NMR profiling of biomolecules at natural abundance using 2D ¹H–¹⁵N and ¹H–¹³C multiplicity-separated (MS) HSQC spectra

Abstract

2D NMR ¹H–X (X = ¹⁵N or ¹³C) HSQC spectra contain cross-peaks for all XH_n moieties. Multiplicity-edited ¹H–¹³C HSQC pulse sequences generate opposite signs between peaks of CH₂ and CH/CH₃ at a cost of lower signal-to-noise due to the ¹³C T₂ relaxation during an additional 1/¹J_CH period. Such CH_n-editing experiments are useful in assignment of chemical shifts and have been successfully applied to small molecules and small proteins (e.g. ubiquitin) dissolved in deuterated solvents where, generally, peak overlap is minimal. By contrast, for larger biomolecules, peak overlap in 2D HSQC spectra is unavoidable and peaks with opposite phases cancel each other out in the edited spectra. However, there is an increasing need for using NMR to profile biomolecules at natural abundance dissolved in water (e.g., protein therapeutics) where NMR experiments beyond 2D are impractical. Therefore, the existing 2D multiplicity-edited HSQC methods must be improved to acquire data on nuclei other than ¹³C (i.e. ¹⁵N), to resolve more peaks, to reduce T₂ losses and to accommodate water suppression approaches. To meet these needs, a multiplicity-separated ¹H–X HSQC (MS-HSQC) experiment was developed and tested on 500 and 700 MHz NMR spectrometers equipped with room temperature probes using RNase A (14 kDa) and retroviral capsid (26 kDa) proteins dissolved in 95% H₂O/5% D₂O. In this pulse sequence, the 1/¹J_XH editing- period is incorporated into the semi-constant time (semi-CT) X resonance chemical shift evolution period, which increases sensitivity, and importantly, the sum and the difference of the interleaved ¹J_XH-active and the ¹J_XH-inactive HSQC experiments yield two separate spectra for XH₂ and XH/XH₃. Furthermore we demonstrate improved water suppression using triple xyz-gradients instead of the more widely used z-gradient only water-suppression approach.

1. Introduction

There is an increasing need to fingerprint therapeutic biomolecules by solution NMR in formulated drug products at natural abundance [1–8]. Two-dimensional hetero-nuclear ¹H–X (X = ¹⁵N or ¹³C) spectral patterns are sensitive to chemical structure and therefore can detect structural change at the level of individual nuclei. These methods are superior to homo-nuclear two-dimensional spectra because of their increased chemical shift dispersion, less homo-nuclear J-splitting, and concomitant reduced peak overlap [2]. Two-dimensional methods have become even more powerful with the advent of multiplicity-edited ¹H–¹³C HSQC, HMQC, HMBC and H2BC experiments [9–19]. Multiplicity-edited pulse sequences use a dedicated period of 1/¹J_CH to edit CH₂ and CH/CH₃ peaks into the opposite phases in the same spectrum. Most edited experiments were developed for and applied to small molecules at ¹³C natural abundance dissolved in deuterated solvents and are therefore multiple-quantum based experiments. Additionally, generally, small molecules contain fewer overlapping resonances and have longer T₂ values compared to larger protein molecules.

For larger molecules like proteins, HSQC-based experiments have higher sensitivity than HMQC-based experiments, because, due to the transverse ¹H magnetization during t₁, the HMQC-based experiments have additional T₂ decay [20]. Interestingly, among all published ¹³C multiplicity-edited 2D experiments only the HSQC version was demonstrated on ubiquitin dissolved in D₂O [16].

Experimental protein NMR spectra are typically collected in H₂O, where some protein backbone amide NH peaks overlap with side-chain NH₂ peaks. In addition, CH_n peaks also overlap with CH_m (where n ≠ m) peaks, especially in spectra where peaks are folded. Finally, XH_n peaks in un-folded peptides and proteins may overlap with peaks from un-structured regions of a folded protein. Unresolvable 2D cross-peaks, though edited, will adversely affect chemical shift assignment and, concomitantly, chemometric analysis of 2D-NMR data acquired on biomolecules at natural abundance. Thus, to resolve potentially overlapped XH_n peaks, we present a multiplicity-separated (MS) HSQC experiment. Here, the most important feature of multiplicity-separation is achieved by separately collecting the ¹J_XH-active and the ¹J_XH-inactive spectra in an interleaved fashion; then, the sum and the difference are calculated directly to yield separation. This new approach absorbs the previously separate editing period of 1/¹J_XH into a modified semiconstant time (semi-CT) chemical shift evolution period, reducing the total time that X magnetization spends in the transverse plane, and minimizing signal loss due to T₂ relaxation. In addition, water suppression can be readily achieved in 95% H₂O/5% D₂O using either flip-back pulses [21] or triple-axis gradients. Herein, we demonstrate the new experiments on a 14-kDa natural abundant RNase A and a 26-kDa ¹⁵N-labeled retroviral capsid within a reasonable amount of spectrometer time (1–2.5 days for the natural abundance sample) using a room-temperature probe-head.

2. Results and discussion

2.1. The ¹J_XH-active and the ¹J_XH-inactive semi-CT HSQC pulse sequences

The original multiplicity-edited ¹H–¹³C HSQC pulse sequence [16,17], commonly found in vendor pulse sequence libraries, has the signature spin chemical shift and ¹J_CH evolution shown in Eq. (1). Eq. (1) depicts the ¹³C antiphase H_zC_y magnetization evolution between the first and the last INEPT periods,

$${\mathrm{H}}_z{\mathrm{C}}_y\stackrel{t_1}{\to }{\mathrm{H}}_z{\mathrm{C}}_{y}cos{\omega }_{\mathrm{C}}{t}_1\stackrel{1/J_{\text{CH}}}{\to }{\mathrm{H}}_z{\mathrm{C}}_{y}cos{\omega }_{\mathrm{C}}{t}_1{cos}^n\pi$$ (1)

where t₁ is the ¹³C chemical shift evolution period, ω_C is the ¹³C Larmor frequency and n is the number of ¹H attached to ¹³C. Depending on the value of n, the sign of the H_zC_y magnetization will be positive for CH₂ and negative for CH/CH₃ cross-peaks. The maximum amount of time that ¹³C magnetization spends in the transverse plane is t_1max + 1/¹J_CH.

To keep the editing feature and reduce T₂ losses, we incorporated an extra 1/¹J_XH time, 8–11 ms, into the t₁ chemical shift evolution period, i.e., t_1max of 40–50 ms. The semi-CT method is selected for this purpose. The semi-CT was originally used for ¹H chemical shift evolution during INEPT transfers to ¹³C in order to minimize ¹H T₂ losses [22,23]. Here, we modify the semi-CT during the hetero-nuclear X chemical shift evolution to generate the multiplicity- separated spectra. Briefly two semi-CT HSQC pulse sequences were run in an interleaved fashion. The difference between the two sequences lies in where the ¹H 180° pulse is applied (Fig. 1). In the ¹J_XH-active experiment, the ¹H 180° pulse is applied at the red position and properly decrementing t_a and incrementing t_b and t_c (Table 1), the net ¹J_XH evolution time is maintained constant as 1/¹J_XH (Eq. (2)). Therefore, the editing feature is retained and the t₁ chemical shift labeling is properly maintained. The maximum time that hetero-nuclear X magnetization spends in the transverse plane is t_1max + (1/¹J_XH)/(N + 1) ≈ t_1max, where N is the number of real points in t₁ acquisition.

Fig. 1.

The pulse sequences for the ¹J_XH-active/inactive semi-CT HSQC experiments, which can be either the standard HSQC (A) or the sensitivity-enhanced version (B). Narrow and wide rectangles correspond to hard pulses with flip angles of 90° and 180°, respectively. Filled and open bells are 90° sinc-shaped water-selective pulses with a duration of 1.5–2.0 ms. The pulses in red and blue are only applied for the ¹J_XH-active and the ¹J_XH-inactive experiment, respectively. The open rectangles denoted CPD are low power decoupling scheme in either WALTZ-16 for ¹⁵N or GARP for ¹³C. Pulses are x-phase by default. Phase cycles are listed as follows, ϕ1 = x; ϕ2 = x, −x; ϕ3 = (x)₄, (−x)₄; ϕ4 = (x)₂, (−x)₂; ϕ5 = (x)₈, (−x)₈; ϕ_rec = (x, −x)₂, (−x, x)₂ for (A) and ϕ1 = x; ϕ2 = y, −y; ϕ3 = (x)₄, (−x)₄; ϕ4 = (x)₂, (−x)₂; ϕ5 = (y)₂, (−y)₂; ϕ_rec = (x, −x, −x, x, −x, x, x, −x) for (B). The fixed delay durations are listed as follows, τ₁ = 1/(4¹J_XH), τ₂ = 1.25 ms and τ₃ = 1/(8¹J_XH). The varied delays t_a–d during the X spin chemical shift evolution are listed in Table 1. All gradient pulses are sine-shaped and along z-axis by default except for the xyz gradient pulses G1, G2, G4 and G5 of (B). The duration and strength for gradient pulses are as follows, G1 = 1 ms, 10 G/cm; G2 = 1 ms, 5 G/cm; G3 = 0.5 ms, 10 G/cm for (A) and G1(xyz) = 5 ms, 12.5 G/cm; G2(xyz) = 1 ms, 40 G/cm; G3 = 0.5 ms, 5 G/cm; G4(xyz) = 1 ms, 20 G/cm; G5(xyz) = 1 ms, 30 G/cm for (B). Quadrature detection in (A) was achieved via States-TPPI method such that phase ϕ1 and ϕ2 of the X pulses were incremented 90° for every FID in t₁ acquisition. Quadrature detection in (B) was achieved via the Echo-Antiecho method such that the second FID for each increment of t₁ was collected with signs of gradient pulses G2 and phase ϕ5 of the 180° pulse switched simultaneously. The ¹H carrier frequency was at water resonance of 4.76 ppm and the ¹⁵N and ¹³C carrier frequencies were at 117 ppm and 57.0 ppm, respectively. The spectral width of ¹H was 12 ppm and the number of direct time domain complex points was 1024. The ¹⁵N acquisition time was 42.1 ms and the spectral width was 30 ppm. The ¹³C acquisition time was 50.7 ms and the spectral width was 14 ppm. The ¹J_XH-active and the ¹J_XH-inactive spectra were collected in interleaved manner by switching the position of the 180° ¹H pulse between the red and the blue. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 1.

The varied delays in the ¹J_XH-active/inactive semi-CT HSQC experiments.

Varied delay	In-/de-crement value	Resulting value
ta	Δta = (1/21JXH)/(N + 1)a	1/(21JXH) − Δta
tb	Δtb = Δt1/2 + Δtab	Δtb
tc	Δtc = Δt1/2	1/(21JXH) + Δtc
td	Δtd = Δt1/2	Δtd

^aN is the number of real points in the indirect t₁ dimension.

^bΔt₁ = 1/SW, where SW is the spectral width of the X (¹⁵N or ¹³C) spin.

$$\begin{gathered} {\mathrm{H}}_z{\mathrm{X}}_y\stackrel{t_a}{\to }{\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}{t}_a{cos}^n\pi J_{\text{XH}}{t}_a\stackrel{180°\mathrm{X},t_b}{\to }{\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}(t_a-t_b) \\ \times {cos}^n\pi J_{\text{XH}}(t_a-t_b)\stackrel{180°\mathrm{H},t_c}{\to }{\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}(t_a-t_b-t_c) \\ \times {cos}^n\pi J_{\text{XH}}(t_a-t_b+t_c) \\ ={\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}(\mathrm{\Delta }{t}_1){cos}^n\pi J_{\text{XH}}(1/J_{\text{XH}}) \end{gathered}$$ (2)

In parallel the ¹J_XH-inactive experiment has the ¹H 180° pulse being applied at the blue position (Fig. 1) with a new introduction of variable delay t_d, which is incremented within the t_c period (Table 1). The modification results in the net ¹J_XH evolution time of 0 (Eq. (3)), ensuring the ¹H-decoupling throughout the full t₁ period. The ¹J_XH-inactive spectrum yields the same sign for all XH_n cross-peaks. The t₁ chemical shift labeling is identical to the ¹J_XH-active spectrum.

$$\begin{gathered} {\mathrm{H}}_z{\mathrm{X}}_y\stackrel{t_a}{\to }{\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}{t}_a{cos}^n\pi J_{\text{XH}}{t}_a\stackrel{180°\mathrm{X},t_b+t_c-t_d}{\to }{\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}(t_a \\ -t_b-t_c+t_d){cos}^n\pi J_{\text{XH}}(t_a-t_b-t_c+t_d)\stackrel{180°\mathrm{H},t_d}{\to }{\mathrm{H}}_z{\mathrm{X}}_y \\ \times cos{\omega }_{\mathrm{X}}(t_a-t_b-t_c+t_d-t_d){cos}^n\pi J_{\text{XH}}(t_a-t_b+t_c+2t_d) \\ ={\mathrm{H}}_z{\mathrm{X}}_{y}cos{\omega }_{\mathrm{X}}(\mathrm{\Delta }{t}_1){cos}^n\pi J_{\text{XH}}(0) \end{gathered}$$ (3)

2.2. MS-HSQC spectra

Since both the ¹J_XH-active and the ¹J_XH-inactive experiments have the identical pulse execution time or T₂ loss, the post-acquisition sum and difference of the two at a 1:1 ratio yields the XH₂ only and the XH/XH₃ only spectrum (Figs. 2–4 and S2) without any need to scale relative intensities or further correct phases. The approach of directly adding and subtracting to obtain the desired spectra is similar to the in-phase-anti-phase (IPAP) HSQC experiments for residual dipolar coupling (RDC) measurement [24].

Fig. 2.

The sub-region of the ¹H–¹⁵N MS-HSQC spectra of RNase A at natural abundance dissolved in 95% H₂O/5% D₂O. The positive and the negative peaks are shown in black and red, respectively. The spectra of the ¹J_NH-active (A) and the ¹J_NH-inactive (B) were collected using the pulse sequence in Fig. 1A. The post-acquisition difference (C) and sum (D) spectra showed multiplicity separation between peaks of NH and NH₂, e.g., peaks a and b within the dashed box. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 4.

The ¹H–¹³C MS-HSQC spectra for moieties of CH/CH₃ of RNase A at natural abundance dissolved in 95% H₂O/5% D₂O. The positive and the negative peaks are shown in black and red, respectively. The two spectra were collected with identical experimental parameters except for the use of the xyz-axis gradient pulses (A) and the z-axis only ones (B). The red dashed-circles indicate CH₂ cross-peaks that were not completely removed due to ¹J_CH mis-match. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

For experiments on proteins at natural abundance the current multiplicity-separation is an efficient and practical choice because chemical shift assignments from MS-HSQC spectra are more interpretable than those from multiplicity-edited spectra and an improved S/N is obtained relative to the standard multiplicity edited HSQC sequence. We tested both ¹H–¹⁵N (Figs. 1A and 2) and ¹H–¹³C (Figs. 1B, 4A and S2) MS-HSQC experiments on an RNase A sample at natural abundance dissolved in 95% H₂O/5% D₂O. For example, the NH peak a of RNase A, would commonly be treated as a minor impurity peak in the ¹J_NH-active (edited) spectrum (Fig. 2A) or as a tail of peak b in the ¹J_NH-inactive (regular) spectrum (Fig. 2B). Only under the separation method we can confidently identify peak a as an NH peak (Fig. 2C) and peak b as an NH₂ peak with correct intensity (Fig. 2D).

The ¹H–¹⁵N MS-HSQC experiments were also demonstrated on a larger ¹⁵N-labeled 26-kDa retroviral capsid protein (Fig. 3). The capsid protein shape is highly asymmetric, with an anisotropy of 1.8. Indeed, its averaged ¹⁵N T₂ relaxation rate and Stokes radius correspond to that of a 32 kDa symmetric protein [25]. Table 2 shows sensitivity comparisons of the standard HSQC, the standard multiplicity-edited HSQC and the new semi-CT HSQC (Table 2). Due to the reduction in the time that ¹⁵N magnetization spends in the transverse plane, the MS-HSQC yielded an average of 10% increase in peak S/N from the standard edited HSQC (Table 2 and Fig. S1). The MS-HSQC is at 85% of the S/N of the standard HSQC spectra, indicating a protein sample should be roughly 15% more concentrated to obtain S/N similar to a standard HSQC, if an MS-HSQC spectrum is desired. Based on the comparison we expect the MS-HSQC is generally applicable to a similar molecular mass range as the standard HSQC.

Fig. 3.

The ¹H–¹⁵N MS-HSQC spectra of a ¹⁵N-labeled 26 kDa retroviral capsid protein dissolved in 95% H₂O/5% D₂O. The post-acquisition difference (A) and sum (B) spectra showed multiplicity separation between peaks of NH and NH₂.

Table 2.

Feature comparison among the three ¹H–¹⁵N HSQC spectra collected on a 26 kDa ¹⁵N-labeled protein.

Type	Pulse program	Cross-peak sign			Averaged S/Na	Experimental timeb
		NH	NH2	Separation
Standard	Bruker hsqcfpf3gpphwg	+	+	No	32.4	3 h 18 min
Standard multiplicity-edited	hsqcfpf3gpphwg with an extra 1/1JNH period after t1	+	−	No	24.6	3 h 21 min
Semi-CT	The 1JNH-active version in Fig. 1A	+	−	Yes	27.1	3 h 12 min

^aA total of 137 isolated cross-peaks were picked on all spectra and peak S/N were calculated using Sparky (T.D. Goddard and D.G. Kneller, University of California, San Francisco). The averaged S/N are shown.

^bThe sample is a 0.1 mM ¹⁵N-labeled retroviral capsid protein (26 kDa). All HSQC spectra were collected on a Bruker 500 MHz spectrometer equipped with a room-temperature probe.

Natural abundance ¹H–¹³C MS-HSQC experiments were performed on the RNase A sample. The ¹J_CH-active and the ¹J_CH-inactive ¹H–¹³C HSQC spectra were collected with CH₂/CH₃ peaks folded (Figs. S2A and S2B). The difference and sum yielded the CH/CH₃ only (Figs. S2C and 4A) and the CH₂ only spectra (Fig. S2D). Importantly, the ¹J_CH values vary among CH_n moieties, i.e., the ¹J_CH values of the sp² ring CH are 220–155 Hz, the ¹J_CH of the sp³ groups are 140–150 Hz for CH, 135–145 Hz for CH₂ and 125–135 Hz for CH₃, and are directly correlated to their respective ¹³C chemical shifts [26]. Since all amino acid aromatic groups have CH only, a standard HSQC can be collected for them without any need of multiplicity separation. For aliphatic CH_n moieties, we have used the value of 125 Hz to match CH₃ peaks. Notably, the CH₂ only spectra contain undetectable CH/CH₃ peaks at a S/N threshold of 7.8 (Fig. S2D). However, there are still 3 CH₂ clusters/peaks with attenuated intensity in CH/CH₃ sub-spectra (red-dashed circles in Figs. 4 and S2C). In practice the ¹J_CH value might be set higher to 135 Hz, closer to the CH₂ ¹J_CH values to minimize the imperfect cancelation. Another alternative approach demonstrated by Heikkinen et al. is to add several MS-HSQC spectra with varied ¹J_CH values, because the sum of these spectra are less sensitive to ¹J_CH variations [27].

2.3. Water suppression

The MS-HSQC methods developed here (Fig. 1) also minimize sample manipulation steps, (e.g., freeze-drying a biomolecule sample and re-dissolving in D₂O for ¹H–¹³C HSQC data) because all NMR experiments can be performed in 95% H₂O/5% D₂O. Water suppression was demonstrated on two general versions of HSQC, the standard HSQC [28] (Fig. 1A) and the sensitivity-enhanced version [29,30] (Fig. 1B). For the standard HSQC experiment, we used low-power shaped pulses to achieve a more complete water flip-back scenario, where water magnetization has been kept along +z throughout the pulse sequence except for the first INEPT transfer, where water magnetization was along the y-axis. The complete +z-flip-back is more robust than the typical flip-back pulse used in vendor libraries. Specifically, in typical flip-back experiments, the water magnetization is aligned along the –z-axis for a period of t₁/2, which increases the likelihood of radiation damping, especially at longer t₁ acquisition times.

Here, for the sensitivity-enhanced HSQC (Fig. 1B) water suppression was achieved by using xyz-axis gradients after the first INEPT and during the coherence selection without using any shaped pulses. This allows the observation of C_α–H_α correlation in the ¹H–¹³C MS-HSQC spectrum (Fig. 4A), which can be obscured by poor water suppression. The xyz-axis gradient is essential since the same HSQC experiment, when performed without using xy-gradient but otherwise identical experimental parameters, yielded poor water suppression spectrum which precluded the observation of C_α–H_α correlations (Fig. 4B).

3. Conclusion

Here, the application of a new MS-HSQC experiment on biomolecules at natural abundance dissolved in 95% H₂O/5% D₂O was demonstrated. The new experiment generates separated spectra containing XH₂ peaks only or XH/XH₃ peaks only with higher sensitivity than the multiplicity-edited HSQC [14–16]. The separation feature is not an issue for ¹⁵N/¹³C-labeled proteins, e.g., a 3D HNCO spectrum can be used to separate the side-chain NH₂ peaks and reveal any backbone NH peaks that may be hidden underneath them. However, for proteins at natural abundance, the simplified 2D MS-HSQC can improve the accuracy of comparisons between molecules made by chemometric analysis of specific chemical structures (e.g., backbone NH vs. side-chain NH₂) of biomolecules in their native state [31]. This is crucial when profiling protein therapeutics for backbone structure as the side-chain NH₂ peaks add no information to such an evaluation and are more susceptible to formulation differences. Thus, the separated HSQC spectra with good water suppression should be amenable to NMR analysis of biotherapeutics in their formulation buffers.

The experiment described here will also work on isotope-labeled molecules for collecting ¹H–¹⁵N HSQC spectra (with the decoupling of C_α/C′ if ¹³C-enriched). Similar to protein spectra in natural abundance, this approach can resolve the ambiguities in chemical shift assignments and serve to improve resolution in 2D spectra, which in turn will guide 3D assignments with higher confidence. Finally, the MS-HSQC can reveal more interacting NH peaks in titration studies if such peaks happen to be buried by the NH₂ resonances, and when combined with spin relaxation HSQC, allow a more complete set of measurements on the structure and dynamics of a biomolecule.

4. Materials and methods

4.1. Protein sample preparation

The RNase A powder at natural abundance (GE Healthcare Bio- Sciences, Pittsburgh, PA) was dissolved in 2− diluted PBS buffer (Quality Biological, Gaithersburg, MD) to the desired concentration of 1.7 mM. The ¹⁵N-labeled retroviral capsid protein, 0.1 mM in PBS buffer, was prepared as described [32]. For both samples, 570 μl protein solution was mixed with 30 μl D₂O (Cambridge Isotope Laboratories, Tewksbury, MA) then 550 μl of these solutions were loaded into 5-mm Wilmad 535-pp tubes (Wilmad-LabGlass, Vineland, NJ).

4.2. NMR spectroscopy

All NMR experiments were performed at 25 °C. The experimental temperature was calibrated using pure methanol. The pulse sequence details and acquisition parameters are in Fig. 1 caption and Table 1. We include the pulse sequence codes (Supplementary material) of Fig. 1A and B using the templates of the Bruker pulse programs hsqcfpf3gpphwg and hsqcetgpsi, respectively. For the ¹H–¹⁵N HSQC experiment, which was performed on a Bruker Avance III, 500 MHz spectrometer, equipped with a room temperature QXI probe and a z-axis gradient, the total experimental time is 63 h using the pulse sequence in Fig. 1A. For the ¹H–¹³C HSQC experiment, which was performed on a Bruker Avance II 700 MHz spectrometer, equipped with a room temperature QXI probe and a xyz-axis gradient, the total experimental time is 21 h using the pulse sequence in Fig. 1B.

4.3. Data processing

The interleaved raw FID file was separated then added and subtracted using NMRPipe [33] scripts (Supplementary material). The resulting time domain FID were polynomial filtered, apodized with a cosine function and zero-filled to 4 k data points before Fourier transform along both dimensions. The both frequency domain were baseline corrected with a polynomial function.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jmr.2014.11.011.