Relative contribution of speciﬁc sources of systematic errors and analytical imprecision to metabolite analysis by HPLC–ECD ? Yevgeniya I. Shurubor a , Wayne R. Matson b , Rolf J. Martin a,c , and Bruce S. Kristal a,d, * a Dementia Research Service, Burke Medical Research Institute, 785 Mamaroneck Ave, White Plains, NY 10605, USA b ESA Inc., 22 Alpha Road, Chelmsford, VA Building 70, 100 Springs Road, Bedford, MA 01824, USA c Lane College, 545 Lane Ave, Jackson, TN 38301, USA d Department of Neuroscience, Cornell University Medical College, 1300 York Ave, NY 10021, USA Received 1 February 2005; accepted 17 March 2005 Objective interpretation of metabolomics data requires understanding both analytical and biological measurement errors. Here we address analytical measurement errors, the sources of these errors, and how this variability can impact metabolomic proﬁles. Sources considered include room temperature exposure (which could affect sample stability), spiking with authentic standards, the number of study replicates, the overall temporal design of the experimental series, and the complexity of the biological matrix of the samples (individual or pooled sera). The study focused on the analysis of 80 rat sera metabolites by HPLC coupled with coulometric array detectors. Time delay and room temperature exposure had minimal effects on the total relative metabolite concentrations and variability (mean: 94–98% of control, CV median : ±5–7%), but the concentrations of some speciﬁc metabolites were signiﬁcantly altered. Changes observed in the concentrations of speciﬁc metabolites ranged as high as ±7-fold, with changes in variability ranging from 0.3% to 68%. Spiked samples demonstrated more complex behavior when allowed to decay over time than did control samples. The spiking of sera and standards with 43 known metabolites increased variability of the apparent concentrations of metabolites up to 24% as opposed to 3% in pure sera. Increased variability was metabolite-speciﬁc. In both pure and spiked sera, 80–95% of metabolites had CVs equivalent to standard analytical CVs for these metabolites. Experimental design, number of replicates, and complexity of the biological matrix had comparable eﬀects. These results suggest that, under carefully controlled conditions, these analytical issues are not signiﬁcant sources of variability relative to biological variation for most metabolites. KEY WORDS: HPLC; dietary restriction; biomarkers; caloric restriction; analytical parameters. 1. Introduction The identiﬁcation of variables that convey informa- tion about samples or classes of interest, a process known as feature selection, is a central problem in modern-omics approaches (German et al., 2002; Olden, 2004). The nature of the problem diﬀers with the inherent complexity of the speciﬁc question being addressed. When multiple easily-assessed state markers (markers that by themselves identify class) exist, the choice of features can be driven by which is/are the fastest and/or cheapest to analyze (Kristal et al., 2005). If a single state marker is identiﬁed, that marker should clearly be used. If not, the use of complex proﬁles based on subtle concentration diﬀerences of multiple markers might be required; these proﬁles are associated with a series of diﬀerent concerns. One major concern is the interplay between analytical and biological variability. The use of metabolomics to understand questions of nutritive status provides a clear example of the com- plexity of interactions between analytical and biological variability. We are interested in using metabolomic proﬁles to validate dietary intake biochemically and to determine individual predisposition to nutrition-related diseases such as diabetes, coronary diseases, stroke, and cancer. Overall, organism-level homeostasis is similar between multiple individuals in a population, even as diets change within ranges that are associated with long term risk of diﬀerent diseases. Thus, the choice of spe- ciﬁc serum metabolites to comprise such metabolomic proﬁles is critical (Prentice et al., 2004). In general, assessed variability of metabolites in sera reﬂects a combination of biological and analytical variability. Biological variability depends on genetic and environ- mental factors, health status, individual behaviors, and eﬀects introduced by the speciﬁc choice of sampling schedules. Analytical variability results from issues in sample acquisition and storage, laboratory-related errors including methodological errors, instrument imprecision, inconsistent or impure reagents, and matrix eﬀects (Blanck et al., 2003), as well as errors related to ? This work was supported by NIA-AG15354. *To whom correspondence should be addressed. E-mail: bkristal@burke.org Metabolomics Vol. 1, No. 2, April 2005 (Ó 2005) 159 DOI: 10.1007/s11306-005-4431-8 1573-3882/05/0400–0159/0 Ó 2005 Springer ScienceþBusiness Media, Inc.