach with structurally homogeneous and heterogeneous datasets providing satisfactory results with both type of data.

Common lines modeling for reference free Ab-initio reconstruction in cryo-EM

We consider the problem of estimating an unbiased and reference-free ab initio model for non-symmetric molecules from images generated by single-particle cryo-electron microscopy. The proposed algorithm finds the globally optimal assignment of orientations that simultaneously respects all common lines between all images. The contribution of each common line to the estimated orientations is weighted according to a statistical model for common lines' detection errors. The key property of the proposed algorithm is that it finds the global optimum for the orientations given the common lines. In particular, any local optima in the common lines energy landscape do not affect the proposed algorithm. As a result, it is applicable to thousands of images at once, very robust to noise, completely reference free, and not biased towards any initial model. A byproduct of the algorithm is a set of measures that allow to asses the reliability of the obtained ab initio model. We demonstrate the algorithm using class averages from two experimental data sets, resulting in ab initio models with resolutions of 20Å or better, even from class averages consisting of as few as three raw images per class.

Cryo-EM structures and atomic model of the HIV-1 strand transfer complex intasome

Like all retroviruses, HIV-1 irreversibly inserts a viral DNA (vDNA) copy of its RNA genome into host target DNA (tDNA). The intasome, a higher-order nucleoprotein complex composed of viral integrase (IN) and the ends of linear vDNA, mediates integration. Productive integration into host chromatin results in the formation of the strand transfer complex (STC) containing catalytically joined vDNA and tDNA. HIV-1 intasomes have been refractory to high-resolution structural studies. We used a soluble IN fusion protein to facilitate structural studies, through which we present a high-resolution cryo-electron microscopy (cryo-EM) structure of the core tetrameric HIV-1 STC and a higher-order form that adopts carboxyl-terminal domain rearrangements. The distinct STC structures highlight how HIV-1 can use the common retroviral intasome core architecture to accommodate different IN domain modules for assembly.

Addressing preferred specimen orientation in single-particle cryo-EM through tilting

We present a strategy for tackling preferred specimen orientation in single-particle cryogenic electron microscopy by employing tilts during data collection. We also describe a tool to quantify the resulting directional resolution using 3D Fourier shell correlation volumes. We applied these methods to determine the structures at near-atomic resolution of the influenza hemagglutinin trimer, which adopts a highly preferred specimen orientation, and of ribosomal biogenesis intermediates, which adopt moderately preferred orientations.

Modular Assembly of the Bacterial Large Ribosomal Subunit

The ribosome is a complex macromolecular machine and serves as an ideal system for understanding biological macromolecular assembly. Direct observation of ribosome assembly in vivo is difficult, as few intermediates have been isolated and thoroughly characterized. Herein, we deploy a genetic system to starve cells of an essential ribosomal protein, which results in the accumulation of assembly intermediates that are competent for maturation. Quantitative mass spectrometry and single-particle cryo-electron microscopy reveal 13 distinct intermediates, which were each resolved to ∼4-5 Å resolution and could be placed in an assembly pathway. We find that ribosome biogenesis is a parallel process, that blocks of structured rRNA and proteins assemble cooperatively, and that the entire process is dynamic and can be "re-routed" through different pathways as needed. This work reveals the complex landscape of ribosome assembly in vivo and provides the requisite tools to characterize additional assembly pathways for ribosomes and other macromolecular machines.

Glycan shield and epitope masking of a coronavirus spike protein observed by cryo-electron microscopy

Cryo-EM and mass spectrometry analyses of the spike glycoprotein trimer from coronavirus HcoV-NL63 reveal an extensive glycan shield that covers the protein surface, including an epitope targeted by neutralizing antibodies against several coronaviruses.

The threat of a major coronavirus pandemic urges the development of strategies to combat these pathogens. Human coronavirus NL63 (HCoV-NL63) is an α-coronavirus that can cause severe lower-respiratory-tract infections requiring hospitalization. We report here the 3.4-Å-resolution cryo-EM reconstruction of the HCoV-NL63 coronavirus spike glycoprotein trimer, which mediates entry into host cells and is the main target of neutralizing antibodies during infection. The map resolves the extensive glycan shield obstructing the protein surface and, in combination with mass spectrometry, provides a structural framework to understand the accessibility to antibodies. The structure reveals the complete architecture of the fusion machinery including the triggering loop and the C-terminal domains, which contribute to anchoring the trimer to the viral membrane. Our data further suggest that HCoV-NL63 and other coronaviruses use molecular trickery, based on epitope masking with glycans and activating conformational changes, to evade the immune system of infected hosts.

Cryo-EM reveals a novel octameric integrase structure for betaretroviral intasome function

Retroviral integrase catalyses the integration of viral DNA into host target DNA, which is an essential step in the life cycle of all retroviruses. Previous structural characterization of integrase-viral DNA complexes, or intasomes, from the spumavirus prototype foamy virus revealed a functional integrase tetramer, and it is generally believed that intasomes derived from other retroviral genera use tetrameric integrase. However, the intasomes of orthoretroviruses, which include all known pathogenic species, have not been characterized structurally. Here, using single-particle cryo-electron microscopy and X-ray crystallography, we determine an unexpected octameric integrase architecture for the intasome of the betaretrovirus mouse mammary tumour virus. The structure is composed of two core integrase dimers, which interact with the viral DNA ends and structurally mimic the integrase tetramer of prototype foamy virus, and two flanking integrase dimers that engage the core structure via their integrase carboxy-terminal domains. Contrary to the belief that tetrameric integrase components are sufficient to catalyse integration, the flanking integrase dimers were necessary for mouse mammary tumour virus integrase activity. The integrase octamer solves a conundrum for betaretroviruses as well as alpharetroviruses by providing critical carboxy-terminal domains to the intasome core that cannot be provided in cis because of evolutionarily restrictive catalytic core domain-carboxy-terminal domain linker regions. The octameric architecture of the intasome of mouse mammary tumour virus provides new insight into the structural basis of retroviral DNA integration.

Cryo-electron microscopy structure of a coronavirus spike glycoprotein trimer

The tremendous pandemic potential of coronaviruses was demonstrated twice in the past few decades by two global outbreaks of deadly pneumonia. Entry of coronaviruses into cells is mediated by the transmembrane spike glycoprotein S, which forms a trimer carrying receptor-binding and membrane fusion functions. S also contains the principal antigenic determinants and is the target of neutralizing antibodies. Here we present the structure of a mouse coronavirus S trimer ectodomain determined at 4.0 Å resolution by single particle cryo-electron microscopy. It reveals the metastable pre-fusion architecture of S and highlights key interactions stabilizing it. The structure shares a common core with paramyxovirus F proteins, implicating mechanistic similarities and an evolutionary connection between these viral fusion proteins. The accessibility of the highly conserved fusion peptide at the periphery of the trimer indicates potential vaccinology strategies to elicit broadly neutralizing antibodies against coronaviruses. Finally, comparison with crystal structures of human coronavirus S domains allows rationalization of the molecular basis for species specificity based on the use of spatially contiguous but distinct domains.

Bayesian inference of initial models in cryo-electron microscopy using pseudo-atoms

Single-particle cryo-electron microscopy is widely used to study the structure of macromolecular assemblies. Tens of thousands of noisy two-dimensional images of the macromolecular assembly viewed from different directions are used to infer its three-dimensional structure. The first step is to estimate a low-resolution initial model and initial image orientations. This is a challenging global optimization problem with many unknowns, including an unknown orientation for each two-dimensional image. Obtaining a good initial model is crucial for the success of the subsequent refinement step. We introduce a probabilistic algorithm for estimating an initial model. The algorithm is fast, has very few algorithmic parameters, and yields information about the precision of estimated model parameters in addition to the parameters themselves. Our algorithm uses a pseudo-atomic model to represent the low-resolution three-dimensional structure, with isotropic Gaussian components as moveable pseudo-atoms. This leads to a significant reduction in the number of parameters needed to represent the three-dimensional structure, and a simplified way of computing two-dimensional projections. It also contributes to the speed of the algorithm. We combine the estimation of the unknown three-dimensional structure and image orientations in a Bayesian framework. This ensures that there are very few parameters to set, and specifies how to combine different types of prior information about the structure with the given data in a systematic way. To estimate the model parameters we use Markov chain Monte Carlo sampling. The advantage is that instead of just obtaining point estimates of model parameters, we obtain an ensemble of models revealing the precision of the estimated parameters. We demonstrate the algorithm on both simulated and real data.

Single-particle cryoEM analysis at near-atomic resolution from several thousand asymmetric subunits

A single-particle cryoEM reconstruction of the large ribosomal subunit from Saccharomyces cerevisiae was obtained from a dataset of ∼75,000 particles. The gold-standard and frequency-limited approaches to single-particle refinement were each independently used to determine orientation parameters for the final reconstruction. Both approaches showed similar resolution curves and nominal resolution values for the 60S dataset, estimated at 2.9 Å. The amount of over-fitting present during frequency-limited refinement was quantitatively analyzed using the high-resolution phase-randomization test, and the results showed no apparent over-fitting. The number of asymmetric subunits required to reach specific resolutions was subsequently analyzed by refining subsets of the data in an ab initio manner. With our data collection and processing strategies, sub-nanometer resolution was obtained with ∼200 asymmetric subunits (or, equivalently for the ribosomal subunit, particles). Resolutions of 5.6 Å, 4.5 Å, and 3.8 Å were reached with ∼1000, ∼1600, and ∼5000 asymmetric subunits, respectively. At these resolutions, one would expect to detect alpha-helical pitch, separation of beta-strands, and separation of Cα atoms, respectively. Using this map, together with strategies for ab initio model building and model refinement, we built a region of the ribosomal protein eL6, which was missing in previous models of the yeast ribosome. The relevance for more routine high-resolution structure determination is discussed.

Cryogenic electron microscopy and single-particle analysis

About 20 years ago, the first three-dimensional (3D) reconstructions at subnanometer (<10-Å) resolution of an icosahedral virus assembly were obtained by cryogenic electron microscopy (cryo-EM) and single-particle analysis. Since then, thousands of structures have been determined to resolutions ranging from 30 Å to near atomic (<4 Å). Almost overnight, the recent development of direct electron detectors and the attendant improvement in analysis software have advanced the technology considerably. Near-atomic-resolution reconstructions can now be obtained, not only for megadalton macromolecular complexes or highly symmetrical assemblies but also for proteins of only a few hundred kilodaltons. We discuss the developments that led to this breakthrough in high-resolution structure determination by cryo-EM and point to challenges that lie ahead.

Structural basis for translational surveillance by the large ribosomal subunit-associated protein quality control complex

All organisms have evolved mechanisms to manage the stalling of ribosomes upon translation of aberrant mRNA. In eukaryotes, the large ribosomal subunit-associated quality control complex (RQC), composed of the listerin/Ltn1 E3 ubiquitin ligase and cofactors, mediates the ubiquitylation and extraction of ribosome-stalled nascent polypeptide chains for proteasomal degradation. How RQC recognizes stalled ribosomes and performs its functions has not been understood. Using single-particle cryoelectron microscopy, we have determined the structure of the RQC complex bound to stalled 60S ribosomal subunits. The structure establishes how Ltn1 associates with the large ribosomal subunit and properly positions its E3-catalytic RING domain to mediate nascent chain ubiquitylation. The structure also reveals that a distinguishing feature of stalled 60S particles is an exposed, nascent chain-conjugated tRNA, and that the Tae2 subunit of RQC, which facilitates Ltn1 binding, is responsible for selective recognition of stalled 60S subunits. RQC components are engaged in interactions across a large span of the 60S subunit surface, connecting the tRNA in the peptidyl transferase center to the distally located nascent chain tunnel exit. This work provides insights into a mechanism linking translation and protein degradation that targets defective proteins immediately after synthesis, while ignoring nascent chains in normally translating ribosomes.

A combined quantitative mass spectrometry and electron microscopy analysis of ribosomal 30S subunit assembly in E. coli

Ribosome assembly is a complex process involving the folding and processing of ribosomal RNAs (rRNAs), concomitant binding of ribosomal proteins (r-proteins), and participation of numerous accessory cofactors. Here, we use a quantitative mass spectrometry/electron microscopy hybrid approach to determine the r-protein composition and conformation of 30S ribosome assembly intermediates in Escherichia coli. The relative timing of assembly of the 3' domain and the formation of the central pseudoknot (PK) structure depends on the presence of the assembly factor RimP. The central PK is unstable in the absence of RimP, resulting in the accumulation of intermediates in which the 3'-domain is unanchored and the 5'-domain is depleted for r-proteins S5 and S12 that contact the central PK. Our results reveal the importance of the cofactor RimP in central PK formation, and introduce a broadly applicable method for characterizing macromolecular assembly in cells.

Efficient initial volume determination from electron microscopy images of single particles

MOTIVATION

Structural information of macromolecular complexes provides key insights into the way they carry out their biological functions. The reconstruction process leading to the final 3D map requires an approximate initial model. Generation of an initial model is still an open and challenging problem in single-particle analysis.

RESULTS

We present a fast and efficient approach to obtain a reliable, low-resolution estimation of the 3D structure of a macromolecule, without any a priori knowledge, addressing the well-known issue of initial volume estimation in the field of single-particle analysis. The input of the algorithm is a set of class average images obtained from individual projections of a biological object at random and unknown orientations by transmission electron microscopy micrographs. The proposed method is based on an initial non-lineal dimensionality reduction approach, which allows to automatically selecting representative small sets of class average images capturing the most of the structural information of the particle under study. These reduced sets are then used to generate volumes from random orientation assignments. The best volume is determined from these guesses using a random sample consensus (RANSAC) approach. We have tested our proposed algorithm, which we will term 3D-RANSAC, with simulated and experimental data, obtaining satisfactory results under the low signal-to-noise conditions typical of cryo-electron microscopy.

AVAILABILITY

The algorithm is freely available as part of the Xmipp 3.1 package [http://xmipp.cnb.csic.es].

CONTACT

jvargas@cnb.csic.es

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.