Statistical two-dimensional correlation spectroscopy of urine and serum from metabolomics data

Statistical two-dimensional correlation spectroscopy combined with pattern recognition is demonstrated for coanalysis of NMR spectroscopic data from different sources. The urine and serum 1H NMR spectra from metabolomics datasets of diabetes and hyperthyroidism are taken as examples. The intrinsic covariance of certain molecules between urine and serum spectra is identified. The highly urine-serum-correlated metabolites are further analyzed by using the projection to latent structure discriminant analysis (PLS-DA) method. To illustrate the applicability of the method, the metabolomics datasets of diabetes and hyperthyroidism are imported separately to calculate the corresponding two-dimensional urine-serum correlation coefficient matrixes. The results show that creatinine (δ 4.08) and succinate (δ 2.45) are found to be highly correlated between urine and serum from diabetes patients, and choline (δ 3.21) and pyruvate (δ 2.33) are highly correlated between urine and serum from hyperthyroidism patients. This study offers a new angle of view for interpreting metabolomics data and demonstrates the potential of the correlation analysis of spectra from different biological sources as a new systems biology tool.

► Urine and serum NMR data were coanalyzed by statistical correlation spectroscopy.
► Intrinsic covariance of molecules between urine and serum spectra was identified.
► The ‘highlighted’ metabolites were analyzed by PLS-DA regression coefficient plot.
► This work provides a new interpretation of metabolomic data.