The notion that the free energy space of the conformations of a functional protein could be simplified as a landscape populated with local minima1–3 has re-focused our thinking about the entire range of biological functions. Catalytic and signaling processes could now be related to the dynamic transitions between those local minima. This special collection includes discussions of technical advances that can furnish structural insights into structures at those local minima and the transition states between them, how the amino acid sequences of proteins determine the free energy differences between them, and the timescales on which they interconvert. These three kinds of information form the time-honored framework for mechanistic inference.
The initial hope of time-resolved crystallography was that “movies” could be made of a reaction in progress.4 It soon became clear that such movies were space and time averages over many unit cells and along a heterogeneous time progression through a transition state. Thus, they would not form time-resolved snapshots along one trajectory. Rather, such snapshots would capture the average electron density of ensembles in variable sets of states. These may or may not represent a series of quasi-stable stages. In serial femtosecond crystallography (SFX), singular value decomposition (SVD) is a mathematical technique used to analyze time-resolved diffraction data by separating the signal containing relevant structural information from noise, allowing researchers to extract meaningful structural changes occurring within a system on very fast timescales, typically on the order of femtoseconds by identifying the most significant components of the dataset.5–7 These “eigen structures” can then be interpreted, albeit somewhat indirectly, as parts of a movie-like trajectory.
An alternative approach is to use manifold-embedding methods to infer a one-dimensional trajectory from heterogeneous states. This has been achieved using the heterogeneity of cryo-electron microscopy images for several systems.8,9 Others use different computational methods to extract conformational variability from native cryo-EM data,10 or employ freeze-quench methods to trap molecules in various states prior to imaging to provide data on conformational ensembles that evolve along a trajectory.11
Despite these recent developments, motion in proteins is still only rarely visualized directly. Rather, it is most often inferred. Several of the papers in this collection address ways to enhance the accuracy of these inferences. Kim et al. describe the application of an enhanced ensemble refinement technique to the crystal structures of three differently liganded forms of adenylate kinase.12 The three crystal structures each have multiple trimeric assemblies, leading to a total of 24 independent crystallographic estimates for the averaged structure. The structural deviations within and between identically liganded subunits provide an elegant demonstration of the improvement afforded by using a tensorial representation of approximately rigid bodies to fit the anisotropic thermal motion detected by x-rays.
Shen and Bax use NMR RDC (residual dipolar coupling) measurements to evaluate quantitative interpretations of molecular motion from ensemble refinements and multiple x-ray structures.13 Although NMR offers a rich variety of potentially useful information on protein motions, the RDC experiment appears to provide the most reliably useful path to exploiting the synergy between diffraction and resonance measurements. In particular, for two quite different examples—the SARS-CoV-2 main protease, Mpro, and ubiquitin, the authors report that time-averaged bond-vector orientations and may offer the opportunity to correctly weight dynamic contributors to ensemble refinement.
These examples of how to document distinct protein conformations pose the following question: how to relate such structural changes to mechanism? Defining structural change is the first requirement for a mechanism. What remains is to show how the equilibria and rates for transitions between them frame a coherent narrative. Both equilibria and rates depend on the underlying thermodynamics of both polymer and solvent. Protein melting profiles can provide some of those data.
Two of the studies in the collection update the data available from unfolding studies. In the first, Weinreb et al. describe the use of Thermofluor for high-throughput melting studies.14, Geobacillus stearothermophilus tryptophanyl-tRNA synthetase may be a unique example. Specific side chains form the barrier separating the three states. They used mutations to these residues designed to reduce stability differences between states.14–18 They measured melting curves for the 16 conformational switch variants. Two aromatic residues assume distinct configurations along the path.18 Aromatic side chain motions may be the slowest of all side chain rearrangements.19,20 The authors used the stabilities of different variants to demonstrate coupling between the one of the two aromatic residues and the relative stabilities of the 16 variants.
The second paper15 describes melting profiles of apo and fully liganded TrpRS. Microcalorimetry confirmed the melting profiles observed in the first paper.14 The authors also describe a full free-energy surface obtained by replica exchange discreet molecular dynamics for fully liganded TrpRS. Remarkably, the calorimetric, fluorometric, and computational melting curves all show a dominant formation of a molten globule state between ground and fully melted states. Calorimetry also provided the enthalpy change required to convert the thermal shifts between the 16 combinatorial variants into true free energy changes due to mutation.
Wand21 addresses conformational entropy, the commonly neglected axis of the protein free energy landscape. He points out that the full extent and variability of conformational entropy—represented by the breadth of the native state ensemble—is largely undocumented and ripe for mining. To emphasize the importance of conformational entropy to protein function he decomposes 28 examples of ligand binding for which the range of binding free energy changes is ∼−10 to −50 kJ/mol whereas the range of contributions to binding from conformational entropy changes is +90 to −45 kJ/mol. Finally, he outlines the potential for using NMR relaxation measurements for fast side chain motion in limited regions of a protein to determine relative entropy changes for the entire protein during functional conformational changes.
A fervent hope in the structural biology community with the advent of new x-ray lasers and cryo-EM methods is that we will be able to routinely visualize transitions between conformations during a catalyzed reaction. The collection has two contributions in this area. Schmidt22 reviews recent progress in the cryo-EM arena and Gao and coworkers23 describe an example of how much closer we have come to that goal using x-ray sources and crystals. Their discussion of how transient, in situ crystallography has provided crucial details of the two-metal mechanism for DNA polymerase that were not evident from comparing static crystal structures. This paper shows how time-resolved crystallography can be used with ordinary synchrotrons if the time of the reactions is seconds or minutes.
Nam and Wolf-Watz24 see the future of protein dynamics studies as very bright, offering a perspective on the past, present, and future of our quest to understand proteins to the point where we can engineer them from scratch to have a designated catalytic or other function. The authors point out that the protein folding problem is sufficiently described by pairwise residue information, but also argue that propagated dynamical modes and function in enzymes are dependent on higher order networks and pose a significantly more complex problem.
It is clear that experiments observing the dynamics of proteins at the atomic level are still woefully scarce. We, collectively, are still quite inept at designing proteins that do anything besides just fold or maybe bind some ligand.25 Progress is slow partly because the dynamic components of structure in function cannot yet be modeled precisely and it is difficult to generalize their roles. In one example case, reducing the magnitude of the dynamic behavior helped make a better catalyst,26 and in another case, dialing up the dynamic components increases functionality.27 Of course, all Nature cares about is the holistic contribution to fitness, so it is not surprising that the role dynamics play in different systems should be idiosyncratic. The authors hope that studies on the dynamics of proteins continue to find support in order that we can move past static snapshots and abstract protein energy landscapes to a solid understanding of their principles of operation and purposeful design.
