Application of the FLIPSY Pulse Sequence for Increased Sensitivity in

Anal. Chem. 2008, 80, 3365-3371

Application of the FLIPSY Pulse Sequence for Increased Sensitivity in 1H NMR-Based Metabolic Profiling Studies Michael Lauridsen,*,† Anthony D. Maher,§ Hector Keun,§ John C. Lindon,§ Jeremy K. Nicholson,§ Nils T. Nyberg,‡ Steen H. Hansen,† Claus Cornett,† and Jerzy W. Jaroszewski‡

Department of Pharmaceutics and Analytical Chemistry and Department of Medicinal Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark, and Biomolecular Medicine, SORA Division, Faculty of Medicine, Imperial College London, Sir Alexander Fleming Building, South Kensington, London SW7 2AZ, U.K.

Metabolite profiling relies on optimal precision of the acquired data, which requires, among others, a high signal-to-noise ratio (S/N). In addition, increased S/N will increase the likelihood of identification of new biomarkers. Here we introduce, for the first time in metabolite profiling studies by 1H NMR, an approach to enhance the precision of multivariate regression models by use of the FLIPSY (flip angle adjustable one-dimensional NOESY) pulse sequence, augmented by a homospoil pulse after the presaturation period to provide superior baseline quality. Unlike NOESYPRESAT, the standard one-dimensional (1D) sequence generally used in metabonomic studies, FLIPSY incorporates a variable flip angle, allowing use of the Ernst angle for excitation and thus optimization of S/N ratios according to spin lattice relaxation times (T1) of individual resonances. T1 values of metabolites present in human urine were determined by inversion-recovery experiments and subsequently used in calculations of optimal experimental conditions. Comparison of human urine analysis by the FLIPSY and NOESYPRESAT demonstrated an increase of S/N ratio in the former case that amounts to approximately 7% when measured for the hippurate doublet at δ 7.84. An orthogonal projection to latent structures discriminant analysis (O-PLS-DA) model exhibited superior discrimination between controls and simulated phenylketonuria urines when using data generated by the FLIPSY as compared to NOESYPRESAT. In recent decades, metabolic profiling of biofluids, such as urine and blood plasma, by 1H nuclear magnetic resonance (1H NMR) spectroscopic and mass spectrometric (MS) techniques has contributed significantly to life sciences and drug development as a systems biology approach.1-3 1H NMR spectra are informa* Corresponding author. E-mail: [email protected]. Fax: +45 3530 6001. † Department of Pharmaceutics and Analytical Chemistry, University of Copenhagen. ‡ Department of Medicinal Chemistry, University of Copenhagen. § Imperial College London. (1) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Pharm. Res. 2006, 23, 10751088. 10.1021/ac702563u CCC: $40.75 Published on Web 03/26/2008

© 2008 American Chemical Society

tion-rich, providing the user with highly reproducible “snapshots” of the metabolic content of a biofluid in a single step. This contrasts with MS approaches, which require a preceding chromatographic step for resolution of metabolites. In addition, 1H NMR spectroscopy is associated with a very high analytical reproducibility as compared to MS techniques.4,5 However, the major disadvantage of NMR is a relatively low sensitivity that results in comparatively low sample turnover times. Given the large amount of samples that have to be analyzed in metabolic profiling studies, any sensitivity increase of 1H NMR data acquisition will result in increased throughput, increased data quality, or both. Classification studies aiming to predict class membership by supervised multivariate approaches rely on acquisition of 1H NMR data with the optimal signal-to-noise level (S/N) for discriminating resonances. Many different approaches to improve S/N exist, including use of higher field strengths, cryogenic probe technologies, miniaturization, and more efficient pulse sequences.6-8 Major goals in metabonomic studies include biomarker identification, characterization and prediction of disease states, and assessment of the metabolic consequences of drug administration and other medical interventions.9-11 These rely on precise measurement of metabolites for discrimination between groups. (2) Clayton, A. T.; Lindon, J. C.; Cloarec, O.; Antti, H.; Charuel, C.; Hanton, G.; Provost, J. P.; Le Net, J. L.; Baker, D.; Walley, R. J.; Everett, J. R.; Nicholson, J. K. Nature 2006, 440, 1073-1077. (3) Nicholson, J. K.; Wilson, I. D. Prog. Nucl. Magn. Reson. Spectrosc. 1989, 21, 449-501. (4) Dumas, M. E.; Maibaum, E. C.; Teague, C.; Ueshima, H.; Zhou, B.; Lindon, J. C.; Nicholson, J. K.; Stamler, J.; Elliott, P.; Chan, Q.; Holmes, E. Anal. Chem. 2006, 78, 2199-2208. (5) Keun, H. C.; Ebbels, T. M. D.; Antti, H.; Bollard, M. E.; Beckonert, O.; Schlotterbeck, G.; Senn, H.; Niederhauser, U.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Chem. Res. Toxicol. 2002, 15, 1380-1386. (6) Foxall, P. J.; Spraul, M.; Farrant, R. D.; Lindon, L. C.; Neild, G. H.; Nicholson, J. K. J. Pharm. Biomed. Anal. 1993, 11, 267-276. (7) Keun, H. C.; Beckonert, O.; Griffin, J. L.; Richter, C.; Moskau, D.; Lindon, J. C.; Nicholson, J. K. Anal. Chem. 2002, 74, 4588-4593. (8) Van, Q. N.; Chmurny, G. N.; Veenstra, T. D. Biochem. Biophys. Res. Commun. 2003, 301, 952-959. (9) Holmes, E.; Nicholls, A. W.; Lindon, J. C.; Ramos, M.; Spraul, M.; Neidig, P.; Connor, S. C.; Connelly, J.; Damment, S. J. P.; Haselden, J.; Nicholson, J. K. NMR Biomed. 1998, 11, 235-244. (10) Lindon, J. C.; Holmes, E.; Bollard, M. E.; Stanley, E. G.; Nicholson, J. K. Biomarkers 2004, 9, 1-31.

Analytical Chemistry, Vol. 80, No. 9, May 1, 2008 3365

optimization of S/N difficult. On the other hand, FLIPSY (flip angle adjustable one-dimensional NOESY, Figure 1) enables control of the overall flip angle of the sequence17 and thus optimization of S/N for certain metabolites by application of the Ernst angle.18 Here, we present an application of the FLIPSY17 pulse sequence for increased classification sensitivity between controls and simulated phenylketonuria urines, using sophisticated pattern recognition techniques.

Figure 1. 1D 1H NMR pulse sequences for acquisition of data in the presence of a strong solvent resonance: (a) NOESYPRESAT, (b) FLIPSY, (c) modified FLIPSY. Irradiation at the solvent frequency during relaxation delay and for NOESYPRESAT also during the mixing time is indicated by horizontal bars. Variable flip angles are indicated by hatched bars. Solid bars indicate 90° pulses, open bars indicate 180° pulses, and shaded shape indicates a homospoil fieldgradient pulse. Phase cycles and experimental parameters are presented in the Experimental Section.

The metabolites responsible for the discrimination are not always known beforehand, but situations occur where information about one particular metabolite or a class of metabolites is of particular interest (metabolic profiling), e.g., in diabetes or other metabolic diseases as well as in many toxicological studies. In these cases, optimization of experimental conditions for detection of these particular metabolites is expected to enhance detection and interpretation of the differences between groups, as has been demonstrated previously with a semiselective TOCSY approach.12 Furthermore, an increase in the S/N ratio for these metabolites leads to an improved specificity and sensitivity of chemometric models. Solvent suppression techniques are an essential part of nearly all biofluid NMR experiments. Methods rely on null excitation, relaxation, or selective irradiation, and application of presaturation and pulsed field gradients are encountered frequently.3 The majority of 1H NMR-based metabonomic studies on urine samples employ the popular NOESYPRESAT pulse sequence for solvent peak suppression and signal excitation.13,14 It consists of the first increment of the two-dimensional (2D) NOESY pulse sequence,15,16 supplemented with a low-power continuous-wave radiofrequency irradiation of the solvent (water) resonance during the relaxation delay and mixing time (Figure 1). Although NOESYPRESAT is robust, simple to use, and provides efficient water peak suppression, it requires a 90° flip angle of the read pulse, making (11) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Expert Rev. Mol. Diagn. 2004, 4, 189-199. (12) Sandusky, P.; Raftery, D. Anal. Chem. 2005, 77, 7717-7723. (13) Nicholson, J. K.; Foxall, P. J. D.; Spraul, M.; Farrant, R. D.; Lindon, J. C. Anal. Chem. 1995, 67, 793-811. (14) Lindon, J. C.; Nicholson, J. K.; Everett, J. R. Annu. Rep. NMR Spectrosc. 1999, 38, 1-88. (15) Kumar, A.; Wagner, G.; Ernst, R. R.; Wu ¨ thrich, K. J. Am. Chem. Soc. 1981, 103, 3654-3658. (16) Kumar, A.; Ernst, R. R.; Wu ¨thrich, K. Biochem. Biophys. Res. Commun. 1980, 95, 1-6.

3366 Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

EXPERIMENTAL SECTION Chemicals. Buffer solution (0.33 M, pH 7.4) was prepared from K2HPO4‚3H2O and KH2PO4 in D2O (99.9 atom % deuterium). Sodium 3-trimethylsilylpropionate-d4 (TSP, 99 atom % deuterium) was added to obtain a concentration of 0.3 mM in the buffer. Final adjustment of the pH was carried out with HCl or KOH solutions using a pH meter equipped with a glass combination electrode. Water purified by deionization and membrane filtration (0.22 µm) on a Millipore system was used. Sample Collection and Preparation. Human urine was collected from healthy volunteers and stored at -25 °C until preparation. Samples were diluted with the buffer solution in a ratio of 2:1 (final buffer concentration 0.11 M), centrifuged at 10 000g for 10 min, and 0.6 mL aliquots were transferred to 5 mm NMR tubes and stored frozen (-25 °C) until analysis. Simulated phenylketonuria urine was prepared by adding phenylalanine (46.5 mM in H2O) to the final concentration of 7.75 mM to 14 samples. An equivalent amount of water was added to 15 control samples. NMR Spectroscopy. Longitudinal (T1) relaxation times of metabolites in urine samples were determined at 25 °C using an inversion-recovery experiment employing a modified FLIPSY pulse sequence instead of the 90° read pulse on a Bruker Avance 600 MHz spectrometer operating at 600.13 MHz for 1H, using a TXI probe equipped with a z-gradient and acquiring 64 transients with 64k time-domain points and a spectral width of 6188 Hz. The presaturation period was 3.0 s, and the relaxation delay was 25.3 s (including the acquisition time). The delay between the inversion pulse and the first 90° pulse of the FLIPSY sequence was varied between 2.2 ms and 20 s (a total of 16 experiments). Irradiation of water was applied immediately before the modified 90° FLIPSY pulse train, by using a second channel for presaturation. This allowed for a simultaneous hard pulse for inversion of magnetization, when variable delay was below 3 s, ensuring the same presaturation period in all experiments. NOESYPRESAT and FLIPSY experiments were performed for two pulse repetition periods (FLIPSY 2.7 s and NOESYPRESAT 2.8 s; FLIPSY 3.7 s and NOESYPRESAT 3.8 s) at 27 °C for three human urines collecting 16 transients, each time collecting 32k time-domain points, on a Bruker Avance 600 MHz spectrometer operating at 600.29 MHz for 1H, using a TXI probe equipped with x-, y-, and z-gradients. Irradiation of the solvent resonance was performed during the relaxation delay for NOESYPRESAT and the FLIPSY and furthermore during the mixing time (100 ms) for NOESYPRESAT. To evaluate the effect of a shorter solvent presaturation period, different presaturation delays (2.0, 1.5, 1.0, and 0.5 s) were applied for these experiments keeping the receiver gain, and the pulse repetition period constant at 3.8 s for (17) Neuhaus, D.; Ismail, I. M.; Chung, C. W. J. Magn. Reson., Ser. A 1996, 118, 256-263. (18) Ernst, R. R. Adv. Magn. Reson. 1966, 2, 1-135.

NOESYPRESAT and 3.7 s for the FLIPSY. Cyclops (first pulse x, -x; second pulse x, x, x, x, x, x, x, x, -x, -x, -x, -x, -x, -x, -x, -x; third pulse x, x, -x, -x, y, y, -y, -y; receiver phase x, -x, -x, x, y, -y, -y, y, -x, x, x, -x, -y, y, y, -y) and Exorcycle (first pulse x; second pulse x, y, -x, -y; third pulse x, x, x, x, y, y, y, y, -x, -x, -x, -x, -y, -y, -y, -y; receiver phase x, -x, x, -x, -x, x, -x, x) were applied for phase cycling for NOESYPRESAT and the FLIPSY, respectively. The FLIPSY excitation pulse was adjusted to 61° to provide the optimal S/N ratio per time unit for T1 ) 3.7 s (acquisition time 1.7 s, presaturation delay 1 s) or to 68° for pulse repetition period 3.7 s and T1 ) 3.7 s. Eight “dummy scans” were inserted at the beginning of the pulse sequence to allow steady-state conditions. Exponential line broadening was applied to all spectra. For the classification study, human urine samples were analyzed at 27 °C on the spectrometer system used for the T1 determination. After eight “dummy scans”, 96 transients, each consisting of 41k time-domain data points, were acquired with a flip angle of 62.3° for the FLIPSY, a presaturation delay of 1 s, and an acquisition time of 1.7 s. A sine-shaped homospoil gradient pulse (1 ms, approximately 25 G/cm) was added to the FLIPSY sequence after the presaturation pulse. Following phasing and baseline correction, NOESYPRESAT and FLIPSY data were zerofilled to 32k real data points and apodized with an exponential line-broadening factor of 1.0 Hz prior to Fourier transformation. Assignment of metabolites was performed with reference to previously published data.14,19 NMR Data Preprocessing and Statistical Analysis. Data were manually phased and baseline-corrected in TopSpin ver. 1.2 (Bruker BioSpin, Rheinstetten, Germany) and imported into Matlab ver. 7.0 software (MathWorks, Natick, MA), where interpolation of the spectra onto a common chemical shift axis, removal of the water resonance, and calibration to TSP (δ 0.0) was performed. For calculation of T1 values from inversionrecovery data two approaches was used, the first with manually selected signal ranges and the second with signal ranges corresponding to a moving window of 0.015 ppm (51 data points). For both, the sum of signal intensities of the selected ranges were fitted to the function Mt ) M0[1 - a exp(-t/T1)] (Curve Fitting Toolbox, ver. 1.1.7, MathWorks). Univariate data analysis was performed using Microsoft Excel 2002 on data segmented into integral regions (bins) of 0.01 ppm or using full resolution data sets using Matlab. Principal component analysis (PCA) and orthogonal projection to latent structures discriminant analysis (O-PLS-DA) calculations were performed with Simca P-11 ver. 11.0 software (Umetrics, Umeå, Sweden) using the autofit function on binned (bucketed) and full resolution data sets; the solvent region and the edges of the spectrum were removed from the data set prior to analysis. S/N ratios were calculated by the “sino” command available in TopSpin ver. 1.2 (Bruker BioSpin, Rheinstetten, Germany). Product Operator Calculations. Derivation of product operators for FLIPSY and NOESYPRESAT was performed using POMA20 in Mathematica ver. 6.0 (Wolfram Research, Champaign, (19) Tang, H.; Wang, Y.; Nicholson, J. K.; Lindon, J. C. Anal. Biochem. 2004, 325, 260-272. (20) Guntert, P.; Schaefer, N.; Otting, G.; Wu ¨ thrich, K. J. Magn. Reson., Ser. A 1993, 101, 103-105.

IL) by summing the magnetization after one complete phase cycle without taking into consideration the relaxation effects. RESULTS AND DISCUSSION Theory. Use of a 90° read pulse requires a pulse repetition period corresponding to approximately 5T1 for 99% relaxation of a resonance; thus, a lot of spectrometer time is spent if nearly full relaxation is to be achieved before the next pulse is applied. However, collection of accurate (true) signal intensity data is possible only by this approach. It has been shown that use of a flip angle of 83° and a pulse repetition period equal to 4.5 times the longest T1 value in the sample will allow for the collection of accurate integrals for all resonances in the spectrum in the shortest amount of time.21 If accurate integrals are not necessary for all resonances in the spectrum, but only precise integrals are needed, as in metabonomic studies,7 one should optimize the experiment for maximal S/N. Application of the Ernst angle (Ropt)19 (eq 1, where PR is the pulse repetition period) allows a shorter pulse repetition period and thereby provides optimal S/N per time unit. However, as a result, lower integral accuracy will result for resonances with longer relaxation times.

cos(Ropt) ) e-PR/T1

(1)

To calculate the theoretical gain in S/N we rely on eq 2, derived from the Bloch equations21 and defining the recovery (RC) of magnetization aligned with the external magnetic filed (Mz) after one transient following a read pulse with flip angle R.

RC )

(ePR/T1 - 1) ePR/T1 - cos R

(2)

The Bloch equation (eq 3) alone does not provide information about S/N per time unit as it only reveals the amount of magnetization recovered after time t.

Mt ) M0 1 - e-t/T1

(3)

S/N as a function of R and PR can now be calculated by application of eq 4.21

S/N ) RC sin(R)/xPR

(4)

The graphical representation of the theoretical (calculated) S/N as a function of flip angle is shown in Figure 2. Optimization of S/N thus requires knowledge of the relaxation parameter T1, which can be determined using the well-known inversionrecovery method. The duration of PR in NOESYPRESAT and the FLIPSY experiments depends mainly on the variables presaturation time and the acquisition time. The presaturation time must be long enough to allow adequate water peak suppression, and the acquisition time must be long enough to allow for T2 relaxation to occur largely to completion so that truncation of decaying magnetization is avoided. Absence in the FLIPSY sequence of the (21) Traficante, D. D. Concepts Magn. Reson. 1992, 4, 153-160.

Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

3367

Figure 2. Theoretical S/N ratio as a function of flip angle. Calculations performed for PR ) 2.8 s and T1 ) 3.6 s. Table 1. Experimental T1 Relaxation Times of Selected Metabolites in Human Urine

metabolite

δ

multiplicity

TSP lactate R-hydroxyisobutyrate alanine succinate citrate citrate dimethylamine creatinine phenylalanine

0.00 1.34 1.36 1.49 2.41 2.55 2.68 2.73 3.05 3.13

phenylalanine

3.30

glycine hippurate phenylalanine

3.57 3.97 4.01

unassigned creatinine unassigned urea phenylalanine phenylalanine phenylalanine hippurate hippurate hippurate

4.04 4.06 4.44 5.79 7.33 7.38 7.43 7.55 7.64 7.83

singlet doublet singlet doublet singlet doublet doublet singlet singlet double doublet double doublet singlet doublet double doublet tripletc singlet singlet singlet doubletd multipletd multipletd tripletd tripletd doubletd

T1 (s)

standard deviation (s)

3.15 1.41 1.41 1.24 2.54 0.81 0.81 4.53 2.63 1.04

0.12 0.24 0.19 0.29 0.28 0.1 0.11 0.54 0.12 0.07

1.02

0.04

2.50 1.59b 2.00

0.23 0.04 0.08

5.11 2.04 3.22 0.26 2.54 3.34 3.01 2.96 3.81 3.07

0.48 0.07 0.69 0.02 0.05 0.08 0.04 0.05 0.18 0.08

T1 (s)a 5.83 2.12 1.85 0.88 3.96

2.12

3.07

4.76

a From ref 22, values determined at 37 °C. b Uncertain value due to signal overlap. c Multiplicity uncertain. d Apparent multiplicity of AA′BB′C pattern.

mixing time period present in NOESYPRESAT (usually 100 ms) eliminates the T1 relaxation during this period (approximately 2.8% for T1 ) 3.6 s) and shortens the total pulse repetition period. T1 Relaxation Times of Metabolites in Human Urine. The T1 relaxation time values of metabolites in human urine were determined from inversion-recovery data employing peak areas (Table 1) and are in accordance with previously reported values when considering the different temperatures at which the data were acquired.22 Data for T1 calculations were obtained by using integration either of segments (bins) with fixed sizes (51 data points) or manually selected segments with variable sizes repre(22) Munasinghe, J. P.; Colebrook, L. D.; Attard, J. J.; Carpenter, T. A.; Hall, L. D. Magn. Reson. Chem. 1998, 36, 116-123.

3368 Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

senting entire peaks of selected resonances. Both approaches were in good accordance with each other suggesting that reliable T1 values are obtained. Small differences in the T1 values obtained using these two approaches for certain peaks (e.g., the phenylalanine multiplet at δ 7.38) suggest the presence of overlapped peaks. Thus, it appears that more reliable results can be obtained using the small bin sizes. The variations of T1 values for main metabolites in human urine are presented graphically in Figure 3. Apparent transverse relaxation time T2* is important for choice of optimal acquisition time. For the TSP peak, T2* was estimated to be 1.3 s from the peak width at half-height (∆ν ) (1/π)T2*). However, for practical purposes the acquisition time was limited to 1.7 s as an exponential line broadening of 1 Hz was applied. Water Peak Suppression. Lowering the presaturation time is important in order to optimize the FLIPSY sequence, as this delay is one of the two major contributors to the total PR. Investigations of spectra of urine samples acquired with different presaturation times indicated that this strategy can be applied to lower the PR without significant loss of S/N (Figure 4). A t test showed that the two extreme presaturation periods (0.5 and 2.0 s) for NOESYPRESAT and the FLIPSY caused no difference between the means of the S/N for both TSP signals and signals of aromatic resonances (P > 0.05). Thus the presaturation time can be lowered substantially without dramatic loss of S/N, allowing a reduction of the PR and sample heating. However, decreasing the presaturation time caused increase of the residual water signal. As a consequence the S/N per time unit may decrease due to a reduction of the receiver gain to prevent overload of the digitizer. Nevertheless, increasing the power level during presaturation ameliorated this effect (Figure 5). Complete removal of the residual water signal was not achieved in any FLIPSY experiments, despite the increase in power level during presaturation. However, receiver gain was not limited by the solvent signal. Modification of the Pulse Sequence. The original FLIPSY pulse sequence17 was augmented by a 1 ms homospoil fieldgradient pulse after the presaturation to dephase any residual transverse magnetization. This improved the baseline around the residual water signal rendering the spectrum easier to phase and providing better or at least equal baseline characteristics as compared to the original FLIPSY and NOESYPRESAT (Figure 5). Theoretical Increase in Relative S/N. We see five advantages of the FLIPSY pulse sequence as compared to the NOESYPRESAT for metabonomics studies: (1) use of a variable flip angle; (2) the absence of a mixing time, shortening the total pulse repetition period; (3) absence of relaxation during mixing time as a result of short interpulse delays; (4) contributions from regions in the sample experiencing suboptimal pulses due to B1 field inhomogeneity have less influence in FLIPSY as the Iz product operator can be described by sin5(R) compared to sin3(R) for NOESYPRESAT (see the Experimental Section for the method of calculation); however, for the same reason, the FLIPSY is not as robust to suboptimal pulses as NOESYPRESAT due to the faster decline of signal in the inhomogeneous region; (5) after the FLIPSY pulse train the magnetization is always closer to the thermal equilibrium (the z-axis) as long as the variable pulse is

Figure 3. Variations of T1 values (shown in red) for signals in the 1H NMR spectrum (600 MHz) of human urine spiked with phenylalanine to a final concentration of 7.75 mM. (a) Full spectrum recorded with inversion-recovery pulse sequence using modified FLIPSY sequence (inversionrecovery delay 20 s, relaxation delay 20 s, 64 transients, 6.2 kHz sweep width). Panels b-d are expansions: the T1 values were calculated for bins with a fixed width of 51 data points (0.015 ppm); only T1 values for signals with the intensity higher than 20% of the largest signal in the fully relaxed spectrum are included for clarity. Blue lines represent the 95% confidence intervals of T1 values. The boxes represent fitted T1 values with 95% confidence intervals as calculated for manually selected segments (see Table 1).

Figure 4. Experimental S/N values with standard deviations (n ) 3) as function of presaturation period. TSP: signal region δ [-0.03; 0.03] and noise region δ [-0.60; -0.90]. Aromatic protons: signal region δ [7.00; 8.00] and noise region δ [9.00; 10.00]. Squared bars: S/N for TSP in NOESYPRESAT spectrum. Black bars: S/N for aromatic protons in NOESYPRESAT spectrum. Dotted bars: S/N for TSP in FLIPSY spectrum. White bars: S/N for aromatic protons in FLIPSY spectrum.

below 90°. For a presaturation time of 1.0 s and acquisition time of 1.7 s, the pulse repetition period is approximately 2.8 s for NOESYPRESAT and 2.7 s for FLIPSY. The theoretical gain in S/N was calculated for aromatic proton signals of human urine as an

example (average T1 ≈ 3.6 s). For PR ) 2.7 s, the optimal flip angle according to eq 1 is 61.8°. The calculated S/N under these conditions (eq 4) is 12.8% higher for FLIPSY than for NOESYPRESAT (90°). These results are tabulated in Table 2. It is clear that an increase of T1 values and a decrease of the pulse repetition period emphasize the advantage of FLIPSY. However, too short pulse repetition periods may lead to an unwanted temperature increase. Variation in T1 values between samples due to matrix effects is potentially disadvantageous. Nevertheless, from a theoretical perspective, a variation of T1 by 10%, i.e., a change of T1 ) 3.6 to 3.2-4.0 s, would decrease S/N by at most 4.9%. It can also be noted that analysis of human plasma or other body fluids containing macromolecular compounds with very short T1 values will benefit from the FLIPSY sequence by elimination of relaxation that occurs during the mixing time in NOESYPRESAT, resulting in an increased S/N ratio. Experimental Increase in Relative S/N. To illustrate the above findings in practice, three human urine samples were analyzed using NOESYPRESAT and FLIPSY with fixed pulse repetition period either 2.8 and 2.7 s, respectively, or 3.8 and 3.7 s, respectively. The peaks influenced to the greatest extent Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

3369

Figure 5. 1H NMR spectra of human urine acquired with the NOESYPRESAT pulse sequence (A), original (ref 17) FLIPSY pulse sequence (B), and modified FLIPSY pulse sequence (C and D) illustrating superior baseline quality in the latter spectrum. The spectra were acquired with the following parameters: pulse repetition period 2.7 and 2.8 s for FLIPSY and NOESYPRESAT, respectively, FLIPSY excitation pulse 62.3°, 32 transients, 20.3 kHz sweep width. For spectrum D the presaturation power level was increased 5.9%. Table 2. Theoretical S/N Values Calculated for Different Combinations of T1 Values and Pulse Repetition Periods (PR) T1 (s)

PR (s)

flip angle (R)

S/N

% increase

3.6 3.6

2.7 2.8

61.8° (Ernst angle) 90° (not optimal)

36.4 32.3

12.8

3.6 3.6

3.7 3.8

69.0° (Ernst angle) 90° (not optimal)

35.7 33.4

6.9

2.0 2.0

2.7 2.8

75.0° (Ernst angle) 90° (not optimal)

46.7 45.0

3.7

are shown in Figure 6. As shown in Table 1, T1 values around 3.6 s primarily correspond to aromatic protons, but also dimethylamine (δ 2.72) and TSP have relative long T1 values leading to increased S/N in the FLIPSY. For the hippurate doublet (δ 7.64 and T1 ) 3.8 s), the measured increase in S/N in the FLIPSY experiment as compared to NOESYPRESAT (approximately 12%) corresponds well to the theoretical estimate (12.8%). Also, the slowly relaxing formic acid proton at δ 8.47 is seen to benefit greatly (43%) from the FLIPSY experiment, illustrating the potential of variable flip angles in metabolite profiling. The 3370 Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

Figure 6. Average histogram with standard deviations (n ) 3) showing percent increase in S/N for selected resonances using FLIPSY (optimized for T1 ) 3.6) vs NOESYPRESAT. White bars: PR ) 3.7 or 3.8 s. Gray bars: PR ) 2.7 or 2.8 s.

experiments with pulse repetition periods of 3.7 and 3.8 s show, as expected, a more moderate improvement in S/N. Classification Sensitivity. As a model study, 14 urine samples were spiked with 7.75 mM phenylalanine to resemble phenylketonuria urines, and 15 other urine samples were used as controls. 1H NMR spectra were acquired using the NOESYPRESAT sequence, the original FLIPSY sequence, and the FLIPSY se-

Table 3. Classification Sensitivity for O-PLS-DA Models NMR experiment

Q2Y

RMSEP

NOESYPRESAT (full resolution data) modified FLIPSY (full resolution data)

0.83 (0.88)b 0.88

0.161 (0.112)b 0.107

NOESYPRESAT (full resolution data; Phe signals only) modified FLIPSY (full resolution data; Phe signals only)

0.89

0.142

0.90

0.136

% decrease in RMSEP 33.4 (4.5)b

number of componentsa 1+1 (1 + 2)b 1+2 1+1

3.9

1+1

a The number of predictive plus orthogonal principal components. Value obtained by forcing an extra orthogonal component into the model.

b

quence modified with the homospoil pulse (Figure 1). The precision of a multivariate model can be assessed by a number of different measures. In accordance with the current norm for reporting precision of chemometric models in metabonomics23 the relative mean standard error of prediction (RMSEP) is used. This value is calculated in the validation step to report the precision of a model constructed with test set- or cross-validation and is used here to evaluate and compare the O-PLS-DA models. In addition, the cumulated Q2Y values are tabulated as a means of comparing the predictive ability of the calculated models and also to show the insufficiency of precision assessment using this parameter. Models were calculated on full resolution data on all 29 samples with a category variable reflecting the addition of phenylalanine (Table 3). Only data for NOESYPRESAT and the modified FLIPSY sequence are shown as the spectra obtained with the original FLIPSY sequence suffered from severe distortions of the baseline. The initial PCA models showed one outlier that was excluded due to a very high concentration of dimethylamine (δ 2.72). Also, spectral segments with the two creatinine resonances (δ 3.05 and δ 4.06) were excluded from the analysis as these signals exhibited large chemical shift variations across the samples. Following these changes of the original data set, O-PLS-DA on full resolution data introduced a decrease of the RMSEP of 33.4%. However, the autofit routine determined only one orthogonal component for the NOESYPRESAT data but two for the modified FLIPSY data. Therefore, in order to create a more even comparison between the two pulse sequences, another orthogonal component was calculated manually for the NOESYPRESAT data, resulting in a decrease of RMSEP of 4.5%. Taking only the phenylalanine resonances into account (variable selection), a decrease of 3.9% was obtained. These results suggest that the FLIPSY approach (23) Danzer, K.; Otto, M.; Currie, L. A. Pure Appl. Chem. 2004, 76, 1215-1225.

does improve data quality as reflected in the above data analysis. Score and loading plots corresponding to data in Table 3 are included as Supporting Information. Although differences between score plots obtained with the two pulse sequences may be difficult to recognize visually, RMSEP values illustrate the superiority of data obtained with the modified FLIPSY sequence. Thus, we recommend the modified FLIPSY pulse sequence as a supplement to the NOESYPRESAT approach, as this experiment has the capability to enhance discrimination when relaxation parameters of the discriminating resonances are available. In addition, when the discriminating resonances are difficult to separate from the noise level in a NOESYPRESAT spectrum, we expect the reduction of the RMSEP to be considerably larger when the same resonances are defined with a better S/N ratio in a FLIPSY spectrum acquired using the Ernst angle for this particular resonance. CONCLUSIONS So far, data acquisition in metabolite profiling has been routinely performed using the NOESYPRESAT pulse sequence. This can result in suboptimal data acquisition efficiency due to the shortcomings of NOESYPRESAT with regard to application of variable flip angles. By contrast, the FLIPSY pulse sequence augmented with a homospoil pulse allows variable read pulse and thus maximization of S/N ratios by use of the Ernst angle, while providing spectra of equivalent quality in terms of water peak suppression and baseline quality. Thus, the application of variable flip angles is proposed as a supplementary technique in 1H NMRbased metabonomics. While the advantage of optimization of S/N ratios in FLIPSY spectra for increased classification sensitivity was in this work demonstrated with model urines, additional advantages are expected for other types of metabonomic samples. However, the advantages rely upon the knowledge of T1 relaxation times, which must be measured or estimated prior to the FLIPSY experiment. ACKNOWLEDGMENT A.D.M. acknowledges funding from the EU FP6 MolPAGE project. NMR equipment in the Copenhagen laboratory was purchased via a Grant from “Apotekerfonden af 1991”. SUPPORTING INFORMATION AVAILABLE 1H NMR spectrum of simulated phenylketonuria urine and O-PLS-DA score and loading plots for the data presented in Table 3. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review February 12, 2008.

December

18,

2007.

Accepted

AC702563U

Analytical Chemistry, Vol. 80, No. 9, May 1, 2008

3371