1H NMR Spectroscopic Analysis

Frozen plasma samples were thawed and then centrifuged at 12,000g for 5 minutes at 4°C to remove insoluble material. Individual samples were prepared by mixing 300µl aliquots with an equal amount of 0.075 M NaH2PO4 buffer (pH 7.4) containing sodium 3-(trimethylsilyl) propionate-2,2,3,3-d4 (TSP) for the chemical shift reference and sodium azide as preservative. This mixture was then transferred to 5mm SampleJet NMR tubes (BrukerBiospin Ltd.) for spectroscopic analysis.

A standard 1D 1H NMR experiment, the Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence (with the form RD-90º-(t-180º-t)n-ACQ) was performed to attenuate peaks from larger and slower moving molecules (such as lipoproteins, and lipids) with a view to identifying low molecular weight metabolites. A relaxation delay (RD) of 4s, a mixing time of 0.01 seconds, a spin-echo delay of 0.3 ms, 128 loops and an FID acquisition time (ACQ) of 3.067s were applied. A total of 32 scans were recorded into 73k data points with a spectral width of 20 ppm. Additionally standard 1D pulse sequence without spin echo modulation and a 2-D J-resolved pulse sequence were carried out following in house protocols for the purpose of metabolite identification (1). The instrument receiver gain (RG) was set at 90.5 for all the experiments.

Data Reduction and Multivariate Analysis

The plasma spectra were referenced to the anomeric proton assigned to α-glucose at δ 5.22 (2) and imported to python or MATLABTM for further analysis. Regions containing only noise (bellow 0.3 and above δ 10) or corresponding to the region containing the water peak (δ 4.3 to 4.925) were removed. Both unsupervised (principal component analysis, PCA) and supervised multivariate data analysis methods (orthogonal partial least squares regression discriminant analysis (OPLS/OPLS-DA (3)) were employed to visualize and interpret experimental differences, using in-house MATLABTM (for partial least squares models) scripts and scikit-learn for PCA(4). For multivariate statistical analysis all data were mean-centered and unit-variance scaled to equally weight changes in low and high abundance metabolites. Models were also fitted using total area normalized and probabilistic quotient normalized data (using the median spectrum as the reference) (5), and with no scaling or pareto scaling (variables scaled by the square root of their standard deviation). Since no qualitative difference was observed in the statistical outcome when compared with the unit-variance scaled and non-normalized models, results presented and their interpretation are made based on univariate and non-normalized data. All analysis reported was performed on the CPMG pulse sequence dataset. For deconvolution and relative metabolite quantification in the spectra the R(6) package BATMAN (BayesianAuTomated Metabolite Analyser for NMR spectra) was employed(7)

Peak fitting

Specific templates for the metabolites were generated, with their chemical shift priors adjusted for ranges observed in this dataset and other human plasma samples. The quality of the fit was assessed visually and through comparison of the results between multiple NMR signals of the same molecule where possible. Since the plasma profiles of the children were compromised to various degrees by drugs, feeds and other interventions, analysis of the global profile was inappropriate. Therefore, we selected 15 metabolites (acetate, acetoacetate, acetone, alanine, citrate, creatine, creatinine, formate, glucose, 3-hydroxybutyrate, isoleucine, lactate, leucine, threonine, and, valine) on the basis of existing literature on the metabolic response to critical illness, and investigated them in detail in our patient cohort. The relative concentrations values for these metabolites were used alongside the cytokine values to obtain the Pearson correlation matrices and to plot the heat maps in python (8-10) the plotting library matplotlib (11)

  1. Dona AC, Jimenez B, Schaefer H, Spraul M, Lewis MR, Pearce JT, Holmes E, Lindon JC, Nicholson JK. Precision high throughput proton nmr spectroscopy of human urine, serum and plasma for large-scale metabolic phenotyping. Analytical chemistry 2014;Sep (epub ahead of print).
  2. Pearce JT, Athersuch TJ, Ebbels TM, Lindon JC, Nicholson JK, Keun HC. Robust algorithms for automated chemical shift calibration of 1d 1h nmr spectra of blood serum. Analytical chemistry 2008;80:7158-7162.
  3. Python language reference. In: foundation Ps, editor.
  4. Trygg J, Wold S. Orthogonal projections to latent structures (o-pls). J Chemom2002;16:119-128.
  5. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M. Scikit-learn: Machine learning in python JMLR 2011;12:2825-2830.
  6. 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 nmrmetabonomics. Analytical chemistry 2006;78:4281-4290.
  7. Team RDC. R: A language and environment for statistical computing. . Vienna, Austria: R Foundation for Statistical Computing; 2013.
  8. Hao J, Astle W, De Iorio M, Ebbels TM. Batman—an r package for the automated quantification of metabolites from nuclear magnetic resonance spectra using a bayesian model Bioinformatics 2012;28:2088-2090.
  9. Van der Walt S, Colbert SC, Varoquax G. The numpy array: A structure for efficient numerical computation. . ComputSciEng2011;13:22-30.
  10. Jones E, Oliphant E, Peterson P. Scipy: Open source scientific tools for python. 2001.
  11. Hunter JD. Matplotlib: A 2d graphics environment. ComputSciEng2007;9:90-95.