|Home | About | Journals | Submit | Contact Us | Français|
We present a time-shared 3D HSQC-NOESY experiment that enables one to simultaneously record 13C- and 15N-dispersed spectra in Ile, Leu and Val (ILV) methyl-labeled samples. This experiment is designed to delineate the two spectra which would otherwise overlap with one another when acquired together. These spectra display nOe correlations in the detected proton dimension, i.e. with maximum resolution. This is in contrast to NOESY-HSQC types of experiments that provide cross-peaks in the indirect dimension with low resolution due to limits in experimental time. The technique is particularly advantageous at high field where even longer experimental times would be required for comparable resolution in NOESY-HSQC experiments. The method is demonstrated at 900 MHz and at 750 MHz on 37 kDa and 31 kDa proteins, respectively. The resolution and time saving provided in this experiment was crucial for solving the structures of these two proteins.
Nuclear magnetic resonance is routinely used nowadays to determine structures of biomolecules. Key to this success is the ability to determine distance constraints from nuclear Overhauser effects measured in NOESY experiments. In large proteins, or more generally, for molecules with crowded NMR spectra, selective labelling of specific nuclei is routinely performed to simplify the spectra and minimize relaxation losses. The most popular strategy relies on using a sample which is selectively protonated and 13C-enriched at the methyls of Leu, Val and Ile (at the δ position) in an otherwise deuterated, 12C background. The molecule is typically uniformly enriched in 15N so that, when in H2O, nOes involving amide protons or methyl protons can be measured with the same sample. This sparse labelling approach eliminates many ambiguities in nOe cross-peak assignments but obviously reduces the number of observable nOes drastically. Therefore it becomes critical to resolve and unambiguously assign a maximum number of the observed NOE cross-peaks. This can be achieved by recording several multidimensional experiments which disperse the nOe cross-peaks of 2D spectra along a third dimension, provided by correlated 13C or 15N nuclei in 3D experiments. Further dispersion of nOe cross peaks can be achieved by including an additional, fourth heteronuclear dimension in 4D experiments. This increase in the number of experiments is accompanied by an increase in precious spectrometer acquisition time. We recently showed how the two 15N- and 13C-edited 3D spectra can be recorded in a single TS-NOESY-PEP-HSQC/TROSY experiment, while a second experiment allows for the four possible 4D spectra to be acquired simultaneously (Frueh et al., 2006). This was achieved by implementing time-shared transfers and evolutions (Sørensen, 1990; Farmer II, 1991; Boelens et al., 1994; Pascal et al., 1994; Jerala and Rule, 1995; Sattler et al., 1995), a method which has recently found many new applications in biomolecular NMR (Uhrín et al., 2000; Xia and Zhu, 2001; Kupce et al., 2003; Frueh et al., 2005; Frueh et al., 2006; Perez-Trujillo et al., 2007; Wurtz et al., 2007; Guo and Tugarinov, 2009). While the 4D experiment leads to unambiguous assignment of nOe cross-peaks and the 3D experiment enables their integration, the latter suffers from limited resolution in the dimension displaying the nOe correlations and part of the improvement obtained from the technique is deteriorated during peak integration due to signal overlap. This limitation results from using NOESY-HSQC based experiments, where the nOe cross-peaks appear in an indirect dimension, which can only achieve a relatively low resolution for typical acquisition times. To overcome this shortcoming, we developed a variant of the 3D time-shared experiment based on the HSQC-NOESY experiment which provides the desired dipolar correlations in the detected dimension to maximize the resolution of nOe cross-peaks. The higher resolution achieved makes up for any reduced sensitivity from the inability to use TROSY efficiently (and an increased length of the pulse sequence for a water suppression scheme) because more accurate volume integrations become possible. Because nOe cross-peaks appear in the detected dimension, the cross-peaks of one pathway (e.g. 15N dispersed) overlap with those of the other pathway (e.g. 13C dispersed), as well as with the strong diagonal peaks. The TS-HSQC-NOESY experiment has thus been designed to delineate the 15N- and 13C-edited spectra that would otherwise overlap. The method is demonstrated on the 37 kDa T-TE di-domain of the enterobactin EntF non-ribosomal peptide synthetase (NRPS) and the 31 kDa regulatory subunit of γ-glutamyl-cysteine synthetase (γGCS). The high resolution provided by this experiment was absolutely crucial for solving these structures (Frueh et al., 2008) and (Leed et al., In preparation).
For both proteins, an expression plasmid (pET30a+ for the point mutant S48A-TTE and pBluescript SK- for γGCS) containing the full length cDNA for the N-terminally His6-tagged protein was transformed into E. coli BL21(DE3) cells. A 15N 2H uniformly labeled sample, with protonated Phe and Tyr residues and with selective protonation and 13C labelling of the methyl carbons of Ile (γ position only), Leu and Val side-chains, was prepared by over-expression in M9 minimal medium in D2O containing 2H glucose and 15NH4Cl. Protons also occur at the β position of Val and at the γ positions of Ile and Leu (except for the γGCS sample used to record the TS-HSQC-NOESY). The cells were initially allowed to grow at 37 °C until one hour before induction (O.D. ~0.4). At this point, 2-oxo-3-(13C-methyl)-4-13C-butanoate (13C -dimethyl-α-ketoisovalerate), 2-oxo-4-13C-butanoate (132 C-methyl-α-ketobutyrate), 15N-phenylalanine and 15N-tyrosine were added to the growth medium. When the optical density reached ~0.6, the bacteria were induced with 1mM IPTG and the medium was cooled to 25 °C and allowed to grow overnight. Recombinant protein samples were purified using Ni-NTA resin (Qiagen) followed by FPLC using a Sephadex gel-filtration column (S75 for S48A-TTE and S200 for γGCS). The S48A-TTE sample was concentrated to a final concentration of 300 μM in 20 mM phosphate (pH 6.7), 150 mM NaCl, 1 mM EDTA and 1 mM DTT in 95% H2O 5% D2O. γGCS was concentrated to 500μM (sample used for the time-shared NOESY-HSQC/TROSY) or 300 μM (sample used for the time-shared HSQC-NOESY). The γGCS NMR buffer contained 50 mM Tris-d11 (pH 7.25), 100 mM NaCl, 5mM MgCl2, and 10 mM DTT-d10. The γGCS sample used for the TS-HSQC-NOESY was prepared with α-keto-acid precursors deuterated at the alpha position of the ketone.
The S48A-TTE data were recorded at 25 °C on a 900 MHz Bruker spectrometer equipped with a cryoprobe®. A mixing time of 200 ms was used in order to detect long-range nOes. The spectral widths were 16.023 ppm for proton (ω3, centred at 4.690 ppm), 35 ppm for nitrogen (ω'1, centred at 118 ppm), 22 ppm for carbon (ω1, centred at 17 ppm) and 5 ppm for the indirect proton dimension (ω2, centred at 8.5 ppm). A data matrix of 1024 × 78 × 120 complex points was acquired for these nuclei, respectively. A recycling delay of 1 s was used, and for each value of t1, t'1, t2, t'2, two fids, each with 8 scans accumulated, were recorded in an interleaved manner with a different phase combination (see below). The total measurement time was 4 days and 17 hours. The spectrum was linearly predicted and zero-filled to a final size of 2048 (ω3) × 156 (ω1/ ω'1) × 240 (ω2) points.
The γGCS spectra for both the TS-NOESY-PEP-HSQC/TROSY and TS-HSQC-NOESY were recorded on a 750 MHz Bruker spectrometer equipped with a cryoprobe®. For each measurement, 1024 (ω3, 1H) × 50 (ω2,13C and 15N) × 152 (ω,1, 1H indirect) complex points were acquired. The TS-NOESYPEP-HSQC/TROSY was acquired with 16 scans and the TS-HSQC-NOESY was acquired with 2×8 scans recorded in an interleaved fashion for two different value of the phase ϕ2. Each experiment lasted 5 days and 15 hours. The spectral widths were 16.023 ppm for proton (ω3, centred at 4.690 ppm), 35 ppm for nitrogen (ω'1, centred at 119.5 ppm) and 20 ppm for carbon (ω1, centred at 18.5 ppm). For the TS-HSQC-NOESY, 5 ppm were used for the indirect proton dimension (ω2, centred at 8.5 ppm), while 13 ppm were used for the TS-NOESY-PEP-HSQC/TROSY. The mixing time was 200 ms in both experiments.
For the two TS-HSQC-NOESY measurements, the two sub-matrices obtained with different phases were extracted from the raw data. The sum of the matrices led to spectra featuring nOes involving amide protons, while their difference led to spectra containing nOes involving methyl protons. The amide and methyl regions of the detected dimension were separated, and the resulting spectra were referenced according to the appropriate heteronuclei. All spectra were processed with the nmrPipe suite(Delaglio et al., 1995) and analysed with CARA(Keller, 2004). All dimensions were zero-filled once and indirect dimensions were linearly predicted.
In short, concatenated INEPT modules allow simultaneous transfers of polarization from amide and methyl protons to nitrogens and carbons, respectively (Figure 1). A joint evolution of 13C and 15N single-quantum coherences (SQC) is tailored to provide optimal resolution in each dimension and minimize relaxation losses (see toggling frame in Figure 1)(Boelens et al., 1994). Nitrogen SQC evolves during t'1 + t1, with time increments Δt'1 = 1/SW(N) − 1/SW(C) and Δt1 = 1/SW(C), while carbon SQC only evolves during t1 with the increment Δt1 = 1/SW(C). A time-shared reverse INEPT is followed by a proton evolution period that encodes for the signals of the source protons. A mixing time is followed by a WATERGATE (Piotto et al., 1992) scheme. The water suppression can be done in this manner since amide and methyl protons resonances are very distinct from the water resonance. The magnetization flow can be summarized by:
where 2 = +1 or −1, depending on the value of ϕ2, and where H denotes all protons dipolar-coupled to the selected amide or methyl protons. For the samples in this study, these protons are the amide HN, the δ-Ile, δ-Leu and γ-Val methyl HMe, the Val Hβ, the Ile and Leu Hγ, all Phe and all Tyr protons. Thus, two 3D spectra featuring either the nOe correlations HMe →, H or HN →, H are obtained simultaneously, where the arrow indicates the transfer of magnetization occurring during the mixing time. The nOe signals appear along the detected dimension and thus benefit from maximum resolution at no cost in experimental time. However, the cross-peaks of one pathway overlap with those of the other pathway in the detected dimension, so that the experiment cannot be used as such (Figure 2a and b). Signals that are dispersed in 13C, i.e. from the HMe →, H pathway, appear in the same region as those that are 15N-edited. This impedes the assignment of nOe signals and renders the experiment impractical. To enable delineation of the two spectra, interleaved experiments are recorded in which the phase ϕ2, which selectively affects one pathway, is alternated. Addition and subtraction of the two datasets then provide the individual spectra of the 13C-HSQC-NOESY and 15N-HSQC-NOESY experiments (see below).
For larger proteins and/or for higher resolutions in the indirect dimensions, TROSY versions of the HN →, H pathway may need to be used. Our previous NOESY-HSQC type experiment was indeed designed to allow for the exploitation of TROSY. Improved versions of the H-N/C time-shared HNTROSY/HC-HSQC building block, as designed by Permi and co-workers (Wurtz et al., 2007) or Tugarinov and co-workers (Guo and Tugarinov, 2009), may be used in lieu of the HSQC block (between a and e in Figure 1). However, these need to be modified to allow for sensitivity improvement when 15N encoding takes place before the mixing time (i.e. in order to preserve N- and P- pathways) as previously developed(Brutscher et al., 1998; Meissner and Sorensen, 1999; Zhu et al., 1999). The experiment then benefits from further sensitivity enhancement in the HN →, H pathway, when compared to NOESY-TROSY, since the native 15N polarization contributes to the detected signal. When designing the time-shared experiment, care must be taken in order to maintain optimized transfers for both pathways, in particular when recording the four FIDs that are necessary for sensitivity enhanced TROSY HN →, N pathway. One drawback is that the water suppression can not be concatenated with the polarization transfers, unlike in NOESY-HSQC derived experiments. This, together with the use of proton shaped-pulses to independently manipulate amide and methyl protons, may deteriorate the advantages of the experiment. Such an experiment may prove useful, however, when used in conjunction with non-uniform sampling (NUS)(Barna et al., 1987; Schmieder et al., 1993; Orekhov et al., 2001), where long values of the evolution times t1 and t2 can be recorded, providing optimal resolution in all dimensions. Recent studies, carried out in parallel to the work presented here, indicate that NUS may be applied reliably to NOESY experiments of large proteins(Hyberts et al., 2009).
The method described above is illustrated below for two different systems: a 37 kDa segment of a non-ribosomal peptide synthetase and a 31 kDa regulatory subunit of an enzyme involved in the synthesis of glutathione.
The amide and methyl spectra first need to be extracted from the data matrix by recombining the two sub-spectra. Figures 2a and 2b show the H/CN projections of the 3D spectra recorded in an interleaved manner. The sum of these spectra leads to the 15N-dispersed HSQC-NOESY spectrum (Figure 2c), while their difference results in the 13C-dispersed spectrum (Figure 2d). Each dataset is then independently referenced according to the appropriate nuclei and the indirect methyl proton dimension is circular shifted to compensate for spectral aliasing. The H/HN and H/HC projections are shown in Figures 2e and 2f, respectively. The resulting 13C- and 15N-dispersed spectra are indistinguishable from those obtained with the individual experiments.
The motivation to develop this method stemmed from difficulties in solving the structure of the 37 kDa T-TE di-domain of the EntF component of the E. coli enterobactin synthetase (Frueh et al., 2008). Initially, distance constraints were identified on a sample of the apo-form of this protein by using the time-shared 3D and 4D experiment described in Frueh et al (Frueh et al., 2006). Indeed one of two proton dimensions of the 4D spectra is in the detected dimension and thus benefits from the ultra-high resolution as in the experiment proposed here. However, the 4D experiment has significantly lower sensitivity than the 3D experiment. Moreover, the three indirect dimensions are constrained to much lower resolutions than in the 3D spectra to maintain a reasonable acquisition time. Thus, while the 4D is immensely beneficial for cross-peak assignment, it is in general not suitable for volume integration which is hence performed in the 3D spectra. Unfortunately, when going to the time-shared 3D NOESY as described in Frueh et al (Frueh et al., 2006) it became apparent that the resolutions along the nOe dimensions were too poor to allow for automated signal integration of many of the constraints that were determined in the 4D spectra. We thus developed the technique presented here to provide maximal resolution in the NOESY dimension, with a reasonable experimental time. This was further motivated by the desire to exploit the advantages of a recently acquired 900 MHz spectrometer. Indeed, for an identical number of points in a given dimension, and thus for the same experimental time, the resolution is actually lower at higher fields due to shorter dwell times. Thus, longer experimental times are needed at high fields to overcome the reduction in dwell time. 2048 complex points are traditionally collected in the detected dimension. When the nOes are in this dimension, this crowded spectrum benefits from a digital resolution of 7 Hz for a spectral width of 16 ppm. To achieve such resolution in an indirect dimension, as in NOESY-HSQC derived experiments, would take 34 minutes per FID. With 8 transients accumulated, this translates to roughly 5 hours per plane for a total time of approximately 21 days, assuming 50 complex points in the heteronuclear dimension. In the time-shared NOESY-HSQC the indirect dimension can be reduced to 5 ppm since it only contains either amide protons (15N-dipersed) or methyl protons (13C-dispersed) so that the maximum time would be 6 days and 7 hours. In addition, a much lower resolution is needed in the proton indirect dimension of HSQC-NOESY experiments. Indeed, this dimension here only contains one signal per residue (amide and Ile methyl) or two signals per residue (Val, Leu methyls). Thus the experimental time can be further reduced to about two days. Note that here we only consider cases where sensitivity is not an issue, and our comparison focuses solely on resolution. We did not account for sensitivity improvements that can be achieved with TSNOESY-HSQC derived experiments (TROSY, PEP, Methyl-TROSY) (Frueh et al., 2006; Wurtz et al., 2007; Guo and Tugarinov, 2009).
The time-shared HSQC-NOESY experiment was recorded at 900 MHz on an S48A mutant of the TTE di-domain. This mutation improves sample homogeneity and limits conformational exchanges in the T-domain (Koglin et al., 2006). The 13C-HSQC-NOESY and 15N-HSQC-NOESY spectra of this protein were used to illustrate Figure 2. Using only methyl to amide and amide to amide nOes (HN →, HN and HN →, HMe), the average rmsd of a structural bundle obtained by integrating these spectra is 5.7 Å (calculated by aligning α carbons in secondary structure elements of 10 structures). This relatively large value results from slowly exchanging amide protons that account for about 30% of the TE domain and which are absent from the calculation. For comparison, the structure of the apo sample, obtained with the 3D TS-NOESY-PEP-HSQC/TROSY had a larger rmsd of 9.2 Å. The improvement obtained with the TS-HSQC-NOESY data is the result of many parameters: 1) many new nOes could be observed as a result of the increase in resolution (nOes which could not be observed in the less sensitive 4D experiment), 2) the majority of the signals could be integrated and did not require arbitrary interpretation of signal overlap, and 3) NOESY cross-peak assignments could be corrected: several ambiguous constraints were erroneously attributed to short-range interactions in the NOESY-HSQC spectra when in fact they consisted of both long-range and short-range nOes or in some cases only long range constraints. These constraints could be unambiguously reassigned in the HSQC-NOESY version which allowed differentiation between two candidates. Thus, 259 long range nOes were observed in the NOESY-HSQC derived spectra, whereas the HSQC-NOESY allowed the identification of 272 long range constraints. The fold obtained with the HSQC-NOESY pair of spectra was then used as a basis to identify more nOes involving aliphatic protons in an 15N-dispersed NOESY-TROSY, recorded on a uniformly protonated sample of S48A-TTE. In the end, we obtained a structural bundle with an rmsd of 1.2 Å (Frueh et al., 2008) (including all data, as well as hydrogen bond and torsion angle constraints). The method presented here was decisive in elucidating the structure of this 37 kDa protein.
To compare the TS-NOESY-HSQC/TROSY and TS-HSQC-NOESY, both experiments were recorded for the 31 kDa regulatory subunit of γ-glutamyl-cysteine synthetase (γGCS). In both experiments, the same number of complex points in the 13C, 15N and 1H dimensions were collected; thus, the total measuring times were identical. However, slow aggregation prevented the use of the same sample for both experiments and thus, the sensitivity cannot be compared in these two spectra. The relatively short lifetime of the samples was another motivation to use the time-shared strategy and obtain 15N and 13C dispersed spectra at once. A quick assessment of the relative sensitivity was performed for a 38 kDa protein by recording H/NC 2D planes of the TS-HSQC-NOESY and the TSNOESY-HN-TROSY/HC-HSQC experiment as improved by (Wurtz et al., 2007). For 39 complex points in the 15N, 13C mixed dimension, the TROSY version provided about 30% more signal. In the discussion that follows, only resolution is considered.
Figure 3 shows representative strips obtained with these two experiments. A simple inspection reveals the advantages of the HSQC-NOESY experiment at high-field. Figure 3a displays amide to amide strips obtained with the NOESY-TROSY spectrum. Only sequential nOe cross-peaks can be observed between residues 202 and 203. The higher resolution of the HSQC-NOESY experiment (Figure 3b) reveals additional long range cross-peaks (with 226 and 228 for 202 and with 180 and 181 for 203) that were obfuscated in the NOESY-TROSY spectrum. The situation is even more dramatic in the methyl to methyl region of the 13C-dispersed spectrum. Most cross-peaks overlap in the NOESY-PEP-HSQC spectrum, and only a few signals can be assigned (e.g. 166 δ1 and 175 δ1 to 166 δ2). The HSQCNOESY spectrum provides five more constraints. In addition the overlap is so severe in the NOESYHSQC that peak integration would be unreliable for structure calculations.
The increased resolution also has implications for automated assignment and peak integration procedures. We selected a region of the 13C-dispersed spectra of γGCS and used the automated peak picking procedure as implemented in nmrPipe (Delaglio et al., 1995). To ensure that the result reflects the increase in resolution and not the sensitivity of each spectrum, a weak cross-peak was used as an internal reference to define a threshold. 172 peaks were observed in the NOESY-PEP-HSQC while 287 could be picked and quantitatively analyzed in the HSQC-NOESY spectrum. Thus, the number of distance constraints was nearly doubled with the HSQC-NOESY.
The time-shared 3D HSQC-NOESY experiment presented here provides high resolution nOe cross-peaks with a simultaneous dispersion in the 13C dimension for methyl protons and in the 15N dimension for amide protons. This increase in resolution, at no cost in experimental time, is especially dramatic at the highest-field NMR spectrometers where more popular NOESY-HSQC-derived experiments would impart a much lower resolution in the dimension featuring nOes. Thus, when relaxation losses and the resulting sensitivity reduction is not an issue, it is preferable to use the TS-HSQC-NOESY rather than the TS-NOESY-HSQC family of experiments(Pascal et al., 1994; Uhrín et al., 2000; Xia and Zhu, 2001; Frueh et al., 2006; Wurtz et al., 2007; Guo and Tugarinov, 2009). This will be particularly valuable for α-helical proteins and unfolded proteins, in general, as well as for large proteins. The experiment is particularly suitable for ILV labelled proteins since all signals are distinct from the water resonance, which can hence be suppressed without affecting detection of nOe cross peaks. For such samples, the optimal resolution then allows one to rescue the poor dispersion of methyl protons. Further improvements may be obtained in the indirect dimensions by using non-uniform sampling (NUS) (Schmieder et al., 1993; Rovnyak et al., 2004; Tugarinov et al., 2005; Hyberts et al., 2007; Hyberts et al., 2009) to optimize the resolution without dramatic increase in measurement time. Since long evolution times are then recorded in the indirect dimensions, TROSY variation, as suggested in the discussion above, should be beneficial, in particular for large proteins. Even without NUS, the use of the TS-HSQC-NOESY allows one to readily exploit the advantages of high-field instruments. This results in a dramatically increased number and improved quantification of distance constraints. This experiment was key in the successful determination of the structures of two large proteins of 37 kDa and 31 kDa. The experiment described here has helped to resolve many ambiguities in the crowded NMR spectra of these systems. The pulse sequences, together with an nmrPipe script to process the data, can be downloaded from http://gwagner.med.harvard.edu.
This research was supported by the National Institutes of Health (grants GM 47467, CA127990 and RR 00995) and a Human Frontier Science Fellowship to AK. AK is currently a Director's fellow at the Los Alamos National Laboratory. We thank Dr. Ruma Banerjee for providing the construct for γGCS. We thank two anonymous reviewers for their comments on water suppression.