Direct Detection of Carbon and Nitrogen Nuclei for High-Resolution Analysis of Intrinsically Disordered Proteins using NMR Spectroscopy

Abstract

Nuclear magnetic resonance spectroscopy (NMR) is a powerful technique for characterizing the structural and dynamic properties of intrinsically disordered proteins and protein regions (IDPs & IDRs). However, the application of NMR to IDPs has been limited by poor chemical shift dispersion in two-dimensional (2D) ¹H-¹⁵N heteronuclear correlation spectra. Among the various detection schemes available for heteronuclear correlation spectroscopy, ¹³C direct-detection has become a mainstay for investigations of IDPs owing to the favorable chemical shift dispersion in 2D ¹³C′-¹⁵N correlation spectra. Recent advances in cryoprobe technology have enhanced the sensitivity for direct detection of both ¹³C and ¹⁵N resonances at high magnetic field strengths, thus prompting the development of ¹⁵N direct-detect experiments to complement established ¹³C-detection experiments. However, the application of ¹⁵N-detection has not been widely explored for IDPs. Here we compare ¹H, ¹³C, and ¹⁵N detection schemes for a variety of 2D heteronuclear correlation spectra and evaluate their performance on the basis of resolution, chemical shift dispersion, and sensitivity. We performed experiments with a variety of disordered systems ranging in size and complexity; from a small IDR (99 amino acids), to a large low complexity IDR (185 amino acids), and finally a ~73 kDa folded homopentameric protein that also contains disordered regions (133 amino acids/monomer). We conclude that, while requiring high sample concentration and long acquisition times, ¹⁵N-detection often offers enhanced resolution over other detection schemes in studies of disordered protein regions with low complexity sequences.

Graphical Abstract

1. Introduction

NMR is a powerful technique for studies of the structural and dynamic properties of biomolecules with atomic resolution in solution. Of particular interest are NMR studies of intrinsically disordered proteins, and disordered protein regions (IDPs and IDRs), because, due to their dynamic features, x-ray crystallography is often not possible or relevant. However, the study of IDPs and IDRs is often limited by poor chemical shift dispersion in heteronuclear correlation spectra [1]. For IDPs and IDRs (in the following, we use the term IDPs to mean both IDPs and IDRs), the absence of highly populated secondary and tertiary structure, extensive dynamics and conformational averaging, and low amino acid sequence complexity, homogenize the chemical environments of amino acid residues, leading to limited chemical shift dispersion and severe resonance overlap. These resolution challenges become more pronounced as the length of the polypeptide increases, thus placing practical limitations on the size of IDPs/IDRs for which full backbone resonance assignments can be made using traditional, ¹H-detection-based, multi-dimensional protein NMR techniques.

The extensive conformational flexibility that underlies poor NMR spectral dispersion is, however, associated with favorable magnetization relaxation properties, which provides opportunities to achieve ultra-high spectral resolution for IDPs. The NMR linewidth, or full-width at half-height (Δ_FWHH), of the Lorentzian lineshape is an important factor affecting the resolution and the signal-to-noise (S/N) ratio of NMR spectra. The value of Δ_FWHH is proportional to the transverse relaxation rate R₂ (Δ_FWHH = 1/πT₂ = R₂/π); however, additional factors contribute to observed resonance linewidths, including conformational exchange, magnetic field (B₀) inhomogeneity, sample heterogeneities, and temperature gradients. The latter sources of resonance broadening can be minimized by using modern NMR spectrometers and probes and through preparation of monodisperse protein samples; given this, the intrinsic relaxation properties of the detected nuclei strongly influence the resolution and sensitivity that can be achieved in NMR spectra. Among the spin ½ nuclei present in proteins, those with small values of the gyromagnetic ratio (γ; e.g., ¹³C and ¹⁵N) exhibit intrinsically lower sensitivity, as given by the relation S₀ ∞ γ_eγ _d ^3/2, where γ_e and γ_d correspond to the γ values for the excitation and detection nuclei, respectively. However, this can be compensated by slower relaxation rates relative to those for ¹H nuclei due to reduced dipole-dipole relaxation associated with low-γ nuclei [2]. These properties were originally exploited in a series of “proton-less” ¹³C direct-detection heteronuclear correlation experiments [3], which have subsequently been widely applied in studies of IDPs. In addition, however, direct detection of ¹⁵N resonances for backbone amide groups (¹⁵N_H) has re-emerged as a means to enhance resolution and sensitivity for systems that experience rapid amide group ¹H relaxation, including high molecular weight proteins [4]. However, the application of ¹⁵N detection to IDPs has not been widely explored. To assess the benefits and limitations of ¹³C and ¹⁵N detection for IDPs, we analyzed ¹H_N, ¹³C′, and ¹⁵N_H resonance line shapes and compared detection schemes for the most commonly used 2D correlation experiments, including 2D ¹H-¹⁵N and ¹³C-¹⁵N correlation experiments. As test cases, we investigated the central, Arf binding IDR of the E3 ubiquitin-protein ligase Hdm2 (Hdm2-ABD; residues 210 to 304) [5], a 182 amino acid, low complexity region of Surfeit locus protein 6 (Surf6-N; residues 1 to 182) [6], and a 130 amino acid region of Nucleophosmin (NPM1) that contains the pentamerization domain (residues 13–119) and two flanking IDRs (N130; residues 1 to 130) [7].

2. Materials and Methods

2.1. Protein Expression and purification

Hdm2-ABD

The Arf binding domain of Hdm2 (residues 210–304) with an N-terminal polyhistidine tag was expressed in Escherichia coli (E. coli) BL21 (λDE3) from the pET28a expression vector (Novagen) as described previously [5]. ¹³C/¹⁵N-labeled Hdm2-ABD was expressed using MOPS-based minimal media [8] containing [¹³C] D-glucose and ¹⁵NH₄Cl (Cambridge Isotope Laboratories). Cultured cells were harvested by centrifugation and lysed in 25 mM Tris HCl (pH 8.0), 500 mM NaCl, 5 mM β-mercaptoethanol (BME), and protease inhibitor cocktail (Sigma) by sonication. Lysates were clarified by centrifugation and Urea was added to the clarified extract to a concentration of 3 M. His-tagged Hdm2-ABD was purified by Ni²⁺-NTA affinity chromatography and eluted with buffer containing 6 M urea and 0.5 M Imidazole. Fractions containing Hdm2-ABD were dialyzed against 25 mM Tris HCl (pH 8.0), 150 mM NaCl, 5 mM BME and treated with thrombin to cleave the His tag. Cleaved Hdm2-ABD was buffer exchanged by dialysis into 25 mM sodium phosphate (pH 7.0), 50 mM NaCl and further purified using anion-exchange chromatography (Q Sepharose; Amersham Pharmacia Biotech, Inc.) using a linear gradient of 0.05′1 M NaCl over 0.1 L. NMR experiments were performed at a protein concentration of 1 mM in 25 mM Sodium Phosphate pH 6.0, 10 mM NaCl, 0.03 % NaN₃, and 10% D₂O.

Surf6-N

Surf6-N (residues 1–182 of human Surf6) was cloned into a pET28a expression vector encoding an N-terminal polyhistidine tag and a tobacco etch virus protease (TEV) cleavage site. E. coli Rosetta2 (DE3) cells were grown at 37 ^oC in MOPS based minimal media containing [¹³C] D-glucose and ¹⁵NH₄Cl (Cambridge Isotope Laboratories) to an optical density at 600 nm of 1.0 and expression was induced by the addition of 0.4 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) followed by incubation at 37 ^oC for an additional 3 hours. Cultured cells were harvested by centrifugation. Surf6-N was purified from inclusion bodies using the following procedure. Cells were lysed in 25 mM Tris HCl (pH 8.0), 500 mM NaCl, 5 mM β-mercaptoethanol (BME), 0.1% Triton-1000, and protease inhibitor cocktail (Sigma). The crude lysates were clarified by centrifugation and the supernatant was removed. The insoluble fraction was resuspended in buffer containing 8M GuHCl and disrupted by sonication. Following an additional round of centrifugation, Surf6-N was purified by Ni²⁺-NTA affinity chromatography and eluted with buffer containing 6 M urea and 0.5 M Imidazole. Fractions containing Surf6-N were dialyzed against 25 mM Tris HCl (pH 8.0), 500 mM NaCl, 5 mM BME to remove the urea and treated with TEV protease to cleave the His tag. Cleaved Surf6-N was mixed with 6M urea buffer and passed over Ni²⁺ resin to remove the His tag and TEV protease. The column flow-through fraction was then concentrated by ultracentrifugation using a 10 kDa MWCO filter (Amicron) and further purified by HPLC using an H₂O/CH₃CN/0.1% trifluoroacetic acid solvent system. NMR experiments were performed at a protein concentration of 1 mM in 25 mM Sodium Phosphate pH 7.0, 500 mM NaCl, 5 mM DTT, 0.03 % NaN₃, and 10% D₂O.

N130

N130 (human NPM1 residues 1–130) was expressed in E. coli as described previously [7]. N130 was purified via Ni²⁺-NTA affinity chromatography, followed by proteolytic removal of the polyhistidine tag by TEV protease followed by HPLC using an H₂O/CH₃CN/0.1% trifluoroacetic acid solvent system. NMR experiments were performed at a protein concentration of 400 μM in 10 mM Tris HCl pH 7.0, 150 mM NaCl, 2 mM DTT, 0.03 % NaN₃, and 10% D₂O.

NMR Spectroscopy

All experiments were collected at 298 K on a Bruker AVANCE NEO spectrometer operating at a proton frequency of 800 MHz equipped with a TXO cryoprobe optimized for ¹³C. For 1D spectra, processing (including Fourier transform and phase correction) and peak integration was performed by global spectral deconvolution using MestReNova software (Mestrelab Research). 2D Spectra were acquired with acquisition times corresponding to >3 X T₂ and >1 X T₂ in the F₂ and F₁ dimensions, respectively. 2D spectra were processed using data points corresponding to 3.14 X T₂ and 1 X T₂ in the F₂ and F₁ dimensions, respectively, and each dimension was zero-filled to the next power of 2. Apodization was performed by applying a cosine function (Shifted-Sine Bell; SSB = 2) and forward linear prediction was applied to complex data using 32 coefficients. 2D spectra were processed in Topspin 4.0 (Bruker). For 2D spectra, peak picking and integration by Gaussian fitting was performed using the program Sparky (UCSF).

3. Results

3.1. Choosing the Optimal Nucleus for Direct Detection; Analysis of Lineshapes in 1D NMR spectra for a Prototypical IDR, Hdm2-ABD

Backbone atoms in IDPs generally experience rapid local motions (e.g., local correlation times of <6 ns] [9]) that are uncorrelated with those in distal protein regions, giving rise to large transverse relaxation times (T₂ values) and sharp resonances. This is exemplified in 1D spectra for ¹³C/¹⁵N–labeled Hdm2-ABD; for example, ΔV_FWHH values for ¹H_N resonances were 4.7 ± 0.6 Hz (Fig. 1A, Fig. S1, Table S1), those for ¹³C′ resonances were 5.4 ± 1.5 Hz (Fig. 1B, Fig. S2, Table S1), and, finally, those for ¹⁵N_H resonances were 1.2 ± 0.3 Hz (Fig. 1C, Fig. S3, Table S1), by far the sharpest. The Δv_FWHH values are consistent with theoretical predictions for ¹H_N, ¹³C′, and ¹⁵N_H transverse relaxation rates for these nuclei in a uniformly ¹³C/¹⁵N-labelled protein at 800 MHz [2] and highlight the potential of ¹⁵N-detection for yielding narrow linewidths and optimal spectral resolution for IDPs.

Figure 1.

1D NMR lineshape analysis for ¹³C/¹⁵N Hdm2-ABD. (A) The amide region of a 1D ¹H spectrum. (B) The carbonyl region of a 1D ¹³C spectrum. (C) A 1D ¹⁵N spectrum showing amide nitrogen resonances. Insets in each panel provide a detailed view of well resolved resonances, as indicated by the black arrows, along with the fits from global spectral deconvolution (blue; see Methods). The black bar corresponds to a width of 10 Hz. It should be noted that the carbonyl region was chosen here to assess the obtainable resolution of ¹³C′ direct detection. ¹³C′ linewidths are influenced by magnetic field strength due to chemical shift anisotropy and should not be used to generally assess all ¹³C linewidths.

3.2. Choosing the Optimal Heteronuclear Correlation Scheme; Comparison of 2D Spectra for Disordered Proteins of Varying Size and Sequence Complexity

We next evaluated the resolution and sensitivity associated with different detection schemes for recording 2D ¹H-¹⁵N and ¹³C-¹⁵N correlation spectra for several IDRs. These two types of 2D spectra were chosen for analysis here because they are the basis for commonly used multi-dimensional NMR experiments used for establishing backbone resonance assignments. For 2D ¹H-¹⁵N correlation spectra, the ¹H-detected ¹H-¹⁵N HSQC [10] (Fig. S4) and ¹⁵N-detected ¹H-¹⁵N INEPT (Fig. S5) experiments were used. For 2D ¹³C-¹⁵N correlation spectra, the ¹³C-detected CON-IPAP experiment [3] (Fig. S6) and ¹⁵N-detected CON experiment [11] (Fig. S7) were used.

The results for 2D ¹H-¹⁵N correlation experiments with Hdm2-ABD, recorded for similar times (~5 hours for each) showed that most non-proline resonances (96 expected resonances) were resolved using either detection scheme, although the average S/N ratio for the ¹H-detected spectrum was >16-fold greater than for the ¹⁵N-detected spectrum (Fig. 2A,C; Table 1). The poor spectral dispersion associated with IDPs, especially in the ¹H dimension of 2D ¹H-¹⁵N correlation spectra, can limit spectral resolution. However, given the relatively narrow ¹⁵N_H resonances for Hdm2-ABD (~1 Hz), spectral resolution for the ¹H-deteced spectrum was sufficient to resolve 88 out of 96 possible resonances. With the acquisition and apodization parameters used for the ¹H-detected 2D spectrum, the resonance linewidths were ~15 Hz in the ¹H dimension and ~6 Hz in the ¹⁵N dimension. When directly detecting ¹⁵N magnetization, 91 of the 96 possible resonances were resolved owing to the enhancement in ¹⁵N_H linewidths, which were ~3 Hz using our standardized acquisition scheme. These results demonstrate that for small, well behaved IDRs like Hdm2-ABD, the majority of resonances may be resolved using 2D ¹H-¹⁵N correlation experiments, with ¹⁵N-detection offering superior resolution although with dramatically reduced sensitivity.

Figure 2.

Comparison of 2D heteronuclear correlation spectra for ¹³C/¹⁵N Hdm2-ABD and ¹³C/¹⁵N Surf6-N. (A) Overlaid ¹H- (red) and ¹⁵N-detected (blue) 2D ¹H-¹⁵N correlation spectra for Hdm2-ABD. The boxed region is shown enlarged with contours in the ¹H-detected spectrum scaled by 0.5 for clarity (A, inset) (B) Overlaid ¹H- (red) and ¹⁵N-detected (blue) 2D ¹H-¹⁵N correlation spectra for Surf6-N. The boxed region is shown enlarged with contours in the ¹H-detected spectrum scaled by 0.5 for clarity. Chemical shift perturbations were observed due to sample heating during decoupling (B, inset) (C) Zoomed view of a region from panel A, as indicated by the black arrow, with 1D projections along 8.08 ppm in the ¹H dimension (dotted line) highlighting the differences in ¹⁵N linewidths. (D) Zoomed view of a region from panel B (black arrow), with 1D projections along 8.57 ppm in the ¹H dimension (dotted line) (E) Overlaid ¹³C- (red) and ¹⁵N-detected (blue) 2D ¹³C’-¹⁵N correlation spectra for Hdm2-ABD. (F) Overlaid ¹³C- (red) and ¹⁵N-detected (blue) 2D ¹³C’-¹⁵N correlation spectra for Surf6-N. In panels E and F, proline resonances are denoted by the boxed regions. (G) Zoomed view of a region from panel E (black arrow), with 1D projections along 171.9 ppm in the ¹³C dimension (dotted line). (H) Zoomed view of a region from panel F (black arrow), with 1D projections along 175.4 ppm in the ¹³C dimension (dotted line). Axes for ¹H and ¹³ C-detected spectra have been rotated for clarity.

Table 1.

Acquisition parameters and observed spectral features for 2D heteronuclear correlation experiments performed on ¹³C/¹⁵N Hdm2-ABD.

Experiment	1H,15N-HSQC	N_HNINEPT	13C-CON	15N-CON
Points (F2 x F1)	4096 x 1500	5120 x 548	1024 x 1500	5120 x 136
Acquisition time (ms)	225 x 264	828 x 68	177 x 264	886 x 24
Scans	8	16	8	124
Recycle Delay (s)	1	1	1	1
Exp Time (hrs:min)	4:36	4:46	9:17	9:13
<FWHH 1HN> (Hz)	15.1 ± 0.9 (n = 33)	15.3 ± 1.0 (n = 41)	--	--
<FWHH 13C′> (Hz)	--	--	4.9 ± 2.1 (n = 41)	21.7 ± 1.8 (n = 35)
<FWHH 15NH> (Hz)	6.1 ± 1.3 (n = 33)	2.9 ± 1.1 (n = 41)	10.2 ± 1.0 (n = 41)	2.8 ± 0.8 (n = 35)
Signal-to-noise	3878	240	169	42
# peaks (observed/possible)	88/96	91/96	96/99	98/99

In contrast to Hdm2-ABD, Surf6-N is a 182 residue-long IDR that exhibits low sequence complexity and self-association that leads to broadening of some resonances. Consequently, the average Δ_FWHH values from 1D NMR lineshape analysis were larger than those measured for Hdm2-ABD, with values of 15.4 ± 1.8 Hz, 6.18 ± 1.2 Hz, and 3.4 ± 2.4 Hz for ¹H_N, ¹³C′, and ¹⁵N_H resonances, respectively (Fig. S8, Table S2). Note that, due to extensive resonance overlap, ¹H_N line widths were determined using a very high resolution 2D ¹H-¹⁵N HSQC spectrum (Fig. S8 A,D). Furthermore, the ¹H-detected 2D ¹H-¹⁵N HSQC spectrum (that used 3.14 X T₂ data points in the ¹H dimension) exhibited extensive resonance overlap (Fig. 2B,D). Resonance linewidths were ~26 Hz in the ¹H_N dimension and ~10 Hz in the ¹⁵N_H dimension, permitting resolution of 142 of the 177 possible resonances. Using ¹⁵N-detection, the resonance linewidths were ~31 Hz in the ¹H_N dimension and ~7 Hz in the ¹⁵N_H dimension, yielding an additional 13 well resolved resonances (Table 2). However, it should be noted that chemical shift perturbations were observed for several resonances as a result of sample heating (Fig. 2B, inset, Fig. S9), presumably due to the high decoupling pulse power applied during detection that was exacerbated by the high salt concentration in the sample buffer used (500 mM NaCl). This demonstrates that for a large IDR, spectral crowding can be partially overcome through the enhanced resolution provided by ¹⁵N-detection. In this case, probably due to the high salt concentration [12] or the differential effects of transient self-association on ¹H_N and ¹⁵N_H line widths, the signal-to-noise advantage of detecting ¹H was only ~6.5 (versus ~16 for the corresponding spectra for Hdm2-ABD) (Table 1). This example illustrates the potential of ¹⁵N-detection to improve resolution in 2D¹H-¹⁵N correlation spectra of challenging IDRs such as Surf6-N.

Table 2.

Acquisition parameters and observed spectral features for 2D heteronuclear correlation experiments performed on ¹³C/¹⁵N Surf6-N.

Experiment	1H,15N-HSQC	N_HNINEPT	13C-CON	15N-CON
Points (F2 x F1)	2048 x 510	2048 x 154	1024 x 510	2048 x 104
Acquisition time (ms)	113 x 92	369 x 21	236 x 92	369 x 24
Scans	16	48	32	296
Recycle Delay (s)	1	1	1	1
Exp Time (hrs:min)	2:40	2:54	12:22	12:25
<FWHH 1HN> (Hz)	26.1 ± 2.0 (n = 35)	30.7 ± 2.1 (n = 31)	--	--
<FWHH 13C′> (Hz)	--	--	7.5 ± 0.8 (n = 34)	20.9 ± 0.8 (n = 34)
<FWHH 15NH> (Hz)	9.7 ± 1.5 (n = 35)	7.3 ± 1.6 (n = 31)	9.7 ± 1.5 (n = 34)	5.8 ± 1.0 (n = 34)
Signal-to-noise	923	143	142	80
# peaks (observed/possible)	142/177	155/177	161/185	162/185

For IDPs, 2D ¹³C-¹⁵N correlation spectra provide enhanced chemical shift dispersion and tolerance to small variations in pH (in comparison with 2D ¹H-¹⁵N correlation spectra) and, importantly, resonances for proline residues [13]. For Hdm2-ABD, 2D ¹³C- ¹⁵N correlation experiments recorded for similar times (~9 hours for each) show that almost all resonances were resolved using either detection scheme, although the average S/N ratio for the ¹³C-detected spectrum was ~4-fold greater than that for the ¹⁵N-detected spectrum (Fig. 2E,G). For the ¹³C-detected spectrum, the average resonance linewidths were ~5 Hz in the ¹³C’ dimension and ~10 Hz in the ¹⁵N_H dimension, yielding 96 of the 99 possible resonances. For the ¹⁵N-detected spectrum, average resonance linewidths were ~3 Hz in the ¹⁵N_H dimension and ~22 Hz in the indirect ¹³C’ dimension. Despite the limited resolution enforced by constant-time editing in the indirect ¹³C dimension, the ¹⁵N-detected ¹³C-¹⁵N correlation spectrum provided the largest number of resolved resonances (98 of 99).

For the ¹³C-detected ¹³C-¹⁵N correlation spectrum for Surf6-N, the average resonance linewidths were ~8 Hz in the ¹³C’ dimension and ~10 Hz in the ¹⁵N_H dimension, yielding 161 of the 185 possible resonances. For the ¹⁵N-detected spectrum, average resonance linewidths were ~6 Hz in the ¹⁵N_H dimension and ~21 Hz in the indirect ¹³C’ dimension, yielding 162 of 185 possible resonances. Furthermore, all the resonances that were observed in the ¹³C-detect ¹³C-¹⁵N correlation spectrum were also observed in the ¹⁵N-detected spectrum, with the exception of two resonances, presumably due to the lower sensitivity of ¹⁵N detection. Therefore, in this case, where ¹⁵N_H and ¹³C’ linewidths are more similar than with Hdm2-ABD, ¹⁵N-detection provides only a slight resolution advantage and a less than 2-fold reduced S/N ratio (Table 2).

Taken together, our results show that for two IDRs that are characterized by rapid backbone conformational fluctuations and correspondingly slow transverse relaxation, ¹⁵N_H linewidths were narrower than ¹H_N and ¹³C′, thus allowing ¹⁵N-detection of 2D correlation spectra to afford the highest resolution at the expense of lower S/N. The S/N differential for ¹H- versus ¹⁵N- detected 2D ¹H-¹⁵N correlation spectra is much greater than for ¹³C- versus ¹⁵N- detected 2D ¹³C-¹⁵N correlation spectra. Therefore, ¹⁵N-detection may be reserved for cases where the highest possible resolution in 2D ¹H-¹⁵N correlation spectra is needed and the protein of interest can be prepared at high concentration (> 1 mM). In contrast, ¹⁵N-detection may be optimally used to record 2D ¹³C-¹⁵N correlation spectra when ¹⁵N_H line widths are significantly narrower than those for ¹³C’ nuclei, in which case the resolution advantages of ¹⁵N-detection can be manifested. Estimates of ¹³C’ and ¹⁵N_H line widths can be made from directly-detected 1D spectra and used to decide which detection strategy, ¹³C versus ¹⁵N, is likely to yield the highest resolution 2D ¹³C-¹⁵N correlation spectra. Based upon our results, we conclude that ¹⁵N-detection should be considered for the analysis of IDPs that exhibit poor resonance dispersion due to large size and/or low sequence complexity.

3.3. Comparison of ¹H- and ¹⁵N-detected TROSY-enhanced 2D ¹H-¹⁵N Correlation Spectra for N130, a High Molecular Weight Folded Protein with Disordered Regions

The development of transverse relaxation-optimized spectroscopy (TROSY) [14] dramatically extended the size limits of detection of 2D ¹H-¹⁵N correlation spectra for large molecular weight proteins. Additionally, the BEST-TROSY strategy that affords improved sensitivity and resolution was exploited in a set of multi-dimensional experiments to obtain sequential assignments for large IDPs [15]. Historically, TROSY has been utilized for ¹H-detection experiments; however, the long T₂ values of ¹⁵N_H resonances have prompted the recent development of ¹⁵N-detected TROSY experiments [4, 12]. Therefore, we tested ¹H- and ¹⁵N-detection to record TROSY-enhanced 2D ¹H-¹⁵N correlation spectra (Fig. S10 and S11, respectively) for the N-terminal oligomerization domain of Nucleophosmin (NPM1; termed N130), a 73.4 kDa homopentamer (residues 13–119, with a T_c value of 53.6 ± 0.9 ns) which also has two short IDRs (residues 1–13 and 120–130, with T_c values of 2.10 ± 0.05 ns for the latter segment) [16]. The first increments of ¹H- and ¹⁵N-detected ¹H-¹⁵N TROSY-enhanced correlation spectra were analyzed to estimate the T₂ values of resonances in the pentamerization domain (Fig. S12, Table S3); the average Δ_FWHH values of TROSY-enhanced ¹H_N and ¹⁵N_H resonances were 22.7 ± 7.0 Hz and 7.6 ± 0.8 Hz, respectively. These values were used to establish acquisition parameters for the two 2D spectra according to our standard scheme (see Methods). For the ¹H-detected TROSY-enhanced 2D ¹H-¹⁵N correlation spectrum, the average resonance linewidths were ~38 Hz and ~23 Hz in the ¹H_N and ¹⁵N_H dimensions, respectively, for residues within the pentamerization domain and ~31 Hz and ~17 Hz, respectively for IDR residues (Fig. 3A,C; Table 3). The corresponding ¹⁵N-detected 2D spectrum yielded slightly higher resolution albeit with lower sensitivity (~3-fold less). For example, the average resonance linewidths were ~43 Hz and ~15 Hz in the ¹H_N and ¹⁵N_H dimensions, respectively, for residues within the pentamerization domain and ~39 Hz and ~12 Hz, respectively for IDR residues (Fig. 3B,D; Table 3). These results illustrate the benefits of ¹⁵N detection for realizing the resonance narrowing effects of TROSY, especially for ¹⁵N_H resonances of the large pentamerization domain. In addition, ¹⁵N_H resonances for residues in the IDRs were narrower in the ¹⁵N-detected 2D spectrum, illustrating the resolution advantage associated with ¹⁵N detection for a large protein comprised of both a folded domain and IDRs.

Figure 3.

Comparison of ¹H- and ¹⁵N-detected TROSY-enhanced 2D ¹H-¹⁵N correlation spectra for ²H/¹⁵N N130. (A) ¹H-detected TROSY-enhanced 2D ¹H-¹⁵N correlation spectrum of N130, where the axes have been rotated for clarity. The small inset reveals a weak resonance (contoured at 4X). (B) ¹⁵N-detected TROSY-enhanced 2D ¹H-¹⁵N correlation spectrum of N130. The boxed regions in (A) and (B) are enlarged in the insets to show spectral regions containing resonances for IDR residues. (C) 1D projection in of ¹⁵N dimension of the ¹H-detected spectrum in (A) showing ¹⁵N_H linewidths. The expanded region of the spectrum, which contains pentamerization domain resonances, is scaled 3x for clarity. (D) 1D projection in the ¹⁵N dimension of the ¹⁵N-detected spectrum in (B) showing ¹⁵N_H linewidths at a scale of 22x. The expanded region of the spectrum containing pentamerization domain resonances is shown at a scale of 66x for clarity.

Table 3.

Acquisition parameters and observed spectral features for 2D heteronuclear correlation experiments performed on ²H/¹⁵N N130.

Experiment	1H-TROSY	15N-TROSY
Points (F2 x F1)	1024 x 218	1024 x 136
Acquisition time (ms)	45 x 42	198 x 14
Scans	64	96
Recycle Delay (s)	1	1
Exp Time (hrs:min)	4:18	4:31
<FWHH 1HN> pentamer (Hz)	37.8 ± 3.2 (n = 20)	43.2 ± 3.9 (n = 20)
<FWHH 1HN> IDR (Hz)	31.4 ± 1.9 (n = 8)	39.1 ± 2.2 (n = 8)
<FWHH 15NH> pentamer (Hz)	22.9 ± 2.6 (n = 20)	15.2 ± 2.4 (n = 20)
<FWHH 15NH> IDR (Hz)	17.4 ± 2.6 (n = 8)	11.6 ± 1.7 (n = 8)
Signal-to-noise	115	40
# peaks (observed/possible)	124/124	124/124

4. Discussion

NMR studies of IDPs at atomic resolution are often challenging due to poor resonance dispersion, especially for backbone amide protons. Further, the general lack of secondary and tertiary structure leads to very similar chemical shift values for sidechain carbons of amino acids of the same type, often limiting the establishment of sequential resonance assignments. The latter factor is exacerbated by the low complexity that is associated with the sequences of many IDPs and IDRs. The advent of cryogenic NMR probes offering high-sensitivity detection of resonances for ¹³C nuclei led to the development of a new generation of multi-dimensional NMR experiments based upon correlation of resonances for ¹³C’ and ¹⁵N_H rather than ¹H and ¹⁵N_H [3]. Because 2D ¹³C-¹⁵N correlation spectra offer improved spectral dispersion for IDPs, and because resonances of often abundant proline residues are detected, derivative multidimensional (≥3D) heteronuclear correlation experiments provided advantages for establishing sequential resonance assignments for IDPs [13, 17–20]. However, those based on sequential linkage of sidechain carbon resonances still suffered from amino acid-type chemical shift degeneracy, as discussed above. To overcome this problem, Showalter and co-workers developed a set of ¹³C-deteted 3D heteronuclear correlation experiments that allowed resonance assignments to be established on the basis of the chemical shifts of sequential ¹⁵N_H nuclei [13, 17]. However, the inclusion of two indirect ¹⁵N dimensions in these experiments was non-ideal with regard to achieving the highest possible resolution due to the time demands of extensive sampling in the indirect ¹⁵N dimensions. Recently, cryogenic NMR probes have become available that offer high sensitivity detection of ¹⁵N_H resonances in addition to those of ¹³C′, offering the opportunity to explore the possible advantages of direct ¹⁵N detection of spectra of IDPs and IDRs for improved resolution. Here, we compared spectral resolution and sensitivity for 2D ¹H-¹⁵N and ¹³C-¹⁵N correlation spectra recorded through direct detection of each of the two correlated nuclei for two protein regions that are entirely disordered (Hdm2-ABD and Surf6-N) and 2D ¹H-¹⁵N correlation spectra for a ~73 kDa folded, pentameric protein with two short IDRs (N130).

Our results for Hdm2-ABD showed that ¹⁵N_H resonances were by far the narrowest (1.2 ± 0.3 Hz ) amongst those also for ¹H_N and 13C′ nuclei (4.7 ± 0.6 Hz and 5.4 ± 1.5 Hz, respectively), and that this resolution advantage was manifested in improved resolution in ¹⁵N-detected 2D ¹H-¹⁵N (Fig. 2A,C) and ¹³C-¹⁵N correlation spectra (Fig. 2E,G). However, direct detection of ¹⁵N resonances was accompanied by significantly reduced S/N ratios in comparison with the corresponding ¹H- and ¹³C-detected 2D spectra. However, with a ~1 mM protein sample, the two ¹⁵N-detected 2D correlation spectra were recorded in reasonable time periods (2D ¹H-¹⁵N spectrum, ~4 ¾ hours; and ¹³C- ¹⁵N spectrum, 9 ¼ hours; Table 1). These measurement times were influenced by the design of our data sampling scheme, which involved linear sampling in the direct dimensions for >3 x T2 and in the indirect dimensions for >1 x T2 (except for the ¹⁵N-detected 2D ¹³C-¹⁵N correlation experiment, which was limited by the need for a constant evolution time for the ¹³C indirect dimension). This scheme was used to afford a level of resolution in both the directly and indirectly detected dimensions that was scaled to the relaxation properties of the relevant nuclei. This allowed for very high resolution in the directly detected dimensions and moderate resolution in the indirectly detected dimensions and also balanced considerations regarding sensitivity and total acquisition time.

Hdm2-ABD is a relatively small and well-behaved IDR and, despite its extensive disorder [21], yielded well resolved 2D ¹H-¹⁵N and ¹³C-¹⁵N correlation spectra that did not provide the opportunity to illustrate the full potential of ¹⁵N-detection for improved resolution. However, the Surf6-N IDR presented greater resolution challenges due to its larger size (185 versus 99 residues for Surf6-N and Hdm2-ABD, respectively), low sequence complexity (17.7% Ala, 15.5% Lys, 12.6% Glu and 11.0% Arg residues), and tendency for self-association (unpublished results, E. Gibbs and R. Kriwacki). In this case, ¹⁵N-detection of the 2D ¹H-¹⁵N correlation spectrum, which was recorded in ~3 hours, exhibited 155 resolved resonances versus 142 in the corresponding ¹H-detected spectrum (Fig. 2B, Table 1). In contrast, nearly identical numbers of resonances were resolved in the ¹³C- and ¹⁵N-detected 2D ¹³C’-¹⁵N correlation spectra for Surf6-N (161 and 162 resonances, respectively). This reflects the limitation on the maximal resolution that can be achieved in the ¹⁵N-detected experiment used here due to constant time evolution in the indirect ¹³C′ dimension (the Δ_FWHH value for ¹³C′ was ~21 Hz in comparison with ~8 Hz in the ¹³C-detected experiment; Table 1). Despite this limitation, these two spectra illustrate the advantage associated with 2D ¹³C-¹⁵N versus ¹H-¹⁵N correlation spectra in resolving resonances for a challenging, moderately sized IDR.

We further evaluated the performance of ¹H- and ¹⁵N-detection of 2D ¹H-¹⁵N correlation spectra for N130, which is comprised of a highly stable, ~70 kDa pentamerization domain flanked by short N- and C-terminal IDRs [7]. As expected [2, 14], ¹⁵N_H linewidths for resonances of residues in both the pentamerization domain and IDRs were much narrower than the corresponding ¹H_N resonances, and this was manifested in enhanced resolution in the ¹⁵N-detected 2D spectrum. However, because N130 is a symmetric pentamer, it exhibits only 124 resonances for non-proline residues and does not present the degree of resonance overlap that would be associated with a single polypeptide chain protein comprised of a similar size folded domain and IDRs. In this latter case, ¹⁵N-detection is likely to be required to achieve the highest possible spectral resolution in 2D ¹H-¹⁵N correlation spectra, as discussed by Wagner and co-workers [2, 4].

In summary, recent technical developments enable direct detection of ¹H, ¹³C, and ¹⁵N nuclei for isotope-labeled disordered proteins in reasonable time periods given that high concentration (~1 mM) can be achieved. This allows optimization of data acquisition to address the poor spectral dispersion associated with NMR studies of IDPs. Our data for several IDRs shows that the ¹⁵N_H resonance of amide groups is usually narrower than amide ¹H_N resonances and ¹³C’ resonances. Consequently, direct detection of ¹⁵N_H resonances will afford 2D ¹H-¹⁵N and ¹³C-¹⁵N correlation spectra with the highest possible resolution. However, this resolution advantage must be balanced against the sensitivity disadvantage associated with the low γ value of ¹⁵N versus that for ¹H and ¹³C. Because linewidths for ¹H_N, ¹³C’, and ¹⁵N_H resonances in IDPs can vary widely, we recommend assessing these linewidths for each protein and to choose a detection scheme accordingly. The amide ¹⁵N_H resonance linewidth will usually be much narrower than the corresponding ¹H_N linewidth, favoring ¹⁵N detection for high resolution when high protein concentration can be achieved. Alternatively, for proteins at lower concentrations (significantly less than 1 mM), ¹H detection of the 2D ¹H-¹⁵N correlation spectrum with extended sampling in the indirect ¹⁵N dimension may give the best compromise between resolution and sensitivity. In contrast to 2D ¹H-¹⁵N correlation spectra, 2D ¹³C-¹⁵N correlation spectra allow detection of resonances for proline residues, and also afford improved spectral dispersion for IDPs due to the broader range of chemical shift values for ¹³C’ and ¹⁵N_H resonances versus ¹⁵H_N resonances. Our results showed that the linewidths of ¹³C’ resonances were 2-fold or less large than those for ¹⁵N_H resonances, reducing the resolution benefits associated with direct ¹⁵N detection. Nonetheless, direct ¹⁵N detection was associated with slightly improved resolution in 2D ¹³C-¹⁵N correlation spectra for Hdm2-ABD and Surf6-N (Tables 1 and 2) and these advantages may be more dramatically manifested in studies of larger IDPs displaying highly crowded 2D ¹³C-¹⁵N correlation spectra. In our experiments, we employed linear data sampling in the indirect dimensions of 2D correlation spectra and advocate optimizing this to match the relaxation behavior of the sampled nuclei. The use of non-uniform sampling in the indirect dimensions and data processing using other than Fourier transformation [22–24] are likely to further extend the resolution benefits of direct ¹⁵N detection and also improve sensitivity, and should be utilized in the future. For ¹⁵N-detected 2D ¹³C-¹⁵N correlation spectra, in particular, resolution enhancements may be afforded by using D₂O buffer [2] and greater sensitivity may be achieved using magnetization transfer pathways starting from H-alpha in analogy to ¹³C-detected ¹³C- ¹⁵ N correlation spectra described previously [18]. In conclusion, as had been demonstrated previously by others [3, 13], 2D ¹³C-¹⁵N correlation spectra afford superior spectral dispersion and resolution for IDPs and the advantages of this 2D fingerprint can be enhanced through ¹⁵N detection when high protein concentration can be achieved.

Highlights.

• NMR detection schemes can be tailored to the unique linewidth features of IDPs.

• ¹⁵N-detection often offers enhanced resolution over ¹H- or ¹³C-detection schemes.

• High resolution ¹⁵N-detection is balanced by intrinsically lower S/N.