Disordered proteins attract increasing attention, due in part to the fact that intrinsic disorder is prevalent in the proteomes of higher order organisms (Ward et al. 2004). In addition, it has been established that the disordered states of proteins are not mere featureless ‘random coils’, but are characterized by secondary structure propensities of the polypeptide backbone, sometimes augmented with specific local interactions between side chains (Dill and Shortle 1991; Mittag and Forman-Kay 2007; Shortle 1996; Wirmer et al. 2005). High-resolution determination of residual structure in these inherently flexible molecules is therefore necessary (Bartlett and Radford 2009). NMR spectroscopy is undoubtedly the most appropriate technique to offer detailed insight into disordered states, being sensitive to the length and time scales characterizing the atomic structure (Bartlett and Radford 2009; Eliezer 2009; Mulder et al. 2009; Wirmer et al. 2005). In our analyses of intrinsically disordered proteins (IDPs) we have been interested in the measurement of scalar coupling constants to probe the local polypeptide backbone structure. To this end we have employed experiments that have been very successfully applied to folded proteins, and for which careful parameterization of three-bond coupling constants as a function of the backbone angle ϕ are available (Billeter et al. 1992; Case et al. 2000; Vuister and Bax 1993). For example, the most effectual and widely used coupling constants to define the backbone geometry, 3JHNHα, can be determined efficiently from a pair of 2D spin-echo difference measurements, obtained as HMQC (Ponstingl and Otting 1998) or HSQC (Petit et al. 2002) variants, recorded on 15N-labeled proteins. 2D 1H-15N correlation spectra of unfolded proteins are, however, severely compromised by resonance overlap, thereby limiting the number of probes available for conformational analysis. Since 3D experiments (Kuboniwa et al. 1994; Vuister and Bax 1993) to measure 3JHNHα include the 2D 1H-15N correlations as ‘diagonal peaks’, overlap in the 2D spectrum comparatively reduces the utility of 3D spectroscopy for the measurement of 3JHNHα in unfolded proteins. In an elegant approach, Lendel and Damberg (2009) very recently showed that improved resolution can be obtained by effectively removing the homonuclear coupling from the proton linewidth in a 3D J-resolved 1H-15N correlation spectrum, and were thereby able to achieve a near-complete analysis of the coupling constants for the natively disordered protein alpha-synuclein. In this paper we present an alternative approach to achieve excellent spectral resolution, which is based on 2D spectroscopy.

We show here that a much more complete set of 3JHNHα coupling constants can be obtained for uniformly (15N,13C)-labeled proteins by recording 13C′, rather than 15N, chemical shift modulation in the indirectly detected domain. The rationale for this strategy was derived from a comparison of 2D 1H-15N and 1H-13C′ correlation spectra (viz. Fig. 1) for the cytoplasmic domain of human neuroligin-3 (hNlg3cyt), obtained with gradient-sensitivity enhanced HSQC (Kay et al. 1992) and HNCO (Yang and Kay 1999) experiments, respectively. Both experiments were recorded with 100 complex time points in the indirect dimension, and processed identically. Figure 1 clearly demonstrates the improvement in resolution that is obtained in the 1H-13C′ spectrum of the unfolded protein, compared to 1H-15N correlation spectroscopy. In addition to the higher apparent spread of the signals in the 1H-13C′ spectrum, the 13C′ line widths (measured at 14.1 T) are also significantly smaller. For example, in the case of hNlg3cyt, the natural line widths measured in 1H-13C′ and 1H-15N spectra were 5.6 and 16.2 Hz, respectively.

Fig. 1
figure 1

2D 1H-15N HSQC (a) and 1H-13C′ H(N)CO (b) correlation spectra for the 139 amino acid construct of the intrinsically disordered protein domain hNlg3cyt

Full size image

With the high resolution of 1H-13C′ correlation spectroscopy in mind we developed a novel pulse sequence to measure 3JHNHα for uniformly 15N, 13C-enriched proteins, which is shown in Fig. 2. The scheme is a combination of HNCO triple resonance spectroscopy (Yang and Kay 1999) and HMQC-based spin-echo difference 3JHNHα experiments (Ponstingl and Otting 1998): through successive INEPT transfer steps the starting amide polarization is transferred to 13C′ for chemical shift encoding. The sequence of events up to point a can be described as H Z  → 2N Z C′ Z , and, after a 90° 13C′ pulse is followed by carbonyl carbon chemical shift encoding. Subsequently, at point b in the sequence, evolution of the anti-phase term −2NYC′Z takes place under the one-bond 15N-13C′ and 15N-1H coupling Hamiltonians simultaneously. After a delay κ = 1/(2JNH) the density operator is proportional to 4HZNXC′Z·cos(πJNC′κ) + 2HZNY·sin(πJNC′κ). A 90° 1H pulse transforms this into −4HYNXC′Z·cos(πJNC′κ)−2HYNY·sin(πJNC′κ), which interrupts any further evolution under the proton-nitrogen coupling, while allowing now for the simultaneous evolution under the one-bond 13C′-15N and three-bond 1HN-1Hα coupling Hamiltonians. At point c in the sequence the refocusing due to JNC′ is complete. Subsequently, after the final 90° 15N pulse, evolution under the mutual scalar coupling for a period κ also completely refocuses the nitrogen-proton anti-phase terms. A non-selective 180° proton pulse in the middle of the multiple quantum (MQ) evolution period ensures that 1H chemical shift evolution is fully refocused. In total, evolution due to the 3JHNHα coupling equals 2T at that point. After the last 90° 1H pulse the only terms present are now HX·cos(πJHNHα2T)−2HZH αY ·sin(πJHNHα2T). Only the in-phase term survives the subsequent echo period, which is comprised of an amide proton-selective REBURP refocusing pulse (Geen and Freeman 1991), combined with a gradient echo. This short additional element is also necessary to suppress any small amount of water magnetization that is not returned to the +Z axis by the last 90° 1H pulse in the sequence. Fourier transformation leads to an absorptive signal at the amide proton chemical shift, with its intensity encoding the three-bond 1HN-1Hα coupling constant. In order to quantify the 3JHNHα coupling constants a reference measurement is performed, during which selective inversion of 1Hα is achieved at time points T/2 (= 1/(4 × 1JNC′)) and 3T/2 (= 3/(4 × 1JNC′)), thereby effectively refocusing the coupling (these pulses are shown as boxed filled domes in Fig. 2). For this purpose, IBURP-2 pulses (Geen and Freeman 1991) were employed, which effect a rapid inversion during the central segment of the pulse. Practically, two experiments are recorded: one with the selective 1Hα pulses (I A) and one without (I B). Residue-specific values for the three-bond coupling constants are calculated from the intensity ratios of the resulting spectral signals as 3JHNHα= (2πT)−1acos(I B/I A).

Fig. 2
figure 2

Pulse sequence of the 13C′-resolved spin-echo difference experiment to measure 3JHNHα. Two versions of the experiment are executed, one with the boxed inversion pulses present (A) and one without (B). Narrow (wide) filled bars indicate 90° (180°) RF pulses applied along the x-axis, unless otherwise indicated. The 1H carrier is centered at the water resonance (4.76 ppm) and proton pulses are applied with a field strength of ω1/2π = 37.3 kHz. Proton decoupling is achieved using a WALTZ-16 decoupling scheme with ω1/2π = 5.0 kHz. The water flip-back pulse at the beginning of the sequence has an EBURP-1 profile (Geen and Freeman 1991) (ω1/2π = 715 Hz, 5.12 ms duration). The filled dome shaped proton pulses are used for the selective inversion of the 1Hα region and have an IBURP-2 profile (ω1/2π = 1.75 kHz, 2.9 ms duration). The open dome shaped proton pulse has a REBURP profile and inverts only the amide region (ω1/2π = 1.55 kHz, 4.01 ms duration) through phase-modulation. The 15N carrier is centered at 117 ppm and nitrogen pulses are applied with a field strength of ω1/2π = 6.0 kHz. Rectangular 90° (180°) 13C pulses are applied with a field of ω1/2π = Δ/√15 (Δ/√3), where Δ is the difference (in Hz) between 13Cα (57 ppm) and 13C′ (176 ppm). Decoupling during acquisition is done using GARP-1, with ω1/2π = 1.25 kHz. Values of the delays are: τ = 2.3 ms, κ = 5.4 ms, T = 16.67 ms. Gradient strengths in G/cm (length in ms): g0 = 8.0 (0.5), g1 = 5.0 (0.5), g2 = 15.0 (2.0), g3 = 20.0 (0.75), g4 = 15.0 (2.0), g5 = 1.0 (variable), g6 = 21.0 (0.1). Phase cycling: ϕ1 = x; ϕ2 = [x,−x]; ϕ3 = [x,x,−x,−x]; ϕ4 = [x,−x]; ϕreceiver = [x,−x,−x,x]. Quadrature detection in F1 is obtained by incrementing ϕ2, according to the States-TPPI protocol

Full size image

A few further aspects of the experiment are worth mentioning at this point. First, during the final part of the pulse sequence, no sensitivity losses incur from the refocusing of −2NYC′Z during the 2T period of 33 ms. The MQ evolution period nicely combines the refocusing period with the period that is required for the three-bond proton-proton coupling evolution, since these two processes require approximately the same amount of time.

Second, 1H composite pulse decoupling is applied during 13C′ evolution, since three-bond couplings to β-protons (2–6 Hz) would otherwise contribute to unresolved line-broadening. Under these conditions narrow lines are observed for 13C′, due to the absence of strong relaxation mechanisms: the dominant source of relaxation, even at moderate static magnetic field strength, is the chemical shift anisotropy (CSA), which is modest at 14.1 T. Since protons are far removed from the carbonyl carbon atom, dipolar relaxation contributions are small. For unfolded proteins a three-fold reduction in the line widths was observed by recording 13C′ rather than 15N-resolved spectra (see above), and this advantage also extends to folded proteins: for calbindin D9k the average carbonyl line width is 6.4 Hz, compared with 20.0 Hz for nitrogen. In addition, the high spectral dispersion displayed by 13C′ arises from the fact that the carbonyl chemical shift is influenced by the neighboring residue types to both sides (Wang and Jardetzky 2002; Wishart et al. 1995; Yao et al. 1997). The favorable relaxation properties and sequence-specific shielding contributions together ensure the high inherent resolution for 1H-13C′ correlation spectroscopy displayed by disordered proteins.

Third, a water flip-back strategy is used to avoid the transfer of water saturation to the solvent-exposed, exchangeable amide protons. Since all amides in the unfolded state are solvent-exposed, care has to be taken to control the state of water throughout the pulse sequence. Failure to establish that the water polarization is very similar in each of the two spin-echo difference experiments could lead to significant systematic errors. A three-pronged approach was taken here: (1) Since radiation damping can rapidly cause the large out-of-equilibrium water polarization to align with the external magnetic field we have limited the time during which it is not aligned along the +Z axis; (2) During the chemical shift evolution period where the water polarization is along –Z, weak gradients are employed to rapidly dephase any magnetization that rotates away from the −Z axis; (3) When the magnetization is in the transverse plane it is kept locked by a strong (ω1/2π~5 kHz) rotating transverse field, which is simultaneously used for composite pulse 1H decoupling. Residual water suppression after the final 90° 1H pulse is achieved by a selective 1HN REBURP refocusing pulse (Geen and Freeman 1991), bracketed by gradients, akin to the WATERGATE procedure (Piotto et al. 1992). Rather than judging the effectiveness of the water flip-back procedure from the degree of water suppression, an optional 90° 1H pulse was included at the end of the pulse sequence to query the longitudinal water polarization prior to acquisition. This was done, by comparing the signal intensity with that obtained after a single 90° 1H pulse. To avoid receiver overflow, the signal was moderated, by placing a 6 dB attenuator between the preamplifier and the observation receiver (i.e. prior to signal digitization). In this way it was established that the flip-back strategy was effective to 95% or better. More importantly, the water state was no different in the experiments that do or do not contain selective 1Hα decoupling pulses. We found that this procedure yielded more reliable estimates of the water signal magnitude, than using a short tapping pulse.

To benchmark the experimental procedure, the new pulse sequence was first applied to the small folded protein calbindin D9k, and measured 3JHNHα values were compared with coupling constants derived from a 3D HNHA experiment (Vuister and Bax 1993). The 3D data set was recorded using 47 × 62 complex data points for the proton and nitrogen domain, corresponding to maximum evolution times of 5.9 and 31.9 ms, respectively. Linear prediction was used in both indirect domains, followed by strongly apodizing window functions. The coupling constants extracted using this procedure were the same as those obtained without linear prediction. The total acquisition time was 30 h. For the 2D measurement 256 complex points were acquired in the 13C′ domain, using a maximum evolution time of 170.7 ms. A pair of experiments (once with and once without modulation due to the three-bond coupling) was recorded in 6 h, and this procedure was repeated four times, in an interleaved fashion, to ensure that any changes would affect both experiments equally, and the data subsequently added together.

As Fig. 3a shows, there is excellent agreement between the coupling constants for this protein, derived from 2D and 3D measurements, except for the presence of a systematic difference between the two datasets. Coupling constants obtained for ten residues for which the selective pulses do not perfectly effect the desired inversion of 1Hα magnetization, or completely refocus the 1HN region were excluded from the analysis, and Fig. 3a. For example, 3JHNHα values for Leu31 and Phe63 appeared seriously underestimated in the 2D measurement. This outcome was anticipated, as the 1Hα chemical shifts for Leu31 and Phe63 are 2.37 and 3.32 ppm, respectively, and fall outside the region for proper inversion.

Fig. 3
figure 3

a Comparison of experimental 3JHNHα coupling constants for the small folded protein calbindin D9k, obtained using 2D 1H-13C′ spin-echo difference and 3D HNHA experiments. Error bars on the 3D data set are similar to the size of the symbols, and are not shown for clarity. b Comparison of the experimental 2D 1H-13C′ spin-echo difference derived 3JHNHα coupling constants for calbindin D9k with those calculated from the 1.6 Å X-ray crystal structure (Svensson et al. 1992) 4ICB, using the Karplus parametrization by Vuister and Bax (Vuister and Bax 1993). c Comparison of experimental 3JHNHα coupling constants for the intrinsically disordered protein domain hNlg3cyt, obtained using 2D 1H-13C′ spin-echo difference and 3D HNHA experiments. Error bars are of similar size as the sample points. d 3JHNHα coupling constants for the intrinsically disordered protein domain hNlg3cyt, obtained with the new pulse sequence. Error bars are smaller than the size of the symbol for most data points. Empty circles show 3D HNHA-derived coupling constants for five additional residues, which give rise to overlapped cross peaks in the 2D 1H-13C′, but not in the 2D 1H-15N spectrum

Full size image

Although differential relaxation of the in-phase and anti-phase terms in the 3D HNHA scalar coupling measurement may lead to the systematic underestimation of the true J values for larger proteins (Harbison 1993; Kuboniwa et al. 1994; Vogeli et al. 2007), the short rotational correlation time for calbindin D9k (τC = 4.2 ns) suggests that a correction for differential relaxation would amount to underestimations of the true J couplings by approximately 5% (Kuboniwa et al. 1994). Even after making this correction it appears that the values from the 3D HNHA experiment underestimate the scalar coupling values by about 0.5 Hz. Interestingly, Wang and Bax (Wang and Bax 1996) required a similar systematic increase of 3D HNHA-derived relaxation-corrected 3JHNHα values for human ubiquitin (0.4 Hz) to obtain satisfactory agreement with values obtained using 2D J-modulated CT-HMQC spectra and HNCA[HA]-E.COSY experiments. It therefore appears that the J couplings measured with the proposed 2D spin-echo difference scheme are robust, even for small folded proteins.

The experimental coupling constants were compared with those calculated from a 1.6 Å X-ray crystal structure 4ICB (Svensson et al. 1992), using the Karplus parametrization by Vuister and Bax (Vuister and Bax 1993). The agreement with the 2D spin-echo difference derived coupling constants is very good, although there are a number of residues for which the coupling constants are further removed from the calculated curve than would be expected from the quality of the data. There exist two explanations for these differences. The first reason is that values for 3JHNHα calculated from a single static structure cannot be expected to faithfully represent experimental data in the case of dynamics. Residues that are known to be flexible from NMR relaxation studies (Kördel et al. 1992) indeed show poor agreement. For example, residues belonging to the termini (Glu4, Lys72-Ser74) and the linker region (Lys41) yield values between 6 and 8 Hz, indicative of averaging, and are more than 1 Hz removed from the values calculated from a static structure. Second, a number of further deviations appear to be the result of true differences between the structures in the crystal lattice and in solution. Calbindin D9k contains two loops, which coordinate a calcium ion each, providing backbone as well as side chain oxygen ligands. The first loop comprises Ala14-Glu27, and the second extends from Asp54 to Glu65. It is observed that data for residues of the four helices nicely fit to the Karplus relation, whereas the experimentally determined J couplings for all other residues that deviate by more than 1 Hz from calculation map to the two calcium binding loops. These are indicated in Fig. 3b. It appears that backbone angles in the loop regions differ by up to 10° between the crystalline and solution states. This assertion is confirmed by 3JC′C′ couplings, which report on the same dihedral angle (data not shown).

Once the methodology was validated, we applied the experiment to the disordered protein domain hNlg3cyt. In a previous study the structural nature of this domain was studied by a host of biophysical methods, including small-angle X-ray scattering, CD spectroscopy and NMR spectroscopy (Paz et al. 2008). The main conclusion of that work was that hNlg3cyt belongs to the family of intrinsically disordered proteins (IDPs), also referred to as natively unfolded proteins. Since no NMR assignments were available at the time of that study, the experimental data were used to derive an overall picture of the protein’s physical state. Meanwhile we have obtained near-complete backbone resonance assignments of the protein, such that a site-specific analysis of order and disorder in the sequence can be undertaken. Because scalar coupling constants are very sensitive to the local order of the peptide bond, they would serve as excellent probes to detect any residual structural propensity along the polypeptide chain.

Using the 3D HNHA experiment, a dataset of 46 × 62 complex data points was collected for the proton and nitrogen domain, corresponding to maximum evolution times of 5.7 and 31.9 ms, respectively. The total measuring time was 30 h. Due to extensive overlap we were only able to obtain 54 3JHNHα values for hNlg3cyt from this dataset. Therefore the pulse sequence of Fig. 2 was employed, using 128 complex data points for the 13C′ dimension, with a maximum evolution time of 160.0 ms. The experiments with and without 1Hα decoupling were repeated four times. The total measuring time was 30 h. Making use of the 2D 1H-13C′ correlation spectra, the number of probes amenable to analysis increased to 78, out of 105 possible (139-residue long sequence, with 10 amino acids in the poly-histidine tag, 16 proline and 8 glycine residues). Of note, for unfolded proteins the clear separation of the narrow 1Hα (4.0–4.8 ppm) and 1HN (7.9–8.7 ppm) chemical shift ranges does not hinder the coupling measurement as it does for folded proteins (see above), and no data needed to be excluded from further analysis. A single outlier was obtained (residue 56), but the reason for this discrepancy is currently unknown.

The 3JHNHα values measured for hNlg3cyt by 3D HNHA and 2D spin-echo difference spectroscopy are compared in Fig. 3c. Since the inter-proton dipolar correlation times in unfolded proteins are much smaller than those of folded proteins, it is unlikely that the effects due to differential relaxation need to be taken into account (Harbison 1993; Kuboniwa et al. 1994; Rexroth et al. 1995). However, a small systematic difference of about 0.4 Hz is still observed, as was the case for calbindin D9k. Such a systematic difference has previously been attributed to an underestimation of the coupling constants using the 3D HNHA experiment, suggesting that the values derived with the new experiment are reliable. In addition, since the 2D spin-echo difference-derived coupling constants agree well with the three-dimensional molecular structure for calbindin D9k, 3JHNHα values for unfolded proteins can be used as reliable probes of polypeptide conformation.

A plot of the measured 3JHNHα for hNlg3cyt as a function of residue number is shown in Fig. 3d. The average value of ca. 7 Hz is similar to values obtained for flexible peptides and other disordered peptides, corroborating earlier results (Paz et al. 2008). However, a few regions deviate from this trend, showing significantly lower values. The interpretation of this result is that conformations with values for the angle ϕ towards −60° are increasingly being populated. Importantly, from 3JHNHα values alone it is not possible to say whether these include canonical α-helical or polyproline type II (PPII) structures. This will await further investigation, and an analysis of additional experimental data is beyond the scope of this Communication, and will be presented elsewhere.

In summary, an experiment is presented to measure 3JHNHα, employing 13C′- instead of 15N-resolved experiments. The sizeable gain in the number of probes that become available for unfolded proteins suggests that the proposed experiment serves as an important complement to characterize the structure and dynamics of disordered states of proteins.