Frontiers in CryoEM Modeling

In recent years, remarkable advances in single-particle cryoelectron microscopy (cryoEM) have enabled the determination of an increasing number of biomolecular assemblies with unprecedented detail and resolution.¹ The emergence of direct detection device (DDD) detectors has driven a revolution in the field, performing digital image alignment to eliminate specimen drift in cryoEM images. This has led to a remarkable increase in resolution that enables the ability to solve increasingly realistic structures with near-Ångström resolution. Thanks to this technical revolution, it is now possible to observe large macromolecular complexes in their “native” aqueous solution. For example, the structures of protein/nucleic acid assemblies, viruses, and organelles have been obtained that reflect their in vivo situation. In addition, relatively smaller membrane protein structures can be determined using cryoEM experiments. The explosion of deposited cryoEM maps at high resolution is now challenging researchers in computational chemistry and biology, calling for dynamic and mechanistic interpretations of experimental data. Emerging theoretical approaches aim at processing, complementing, and interpreting cryoEM data, overcoming issues in map refinement, data processing, and the atomistic simulation of large biomolecules. It is to this end that JCIM invites authors to contribute their cryoEM work in an upcoming Special Issue, which will be published in Spring 2020.

From a historical perspective, computational chemists and biologists have approached cryoEM by developing and applying methods enabling the refinement of the first low-resolution cryoEM maps. The first fitting methods based on rigid-body docking are currently being replaced by Monte Carlo,² Normal Mode Analysis,³ and molecular dynamics (MD) fitting schemes using the EM map as a constraint. Among them, the pioneering package Situs⁴ uses the minimization of density discrepancy, whereas the MD Flexible Fitting⁵ method uses the gradient of electronic density as a penalty function for potential, thereby allowing flexible fitting. Hybrid approaches have been developed that include a rigid fitting stage followed by MD-driven refinement⁶ or a coarse-grained force field to enable flexibility throughout a docking search.⁷ A major challenge for these methods is to obtain refined atomistic models that reliably represent density maps. Moreover, these methods have been designed to refine structures from medium-resolution maps (∼8 Å), while today’s cryoEM detectors provide higher resolution maps. Concomitant with the benefits of higher resolution maps, the resultant distinct structural features can trap the fitted structure into non-native conformations. This problem particularly arises when dealing with heterogeneous structures in which certain regions reach a near-Ångström resolution, whereas other areas of the map remain above ∼6−8 Å resolution. To deal with the increased resolution of cryoEM maps while refining low-resolution regions, enhanced sampling techniques have been proposed for the refinement process.^8,9 In this respect, accelerated MD and metadynamics show promise during MD fitting, as these methodologies help avoid local minima of the potential energy and efficiently explore the conformational space.

The absence of packing constraints in single-particle cryoEM facilitates the observation of an ensemble of coexisting conformational states. Typically, thousands of 2D images of picked particles are processed through clustering to obtain the most representative 3D structures. Such processing can result in artifacts from under-represented conformations or view directions. Moreover, while the 2D and 3D classifications are generally utilized to remove outliers exhibiting conformational heterogeneity, it is difficult to assess how well atomic models represent the density from which they were generated. Along the same lines, atomic models might lack the representation of the physiological architecture of the sample.¹⁰ To tackle these challenges, theoretical approaches are being considered, including novel clustering schemes¹¹ in parallel with Bayesian analysis, which help deal with large amounts of heterogeneous data.¹² Furthermore, modern image processing algorithms use graphics processing units to address image classification and high-resolution refinement,¹³ while MD is being used to explore the conformational space at the 3D level, defining the most probable states to be extracted from 2D representations. Alternative approaches for cryoEM reconstruction have recently emerged that employ network structural similarity metrics and harness graph theory for map reconstruction.¹⁴

Beyond determining the high-resolution structures of increasingly realistic biological systems, a closely related challenge is the determination of the energetic landscapes that govern how such systems interconvert between their functional states.¹⁵ Emerging methods such as manifold embedding offer the tantalizing possibility of determining energetic landscapes directly from the cryoEM data itself.^16,17 In addition, MD simulations offer the promise of exploring such functional pathways. However, simulating such biophysical events at the molecular level and at longer (biologically relevant) time scales challenges the current capabilities of MD.^18,19 While the use of enhanced sampling^20,21 is a straightforward way to tackle the problem, all-atom simulations, including millions of explicit atoms, will require next-generation algorithms exploiting future exascale computing power, which will likely change the perspective of what can be done with computational biophysics.²² Accordingly, this special issue focuses on the most exciting applications of molecular simulations in the field of cryoEM, as well as emerging theoretical and computational approaches aimed at processing, complementing, and interpreting single-particle cryoEM experiments. This issue will give an overview of the methodological advances in the field and the forthcoming challenges, with particular attention to validation and reproducibility issues.

We focus on three main objectives for the special issue, which concerns current challenges in modeling data from single-particle cryoEM. First, articles will be solicited that describe theoretical approaches enabling the refinement of cryoEM maps. Subsequently, manuscripts will be sought that discuss the current efforts of microscopists and computational biophysicists to process data obtained via cryoEM. Finally, the application of theoretical tools to refine cryoEM maps and all-atom simulations of biological systems obtained via cryoEM will round out the special issue. We look forward to receiving your submissions for this Special Issue by November 1, 2019.