High-throughput metabolic state analysis: the missing link in integrated functional genomics of yeasts

Abstract

The lack of comparable metabolic state assays severely limits understanding the metabolic changes caused by genetic or environmental perturbations. The present study reports the application of a novel derivatization method for metabolome analysis of yeast, coupled to data-mining software that achieve comparable throughput, effort and cost compared with DNA arrays. Our sample workup method enables simultaneous metabolite measurements throughout central carbon metabolism and amino acid biosynthesis, using a standard GC-MS platform that was optimized for this purpose. As an implementation proof-of-concept, we assayed metabolite levels in two yeast strains and two different environmental conditions in the context of metabolic pathway reconstruction. We demonstrate that these differential metabolite level data distinguish among sample types, such as typical metabolic fingerprinting or footprinting. More importantly, we demonstrate that this differential metabolite level data provides insight into specific metabolic pathways and lays the groundwork for integrated transcription-metabolism studies of yeasts.

Scheme 1. Metabolite library coverage

We generated a MCF derivatization metabolite library, covering much of central carbon metabolism and nearly all of amino acid biosynthesis pathways. We show the metabolic network of amino acid biosynthesis in S. cerevisiae during aerobic growth, highlighting (in grey) metabolites in our library. Continuous arrows indicate a one-step reaction, and broken arrows indicate a series of biochemical reactions where the numbers indicate the reaction steps in the pathway. This illustrative figure does not include all metabolites present in the library.

Figure 1. Observed metabolites

We detected and identified a slightly different set of metabolites for each data class. Metabolites detected and identified in one class and not another resulted in infinite differential ratios and corresponding statistical confidences. The Förster et al. [22] genome-wide metabolic reconstruction abbreviations are found in Table 1. aer, aerobic; ana, anaerobic; mt, mutant; wt, wild-type.

Figure 2. Sample visualization

GC-MS metabolite data from MCF derivatization successfully distinguishes among strains and cultivation conditions. Projecting intracellular metabolite data from approx. 60 samples into a 3D space reveals distinct clustering of the four data classes. For each sample, we calculated projection values as a linear combination of metabolite values determined by FDA.

Figure 3. Error structure among shake flasks

A histogram of metabolite residuals, x−x, reveals the error structure between samples and shake flasks for wild-type aerobic intracellular metabolites. The residuals do not display a large flask bias relative to the overall sample variability. Residuals from the other data classes behaved similarly. Thus we adopt the equal-means hypothesis among shake flasks and samples from each shake flask were treated as independent repetitions.

Scheme 2. Aerobic mutant growth

We detected the black-highlighted metabolites at higher levels in the mutant (GDH1 deleted, GDH2 over-expressed). In contrast, the grey metabolites were detected at lower levels and significant P values. Continuous arrows indicate one-step reaction, and broken arrows indicate multiple biochemical reactions. The numbers indicate the reaction steps in the pathway.

Figure 4. Metabolic array comparing cell populations

Based upon the metabolite level ratios and statistical significances (P values), we visualized differences among the cell populations. Uppercase metabolite abbreviations indicate intracellular detection, whereas lowercase metabolite abbreviations indicate extracellular detection. The Förster et al. [22] genome-wide metabolic reconstruction abbreviations are found in Table 1.