|Home | About | Journals | Submit | Contact Us | Français|
High-sensitivity proton-detected experiments in solid-state NMR have been recently demonstrated in proton-diluted proteins as well as fully protonated samples under fast magic-angle spinning. One key element for performing successful proton detection is effective solvent suppression achieved by pulsed field gradients (PFG) and/or saturation pulses. Here we report a high-performance solvent suppression method that attenuates multiple solvent signals simultaneously by more than a factor of 10,000, achieved by an optimized combination of homospoil gradients and supercycled saturation pulses. This method, which we call Multiple Intense Solvent Suppression Intended for Sensitive Spectroscopic Investigation of Protonated Proteins, Instantly (MISSISSIPPI), can be applied without a PFG probe. It opens up new opportunities for two-dimensional heteronuclear correlation spectroscopy of hydrated proteins at natural abundance as well as high-sensitivity and multi-dimensional experimental investigation of protein-solvent interactions.
Magic-angle spinning solid-state NMR (MAS SSNMR) has been demonstrated as a powerful tool for protein structure determination and the investigation of protein-solvent interactions [1–7]. However, the majority of experiments so far have been performed by detecting on either 13C or 15N nuclei, which are relatively insensitive due to their low gyromagnetic ratios. High-sensitivity detection on 1H has only been realized recently in SSNMR. The strong proton dipolar coupling problem has been addressed with two approaches: (1) diluting the proton bath, (2) spinning fully protonated samples at ~40 kHz. The first approach yields very high resolution [8–11], and 1H-1H distance restraints as long as 9 Å have been detected, enabling the determination of a high-resolution protein structure . The second approach is important for systems such as membrane proteins where deuteration and back-exchange may be challenging; it also permits all non-exchangeable protons to be investigated in a single experiment . Both approaches result in a common outcome of greatly enhanced sensitivity relative to 13C or 15N detection.
To utilize this enhanced detection sensitivity for analyses of natural abundance proteins is one important application of these developments. Our initial attempts to perform fast MAS proton-detection heteronuclear-correlation (HETCOR) experiments on natural abundance proteins, however, were unsuccessful due to the overwhelming solvent signal intensity. In hydrated protein samples containing roughly 1 μmol of protein and 1 μL of water, the ratio of water to protein signal is 100, and the protein signal will be at least ~100 (or 270) times weaker when performing correlation experiments with 13C (or 15N) at natural abundance. Therefore the water resonance is more than 10,000 times stronger than a 13C or 15N-edited protein proton resonance; thus water and other solvent signals must be suppressed by four orders of magnitude to utilize the full receiver dynamic range for the protein signals. Previously reported solvent suppression schemes yield suppression by two to three orders of magnitude for a single resonance line (such as water), achieved by employing either pulsed field gradients (PFGs)  or saturation pulses . We had extended the method of Paulson and Zilm  to achieve simultaneous suppression of multiple solvent signals, by using a long, moderate-power saturation pulse . Here we report a new high-performance suppression method, which we call Multiple Intense Solvent Suppression Intended for Sensitive Spectroscopic Investigation of Protonated Proteins, Instantly (MISSISSIPPI). This approach combines both gradient and saturation pulses. Since pulsed field gradients are generally not available for all types of solid-state NMR probes, we substitute with homospoil gradients, in which case the Z1 shimming current is maximized for a short duration to apply a small gradient (a few Gauss per centimeter). MISSISSIPPI employs carefully ordered homospoil gradients and saturation pulses with supercycled phases, resulting in attenuation of multiple solvent signals simultaneously by at least a factor of 10,000, opening up new opportunities for spectroscopy of natural abundance proteins and investigation of protein-solvent interactions.
Natural abundance and 2H,13C,15N-labeled β1 immunoglobulin binding domain of protein G (GB1) were prepared according to previously published procedures [12, 14]. Nano-crystalline GB1 samples were precipitated with natural abundance 2-methyl-pentane-2,4-diol (MPD) and isopropanol (IPA) . About 5 mg (0.9 μmol) GB1 and 4 mg solvents were packed into the 1.6-mm MAS rotor, with a rubber disc on each end to maintain hydration .
The SSNMR experiments were performed on a 750 MHz Varian INOVA spectrometer with a FastMASTM probe (Varian, Inc., Palo Alto, CA and Fort Collins, CO). The 1H, 13C, and 15N π/2 pulse widths were 1.6, 2.3, and 3.3 μs with 111, 213, and 295 W input power levels, respectively. Shimming was performed on the water signal of a 50 mM phosphorus buffer to obtain 22 Hz (0.03 ppm) 1H line width. Temperature was regulated with variable temperature gas (dry air) at a set point of -5 °C with 75 scfh flow. The sample temperature was calibrated using ethylene glycol  and determined to be 7.2 and 10.0 °C for MAS rates (νR) of 36 kHz and 39 kHz respectively. During cross polarization (CP) [16, 17], the 1H and 15N nutation frequencies are 2.5 and 1.5 times the MAS rate, respecively. For n.a. GB1, contact times of 1.6 and 0.6 ms were used for the 1H to 15N and 15N to 1H CP steps, respectively; for 2H,13C,15N-GB1 the corresponding contact times were 1.4 and 0.3 ms. Low power XiX decoupling (nutation frequency ~10 kHz, pulse width 5.85/νR) was applied during 15N evolution . WALTZ-16 decoupling (6.4 kHz) was applied on 15N channel during proton acquisition [10, 19]. Recycle delay was 1.5 s unless specified differently. Other experimental details can be found in section 3.1 and figure captions.
Fig. 1 shows the MISSISSIPPI scheme in the context of a 15N-1H HETCOR experiment. In the beginning of the recycle delay, a homospoil pulse is turned on for τH to suppress transverse signals left from the previous scan. After the recycle delay, a π/2 pulse excites the 1H magnetization, which is then transferred to 15N by CP. After t1 evolution, the magnetization is restored to the longitudinal axis by a 15N π/2 pulse, followed by a delay t1max–t1 to maintain a constant duty cycle. A second homospoil pulse is then applied to dephase solvent signals; the parasitic eddy current (caused by the switching of the shimming current) is allowed to decay during τw. Proton saturation pulses are then applied for S for further suppression. The magnetization is flipped to the transverse plane by a 15N π/2 pulse and transferred to 1H via a second CP for data acquisition. Optionally spin-diffusion or radio frequency-driven dipolar recoupling (RFDR) can be applied before acquisition in order to establish correlations among protein signals or between protein and solvent signals .
Each of the homospoil pulses was applied for τH = 20 ms, the maximum allowed by the INOVA system. The delay τw > 40 ms is needed for good suppression and 50 ms was used for data presented subsequently in this report. The RF saturation pulses should exceed a nutation frequency of 20 kHz. For best suppression performance here, we found it necessary to place these saturation pulses as late in the pulse sequence as possible. In our earlier implementation, the saturation pulse precedes any evolution periods to avoid potentially unwanted 15N-15N or 13C-13 C recoupling . This is not a concern with natural abundance samples, and even with 15N-labeled samples, the nutation frequency can be set to the n=1 DARR condition (νR)  without resulting in significant 15N-15N mixing, since this process is extremely slow at such high MAS rates; we confirmed this expectation with 2D 15N-1H spectra of 2H,13C,15N-GB1 (data not shown).For the 13C-1H correlation experiments on uniformly 13C-labeled samples, it is best to avoid the DARR  conditions; setting the presaturation pulses to 0.75νR for 250 ms resulted in only modest recoupling of a very small number of signals (Thr Cβ-Cγ and Lys Cβ–Cγ and Cγ–Cδ), as determined on a uniformly-13C,15N sample of GB1 (data not shown).
Saturation pulses can be applied for a total duration of 100 to 300 ms; 200 ms was used for subsequent data. An even number of saturation pulses with alternating x and y phases were found to perform far better than a single pulse with the same total duration; this is an extension of the method of two pulses (400 μs each with orthogonal phases) used by Ishii to filter out protons attached to 12C . Two and four pulses were found appropriate for 100 and 200 ms, respectively; both four and six pulses had similar performance for τS = 300 ms. As we reported earlier , the relatively long delay (τH+τW+τS) is short compared to the T1 relaxation times of 15N and therefore causes very little decay. For 13C, T1 values may in general be shorter (especially for methyl groups). Although we still found only a ~10% loss over this duration, this issue should be considered when storing magnetization on 13C sites. The exact proton transmitter frequency is not critical for MISSISSIPPI; the typical placement in our studies was 3 ppm.
Fig. 2 demonstrates the MISSISSIPPI results on a natural abundance GB1 sample. The solvent signals from water, MPD and IPA dominate the single pulse excitation spectrum (Fig. 2a). The weak amide proton signals are greatly attenuated in the 15N-edited spectra (Fig. 2b–f) due to the low 15N natural abundance (0.368%). For two transients (Fig. 2b), the water signal is well suppressed (by 7,800 times) but the methyl signals (suppressed by 2,200) are still stronger than the amide signals. The suppression efficiencies improve significantly upon signal averaging with multiple scans. Water and methyl signals are suppressed by 10,300 and 3,600 times for 4 transients, respectively; 12,000 and 6,900 for 8 transients; 18,800 and 13,000 for 16 transients (Fig. 2c); and 22,500 and 13,500 for 32 transients. The suppression factors are accurate when the residual solvent signals are much stronger than the noise level. In cases where the solvent signals are close to noise level, the strongest intensity in the expected solvent resonance region is used as the residual solvent intensity, leading to an underestimate of the suppression efficiency. While the values are not very precise in these cases, the large differences observed between different numbers of transients are reproducible and reliable. Both suppression efficiencies and signal-to-noise ratios can be further increased by truncating the time domain signal to 4T2 (T2 = 0.85 ms for amide proton spin-spin relaxation time), which rejects the late portion of the free induction decay that contains little amide but noise and solvent signals (Fig. 2d,e). For comparison, even after 256 transients of signal averaging, our previous suppression method employing a single saturation pulse (250 ms, 40 kHz) , which was adequate for enriched samples, resulted in strong residual solvent signals that severely interfere with amide signals of natural abundance GB1. For enriched samples, residual solvent signals after MISSISSIPPI suppression can be completely ignored with the protein signals 273 times stronger than in natural abundance samples (Fig. 2g; identical to the first spectrum of Fig. 3a).
The two-step phase cycle allows residual solvent signals to be subtracted between two scans and meanwhile the desired signals to be added; increasing phase steps has not been effective. Amplifier and probe instabilities, phase errors, and quadrature gain and phase imbalance of the receiver account for the remaining solvent intensities in two scans. The observed improvement with increasing signal averaging is likely due to the apparent incoherence of the remaining solvent signals after every two scans. We attribute the better suppression for water than for solvent methyl protons to the relative T2 values (4.7 and 13.7 ms for water and methyl, respectively).
Water molecules are important for protein structure and function . Also protein samples used in X-ray diffraction or SSNMR are often crystallized with the assistance of other molecules (e.g. polyethylene glycol, MPD, IPA); determining the location of these molecules aids in understanding their role in assisting crystal growth and modulating side-chain structure. Several SSNMR studies on protein-water interactions have been performed nevertheless with only 13C or 15N detection [4, 6, 7]; proton detection was not used because of the deleterious residual bulk solvent signals. Next we demonstrate that with MISSISSIPPI, it is possible to detect only solvent signals that are transferred from the protein.
Protein and solvent interactions were examined with 2H,13C,15N-GB1 (back-exchanged with H2O), utilizing 1H-1H mixing in the pulse sequence (Fig. 1). Spectra with mixing times from 0 to 100 ms in 1-ms steps were acquired for spin-diffusion (Fig. 3a) and RFDR (Fig. 3b). With increasing mixing time (τM = 0, 6, 12, 20, 40 ms; Fig. 3), the solvent signals monotonically increase for spin diffusion (Fig. 3a) but maximize around 12 ms for RFDR (Fig. 3b). The maximal achievable solvent intensities are much larger in the spin-diffusion than RFDR experiments. It is also obvious that the amide signals decay much faster in the RFDR experiments. Control experiments that have identical parameters in the 1H channel but with 15N channel turned off are also shown in Fig. 3. The control spectra prove that the recovery of bulk solvent signal during the mixing period can be ignored and that the observed solvent intensities in the normal spectra are transferred from the protein.
The decay of amide signal and rise of solvent signals are shown in Fig. 4. These trajectories were fitted with exponential models and the time constants are listed in Table 1. The amide signals decay much slower for spin-diffusion (decay time constant τd = 63.7 ms) than for RFDR (τd = 7.8 ms). RFDR effectively recouples the dipolar interaction among protons, especially those within the relatively immobile protein molecule. The total signal (integration from 0.9 to 11 ppm) decays extremely slowly for spin-diffusion, with τd = 0.54 s agreeing with amide proton T1 (measured by adding an inversion pulse to the sequence shown in Fig. 1). For RFDR, the total signal decay time constant is only 13.0 ms; the recoupled dipolar interactions expedite the longitudinal relaxation.
The solvent trajectories under spin-diffusion are well fit with a single exponential growth model, with time constant τg = 26.1, 30.6, and 33.8 ms for water, alcohol groups, and methyl groups, respectively. The decay appears to be dictated by T1 relaxation, which is at least 10 times slower and not necessary to be included in the fitting model. On the other hand, the trajectories under RFDR mixing require both exponential growth and exponential decay fit parameters. The growth time constants are τd =7.0, 0.003, and 2.7 ms and the decay time constant τd = 33.7, 25.6, and 34.0 ms for water, alcohol, and methyl, respectively. We attribute the extremely fast initial growth of the alcohol signals to the overlapping Thr hydroxyl protons.
Maximal achievable solvent intensities are significantly higher for spin-diffusion than RFDR from Fig. 4. This can also be seen by comparing the pre-exponential fitting parameter, which corresponds to the saturating intensity in the absence of relaxation. The ratios of pre-exponential parameters are 2.4, 2.8, and 8.7 for water, alcohol, and methyl, respectively.
We have demonstrated that MISSISSIPPI can achieve very efficient multiple solvent suppression in SSNMR by a combination of homospoil gradients and RF saturation pulses. MISSISSIPPI enables the sensitivity benefit of proton detection to be translated to new, challenging applications. The dynamic range afforded by MISSISSIPPI has allowed 1D spectra to be acquired on natural abundance samples and examination of solvent signals correlated to the protein. We envision that MISSISSIPPI will become a widespread building block in future proton-detection SSNMR experiments, including straightforward extensions to additional dimensions for a variety of applications.
This work is supported by the National Institutes of Health R01GM75937 (Roadmap Initiative and NIGMS). The authors appreciate technical advice from Charles Mullen, Mircea Cormos, John Stringer, and Dennis Sandoz (Varian, Inc.), as well as sample preparation assistance from Gautam Shah.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.