Anal. Chem. 2007, 79, 2957-2960
Improvements in Complex Mixture Analysis by NMR: DQF-COSY iDOSY Jacob M. Newman and Alexej Jerschow*
Department of Chemistry, New York University, 100 Washington Square East, New York, New York 10003
Diffusion-ordered NMR spectroscopy (DOSY NMR) is a highly useful tool for the study of complex mixtures via NMR. Often, spectral overlap limits the ability of obtaining cleanly separated subspectra of the components due to inherently instable multiexponential fits or data inversion procedures. Three-dimensional DOSY variants offer the advantage of separating individual peaks in an additional dimension, such that robust monoexponential fits to cross-peaks may be used to determine the diffusion coefficients with higher accuracy. For sensitivity reasons, methods based on proton nuclei are preferable. We show that a double-quantum-filtered COSY-DOSY experiment provides advantages over COSY-DOSY, while high signalto-noise ratios are maintained. We demonstrate the viability of the technique by applying it to a solution of single-stranded DNA oligomers and to a mixture of unprocessed beeswax and decanol. Diffusion-ordered NMR spectroscopy (DOSY)1 is based on the signal attenuation due to the action of a gradient echo. The attenuation can be described by
S(g) ) C exp[-D(γsgδ)2(∆ - δ/3)]
(1)
with D being the diffusion coefficient of the molecule (in m2/s), g the magnetic field gradient used (in T/m), γ the gyromagnetic ratio, ∆ the duration between the start of the dephasing and the start of the rephasing gradient pulse (Figure 1), and δ the duration of the gradient pulses. The shape factor s of the gradient pulses takes into account the deviation of the decay behavior due to a nonrectangular shape (for sine-shaped gradients it can be derived from the Bolch-Torrey equations2 as s ) (2/π)[(∆/(∆ δ/3)](-1/2). A monoexponential fit to the decay of an NMR resonance line as a function of the gradient strength allows one to determine the diffusion coefficient if there is no spectral overlap. When several signals from different molecules overlap, a multiexponential fit is required, which is, however, numerically unstable. Several different methods based on regularization,3,4 maximum entropy,5 multivariate curve resolution,6 the regularized * Corresponding author. E-mail:
[email protected], Fax (212) 260 7905. (1) Morris, K. F.; Johnson, C. S. J. Am. Chem. Soc. 1992, 114, 3139-3141. (2) Callaghan, P. T. Principles of Nuclear Magnetic Resonance; Oxford Science Publications: Oxford, 1995. 10.1021/ac061760g CCC: $37.00 Published on Web 03/01/2007
© 2007 American Chemical Society
Figure 1. (A) COSY-iDOSY and (B) DQF-COSY-iDOSY NMR pulse sequences. The dephasing and rephrasing gradients are labeled Gz, and the coherence selection gradients are labeled Gx.
resolvent transform,7 and neural networks8 have been adapted to help overcome this limitation. Based on the calculation of the diffusion coefficient distribution one can reconstruct a spectrum, which shows a chemical shift axis in one dimension and a diffusion coefficient axis in the second dimension (2D DOSY).1,9 Another commonly used approach to address the numerical instability problem is to increase the spectral frequency separation such that only a single-component fit is required to a single resonance. Along these lines, much of the motivation of the development of 3D DOSY experiments originated from the increased spectral resolution afforded by COSY,10,11 NOESY,12 TOCSY,13 J-spectroscopy,14,15 and HMQC16 through the introduction of an additional dimension. While spectra such as the ones obtained from DOSY-HMQC offer superb resolution, these have to operate under most practical circumstances with low natural (3) Provencher, S. W. Comput. Phys. Commun. 1982, 27, 229-242. (4) Provencher, S. W. CONTIN Users Manual (Version 2); EMBL: Heidelberg, Germany, 1984. (5) Delsuc, M. A.; Malliavin, T. E. Anal. Chem. 1998, 70, 2146-2148. (6) Huo, R.; Wehrens, R.; Buydens, L. M. J. Magn. Reson. 2004, 169, 257269. (7) Armstrong, G. S.; Loening, N. M.; Curtis, J. E.; Shaka, A. J.; Mandelshtam, V. A. J. Magn. Reson. 2003, 163, 139-148. (8) Sebastiao, R. C.; Pacheco, C. N.; Braga, J. P.; Pilo-Veloso, D. J. Magn. Reson. 2006, 182, 22-28. (9) Johnson, C. S., Jr. Prog. Nucl. Magn. Reson. Spectrosc. 1999, 34, 203-256. (10) Nilsson, M.; Gil, A. M.; Delgadillo, I.; Morris, G. A. Chem. Commun. 2005, 1737-1739. (11) Wu, D.; Chen, A.; Johnson, C. S., Jr. J. Magn. Reson., Ser. A 1996, 121, 88-91. (12) Gozansky, E. K.; Gorenstein, D. G. J. Magn. Reson. B 1996, 111, 94-96. (13) Jerschow, A.; Mu ¨ ller, N. J. Magn. Reson. A 1996, 123, 222-225. (14) Lucas, L. H.; Otto, W. H.; Larive, C. K. J. Magn. Reson. 2002, 156, 138145. (15) Nilsson, M.; Gil, A. M.; Delgadillo, I.; Morris, G. A. Anal. Chem. 2004, 76, 5418-5422. (16) Barjat, H.; Morris, G. A.; Swanson, A. G. J. Magn. Reson. 1998, 131, 131138.
Analytical Chemistry, Vol. 79, No. 7, April 1, 2007 2957
Figure 2. (A) COSY and (B) DQF-COSY subspectra of 5′-TTTTTTTTTT-3′ ssDNA extracted from the respective 3D experiments at a diffusion coefficient of 2.0 × 10-10 m2/s. (C) COSY and (D) DQF-COSY subspectra of 5′-GGGGG-3′ ssDNA extracted from the respective 3D experiments at a diffusion coefficient of 1.2 × 10-9 m2/s. The resonances are labeled according to IUPAC DNA numbering.20
abundance (e.g., 13C, 15N), which results in poor sensitivity and long experiment times. This is especially true for complex mixture analysis, where the molecules of interest may be represented with relatively low concentrations. In these cases, proton 3D DOSY experiments become the most appropriate methods. We show in this work that the use of a double-quantum filter (DQF) in the COSY-DOSY experiment greatly enhances the usefulness of this experiment, due to filtering of strong singlet peaks along the diagonal. This approach therefore further increases the resolution afforded by 3D DOSY for low-concentration solutes, in particular for those with signals close to the diagonal. As a result, we increase the validity of the monoexponential fits in these experiments, leading to more robust DOSY experiments. We implement this sequence as an iDOSY10,15 experiment, where the diffusion labeling gradients are intertwined with the COSY delays in the pulse sequence. We demonstrate the improvement of diffusion separation by this method on a mixture of DNA oligomers and an extract of unprocessed beeswax mixed with decanol. 2958
Analytical Chemistry, Vol. 79, No. 7, April 1, 2007
EXPERIMENTAL SECTION Samples. The first sample consisted of 2.9 mM 5′-GGGGG-3′ and 2.2 mM 5′-TTTTTTTTTT-3′ single-stranded DNA (Integrated DNA Technologies) in a phosphate buffer that was 100 mM in NaCl and 10 nM Na2HPO4 in D2O. The second sample consisted of 21 mg of natural unprocessed beeswax obtained from a local art supply store and 0.25 mL of decyl alcohol which were dissolved in 1.5 mL of CD2Cl2. NMR Experiments. All spectra were acquired on a Bruker AV-500 NMR spectrometer, equipped with a 5-mm TXI probe with triple axis gradients. The spectra were processed using Bruker TOPSPIN 1.3 software. The spectra were obtained without temperature control at ambient temperature (23 °C) to reduce convection problems.17-19 The pulse sequences of the DQF(17) (18) (19) (20)
Nilsson, M.; Morris, G. A. J. Magn. Reson. 2005, 177, 203-211. Jerschow, A. J. Magn. Reson. 2000, 145, 125-131. Jerschow, A.; Mu ¨ ller, N. J. Magn. Reson. 1997, 125, 372-375. Lie´becq, C. Biochemical Nomenclature and Related Documents. 2nd ed.; Portland Press: London, 1992.
Figure 3. (A) COSY and (B) DQF-COSY subspectra corresponding to decyl alcohol extracted from the respective 3D experiments of the beeswax mixture at a diffusion coefficient of 2.4 × 10-9 m2/s. (C) COSY and (D) DQF-COSY subspectra of a major unidentified component of beeswax from the respective 3D experiments of the mixture at a diffusion coefficient of 1.0 × 10-9 m2/s.
COSY-iDOSY and the COSY-iDOSY experiments are shown in Figure 1. The 4096 × 512 data points were acquired in the COSY dimension (t2 × t1) to cover 5000 Hz each, and 8 gradient steps were used in the diffusion dimension. The diffusion delay, ∆, was set to 23 ms, and the diffusion gradient pulse length was 3 ms for the oligonucleotide samples, and 19 and 3 ms, respectively, for the beeswax sample. The gradient strength was varied on a linear scale between 1 and 47.5 G/cm maximum values, with sine-shaped waveforms. The 90° pulse length was 7.4 and 9.5 µs for the oligonucleotide and beeswax samples, respectively. The gradient recovery delay τ was 200 µs. The DQF selection gradient pulses (maximum strength 5 and 10 G/cm, respectively) were sineshaped and were applied for a duration of 1 ms along the x direction to avoid interference with the diffusion gradients along z. The recycle delay was 1.5 s. The phase cycle was x, -x for the first pulse and the receiver phase in both the DQF-COSY-iDOSY and the COSY-iDOSY experiments. Sign discrimination in t1 was performed with the States-TPPI method. Two signal accumulations were used (5.1 h experiment time).
In processing the data, a window function of cos2[(π/2)(t/aq)] was used in both COSY dimensions, where t is the time at which the respective data point is acquired and aq is the total acquisition time. The spectrum was zero-filled to 8192 points in the t2 dimension and 1024 points in the t1 dimension. After a twodimensional Fourier transform in magnitude mode, the DOSY dimension was reconstructed on a logarithmic scale using a monoexponential fit to all resonances, and the peaks were reconstructed in the diffusion constant dimension on a grid of 32 data points. The subspectra were obtained by averaging 4 slices centered at the respective diffusion coefficients. RESULTS AND DISCUSSION Figure 2 shows the subspectra extracted from the COSYiDOSY and DQF-COSY iDOSY experiments of the oligonucleotide mixture. For both experiments, two slices from the transformed 3D DOSY spectra are shown at the diffusion coefficients corresponding to the 5- and 10-mer, respectively. While the DQF-COSY iDOSY experiment produces some overall signal reduction due to the nature of the double-quantum filter, it also suppresses the Analytical Chemistry, Vol. 79, No. 7, April 1, 2007
2959
very strong singlet peaks, leading to a better resolution of the signals close to the diagonal. The resolution in the DOSY dimension was sufficient to clearly separate the two DNA oligomers. Another example illustrating the utility of the double-quantum filtration in facilitating spectral interpretation can be seen in Figure 3, where we examine a sample of natural beeswax in a mixture with decanol. The large singlet diagonal signal at 1.3 ppm in the COSY iDOSY obscures a number of cross-peaks that are apparent in the double-quantum-filtered spectrum, while it is well suppressed in the DQF-COSY iDOSY experiment, as are some additional strong diagonal peaks. The two components, decanol and a major fatty acid component of beeswax, are nicely separated in the 2D subspectra despite the spectral crowding and topological similarity of the coupling patterns. We have chosen here the DQF-COSY iDOSY variant of this approach, in which the diffusion delay is simply inserted after the indirect dimension evolution delay in a regular COSY (or DQFCOSY) experiment. This approach proves advantageous in reducing the number of required phase cycling steps over the sequential implementation (COSY-DOSY or DOSY-COSY). The phase cycle frequently sets a lower limit on the experiment time, which is particularly lengthy in a sequential implementation. In our case, the DQF-COSY DOSY implementation was longer by at least a factor of 4 than the iDOSY implementation, which would result in prohibitively long experiment times. In addition, a 50% signal loss is incurred from performing a stimulated echo. The drawbacks of the iDOSY implementation, however, are that strong signal losses may be experienced with samples with short T2 times (on the order of 20 ms and below) and that J-coupling evolution prevents one from producing correctly phased spectra. The phasing problem is also present in the sequential implementation, albeit to a smaller extent. In most cases, however, the improvement in experiment time makes the iDOSY experiment preferable to the standard DOSY.
2960
Analytical Chemistry, Vol. 79, No. 7, April 1, 2007
CONCLUSION We demonstrate in this work an extension to three-dimensional DOSY experiments, which proves to be useful for the analysis of complex mixtures by NMR. Frequently, the diffusion separation is not sufficient to allow one to determine the diffusion coefficients with accuracy from overlapping NMR resonances. Three-dimensional DOSY sequences allow one to reduce the spectral overlap by introducing further frequency dimensions, such that a robust monoexponential fit can produce accurate diffusion coefficients. While good resolution is achievable with heteronuclear implementations, such as HSQC-DOSY, these have sensitivity drawbacks when mixtures with low concentrations are investigated. COSY-DOSY therefore becomes the best alternative. We show here that adding a double-quantum filter greatly enhances the utility of this experiment, since strong singlet peaks are suppressed, thereby increasing the resolution while retaining high signal-to-noise ratios. We show diffusion-separated spectra from a mixture of oligonucleotides, as well as natural beeswax. ACKNOWLEDGMENT This work was supported by U.S. National Science Foundation Grant CHE-0554400 and conducted in a facility constructed with support from Research Facilities Improvement Grant C06 RR16572-01 from the National Center for Research Resources, National Institutes of Health. NYU’s NMR resources were supported by NSF Grant MRI-0116222. A.J. is a member of the New York Structural Biology Center, which is supported by the New York State Office of Science, Technology, and Academic Research and National Institutes of Health Grant P41 FM66354. Received for review September 18, 2006. Accepted January 29, 2007. AC061760G
