NMR methods for structural studies of large monomeric and multimeric proteins

Abstract

NMR structural studies of large monomeric and multimeric proteins face distinct challenges. In large monomeric proteins, the common occurrence of frequency degeneracies between residues impedes unambiguous assignment of NMR signals. To overcome this barrier, non-uniform sampling is used to measure spectra with optimal resolution within reasonable time, new correlation maps resolve previous impasses in assignment strategies, and novel selective methyl labeling schemes provide additional structural probes without cluttering NMR spectra. These advances push the limits of NMR studies of large monomeric proteins. Large multimeric and multi-domain proteins are studied by NMR when individual components can also be studied by NMR and have known structures. The structural properties of large assemblies are obtained by identifying binding surfaces, by orienting domains, and employing limited distance constraints. Segmental labeling and the combination of NMR with other methods have helped popularise NMR studies of such systems.

Introduction

Since its inception nuclear magnetic resonance (NMR) has evolved from a technique devoted to chemical analysis to a powerful and versatile tool for biological studies. Notably, NMR provides structural models of proteins in near physiological conditions and thus offers a complementary alternative to crystallographic studies. Functional NMR studies further contribute to the popularity of the method; NMR can probe protein dynamics, kinetics, and thermodynamics all at atomic resolution. This versatility results from the power of NMR correlation maps that simultaneously report on targeted molecular properties and correlate various nuclei in the protein. Unfortunately, the method has historically been limited to proteins smaller than 25 kDa. The two major obstacles to a universal application of NMR are sensitivity losses and increased spectral complexity, both of which are pronounced in large proteins. The loss in sensitivity originates from NMR relaxation, the process by which the NMR spin systems return to their equilibrium state. In particular, transverse relaxation leads to concomitant losses in signal intensities and signal line-broadening. In a major breakthrough, so-called TROSY (transverse relaxation optimised spectroscopy) techniques have provided a means to combat these adverse effects when combined with isotope labeling [1,2]. Many NMR experiments were adapted to TROSY methods and structural studies of larger systems emerged. However, these studies have most often focused on multimeric proteins, as reflected by depositions in the PDB (Figure 1), highlighting the additional challenges to which large monomeric proteins are subject. As a consequence, this review covers NMR studies of monomeric and multimeric proteins separately, with an emphasis on NMR method development for monomeric proteins and on combining NMR and other biophysical methods for multimeric proteins.

Figure 1.

Molecular-weight distribution of NMR structures in the PDB. Blue: Monomeric proteins, scaled to left axis. Green: homomeric proteins, right axis. Red: heteromeric proteins, right axis. The number of structures in all categories drops off dramatically at molecular weights above 20 kDa reflecting difficulties associated with solving these structures. The inset graph shows the number of structures greater than 50 kDa. These structures are predominantly heteromeric, reflecting a strategy in which structures of each subunit are solved individually and combined to model a complex.

Structures of large single domain monomeric proteins

NMR structure determination necessitates successful assignment of NMR signals, which relies on resolving NMR signals in correlation maps. The abundance of NMR signals in larger proteins results in spectral crowding and higher resolution is needed to prevent signal overlap. Higher-magnetic fields and TROSY help decrease signal overlap, but the high experimental resolution needed to resolve signals requires impractical measurement times. The acquisition time needed for an N-dimensional correlation map scales as the product of the number of points in each indirect dimension. Maximal resolution for a 50 kDa protein at 900 MHz [3] necessitates 6h30 for a 2D HN-TROSY-HSQC spectrum and 28 days for a 3D HNCO (assuming a 1D spectrum takes 1 minute). Many strategies have been suggested to accelerate NMR data acquisition [4]. In particular, the measurement time can be shortened dramatically by recording a sparse subset of points in the indirect dimensions, a method called non-uniform sampling (NUS)[5]. Alternatives to Fourier transformation such as maximum entropy reconstruction (MaxEnt) or multidimensional decomposition (MDD) have been used to produce 3D and 4D spectra[6,7], but suffer from non-linearity in signal intensities or leak-through of adjacent signals in crowded spectra, respectively. Recent processing techniques provide a faithful reproduction of the spectrum and do not require human intervention for optimizing parameters. Forward Maximum Entropy[8] can reproduce crowded protein NOESY spectra without false cross-peaks and maintains linearity in signal amplitudes, but requires extensive processing time. Minimisation of l1-norm, paired with iterative soft threshold [8–11], and variations of compressed sensing [12,13] have been equally successful and benefit from fast processing time. Thus, NUS is now practical for routine acquisition of reliable multidimensional NMR spectra with optimal resolution, which alleviates partial overlap and facilitates unambiguous assignments in larger proteins as shown in Figure 2a-d.

Figure 2.

Overcoming frequency degeneracies in a 53 kDa monomeric protein. Upper right: cartoon representation of a 3D HNCA; the labels refer to panels a,b,c, and d. a) H/N projection of HNCA with uniform (red) and non-uniform sampling (NUS, black). The inset emphasises a region in which four correlations appear as a single signal in the uniform experiment (thick line) but are resolved in the NUS experiment. The cross-hairs denote the frequencies of residue 107. b) H/C projection of the same spectrum. c) H/N plane at the frequency ω_C=ω_C(107), as defined in panel b. Signals only seen in the uniform data result from a leak-through of adjacent planes due to poor resolution in the carbon dimension. d) The corresponding H/C plane further emphasises the advantage of an increased resolution in nitrogen since fewer signals accidentally appear to share the same nitrogen frequency. e) The corresponding H/N plane of the (H)NCA(N)H experiment allows for straightforward identification of sequential residues 107 and 108.

The multitude of atoms in larger proteins results in frequency degeneracies that lead to indistinguishable correlations along various dimensions of NMR spectra, and in these cases traditional NMR experiments fail to provide unambiguous assignment. Here, NUS can be used to design experiments that would otherwise not be feasible for larger proteins. For instance, the 3D double-TROSY (H)NCA(N)H [14] provides correlations between backbone atoms of sequential residues along all three dimensions of the spectrum, ¹H,¹⁵N and ¹³C. Other experiments only provide sequential correlations along a single dimension. As a consequence, a tedious pairwise comparison of correlations can be replaced with visual identification of sequential cross-peaks akin to NOESY cross-peaks (Figure 2e). Here, NUS was used to rescue sensitivity by increasing the number of NMR transients accumulated whilst maintaining high-resolution in all dimensions. The application of NUS to improve sensitivity has recently been discussed in detail[15]. For 4D backbone experiments[16], NUS can overcome severe limitations in resolution that are otherwise needed to record spectra in reasonable time. Without NUS, low resolution induces overlap in each dimension and offsets the benefits of separating signals along additional dimensions.

Another impediment to determining structures of larger proteins is the scarcity of structural constraints. Recent advances have attenuated this limitation: torsion angles can be predicted from assigned backbone resonances with increased reliability [17] and novel experiments are available to measure residual dipolar couplings and provide bond orientations in larger proteins [18–20]. Distance constraints between protons are nevertheless necessary for de novo structure determinations; however, to minimise relaxation, larger proteins are predominantly deuterated, severely reducing the number of distance constraints from the outset. A protocol introduced by Kay and coworkers, which combines methyl-TROSY[2] with selective protonation of Ile (δ), Leu, and Val methyls[21], paved the way for overcoming this limitation. Other methyl groups can now be used as probes: Met [22–25], Ala [26,27], Ile (γ2) [28], and Thr methyls [29–31] can all be selectively labeled. Novel experiments have been designed to assign the methyl resonances of Ala and Ile(γ2) methyls[32]. Met and Thr methyls are assigned by using NOESY or mutagenesis and comparing the resulting spectra. Thus, all methyls that can be found in a protein can be selectively labeled to provide distance constraints.

Even with the accumulation of distance probes in selectively labeled large proteins it remains critical to maximise the assignment of nOe correlations to compensate for the loss of distance constraints when compared to protonated samples. Indeed, the accuracy and precision of NMR structures depends on the number of constraints and on their spatial distribution throughout the molecule; discarding a single constraint may have dramatic consequences for sparsely labeled samples. Non-uniform sampling can be combined with time-shared acquisition, which simultaneously provides several correlation maps with minimal impact on signal to noise[33,34], to measure both ¹⁵N and ¹³C edited 3D NOESY at once. In larger, selectively methyl labeled proteins, all distances involving either methyl or amide protons or both can be obtained with a single experiment that benefits from TROSY [35–38]. Likewise, a time-shared 4D experiment provides four 4D spectra involving methyl and amide protons in a single acquisition[38]. Recent progress in processing NUS data enables its implementation in time-shared NOESY experiments; together, the methods minimise the number of sacrificed constraints and increase the accuracy and precision of NMR structures of large proteins.

The methods described in this section render structure determination by NMR tractable for proteins up to 80 kDa. The exact limit depends on the protein structural and dynamic properties and on its solubility, which will affect the quality of NMR spectra. Fortunately, these effects can be identified upfront with a simple 2D HN-TROSY-HSQC and the sample design, as well as the buffer, can be modified to improve the spectral quality. In the end, recent developments should help structure determinations of large monomeric proteins become more common, although a certain degree of expertise is still required.

Structures of supra-molecular assemblies

The advent of TROSY techniques has readily provided access to studies of large homomeric proteins. The 900 kDa complex between tetradecameric GroEL and heptameric GroES was studied by amide proton and nitrogen HN-correlation maps[39] and the 670 kDa α₇β₇β₇α₇ core particle of the proteasome was studied by methyl spectra [40]. In both cases a divide and conquer approach was central in assigning the resonances. The monomeric units were assigned using conventional 3D or 4D HN-TROSY strategies and comparison with 2D correlation maps of the homomeric protein allowed transfer of the assignments. For GroEL/GroES, HN-correlation maps were used, and only resonances that did not undergo extensive shifts upon oligomerization were considered during subsequent analysis. In the proteasome studies, 3D experiments correlating methyl and beta or gamma carbons were also performed to assign signals that differ in monomer and multimer spectra; near complete assignment of the methyls in the proteasome core particle was achieved. More recent investigations of the folded states of protein substrates inside the proteasome brilliantly illustrated the biological impact of such methods [41].

Heteromeric systems can be studied with a similar divide and conquer strategy but are subject to further limitations. Symmetric homomers benefit from a signal accumulation of their individual units that heteromers lack; a 100 µM sample of a heptamer corresponds to an effective concentration of 700 µM in monomer units. In addition more resonances are likely to be affected by complex formation in heteromers when binding partners are different, thus reducing the number of resonance assignments that can be transferred from isolated monomeric units. For these reasons, the size of the systems that can be studied by NMR is somewhat smaller for heteromeric than for homomeric systems and many techniques need to be combined including non NMR methods[42]. In such cases, NMR is used to orient and position the components of the heteromer with respect to one another. Contacts between domains are identified by chemical shift perturbation[43], saturation transfer[44], or paramagnetic relaxation enhancement (PRE)[45], while domain orientations can be determined either by residual dipolar couplings[46–48] or relaxation analysis[49,50]. PRE was instrumental in determining the solution structure of a 40 kDa di-domain ubiquitin receptor associated with the proteasome, in which a flexible linker prevented traditional structure determination [51]. Likewise, in studies of protein ubiquitination, PRE could demonstrate that a ubiquitin-ligated E2 shifted to a tighter complex upon binding with its E3[52]. RDCs were used to orient two proteins and one RNA fragment of a ternary complex involved in mRNA processing, and PRE resolved the degeneracies in relative unit orientations[53]. Small (or wide) angle X-ray scattering has been particularly useful for determining quaternary models in conjunction with NMR54]. Chemical shift perturbation, RDCs, PRE, and ¹⁵N relaxation have all been used with SAXS, notably for structures with disordered regions [55–58]. CryoEM, X-ray crystallography, mass-spectrometry, and other techniques can all be combined with NMR to provide molecular models of heteromers[59]. These strategies can be extended to multi-domain proteins by using segmental labeling[60]. Each domain can be produced with a different isotopic composition and ligated subsequently either in vitro, when all domains are stable, or in-vivo with sequential induction of protein expression concomitant with alteration of growth media. The development of novel methods of ligation, e.g. by using sortases [61], should help increase the chances of successful segmental labeling for a given system.

Conclusions

The last decade has been relatively rich in NMR studies of larger proteins, and many biological studies have benefited from the unique versatility provided by the method. While the focus of this review has been on structural studies, an important observation emerges from inspecting the examples presented. NMR structural investigations of larger proteins are rarely pursued without accompanying studies exploiting other applications of NMR, such as defining ligand affinity or studying protein dynamics. This observation emphasises the rationale for utilizing NMR in studying larger proteins: the objective is not to compete with X-ray crystallography, which is much more suitable for high-resolution structures, but rather to use NMR for systems in which other functional studies are needed. NMR has the unique advantage of providing an atomic level read out. In the extreme case of supramolecular assemblies (>500 kDa), the “structural” study in fact consists in probing for molecular interactions, including structural rearrangements or modulation of dynamics upon binding events. When a molecular model of heteromers or multidomain proteins needs to be determined in solution, e.g. in presence of transient interactions, NMR is often combined with other techniques to provide a structure. Again, the structural studies are most often accompanied by other NMR studies, such as protein dynamics. The versatility of NMR has been the major driving force for developing means of overcoming challenges in studies of large proteins and the methods described in this review should help promulgate its application to heretofore inaccessible systems.

Highlights.

• Large monomeric proteins suffer from spectral crowding

• Non-uniform sampling of 3D and 4D data optimises spectral resolution

• Selective methyl labeling and tailored NMR experiments provide distance constraints

• Large multimeric proteins are studied with divide and conquer approaches

• Multimeric proteins studied by combining NMR methods with other techniques

Abbreviations

CryoEM

Cryo-electron microscopy

HNCO

a 3D experiment correlating amide protons and nitrogens with carbonyl carbons

HSQC

Heteronuclear Single-Quantum Coherence

MaxEnt

A particular implementation of maximum entropy reconstruction with a modified entropy function used for processing NUS data.

MDD

Multi-Dimensional Decomposition

NOESY

Nuclear Overhauser Effect (nOe) Spectroscopy

NUS

Non-uniform sampling

PDB

Protein Data Bank

PRE

Paramagnetic Relaxation Enhancement

RDC

Residual Dipolar Couplings

SAXS

Small Angle X-ray Scattering