Please, cite this online document as:
Mari S., Mannella V., Quilici G., Musco G., Cobas C., Sykora S.,
   GSD (Global Spectrum Deconvolution) in Metabolomics: Absolute Quantification and GSD Binning,
   Poster at 51st ENC Conference, Daytona Beach (FL, USA), April 18-23, 2010, DOI: 10.3247/SL3Nmr10.002.

Abstract

NMR spectroscopy is an important tool in many metabolomic applications. Its potential capability to handle complex mixtures of metabolites makes it a prime choice in both identification and quantification of the multitude of different species constituting unprocessed biological mixtures. In quantitative metabolomics (targeted metabolomic profiling) all detectable compounds in biosample are identified and quantified by comparing the biosample spectrum to a library of reference spectra of pure compounds [1]. The underlying assumption is that a spectrum is the weighed sum of the spectra of all the individual metabolites which constitute the mixture (spectral additivity).

In practice, the inverse problem of decomposing an experimental spectrum into its component parts corresponding to individual metabolites is arduous due to their large number and the complexity of their spectra, massive overlap of spectral peaks, nontrivial deviations of peak shapes from the ideal Lorentzian profile [2], lack of a suitable orthonormal base in the vector space spanned by sets of Lorentzian peaks, and presence of artifacts such as receiver noise, irregular baseline drifts, and magnetic field inhomogeneity effects. Due to all these factors, attempts to overcome peaks overlap problems by means of conventional deconvolution (fitting) of selected spectral areas have in general limited success.

For this reasons, and because the actual nature of all the metabolites is rarely known in advance, metabolomics often uses alternative statistical evaluation methods, such as multivariate factor analysis [3], which sidestep the need for a complete interpretation of the spectra and a full solution of the inverse problem, while still permitting the correlation of the spectra with specific biological aspects. However, such approaches require integration over predefined intervals (bins) and a meaningful integration of such intricate and artifact-burdened spectra may often be just as arduous as peaks fitting.

Recently, a new algorithm called GSD (Global Spectrum Deconvolution) has been developed [4] and made available in the Mnova software package of Mestrelab. GSD is capable of identifying even poorly resolved spectral peaks and of fitting all recognizable peaks in even a very complex 1D spectrum in a surprisingly short time (typically a dozen seconds for up to 1000 peaks). Moreover, it is fully automatic and objective (no human intervention is required) and produces a table of all detectable spectral peaks and their parameters. Such a table can be then used for various purposes such as generation of artifact-free synthetic spectra (with or without resolution enhancement), stick spectra, artifact-free integrals, as well as accurate binning void of any bin-crossover problems due to the overlapping wings of spectral peaks. Because of these attractive features, GSD is likely to become a very important pre-processing tool for all metabolomic approaches to the evaluation of NMR spectra of whole biosamples.

Here we present the first controlled tests of the applicability of GSD to the multivariate analysis of complex metabolite mixtures such as cell culture media containing serum and mouse urine. GSD allows the user to generate synthetic spectra that can be used in the subsequent steps of multivariate statistical analysis. In particular, we compare the quality of PCA plots obtained by means of conventional binning of experimental spectra (NOESYPR or CPMG) with PCA plots obtained by binning the synthetic spectra as well as direct GSD-binning based just on the GSD peaks table.

In addition, we have taken advantage of GSD to implement under Mnova the quantitative referencing strategy known as PULCON [5]. It is well known that when dealing with biological samples containing lipids and proteins, one can not use an internal standard to perform absolute quantification of metabolites. This is due to the fact that all possible reference compounds (such as TSP or DSS) interact with large biological molecules, making the quantification error prone. By combining the GSD algorithm with a PULCON script we are able to deconvolute overlapped regions and perform absolute quantification even of metabolites whose peaks are buried under these areas. In order to verify the quality of the quantification, we have acquired spectra of a commercial cell culture medium which is a complex but well known mixture of more than 20 compounds. We then compare the metabolite concentrations obtained from experimental spectra by means of the GSD - PULCON algorithm with those declared by the medium vendor.

References and links:

[1] Wishart D.S., Quantitative metabolomics using NMR,
      TrAC 27, 228-237 (2008). DOI: 10.1016/j.trac.2007.12.001.
[2] Sykora S., Cobas C., Peak Shapes in NMR Spectroscopy,
      a presentation at the 2009 MMCE (Magnetic Moments in Central Europe) meeting (Otocec, Slovenia, February 11-15, 2009).
[3] Issaq HG.J., Van Q.N., Waybright T.J., Muschik G.M., Veenstra T.D.,
      Statistical and analytical approaches to metabolomics research,
      J.Sep.Sci. 32, 2183-2199 (2009). DOI: 10.1002/jssc.200900152.
[4] Cobas C., Sykora S., The Bumpy Road towards Automatic Global Spectral Deconvolution (GSD),
      a poster at the 50th ENC Conference (March 29 - April 3, 2009, Asilomar, California, USA). DOI: 10.3247/SL3Nmr09.003.
[5] Wider G., Dreier L., Measuring Protein Concentrations by NMR Spectroscopy,
      JACS 128, 2571-2576 (2006). DOI: 10.1021/ja055336t. See also this link.

Discussions

Your comments are welcome and will appear here