Emerging theoretical approaches to complement single-particle cryo-electron microscopy
The field of structural biology is experiencing a transformative phase thanks to remarkable advances in single-particle cryo-electron microscopy (cryo-EM). In recent years, the emergence of Direct Detection Devices (DDD) has driven a revolution in the field, enabling the determination of an increasing number of biomolecular systems with unprecedented details and resolution (Nogales Nat Methods 2016, 13). The explosion of deposited cryo-EM maps, which range from 5 Å to better than 2 Å resolution, is challenging computational biophysics to rapidly and robustly discern mechanistic implications from the experimental data. Emerging theoretical approaches aim at processing, validating, complementing, and interpreting cryo-EM data, overcoming issues in image analysis, map refinement, and in the simulation of large biomolecules.
For decades, computational biologists have been integral to the cryo-EM community by developing methods that generated interpretable atomic models from low-resolution maps. Initial rigid-body docking algorithms have been replaced by Monte Carlo (Di Maio et al. Nat Methods 2015, 12) and molecular dynamics (MD) fitting schemes. Pioneering packages such as Situs (Wriggers et al. J Struct Biol 1999, 125) and MD Flexible Fitting (Trabuco et al. Structure 2008, 16) use the gradient of the EM density as a penalty function for the potential, while other packages combine rigid-body fitting with a MD-driven refinement (Topf et al. Structure 2008, 16) or employ coarse-grained models to enable flexibility (de Vries & Zacharias Plos One 2012, 7). Today’s cryo-EM maps often contain a range of resolutions, wherein structurally homogeneous regions are resolved to high resolution while conformationally variable regions are resolved to 6 Å or worse. Earlier methods were designed to refine into medium-resolution maps (~8 Å), leading to models of non-native conformations due to limitations of the algorithms. Enhanced sampling techniques during MD fitting reduce the likelihood of becoming trapped in local minima and more efficiently explore the conformational space presented by today’s cryo-EM maps (Goh et al. Annu Rev Biophys 2016, 5).
Cryo-EM has the groundbreaking capacity to produce ensembles of coexisting conformational states enabling to elucidate molecular function. In pursuit of high-resolution structures, 2D and 3D classification is generally utilized to remove particle data exhibiting conformational heterogeneity, which questions how well atomic models represent not only the density from which they were generated, but the physiological architecture of the sample (Herzik et al. Structure 2019, 344). As such, theoretical approaches relying on Bayesian analysis are under development to deal with a large amount of data (Cossio et al. Comput Phys Commun 2017, 210), while image-processing software are being used to assess data classification (Kimanius et al. eLIFE 2016, 5). In parallel, MD is considered to explore the 3D conformational space, defining the most probable states to be interpolated with the 2D representations.
Finally, increasingly realistic biological systems challenge MD by requiring simulations of atomistic level biophysical events at longer time scales (He et al. Nature 2016, 533). In this respect, all-atom simulations including millions of explicit atoms demand next-generation algorithms and future exascale computing power, which will likely change the perspectives of computational biophysics.
Alessandra Magistrato ( CNR-IOM@SISSA ) - Organiser
Joanna Trylska ( University of Warsaw ) - Organiser
Michael Feig ( Michigan State University, East Lansing ) - Organiser
Gabriel Lander ( Scripps Research ) - Organiser
Giulia Palermo ( University of California Riverside ) - Organiser