Functional protein dynamics on uncharted time scales detected by nanoparticle-assisted NMR spin relaxation

Nanoparticle-assisted protein NMR opens window to observation of novel functional dynamics in the nano- to microsecond range.

Abstract

Protein function depends critically on intrinsic internal dynamics, which is manifested in distinct ways, such as loop motions that regulate protein recognition and catalysis. Under physiological conditions, dynamic processes occur on a wide range of time scales from subpicoseconds to seconds. Commonly used NMR spin relaxation in solution provides valuable information on very fast and slow motions but is insensitive to the intermediate nanosecond to microsecond range that exceeds the protein tumbling correlation time. Presently, very little is known about the nature and functional role of these motions. It is demonstrated here how transverse spin relaxation becomes exquisitely sensitive to these motions at atomic resolution when studying proteins in the presence of nanoparticles. Application of this novel cross-disciplinary approach reveals large-scale dynamics of loops involved in functionally critical protein-protein interactions and protein-calcium ion recognition that were previously unobservable.

INTRODUCTION

Nuclear magnetic resonance (NMR) spin relaxation measurements of proteins offer a wealth of information at atomic resolution about internal motional amplitudes and time scales under physiological conditions (1–4). This dynamics information provides critical mechanistic and thermodynamic insights into protein function involving loop motions, interdomain dynamics, recognition dynamics with small ligands, nucleic acids or other proteins, and partial unfolding/refolding events (5–13).

Every type of NMR parameters—such as scalar J-couplings, residual dipolar couplings, and average chemical shifts—depends in a unique way on both molecular structure and dynamics (1). In the case of spin relaxation parameters, the observable range of intramolecular protein motions covers fast time scales in the picosecond to low nanosecond range via longitudinal spin relaxation R₁, transverse spin relaxation R₂, and the heteronuclear Overhauser enhancement (NOE) experiments (1). By contrast, the chemical exchange and conformational exchange regime on tens of microseconds to seconds is covered by rotating frame relaxation experiments [R_1ρ, Carr-Purcell-Meiboom-Gill (CPMG), and chemical exchange saturation transfer] (14–17). R₁, R₂, and NOE relaxation data represent the convolution of overall rotational tumbling and intramolecular dynamics, which renders motions unobservable if they take place on time scales slower than the overall tumbling correlation time τ_P, which is typically around 10 ns. Thus, the intermediate time scale regime between low nano- and microsecond motions represents a critical gap in our ability to directly observe protein dynamics. The mere existence and atomic-detail character of these motions and their role in protein function are therefore largely uncharted territory.

As described in the paper, this situation can be addressed by studying protein dynamics in the presence of aqueous colloidal dispersions of synthetic nanoparticles (NPs) (Fig. 1). With their much larger size, NPs have tumbling correlation times τ_NP >> τ_P into the hundreds of nanosecond to microsecond range (18). For proteins that transiently interact with the NP surface and are in rapid exchange between a free and an NP-bound state, their spin relaxation reflects dynamics on the much broader picosecond to τ_NP range, thereby offering an unobstructed view of pico- to microsecond motions.

Fig. 1. Protein dynamics into the hundreds of nanosecond to microsecond range accessible to NP-assisted solution NMR.

(A) Protein molecules are in fast exchange between their rapidly tumbling free state and a slowly tumbling NP-bound state giving rise to effective transverse spin relaxation rates $R_2^{\text{NP}}$ versus $R_2^{\text{free}}$in the absence of NPs. (B) Simulated dependence of $R_2^{\text{free}}$ in the absence of NPs (blue) and $\mathrm{\Delta }{R}_2=R_2^{\text{NP}}-R_2^{\text{free}}$ in the presence and absence of NPs (red) on the internal correlation time τ_i and motional restriction (S² order parameter), which demonstrates the wide range of time scales sensitively probed by ΔR₂. The blue and red curves were normalized by setting their maximal values to 1.0.

RESULTS

Transverse R₂ spin relaxation experiments (1) are particularly sensitive to the presence of NPs. Consider a protein with an overall tumbling correlation time τ_P, which is intermittently bound to an NP with correlation time τ_NP with an exchange rate that is fast on the NMR chemical shift time scale (>10³ s⁻¹) but slow on the molecular tumbling time scale (<10⁷ to 10⁸ s⁻¹). The effective transverse relaxation rate of a protein ¹⁵N spin is then

$$R_2^{\text{NP}}=p R_2^{\text{bound}}+(1-p)R_2^{\text{free}}$$ (1)

where p and 1 − p are the bound and free protein populations with transverse relaxation rates $R_2^{\text{bound}}$ and $R_2^{\text{free}}$, respectively. R₂ is dominated by the spectral density of motion at zero frequency R₂ = cJ(0), where c is a constant (see the Supplementary Materials). When internal dynamics is represented for each ¹⁵N site in a model-free way with an S² order parameter and an internal correlation time τ_i (19), $J^{\text{free}}(0)=S^2{\mathrm{\tau }}_{\mathrm{P}}+(1-S^2)\frac{{\mathrm{\tau }}_{\mathrm{P}}{\mathrm{\tau }}_{\mathrm{i}}}{{\mathrm{\tau }}_{\mathrm{P}}+{\mathrm{\tau }}_{\mathrm{i}}}$ and $J^{\text{bound}}(0)=S^2{\mathrm{\tau }}_{\text{NP}}+(1-S^2)\frac{{\mathrm{\tau }}_{\text{NP}}{\mathrm{\tau }}_{\mathrm{i}}}{{\mathrm{\tau }}_{\text{NP}}+{\mathrm{\tau }}_{\mathrm{i}}}$, where ${\mathrm{\tau }}_{\mathrm{i},\mathrm{P}}^{-1}={\mathrm{\tau }}_{\mathrm{P}}^{-1}+{\mathrm{\tau }}_{\mathrm{i}}^{-1}$ and ${\mathrm{\tau }}_{\mathrm{i},\mathrm{N}\mathrm{P}}^{-1}={\mathrm{\tau }}_{\mathrm{N}\mathrm{P}}^{-1}+{\mathrm{\tau }}_{\mathrm{i}}^{-1}$ (used below). S² is a general measure of the motional restriction of a ¹⁵N-¹H bond vector varying between 0 (highly mobile) and 1 (static). The difference of R₂ in the presence and absence of NPs, ΔR₂, can then be expressed as

$$\mathrm{\Delta }{R}_2/(c p)=(R_2^{\text{NP}}-R_2^{\text{free}})/(c p)=S^2({\mathrm{\tau }}_{\text{NP}}-{\mathrm{\tau }}_{\mathrm{P}})+(1-S^2)({\mathrm{\tau }}_{\mathrm{i},\text{NP}}-{\mathrm{\tau }}_{\mathrm{i},\mathrm{P}})$$ (2)

If ∣τ_NP − τ_P∣ ≫ ∣τ_i,NP − τ_i,P∣, which applies when τ_i < τ_NP/10 and S² is nonzero, then Eq. 2 reduces to

$$S^2\cong \mathrm{\Delta }{R}_2/(c p {\mathrm{\tau }}_{\text{NP}})$$ (3)

It follows that the site-specific S² order parameters can be directly extracted from experimental ΔR₂ values whereby the global scaling factor (c pτ_NP)⁻¹ is identical for all residues. Since ΔR₂-derived S² reflects the cumulative effect of all internal motions with correlation times τ_i < τ_NP/10, it exceeds the time scale range τ_i < τ_P of standard model-free S² values determined in the absence of NPs by several orders of magnitude (Fig. 1B). The validity range of Eq. 3 is depicted in fig. S2.

NP-assisted NMR relaxation is demonstrated for the two globular proteins colicin E7 immunity protein (Im7) and calcium-binding domain 1 of the canine sodium-calcium exchanger NCX1 (CBD1). Im7 forms a four-helix bundle and binds to the deoxyribonuclease (DNase) domain of bacteriocin colicin E7 (ColE7), thereby inhibiting its strong toxic effect (20). Because Im7 exhibits an on-pathway folding intermediate, it has served as a model system for studying protein folding (21, 22). CBD1 is a globular domain with a β-sandwich fold and is part of the large cytosolic loop of the sodium-calcium exchanger (NCX) (23). CBD1, which is studied here in the absence of Ca²⁺, can bind up to four Ca²⁺ ions that produce an allosteric response that allows the exchange of intracellular Ca²⁺ with extracellular Na⁺ ions across the transmembrane domain of NCX (24).

Backbone ¹⁵N R₂ relaxation parameters were measured in the presence and absence of submicromolar to low micromolar anionic silica NPs (SNPs) of 20 nm diameter (Figs. 2A and 3A) with minimal effect on solvent viscosity (25). Because of the transient interactions of the protein molecules with SNPs, their average overall tumbling is slowed down, which explains why R₂ in the presence of SNPs exceeds that of the free state. Several residues display enhanced R₂ relaxation, which is caused by chemical exchange R_ex on the millisecond time scale and remains largely unaffected by the presence of SNPs. When focusing on ΔR₂ profiles (Fig. 2, B and C), R_ex effects cancel out, allowing an interpretation of ΔR₂ solely in terms of S² order parameters (Eq. 3). Conformationally rigid regions with large ΔR₂ belong to regular secondary structures and certain loops. At the same time, both proteins show large amounts of dynamics in the N- and/or C-terminal tails and in selected loop regions, manifested in a substantial decrease of their ΔR₂ values. To obtain a quantitative measure of the dynamics, ΔR₂ values were converted to S²(ΔR₂) order parameters (Eq. 3) by global scaling so that rigid secondary structures have average S²(ΔR₂) values of 0.85 (see the Supplementary Materials). This allows a direct comparison with traditional model-free S²(MF) values (blue circles in Figs. 2C and 3C) (22, 26) derived from standard ¹⁵N R₁, R₂, and NOE data of free protein reporting on picosecond to low nanosecond motions only.

Fig. 2. Dynamics of Im7 protein by backbone ¹⁵N-NMR spin relaxation and molecular dynamics (MD) simulations.

(A) ¹⁵N-R₂ relaxation rates measured in the absence (gray) and presence (black) of NPs. Data points with asterisks (*) indicate substantial chemical exchange R_ex effects. (B) R₂ differences (ΔR₂) of (A) with secondary structure of Im7 indicated at the bottom (4 α helices and 3₁₀ helix at N terminus). Experimental uncertainty (1 SD) is depicted by the shaded red area based on five independently measured ΔR₂ profiles (see fig. S6). (C) Comparison of ΔR₂-derived S² (red circles) with standard model-free S² order parameters (blue circles) and S² values determined from 1-μs MD trajectory with variable averaging time window (from 250 ps to 1 μs). (D) S²(ΔR₂) values mapped on three-dimensional (3D) crystal structure [Protein Data Bank (PDB) code 1AYI] show loops and tails that undergo substantial dynamics on pico- to microsecond time scales. N.A., not available.

Fig. 3. Dynamics of CBD1 protein domain from backbone ¹⁵N-NMR spin relaxation and MD simulations.

(A) ¹⁵N-R₂ relaxation rates measured in the absence (gray) and presence (black) of NPs. Gly³⁹⁹ (*) shows substantial chemical exchange R_ex. (B) R₂ differences (ΔR₂) of (A) with secondary structure of CBD1 indicated at the bottom (nine β strands). (C) Comparison of ΔR₂-derived S² (red circles) with standard model-free S² order parameters (blue circles) and S² values determined from 1-μs MD trajectory with variable averaging time window (from 250 ps to 1 μs). (D) S²(ΔR₂) values mapped on the 3D crystal structure of CBD1 (PDB code 2DPK) show loops and tails that undergo substantial dynamics on pico- to microsecond time scales.

For Loop II of Im7, the S²(ΔR₂) and S²(MF) profiles reflect a very similar degree of mobility with minima around 0.5, which indicates that relevant loop conformations are mostly explored on the fast picosecond to low nanosecond time scale. By contrast, Loop I has S²(ΔR₂) values that are substantially lower than S²(MF), revealing the presence of fast dynamics with S²(MF) > 0.53 and additional slower motions on the nano- to microsecond range with S²(ΔR₂) between 0.32 and 0.53. The S²(ΔR₂) profile for Loop I is wider than the S²(MF) profile and reaches into the C terminus of Helix I (residues Lys²⁴-Val²⁷). This fraying of Helix I is consistent with submicrosecond folding and unfolding of the last helical turn observed in the molecular dynamics (MD) simulation (Fig. 2C). These nano- to microsecond motions were missed in previous studies based on spin relaxation data in the absence of SNPs. Further fraying and partial unfolding occur in a lowly populated folding intermediate, where the stabilizing Glu²¹-Lys²⁴ salt bridge is absent and Glu²⁵ exhibits random-coil behavior (27).

CBD1 has a total of eight loops connecting the nine β strands (strands A to G). Loop F-G, which is missing in the x-ray crystal structure [Protein Data Bank (PDB) code 2DPK], is most flexible with S²(ΔR₂) values <0.2 that are in excellent agreement with the corresponding S²(MF) values. It suggests that this loop is highly dynamic with dominant correlation times on the pico- to nanosecond time scale that are fully reflected by spin relaxation data both in the presence and absence of SNPs. Loop E-F, which is located at the other end of the protein (Fig. 3D), has a distinctly different behavior. According to standard ¹⁵N-relaxation analysis, it is only moderately flexible with S²(MF) > 0.60. However, S²(ΔR₂) values dip as low as 0.35, reflecting the presence of substantial amounts of additional dynamics into the hundreds of nanosecond range. Loop E-F therefore probes a much broader ensemble of conformations than suggested by traditional ¹⁵N-relaxation data alone. Similarly, the C-terminal residues of CBD1 also exhibit S²(ΔR₂) < S²(MF) caused by dynamics on both the pico- to nanosecond and nano- to microsecond time scales.

Independent corroborating evidence of the presence and location of fast and slow time scale dynamics can be gleaned from extended MD computer simulations. MD trajectories in explicit water solvent were computed and analyzed for both protein systems. ¹H-¹⁵N S²(MD) values, computed from MD trajectories using 12 different isotropic reorientational eigenmode dynamics (iRED) (28) time-averaging windows τ_iRED ranging from 250 ps to 1 μs, are plotted in Figs. 2C and 3C. Loops I and II in Im7 and Loops E-F and F-G in CBD1 all show a steady drop of S²(MD) when averaging over slower time scales (longer τ_iRED windows), whereas secondary structures and other loops remain notably rigid. Adequate sampling of Loop I conformations is only achieved when considering the full-length trajectory, whereas Loop II samples most of the relevant conformations already within τ_iRED ~25 ns, which corresponds to the time scale window accessible by ¹⁵N spin relaxation of the free protein (28). Similarly, for Loop E-F and the C terminus, the entire 1-μs trajectory length is required to reach good agreement with experimental S²(ΔR₂), supporting the experimental finding that this loop displays large amplitude motions on time scales inaccessible by standard NMR relaxation methods.

The fast and slow dynamics of Im7 Loop I and CBD1 Loop E-F were further analyzed by principal component analysis (PCA) in backbone dihedral angle space (29). The score plots of Fig. 4 (A and C) show that both loops transition between multiple conformational clusters with representative cluster snapshots depicted in Fig. 4 (B and D). The diverse nature of the conformational loop ensembles reveals multiple possible interaction modes with partner molecules.

Fig. 4. Visualization of complex loop motions by 1-μs MD trajectories.

(A and C) Backbone dihedral angle–based PCA score plots of (A) Loop I of Im7 and (C) Loop E-F of CBD1 display multiple distinct conformational loop clusters. (B and D) 3D visualization of three cluster centers for each protein indicated by colored diamond (♦) symbols in score plots.

DISCUSSION

Despite their potential significance for biological function, observation of internal protein motions on the nano- to microsecond range has been a major challenge in the past. To make this motional regime accessible, rotational tumbling is sometimes slowed down by increasing solvent viscosity through the addition of ethylene glycol or glycerol. However, this also tends to stiffen or slow down internal motions, keeping slower motions mostly out of range of NMR relaxation experiments (30, 31). Alternatively, increasing the size of the molecule, as was shown for RNA, can slow down tumbling and open up observation of slower nanosecond motions (13). A recent NMR relaxometry approach, measuring R₁ over a wide range of magnetic field strengths (0.33 to 22.3 T), could access internal protein motions into the low nanosecond range (32).

As demonstrated here, the use of slowly tumbling NPs to which a biomolecule can bind in fast exchange offers a general solution to this long-standing challenge. Rapid exchange has been observed in NMR-based NP binding studies of globular (33) and intrinsically disordered proteins (34). The binding equilibrium between NPs and proteins can be shifted by adjusting the NP concentration in the submicromolar to low micromolar range to optimize NMR line broadening (ΔR₂ < 10 s⁻¹ with p ≈ 0.01), allowing ΔR₂ measurements with high accuracy.

For both CBD1 and Im7, NP-assisted spin relaxation reveals previously unknown slow time scale dynamics of loops displaying low S²(ΔR₂) values that are directly involved in protein function through electrostatic interactions with their binding partners. Im7 has pico- to femtomolar affinity to its target colicin protein partner (35), whereby acidic residues Asp³¹ and Asp³⁵ of Im7 Loop I form strong salt bridges with basic colicin residues Arg⁵²⁰, Lys⁵²⁵, and Lys⁵²⁸ (fig. S7). For CBD1, Ca²⁺ ions bind to negatively charged side chains of Asp⁴⁴⁶ and Asp⁴⁴⁷ of Loop E-F, which allosterically triggers the exchange of Ca²⁺ versus Na⁺ ions through the transmembrane domain (24). The central role of electrostatic interactions in biological complex formation is well established (36). The dynamic nature and high plasticity of the interacting loops identified here are likely to help fine-tune these interactions and optimize the affinity and specificity with interacting partner proteins and metal ions.

Low S²(ΔR₂) values are reporters of previously unknown contributions to the conformational entropy, analogous to S²(MF) (12, 37), allowing a more quantitative and complete understanding of the thermodynamics of protein-protein and protein-ligand interactions. As demonstrated here (Figs. 2C and 3C), S²(ΔR₂) provide powerful benchmarks for the testing (and potential improvement) of computational models, in particular MD simulation protocols and their underlying force fields. As MD trajectories now routinely extend into the hundreds of nano- to microsecond range, there is a pressing need for experimental benchmarks (38) that permit evaluation of these trajectories for which S²(ΔR₂) profiles are highly suitable.

Because of their spherical shape, SNPs tumble isotropically along with the proteins bound to them (Fig. 1A), which further simplifies the model-free interpretation of ΔR₂. It assumes that protein-NP interactions do not significantly affect protein structure and dynamics, similar to alignment media used for NMR residual dipolar coupling measurements (39, 40). This should hold when a globular protein interacts with NPs mostly in a nonspecific manner, for example, by making contacts with many different surface sections. Specific binding, for example, to a mobile loop, may systematically alter its mobility in the presence of NPs. By comparing ΔR₂, profiles measured for different types of NPs could help identify these situations. For Im7, few residues (Asn⁷⁹, Gly⁸⁰, Gly⁸³, Gln⁸⁶, and Gly⁸⁷) display S²(ΔR₂) > S²(MF). This modest S² inversion may reflect chemical exchange contributions with the SNPs, although no line broadening effects were observed in these regions. Because these residues are neither cationic nor hydrophobic, it is unlikely that these residues are limited in their mobility because of direct interactions with the SNPs.

We used anionic SNPs, but other types of NPs should work similarly well for the proteins studied here and other biomolecules. Anionic SNPs have partially dissociated silanol groups at their surface, which give rise to attractive and repulsive electrostatic interactions with charged protein residues. SNPs can also display hydrophobic interactions that are presumably mediated by siloxane groups at the NP surface, but the details of this interaction mechanism are not fully understood (41, 42).

The cross-disciplinary NP-assisted relaxation method introduced here uncovers previously undetected motions on submicrosecond time scales of protein regions that play key roles in the function of these systems by mediating receptor and ligand interactions. It seems likely that protein motions on these time scales are widespread. NP-assisted spin relaxation enables their comprehensive characterization at atomic resolution and shed new light on biomolecular function.

MATERIALS AND METHODS

The full description of Materials and Methods can be found in the Supplementary Materials. A brief summary is provided here.

Sample preparation

Uniformly ¹⁵N-labeled and/or ¹³C-labeled proteins Im7 and CBD1 were overexpressed in Escherichia coli with final NMR concentrations of 500 and 400 μM, respectively. For SNP-containing samples, Bindzil 2040 colloidal SNPs (AkzoNobel) with a 20 nm diameter were dialyzed in pH 7.0 buffers and mixed directly with proteins. All samples were stable over the entire course of the NMR data acquisition.

NMR spectroscopy

NMR experiments were performed on Bruker AVANCE III HD spectrometers operating at 850-MHz ¹H frequency (19.97 T) for resonance assignments and relaxation measurements at 298 K (for Im7) and 306 K (for CBD1). ¹⁵N spin relaxation rates (R₁ and R_1ρ) for protein samples both in the absence and presence of SNPs, as well as a {¹H}-¹⁵N steady-state NOE experiment in the absence of SNPs, were measured and analyzed using standard two-dimensional NMR ¹⁵N relaxation methods (43, 44).

MD simulations

MD simulations were performed by standard methods using the GROMACS 5.1.2 package (45) with initial structures of Im7 and CBD1 built on the basis of crystal structures (PDB codes, 1CEI and 2DPK). AMBER ff99SBnmr1 protein force field (46) together with the TIP3P explicit water model (47) was used. The integration time step was set to 2 fs, and Na⁺ ions were added to neutralize the total charge of the system. Particle mesh Ewald summation with a grid spacing of 1.2 Å was used to calculate long-range electrostatic interactions. After equilibration, the production run was performed in the NPT ensemble at 300 K and 1 atm for 1 μs. Amide order parameters S²(MD) were back-calculated from MD trajectories using the iRED method with varying lengths of the time-averaging window (48). PCA was performed on the backbone dihedral angles of loop residues.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/5/8/eaax5560/DC1

Supplementary Materials and Methods

Fig. S1. Simulated dependence of $R_2^{\text{free}}$ and ΔR₂ on internal correlation time τ_i and S² order parameter.

Fig. S2. Range of validity of Eq. 3 for the extraction of S² from ΔR₂.

Fig. S3. Experimental ¹⁵N spin relaxation parameters of Im7 in the absence of NPs.

Fig. S4. Experimental ¹⁵N spin relaxation parameters of CBD1 in the absence of NPs.

Fig. S5. Comparison between NMR S²(ΔR₂) and x-ray B-factors of backbone nitrogen atoms in crystal structures.

Fig. S6. Dependence of ΔR₂ values on SNP concentration.

Fig. S7. Mapping of experimental S²(ΔR₂) onto the structural model of Im7 when bound to the DNase domain of colicin E7.

References (49–63)