Research

     Over the past decade, cryo-electron microscopy (cryo-EM) has emerged as one of the most important technologies for high-resolution structural studies of biomacromolecules, finding its widespread applications in diverse research fields ranging from structural biology to drug discovery. To further expand the capabilities of cryo-EM, a multitude of advancements are required in various areas, including specimen preparation, image acquisition, data processing, and structural modeling. These developments are necessary to improve the throughput of this technique, attain better resolution and quality of the reconstructed maps, and obtain greater structural information across higher dimensions and functional states. Furthermore, the advent of artificial intelligence (AI) technology has opened up new possibilities for the development of cryo-EM techniques. In this special issue of the Journal of Molecular Biology, we have collected four research articles and five review papers that reflect the cutting-edge frontier of this field.

     The basic workflow of cryo-EM single particle analysis includes specimen vitrification, cryo-EM data collection, image processing and 3D reconstruction, and interpretation and modeling of the final map. In recent years, the presence of the air-water interface, which arises during the process of specimen vitrification, has been identified as a significant bottleneck in cryo-EM. This interface is known to induce preferential orientation or degradation/disassociation of biomacromolecular particles, which severely hinders subsequent cryo-EM image processing and structural analysis. Liu and Wang have reviewed recent efforts to comprehend the nature and impact of air-water interface and have presented several potential solutions to mitigate its effects.1 They specifically focused on the use of supporting film solutions, including the graphene grids.

     A fundamental assumption of single-particle analysis for ideal 3D reconstruction is that all particle images are projections of structurally identical biomacromolecules. However, this assumption is not rigorously accurate due to the flexibility of biomacromolecules, resulting in reduced overall resolution in the final 3D reconstruction and non-uniform resolution in different regions. To address this issue, Herreros et al. developed a new reconstruction algorithm called ZART (Zernike3D Algebra Reconstruction Technique).2 The ZART method uses Zernike3D expansion to represent a deformable cryo-EM map, which can describe the deformation of biomacromolecules and perform non-rigid alignments to correct motion of different regions during 3D reconstruction in real space. In their study, the application of the ZART algorithm resulted in improved clarity of the cryo-EM map at the flexible region, which was previously blurry using other software.

     The dynamics and flexibility of biomacromolecule lead to non-uniform resolution of cryo-EM maps. Therefore, in addition to the standard assessment of global resolution by the gold standard Fourier Shell Correlation (FSC) criteria, tools like ResMap, MonoRes and DeepRes have been developed to quantify the local resolution of a cryo-EM map. Dai et al. introduced a new AI-based algorithm, called CryoRes, to address the same issue without the need of half maps or accurate masks.3 Unlike the previous AI-based method DeepRes, which was trained on synthetic datasets, CryoRes was developed by training experimental datasets.

     Interpreting a cryo-EM map and building the atomic model is the final step of the cryo-EM workflow. The accuracy of model building is essential to all subsequent structure-based functional studies and applications in pharmaceutics. However, this task becomes highly challenging and arduous when the complexity of biomacromolecular machines increases, or the resolution of cryo-EM map is lower than 4.0 angstroms, or there are additional densities in cryo-EM map representing previously unidentified components. The emergence of AI-based modeling approach has enabled the development of efficient tools to interpret cryo-EM map into structural models accurately and quickly, making previously impossible modeling tasks possible. Si et al. conducted a systematic review of this topic and especially summarized their recent efforts on the development of DeepTracer and DeepTracer-ID for automatic modeling and recognition of new subunits of biomacromolecular complexes based on cryo-EM maps.4

     The structural dynamics of biomacromolecules can be analyzed from experimental cryo-EM data, which is another important direction and frontier of the field, aside from improving resolution. While the currently widely used 3D classification algorithms implemented in RELION and cryoSPARC have proven to be efficient in separating discrete conformations of biomacromolecules from a large number of cryo-EM particle images, they are not capable of detecting continuous conformational changes of biomacromolecules. In this special issue, Toader et al. discussed a general mathematical framework and systematically surveyed recent developments to extract continuous conformations from cryo-EM datasets, including manifold learning, normal mode method, linear and nonlinear models.5 Meanwhile, Chen et al. reported a new nonlinear model algorithm, Deep Gaussian Mixture Model (Deep GMM), to study the continuous conformation of biomacromolecules with a scalable ability.6 In contrast to the previously reported nonlinear model CryoDRGN that is based on voxelized representation, their GMM model utilizes pseudo-atomic representation, which greatly reduces the need for computational resources. Besides, Vuillemot et al. developed another new algorithm MDSpace to utilize molecular dynamics (MD) simulation to perform flexible fitting from 3D structural models into 2D cryo-EM particle images and derive the continuous conformational landscape,7 providing a new way to study the structural dynamics of biomacromolecules.

     The future of structural biology research aims to perform in situ structural study in the cellular context. Recent advances in cryo-electron tomography have transformed this goal into a reality, shaping a new era of structural biology. However, one of the major challenges in cryo-electron tomography data processing is to identify target biomacromolecules accurately and efficiently from the noisy and distorted tomograms within the crowded cellular context. Kim et al. reviewed the recent computational methods to recognize and analyze various biomacromolecular particles from tomograms and highlighted the use of AI methods in this area.8

     The continuous development of cryo-EM technology has enabled researcher to study the structures and functions of more complex and challenging super biomacromolecular machines, such as the nuclear pore complex (NPC), the centrosome, the centromere-kinetochore complex, and more. In this special collection, Tai et al. reviewed the history of cryo-EM studies on NPC structures, from in vitro to in vivo, and particularly highlighted the latest breakthroughs in achieving sub-nanometer resolution by taking the advantages of latest development on cryo-EM technology and successful applications of AI modeling techniques,9 providing insight into the future direction of structural biology.

     By compiling a selection of reviews and research papers from several esteemed experts in the field, we hope to offer an insightful picture of the exciting frontiers and promising directions of modern cryo-EM and its potential applications in various areas. We express our sincerest appreciations to all those who have contributed to this special issue.

Article Link: https://www.sciencedirect.com/science/article/pii/S0022283623001602?via%3Dihub