Integrative Approaches for Characterizing Protein Dynamics: NMR, CryoEM and Computer Simulations

Abstract

Proteins are inherently dynamic and their internal motions are essential for biological function. Protein motions cover a broad range of timescales: 10⁻¹⁴–10 sec, spanning from sub-picosecond vibrational motions of atoms via microsecond loop conformational rearrangements to millisecond large amplitude domain reorientations. Observing protein dynamics over all timescales and connecting motions and structure to biological mechanisms requires integration of multiple experimental and computational techniques. This review reports on state-of-the-art approaches for assessing dynamics in biological systems using recent examples of virus assemblies, enzymes, and molecular machines. By integrating NMR spectroscopy in solution and the solid state, cryo electron microscopy, and molecular dynamics simulations, atomistic pictures of protein motions are obtained, not accessible from any single method in isolation. This information provides fundamental insights into protein behavior that can guide the development of future therapeutics.

Introduction

Proteins are the major components of the molecular machinery that keep cells and organisms alive, performing their tasks in a spatially and temporally organized fashion. Thus, their structures and dynamics have evolved to ensure optimal function [1,2]. Motions are a pivotal intrinsic property of proteins, exploited by nature to fine-tune their reactivities and interactions, when they perform their activities [3-5]. Alas, protein dynamics are generally ignored in the traditional representations of static average structures, despite their fundamental importance for understanding mechanisms and regulation of biological processes.

Protein motions span over 15 orders of magnitude in timescale, ranging from subpicoseconds to milliseconds or seconds [6]. Subpicosecond and picosecond dynamics are associated with vibrational motions of atoms, pico- to nanosecond motions are connected to side chain reorientations, while slower motions involve the collective movement of different parts of a protein, ranging from covalent bond isomerization (disulfides, cis-trans proline) to local or loop conformational changes to reorientation of domains [7]. Functionally relevant conformational rearrangements in proteins are often associated with changes in dynamics [8], both in frequencies and amplitudes. They can also occur in regions distal to the structural change by allosteric mechanisms [9,10].

Among the multitude of current methods for atomic-level structure determination, nuclear magnetic resonance (NMR), cryogenic electron microscopy (cryo-EM) and X-ray crystallography provide the spatial arrangements of atoms in a protein. The resulting protein structural models can be complemented by computationally assessing dynamics using large-scale molecular dynamics (MD) simulations [11-13]. Dynamics can also be explored directly by NMR: timescales from pico- to milliseconds can be examined through relaxation measurements, lineshape analysis, and exchange experiments [14,15]. By contrast, in cryo-EM and X-ray crystallography, information on timescales of motions is absent, although data on conformational distributions present in the density maps are often interpreted as reflecting possible motions [16-18]. Timescales of motions accessible by MD simulations critically depend on the length of simulations [19] and usually do not exceed milliseconds for all-atom simulations on large systems [20-23]. Furthermore, many large-amplitude conformational changes are not accessible by MD simulations, and time-independent approaches, such as normal mode analysis, principal component analysis or time-structure independent component analysis, are necessary [24-26]. Gaining information about the entire range of biologically relevant timescales of protein motions requires integration of all possible experimental and computational techniques to visualize dynamics and connect motions to function.

In this review, we will highlight recent impactful studies of protein dynamics using an integrative approach encompassing NMR, cryo-EM and MD simulations, reporting on functionally significant motions in a number of complex biological systems, including enzymes and large molecular assemblies. Such integrative studies are still relatively rare, and in the second part of the review investigations integrating NMR and cryo-EM will be highlighted.

Integration of NMR, cryo-EM and MD simulations

Transcription factor II Human (TFIIH) is a multiprotein complex involved in regulation of transcription processes in the cells. Its core complex consists of seven subunits: XPB, XPD, p62, p52, p44, p34 and p8. Greber et al. determined the structure of the TFIIH core complex by cryo-EM [27,28]. However, the functionally important N-terminal pleckstrin homology domain (PH-D) of p62 (residues 1-103), involved in stabilizing the assembly and interacting with crucial components in the transcriptional pathway, was not observed. Therefore, the Nogales group hypothesized that it was disordered. In a recent study, Okuda et al. employed solution NMR to elucidate the structure and dynamics of the PH-D linked to the structured BTF-2-like transcription factor and synapse-associated DOS2-like proteins domain (BSD1) (residues 104-158) which was visible by cryo-EM [29]. The NMR study demonstrated that PH-D is not disordered, exhibiting the canonical fold of seven β-strands and an α-helix. In contrast, the linker was dynamic on the millisecond timescale and therefore mediated transient interactions of the PH-D with other domains as shown in Figure 1. Taking all the experimental results into account made it possible for the Nishimura group to generate a dynamic structural model, by integrating the previous cryo-EM studies by the Nogales group and their own NMR data, in an MD-based refinement highlighting interdomain linker motions as well as the overall dynamic behaviour and transient interactions of the PH domain with other TFIIH core complex subunits.

Figure 1. ∣. Structural model of human TFIIH with a superposition of 275 structural models of human the p62 PH domain.

Reproduced from [29].

Fused in Sarcoma (FUS) is an abundant nuclear protein involved in gene transcription and regulation, RNA metabolism and DNA repair [30]. Its low sequence complexity (LC) N-terminal domain (FUS-LC-N, residues 2-108) folds into a cross-β core and is responsible for FUS intracellular aggregation with amyloid-like fibrils formation [31], while the functionally important C-terminal domain (FUS-LC-C, residues 111-214) [32] appears to be disordered in full-length FUS-LC fibrils. In isolation, FUS-LC-C can form a two-fold symmetric fibril core, different from the one formed by full length FUS-LC, as was established in a recent study by the Tycko group integrating cryo-EM, magic angle spinning (MAS) NMR spectroscopy and MD simulations [33]. Only a fraction of residues of FUS-LC-C, namely residues 112-150 in three β-strands, were visible by cryo-EM. The overall sheet arrangement has lower curvature than that in FUS-LC and FUS-LC-N fibrils, explaining the difference in the core structure. Residues 150-214 are disordered, with several being dynamic on a sub-microsecond timescale, suggesting the presence of multiple subspecies, possibly relevant to full-length FUS-LC fibrils. The authors propose that in the full-length protein the FUS-LC-N cross-β core structure is more compact, energetically favorable, and folds faster.

The transient ribosome-nascent chain complex (RNC) is formed at the ribosome’s exit tunnel to assist in co-translational folding of a protein. A recent study of 70S E.coli ribosomes in complex with nascent chains (NC) of immunoglobulin-like filamin domain (FLN5) by Ahn et al. [34] investigated the role of large ribosomal subunit proteins uL22, uL23 and uL24 that possess loops protruding into the exit tunnel as shown in Figure 2. Using a combination of cryo-EM, NMR and MD simulations showed that truncation of the uL22 protein loops had no significant impact on FLN5 folding, suggesting that NCs are pointing into the vestibule region of the tunnel. NMR data revealed that the uL24 loop is more dynamic than the uL23 loop, with both loops enhancing the stability of the complex, compared to the loop-truncated constructs. Cryo-EM structures of the complex using 70S variants revealed electrostatic loop interactions with neighboring RNA. MD simulations based on the experimental NMR data led the authors to suggest a reduction in affinity of 70S devoid of the loops for the NC, influencing early folding events. The authors propose that the underlying dynamics in the ribosome create an interesting interplay: while the uL23 and uL24 loops only weakly interact with the newly synthesized protein, uL23 influences the motional modes of the NC and interferes with premature folding while uL24 guides its path through the energy landscape and permits productive folding of FLN5.

Figure 2. ∣. Ribosomal proteins at the exit tunnel.

Structure of the 70 S E. coli ribosome with the exit tunnel expansion. Reproduced from [34].

Human Immunodeficiency Virus Type 1 (HIV-1) is the causative agent of acquired immunodeficiency syndrome (AIDS) that has globally claimed at least 40 million lives since 1981. HIV-1 capsid (CA) protein, a promising therapeutic target due to its key functional roles in the viral life cycle, forms a fullerene-like core of variable curvature assembled from ~250 CA hexamers and 12 pentamers [35]. Inherent pleomorphism renders mature capsid a challenging target for atomic-level structural characterization, although in-vitro CA assemblies of various morphologies recapitulate the salient capsid’s structural features and have been studied extensively by multiple groups [36-42].

Early studies of CA assemblies by solution and MAS NMR spectroscopy, X-ray crystallography, cryo-EM and all-atom MD simulations demonstrated that capsids are remarkably dynamic with motions occurring on nano- to millisecond timescales [37,43-46]. Recently a structure of CA tubular assemblies with dynamic details was determined, integrating MAS NMR, low-resolution cryo-EM and MD simulations, as shown in Figure 3 [42]. This study provided atomic-level dynamic and conformational information on the functionally important β-hairpin (residues 1-13), the Cyclophilin A (CypA) binding loop (residues 83-100), as well as the interhexamer dimer and trimer interfaces. Distinct conformational clusters and their relative populations were derived by integrating MAS NMR experiments and data-guided MD simulations with rigorous model-free statistical analysis.

Figure 3. ∣. HIV-1 CA protein assemblies dynamics.

with a) illustrating the molecular surface of a HIV-1 CA tube. The NTD and CTD are colored in purple and gray, respectively, and b) detailing MAS-NMR derived distances for dynamic parts of CA CypA loop (top) and β-hairpin (bottom). In c) representative structural clusters of the flexible linker region (top) and the cyclophilin A (CypA) binding loop (bottom) are provided. Reproduced with permission from [42].

Integration of NMR and cryo-EM

The caseinolytic protease proteolytic subunit (ClpP) is a serine protease necessary for cellular proteostasis. Several bacteria, including Mycobacterium tuberculosis, rely on this protease in their life cycle, rendering ClpP an attractive drug target for combatting drug resistance. M. tuberculosis possesses two enzymes, MtClpP1 and MtClpP2 that form homomultimeric MtClpP1P2 complexes, whose structures are ill understood. A recent study by the Kay group combining solution NMR and cryo-EM explored structure-dynamics-function relationship in apo MtClpP1P2 and its complex with the small molecule activator benzoyl-leucyl-leucine (Bz-LL) that is necessary for catalytic activity, as well as with cyclic acyldepsipeptidase (ADEP), that causes dysregulation of activity [47]. This study showed that the heptamers of MtClpP2 are more dynamic at the N-terminal β-hairpins, compared to MtClpP1, and that heptamers convert slowly into a tetradecameric MtClpP1P2 complex. ADEP binding rigidifies the N-terminal β-hairpins and, by quenching the dynamics, activates MtClpP2, while MtClpP1 remains catalytically inactive (Figure 4). MtClpP1’s activation requires Bz-LL, which rigidifies the heptamer and facilitates β-sheet formation at the inter-heptameric interface, thereby generating the active enzymatic catalytic site. While cryo-EM yielded major structural information and NMR provided both structural and dynamic details, the combination of both methods allowed to establish a comprehensive structure-dynamics-function relationships.

Figure 4. ∣. Dynamics associated with the caseinolytic protease.

The cryo-EM structure of apo MtClpP1P2 is shown on the left and the crystal structure of Bz-LL-bound MtClpP1P2 is shown on the right. Reproduced from [47].

Tailed bacteriophages possess icosahedral capsid heads that contain their double stranded DNA genome and tails of differing lengths, which attach to the surfaces of their bacterial hosts [48]. The tail of the SPP1 phage is long and flexible, formed by gp17.1. Monomers of gp17.1 self-polymerize into tubes permitting their characterization by cryo-EM and MAS NMR. A recent study by the Lange group [49] showed that this SPP1 tail-tube comprises an inner β-barrel that is rigid on the nano- to millisecond timescale, while loops in the hinge regions (residues 40-59) and the C-arm (residues 143-176) are dynamic, thereby allowing the tail to be flexible. A hybrid structure of the tail tube was obtained by integrating structural restraints from MAS NMR with medium-resolution cryo-EM density maps illustrating details inaccessible otherwise. Based on their observation, the authors proposed a simplified two-state model for tube binding.

Calmodulin-regulated spectrin-associated proteins (CAMSAP)/Patronin comprise proteins associated with the microtubule (MT)-based cytoskeleton of eukaryotic cells. The characteristic CKK domain of these proteins interacts with MTs, assisting with stabilization and growth. In a recent report, Atherton et al. combined a MAS NMR structural and dynamics study of CKK with a cryo-EM structure of MTs in the CKK/MT complex. The preferential binding of human CAMSAP1 CKK (HsCKK) to MTs minus end was observed and compared with Naegleria gruberi (NgCKK), which strongly interacts with the entire mammalian MT lattice [50]. HsCKK is rigid, and MTs accommodate HsCKK via small adjustments along the binding interface, thereby resulting in preferential minus end recognition. In contrast, NgCKK appears to be more dynamic and is less discriminate in its MT interaction. The authors propose that these distinct dynamics are associated with differences in binding affinity to tubulin subunits and MT, thereby resulting in the preferential binding of HsCKK to the minus end of MTs. In this case, both methods contribute unique information, as cryo-EM lacked atomic resolution and could not provide dynamics information for the protein, while NMR could not reveal the MTs binding site that is remodelled by HsCKK and the binding specificity.

Future outlook

With further advances in integrative structural approaches that also incorporate dynamics, we anticipate that in the upcoming years many more systems will become amenable to atomic-level studies, and such investigations will expand tremendously. Rapid advances in NMR instrumentation, including magnets operating at magnetic fields of 28.2 T [51], faster spinning probes [52] and CPMAS CryoProbes [53], high-field dynamic nuclear polarization (DNP) instrumentation and new polarizing agents [54] will enable characterization of motions and conformational ensembles for a wide range of biological assemblies, including very large systems [55,56] as well as those that possess a large degree of disorder [57,58]. Specifically, ultrahigh magnetic fields will give access to enhanced resolution and sensitivity of NMR experiments allowing for characterization of larger structures as well as multiple conformations and minor species. Faster spinning NMR probes will bring better sensitivity with lower sample amounts and improved spectral resolution without the necessity for sample deuteration. CPMAS CryoProbes can boost sensitivity by the factor of 3 to 4 making experiments 10-15 times faster. For example, this technique made possible the study of the previously inaccessible conventional kinesin (KIF5B) motor domain complex with microtubules [59]. DNP is uniquely positioned to reveal atomic-level detail on systems, in which dynamic disorder interferes with room temperature MAS NMR, X-ray or cryo-EM measurements, by providing information on conformational distributions and minor conformers. Studying protein dynamics in native environments, such as cellular contexts, will likely become easier by NMR, with drug screening already available via in-cell NMR [60-62].

Integrating cryo-EM with MD simulations and deep learning approaches will also rapidly progress and will permit to maximize the information content [63,64]. The amount of generated data points can be reduced with AI-driven sampling, while the information extraction from every experiment, simulation or density map becomes faster and less involved with deep learning implementation based on pattern recognition. Already, single particle analysis of cryoEM maps allows for extraction of high-resolution snapshots along the continuous space of conformational dynamics [65] with time-resolved cryo-EM becoming an exciting new addition to the repertoire of methods to illuminate dynamics [66].

Thanks to the recent advances in computational hardware, MD simulations of very large systems are now possible [67], and timescales longer than microseconds will soon be accessible. Finally, the development of the adaptive force fields and user-friendly interfaces is hoped to make MD an easy-to-use integrative tool for all structural biologists [68,69]. Adaptive force fields will make MD simulations faster and less involved as nowadays even a small-molecule parametrization for MD calculations is a significant effort [70]; incorporation of non-canonical amino acids in simulations is another example of non-trivial modern task [71]. Adding broad MD simulation capabilities to a cloud-based web server, as it was done for example with docking [72], would have a tremendous impact on the field.