- Email: [email protected]
Spectroscopic correlation analysis of NMR-based metabonomics in exercise science
Analytica Chimica Acta 652 (2009) 173–179
Contents lists available at ScienceDirect
Analytica Chimica Acta journal homepage: www.elsevier.com/locate/aca
Spectroscopic correlation analysis of NMR-based metabonomics in exercise science Gemma M. Kirwan a , Vernon G. Coffey b , Julie O. Niere a , John A. Hawley b , Michael J. Adams a,∗ a b
Applied Chemistry, School of Applied Sciences, RMIT University, Melbourne, Vic., Australia Exercise Metabolism Group, School of Medical Sciences, RMIT University, Bundoora, Vic., Australia
a r t i c l e
i n f o
Article history: Received 8 May 2009 Received in revised form 3 July 2009 Accepted 3 July 2009 Available online 8 July 2009 Keywords: Nuclear magnetic resonance Metabonomics Correlation spectroscopy Statistical total correlation spectroscopy Covariance matrix Chemometrics
a b s t r a c t Spectroscopic studies of complex clinical fluids have led to the application of a more holistic approach to their chemical analysis becoming more popular and widely employed. The efficient and effective interpretation of multidimensional spectroscopic data relies on many chemometric techniques and one such group of tools is represented by so-called correlation analysis methods. Typical of these techniques are two-dimensional correlation analysis and statistical total correlation spectroscopy (STOCSY). Whilst the former has largely been applied to optical spectroscopic analysis, STOCSY was developed and has been applied almost exclusively to NMR metabonomic studies. Using a 1 H NMR study of human blood plasma, from subjects recovering from exhaustive exercise trials, the basic concepts and applications of these techniques are examined. Typical information from their application to NMR-based metabonomics is presented and their value in aiding interpretation of NMR data obtained from biological systems is illustrated. Major energy metabolites are identified in the NMR spectra and the dynamics of their appearance and removal from plasma during exercise recovery are illustrated and discussed. The complementary nature of two-dimensional correlation analysis and statistical total correlation spectroscopy are highlighted. © 2009 Elsevier B.V. All rights reserved.
1. Introduction Metabonomics is the measurement of the dynamic metabolic response of living systems to stimuli or modification [1]. The field of study developed from the application of NMR spectroscopy and mass spectrometry as tools for the study of complex biofluids [2,3]. Metabonomics encompasses comprehensive metabolomic profiling in whole organisms and systematic changes due to, for example, diet, lifestyle, and pharmaceutical interventions. In recent years the use and application of NMR-based metabonomics has grown rapidly and NMR techniques are commonly used in profiling urine and blood plasma [4–7], as well as other biofluids such as human cerebrospinal fluid [8], seminal fluid, synovial fluid, digestive fluids, etc. [2–4,9]. NMR is a powerful technique for biochemical analysis and such applications are in contrast to the more traditional use of NMR spectroscopy as a tool for chemical structural elucidation. The spectra provide information on a wide range of low molecular weight metabolites pertaining to biochemical status and physiological processes. Lindon and Nicholson have recently reviewed the field and summarised the status of metabonomics [10]. 1 H NMR spectra of biofluids, such as urine and plasma, may contain thousands of signals arising from many hundreds of
∗ Corresponding author. Tel.: +61 3 9925 3358. E-mail address: [email protected] (M.J. Adams). 0003-2670/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2009.07.011
molecules, and a limiting factor in retrieving and understanding the relevant information from biological NMR spectra is their complexity [6]. One common approach to handling this complexity is to consider the NMR signal intensity data from a series of samples as a multi-sample array of metabolite concentrations and to treat it as a statistical object for data reduction and pattern recognition analysis [11]. With increasing spectral resolution (500–950 MHz fields are commonly employed) the complexity of the spectra increases and efficient and effective preprocessing and multivariate analysis methods are of paramount importance. Signal alignment of spectra recorded from a series of similar samples presents a serious challenge. Averaging, or ‘binning’, data in a pre-selected window of frequencies has been one commonly used technique to reduce the effect of misalignment [3,12] but this can lead to loss of resolution and information. As a result, whole-spectrum analysis is gaining favour with sophisticated spectra alignment algorithms being employed [13–15]. Data analysis in metabonomics makes extensive use of factor analysis methods. Unsupervised pattern recognition using principal components analysis is common and when a dependant variable is present, for example a known concentration of a species or time of recording of each of a dynamic series of spectra, supervised data analysis methods are employed, of which partial least squares (PLS) is the most common [16]. These chemometric techniques are applied via decomposition of a suitable dispersion matrix, of which
174
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
the covariance matrix, or related correlation coefficient matrix, is the most common. Since the covariance matrix summarises the statistical interaction between measured variables it is not surprising that many multivariate analysis techniques use covariance, or correlation coefficient, to inform subsequent data analysis. Correlation analysis in NMR studies was pioneered in the 1970s, and two-dimensional NMR COSY (Correlation Spectroscopy) experiments allow determination of connectivity in a molecule by identifying spin–spin proton coupling [17]. In seeking to apply a similar concept to other spectroscopic techniques, particularly infrared spectroscopy, Noda recognised that the RF pulse sequences used in two-dimensional NMR may be regarded as external perturbations applied to stimulate the system and he developed perturbation-based two-dimensional IR correlation spectroscopy using an externally applied oscillating mechanical perturbation of a sample. To overcome the limitation that the perturbation should be simple sinusoidal, Noda expanded the concept and mathematical basis of the technique to handle an arbitrary form of variable dependence [18] and developed what is referred to as generalized two-dimensional correlation spectroscopy which has been applied to many analytical techniques and a wide variety of analyses, including NMR spectroscopy [19]. For example, in a dynamic series of NMR spectra a significant correlation coefficient value between resonance peaks indicates them as arising from the same or related (in terms of process dynamics) molecular species. NMR-based metabonomics has largely been pioneered by Nicholson and co-workers who developed a technique for aiding identification and assignment of multiple NMR peaks from the same molecule in a complex biofluids mixture [20,21]. The method, Statistical Total Correlation Spectroscopy (STOCSY), displays colinearity of intensity variables from a series of NMR spectra, so that correlations between resonances arising from the same molecule can be identified [7,22]. Sasic has recently applied two-dimensional correlation analysis of 1 H NMR metabonomics data from rat urine collected at specific time intervals following administration of a trial drug [23]. Both binned and high-resolution data were examined, and covariance and correlation coefficient maps produced to aid interpretation of the data. In the current study, human blood plasma samples, collected from subjects following vigorous exercise and ingestion of a controlled diet, have been investigated by 1 H NMR spectroscopy and the data subjected to STOCSY and generalised two-dimensional correlation analysis. Recently, Pederson et al. have reported the effects of glycogen-depleting exercise and subsequent carbohydrate and caffeine ingestion on rates of post-exercise muscle glycogen accumulation [24]. Plasma samples collected during a comparable study replicating the methods of Pedersen and co-workers [24] are analysed here to illustrate the merits and features of two-dimensional correlation analysis and demonstrate the potential for 1 H NMR analysis in exercise biochemistry.
Noda treated the correlation between signals as a complex number comprising two orthogonal components known as the synchronous, , and asynchronous, , correlation intensity [18]. These components are defined by, ˚ = X¯ T · X¯
(1)
= X¯ T · N · X¯
(2)
where X¯ is a matrix of corrected spectra, recorded at i = 1 to n sequential time intervals over ı = 1–m spectral variables (chemical shift values). Generally, X¯ is the matrix of mean-centred spectra and ˚ is thus the covariance matrix indicating the simultaneous changes of the spectral intensities observed at any pair of spectral variables. Visual inspection of ˚ will indicate correlated, positive and negative, and uncorrelated spectral features. The diagonal vector of ˚ is referred to as the autocorrelation spectrum comprising the so-called auto-peaks. For mean-centred data this diagonal is simply the vector of variance values for the series of spectra. The asynchronous matrix, , represents sequential or unsynchronised variations and is obtained from the cross product between the original mean-centred data and orthogonalised data. This is achieved in Eq. (2) by the Hilbert-Noda matrix, N (size n × n) that serves to extract out of phase portions of the signals. It is defined by
Ni,k =
0 1/(i − k),
if i = k otherwise
(3)
The asynchronous matrix, , serves to indicate sequential, but not simultaneous, changes of spectral intensities measured at two different spectral variables. The presence of a non-zero element in means that that the spectral dynamics at the pair of variables are not in linear relationships [25]. The diagonal of is thus zero by definition. Noda has provided a graphical interpretation of and and their complementary nature [26]. It is usual to display both ˚ and as two-dimensional contour plots of covariance between the measured intensities at the spectral variables. However, visual interpretation of contour maps of the covariance, synchronous and asynchronous, matrices is difficult with high-resolution spectra, such as the 1 H data used here. The matrices are relatively large (typically a few thousand elements square) and sparse, and relationships between molecular species may be far apart on the chemical shift axis. This problem is overcome by selecting discrete slices from the matrices. A slice, (ı1,ı) or (ı1,ı), defines the synchronous or asynchronous relationships between a single peak (characteristic of a selected molecule appearing at ␦1) and all other peaks, across all ı. The selected molecule, or peak, is sometimes referred to as the ‘driver’ peak [27]. The covariance matrix, ˚, as defined in Equation 1 also lies at the heart of STOCSY, which in addition makes use of the correlation coefficient matrix between spectral variables, R. The correlation coefficient between recorded spectra and the dependant variable, time, (ı,t), can also be useful.
2. Theory
R = X˜ T · X˜
(4)
Two-dimensional correlation spectroscopy provides graphical representation of the quantitative relationships between signal intensity at all pairs of spectral variable over the range of the dependant, perturbation variable. In the case of a dynamic series of spectra this perturbation variable is represented by the known time intervals at which discrete samples are collected or, in a continuous study, the spectra are recorded during the experiment. The methods consider not the specific individual spectra themselves but the dynamics and changes associated with spectral features. Changes in data are visualised and investigated by examination of spectral maps.
(ı,t) = X˜ T · t˜
(5)
where X˜ represents the matrix of standardised (auto-scaled) spectral intensity values and t˜ the standardised time variable. The values from a slice, R (ı1, ı), from the correlation coefficient matrix indicate the degree of correlation between the selected variable and all others and have been displayed by colour coding the correlation coefficient values and projecting these on the NMR spectrum or covariance slice [20]. In this study, the correlation coefficient slices are displayed as separate plots for clarity alongside the covariance, synchronous and asynchronous, slice plots.
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Two-dimensional correlation analysis takes advantage of the multi-collinearity of the intensity variables in a set of NMR spectra to generate a pseudo-two-dimensional NMR spectrum that displays the correlation among the intensities of the various peaks across the whole-spectrum. Although the values of correlation coefficients in R will be subject to noise and will suffer from peak overlap, a set of spectra with a changing amount of some molecule will show high correlation between the resonance variables for the molecule. The methods are not limited to the usual connectivities that are deducible from more conventional techniques, but can also highlight molecules that are involved in the same pathway because of high biological covariance [10]. 3. Experimental 3.1. Samples The protocol used in the present study has been described in detail previously [24]. In brief, the evening before an experimental trial, endurance-trained subjects reported to the laboratory to perform a bout of intermittent exhaustive cycling. After this muscle glycogen-depleting ride, subjects were then provided with a low-carbohydrate meal. Following ∼10 h overnight fast, subjects reported to the lab the next morning and cycled until volitional fatigue at a power output corresponding to ∼70% peak O2 uptake. Subjects consumed a total of 4 g carbohydrate kg−1 body mass during the 4 h recovery period by ingesting carbohydrate at cessation of exercise and again after 60 min, 120 min, and 180 min recovery. In addition, a total of 6 mg caffeine kg−1 body mass was administered in two equal doses immediately after exercise and after 2 h recovery. Blood samples (5 mL) were taken at rest, immediately following exercise, and at regular intervals (30 min, 60 min, 90 min, 120 min, 180 min, 240 min) during the 4 h passive recovery. Five millilitres of whole blood were placed into a tube containing fluoride EDTA, mixed, and centrifuged at 4000 rpm for 8 min at 273K. The plasma samples were stored at 193K. 3.2. NMR analysis EDTA-plasma (300 L), prepared as described above, was added to 400 L of 0.9% saline solution (30%, D2 O). Samples were then placed in 5 mm NMR tubes (Wilmad, NJ, USA). CPMG spin echo [28] spectra, with pre-saturation of the water peak, were measured on a Varian 500 MHz NMR spectrometer using a 5 mm inverse detection probe. Each spectrum was acquired over 128 transients with 32 k data points, a spectrum width 6492.5 Hz, acquisition time 1.99 s and relaxation delay 6.00 s (to ensure complete relaxation between transients). All spectra were manually phase and baseline corrected using MATNMR [29]. Peak assignments were based on those reported previously [30]. The water and EDTA regions [31] were removed manually prior to aligning spectra and further data analysis. All spectra were normalised to a constant integrated intensity of 100 units. 3.3. Data processing Data manipulation and analysis algorithms were implemented using Matlab (V7.4, Mathworks Inc., MA, USA), the PLS Toolbox (V 4.0.2 Eigenvector Research Inc. WA, USA) and in-house developed algorithms. In this study the spectra were aligned, smoothed and processed prior to multivariate analysis using the temporal structure within the data, as discussed previously [14]. Briefly, spectra were aligned using selected spectral regions shifted to maximise correlation with
175
identical segments from spectra recorded at previous times. Following alignment each spectrum was smoothed using a 9-point quadratic Savitsky-Golay polynomial filter. 4. Results and discussion In the present study we present novel information using the application of correlation analysis techniques to NMR-based spectra of plasma from human subjects undertaking an exercisedietary regimen. As might be expected, significant differences were observed comparing data from different subjects commensurate with inter-individual variation in the circulating base metabolic substrate levels and differences in the individual response to exhaustive exercise and subsequent recovery. Accordingly, we present results with specific reference to data and observed trends associated with a single subject. Fig. 1a illustrates the average processed 1 H NMR spectrum of blood plasma from the subject. The peaks associated with major metabolites are identified [30]. Fig. 1b illustrates the correlation coefficient vector, (ı1,t) from Eq. (5), of the post-exercise NMR spectra over time. The diagram is simplified by displaying only those NMR peaks with intensity values greater than 10% of maximum peak intensity and with absolute correlation coefficient values above 0.5. Amino acids valine (1.02 ppm), alanine (1.44 ppm), lysine (1.46 ppm) and an unidentified metabolite (3.34 ppm) have a positive relationship with recovery (time), whilst 3-hydroxybutyrate (1.17 ppm), glutamine (2.36 ppm), acetoacetate 2.20 ppm), and other unidentified metabolites (3.4 – 3.9 ppm), display negative correlation. These results are in general agreement with other independent studies [32,33]. The region between 3.4 ppm and 3.9 ppm is complex and many of these peaks are attributable to glucose and amino acids [30]. Since the isolated -glucose peak at 4.61 ppm displays low linear correlation with recovery time, it can be assumed that those peaks between 3.4 ppm to 3.9 ppm having strong negative correlation are most likely due to amino acids rather than glucose. Further information was obtained from the covariance maps of post-exercise NMR spectra. Fig. 2 shows the synchronous, asynchronous, and correlation slices derived from the -glucose peak at 4.61 ppm. Effective use of the correlation coefficient slice can present practical problems since, by definition, correlation coefficient indicates the degree of linear relationship between variables, independent of variable magnitude. Thus a complete correlation coefficient plot can be dominated by spurious and analytically meaningless high correlations with, for example, background signals of minor intensity. To overcome this and to focus attention on significant trends in the spectra, we have elected to display only those spectral features that are greater than 10% of maximum peak intensity and that also exhibit absolute correlation coefficient values greater than 0.5. From the synchronous slice (Fig. 2a), the change in glucose concentration during post-exercise recovery relates negatively with the concentration of lactate, ketones and amino acids. The correlation coefficient slice (Fig. 2b) confirms the negative glucose-lactate, glucose-ketones and glucose-amino acids correlations. The similar behaviour of the assumed methyl (1.31 ppm) and methine (4.09 ppm) protons of lactate helps to confirm the assignment of these resonance bands. Both slices show positive correlation between the selected driver peak at 4.61 ppm and the alpha and beta glucose signals between 3.5 ppm and 3.9 ppm. The interpretation of the asynchronous data (Fig. 2c) relies on applying ‘Noda’s rule’ [14,18] and provides information on the rate of change of species concentrations over the time period of observation. Briefly, for positively correlated pairs of variables the sign of an asynchronous peak is positive if the intensity change at a variable occurs before that of a second variable in the sequential order,
176
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Fig. 1. (a) Average pre-processed 500 MHz 1 H NMR spectrum of human blood plasma. The resonance assignments are: [1] lipids, [2] valine/isoleucine, [3] 3-hydroxybutyrate, [4] lipid, [5] threanine, [6] lactate, [7] alanine, [8] lysine, [9] arginine/lysine, [10] proline, [11] glutamine, [12] acetoacetate, [13] 3-hydroxybutyrate/glutamine, [14] 2oxoglutarate, [15] glucose/amino acids, [16] creatinine, [17] lactate, [18] -glucose. (b) A correlation vector of 1 H NMR resonances against post-exercise recovery time. Resonance assignments are as for Fig. 1(a).
and the sign becomes negative if the change at the first variable occurs later. This sign rule is reversed if the synchronous correlation intensity at the same coordinate is negative. In the asynchronous slice for glucose there are positive values for the ketones (acetoacetate and 3-hydroxybutyrate). Acetoacetate is
negatively correlated to glucose (Fig. 2b), indicating that the ketone is removed faster than glucose appears. The synchronous, correlation coefficient, and asynchronous relationship between glucose and lactate are negative suggesting that glucose rate of appearance is more rapid than lactate’s rate of concentration decrease in plasma
Fig. 2. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 4.61 ppm -glucose resonance. Resonance assignments are as for Fig. 1(a).
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
177
Fig. 3. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 2.20 ppm acetoacetate resonance. Resonance assignments are as for Fig. 1(a).
during recovery from exhaustive exercise when carbohydrate is consumed. With acetoacetate (2.2 ppm) as the driver peak the synchronous and correlation coefficient slices (Fig. 3) show a positive relationship with 3-hydroxybutyrate (a by-product from acetoacetate),
lactate and some amino acids, and a negative correlation with glucose. During high intensity exercise and high rates of fatty acid oxidation the liver generates large amounts of acetyl-CoA (Fig. 5) that results in ketogenesis, with production of acetoacetate, 3-hydroxybutyrate, and acetone [34,35]. Acetoacetate has
Fig. 4. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 1.46 ppm alanine resonance. Resonance assignments are as for Fig. 1(a).
178
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Fig. 5. Summary of production and transport of some key metabolites in human plasma observable with 1 H NMR spectroscopy.
been observed to rise to large concentrations as a result of exercising and fasting [34,35]. The asynchronous slice driven from acetoacetate shows a near-zero value for 3-hydroxybutyrate indicating a similar rate of change of concentration as observed for acetoacetate. Finally, in Fig. 4 the alanine (1.46 ppm) driven data are presented. Alanine exhibits positive correlation and synchronous values with lactate and negative values with glucose. This is not surprising given that both alanine and lactate can act as substrates in gluconeogenesis (Cori cycle and alanine cycle) [36]. In addition, the positive asynchronous relationship of both glucose and lactate with the alanine driver indicates the glucose rate of change of concentration in plasma is more rapid than alanine, which in turn changes more quickly than lactate after exhaustive exercise. Exercise and nutrient ingestion represent potent stimuli with the capacity to disrupt homeostasis and alter whole body metabolism. In the present study we have employed NMR-based spectra to provide novel data regarding the dynamic changes in plasma metabolite concentration in response to exercise to fatigue and nutrient provision during recovery in humans. There is a paucity of information relating to metabonomics in humans and more work is needed to establish dynamic responses to exercise and nutrients. Nonetheless, the schematic diagram shown in Fig. 5 summarises important putative metabolic changes accompanying endurance exercise and may provide some rationale for the findings of our study. Three main biological processes can be identified: (1) Glucose is a primary energy source of muscle tissue and muscle glycogen is converted to glucose and consumed for energy during exercise. With extended exercise duration glucose is transported from the liver to muscle via the bloodstream to provide additional energy for muscle contraction. In contrast, lactate is a by-product of anaerobic metabolism and its plasma concentration increases following high intensity exercise. Lactic acid can be used as substrate for energy in skeletal muscle or transported to the liver where it is re-cycled (Cori cycle). The relationship between glucose and lactic acid is well documented [35,37]. (2) Exercise to exhaustion would be expected to release ketones into the blood to serve as a temporary and rapid form of muscle substrate [35,38]. (3) Protein turnover (synthesis versus degradation) is a continuous process that is elevated during/after exercise and is characterised by changes in the rate of appearance and disappearance of amino acids in the bloodstream. Amino acids account for a relatively small percentage contribution of energy for contractile activity but dynamic
changes in amino acid metabolism may be an important key to understanding regulation of metabolic adaptations during recovery from physical exertion [32]. 5. Conclusions 1 H NMR spectroscopy is an effective technique for both metabolite fingerprinting and metabolite profiling applications in human blood plasma samples. The use of NMR spectral analysis and the development and implementation of consistent procedures and protocols for data collection and data processing will further enhance the use of the technique. In an early review of the application of NMR spectroscopy in human biochemical energetics, Radda anticipated the rapid growth of NMR-based metabonomics and the increasing pressure to establish empirical correlations between metabolic patterns and subject management [39]. Whilst acknowledging this as a practical and valid approach Radda also emphasized the need to elucidate the fundamental biochemical questions in human pathology. Indeed, such understanding can be gained by application of efficient and effective aids to spectral interpretation, such as the correlation-based methods discussed here. Two-dimensional correlation analysis techniques, including STOCSY, are based on the observed covariance between all pairs of NMR resonances. They provide visual summaries of the dynamic changes occurring in a series of spectra and can serve as valuable aids in interpreting complex NMR spectra providing information on intramolecular links as well as information on molecular biochemical pathways. NMR has much to offer as a source of biochemical information, and in this study human plasma samples taken during recovery from exercise have been examined. The simultaneous collection of data relating to a number of energy metabolites has been achieved and the metabolites and their dynamics during subject recovery elucidated.
Acknowledgements This study was supported in part by a research grant from Glaxo SmithKline(U.K.) to J. A. Hawley. References [1] J.K. Nicholson, J.C. Lindon, E. Holmes, Xenobiotica 29 (1999) 181.
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179 [2] W. Griffiths, Metabolomics, Metabonomics and Metabolite Profiling, Royal Soc. Chem, Cambridge (UK), 2008. [3] J. Lindon, J. Nicholson, E. Holmes, Handbook of Metabonomics and Metabolomics, Elsevier, Amsterdam, 2007. [4] J. Lindon, E. Holmes, J. Nicholson, Pharm. Res. 23 (2006) 1075. [5] J. Lindon, J. Nicholson, Trends Anal. Chem. 27 (2008) 194. [6] J.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Chem. 75 (2003) 384A. [7] A. Maher, S. Zirah, E. Holmes, J. Nicholson, Anal. Chem. 79 (2007) 5204. [8] N.M. Jukarainen, S.P. Korhonen, M.P. Laakso, M. Korolainen, M. Niemetz, P. Soininen, K. Tuppurainen, J. Vepsalainen, R. Laatikainen, Metabolomics 4 (2008) 150. [9] K. Solanky, Anal. Biochem. 323 (2003) 197. [10] J.C. Lindon, J.K. Nicholson, Ann. Rev. Anal. Chem. 1 (2008) 45. [11] E. Holmes, H. Antti, Analyst 127 (2002) 1549. [12] R.A. Davis, A.J. Charlton, J. Godward, S.A. Jones, M. Harrison, J. Wilson, Chemom. Intell. Lab. Syst. 85 (2007) 144. [13] J. Forshed, I. Schuppe-Koistinen, S.P. Jacobsson, Anal. Chim. Acta 487 (2003) 189. [14] G. Kirwan, S. Clark, N.W. Barnett, J. Niere, M.J. Adams, Anal. Chim. Acta 629 (2008) 128. [15] G.C. Lee, D.L. Woodruff, Anal. Chim. Acta 513 (2004) 413. [16] S. Wold, M. Sjöström, L. Eriksson, Chemom. Intell. Lab. Syst. (2001). [17] T.D.W. Claridge, High-resolution NMR techniques in organic chemistry, Elsevier, 1999. [18] I. Noda, Y. Ozaki, Two-Dimensional Correlation Spectroscopy, J. Wiley and Sons Ltd, Chichester (UK), 2004. [19] C.D. Eads, I. Noda, J. Am. Chem. Soc. 124 (2002) 1111. [20] O. Cloarec, M.E. Dumas, A. Craig, R.H. Barton, J. Trygg, J. Hudson, C. Blancher, D. Gauguier, J.V.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Chem. 77 (2005) 1282. [21] J. Nicholson, E. Holmes, O. Cloarec, D. Crockford, J. Lindon, M. Rantalainen, United States Patent Application 0070043518, (2008) 52.
179
[22] L. Smith, A. Maher, O. Cloarec, M. Rantalainen, H. Tang, P. Elliott, J. Stamler, J. Lindon, E. Holmes, J. Nicholson, Anal. Chem. 79 (2007) 5682. [23] S. Sasic, Appl. Spectro. 62 (2008) 840. [24] D. Pedersen, S. Lessard, V. Coffey, E. Churchley, A. Wootton, T. Ng, M. Watt, J. Hawley, J. Appl. Physiol. 105 (2008) 7. [25] S.S.A. Muszynski, Y. Ozaki, J. Phys. Chem. 104 (2000) 6380. [26] I. Noda, Vib. Spectro. 36 (2004) 261. [27] C. Johnson, T. Athersuch, I. Wilson, L. Iddon, X. Meng, A. Stachulski, J. Lindon, J. Nicholson, Anal. Chem. 80 (2008) 4886. [28] S. Meiboom, D. Gill, Rev. Sci. Instrum. 29 (1958) 688. [29] J.D. van Beek, J. Magn. Reson. 187 (2007) 19. [30] J. Nicholson, P. Foxall, M. Spraul, R.D. Farrant, J. Lindon, Anal. Chem. 67 (1995) 793. [31] J.K. Nicholson, M.J. Buckingham, P.J. Sadler, Biochem. J. 211 (1983) 605. [32] M. Di Pasquale, Amino acids and proteins for the athlete: The anabolic edge, CRC Press, Boca Raton (USA), 2007. [33] R. Koopman, M. Beelen, T. Stellingwerff, B. Pennings, W. Saris, A. Kies, H. Kuipers, L. Van Loon, AJP: Endocrinol. Metab. 293 (2007) E833. [34] S. Coolen, C. Daykin, J. van Duynhoven, F. van Dorsten, F. Wulfert, J. Mathot, M. Scheltinga, O. Stroosma, H. Vader, M. Wijnen, NMR Biomed. 21 (2008) 686. [35] R.E.C. Wildman, D.M. Medeiros, Advanced Human Nutrition, CRC Press, Boca Raton (USA), 2000. [36] G. Perriello, R. Jorde, N. Nurjhan, M. Stumvoll, Am. J. Physiol.—Endocrinol. Metab. (1995) E443. [37] E. Blomstrand, B. Saltin, J. Physiol. 514 (1999) 293. [38] I. Wolinsky, Nutrition in Exercise and Sport, 3rd Ed, CRC Press, Boca Raton (USA), 1998. [39] G.K. Radda, Science 233 (1986) 640.
Contents lists available at ScienceDirect
Analytica Chimica Acta journal homepage: www.elsevier.com/locate/aca
Spectroscopic correlation analysis of NMR-based metabonomics in exercise science Gemma M. Kirwan a , Vernon G. Coffey b , Julie O. Niere a , John A. Hawley b , Michael J. Adams a,∗ a b
Applied Chemistry, School of Applied Sciences, RMIT University, Melbourne, Vic., Australia Exercise Metabolism Group, School of Medical Sciences, RMIT University, Bundoora, Vic., Australia
a r t i c l e
i n f o
Article history: Received 8 May 2009 Received in revised form 3 July 2009 Accepted 3 July 2009 Available online 8 July 2009 Keywords: Nuclear magnetic resonance Metabonomics Correlation spectroscopy Statistical total correlation spectroscopy Covariance matrix Chemometrics
a b s t r a c t Spectroscopic studies of complex clinical fluids have led to the application of a more holistic approach to their chemical analysis becoming more popular and widely employed. The efficient and effective interpretation of multidimensional spectroscopic data relies on many chemometric techniques and one such group of tools is represented by so-called correlation analysis methods. Typical of these techniques are two-dimensional correlation analysis and statistical total correlation spectroscopy (STOCSY). Whilst the former has largely been applied to optical spectroscopic analysis, STOCSY was developed and has been applied almost exclusively to NMR metabonomic studies. Using a 1 H NMR study of human blood plasma, from subjects recovering from exhaustive exercise trials, the basic concepts and applications of these techniques are examined. Typical information from their application to NMR-based metabonomics is presented and their value in aiding interpretation of NMR data obtained from biological systems is illustrated. Major energy metabolites are identified in the NMR spectra and the dynamics of their appearance and removal from plasma during exercise recovery are illustrated and discussed. The complementary nature of two-dimensional correlation analysis and statistical total correlation spectroscopy are highlighted. © 2009 Elsevier B.V. All rights reserved.
1. Introduction Metabonomics is the measurement of the dynamic metabolic response of living systems to stimuli or modification [1]. The field of study developed from the application of NMR spectroscopy and mass spectrometry as tools for the study of complex biofluids [2,3]. Metabonomics encompasses comprehensive metabolomic profiling in whole organisms and systematic changes due to, for example, diet, lifestyle, and pharmaceutical interventions. In recent years the use and application of NMR-based metabonomics has grown rapidly and NMR techniques are commonly used in profiling urine and blood plasma [4–7], as well as other biofluids such as human cerebrospinal fluid [8], seminal fluid, synovial fluid, digestive fluids, etc. [2–4,9]. NMR is a powerful technique for biochemical analysis and such applications are in contrast to the more traditional use of NMR spectroscopy as a tool for chemical structural elucidation. The spectra provide information on a wide range of low molecular weight metabolites pertaining to biochemical status and physiological processes. Lindon and Nicholson have recently reviewed the field and summarised the status of metabonomics [10]. 1 H NMR spectra of biofluids, such as urine and plasma, may contain thousands of signals arising from many hundreds of
∗ Corresponding author. Tel.: +61 3 9925 3358. E-mail address: [email protected] (M.J. Adams). 0003-2670/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2009.07.011
molecules, and a limiting factor in retrieving and understanding the relevant information from biological NMR spectra is their complexity [6]. One common approach to handling this complexity is to consider the NMR signal intensity data from a series of samples as a multi-sample array of metabolite concentrations and to treat it as a statistical object for data reduction and pattern recognition analysis [11]. With increasing spectral resolution (500–950 MHz fields are commonly employed) the complexity of the spectra increases and efficient and effective preprocessing and multivariate analysis methods are of paramount importance. Signal alignment of spectra recorded from a series of similar samples presents a serious challenge. Averaging, or ‘binning’, data in a pre-selected window of frequencies has been one commonly used technique to reduce the effect of misalignment [3,12] but this can lead to loss of resolution and information. As a result, whole-spectrum analysis is gaining favour with sophisticated spectra alignment algorithms being employed [13–15]. Data analysis in metabonomics makes extensive use of factor analysis methods. Unsupervised pattern recognition using principal components analysis is common and when a dependant variable is present, for example a known concentration of a species or time of recording of each of a dynamic series of spectra, supervised data analysis methods are employed, of which partial least squares (PLS) is the most common [16]. These chemometric techniques are applied via decomposition of a suitable dispersion matrix, of which
174
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
the covariance matrix, or related correlation coefficient matrix, is the most common. Since the covariance matrix summarises the statistical interaction between measured variables it is not surprising that many multivariate analysis techniques use covariance, or correlation coefficient, to inform subsequent data analysis. Correlation analysis in NMR studies was pioneered in the 1970s, and two-dimensional NMR COSY (Correlation Spectroscopy) experiments allow determination of connectivity in a molecule by identifying spin–spin proton coupling [17]. In seeking to apply a similar concept to other spectroscopic techniques, particularly infrared spectroscopy, Noda recognised that the RF pulse sequences used in two-dimensional NMR may be regarded as external perturbations applied to stimulate the system and he developed perturbation-based two-dimensional IR correlation spectroscopy using an externally applied oscillating mechanical perturbation of a sample. To overcome the limitation that the perturbation should be simple sinusoidal, Noda expanded the concept and mathematical basis of the technique to handle an arbitrary form of variable dependence [18] and developed what is referred to as generalized two-dimensional correlation spectroscopy which has been applied to many analytical techniques and a wide variety of analyses, including NMR spectroscopy [19]. For example, in a dynamic series of NMR spectra a significant correlation coefficient value between resonance peaks indicates them as arising from the same or related (in terms of process dynamics) molecular species. NMR-based metabonomics has largely been pioneered by Nicholson and co-workers who developed a technique for aiding identification and assignment of multiple NMR peaks from the same molecule in a complex biofluids mixture [20,21]. The method, Statistical Total Correlation Spectroscopy (STOCSY), displays colinearity of intensity variables from a series of NMR spectra, so that correlations between resonances arising from the same molecule can be identified [7,22]. Sasic has recently applied two-dimensional correlation analysis of 1 H NMR metabonomics data from rat urine collected at specific time intervals following administration of a trial drug [23]. Both binned and high-resolution data were examined, and covariance and correlation coefficient maps produced to aid interpretation of the data. In the current study, human blood plasma samples, collected from subjects following vigorous exercise and ingestion of a controlled diet, have been investigated by 1 H NMR spectroscopy and the data subjected to STOCSY and generalised two-dimensional correlation analysis. Recently, Pederson et al. have reported the effects of glycogen-depleting exercise and subsequent carbohydrate and caffeine ingestion on rates of post-exercise muscle glycogen accumulation [24]. Plasma samples collected during a comparable study replicating the methods of Pedersen and co-workers [24] are analysed here to illustrate the merits and features of two-dimensional correlation analysis and demonstrate the potential for 1 H NMR analysis in exercise biochemistry.
Noda treated the correlation between signals as a complex number comprising two orthogonal components known as the synchronous, , and asynchronous, , correlation intensity [18]. These components are defined by, ˚ = X¯ T · X¯
(1)
= X¯ T · N · X¯
(2)
where X¯ is a matrix of corrected spectra, recorded at i = 1 to n sequential time intervals over ı = 1–m spectral variables (chemical shift values). Generally, X¯ is the matrix of mean-centred spectra and ˚ is thus the covariance matrix indicating the simultaneous changes of the spectral intensities observed at any pair of spectral variables. Visual inspection of ˚ will indicate correlated, positive and negative, and uncorrelated spectral features. The diagonal vector of ˚ is referred to as the autocorrelation spectrum comprising the so-called auto-peaks. For mean-centred data this diagonal is simply the vector of variance values for the series of spectra. The asynchronous matrix, , represents sequential or unsynchronised variations and is obtained from the cross product between the original mean-centred data and orthogonalised data. This is achieved in Eq. (2) by the Hilbert-Noda matrix, N (size n × n) that serves to extract out of phase portions of the signals. It is defined by
Ni,k =
0 1/(i − k),
if i = k otherwise
(3)
The asynchronous matrix, , serves to indicate sequential, but not simultaneous, changes of spectral intensities measured at two different spectral variables. The presence of a non-zero element in means that that the spectral dynamics at the pair of variables are not in linear relationships [25]. The diagonal of is thus zero by definition. Noda has provided a graphical interpretation of and and their complementary nature [26]. It is usual to display both ˚ and as two-dimensional contour plots of covariance between the measured intensities at the spectral variables. However, visual interpretation of contour maps of the covariance, synchronous and asynchronous, matrices is difficult with high-resolution spectra, such as the 1 H data used here. The matrices are relatively large (typically a few thousand elements square) and sparse, and relationships between molecular species may be far apart on the chemical shift axis. This problem is overcome by selecting discrete slices from the matrices. A slice, (ı1,ı) or (ı1,ı), defines the synchronous or asynchronous relationships between a single peak (characteristic of a selected molecule appearing at ␦1) and all other peaks, across all ı. The selected molecule, or peak, is sometimes referred to as the ‘driver’ peak [27]. The covariance matrix, ˚, as defined in Equation 1 also lies at the heart of STOCSY, which in addition makes use of the correlation coefficient matrix between spectral variables, R. The correlation coefficient between recorded spectra and the dependant variable, time, (ı,t), can also be useful.
2. Theory
R = X˜ T · X˜
(4)
Two-dimensional correlation spectroscopy provides graphical representation of the quantitative relationships between signal intensity at all pairs of spectral variable over the range of the dependant, perturbation variable. In the case of a dynamic series of spectra this perturbation variable is represented by the known time intervals at which discrete samples are collected or, in a continuous study, the spectra are recorded during the experiment. The methods consider not the specific individual spectra themselves but the dynamics and changes associated with spectral features. Changes in data are visualised and investigated by examination of spectral maps.
(ı,t) = X˜ T · t˜
(5)
where X˜ represents the matrix of standardised (auto-scaled) spectral intensity values and t˜ the standardised time variable. The values from a slice, R (ı1, ı), from the correlation coefficient matrix indicate the degree of correlation between the selected variable and all others and have been displayed by colour coding the correlation coefficient values and projecting these on the NMR spectrum or covariance slice [20]. In this study, the correlation coefficient slices are displayed as separate plots for clarity alongside the covariance, synchronous and asynchronous, slice plots.
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Two-dimensional correlation analysis takes advantage of the multi-collinearity of the intensity variables in a set of NMR spectra to generate a pseudo-two-dimensional NMR spectrum that displays the correlation among the intensities of the various peaks across the whole-spectrum. Although the values of correlation coefficients in R will be subject to noise and will suffer from peak overlap, a set of spectra with a changing amount of some molecule will show high correlation between the resonance variables for the molecule. The methods are not limited to the usual connectivities that are deducible from more conventional techniques, but can also highlight molecules that are involved in the same pathway because of high biological covariance [10]. 3. Experimental 3.1. Samples The protocol used in the present study has been described in detail previously [24]. In brief, the evening before an experimental trial, endurance-trained subjects reported to the laboratory to perform a bout of intermittent exhaustive cycling. After this muscle glycogen-depleting ride, subjects were then provided with a low-carbohydrate meal. Following ∼10 h overnight fast, subjects reported to the lab the next morning and cycled until volitional fatigue at a power output corresponding to ∼70% peak O2 uptake. Subjects consumed a total of 4 g carbohydrate kg−1 body mass during the 4 h recovery period by ingesting carbohydrate at cessation of exercise and again after 60 min, 120 min, and 180 min recovery. In addition, a total of 6 mg caffeine kg−1 body mass was administered in two equal doses immediately after exercise and after 2 h recovery. Blood samples (5 mL) were taken at rest, immediately following exercise, and at regular intervals (30 min, 60 min, 90 min, 120 min, 180 min, 240 min) during the 4 h passive recovery. Five millilitres of whole blood were placed into a tube containing fluoride EDTA, mixed, and centrifuged at 4000 rpm for 8 min at 273K. The plasma samples were stored at 193K. 3.2. NMR analysis EDTA-plasma (300 L), prepared as described above, was added to 400 L of 0.9% saline solution (30%, D2 O). Samples were then placed in 5 mm NMR tubes (Wilmad, NJ, USA). CPMG spin echo [28] spectra, with pre-saturation of the water peak, were measured on a Varian 500 MHz NMR spectrometer using a 5 mm inverse detection probe. Each spectrum was acquired over 128 transients with 32 k data points, a spectrum width 6492.5 Hz, acquisition time 1.99 s and relaxation delay 6.00 s (to ensure complete relaxation between transients). All spectra were manually phase and baseline corrected using MATNMR [29]. Peak assignments were based on those reported previously [30]. The water and EDTA regions [31] were removed manually prior to aligning spectra and further data analysis. All spectra were normalised to a constant integrated intensity of 100 units. 3.3. Data processing Data manipulation and analysis algorithms were implemented using Matlab (V7.4, Mathworks Inc., MA, USA), the PLS Toolbox (V 4.0.2 Eigenvector Research Inc. WA, USA) and in-house developed algorithms. In this study the spectra were aligned, smoothed and processed prior to multivariate analysis using the temporal structure within the data, as discussed previously [14]. Briefly, spectra were aligned using selected spectral regions shifted to maximise correlation with
175
identical segments from spectra recorded at previous times. Following alignment each spectrum was smoothed using a 9-point quadratic Savitsky-Golay polynomial filter. 4. Results and discussion In the present study we present novel information using the application of correlation analysis techniques to NMR-based spectra of plasma from human subjects undertaking an exercisedietary regimen. As might be expected, significant differences were observed comparing data from different subjects commensurate with inter-individual variation in the circulating base metabolic substrate levels and differences in the individual response to exhaustive exercise and subsequent recovery. Accordingly, we present results with specific reference to data and observed trends associated with a single subject. Fig. 1a illustrates the average processed 1 H NMR spectrum of blood plasma from the subject. The peaks associated with major metabolites are identified [30]. Fig. 1b illustrates the correlation coefficient vector, (ı1,t) from Eq. (5), of the post-exercise NMR spectra over time. The diagram is simplified by displaying only those NMR peaks with intensity values greater than 10% of maximum peak intensity and with absolute correlation coefficient values above 0.5. Amino acids valine (1.02 ppm), alanine (1.44 ppm), lysine (1.46 ppm) and an unidentified metabolite (3.34 ppm) have a positive relationship with recovery (time), whilst 3-hydroxybutyrate (1.17 ppm), glutamine (2.36 ppm), acetoacetate 2.20 ppm), and other unidentified metabolites (3.4 – 3.9 ppm), display negative correlation. These results are in general agreement with other independent studies [32,33]. The region between 3.4 ppm and 3.9 ppm is complex and many of these peaks are attributable to glucose and amino acids [30]. Since the isolated -glucose peak at 4.61 ppm displays low linear correlation with recovery time, it can be assumed that those peaks between 3.4 ppm to 3.9 ppm having strong negative correlation are most likely due to amino acids rather than glucose. Further information was obtained from the covariance maps of post-exercise NMR spectra. Fig. 2 shows the synchronous, asynchronous, and correlation slices derived from the -glucose peak at 4.61 ppm. Effective use of the correlation coefficient slice can present practical problems since, by definition, correlation coefficient indicates the degree of linear relationship between variables, independent of variable magnitude. Thus a complete correlation coefficient plot can be dominated by spurious and analytically meaningless high correlations with, for example, background signals of minor intensity. To overcome this and to focus attention on significant trends in the spectra, we have elected to display only those spectral features that are greater than 10% of maximum peak intensity and that also exhibit absolute correlation coefficient values greater than 0.5. From the synchronous slice (Fig. 2a), the change in glucose concentration during post-exercise recovery relates negatively with the concentration of lactate, ketones and amino acids. The correlation coefficient slice (Fig. 2b) confirms the negative glucose-lactate, glucose-ketones and glucose-amino acids correlations. The similar behaviour of the assumed methyl (1.31 ppm) and methine (4.09 ppm) protons of lactate helps to confirm the assignment of these resonance bands. Both slices show positive correlation between the selected driver peak at 4.61 ppm and the alpha and beta glucose signals between 3.5 ppm and 3.9 ppm. The interpretation of the asynchronous data (Fig. 2c) relies on applying ‘Noda’s rule’ [14,18] and provides information on the rate of change of species concentrations over the time period of observation. Briefly, for positively correlated pairs of variables the sign of an asynchronous peak is positive if the intensity change at a variable occurs before that of a second variable in the sequential order,
176
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Fig. 1. (a) Average pre-processed 500 MHz 1 H NMR spectrum of human blood plasma. The resonance assignments are: [1] lipids, [2] valine/isoleucine, [3] 3-hydroxybutyrate, [4] lipid, [5] threanine, [6] lactate, [7] alanine, [8] lysine, [9] arginine/lysine, [10] proline, [11] glutamine, [12] acetoacetate, [13] 3-hydroxybutyrate/glutamine, [14] 2oxoglutarate, [15] glucose/amino acids, [16] creatinine, [17] lactate, [18] -glucose. (b) A correlation vector of 1 H NMR resonances against post-exercise recovery time. Resonance assignments are as for Fig. 1(a).
and the sign becomes negative if the change at the first variable occurs later. This sign rule is reversed if the synchronous correlation intensity at the same coordinate is negative. In the asynchronous slice for glucose there are positive values for the ketones (acetoacetate and 3-hydroxybutyrate). Acetoacetate is
negatively correlated to glucose (Fig. 2b), indicating that the ketone is removed faster than glucose appears. The synchronous, correlation coefficient, and asynchronous relationship between glucose and lactate are negative suggesting that glucose rate of appearance is more rapid than lactate’s rate of concentration decrease in plasma
Fig. 2. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 4.61 ppm -glucose resonance. Resonance assignments are as for Fig. 1(a).
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
177
Fig. 3. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 2.20 ppm acetoacetate resonance. Resonance assignments are as for Fig. 1(a).
during recovery from exhaustive exercise when carbohydrate is consumed. With acetoacetate (2.2 ppm) as the driver peak the synchronous and correlation coefficient slices (Fig. 3) show a positive relationship with 3-hydroxybutyrate (a by-product from acetoacetate),
lactate and some amino acids, and a negative correlation with glucose. During high intensity exercise and high rates of fatty acid oxidation the liver generates large amounts of acetyl-CoA (Fig. 5) that results in ketogenesis, with production of acetoacetate, 3-hydroxybutyrate, and acetone [34,35]. Acetoacetate has
Fig. 4. (a) Synchronous, (b) correlation coefficient, and (c) asynchronous slices associated with the 1.46 ppm alanine resonance. Resonance assignments are as for Fig. 1(a).
178
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179
Fig. 5. Summary of production and transport of some key metabolites in human plasma observable with 1 H NMR spectroscopy.
been observed to rise to large concentrations as a result of exercising and fasting [34,35]. The asynchronous slice driven from acetoacetate shows a near-zero value for 3-hydroxybutyrate indicating a similar rate of change of concentration as observed for acetoacetate. Finally, in Fig. 4 the alanine (1.46 ppm) driven data are presented. Alanine exhibits positive correlation and synchronous values with lactate and negative values with glucose. This is not surprising given that both alanine and lactate can act as substrates in gluconeogenesis (Cori cycle and alanine cycle) [36]. In addition, the positive asynchronous relationship of both glucose and lactate with the alanine driver indicates the glucose rate of change of concentration in plasma is more rapid than alanine, which in turn changes more quickly than lactate after exhaustive exercise. Exercise and nutrient ingestion represent potent stimuli with the capacity to disrupt homeostasis and alter whole body metabolism. In the present study we have employed NMR-based spectra to provide novel data regarding the dynamic changes in plasma metabolite concentration in response to exercise to fatigue and nutrient provision during recovery in humans. There is a paucity of information relating to metabonomics in humans and more work is needed to establish dynamic responses to exercise and nutrients. Nonetheless, the schematic diagram shown in Fig. 5 summarises important putative metabolic changes accompanying endurance exercise and may provide some rationale for the findings of our study. Three main biological processes can be identified: (1) Glucose is a primary energy source of muscle tissue and muscle glycogen is converted to glucose and consumed for energy during exercise. With extended exercise duration glucose is transported from the liver to muscle via the bloodstream to provide additional energy for muscle contraction. In contrast, lactate is a by-product of anaerobic metabolism and its plasma concentration increases following high intensity exercise. Lactic acid can be used as substrate for energy in skeletal muscle or transported to the liver where it is re-cycled (Cori cycle). The relationship between glucose and lactic acid is well documented [35,37]. (2) Exercise to exhaustion would be expected to release ketones into the blood to serve as a temporary and rapid form of muscle substrate [35,38]. (3) Protein turnover (synthesis versus degradation) is a continuous process that is elevated during/after exercise and is characterised by changes in the rate of appearance and disappearance of amino acids in the bloodstream. Amino acids account for a relatively small percentage contribution of energy for contractile activity but dynamic
changes in amino acid metabolism may be an important key to understanding regulation of metabolic adaptations during recovery from physical exertion [32]. 5. Conclusions 1 H NMR spectroscopy is an effective technique for both metabolite fingerprinting and metabolite profiling applications in human blood plasma samples. The use of NMR spectral analysis and the development and implementation of consistent procedures and protocols for data collection and data processing will further enhance the use of the technique. In an early review of the application of NMR spectroscopy in human biochemical energetics, Radda anticipated the rapid growth of NMR-based metabonomics and the increasing pressure to establish empirical correlations between metabolic patterns and subject management [39]. Whilst acknowledging this as a practical and valid approach Radda also emphasized the need to elucidate the fundamental biochemical questions in human pathology. Indeed, such understanding can be gained by application of efficient and effective aids to spectral interpretation, such as the correlation-based methods discussed here. Two-dimensional correlation analysis techniques, including STOCSY, are based on the observed covariance between all pairs of NMR resonances. They provide visual summaries of the dynamic changes occurring in a series of spectra and can serve as valuable aids in interpreting complex NMR spectra providing information on intramolecular links as well as information on molecular biochemical pathways. NMR has much to offer as a source of biochemical information, and in this study human plasma samples taken during recovery from exercise have been examined. The simultaneous collection of data relating to a number of energy metabolites has been achieved and the metabolites and their dynamics during subject recovery elucidated.
Acknowledgements This study was supported in part by a research grant from Glaxo SmithKline(U.K.) to J. A. Hawley. References [1] J.K. Nicholson, J.C. Lindon, E. Holmes, Xenobiotica 29 (1999) 181.
G.M. Kirwan et al. / Analytica Chimica Acta 652 (2009) 173–179 [2] W. Griffiths, Metabolomics, Metabonomics and Metabolite Profiling, Royal Soc. Chem, Cambridge (UK), 2008. [3] J. Lindon, J. Nicholson, E. Holmes, Handbook of Metabonomics and Metabolomics, Elsevier, Amsterdam, 2007. [4] J. Lindon, E. Holmes, J. Nicholson, Pharm. Res. 23 (2006) 1075. [5] J. Lindon, J. Nicholson, Trends Anal. Chem. 27 (2008) 194. [6] J.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Chem. 75 (2003) 384A. [7] A. Maher, S. Zirah, E. Holmes, J. Nicholson, Anal. Chem. 79 (2007) 5204. [8] N.M. Jukarainen, S.P. Korhonen, M.P. Laakso, M. Korolainen, M. Niemetz, P. Soininen, K. Tuppurainen, J. Vepsalainen, R. Laatikainen, Metabolomics 4 (2008) 150. [9] K. Solanky, Anal. Biochem. 323 (2003) 197. [10] J.C. Lindon, J.K. Nicholson, Ann. Rev. Anal. Chem. 1 (2008) 45. [11] E. Holmes, H. Antti, Analyst 127 (2002) 1549. [12] R.A. Davis, A.J. Charlton, J. Godward, S.A. Jones, M. Harrison, J. Wilson, Chemom. Intell. Lab. Syst. 85 (2007) 144. [13] J. Forshed, I. Schuppe-Koistinen, S.P. Jacobsson, Anal. Chim. Acta 487 (2003) 189. [14] G. Kirwan, S. Clark, N.W. Barnett, J. Niere, M.J. Adams, Anal. Chim. Acta 629 (2008) 128. [15] G.C. Lee, D.L. Woodruff, Anal. Chim. Acta 513 (2004) 413. [16] S. Wold, M. Sjöström, L. Eriksson, Chemom. Intell. Lab. Syst. (2001). [17] T.D.W. Claridge, High-resolution NMR techniques in organic chemistry, Elsevier, 1999. [18] I. Noda, Y. Ozaki, Two-Dimensional Correlation Spectroscopy, J. Wiley and Sons Ltd, Chichester (UK), 2004. [19] C.D. Eads, I. Noda, J. Am. Chem. Soc. 124 (2002) 1111. [20] O. Cloarec, M.E. Dumas, A. Craig, R.H. Barton, J. Trygg, J. Hudson, C. Blancher, D. Gauguier, J.V.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Chem. 77 (2005) 1282. [21] J. Nicholson, E. Holmes, O. Cloarec, D. Crockford, J. Lindon, M. Rantalainen, United States Patent Application 0070043518, (2008) 52.
179
[22] L. Smith, A. Maher, O. Cloarec, M. Rantalainen, H. Tang, P. Elliott, J. Stamler, J. Lindon, E. Holmes, J. Nicholson, Anal. Chem. 79 (2007) 5682. [23] S. Sasic, Appl. Spectro. 62 (2008) 840. [24] D. Pedersen, S. Lessard, V. Coffey, E. Churchley, A. Wootton, T. Ng, M. Watt, J. Hawley, J. Appl. Physiol. 105 (2008) 7. [25] S.S.A. Muszynski, Y. Ozaki, J. Phys. Chem. 104 (2000) 6380. [26] I. Noda, Vib. Spectro. 36 (2004) 261. [27] C. Johnson, T. Athersuch, I. Wilson, L. Iddon, X. Meng, A. Stachulski, J. Lindon, J. Nicholson, Anal. Chem. 80 (2008) 4886. [28] S. Meiboom, D. Gill, Rev. Sci. Instrum. 29 (1958) 688. [29] J.D. van Beek, J. Magn. Reson. 187 (2007) 19. [30] J. Nicholson, P. Foxall, M. Spraul, R.D. Farrant, J. Lindon, Anal. Chem. 67 (1995) 793. [31] J.K. Nicholson, M.J. Buckingham, P.J. Sadler, Biochem. J. 211 (1983) 605. [32] M. Di Pasquale, Amino acids and proteins for the athlete: The anabolic edge, CRC Press, Boca Raton (USA), 2007. [33] R. Koopman, M. Beelen, T. Stellingwerff, B. Pennings, W. Saris, A. Kies, H. Kuipers, L. Van Loon, AJP: Endocrinol. Metab. 293 (2007) E833. [34] S. Coolen, C. Daykin, J. van Duynhoven, F. van Dorsten, F. Wulfert, J. Mathot, M. Scheltinga, O. Stroosma, H. Vader, M. Wijnen, NMR Biomed. 21 (2008) 686. [35] R.E.C. Wildman, D.M. Medeiros, Advanced Human Nutrition, CRC Press, Boca Raton (USA), 2000. [36] G. Perriello, R. Jorde, N. Nurjhan, M. Stumvoll, Am. J. Physiol.—Endocrinol. Metab. (1995) E443. [37] E. Blomstrand, B. Saltin, J. Physiol. 514 (1999) 293. [38] I. Wolinsky, Nutrition in Exercise and Sport, 3rd Ed, CRC Press, Boca Raton (USA), 1998. [39] G.K. Radda, Science 233 (1986) 640.