Anal. Chem. 2008, 80, 9288–9297
Urinary Metabolite Quantification Employing 2D NMR Spectroscopy Wolfram Gronwald,*,† Matthias S. Klein,† Hannelore Kaspar,† Stephan R. Fagerer,† Nadine Nu¨rnberger,† Katja Dettmer,† Thomas Bertsch,‡ and Peter J. Oefner† Institute of Functional Genomics, University of Regensburg, Josef-Engert-Strasse 9, 93053 Regensburg, Germany, and Institute of Clinical Chemistry, Laboratory Medicine and Transfusion Medicine, Klinikum Nuernberg, Prof. Ernst-Nathan-Strasse 1, 90419 Nuernberg, Germany Two-dimensional (2D) nuclear magnetic resonance (NMR) spectroscopy is a fairly novel method for the quantification of metabolites in biological fluids and tissue extracts. We show in this contribution that, compared to 1D 1H spectra, superior quantification of human urinary metabolites is obtained from 2D 1H-13C heteronuclear single-quantum correlation (HSQC) spectra measured at natural abundance. This was accomplished by the generation of separate calibration curves for the different 2D HSQC signals of each metabolite. Lower limits of detection were in the low to mid micromolar range and were generally the lower the greater the number of methyl groups contained in an analyte. The quantitative 2D NMR data obtained for a selected set of 17 urinary metabolites were compared to those obtained independently by means of gas chromatography-mass spectrometry and liquid chromatography-mass spectrometry of amino acids and hippurate as well as enzymatic and colorimetric measurements of creatinine. As a typical application, 2D-NMR was used for the investigation of urine from patients with inborn errors of metabolism. Nuclear magnetic resonance (NMR) is a powerful tool for metabolite identification and quantification in biological fluids and tissue extracts. Signal volumes scale linearly with concentration and, in most cases, are independent of the chemical properties of the investigated molecules. Additional advantages are that no prior chemical derivatization and only very limited sample pretreatment are required. One major application of metabolomics is the analysis of biological fluids, such as human urine. Due to the need to maintain homeostasis, human urine is very complex in composition containing hundreds to thousands of different compounds even in healthy individuals.1 This in turn leads to an even larger number of signals in the corresponding NMR spectra that results in considerable signal overlap especially when only one-dimensional 1 H spectra are acquired. As a consequence, the high signal number hampers accurate metabolite identification and quantifica* To whom correspondence should be addressed. E-mail: wolfram.gronwald@ klinik.uni-regensburg.de. † University of Regensburg. ‡ Klinikum Nuernberg. (1) Holmes, E.; Foxall, P. J. D.; Spraul, M.; Farrant, R. D.; Nicholson, J. K.; Lindon, J. C. J. Pharm. Biomed. Anal. 1997, 15, 1647–59.
9288
Analytical Chemistry, Vol. 80, No. 23, December 1, 2008
tion considerably. A mathematical solution to this problem is to fit overlapped experimental signals to signals modeled from pure compound spectra.2 Experimentally, spreading the signals over two or more dimensions will also reduce overlap. The advantages of multidimensional NMR have been recognized in several metabolomic studies.1,3-5 However, in most instances, multidimensional spectra have been acquired solely for metabolite identification and not for quantification. Only recently, the superior resolution of 2D spectra has been utilized for the quantification of metabolites.6-8 The long acquisition times required in the past for 2D spectra had rendered them impractical for high-throughput metabolomic studies. Further, 2D cross-peak intensities depend on a larger number of different factors, such as structuredependent J-coupling values and relaxation times, mixing times, evolution times, and uneven excitation profiles, which must be taken into account when they are used for quantification. Two-dimensional 1H-13C heteronuclear single-quantum correlation (HSQC) spectra offer the advantage of large signal dispersion in the indirect carbon dimension. Recently, 2D 1H-13C HSQC spectra measured at natural abundance were applied to metabolite quantification in synthetic mixtures and plant extracts.7 Here we have employed 2D 1H-13C HSQC spectra to develop an improved procedure for the reliable and accurate quantification of urinary metabolites. We show that, in contrast to 1D NMR spectroscopy, separate calibration curves for the individual signals of a compound are required for its precise and accurate quantification. This is due to factors, such as nonuniform relaxation times and varying transfer efficiencies of the insensitive nuclei enhanced by polarization transfer (INEPT) steps of the HSQC spectra. In addition, we show the importance of taking effects of the biological matrix into account. Finally, we have validated the quantitative NMR data obtained for a selected group of urinary compounds (2) Weljie, A. M.; Newton, J.; Mercier, P.; Carlson, E.; Slupsky, C. M. Anal. Chem. 2006, 78, 4430–42. (3) Adosraku, R. K.; Ghoi, G. T. Y.; Constantiou-Koktos, V.; Anderson, M. M.; Gibbons, W. A. J. Lipid Res. 1994, 35, 1925–31. (4) Tang, H.; Wang, Y.; Nicholson, J. K.; Lindon, J. C. Anal. Biochem. 2003, 325, 260–72. (5) Zheng, M.; Lu, P.; Liu, Y.; Pease, J.; Usuka, J.; Liao, G.; Peltz, G. Bioinformatics 2007. (6) Hu, F.; Furihata, K.; Kato, Y.; Tanokura, M. J. Agric. Food Chem. 2007, 4307–11. (7) Lewis, I. A.; Schommer, S. C.; Hodis, B.; Robb, K. A.; Tonelli, M.; Westler, W. M.; Sussman, M. R.; Markley, J. L. Anal. Chem. 2007, 79, 9385–90. (8) Shanaiah, N.; Desilva, M. A.; Gowda, G. A. N.; Raftery, M. A.; Hainline, B. E.; Raftery, D. Proc. Natl. Acad. Sci. U. S. A. 2007, 104, 11540–4. 10.1021/ac801627c CCC: $40.75 2008 American Chemical Society Published on Web 10/29/2008
by means of gas chromatography-mass spectrometry (GC–MS)9 and liquid chromatography-tandem mass spectrometry (LC-MS/ MS), except for creatinine, the urinary levels of which were determined enzymatically and colorimetrically. EXPERIMENTAL SECTION Sample Preparation. Four hundred microliters of urine, which had been collected from volunteers or provided by the Zentrum fu¨r Stoffwechseldiagnostik Reutlingen GmbH (Reutlingen, Germany), were mixed with 200 µL of phosphate buffer, pH 7.4, and 50 µL of D2O containing 29.02 mM 3-trimethylsilyl-2,2,3,3tetradeuteropropionate (TSP) as internal standard.10 For the analysis of standard mixtures of known concentration, a certified mixture of 17 amino acids in 0.1 M HCl was purchased from National Institute of Standards and Technology (NIST) (Gaithersburg, MD). For the NMR measurements, 400 µL of the undiluted, 1:5 and 1:10 dilutions of the NIST standard with water were treated as described above. NMR Spectroscopy. NMR experiments were carried out on a Bruker Avance III 600 MHz spectrometer employing a tripleresonance (1H, 13C 31P, 2H lock) cryogenic probe equipped with z-gradients and an automatic sample changer. For each sample, the probe was automatically locked, tuned, matched, and shimmed. In each case, a standard shim file specifically optimized for urine samples was used as a starting point for the automated shimming procedure. All spectra were measured at 298 K, and each sample was allowed to equilibrate for 300 s in the magnet before measurement. The temperature unit was calibrated using a deuterated methanol sample. 1D 1H and 2D 1H-13C HSQC spectra of each sample were automatically collected using the Bruker automated acquisition suite ICON-NMR. All spectra were acquired without spinning. 1D 1H NMR spectra were obtained using a 1D NOESY pulse sequence with presaturation during relaxation and mixing time for water suppression and additional spoil gradients. For each spectrum, a total of 128 scans was collected into 64k data points using a relaxation delay of 4 s, an acquisition time of 2.66 s, and a mixing time of 0.01 s. The spectral width was set to 20 ppm for each 1D spectrum and four dummy scans were applied prior to each measurement. Spectra were automatically Fourier transformed and phase corrected, applying a line broadening of 0.3 Hz and zero filling to 128k points. A flat baseline was obtained by using the “baseopt” option of TopSpin2.1, which performs a correction of the first points of the FID. For the 2D 1H-13C HSQC spectra, water suppression was achieved using presaturation during the relaxation delay. For each 2D spectrum, 2048 × 128 data points were collected using 8 scans per increment, a relaxation delay of 3 s, an acquisition time of 0.14 s, and 16 dummy scans. This resulted in a total acquisition time of 56 min per spectrum. The spectral widths were set to 12 and 165 ppm in the proton and carbon dimensions, respectively. For initial assignment of the metabolites, one high-resolution 2D 1 H-13C HSQC spectrum with 2048 × 512 data points and 64 scans per increment was collected. Initial assignments were further validated using long-range proton-carbon couplings obtained from a high-resolution 2D 1H-13C heteronuclear multiple bond (9) Kaspar, H.; Dettmer, K.; Gronwald, W.; Oefner, P. J. J. Chromatogr., B 2008, 870, 222–32. (10) Beckonert, O.; Keun, H. C.; Ebbels, T. M. D.; Bundy, J.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Nat. Protocols 2007, 2, 2692–702.
correlation (HMBC) spectrum. For this spectrum, 2048 × 512 data points and 72 scans were acquired. The total acquisition times for these HSQC and HMBC spectra, which were recorded only once, were 15 h 25 min and 17 h 42 min, respectively. 2D spectra were semiautomatically processed employing a 90° shifted squared sine-bell window function in both dimensions. For increased resolution in the indirect carbon dimension, the number of data points was doubled prior to Fourier transform using complex forward linear prediction. All 2D spectra were manually phase corrected, and a polynomial baseline correction was applied excluding the region around the water artifact. All 1D and 2D spectra were chemical shift referenced relative to the TSP signal. Initial Metabolite Assignment. First, a manual metabolite assignment was performed on a representative high-resolution 2D 1 H-13C HSQC spectrum aided by the corresponding 1D 1H and 2D 1H-13C HMBC spectra. AMIX3.8.1 (Bruker BioSpin GmbH) was used starting with manual picking of peaks in the high-resolution HSQC spectrum. To assign the signals, the spectrum was manually overlaid with reference spectra of pure compounds. The reference spectra were taken from the commercially available Bruker Biofluid Reference Compound Database BBIOREFCODE 2-0-0 that contains reference 1D 1H, 2D 1H-13C HSQC, and 1H-13C HMBC spectra of metabolite standards measured under a variety of different experimental conditions (e.g., pH values and solvents). For 1-methylhistidine, reference spectra information was taken from the Biological Magnetic Resonance Data Bank (BMRB).11 All assigned metabolite signals were further validated based on longrange 1H-13C couplings obtained from the high-resolution HMBC spectrum. A summary of the obtained chemical shift data is provided in Supporting Information Table S2. The obtained chemical shift information of each identified metabolite signal was stored in a so-called knowledge base to guide the following metabolite quantification process. To define the regions where the individual metabolite signals in a series of measured spectra are expected, individual chemical shift ranges were specified for each metabolite signal by manually analyzing interspectra chemical shift variations in a subset of the measured spectra. Depending on the analyzed type of spectrum, individually optimized chemical shift ranges were determined. In addition, the knowledge base describes the individual signals of a compound in terms of multiplet patterns, couplings, relative intensities, and masses. Coupling values with appropriate error bounds were obtained by analyzing the pure reference compound spectra together with the experimentally measured spectra of our series. Note that multiplet patterns caused by J-couplings are only observable in the 1D spectra. In addition, the knowledge base allows the exclusion of overlapping signals located in crowded regions of the real urine spectra from the following quantification process. Overlapping signals were determined by manually analyzing a subset of the measured spectra. In Supporting Information Table S2, the metabolite signals used for quantification are listed. The knowledge base is basically a simple ASCII text file containing the above information. Automated Analysis. The next steps of the analysis were performed with the Analytical Profiler module of AMIX3.8.1. The (11) Seavey, B. R.; Farr, E. A.; Westler, W. M.; Markley, J. L. J. Biomol. NMR 1991, 1, 217–36.
Analytical Chemistry, Vol. 80, No. 23, December 1, 2008
9289
Table 1. Limits of Detection (LOD), Lower Limits of Quantification (LLOQ), Number of Signals Used for Metabolite Assignment and Quantification, Relative Standard Deviations of Quantitative Measurements, and Average Concentrations, Standard Deviations, and Ranges for Selected Urinary Metabolites
metabolite acetate alanine 3-aminoisobutyrate argininec betaine citrate creatinine ethanolamine glutamine glycine hippurate histidine lysine 1-methylhistidine 3-methylhistidine taurine trimethyl-N-oxide
number LOD LLOQ of RSDb (µM) (µM) signalsa (%) 78 39 78 312 20 78 20 78 312 78 312 78 39 78 78 20 20
78 78 312 312 78 312 78 312 312 156 312 156 312 312 78 312 39
1(1) 2(1) 3(1) 4(2) 2(2) 2(2) 2(2) 2(2) 3(2) 1(1) 4(3) 4(1) 5(4) 3(3) 5(3) 2(2) 1(1)
urinary conc (µM) mean ± SD (range)
9.6 367 ± 284 (88-931) 20.6 194 ± 84 (86-465) 6.9 1505 ± 437 (898-2037) 19.6 176 ± 105 (37-637) 14.8 1421 ± 593 (152-2673) 2.0 7069 ± 2748 (3069-13593) 14.8 363 ± 47 (176-496) 19.4 431 ± 121 (314-666) 4.8 953 ± 472 (355-1976) 4.6 2140 ± 1434 (372-5392) 10.4 626 ± 264 (165-1297) 13.4 544 ± 159 (90-832) 14.0 215 ± 109 (78-569) 13.1 450 ± 393 (80-1640) 3.7 868 ± 502 (224-2571) 3.4 650 ± 745 (61-2419)
a Number of groups (e.g., CH, CH2, CH3) used for metabolite identification and quantification (the latter in parentheses). b Average of 6 urine triplicates. c Arginine could not be detected in any sample.
1D 1H and 2D 1H-13C HSQC spectra corresponding to the series of measured urine samples were automatically peak-picked using an automated noise level calculation. For this purpose, regions that contained no signals in all spectra were manually defined. Using the information stored in the knowledge base, metabolite signals were identified and integrated. For 1D spectra, multiplet information stored in the knowledge base was used in the integration process to ensure that only the desired signals within the specified chemical shift ranges were integrated. For 2D spectra, multiplet information was not applicable. Here, information from the reference compound spectra database, such as the expected number of signals in a specific region, were used instead. Next, relative integrals with respect to the reference TSP signal were calculated for the individual metabolite signals. In this process, the number of atoms contributing to a signal was taken into account. The use of relative integrals automatically corrects for machine-dependent sensitivity variations between experiments and has the additional advantage that calibration curves have to be determined only once. Finally, in-house routines were used for absolute quantification, employing individual calibration curves for each metabolite signal. For this purpose, a 10 mM standard stock solution was serially diluted to yield final concentrations of 10 000, 5000, 2500, 1250, 625, 312.5, 156.2, 78.1, 39, and 19.5 µM, respectively, and the corresponding spectra were acquired. For creatinine, the concentration range was extended to 15 mM. Table 1 shows the number of signals that were used for compound identification. It equals the number of measured calibration curves. However, due to overlap of signals even in the 2D spectra, only a subset of these signals could be used for quantification. Due to metabolites inconsistently present in urine, in a few rare cases additional overlap was observed leading to drastically increased integral values for certain metabolite signals. As a consequence, outlier signals that deviated from the median of a compound by more than 50% were automatically excluded from further analysis. 9290
Analytical Chemistry, Vol. 80, No. 23, December 1, 2008
Following absolute quantification of individual signals, the corresponding averages for each metabolite were calculated. The limit of detection (LOD), which was defined as a signalto-noise ratio of 3.5 for the strongest signal of a compound, was obtained from the calibration samples (Table 1). For calculating the lower limit of quantification (LLOQ), the lowest three points of each calibration curve were measured in triplicate and the corresponding relative standard deviations (RSDs) were determined. Following the recommendation of the FDA guide for Bioanalytical Method Validation,12 the LLOQ was defined as the concentration value that could be determined with an RSD < 20%. The LODs and LLOQs obtained are a function of the NMR acquisition time. Therefore, lower values may be obtained by increasing the acquisition time. GC–MS Measurements. An Agilent model 6890 GC (Agilent, Palo Alto, CA), which was equipped with a model 5975 Inert XL mass selective detector, a PTV injector (Gerstel, Muehlheim, Germany), and a MPS-2 Prepstation sample robot, was used for the quantitative analysis of amino acids in urine samples. Analysis was performed as recently described.9 Briefly, amino acids were derivatized with propyl chloroformate in diluted urine, followed by extraction of the derivatives with isooctane and injection of a 2.5-µL aliquot of the organic extract into the PTV. The column used for GC analysis was a ZB-AAA (Phenomenex Inc., Torrence, CA), 15 m × 0.25-mm i.d., 0.1-µm film thickness. LC-MS/MS Measurements. Amino acid analysis by LC-MS/ MS was performed using the iTRAQ kit for labeling free amino acids in physiological fluids (Applied Biosystems, Framingham, MA). An aliquot of 40 µL of urine and 10 µL of 10% sulfosalicylic acid (containing 400 pmol/µL of norleucine) was transferred to a 96-well plate and mixed for 10 s. Then the plate was centrifuged for 2 min at 4000g. Afterward, 10 µL of supernatant was mixed with 40 µL of labeling buffer (0.45 M borate buffer, pH 8.5 containing 20 pmol/µL of norvaline). Ten microliters of the diluted supernatant was taken, mixed with 5 µL of iTRAQ reagent 115, and incubated at room temperature for 60 min. Then 5 µL of 1.2% hydroxylamine solution was added to each well, and the samples were dried in a vacuum concentrator (CombiDancer, Zinsser Analytic, Frankfurt, Germany) and reconstituted with 32 µL of iTRAQ reagent 114-labeled amino acids. Chromatographic separation was achieved using an Agilent 1200 HPLC system. The 150 × 4.6 mm i.d, C18, 5-µm HPLC column (Applied Biosystems) was kept in a column oven at 50 °C. Separation was carried out with solvents A (0.1% formic acid and 0.01% heptafluorobutyric acid in HPLC water) and B (the same as A, but in acetonitrile), respectively. The column was equilibrated in 98% A and the gradient was 98-72% A over 10 min, 72-0% over 0.1 min, hold at 100% B for 5.9 min. A flow rate of 800 µL min-1 was used, and the injected sample volume was 2 µL. The 4000 QTRAP mass spectrometer with turbo ion spray (Applied Biosystems) was operated in positive mode using the following parameters: ion spray voltage (IS) 5000 V; auxiliary gas temperature (TEM) 450 °C; curtain gas (CUR), nebulizer gas (GS1), and auxiliary gas (GS2) were set at 20, 40, and 40 arbitrary units, respectively; collision gas (CAD) was set at medium. The entrance potential (EP), declustering potential (DP), collision energy (CE), and (12) Guidance for Industry. Bioanalytical Method Validation, Center for Drug Evaluation and Research, U.S. Department of Health and Human Services: Rockville, MD, 2001.
collision cell exit potential (CXP) were set at 10, 50, 30, and 12 V, respectively. Quantitative determination was performed by multiple reaction monitoring using one transition each for the analyte and the corresponding internal standard from the precursor ion to the reporter ions (m/z 115 for the analyte and m/z 114 for the internal standard). Processing of the mass spectra was performed using Analyst 1.4.2. The analyte concentrations were calculated from the response (analyte area divided by internal standard area) using individual calibration curves for each amino acid. The calibration curves were obtained by measuring at least 10 concentrations between 2 and 750 µM (blank samples not included) of all compounds except ethanolamine and taurine (4 concentrations between 50 and 500 µM). Enzymatic and Colorimetric Determination of Creatinine. Creatinine was measured enzymatically with the creatinine-PAP method (Biomed, Oberschleissheim, Germany) using the Duacal Multicalibrator from the same manufacturer. The lower limit of detection was 18 µM and the linear measuring range 796-53 040 µM. Creatinine was also measured with the kinetic Jaffe method (Olympus, Hamburg, Germany) using the Olympus Multicalibrator. The lower limit of detection was 0.1 µM and the linear measuring range 88-35 360 µM. Both assays were performed on the AU 2700 Clinical Chemistry System (Olympus, Hamburg, Germany). RESULTS AND DISCUSSION Calibration. A total of 17 urinary metabolites, including several amino acids, betaine, citrate, creatinine, hippurate, and taurine (Table 1), were quantified in both 1D and 2D spectra relative to the intensity of the TSP reference signal. For 1D 1H spectra, calibration curves for all compounds and their respective signals should be very similar provided that a sufficiently long relaxation delay had been applied. To verify this expectation, calibration curves were generated for all nonoverlapping signals of all quantified metabolites at multiple concentration values. A typical example obtained from 1D 1H spectra is shown for creatinine in Figure 1A. The curves were measured for the two signals corresponding to the creatinine methylene (H2A/H2B) and methyl group (H6A/H6B/H6C), respectively. As can be seen, both curves are almost identical and show a linear behavior with r2 values of 0.999 when fitted to a straight line. In the case of 2D spectra, multidimensional peak intensities are influenced by a number of factors such as nonuniform relaxation, different transfer efficiencies of the INEPT step, and uneven excitation profiles. This results in calibration curves of markedly different slopes for the two signals of creatinine (Figure 1B). Similar results were obtained for most other metabolites (data not shown). Consequently, if 2D spectra are used for quantification, it is necessary to calibrate each signal individually to obtain in the next step a reliable average over all signals used for quantification of a compound. Limits of Detection and Lower Limits of Quantification. As described in the Experimental Section, LODs for the 2D 1 H-13C HSQC spectra were calculated from calibration standards (Table 1). LODs ranged from 20 to 312 µM at an S/N of 3.5. The lowest LODs were obtained for analytes containing multiple methyl groups, the protons of which give rise in most instances to one sharp singlet signal. The LLOQ values (Table 1) were determined by measuring the lowest concentration points of the calibration curves in
triplicate and defining the LLOQ value of each compound as the minimal concentration yielding a RSD
