Statistical Integration of 1H NMR and MRS Data from Different

ARTICLE pubs.acs.org/jpr

Statistical Integration of 1H NMR and MRS Data from Different Biofluids and Tissues Enhances Recovery of Biological Information from Individuals with HIV-1 infection Anthony D Maher,*,†,‡ Lucette A. Cysique,§ Bruce J. Brew,§,|| and Caroline D. Rae†,§ †

Neuroscience Research Australia, Barker St, Randwick 2031, Australia School of Medical Sciences, University of New South Wales, NSW 2052, Australia § Brain Sciences, University of New South Wales, NSW 2052, Australia Department of Neurology, St. Vincent’s Hospital, Darlinghurst, NSW, 2010, Australia

)

‡

ABSTRACT: Nuclear magnetic resonance (NMR) spectroscopy is widely used in metabonomics studies, but optimal recovery of latent biological information requires increasingly sophisticated statistical methods to identify quantitative relationships within these often highly complex data sets. Statistical heterospectroscopy (SHY) extracts latent relationships between NMR and mass spectrometry (MS) data from the same samples. Here we extend this concept to identify novel metabolic correlations between different biofluids and tissues from the same individuals. We acquired NMR data from blood plasma and cerebrospinal fluid (CSF) (N = 19) from HIV-1-infected individuals, who are known to be susceptible to neuropsychological dysfunction. We compared two computational approaches to SHY, namely the Pearson’s product moment correlation and the Spearman’s rank correlation. High correlations were observed for glutamine, valine, and polyethylene glycol, a drug delivery vehicle. Orthogonal projections to latent structures (OPLS) identified metabolites in blood plasma spectra that predicted the amounts of key CSF metabolites such as lactate, glutamine, and myo-inositol. Finally, brain metabolic data from magnetic resonance spectroscopy (MRS) measurements in vivo were integrated with CSF data to identify an association between 3-hydroxyvalerate and frontal white matter N-acetyl aspartate levels. The results underscore the utility of tools such as SHY and OPLS for coanalysis of high dimensional data sets to recover biological information unobtainable when such data are analyzed in isolation. KEYWORDS: metabonomics, metabolomics, statistical total correlation spectroscopy (STOCSY), nuclear magnetic resonance spectroscopy (NMR), cerebrospinal fluid, blood plasma, statistical heterospectroscopy (SHY), orthogonal projections to latent structures (OPLS)

’ INTRODUCTION Metabonomics measures the multiparametric response of an organism to a stimulus, typically employing analytical technologies such as nuclear magnetic resonance spectroscopy (NMR) or mass spectrometry (MS) to obtain metabolic “fingerprints” from biological fluids such as urine or blood plasma.1,2 Recent studies have given unique insights into human epidemiology through metabolome-wide association studies3 and offered new avenues toward personalized healthcare.4 Part of the appeal of the socalled ‘omics sciences is that they provide a data-driven approach to discovery, which is complementary to traditional hypothesisdriven research.5 The high dimensionality and complexity of 1H NMR metabonomic data requires increasingly sophisticated multivariate statistical techniques for extraction of latent biological information. In recent years orthogonal projections to latent structures (OPLS) has proven a valuable tool for interpreting the quantitative relationships between NMR data and a given biological phenotype.6,7 A feature of 1H NMR-based metabonomic data is r 2011 American Chemical Society

the multicolinearity of the intensity variables within a set of spectra that allow detection of intra- and intermolecular statistical correlations.8 Statistical Total Correlation Spectroscopy (STOCSY) has been used for both molecular assignment and biomarker identification in toxicology studies.9-11 This concept has now evolved to include statistical integration of data from different NMR experiments,12,13 and from different spectroscopic sources, such as UPLC-MS with NMR.14,15 Termed Statistical Heterospectroscopy (SHY), this approach has shown promise as an analytical tool for enhancing feature extraction from different data types acquired from the same samples. This concept can be extended to extract biological information from different types of samples acquired from the same organisms. Some recent work has indicated the potential for this type of biological integration in animal models.16,17

Received: October 11, 2010 Published: January 18, 2011 1737

dx.doi.org/10.1021/pr1010263 | J. Proteome Res. 2011, 10, 1737–1745

Journal of Proteome Research There is considerable interest in application of this technology to neurological disorders, largely driven by the current lack of specific markers available to identify patients at risk of dementias such as Alzheimer’s disease.18 Cerebrospinal fluid (CSF) functions as a cushion protecting the brain from traumatic shock, but also plays an important role in delivering substances to and from the brain,19 and offers a window into understanding the metabolic basis of neurological disorders. CSF directly interacts with the brain through gap junctions, which are highly permeable to most metabolites compared to the tight junctions that form the blood-brain barrier (BBB),20 one of two discrete barrier systems along with the blood-CSF barrier. The surface area of the BBB is at least 5000 greater than the blood-CSF barrier and is the major diffusion barrier between the blood and the brain.21 The first detailed 1H NMR analysis of CSF was published in the 1980s,22 and most of the major metabolites have been assigned.23-25 1H NMR metabonomic analyses of CSF have indicated it may potentially be used for rapid diagnosis of meningitis and ventriculitis26 and identification of the early signs of schizophrenia in susceptible patients.27 In this report, we develop the necessary methodology for statistical integration of heterogeneous quantitative spectroscopic data that can effectively extract biological information unapparent by analyzing the same data in isolation. We have obtained blood plasma and CSF samples from HIV-infected individuals who were clinically stable and undergoing highly active antiretroviral therapy (HAART) for at least 6 months. We have chosen this cohort because it is generally difficult to collect CSF from otherwise healthy individuals, but also because HIVþ infected individuals suffer neurological complications, and several studies have confirmed abnormalities in BBB function in these patients, even in the presence of HAART.28 Although there is significant compartmentalization of HIV within the CNS,29 the CSF acts as the “brain’s urine”19 and will ultimately interact with the blood, and so understanding the metabolic relationships between the blood and the CSF may offer new insights into the neurological disease status of HIVþ individuals.

’ MATERIALS AND METHODS This study complied with Australian National Health and Medical Research Council guidelines on human research and was approved by St. Vincent’s Hospital and The University of New South Wales Human Research Ethics Committees. Informed consent was obtained from all participants. Thirty-seven participants (all males, median age 58.5 years, range 45-74 years) were enrolled as part of a prospective study investigating the effect of HIV and aging on brain functions in HIVþ individuals with stable disease. By design, these individuals were selected as being older than 44 years, with a nadir CD4 cells count e350 (count/mm3) and an estimated HIV duration of at least five years. All participants were examined with standard neuropsychological testing involving the assessment of attention/working memory; learning/memory, psychomotor speed, motor-coordination, mental flexibility and language. All were screened for other neurological conditions and psychiatric conditions on the psychotic axis. Depressive (assessed with the Beck Depression Inventory-II), cognitive complaints and everyday functioning independence were recorded. Adherence was also recorded and was found to be high to very high (>90%). Substance use was recorded and individuals with current substance use disorders were excluded. Thirteen percent reported

ARTICLE

clinically significant levels of depressive complaints on the Beck Depression Inventory-II but were not excluded. CSF was collected by lumbar puncture from 20 of the 37 participants. The lumbar puncture procedure is highly invasive, and participants were given the option to decline without being excluded from the remainder of the study. There was one CSF sample for which no blood plasma sample was available, resulting in a total of 19 matched CSF-blood plasma pairs. Lumbar punctures were performed by a qualified neurologist. After collection the sample was heat deactivated for HIV at 56 °C for 1 h and immediately frozen at -20 °C until analysis. A total of 1 mL of CSF was transported for NMR. Blood was collected by venipuncture from the cubital fossa into heparinized vacutainers. Cells were separated from plasma by centrifugation. One mL of plasma was aliquoted for NMR and immediately frozen at -20 °C. This was shipped on dry ice and stored at -80 °C until analysis. HIVdisease related biochemical measurements were obtained (CD4þT cell counts, plasma and CSF HIV RNA). Other nonspecific disease biomarkers were obtained such as plasma and CSF lactate. Ninety-seven percent of participants had HIVRNA suppressed ( 0.001 were set to 0. This cutoff value is equivalent to a Bonferroni correction for 50 independent tests and is equivalent to that used in previous correlation analysis publications.14,15,35 Given that a recent NMR study of CSF indicated approximately 53 compounds could be detected in CSF, we consider this a reasonable value for the current work. SHY models for CSF and blood plasma data were from 19 matched pairs, while for CSF to MRS data 18 matched pairs were available. Orthogonal projections to latent structures (OPLS) models were calculated using the nonlinear iterative partial least-squares (NIPALS) algorithm36 with relevant modifications7 coded into Matlab. The Q2 parameter was computed by 7-fold cross validation and is a measure of the ability of the model to predict new values.37

’ RESULTS AND DISCUSSION Statistical Heterospectroscopy

Figure 1 shows an expanded region of a typical 1D 1H NMR spectra from a blood plasma and CSF sample from a patient with HIV-1 infection. Some of the major metabolites have been assigned based on known chemical shifts and J-coupling patterns. A number of metabolites are common to both biofluids, including glucose, creatine, citrate, pyruvate and several amino acids. A key difference is the presence of broad peaks from lipoproteins in blood plasma, but selective observation of smaller molecules was facilitated by use of relaxation editing pulse sequences (CarrPurcell-Meiboom-Gill, CPMG38) shown here. Standard 1H NMR spectra from CSF also contained broad resonances from proteins, but these were of negligible intensity relative to most metabolite peaks. Petroff et al first suggested that the measurement of paired serum and CSF samples by 1H NMR would permit “the rapid measurement of transport across the blood-brain barrier of a large range of dissimilar metabolites, and that such measurements could lead to further understanding of the biochemical condition of the brain in a variety of clinically important physiological states”.22 Statistical heterospectroscopy (SHY)14 provides an ideal approach to this kind of exploratory analysis, potentially offering unique insights into the dynamic interactions between blood and CSF metabolites. Previous applications of SHY have focused on integration of data from the same samples acquired on different analytical platforms (NMR with UPLCMS). Here, for the first time we demonstrate the utility of this approach through integration of data from the same individuals across different biofluids or tissues. A SHY matrix was initially constructed between 1H NMR data from 19 blood plasma (acquired using the CPMG pulse sequence) and 19 corresponding CSF samples from the same patients using the Pearson’s correlation coefficient. In Figure 2A, the output of a region of this “spectrum” between δ0.75 and δ1.25 has been displayed with the magnitude of the correlations indicated by the colormap on the right-hand side. Correlations with p > 0.001 were set to 0, and the blue traces above and to the left of this figure show the mean 1739

dx.doi.org/10.1021/pr1010263 |J. Proteome Res. 2011, 10, 1737–1745

Journal of Proteome Research

ARTICLE

Figure 2. Output from both Pearson’s and Spearman’s SHY analysis of CSF and blood plasma NMR data from 19 patients with HIV-1 infection. (A and B) Results from SHY computed using Pearson’s and Spearman’s correlation coefficients, respectively. Blue traces at the left and to the above of the A and B represent mean spectra from CSF and blood plasma, respectively. The colorbars on the right represent the correlation coefficient. PG = propylene glycol, val = valine. (C and D) X-Y scatter plots of intensities from selected resonances from valine and PG, respectively. Correlation coefficients (r) and p-values are also given.

spectra for the blood plasma and CSF data, respectively. The output showed high statistical correlations for valine (val) between the two biofluids, along with propylene glycol (PG), a drug delivery vehicle. Correlations between broad spectral features corresponding to lipoprotein and other protein signals were also apparent. By contrast, Figure 2B shows the output from the same SHY matrix computed using the nonparametric Spearman’s rank order correlation, a special case of Pearson’s correlation that assesses the relationship between two sets of ranked variables. The main difference between the plots is the absence of correlation observed for PG when computed using Spearman’s rank correlation. To explain these observations, it is instructive to plot the variable intensities at the sites of maximum correlation against each other. Figure C and D, respectively, plot the intensities variables corresponding to valine and PG in each biofluid. The Pearson’s and Spearman’s correlations are also given, with their associated probabilities. Figure 2C suggests a linear relationship between the biofluids for valine, with high correlations and statistical significance. In Figure 2D, the relationship between blood and CSF PG levels (at δ1.14 and δ1.139, respectively) was dominated by one participant with elevated high blood and CSF PG. Although this resulted in a significant

Pearson’s correlation, the Spearman’s rank correlation was much lower and less robust. PG is an effective drug delivery vehicle because of its rapid transfer across the BBB,22 and this participant’s medication included Kaletra (lopinavir and ritonavir), some formulations of which are known to contain PG, although the specific source of PG was not identified in this case. This result highlights a number of important aspects to interpretation of SHY data. When analyzed in isolation, such “outlier” peaks may have been interpreted as unexplained artifacts, but analysis by Pearson’s SHY reveal latent biological relationships relating to translocation across the BBB. When comparing computational methods, we can consider the Spearman’s F to be more robust, with high correlations reflective of endogenous relationships, while Pearson’s r is additionally sensitive to exogenous perturbations, and thus both approaches should be employed in a comprehensive analysis. A recent study of elderly people revealed an association between blood plasma metabolic profiles and mild cognitive impairment (MCI), an indicator of risk for Alzheimer’s disease (AD). One of the metabolites associated with MCI was valine.39 Here, we have shown a high correlation between blood and CSF valine for patients with HIV-1 infection. Under normal 1740



ARTICLE

Figure 3. Further output from both Pearson’s and Spearman’s SHY analysis of CSF and blood plasma NMR data from 19 patients with HIV-1 infection. (A and B) Region expanded around resonances from glutamine computed using Pearson’s and Spearman’s correlation, respectively. (C and D) Expandsion around the regions containing DMSO and TMAO computed using Pearson’s and Spearman’s correlation, respectively. “gln” = glutamine, DMSO2 = dimethylsulfone, TMAO = trimethylamine-N-oxide, MI = myo-inositol, MA = malonic acid, MAcorr = unidentified metabolite.

conditions, astrocytes have a high capacity for metabolism of branched chain amino acids, with ultimate conversion to tricarboxylic acid (TCA) cycle intermediates.40 These amino acids are elevated where brain metabolism is disrupted, such as in cases of complex fatigue,41 implicating them as potential nonspecific markers of slowed energy metabolism. Figure 2 A and B also show correlations between blood lipoprotein signals and wish signals near δ0.9 in the CSF spectra. Close inspection of the chemical shift regions indicated the apex of the correlation peaks corresponded to broad features in the CSF spectra, likely to be resonances from albumin, the most abundant protein in CSF.42 Lipoproteins are also found in the CSF but in levels below the detection limit for NMR.43 The appearance of these correlations is interesting given the relationship between lipoprotein deposition and vascular dementia, but the existence of a causal relationship will require further investigation. Figure 3 shows some other regions from both types of SHY. Figure 3A and B indicate a high correlation between glutamine in CSF and plasma by both Pearson’s and Spearman’s methods, respectively. Glutamine is the most common amino acid in CSF, and has many physiological roles. In the CNS glutamine is the precursor and degradation product of glutamate, the major excitatory neurotransmitter, acting as a vehicle for return of glutamate to the neurons in the “glutamate-glutamine cycle” as

well as the major source of the inhibitory neurotransmitter GABA.44 Outside the CNS glutamine also plays an important role as a nontoxic carrier of nitrogen and in interorgan nutrient balance in the so-called “Intercellular Glutamine Cycle”.45 Transport of glutamine across the BBB and blood-CSF barrier is facilitated by a number of transporters, primarily by system N. These may function either to transport glutamine from blood to brain or to clear glutamine from the CSF to blood. Xiang et al have suggested that the net function of these transporters is clearance of glutamine (and hence nitrogen) from the brain to the blood via the CSF.46 Thus our findings suggest blood glutamine levels may be a potential indicator of CSF and hence brain glutamine clearance in HIVþ patients. A number of strong correlations were also observed in the region between δ3.10 and δ3.35, shown in Figure 3C and D (Pearson’s and Spearman’s, respectively). The correlation observed at δ3.15 in both biofluids was from a relatively narrow singlet, and we have assigned this to dimethyl sulfone (DMSO2). Although in CSF this peak is coresonant with a triplet from citrulline, STOCSY analysis was able to discriminate between this triplet and the singlet from DMSO2 (data not shown). DMSO2 may originate either from dietary sources, bacterial metabolism or endogenous methanethiol metabolism, and has been observed previously by NMR in both CSF and blood plasma.47 We also observed high correlations between the 1741



ARTICLE

Table 1. Results from OPLS Models of CSF Resonance Intensities Regressed against Blood Plasma NMR Dataa metabolite

a

selected resonance intensity (ppm) experiment maximum Q2

number of orthogonal components blood metabolites identified from loadings þvaline

valine

1.04

CPMG

0.203

4

ethanol

1.19

CPMG

0.0475

4

n/a

lactate

1.33

1D

0.4898

4

-Choline (in lipid), þMI

alanine

1.49

1D

0.1565

3

n/a

acetate

1.92

DE

0.1477

4

n/a

NAA

2.05

DE

0.2908

4

-total lipid content

aoetoacetate

2.24

CPMG

0.3268

4

þacetoacetate, þglucose

pyruvate glutamine

2.38 2.46

CPMG 1D

-0.0512 0.4613

4 4

n/a þglutamine, þalbumin

citrate

2.56

1D

0.3136

2

-glyceryl of lipids, -choline (lipid)

creatine

3.05

1D

0.409

4

-HDL(chol), -LDL, -choline (lipid)

creatinine

3.05

CPMG

0.574

4

þcreatinine, þTMAO

TMAO

3.27

CPMG

0.2592

4

þTMAO, þcreatinine

MI

3.30

CPMG

0.2398

4

þMI, þTMAO, þacetate, þcreatinine

“Maximum” Q2 indicates the highest Q2 value achieved after seven-fold cross validation with sequential removal of orthogonal components.

biofluids for a peak at δ3.11. Inspection of the spectra indicated this was a narrow singlet appearing in several paired CSF-plasma samples, and STOCSY analysis revealed this was highly correlated with another narrow resonance at δ3.30. On the basis of chemical shift, J-coupling, pattern and line width, we have assigned the resonance at δ3.11 to malonic acid (MA); while the resonance at δ3.30 (here denoted MAcorr) is yet to be assigned, it is likely to be in the same biochemical pathway as MA and is the subject of ongoing research. The absence of correlations observed in SHY can be just as informative as their presence. The NMR spectra from a number of CSF samples indicated the presence of alcohols including ethanol and methanol. Ethanol is readily detected in biological fluids following consumption of alcoholic beverages, and even penetrates as far as seminal fluid,48 but there is evidence that it can be produced endogenously.49 Interestingly, we did not observe any correlation between the biofluids for ethanol, despite its known rapid translocation across the BBB.50 Although detected in several CSF samples it was not present in any corresponding blood plasma samples. We found no relationship between reported alcohol consumption and presence of ethanol in the spectra, providing further evidence that ethanol may be endogenously produced in CSF. Furthermore, we observed methanol (singlet at δ3.36) in all our CSF samples, but no correlation between blood and CSF for this molecule. This observation is consistent with a previous report that identified methanol in 100% of CSF samples from patients screened for meningitis.19 Methanol may be produced by breakdown of ethanol by gut microflora or through C1 metabolism in the liver.51 There is also evidence that methanol may be produced by enzymatic breakdown of S-adenosylmethionine in the pituitary gland.52 Whether this pathway produces methanol in sufficient quantities to explain our observations remains to be established. Prediction of CSF Metabolite Concentrations from Blood Plasma NMR Data

Although analysis of CSF gives valuable information regarding brain metabolic status, collection by lumbar puncture is highly invasive, requiring local anesthetic. It would be of more clinical benefit if concentrations of key metabolites in CSF could be “predicted” from a more readily collected biofluid such as blood plasma. To investigate this, we have applied orthogonal

projections to latent structures (OPLS) to regress the intensity (which is directly proportional to the concentration) of known resonances in the CSF data (often denoted the “Y” matrix) against blood plasma NMR data (the “X” matrix). The parameter Q2Y was calculated for each model by 7-fold cross validation. Table 1 lists the metabolites selected, along with the chemical shift of the peak used to generate the intensity vector. The “Maximum Q2” was the largest Q2Y value computed across all three NMR data types acquired from blood plasma, indicated in the “Experiment” column. The final column lists the metabolites that were most influential in the model with the highest Q2Y, identified by inspection of the OPLS loadings, if Q2Y exceeded 0.2. CSF metabolites such as valine, myo-inositol and glutamine, identified by SHY as correlated with each other across the two biofluids, were also predicted from blood plasma data using OPLS. Despite being observed in several samples, no predictable model could be generated to predict ethanol in CSF, again suggesting its presence may be the result of endogenous production rather than prior consumption. CSF lactate (measured by NMR) was correlated with myoinositol and inversely correlated with the lipid-based choline resonance at δ3.21, which comes mostly from the choline in HDL.53 CSF lactate generally reflects the ability to clear lactate from the brain. It is often used as an indicator of cerebral anoxia or hypoxia,54 and correlates with cognitive decline in dementia patients.55 Recent evidence suggests blood HDL-cholesterol is associated with decline in memory in midlife.31 Thus, the inverse relationship between CSF lactate and lipid choline in blood may be indicative of a common physiological source rather than a direct link across the BBB. Several other key CSF metabolites were predicted from the blood plasma CPMG data, such as CSF creatine and creatinine, indicators of brain metabolism,56 and citrate, which is associated with depression.57 Although the sample size of the present study was small, these results indicate that expansion of these data would permit the construction of robust models for predicting neurological changes from the less invasive approach of blood metabolic profiling. Integration of MRS and NMR Data 1

H MRS permits the direct quantitative measurement of brain metabolites in vivo. Spectra were acquired from three brain 1742



ARTICLE

Figure 4. Output from Pearson’s SHY matrix constructed from MRS-derived metabolite ratio data with CSF NMR data. The NMR spectral regions around the resonances from 3-hydroxyisovalerate have been expanded to reveal correlations with FWM metabolites.

regions from each participant in this study, providing an opportunity to logically extend the SHY concept to investigate the covariance between biofluid and cerebral metabolites. We constructed Pearson’s and Spearman’s SHY matrices from both blood plasma and CSF data with MRS data set consisting of 26 variables corresponding to metabolite ratios from the frontal white matter (FWM), posterior cingulate gyri (Pcing) and right caudate (RT Caud). The most significant result was the observation of high correlations between CSF 3-hydroxyisovalerate (3HV) with FWM choline/NAA ratio, and anticorrelation with FWM NAA/creatine ratio. 3HV is most likely a product of leucine degradation, and has been observed previously to be associated with decreased NAA/choline ratio.58 Our results are consistent with those observations, as we have observed an inverse relationship between 3HV and FWM NAA, implicating this molecule as a potential biomarker for disrupted neuronal metabolism in these patients. We found no significant correlations between blood plasma and MRS profiles, possibly because of the greater “physiological distance” between the brain and the blood plasma, compared to the more closely interacting CSF (Figure 4).

’ CONCLUSIONS In this paper, we have demonstrated the first statistical integration of magnetic resonance data from biofluids from the same individuals, affording the opportunity to extract unique biological information about the distribution of metabolites within an organism. By analyzing blood and CSF, we can begin to build a model for the nature of the interactions between the central nervous system and the rest of the body that can serve as a template for the application of metabonomic studies to neurological disorders. Although the application of metabonomics to address problems in neuroscience is increasing, it is often difficult to interpret results from studies that have measured metabolites in biofluids such as blood plasma or urine in the context of the

underlying etiology of disease. This is due to biocomplexity and the “tip of the iceberg” information revealed by NMR.35 Our detailed analysis of the metabolic relationships between blood plasma and CSF can go some way toward interpreting results from other studies of neurological disorders.

’ AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected], ph þ61 2 9399 1136, fax þ61 2 9399 1005. Present Addresses

Department of Primary Industries, 1 Park Drive, Bundoora, Victoria 3083, Australia. E-mail: [email protected]

’ ACKNOWLEDGMENT This work was supported by the Australian National Health and Medical Research Council (Project grant 568746 (CIA-LC) and CR, and Fellowship to C.D.R.) and by Brain Sciences UNSW (Fellowship to L.A.C.). The MRUI software package was kindly provided by the participants of the EU Network programmes: Human Capital and Mobility, [CHRX-CT94-0432] and Training and Mobility of Researchers, [ERB-FMRX-CT970160]. We are grateful to Kirsten Moffat and radiography staff, the nursing research staff and the Immunology department at St Vincent’s Hospital, Sydney and to the staff of the UNSW Mark Wainwright Analytical Centre, UNSW for expert technical assistance. We also thank Dr. Nicholas Davies, Welcome Trust Neurology Fellow at St. Vincent’s Hospital who performed the lumbar puncture and Dr. Louise Pemberton for managing the laboratory database at the St. Vincent’s Hospital Applied Medical Research Center. ’ REFERENCES (1) Holmes, E.; Wilson, I. D.; Nicholson, J. K. Metabolic phenotyping in health and disease. Cell 2008, 134 (5), 714–7. 1743


Journal of Proteome Research (2) Goodacre, R.; Vaidyanathan, S.; Dunn, W. B.; Harrigan, G. G.; Kell, D. B. Metabolomics by numbers: acquiring and understanding global metabolite data. Trends Biotechnol. 2004, 22 (5), 245–52. (3) Holmes, E.; Loo, R. L.; Stamler, J.; Bictash, M.; Yap, I. K.; Chan, Q.; Ebbels, T.; De Iorio, M.; Brown, I. J.; Veselkov, K. A.; Daviglus, M. L.; Kesteloot, H.; Ueshima, H.; Zhao, L.; Nicholson, J. K.; Elliott, P. Human metabolic phenotype diversity and its association with diet and blood pressure. Nature 2008, 453 (7193), 396–400. (4) Clayton, T. A.; Lindon, J. C.; Cloarec, O.; Antti, H.; Charuel, C.; Hanton, G.; Provost, J. P.; Le Net, J. L.; Baker, D.; Walley, R. J.; Everett, J. R.; Nicholson, J. K. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 2006, 440 (7087), 1073–7. (5) Kristal, B. S.; Shurubor, Y. I. Metabolomics: Opening Another Window into Aging. Sci. Aging Knowl. Environ. 2005, 2005 (26), pe19. (6) Trygg, J.; Holmes, E.; Lundstedt, T. Chemometrics in metabonomics. J. Proteome Res. 2007, 6 (2), 469–79. (7) Trygg, J.; Wold, S. Orthogonal projections to latent structures (O-PLS). J. Chemomet. 2002, 16 (3), 119–28. (8) Noda, I.; Dowrey, A. E.; Marcott, C.; Story, G. M.; Ozaki, Y. Generalized two-dimensional correlation spectroscopy. Appl. Spectrosc. 2000, 54 (7), 236A–48A. (9) Cloarec, O.; Dumas, M. E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E.; Nicholson, J. Statistical total correlation spectroscopy: an exploratory approach for latent biomarker identification from metabolic 1H NMR data sets. Anal. Chem. 2005, 77 (5), 1282–9. (10) Holmes, E.; Cloarec, O.; Nicholson, J. K. Probing latent biomarker signatures and In Vivo pathway activity in experimental disease states via statistical total correlation spectroscopy (STOCSY) of biofluids: application to HgCl2 toxicity. J. Proteome Res. 2006, 5 (6), 1313–20. (11) Smith, L. M.; Maher, A. D.; Cloarec, O.; Rantalainen, M.; Tang, H.; Elliott, P.; Stamler, J.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. Statistical correlation and projection methods for improved information recovery from diffusion-edited NMR spectra of biological samples. Anal. Chem. 2007, 79 (15), 5682–9. (12) Coen, M.; Hong, Y. S.; Cloarec, O.; Rhode, C. M.; Reily, M. D.; Robertson, D. G.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Heteronuclear 1H-31P statistical total correlation NMR spectroscopy of intact liver for metabolic biomarker assignment: application to galactosamineinduced hepatotoxicity. Anal. Chem. 2007, 79 (23), 8956–66. (13) Keun, H. C.; Athersuch, T. J.; Beckonert, O.; Wang, Y.; Saric, J.; Shockcor, J. P.; Lindon, J. C.; Wilson, I. D.; Holmes, E.; Nicholson, J. K. Heteronuclear 19F-1H statistical total correlation spectroscopy as a tool in drug metabolism: study of flucloxacillin biotransformation. Anal. Chem. 2008, 80 (4), 1073–9. (14) Crockford, D. J.; Holmes, E.; Lindon, J. C.; Plumb, R. S.; Zirah, S.; Bruce, S. J.; Rainville, P.; Stumpf, C. L.; Nicholson, J. K. Statistical heterospectroscopy, an approach to the integrated analysis of NMR and UPLC-MS data sets: application in metabonomic toxicology studies. Anal. Chem. 2006, 78 (2), 363–71. (15) Crockford, D. J.; Maher, A. D.; Ahmadi, K. R.; Barrett, A.; Plumb, R. S.; Wilson, I. D.; Nicholson, J. K. 1H NMR and UPLC-MS(E) statistical heterospectroscopy: characterization of drug metabolites (xenometabolome) in epidemiological studies. Anal. Chem. 2008, 80 (18), 6835–44. (16) Saric, J.; Li, J. V.; Utzinger, J.; Wang, Y.; Keiser, J.; Dirnhofer, S.; Beckonert, O.; Sharabiani, M. T.; Fonville, J. M.; Nicholson, J. K.; Holmes, E. Systems parasitology: effects of Fasciola hepatica on the neurochemical profile in the rat brain. Mol. Syst. Biol. 2010, 6, 396. (17) Yap, I. K. S.; Li, J. V.; Saric, J.; Martin, F.-P.; Davies, H.; Wang, Y.; Wilson, I. D.; Nicholson, J. K.; Utzinger, J.; Marchesi, J. R.; Holmes, E. Metabonomic and Microbiological Analysis of the Dynamic Effect of Vancomycin-Induced Gut Microbiota Modification in the Mouse. J. Proteome Res. 2008, 7 (9), 3718–3728. (18) Quinones, M. P.; Kaddurah-Daouk, R. Metabolomics tools for identifying biomarkers for neuropsychiatric diseases. Neurobiol. Dis. 2009, 35 (2), 165–176.

ARTICLE

(19) Wishart, D. S.; Lewis, M. J.; Morrissey, J. A.; Flegel, M. D.; Jeroncic, K.; Xiong, Y.; Cheng, D.; Eisner, R.; Gautam, B.; Tzur, D.; Sawhney, S.; Bamforth, F.; Greiner, R.; Li, L. The human cerebrospinal fluid metabolome. J. Chromatogr., B 2008, 871 (2), 164–73. (20) Smith, D. E.; Johanson, C. E.; Keep, R. F. Peptide and peptide analog transport systems at the blood-CSF barrier. Adv. Drug Delivery Rev. 2004, 56 (12), 1765–91. (21) Pardridge, W. M. Brain metabolism: a perspective from the blood-brain barrier. Physiol. Rev. 1983, 63 (4), 1481–535. (22) Petroff, O. A.; Yu, R. K.; Ogino, T. High-resolution proton magnetic resonance analysis of human cerebrospinal fluid. J. Neurochem. 1986, 47 (4), 1270–6. (23) Sweatman, B. C.; Farrant, R. D.; Holmes, E.; Ghauri, F. Y.; Nicholson, J. K.; Lindon, J. C. 600 MHz 1H-NMR spectroscopy of human cerebrospinal fluid: effects of sample manipulation and assignment of resonances. J. Pharm. Biomed. Anal. 1993, 11 (8), 651–64. (24) Wevers, R. A.; Engelke, U.; Wendel, U.; Dejong, J. G. N.; Gabreels, F. J. M.; Heerschap, A. Standardized method for highresolution 1H NMR of cerebrospinal-fluid. Clin. Chem. 1995, 41 (5), 744–51. (25) Wishart, D. S.; Lewis, M. J.; Morrissey, J. A.; Flegel, M. D.; Jeroncic, K.; Xiong, Y. P.; Cheng, D.; Eisner, R.; Gautam, B.; Tzur, D.; Sawhney, S.; Bamforth, F.; Greiner, R.; Li, L. The human cerebrospinal fluid metabolome. J. Chromatogr., B 2008, 871 (2), 164–73. (26) Coen, M.; O’Sullivan, M.; Bubb, W. A.; Kuchel, P. W.; Sorrell, T. Proton Nuclear Magnetic Resonance-Based Metabonomics for Rapid Diagnosis of Meningitis and Ventriculitis. Clin. Infec. Dis. 2005, 41 (11), 1582–90. (27) Holmes, E.; Tsang, T. M.; Huang, J. T.; Leweke, F. M.; Koethe, D.; Gerth, C. W.; Nolden, B. M.; Gross, S.; Schreiber, D.; Nicholson, J. K.; Bahn, S. Metabolic profiling of CSF: evidence that early intervention may impact on disease progression and outcome in schizophrenia. PLoS Med. 2006, 3 (8), e327. (28) Berger, J. R.; Avison, M. The blood brain barrier in HIV infection. Front. Biosci. 2004, 9, 2680–5. (29) Harrington, P. R.; Schnell, G.; Letendre, S. L.; Ritola, K.; Robertson, K.; Hall, C.; Burch, C. L.; Jabara, C. B.; Moore, D. T.; Ellis, R. J.; Price, R. W.; Swanstrom, R. Cross-sectional characterization of HIV-1 env compartmentalization in cerebrospinal fluid over the full disease course. AIDS 2009, 23 (8), 907–15. (30) Vanhamme, L.; van den Boogaart, A.; Van Huffel, S. Improved Method for Accurate and Efficient Quantification of MRS Data with Use of Prior Knowledge. J. Magn. Reson. 1997, 129 (1), 35–43. (31) Singh-Manoux, A.; Gimeno, D.; Kivimaki, M.; Brunner, E.; Marmot, M. G. Low HDL cholesterol is a risk factor for deficit and decline in memory in midlife - The Whitehall II study. Arterioscler. Thromb. Vasc. Biol. 2008, 28 (8), 1556–62. (32) Beckonert, O.; Keun, H. C.; Ebbels, T. M.; Bundy, J.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat. Protoc. 2007, 2 (11), 2692–703. (33) Wu, D. H.; Chen, A. D.; Johnson, C. S. An Improved DiffusionOrdered Spectroscopy Experiment Incorporating Bipolar-Gradient Pulses. J. Magn. Reson., Ser. A 1995, 115 (2), 260–64. (34) Dieterle, F.; Ross, A.; Schlotterbeck, G.; Senn, H. Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics. Anal. Chem. 2006, 78 (13), 4281–90. (35) Ala-Korpela, M. Critical evaluation of 1H NMR metabonomics of serum as a methodology for disease risk assessment and diagnostics. Clin. Chem. Lab. Med. 2008, 46 (1), 27–42. (36) Geladi, P.; Kowalski, B. R. Partial least-squares regression: a tutorial. Anal. Chim. Acta 1986, 185, 1–17. (37) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Wold, S. Multiand Megavariate Data Analysis: Principles and Applications; Umetrics: Umea, 2001. (38) Carr, H. Y.; Purcell, E. M. Effects of Diffusion on Free Precession in Nuclear Magnetic Resonance Experiments. Phys. Rev. 1954, 94 (3), 630. 1744



ARTICLE

(39) Tukiainen, T.; Tynkkynen, T.; Makinen, V. P.; Jylanki, P.; Kangas, A.; Hokkanen, J.; Vehtari, A.; Grohn, O.; Hallikainen, M.; Soininen, H.; Kivipelto, M.; Groop, P. H.; Kaski, K.; Laatikainen, R.; Soininen, P.; Pirttila, T.; Ala-Korpela, M. A multi-metabolite analysis of serum by 1H NMR spectroscopy: early systemic signs of Alzheimer’s disease. Biochem. Biophys. Res. Commun. 2008, 375 (3), 356–61. (40) Murín, R.; Mohammadi, G.; Leibfritz, D.; Hamprecht, B. Glial Metabolism of Valine. Neurochem. Res. 2009, 34 (7), 1195–203. (41) Jin, G.; Kataoka, Y.; Tanaka, M.; Mizuma, H.; Nozaki, S.; Tahara, T.; Mizuno, K.; Yamato, M.; Watanabe, Y. Changes in plasma and tissue amino acid levels in an animal model of complex fatigue. Nutrition 2009, 25 (5), 597–607. (42) Thompson, E. J. Proteins of the Cerebrospinal Fluid: analysis and interpretation in the Diagnosis and Treatment of Neurological Disease; Elsevier Academic Press: San Diego, CA, 2005. (43) Albers, J.; Tollefson, J.; Wolfbauer, G.; Albright, R. Cholesteryl ester transfer protein in human brain. Int. J. Clin. Lab. Res. 1992, 21 (2), 264–6. (44) Rae, C.; Hare, N.; Bubb, W. A.; McEwan, S. R.; Br€oer, A.; McQuillan, J. A.; Balcar, V. J.; Conigrave, A. D.; Br€oer, S. Inhibition of glutamine transport depletes glutamate and GABA neurotransmitter pools: further evidence for metabolic compartmentation. J. Neurochem. 2003, 85 (2), 503–14. (45) McGivan, J. D., Metabolism of glutamine and Glutamate in the liver - regulation and physiological significance. In Glutamine and Glutamate in Mammals; Elling, K., Ed.; CRC Press: Boca Raton, FL, 1998; Vol. 1, pp 183-201. (46) Xiang, J.; Ennis, S. R.; Abdelkarim, G. E.; Fujisawa, M.; Kawai, N.; Keep, R. F. Glutamine transport at the blood-brain and bloodcerebrospinal fluid barriers. Neurochem. Int. 2003, 43 (4-5), 279–88. (47) Engelke, U. F.; Tangerman, A.; Willemsen, M. A.; Moskau, D.; Loss, S.; Mudd, S. H.; Wevers, R. A. Dimethyl sulfone in human cerebrospinal fluid and blood plasma confirmed by one-dimensional 1 H and two-dimensional 1H-13C NMR. NMR Biomed. 2005, 18 (5), 331–6. (48) Lynch, M. J.; Masters, J.; Pryor, J. P.; Lindon, J. C.; Spraul, M.; Foxall, P. J.; Nicholson, J. K. Ultra high field NMR spectroscopic studies on human seminal fluid, seminal vesicle and prostatic secretions. J. Pharm. Biomed. Anal. 1994, 12 (1), 5–19. (49) Meshitsuka, S.; Morio, Y.; Nagashima, H.; Teshima, R. 1H NMR studies of cerebrospinal fluid: endogenous ethanol in patients with cervical myelopathy. Clin. Chim. Acta 2001, 312 (1-2), 25–30. (50) Abbott, N. J.; Ronnback, L.; Hansson, E. Astrocyte-endothelial interactions at the blood-brain barrier. Nat. Rev. Neurosci. 2006, 7 (1), 41–53. (51) Haffner, H. T.; Graw, M.; Besserer, K.; Blickle, U.; Henssge, C. Endogenous methanol: variability in concentration and rate of production. Evidence of a deep compartment? Forensic Sci Int. 1996, 79 (2), 145–54. (52) Axelrod, J.; Daly, J. Pituitary Gland: Enzymic Formation of Methanol from S-Adenosylmethionine. Science 1965, 150 (3698), 892–3. (53) Nicholson, J. K.; Foxall, P. J.; Spraul, M.; Farrant, R. D.; Lindon, J. C. 750 MHz 1H and 1H-13C NMR spectroscopy of human blood plasma. Anal. Chem. 1995, 67 (5), 793–811. (54) Gladden, L. B. Lactate metabolism: a new paradigm for the third millennium. J. Physiol. 2004, 558 (Pt 1), 5–30. (55) Gottfries, C. G.; Kjallquist, A.; Ponten, U.; Roos, B. E.; Sundbarg, G. Cerebrospinal Fluid pH and Monoamine and Glucolytic Metabolites in Alzheimer’s Disease. Brit. J. Psych. 1974, 124 (580), 280–7. (56) Ågren, H.; Niklasson, F. Creatinine and creatine in CSF: indices of brain energy metabolism in depression. J. Neural Transmiss. 1988, 74 (1), 55–9. (57) Mellerup, E. T.; Rafaelsen, O. J. Depression and cerebrospinal fluid citrate. Acta Psych. Scand. 1981, 63 (1), 57–60. (58) Engelke, U. F.; Kremer, B.; Kluijtmans, L. A.; van der Graaf, M.; Morava, E.; Loupatty, F. J.; Wanders, R. J.; Moskau, D.; Loss, S.; van den Bergh, E.; Wevers, R. A. NMR spectroscopic studies on the late onset form of 3-methylglutaconic aciduria type I and other defects in leucine metabolism. NMR Biomed. 2006, 19 (2), 271–8. 1745


Statistical Integration of 1H NMR and MRS Data from Different

Recommend Documents