Isotopic ratio outlier analysis improves metabolomics prediction of nitrogen treatment in maize.

ABSTRACT

We evaluated Isotope Ratio Outlier Analysis (IROA) as a metabolome-wide internal standard approach to improve the quality of LC/MS data collected from a large-scale greenhouse experiment designed to metric the ability of metabolomics to model quantitatively nitrogen treatments. We further looked at how IROA would be incorporated into a metabolomics workflow. For this we compared IROA processed data with that generated without the benefit of metabolome-wide internal standards using our current tool, Genedata Expressionist, from the same raw LC/MS data files. In our experiment, 367 maize plants were grown from kernel in a greenhouse under controlled conditions. Plants were treated from germination on with varying concentrations of nutrient nitrogen as one (treatment) variable. A second variable was the presence of one of two transgenes. Metabolomics analysis of leaves was performed by LC/MS positive and negative electrospray ionization modes, and raw data were processed with both our routine and IROA protocols. IROA data analysis detected 184 metabolites in each ionization mode. Analysis without IROA yielded 281 metabolites in positive ionization mode and 172 in negative ionization mode. Data from both protocols were normalized for sample dry weight, location in the greenhouse, extraction batch, sample run order, and internal standard. Normalized results were subjected to partial least squares (PLS) analysis to model the relationship between the metabolome and nitrogen treatment. Without IROA, regression coefficients of 0.819 and 0.849 for positive and negative modes, respectively were achieved. The IROA protocol improved on the values, yielding regression coefficients of 0.876 and 0.879 for positive and negative modes, respectively. In addition, IROA corrected for detector saturation for several high abundant peaks. Our experiment demonstrates that incorporating IROA into an LC/MS metabolomics experiment improves data quality and facilitates more precise modeling of a biological response.