Gene Expression Profiling and the Use of Genome-Scale In Silico Models ofEscherichia colifor Analysis: Providing Context for Content

Abstract

2One of the most widely used high-throughput technologies is the oligonucleotide microarray. From the initial development of microarrays, high expectations were held for their use to aid in answering biological questions, due to their ability to measure mRNA abundances on a genome scale. However, accumulating experience is revealing that even when questions of sample preparation, data processing, and dealing with the inherently noisy data (81) are set aside, the large amount of data generated has proven difficult to analyze and interpret (12). It is also often challenging to narrow down specific novel findings based solely on expression profiling data.Here, we present a downloadable compendium of gene expression profiles for Escherichia coli and discuss the experience from one lab in which expression profiling data have been employed in a myriad of studies of E. coli. We will try to address two classes of expression profiling data usage: (i) how expression profiling can be analyzed using more traditional statistical methods to provide biological understanding and (ii) how genome-scale models form a context within which expression profiling data content increases in value. THE DATAWe present a database of 213 expression profiles produced in our laboratory and used in many of the case studies reviewed here. These profiles represent measurements for about 70 combinations of variations in experimental conditions and genetic variations in E. coli K-12 MG1655. In this set of experiments, the following parameters were varied: (i) the carbon source, (ii) the terminal electron acceptor, (iii) the temperature, (iv) the number of days the bacteria were grown at midlog phase, and (v) the genotype (wild-type and gene deletion strains were used). These data and the corresponding minimum information about a microarray experiment (9) may be freely downloaded at http://systemsbiology.ucsd.edu/In_Silico _Organisms/E_coli/E_coli_expression2. LEVELS OF ANALYSISMuch insight into the function of E. coli has been gained over the past decade through the statistical analysis of cDNA and oligonucleotide arrays. Methods such as the genome-scale analysis of differential gene expression patterns (21), clustering and classification (17), and gene set enrichment (76) have been successfully deployed. Overall, our experience shows that informative results can also be obtained if data are analyzed in the context of a genome-scale network reconstruction and with the use of an in silico model. Furthermore, greater success in data analysis has been achieved on a fine-grain level for more specific hypothesis generation and assessment than on a coarse-grain genome-scale level. The lessons learned from the analysis of this data set can be classified by the granularity of the analysis and the level of use of the in silico model.(i) Genome-scale analysis without using a modeling framework. E. coli grown to mid-log phase on various carbon sources and/or with various gene knockouts has been shown to usually evolve a growth phenotype as predicted computati...

Adaptive Evolution ofEscherichia coliK-12 MG1655 during Growth on a Nonnative Carbon Source,l-1,2-Propanediol

“…Although the POR catalyzes the oxidation of L-1,2-PDO to L-lactaldehyde, L-1,2-PDO cannot be utilized by wild-type (WT) E. coli as a sole carbon source under aerobic conditions because this compound cannot induce expression of the fuc regulon (11). Indeed, the fuc regu-lon was not expressed under any conditions when a database of 213 expression profiles produced in our laboratory was examined (38). Furthermore, even if the POR is expressed, it is oxidatively inactivated by a metal-catalyzed oxidation (MCO) mechanism (7).…”

Systems biology definition of the core proteome of metabolism and expression is consistent with high-throughput data

“…First, using a compendium containing negative control probes (22,23), we found that 86% of the core proteome was always expressed across 69 experiments, which was significant (hypergeometric test, P = 8.0 × 10…”

Identifying radiation exposure biomarkers from mouse blood transcriptome

“…One approach for gaining insight into the mechanisms underlying a transcriptomic response is to analyse the data in the context of a genome-scale metabolic network reconstruction (Duarte et al, 2007; Lewis et al, 2009). Genome-scale network reconstructions may be converted to mathematical models to visually or computationally analyse transcriptome data (Lewis et al, 2009).…”

“…Genome-scale network reconstructions may be converted to mathematical models to visually or computationally analyse transcriptome data (Lewis et al, 2009). The most common visualisation methods involve simply overlaying the expression data on the pathways within the model and visually determining what sections of the network are associated with negative or positive changes in gene expression.…”