Quantitative 1H NMR Spectroscopy of Blood Plasma Metabolites

Quantitative Analysis of Blood Plasma Metabolites Using Isotope Enhanced NMR Methods. G. A. Nagana .... NMR spectroscopy – a modern analytical tool ...
7 downloads 14 Views 93KB Size
Anal. Chem. 2003, 75, 2100-2104

Quantitative 1H NMR Spectroscopy of Blood Plasma Metabolites Robin A. de Graaf*,† and Kevin L. Behar‡

Magnetic Resonance Research Center, Departments of Diagnostic Radiology and Psychiatry, Yale University School of Medicine, New Haven, Connecticut 06520

The absolute quantification of blood plasma metabolites by proton NMR spectroscopy is complicated by the presence of a baseline and broad resonances originating from serum macromolecules and lipoproteins. A method for spectral simplification of proton NMR spectra of blood plasma is presented. Serum macromolecules and metabolites are completely separated by utilizing the large difference in translational diffusion coefficients in combination with diffusion-sensitized proton NMR spectroscopy. The concentration of blood plasma metabolites can be quantified by using formate as an internal concentration reference. The results are compared with those obtained with ultrafiltration, a traditional method for separating macromolecules and metabolites, and demonstrate an excellent correlation between the two methods. The general nature of diffusion-sensitized NMR spectroscopy allows application on a wide range of biological fluids. Blood plasma is a complex mixture of lipoprotein particles, proteins, low molecular weight metabolites, and electrolytes. The biochemical analysis of blood plasma can provide a wealth of information in a range of diseases, including diabetes1, cancer,2,3 and inborn errors of metabolism.4 Proton nuclear magnetic resonance (NMR) spectroscopy has been widely used to study human and animal blood plasma and other biological fluids including urine, bile, and cerebrospinal fluid3. However, quantitative NMR spectroscopic analysis of blood plasma is often hampered by the presence of broad resonances arising from lipoproteins and macromolecules such as albumin. As a consequence, these high molecular weight molecules are often removed prior to 1H NMR by traditional physicochemical methods such as ultrafiltration, chromatography, extraction, or precipitation.5 Many of these methods are time-consuming, may result in (selective) * Corresponding author. Tel: (203) 785-6203. Fax: (203) 785-6643. E-mail: [email protected]. † Department of Diagnostic Radiology. ‡ Department of Psychiatry. (1) Nicholson, J. K.; O’Flynn, M. P.; Sadler, P. J.; Macleod, A. F.; Juul, S. M.; Sonksen, P. H. Biochem. J. 1984, 217, 365-375. (2) Fossel, E. T.; Carr, J. M.; McDonagh, J. N. Engl. J. Med. 1986, 315, 13691376. (3) Nicholson, J. K.; Wilson, I. D. Prog. NMR Spectrosc. 1989, 21, 449-501. (4) Wevers, R. A.; Engelke, U.; Heerschap, A. Clin. Chem. 1994, 40, 12451250. (5) Daykin, C. A.; Foxall, P. J.; Connor, S. C.; Lindon, J. C.; Nicholson, J. K. Anal. Biochem. 2002, 304, 220-230.

2100 Analytical Chemistry, Vol. 75, No. 9, May 1, 2003

loss of metabolites, can introduce contaminants, or give incomplete removal of macromolecules.5 Furthermore, a physicochemical separation of low and high molecular weight molecules always precludes the study of a variety of possible interactions, such as metal complexation, micellar compartmentation, and chemical and physical exchange processes.3 As an alternative to blood plasma sample preparation by traditional physicochemical methods, the NMR properties of the various nuclear spin systems can be utilized to achieve spectral simplification and absolute or relative quantification. The NMR parameters relating to longitudinal T1 and transverse T2 relaxation are natural candidates for spectral simplification, as has been demonstrated for T2 relaxation with a Carr-Purcell-MeiboomGill (CPMG) sequence.6,7 However, physical properties not directly related to NMR, such as translational diffusion, can also be utilized because any NMR experiment is readily sensitized to translational diffusion by incorporating pairs of bipolar magnetic field gradients, as first proposed by Stejskal and Tanner.8 Here we evaluate the utility of diffusion NMR to separate molecules in blood plasma based on translational mobility. Furthermore, a comparison between the macromolecular baseline and absolute metabolite concentrations obtained by diffusion NMR and ultrafiltration will be made to demonstrate the quantitative nature of spectral simplification by diffusion NMR. MATERIALS AND METHODS Animal Preparation. Three male Sprague-Dawley rats (170 ( 10 g, mean ( SD) were prepared in accordance to the guidelines established by the Yale Animal Care and Use Committee. Following an overnight fast (12-16 h), the animals were tracheotomized and ventilated with a mixture of 70% nitrous oxide and 28.5% oxygen under 1.5% halothane anesthesia. A femoral artery was cannulated for monitoring of blood gases (pO2, pCO2), pH, and blood pressure. Physiological variables were maintained within normal limits by small adjustments in ventilation (pCO2 ) 33-45 mmHg; pO2 > 120 mmHg; pH ) 7.33-7.48; blood pressure 90-110 mmHg). A femoral vein was cannulated for infusion of [2-13C]-acetate (Cambridge Isotopes, Andover, MA). After all surgery was completed, anesthesia was maintained by 0.3-0.8% halothane in combination with 70% nitrous oxide. The core temperature was measured with a rectal thermosensor and was maintained at 37 ( 1 °C by means of a heated water pad. Blood (6) Carr, H. Y.; Purcell, E. M. Phys. Rev. 1954, 94, 630-638. (7) Meiboom, S.; Gill, D. Rev. Sci. Instrum. 1958, 29, 688-691. (8) Stejskal, E. O.; Tanner, J. E. J. Chem. Phys. 1965, 42, 288-292. 10.1021/ac020782+ CCC: $25.00

© 2003 American Chemical Society Published on Web 03/28/2003

samples of 250 µL were taken prior to [2-13C]-acetate infusion and at every following 20 min after the start of infusion. Blood plasma was obtained by centrifugation of whole blood at 8000 rpm for 2 min and the supernatant frozen in liquid nitrogen. NMR Spectroscopy. All NMR experiments were performed on a Bruker Avance spectrometer (Bruker Instruments, Billerica, MA) operating at 500.13 MHz for 1H and equipped with a 5-mm triple resonance probe incorporating triple-axis gradient coils. The maximum magnetic field gradient strengths were 500 mT/m for the X and Y directions and 700 mT/m for the Z direction. The NMR sequence is based on a standard stimulated-echo method with an echo time TE of 8 ms. Diffusion-sensitization was achieved by placing balanced trapezoidal magnetic field gradients in either TE/2 period. The amount of signal loss due to diffusion between the magnetic field gradients is given by

ln(S(G)/S(0)) ) -bD, where b ) γ2G2[δ2(∆ - δ/3) + 3/30 + δ2/6] (1)

where D represents the translational diffusion coefficient, γ is the gyromagnetic ratio, and G is the magnetic field gradient strength. The magnetic field gradient length δ, ramp length , and separation ∆ were set to 2.8, 0.7, and 75 ms, respectively. To achieve sufficient suppression of small metabolites with high diffusion coefficients, the magnetic field gradient strength was adjusted. Typically, the diffusion factor b was set to 10 000 s/mm2. All experiments were performed at 298 K. Water suppression was achieved with a chemical shift selective excitation technique9 executed with adiabatic rf pulses.10 To reduce the viscosity, 250 µL of plasma was diluted with 250 µL of concentration reference containing 5 mM formic acid and 1 mM sodium 3-trimethylsilyl[2,2,3,3-D4]-propionate (TSP-d4, Sigma, St. Louis, MO) in deuterium oxide (Cambridge Isotopes, Andover, MA). The 100-µL plasma samples obtained during the [2-13C]-acetate infusions were diluted to 500 µL. All NMR spectra were acquired with 32 averages and a repetition time TR ) 10 000 ms. Postacquisition processing included zero-filling to 64K points followed by apodization (0.5Hz line broadening) and Fourier transformation. Ultrafiltration was performed using Nanosep centrifugal devices (Pall Life Sciences, Ann Arbor, MI) with a 10 000 molecular weight cutoff. Blood plasma (100 µL) was diluted with 400 µL of concentration reference containing 5 mM formic acid and 1 mM TSP-d4 in deuterium oxide and centrifuged at 14000g and 277 K for up to 3 h. To reduce glycerol (membrane preservative) contamination of the filtrate to 100-fold, leaving only resonances from macromolecules. When the macromolecular baseline spectrum is scaled up by ∼20%, to account for signal loss due to diffusion, and subtracted from the total spectrum (Figure 1A), the difference spectrum (Figure 1C) contains only resonances from low molecular weight metabolites. Figure 2 shows a comparison between the macromolecular baselines obtained by ultrafiltration and diffusion NMR. Both 1H NMR spectra display the same number of macromolecular resonances at approximately the same relative intensities. Note that, even after three wash cycles (of 2 h each), the ultrafiltration method did not result in complete separation of all low molecular weight metabolites from the macromolecules as can be judged from the sharp resonances (arrows). Diffusion NMR is capable of suppressing the metabolite resonances by >100-fold in less than 5 min. Figure 3 shows a comparison between the absolute concentrations of blood plasma metabolites after the macromolecular baseline has been removed by ultrafiltration or diffusion NMR. To maintain the experimental conditions as similar as possible, the samples obtained with ultrafiltration were also measured with the diffusion NMR sequence (low diffusion sensitization). Most metabolites, including acetate, alanine, creatinine, β-hydroxybutyrate, glucose, lactate, and valine, showed an excellent correlation between the two methods. Regression analysis gave a linear correlation (slope 0.978, intercept 0.054 mM) with a correlation coefficient of 0.9989. However, noticeable exceptions are glycerol and citrate. When the ultrafiltration membranes were not prewashed at least three times, the resulting metabolite spectrum would be contaminated by glycerol (15, 2.0, 0.3, and 0.04 mM after zero to three washes, respectively). Since diffusion NMR did not require any sample preparation, glycerol resonances were never observed. On the other hand, the strongly coupled citrate multiplet at 2.6 ppm was consistently observed in all plasma samples studied with diffusion NMR. However, in ∼50% of the plasma samples prepared by ultrafiltration, the citrate resonances were not detected. Even though all samples contained two concentration references, formate and TSP-d4, only formate was used for absolute quantification, since TSP is partially invisible due to interactions with plasma proteins, in particular albumin.11 DISCUSSION Here we have presented a method for quantitative 1H NMR spectroscopy of blood plasma. The technique utilizes the difference in translational diffusion coefficients between metabolites and macromolecules to remove the macromolecular baseline without affecting the metabolite resonances, thereby allowing quantitative 1H NMR spectroscopy of blood plasma metabolites or plasma macromolecules and lipoproteins. (11) Kriat, M.; Confort-Gouny, S.; Vion-Dury, J.; Sciaky, M.; Viout, P.; Cozzone, P. J. NMR Biomed. 1992, 5, 179-184.

Analytical Chemistry, Vol. 75, No. 9, May 1, 2003

2101

Figure 1. (A) 1H NMR spectrum of blood plasma acquired with a low sensitivity toward diffusion (b ) 4.1 s/mm2), overlaid with a spectrum obtained with high diffusion sensitivity (b ) 10 000 s/mm2, gray). (B) 1H NMR spectrum of blood plasma acquired with a high sensitivity toward diffusion (b ) 10 000 s/mm2). The macromolecule spectra in (A) and (B) are identical. (C) Difference spectrum between (A) and (B). Abbreviations are given for acetate (Ace), acetoacetate (AcA), aceton (Ac), alanine (Ala), β-hydroxybutyrate (BHB), citrate (Cit), creatinine (Crn), glucose (Glc), isoleucine (Ile), lactate (Lac), leucine (Leu), and valine (Val).

The application of diffusion NMR for spectral simplification of blood plasma and other complex mixtures has been described previously,12,13 but this report is the first that establishes the quantitative nature of diffusion NMR to measure blood plasma metabolites. There are several advantages of the presented diffusion NMR approach over more traditional physicochemical separation methods. First, diffusion NMR does not require any significant sample preparation since the metabolite-macromolecule separation is performed during the NMR experiment. This can lead to considerable time savings. In the case of ultrafiltration, sample preparation could take up to 6 h. In contrast, the diffusion (12) Stilbs, P. Anal. Chem. 1981, 53, 2135-2137. (13) Liu, M.; Nicholson, J. K.; Lindon, J. C. Anal. Chem. 1996, 68, 3370-3376.

2102 Analytical Chemistry, Vol. 75, No. 9, May 1, 2003

NMR approach requires the acquisition of one additional NMR spectrum (with a high diffusion sensitization), which was typically obtained within several minutes. Second, the minimal sample preparation required for diffusion NMR effectively eliminates contaminants that may be introduced during traditional plasma deproteinization methods, e.g., glycerol in the case of ultrafiltration. It can also eliminate selective signal loss, as was occasionally observed for citrate. Third, intact blood plasma samples allow the measurement of many important physicochemical interactions such as compartmentation and exchange. Furthermore, the diffusion technique also allows the direct study of serum macromolecules and lipoproteins, the exact composition of which has been associated with the presence of malignant tumors.14,15

Figure 2. 1H NMR spectra of the macromolecular baseline obtained by ultrafiltration or by diffusion NMR (b ) 10 000 s/mm2). Both spectra show the same number of resonances with roughly the same intensities. Note that the ultrafiltration method (with three wash cycles) does not completely separate the macromolecules from the metabolites as evident from the sharp resonances (arrows).

Potential complications of the diffusion NMR approach to separate metabolite and macromolecular resonances are heterogeneity in the translational diffusion coefficients among the various macromolecular resonances and NMR-invisible metabolite pools due to metabolite-macromolecule interactions. Heterogeneous macromolecular diffusion coefficients will lead to different diffusion losses among the various macromolecular resonances during high diffusion weighting and thus to an incomplete subtraction of the macromolecular baseline. To investigate this potential complication, the absolute translational diffusion coefficients of eight macromolecular resonances were measured. The average macromolecular diffusion coefficient was determined as (0.186 ( 0.025) × 10-4 mm2/s with a range of (0.139-0.215) × 10-4 mm2/s. With a diffusion b value of 10 000 s/mm2, as used on the plasma samples, the macromolecular resonances would be suppressed (14) Ala-Korpela, M. Prog. NMR Spectrosc. 1995, 27, 475-554. (15) Wieczorek, A. J.; Rhyner, C.; Block, L. H. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 3455-3459.

by 13-19%, leading to a correction factor of 1.15-1.24, in excellent agreement with the average scaling factor used on the plasma samples. The spread of macromolecular diffusion coefficients and associated scaling factors leads to a suppression of macromolecular resonances of >96%, which is significantly better than obtained with many of the traditional preparation methods.5 In contrast to the low macromolecular diffusion coefficients, the average metabolite diffusion coefficient was calculated as (5.90 ( 1.15) × 10-4 mm2/s with a range of (4.71-8.42) × 10-4 mm2/s. Therefore, at a b value of 10 000 s/mm2, all metabolite resonances are suppressed by >100-fold. The diffusion coefficients of the macromolecules are relatively low when compared to previously published values. For example, Liu et al.16 derived macromolecular diffusion coefficients ranging between 0.70 and 2.5 × 10-4 mm2/s at 310 K. However, Zhang et al.17 reported in a later publication values ranging from 0.18 × 10-4 to 0.69 × 10-4 mm2/s for lipoproteins, (16) Liu, M.; Nicholson, J. K.; Parkinson, J. A.; Lindon, J. C. Anal. Chem. 1997, 69, 1504-1509.

Analytical Chemistry, Vol. 75, No. 9, May 1, 2003

2103

Figure 3. Comparison between the absolute concentrations of blood plasma metabolites obtained with the ultrafiltration and diffusion NMR methods. The blood plasma samples were taken 2 h following the infusion of [2-13C]-acetate (n ) 3). Formic acid was used as an internal concentration standard. Closed and open circles represent total () [2-12C] + [2-13C]) acetate and glucose, respectively. Closed diamonds indicate lactate, β-hydroxybutyrate and [2-12C]-acetate and open diamonds represent creatinine. The solid line represents the best fit to the data (slope 0.978, intercept 0.054 mM) with a correlation coefficient of 0.9989.

glycoproteins, and serum albumin at 310.2 K. The discrepancy between the two reports was explained by an overestimation of the diffusion coefficients in the presence of thermal convection gradients. Zhang et al.17 have shown that thermal convection gradients are negligible at room temperature (298 K) and should therefore have an insignificant effect on the current macromolecular diffusion coefficients. The interaction between metabolites and macromolecules can potentially lead to (partial) NMR invisibility of the metabolite pools. This is a complication inherent to NMR spectroscopy and should be investigated for each metabolite in a particular biological sample. However, the excellent correlation between the absolute concentrations found with diffusion NMR and ultrafiltration indicates that diffusion NMR is not more sensitive to this potential complication than traditional physicochemical methods. Furthermore, because the intact sample is used with diffusion NMR, the option exists to study a potential NMR-invisible pool by magnetization-transfer techniques.18,19 Utilizing differences in translational diffusion coefficients allows a clear differentiation of macromolecule and metabolite resonances. Besides smaller translational diffusion coefficients, large macromolecules also have longer rotational correlation times than small metabolites, resulting in shorter T1 and especially T2 (17) Zhang, X.; Li, C. G.; Ye, C. H.; Liu, M. L. Anal. Chem. 2001, 73, 35283534. (18) Wolff, S. D.; Balaban, R. S. Magn. Reson. Med. 1989, 10, 135-144. (19) de Graaf, R. A.; van Kranenburg, A.; Nicolay, K. Magn. Reson. Med. 1999, 41, 1136-1144. (20) Rabenstein, D. L.; Nakashima, T.; Bigam, G. J. Magn. Reson. 1979, 34, 669-674. (21) Hofmann, L.; Slotboom, J.; Boesch, C.; Kreis, R. Magn. Reson. Med. 2001, 46, 855-863.

2104 Analytical Chemistry, Vol. 75, No. 9, May 1, 2003

relaxation time constants. Several publications have used the differences in T1 and T2 relaxation between macromolecules and metabolites to achieve spectral simplification.13,20 However, several problems are encountered when relaxation is used for spectral simplification. Differences in T1 relaxation are commonly observed with an inversion recovery method in which the macromolecule (or metabolite) resonances are “nulled” after a specific delay, T1 ln(2), following a nonselective inversion. This method gives incomplete suppression of macromolecular resonances when there is a range of different macromolecular T1 relaxation time constants, since only resonances with a specific T1 relaxation time constant can be “nulled” for a given inversion delay. Furthermore, a range of metabolite T1 relaxation time constants will result in selective suppression of resonances, thereby hampering absolute quantification. An alternative to selective “nulling” of macromolecular resonances is the quantitative determination of the longitudinal relaxation time constant for each frequency in the NMR spectrum, as previously demonstrated for human brain spectroscopy.21 The resulting distribution of longitudinal relaxation times can be segmented (thresholded) to produce metabolite and macromolecule NMR spectra. Even though this approach can potentially give good separation between macromolecule and metabolite resonances, it is very time-consuming, especially in the presence of multiexponential relaxation when many inversion recovery delays have to be sampled. Spectral simplification based on T2 relaxation is complicated by J-coupling evolution, diffusion, and selective signal loss due to different T2 relaxation times. Though the first two effects can be minimized with a CPMG sequence, absolute quantification is still cumbersome due to signal loss during the long echo time. Absolute metabolite quantification after spectral editing based on diffusion does not suffer from the confounding factors associated with T1 and T2 relaxation, since two separate NMR spectra are acquired. In one acquisition, the metabolites are minimally affected by T1 and T2 relaxation and diffusion, such that the spectrum can be directly used for quantification. The second acquisition, in which all metabolites are removed by diffusion, is used to subtract the macromolecule baseline in the first spectrum thereby simplifying, but not compromising, the quantification process. ACKNOWLEDGMENT The authors thank Bei Wang for expert assistance with the animal preparations. Dr. Anant Patel is thanked for his assistance with the ultrafiltration, Dr. Douglas Rothman for useful discussions, and Terry Nixon and Scott McIntyre for maintenance of the NMR system. The research was supported by NIH grants RO1-NS41947 (R.A.G.), RO1-NS34813 and PO1-HD32573 (K.L.B). Additional support was provided through the Yale Mouse Metabolic Phenotyping Center (http://mouse.yale.edu) in affiliation with the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grant, U24 DK59635. Received for review December 30, 2002. Accepted February 25, 2003. AC020782+