Regulatory network of Escherichia coli: consistency between literature knowledge and microarray profiles

Abstract

The transcriptional network of Escherichia coli may well be the most complete experimentally characterized network of a single cell. A rule-based approach was built to assess the degree of consistency between whole-genome microarray experiments in different experimental conditions and the accumulated knowledge in the literature compiled in RegulonDB, a data base of transcriptional regulation and operon organization in E. coli. We observed a high and statistical significant level of consistency, ranging from 70%-87%. When effector metabolites of regulatory proteins are not considered in the prediction of the active or inactive state of the regulators, consistency falls by up to 40%. Similarly, consistency decreases when rules for multiple regulatory interactions are altered or when "on" and "off" entries were assigned randomly. We modified the initial state of regulators and evaluated the propagation of errors in the network that do not correlate linearly with the connectivity of regulators. We interpret this deviation mainly as a result of the existence of redundant regulatory interactions. Consistency evaluation opens a new space of dialogue between theory and experiment, as the consequences of different assumptions can be evaluated and compared.

Figure 1

Summary of consistency in each condition. The figure describes the consistency between knowledge extracted from particular experiments written in terms of our multiple rules and microarray data. We show four measures, the first one with the protein state depending on its effector prediction; the second, assuming no CRP rule (NCRP), the third, assuming that all regulatory genes on are active (NC) for no conformation, and the last one considering a combination between the second and the third assumption (NC NCRP). In all cases, off regulatory genes were assumed as inactive.

Figure 2

Connectivity across simple regulons and complex regulons. Each box represents a regulon. Simple regulons are alphabetically ordered at left. Complex regulons are ordered in increasing numbers of regulatory proteins. Connections depart only from simple regulons and arrive at complex regulons that share the same protein. In a few cases, the number of connections of a complex regulon is smaller than the number of its proteins. This is because not every regulator has a simple regulon.

Figure 3

Consistency vs. connectivity. The graph shows the way in which an incorrect predicted state on a large connected regulatory protein influences the level of consistency. We show the cases for four proteins in the four conditions tested. Regulators are ordered from the least to the most connected. Fis coregulating with 15 proteins, ArcA with 18, FNR related 20, and CRP coregulating with 47 proteins.