Search

The value of the multivariate data analysis tools principal component analysis (PCA) and principal component discriminant analysis (PCDA) for prioritizing leads generated by microarrays was evaluated. To this end, Pseudomonas putida S12 was grown in independent triplicate fermentations on four different carbon sources, i.e. fructose, glucose, gluconate and succinate. RNA isolated from these samples was analysed in duplicate on an anonymous clone-based array to avoid bias during data analysis. The relevant transcripts were identified by analysing the loadings of the principal components (PC) and discriminants (D) in PCA and PCDA, respectively. Even more specifically, the relevant transcripts for a specific phenotype could also be ranked from the loadings under an angle (biplot) obtained after PCDA analysis. The leads identified in this way were compared with those identified using the commonly applied fold-difference and hierarchical clustering approaches. The different data analysis methods gave different results. The methods used were complementary and together resulted in a comprehensive picture of the processes important for the different carbon sources studied. For the more subtle, regulatory processes in a cell, the PCDA approach seemed to be the most effective. Except for glucose and gluconate dehydrogenase, all genes involved in the degradation of glucose, gluconate and fructose were identified. Moreover, the transcriptomics approach resulted in potential new insights into the physiology of the degradation of these carbon sources. Indications of iron limitation were observed with cells grown on glucose, gluconate or succinate but not with fructose-grown cells. Moreover, several cytochrome- or quinone-associated genes seemed to be specifically up- or downregulated, indicating that the composition of the electron-transport chain in P. putida S12 might change significantly in fructose-grown cells compared to glucose-, gluconate- or succinate-grown cells.

A general method is presented for combining mass spectrometry-based metabolomics data. Such data are becoming more and more abundant, and proper tools for fusing these types of data sets are needed. Fusion of metabolomics data leads to a comprehensive view on the metabolome of an organism or biological system. The ideas presented draw upon established techniques in data analysis. Hence, they are also widely applicable to other types of X-omics data provided there is a proper pretreatment of the data. These issues are discussed using a real-life metabolomics data set from a microbial fermentation process.

Osteoarthritis (OA), one of the most common diseases among the elderly, is characterized by the progressive destruction of joint tissues. Its etiology is largely unclear and no effective disease-modifying treatment is currently available. Metabolic fingerprinting provides a novel tool for the identification of biomarkers. A metabolic fingerprint consists of a typical combination of metabolites in a biological fluid and is identified by a combination of [.sup.1]H NMR spectroscopy and multivariate data analysis (MVDA). The current feasibility study was aimed at identifying a metabolic fingerprint for OA and applying this in a nutritional intervention study. Urine samples were collected from osteoarthritic male Hartley guinea pigs (n = 44) at 10 and 12 mo of age, treated from 4 mo onward with variable vitamin C doses (2.5-3, 30 and 150 mg/d) and from healthy male Strain 13 guinea pigs (n = 8) at 12 mo of age, treated with 30 mg vitamin C/d. NMR measurements were performed on all urine samples. Subsequently, MVDA was carried out on the data obtained using NMR. An NMR fingerprint was identified that reflected the osteoarthritic changes in guinea pigs. The metabolites that comprised the fingerprint indicate that energy and purine metabolism are of major importance in OA. Metabolic fingerprinting also allowed detection of differences in OA-specific metabolites induced by different dietary vitamin C intakes. This study demonstrates the feasibility of metabolic fingerprinting to identify disease-specific profiles of urinary metabolites. NMR fingerprinting is a promising means of identifying new disease markers and of gaining fresh insights into the pathophysiology of disease. J. Nutr. 133: 1776-1780, 2003. KEY WORDS: * metabolic fingerprinting * [.sup.1]H NMR spectroscopy * multivariate data analysis * osteoarthritis * vitamin C

Genomics is based on the ability to determine the transcriptome, proteome, and metabolome of a cell. These technologies only have added value when they are integrated and based on robust and reproducible workflows. This chapter describes the experimental design, sampling, sample pretreatment, data evaluation, integration, and interpretation. The actual generation of the data is not covered in this chapter since it is highly depended on available equipment and infrastructure. The enormous amount of data generated by these technologies are integrated and interpreted inorder to generate leads for strain and process improvement. Biostatistics are becoming very important for the whole work flow therefore, some general recommendations how to set up experimental design and how to use biostatistics in enhancing the quality of the data and the selection of biological relevant leads for strain engineering and target identification are described.

Microbial production strains are currently improved using a combination of random and targeted approaches. In the case of a targeted approach, potential bottlenecks, feed-back inhibition, and side-routes are removed, and other processes of interest are targeted by overexpressing or knocking-out the gene(s) of interest. To date, the selection of these targets has been based at its best on expert knowledge, but to a large extent also on 'educated guesses' and 'gut feeling'. Therefore, time and thus money is wasted on targets that later prove to be irrelevant or only result in a very minor improvement. Moreover, in current approaches, biological processes that are not known to be involved in the formation of a specific product are overlooked and it is impossible to rank the relative importance of the different targets postulated. Metabolomics, a technology that involves the non-targeted, holistic analysis of the changes in the complete set of metabolites in the cell in response to environmental or cellular changes, in combination with multivariate data analysis (MVDA) tools like principal component discriminant analysis and partial least squares, allow the replacement of current empirical approaches by a scientific approach towards the selection and ranking of targets. In this review, we describe the technological challenges in setting up the novel metabolomics technology and the principle of MVDA algorithms in analyzing biomolecular data sets. In addition to strain improvement, the combined metabolomics and MVDA approach can also be applied to growth medium optimization, predicting the effect of quality differences of different batches of complex media on productivity, the identification of bioactives in complex mixtures, the characterization of mutant strains, the exploration of the production potential of strains, the assignment of functions to orphan genes, the identification of metabolite-dependent regulatory interactions, and many more microbiological issues.