Bioinformatics Resources for the Study of Gene Regulation in Bacteria

Genomics, which has been identified as the science of the century, is dramatically changing the historically weak relationship between experimental and theoretical biology. The addition to the Journal of Bacteriology of a section for computational biology marks a turning point in the history of this dialogue. This minireview is focused on the computational biology of gene regulation in bacteria, defined as the extensive use of bioinformatics tools to increase our understanding of the regulation of gene expression.

The study of gene regulation has been radically affected by the elucidation of full-genome DNA sequences and the subsequent development of high-throughput methodologies for deciphering their expression. Before the genomics era, most research was focused on individual biological systems. A large number of our colleagues, contributors to this journal, have devoted much of their academic careers to the understanding of individual regulatory units describing operons, regulators, and promoters and their roles in the physiology of the cell. These contributions have provided fundamental information to support the most recent efforts for the integrative knowledge of the cell that all the genomics sciences are achieving.

Genomics offers for the microbiologist studying gene regulation the opportunity to understand individual systems in the context of the whole cell. These integrative sciences have also changed the landscape of data available for new discoveries in the evolution of gene regulation. The major challenge of the genomics era is dealing with large amounts of data at all molecular levels and being able to generate integrated biological knowledge from the data. Bioinformatics is essential to progress in this direction, as it provides what is necessary to deal with large amounts of data: databases, algorithms to generate genomic answers to standard questions, overviews, and navigation capabilities, as well as statistical methods to perform and validate analyses.

Current knowledge of gene regulation in prokaryotes is quite diverse, from the constantly increasing number of full genome sequences for which very little experimental work has been performed, including the many genomes that cannot yet be grown in the laboratory or the little-studied archaea, to the few highly characterized bacterial genomes, such as those of Bacillus subtilis and Pseudomonas aeruginosa, with Escherichia coli K-12 being by far the best-known bacterium and free-living organism. Figure 1 shows this strongly unequal distribution of knowledge, with the number of publications on gene regulation, as an example, for the different microbial genomes.

FIG. 1.

Number of published particles per organism. We searched through PubMed by using a collection of keywords that we regularly use to gather information for RegulonDB across different bacteria. The profile requires the name of the organism to be in the title.

The first exhaustive historical set of regulated promoters and their associated transcription factors (TFs) and TF DNA-binding sites (TFBSs) gathered around 120 σ⁷⁰ and σ⁵⁴ promoters of E. coli K-12 (16, 29). This information and the experience obtained were the seeds for what is now RegulonDB (http://regulondb.ccg.unam.mx/), the original source of expert curated knowledge on the regulation of transcription initiation and operon organization in E. coli K-12. It contains what is currently the major electronically encoded regulatory network for any free-living organism (25). This information is also contained in EcoCyc, the E. coli model organism database, with added curated knowledge on metabolism and transport (http://ecocyc.org/). We estimate that currently around 25% of the interactions of the full cellular regulatory network of transcription initiation have been assembled. RegulonDB should not be conceived of as a database, but as an environment for genomic regulatory investigations, linked to bioinformatics tools that facilitate analyses of upstream regions, together with data sets and tools for microarray analyses and, more recently, direct access to full papers supporting its knowledge. We are not only curating up-to-date original papers, but have also initiated “active annotation,” to use Jean-Michelle Claverie's terminology, to more precisely experimentally map promoters by using a high-throughput strategy.

This minireview has been organized taking into account the fact that most of the readers of the Journal of Bacteriology are experimentalists. We start with a few examples of lessons on gene regulation from bioinformatics, focusing on promoters, their definition and regulation, and operon structure. The second section summarizes how RegulonDB has been useful to experimentalists, as well as its role as the “gold standard” for implementing bioinformatics predictive methods, topological analyses of the network, and models of the cell. The last section offers a compendium of links to bioinformatics resources on gene regulation in bacteria, illustrating their usage with flowcharts associated with questions on the regulation of gene expression that are entailed in the use of RegulonDB and associated bioinformatics resources. We illustrate the very efficient text analysis obtained using Textpresso to access more than 2,400 full papers that support the E. coli regulatory network. A cautionary note: the examples of the three sections are biased toward cases in E. coli and its RegulonDB regulatory database. This bias is natural, since first of all, our direct experience is with RegulonDB, and second, especially for bacterial-gene regulation (Fig. 1), we may well quote Fred Neidhardt (87): “Not everyone is mindful of it, but all cell biologists have two organisms of interest: the one they are studying and Escherichia coli!”

LESSONS FROM OVERVIEWS IN GENE REGULATION: PROMOTER DEFINITION, PROXIMAL-SITE REQUIREMENT FOR RNAP-σ⁷⁰, AND OPERON STRUCTURE

Historically, knowledge of gene regulation started with a model of the lac operon, its cis-regulatory elements, and the notion of allosterism in E. coli (43, 65). Quickly, this model, based on repression, had to be expanded to accommodate the more elaborated positive mechanisms of gene regulation. Since then, we have witnessed a gradual expansion in the diversity of knowledge about the molecular anatomy of gene regulation and the rich mechanisms that together compose the decision machinery of the cell.

The discovery of a conserved motif in E. coli promoters, the −10 box, also called the Pribnow box, is a striking example of how a pattern that was visually discovered in as little as seven DNA sequences has remained valid (75). This was an early contribution of bioinformatics to the study of gene regulation. Certainly, the identification of a promoter as the physical region for the binding of the RNA polymerase (RNAP) results from the combination of transcription initiation experimental mapping and pattern recognition, initially by visual inspection and now performed by multiple-alignment methods (37, 39). Promoters in this sense come from a combination of experimental and bioinformatics evidence.

We should keep in mind, though, that gathering a large collection of data in biology does not guarantee that we can make sense of it or that new knowledge will emerge. We illustrate this by showing what we have learned from an exhaustive genomic collection of components of the regulatory network, that is, regulated promoters and the relative distances to their corresponding activator and repressor binding sites, and with a second example demonstrating how a very simple analysis of operon structure enabled the development of a method to predict operons in E. coli and then in any bacterial genome.

As mentioned above, in 1991, we gathered knowledge about around 120 promoters in E. coli K-12 (16, 29), and we have continued since then, increasing the data set (see http://regulondb.ccg.unam.mx/html/Database_summary.jsp for the history of the accumulated curated knowledge). One of the clearest lessons from that collection effort was the conclusion that regulation of transcription initiation in the case of the σ⁷⁰ holoenzymes (Eσ⁷⁰) always requires a proximal DNA site (a proximal site is defined in terms of its position relative to the transcription initiation site, so that a direct contact of the TF with RNAP is assumed) (53, 74). Seventeen years later, version 6.2 of RegulonDB (July 2008) has 1,754 promoters. Of these, 697 are σ⁷⁰ promoters, 421 of which have at least one characterized binding site for a TF. This collection of promoters has 1,382 associated binding sites with coordinate positions. The definition of proximal sites in that initial review was from −65 to +20 (16); however, single activation and coactivation were later reported from −90 with cyclic AMP receptor protein (CRP) (12, 13, 116). We also learned, after 1991, that the flexibility of the α C-terminal domain of RNAP expands the proximal region to around −100, supporting direct contact with CRP (5, 57). Thus, in principle, the range of positions enabling direct contact with RNAP can be set from −95 to +20. Analyses with current data show that only 26 promoters, accounting for less than 5.9% of all regulated promoters, currently lack sites within this proximal range. In principle, there should be no promoter subject to regulation from only remote positions, other than σ⁵⁴ promoters (67), and we will not discuss them in detail here. The distribution of proximal sites in this range is shown in Fig. 2. We can observe the same general tendencies discussed in the 1991 review, with repressors distributed across all of the proximal region, with the downstream −30 to +20 interval being dominant. Repressors prevent RNAP from interacting with the proximal region, between −30 and +20, or from preventing interactions of activators at −40, −50, and −60 central positions. The purB gene, cotranscribed with hflD, is repressed by PurR with only one operator located at +892.5. It could interfere with transcription initiation or act as a roadblock and obstruct the progress of the transcribing polymerase (30, 33).

FIG. 2.

Distribution of TFBSs. RegulonDB version 6.2 has 697 σ⁷⁰ promoters, 421 of which have at least one characterized binding site for a TF. The figure displays the distribution of central positions of activator and repressor DNA-binding sites in the −95 to +20 interval. The percentage of promoters was divided by the number of activator or repressor DNA-binding sites with the center position within each interval of 10 bp. This figure can be compared to Fig. 2 in reference 16.

As shown in Fig. 2, all activators, not only CRP, tend to prefer interacting near positions −40 and −70. The strongly diminished occupancy around −50 has been shown to be due to the binding of the α C-terminal domain of RNAP (4, 83). On the other hand, some activators overlap the promoter and can bind downstream of the +1 site. TFs of the MerR family activate transcription binding at a palindrome located between the −35 and −10 elements of the promoter (71, 93). Other regulators activate transcription through auxiliary binding sites located between −10 and +20 (38). A striking, well-documented case is activation from +1848 by IHF, a TF known to bend the DNA to favor transcription (1).

Understanding gene regulation requires a detailed knowledge of operon structure. Thanks to the efforts of Mary Berlyn, around 1998 we populated RegulonDB with an initial set of putative operons that we have since been expanding using both experimental information and predictions based on the rules learned from the original sets. Figure 3 shows the distribution of intergenic distances within operons (572 in 2002 versus 1,839 in 2008) and its contrasting set of distances of genes at boundaries of operons in the same direction of transcription (346 in 2002 versus 1,311 in 2008). This striking, clear-cut distribution of very short distances within operons as opposed to intergenic regions upstream of operons was the basis for the prediction of operons even for genes without an assigned function in the E. coli genome, and subsequently in many bacterial genomes (66, 84). As mentioned below, a high-quality curated corpus of knowledge such as this one has supported the development of bioinformatics methods capable of predicting many aspects of the regulatory network.

FIG. 3.

Intergenic distances of genes within and at transcription unit boundaries. The sharply different distributions of these distances enabled the use of a direct method to predict transcription units in the complete E. coli genome. This figure is very similar to Fig. 3 in reference 84.

We recall how the definitions of promoter elements are currently the basis of more elaborate bioinformatics methods for pattern recognition. We briefly discuss two examples in which gathering large amounts of data for individual cases has been fruitful in studying the biology of gene regulation: the examples of (i) promoters and their regulation and (ii) operon structure. It is true, though, that these are large collections of “the same types of stamps.” The major challenge in genomics, as we have said, is nonetheless a different one, that of integrating a particular system or gene and its product with expression patterns of similarly regulated genes as the cellular environment changes, for instance.

HOW AN ELECTRONIC CORPUS ON GENE REGULATION HAS BEEN USEFUL FOR EXPERIMENTALISTS

Here, we illustrate how the experimental and bioinformatics scientists studying gene regulation, not only in E. coli but also in many other organisms, have made extensive use of RegulonDB to gain insights into several aspects of gene regulation. We note that RegulonDB contains detailed, accurate, and up-to-date information about operon organization, regulatory DNA sites for TFs, promoters, terminators, and RNA regulatory elements that have been both experimentally determined and predicted. Together, these elements constitute the known transcriptional regulatory network of E. coli. The primary source of information for this section was obtained from questions addressed to the database by users and a search of the literature for articles citing or making use of RegulonDB. We are certain that these questions are valid for other databases and for any other bacterial species. Table 1 provides a list of selected articles published since the creation of RegulonDB, emphasizing the purposes for which they have used RegulonDB. People studying gene expression or modulon architectures by using microarrays or proteomics data, for many wild-type and mutant strains grown under different conditions, have taken into account the operon structure in E. coli to make sense of experimental data. For instance, in work by Yooseph et al. (114), the 7.7 million Global Ocean Sampling sequences were analyzed using the collection of transcription units in RegulonDB for their statistical analysis and to identify same-operon gene pairs. RegulonDB has also been used for the identification of regulatory binding sites and determination of how this information correlates with gene expression in wild-type and TF mutant strains. Promoter and TF binding site mapping by using genomic strategies (chromatin immunoprecipitation [ChIP]-chip) have also relied on this database as the source of primary information, or even proteomics data. In the genome-wide location of RNAP promoter sites using ChIP-chip, the 961 identified promoters in RegulonDB were used to set the 26% negative-detection rate (35).

TABLE 1.

Uses of RegulonDB

Description	Reference
Experimental
Gene expression analysis (microarray)
Operons improve estimation for cDNA microarrays	113
Large-scale validation of regulation from expression profiles	22
Cross talk between the plasmid and the chromosome	32
Microarray of a Shewanella oneidensis etrA mutant	3
Clustering gene expression data with error information	100
Affymetrix microarray coexpression of genes in operons	31
Quantitative description of large-scale microarray data	86
Analysis of the NsrR regulon	23
The Rcs phosphorelay and intrinsic antibiotic resistance	52
Growth defects and cross-regulation of gene expression	92
σS-dependent genes and their promoters	50
DNA adenine methyltransferase and gene regulation	89
The CRP regulon: in vitro and in vivo transcriptional profiling	115
Genome-wide expression analysis of FNR regulation	46
Microarrays of genes in quorum sensing	19
Transcriptome analysis of E. coli	101
An extended regulon of the methionine repressor	62
Transcriptome polymorphism in E. coli/Shigella species	54,88
Global RNA half-life analysis and patterns of transcript degradation
Early osmostress gene expression using microarrays	111
Transcriptome determination of transcription regulators	47
Constraint-based in silico models of E. coli	80
Analysis of ChIP-chip data
RNA polymerase binding sites by ChIP-chip	35
Analysis of protein abundance
Protein abundance profiling of the cytosol	42
Analysis regulatory mechanisms
Outer membrane vesicle and membrane instability	64
DNA sequence annotation
The symbiotic plasmid of Rhizobium etli CFN42	27
Analysis of specific biological systems
Mutant release factor 1 and 16S rRNA maturation	45
The PTS system of Vibrio fischeri	108
Analysis of gene expression dynamics
SoxRS-dependent transcriptional networks	7
Metagenomics
The Global Ocean Sampling: expanding protein families	114
Bionformatics
Regulatory-network prediction
Prediction of new members of regulons	96
Predicting transcriptional regulatory interactions	107
Promoter prediction
Recognition and prediction of σ70 promoters	56
Improved prediction of transcription start sites	28
Improving promoter prediction in E. coli	11
Transcription factor prediction	73
Predicted transcriptional regulators in E. coli
DNA-binding site prediction
Genomic prediction of transcriptional regulatory sites	98
Binding sites in bacterial genomes	55
TF binding sites in E. coli and Streptomyces coelicolor	51
Operon prediction
Operon prediction in Pyrococcus furiosus	103
Operon prediction without expt	6
A Bayesian network approach to operon prediction	8
Analysis of promoters
Computational promoter analyses in 32 genomes	91
Orthologous regulatory identification
Orthologous TFs have different functions	77
Transcription in archaea	49
Architecture of operons, promoters, or DNA-binding sites
Gene expression with combinatorial promoters	17
Positional distribution of DNA motifs in promoter regions	14
Transcriptional units and operons in Bacillus and E. coli	70
Evolutionary analysis of elements involved in transcription
Evolution of noncoding DNA in prokaryotic genomes	82
Evolution of DNA regulatory regions for proteogamma bacteria	78
Co-volution of TFs depends on mode of regulation	36
Prediction of gene/protein biological function
Gene Function Predictor based on context analysis	21
Protein function and linkages based on genome organization	95
Inference of functional relationships from predicted operons	44
Inferring the role of transcription factors
Inferring the role of TFs in regulatory networks	106
Source for other databases
Multidimensional annotation of the E. coli K-12 genome	48
Regulatory interactions in gammaproteobacterial genomes	72
Modeling and topology of the network
Modeling the overall dynamics of the network
Genetic and metabolic regulatory networks of Bacillus	26
Reconstruction of microbial transcriptional regulatory networks	34
Metabolic network and connectivity analyses
Hierarchy and modularity in metabolic networks	79
Low-degree metabolites in biological networks	85
Regulatory network analysis
Logical types of network control in gene expression	63
Dynamics in regulatory networks from four kingdoms	2
Gene expression in negatively autoregulated circuits	61
Novel topological concepts
Network motifs in the regulation network of E. coli	90

The corpus of knowledge on gene regulation has been essential for diverse implementations relying on bioinformatics. As depicted in Table 1, this corpus provides the means to generate and test predictions for a large collection of regulatory elements, such as promoters, TFs, and TFBSs; complete genomic repertoires of TFs and operons; and even unidentified network interactions. It has also served the purpose of modeling the overall dynamics of the network and for proposing novel biological concepts, such as the network motifs reported by the group of Uri Alon (90) or the notion of hierarchical and modular networks described by Ravasz and colleagues (79).

Annotated genomes and ingenious ways to transfer knowledge, with all the associated risks of assumptions of orthologous relationships, enable us nonetheless to estimate that the wiring of the regulatory network is different across organisms, especially among bacteria (59, 60). We know better now that the evolutionary origin of regulatory interactions in bacteria depends on gene duplication and specialization, operon reorganization, binding-site duplications, and horizontal gene transfer (76, 97).

A COMPENDIUM OF BIOINFORMATICS RESOURCES FOR STUDYING BACTERIAL GENE REGULATION AND INTRODUCTORY PROTOCOLS

Designing a representation of the rich and variable knowledge of biological systems in order to encode it into a formal database management system, together with the corresponding data-gathering and curation processes, is one of the major infrastructure-building efforts that characterize computational biology. This is apparent if one examines the year's first volume of Nucleic Acids Research, which is devoted to databases. We did not find an integrated collection of databases and bioinformatics tools devoted to gene regulation; therefore, we have gathered here an exhaustive selection of resources specifically dealing with gene regulation in bacteria. Based on the compendium gathered by Galperin in 2008 (24) plus Ecoli Hub (http://www.ecolicommunity.org/) and BIOPAX pathway databases (94), we identified approximately 100 different resources (from 240) that were directly related to prokaryotic-gene regulation. Our compendium groups sites, among others, for TFs and gene regulation (e.g., devoted specifically to the AraC/XylR families [102] or to TFs [110]); for RNAs; and for biological pathways and regulatory networks, microarray databases, and some other related themes, such as signal transduction pathways (http://genomics.ornl.gov/mist/), protein-protein interactions, genome databases, the published literature, and metadatabases. Users should be aware that we did not specify for each resource the date of its last update, which is quite variable. This compendium should be a useful resource for those interested in searching or analyzing regulatory features across bacterial genomes. The name of the site and its URL address, together with a short description and a list of tools available, can now be accessed at http://regulondb.ccg.unam.mx/Additional_resources.jsp.

Several resources have easily implemented documentation, tutorials, and demonstrations to help the user. Even though bioinformatitians devoted to database construction and maintenance invest important efforts in the design and implementation of user-friendly interfaces, it is not uncommon for first-time users to have trouble finding the best way to address their questions of interest. Let us take the following question as an example: “If I have a gene, how can I find out all that is known about its regulation and operon organization?” This simple question generates a rather complex and rich answer, including the sequences, coordinates, and regulatory effects of every single TFBS and all promoters of the gene, as well as its operon organization. Figure 4 shows a flowchart that explains how this information can be obtained in a few navigation steps, which in this case all occur within RegulonDB. Many other resources and databases display this same information differently. For instance, the PRODORIC genome browser has the ability to zoom in to the level of the DNA sequence, showing the binding sites and promoters (69). RegulonDB is linked to Gene Expression Tools (GetTools) (40), as well as to the Regulatory Sequences Analysis Tools (RSAT) website (99), and contains a suite of tools built to predict and analyze regulatory regions for 663 available microbial genomes, in addition to 62 eukaryotic ones. These tools are designed to answer questions related to groups of genes suspected to be coregulated, a very common subject whenever a research group has done a microarray experiment, or a group of genes from a ChIP-chip (or ChIP-sequence) experiment. Figure 5 shows a flowchart for a given set of genes from a ChIP-chip experiment with LexA (109). RSAT generates a collection of upstream sequences given the gene set as input, while RegulonDB contains a position-specific matrix that was derived from the collection of experimentally characterized binding sites (TFBSs). These matrices can be used to scan sequences in order to predict putative target TFBSs in the whole set of upstream regions of the genome and can then be displayed in a graph. The RSAT team has just published several protocols for a variety of similar questions. The main goal of these introductory protocols is to illustrate and motivate the use of bioinformatics resources for the study of gene regulation by illustrative questions, showing in comprehensible flowcharts an easy way to answer them. In work reported by Defrance et al. (18), flowcharts and protocols describe in detail how to discover the TFBSs common to a regulon obtained from RegulonDB, or any other bacterial database, in fact. The whole interaction network of E. coli reported in RegulonDB can also be analyzed (10). Given a set of names of commonly expressed genes from, for instance, a microarray or ChIP-chip experiment with any bacterial genome, using RSAT one can obtain the upstream regions and search for common motifs or TF binding sites and cis-regulatory modules (104).

FIG. 4.

Flowchart for gathering all regulation information for a single gene. Navigation options are shown, starting from the main page of RegulonDB with the name of a gene, melA or melR in this example. The MelR-CRP complex regulon is shown.

FIG. 5.

Flowchart for ChIP-chip data and genes with similar DNA-binding site motifs. The example uses as input a set of genes from a ChIP-chip experiment with LexA (109). RSAT (99, 105) was used to obtain the collection of upstream sequences, given the ChIP-chip gene set. The position-specific matrix (PSSM) for LexA was obtained by selecting Downloads → Data sets → Matrix alignment from the RegulonDB main menu. Then, it was pasted into RSAT to run a matrix scan. This program will search, given a threshold, for predicted sites in the complete set of upstream regions of the genome, and the results can be automatically obtained in a graphic display by using the feature map program.

EcoCyc contains a large collection of graphic and text displays, including a genome browser that shows genes by gene ontology class, operons, and all elements of gene regulation in a region of the genome, and it provides a network display of regulatory interactions, in addition to the Omics viewers that enable the user to display pathways and regulatory networks based on an input file, for example (48). Graphic displays of a set of genes can be obtained for several genomes with the PRODONET tool (http://www.prodonet.tu-bs.de/). For many bacteria, given a set of genes, their functional classes can be obtained (http://www.jprogo.de/). Several graphics tools are available to show the genomic context of a gene within its genome, as well as the contexts of its orthologs within several related genomes (see, for example, GeConT [15] or http://img.jgi.doe.gov/cgi-bin/pub/main.cgi). The number of plausible elaborated questions is certainly quite large. We invite the interested reader to use the compendium of resources and to see additional flowcharts by visiting RegulonDB at http://regulondb.ccg.unam.mx/Flow_charts.jsp. We believe these will serve as examples that the user can modify and use to find, intuitively, equivalent usages in other bacterial databases.

ACCESS TO SPECIFIC GENE REGULATION LITERATURE AND FULL PAPERS REQUIRES A SEPARATE MENTION

Textpresso, a powerful text-mining engine for studying the scientific literature (68), was implemented for RegulonDB (http://regulondb.ccg.unam.mx/Textpresso/), and 2,472 full-text papers, 3,125 abstracts, and more than 4,200 curator notes can be directly searched (81). This valuable tool allows the experimental researcher to search through categories, keywords, and ontology classes with the specific gene, promoter, operon, or TF of interest through the knowledge space of full papers that support the electronically encoded transcriptional regulatory network of E. coli.

CONCLUSIONS

As we have mentioned, genomics has changed the focus from individual systems to an understanding of the whole cell. The study of gene regulation, as well as almost any other aspect of modern molecular and cellular biology, requires bioinformatics tools and methods to manipulate and analyze the large amounts of available information to eventually generate a more integrated perspective and knowledge of the cell. The global analysis of TFBSs and their positions relative to transcription initiation illustrates a vivid combination of details of individual systems and the search for a unified understanding based on the need for bound TFs to interact with RNAP. This is one example, among many others, of what genomic perspectives offer as opportunities and challenges. To paraphrase Whitehead (112), integration of gene regulation involves both the cautious gathering of details and a passion for understanding.

The bioinformatics infrastructure for microbial-gene regulation involves an important effort to maintain and update the constantly increasing body of knowledge in this field resulting from the accumulation of experiments performed in many laboratories through the years. In fact, this paper celebrates the 10 years since the first publication of RegulonDB (41). Of course, no matter how much human effort databases involve, they remain tips of icebergs compared to what is found in each paper. New methodologies are helping to manage in more intelligent ways the large amounts of information, such as computational access to query a full-text specific corpus of literature or the scientific community's participation in EcoliWiki within the EcoliHub (http://www.ecolicommunity.org/), and many others.

This review will fulfill its purpose if it facilitates the appreciation by experimentalists of the usefulness of databases and programs devoted to the study of gene regulation. The emphasis of this review has been on the use of bioinformatics resources, not on their implementation or the challenges ahead, which are numerous. For instance, what is the best way to curate experiments as new high-throughput technologies emerge? There is a lot of work to be done in order to expand the prominent databases beyond the level of transcription initiation and to integrate the evolving knowledge about other levels of gene regulation, for instance, those of single molecules and single-cell experiments (20). Furthermore, feedback loops of the type supporting multistationarity and stochasticity (20, 58) or those responsible for the return of systems to their initial state, have not yet been systematically described in databases (however, see Goelzer et al. [26] for further information). A major challenge, given the increasing number of sequenced genomes, will be to generate predictive tools that can expand our knowledge of a few bacteria to a similar extent for the many more organisms for which there is currently much less experimental support (9).

The study of the regulation of gene expression of bacterial systems will no doubt make important contributions in this century of genomics, both to the understanding at the molecular level of the capabilities of the basic unit of life, a single cell, and in the many potential technological applications. Multidisciplinary teams and future, younger generations are welcome to this enterprise.