Rationale

Current efforts in whole-genome sequencing have led to a rapidly increasing number of publicly available bacterial genome sequences [1, 2]. Novel technologies, such as genome-wide transcriptional profiling using DNA microarrays, enables the study of the transcriptional regulation of various processes in these sequenced microorganisms, which can, subsequently, lead to the identification of the regulatory networks involved [3–6]. Bioinformatics tools that enable one to predict and/or identify transcription regulatory elements and terminator sites are publicly available [7–14].

Graphical representations have proved very useful for the efficient interpretation of large amounts of biological data (for example, metabolic pathway and gene regulatory network visualization [15–17], transcriptome data analysis and/or clustering [18, 19]). Our group investigates metabolic pathways and gene regulatory networks of different Gram-positive bacteria. For easy and rapid interpretation of transcriptome data, we required software that enables us to project this onto a linear bacterial genome map, together with additional data (that is, terminator and regulator binding sites). Zimmer and co-workers have previously visualized transcriptome data (displayed as spots) in gene order [20]. However, their program does not allow the inclusion of data on transcription regulatory and terminator sites or other customized data. Visualization of such information would facilitate the interpretation of transcriptomes by displaying which genes are coexpressed in a transcriptional unit (an operon [21]), or are transcribed via readthrough from the neighboring gene (or genes), or lead to the formation of antisense RNA. The possibility of adding putative binding sites for transcriptional regulators onto the genome map would be a quick and convenient way to assess the biological relevance of such operator sites. Furthermore, visual analysis can be preferable over a statistical (mathematical) approach, as relevant data can easily be ignored if too high cutoff settings are applied. We screened several powerful commercial and public-domain software packages for transcriptome data visualization (GenVision (DNAStar, Madison, WI), GeneSpring (Silicon Genetics, Redwood City, CA), Kyoto Encyclopedia of Genes and Genomes (KEGG) [15], EcoCyc [16] and TM4 [19]), but none of these fulfilled our needs. We therefore developed the Microsoft Windows-based program Genome2D.

Genome2D

Genome2D was programmed in Borland Delphi 6 and compiled to a Microsoft Windows 9x/NT/2000/XP application. With its graphical user interface the program is easy to use for non-experts and is easily accessible because of its low system requirements; it can be installed on a standard local Windows personal computer, making it fast and safe (when confidentiality is required). The object-oriented programming environment of Delphi makes it easy to extend Genome2D. The CADSys 4 library version 4.2 was used for two-dimensional visualization of genomes. This library extends the Delphi vectorial graphics support to include 2D/3D CAD-like functions in applications.

The most prominent feature of Genome2D is a drawing module that generates comprehensive bacterial genome maps, in a single window screen, that can include specific genetic elements such as transcription terminators or regulator binding sites (Figure 1). The user can easily prepare figures for use in printed or digital format.

Figure 1
figure 1

Genome2D visualization of the genomic organization of L. lactis IL1403 (GenBank annotation: AE0051576). The figure displays a partial, detailed view in which putative terminators, determined using the TIGR software package TransTerm, are shown as stem-loop structures [11, 46]. Predicted promoter elements (-35 boxes in green; -10 boxes in blue) and cre-boxes (in red) are shown. See text for more details.

Full size image

Display of DNA microarray data in Genome2D is done by coloring the selected genes using a simple input file - that is, a tab-delimited text file with one column containing the names of the genes to be colored (corresponding to the gene names from the annotation file), and a second column with the color codes (black, white, red, yellow, fuchsia, green, lime, blue or aqua), or values, such as gene-expression ratios, on the basis of which color shades are assigned. A defined number of datasets from a complex transcriptome analysis experiment (for example, time-course measurements) can be loaded as separate input files, after which the data can be shown in animation, a feature that, to our knowledge, is not present in existing software. Clearly, the input files are not restricted to transcriptome data, and different kinds of datasets can be projected, such as from proteome analysis.

An umbrella for analysis tools

In addition to its visualization capabilities, Genome2D serves as a platform for different bioinformatics tools, such as data-extraction and conversion algorithms, which are summarized in Table 1. The combination of visualization and information extraction allows subsequent rounds of analyses, and thus an increase in data complexity, making Genome2D a powerful tool in the investigation of bacterial genomics data, especially from transcriptome and proteome analyses. Newly developed algorithms or tools can be easily implemented within the framework of the program.

Table 1 Features of Genome2D*
Full size table

Applications

Genome2D can be used for all annotated bacterial genome sequences. In our group, Genome2D is commonly used for the analysis of genomics data from Bacillus cereus, Bacillus subtilis, Lactococcus lactis, Lactobacillus plantarum and Streptococcus pneumoniae. We will illustrate the strength of Genome2D in visualization of transcriptome data hereafter, using the genomes of B. subtilis 168 [22] and L. lactis IL1403 [23] as examples.

The power of visualization

There are a number of benefits of visual inspection of transcriptome data compared with statistical analyses, which we will show here using published transcriptome data [24]. Most important, visualization can help in discerning true low-level gene activation. For instance, groES was classified as a ComK-regulated gene, as it met the stringent cutoff set in the analysis of Hamoen and co-workers [24]. However, groEL failed to meet these criteria. It has been shown that groES and groEL are part of a single operon in B. subtilis [25]. When the transcriptome data are visualized in Genome2D, one can see that groEL actually shows some level of activation, suggesting that groEL and groES are indeed activated as an operon (Figure 2a). Choosing cutoff values to define the set of regulated genes is a rather arbitrary process. Moreover, the statistical value of expression data in transcriptome studies is based on a limited number of data points, and it is therefore not surprising that several possibly relevant genes, such as groEL, will be missed. Another example is given in Figure 2b. yvrP, yvrN and yvrM were found to be ComK-activated, whereas yvrO did not meet the criteria [24]. Visualization in Genome2D reveals that yvrO is also slightly activated, and allows the conclusion that all four genes are likely to form a ComK-dependent operon (Figure 2b).

Figure 2
figure 2

Demonstration of the power of visualization in transcriptome analyses. The dataset used is from Hamoen and colleagues [24]. The strength of up- or downregulation is depicted by the intensity of the color. Stem-loop structures indicate annotated terminators. (a,b) Probable cases of low-level activation. Genes are colored on the basis of expression ratios from DNA macroarray experiments [24], without applying a stringent cutoff. Red shades indicate ComK-dependent activation, whereas green shows downregulation. Gray shades indicate ratios of around 1. Stem-loop structures are used to depict annotated terminators. K-boxes are shown by vertical red lines. (c,d) Putative cases of transcriptional readthrough. Red shades indicate significant ComK-dependent expression. K-boxes are depicted by vertical green lines. Gray genes are not significantly ComK-dependent. (e,f) Probable cases in which antisense RNA has a role (colors and symbols identical to (c) and (d)).

Full size image

Second, visualized transcriptome data can reveal putative transcriptional readthrough. For example, in the study of Hamoen and colleagues [24] mentioned above, thresholds of significance were partly based on the prior knowledge that limited readthrough from the comF operon occurs into the yvyF, flgM and yvyG genes [26]. This becomes apparent also in the Genome2D visualization of the data from Hamoen and colleagues [24]: the comF operon and downstream-located genes show differential levels of ComK-induction (Figure 2c). Extending this notion, one can predict that the reported ComK-dependent activation of spoIIB/maf/ysxA (radC) and yqzE is due to readthrough from comC and the comG operon, respectively (Figure 2d) [24].

Third, it has been reported that the use of double-stranded amplicons in DNA array studies might lead to the detection of antisense RNA, the biological significance of which is unclear [24]. Genome2D helps in the identification of putative antisense RNA detection by showing whether activated genes are located in reverse orientation downstream of activated genes. In the case of comE, it is known that the comER gene is not transcribed during competence [27]. However, in several array studies this gene appeared to be strongly activated by ComK [24, 28, 29]. From Figure 2e, it is apparent that this activation is due to the hybridization of antisense mRNA. Similarly, the observed expression of yhxD and several genes from the yck/nucA-nin/tlpC area may be instances of antisense RNA detection (Figure 2f). These observations cannot be made by normal statistical analyses without visual inspection.

Data extraction and analysis

To our knowledge no software is available in the public domain that allows information extraction and analysis in the way Genome2D does. To correlate expression with the activity of specific transcription factors more quantitatively, we incorporated several algorithms into Genome2D. The type of analyses that can be performed with these algorithms are exemplified below, using the analysis of the ComK-regulon in B. subtilis and the in silico prediction of the CcpA-regulon of L. lactis.

Hamoen and colleagues [24] used Genome2D to correlate the occurrence of ComK-binding sites (K-boxes) to ComK-dependent expression of genes, with the aim of testing whether the presence of a K-box upstream of a gene can be used to predict ComK-activation. They assigned genes to putative operons using a widely used algorithm [30] incorporated into Genome2D ('Add First Gene of Operon to Gene List'). Furthermore, they identified all K-boxes in the B. subtilis genome (Box searches are available through the Search menu) and located the closest upstream box for all genes and operons ('Add Nearest Box to Gene List'). Finally, the program is able to link predicted binding sites to the genes located closest to the box ('Add Nearest Gene to Boxlist'). Using these and additional algorithms, the authors showed that the predictive value of a K-box can be significantly improved by taking into account genome organization, additional ComK-binding motifs, and binding sites for RNA polymerase [24].

Prediction of the CcpA regulon in L. lactis

As is the case in other bacteria, many L. lactis ssp. lactis IL1403 genes are of unknown function [23, 31]. Prediction of gene regulation can implicate unknown proteins in certain cellular processes and, by directing genetics approaches, can help to assign functions. This is illustrated by the prediction of the CcpA-regulon (sugar catabolism control) in L. lactis IL1403 using Genome2D. We searched for and visualized putative CcpA-binding sites and promoter elements in the genome of L. lactis. Using the search-module in Genome2D (<Search>, 'Make Trained Set') and a list of 36 catabolite-responsive element (cre-box) sequences from several Gram-positive bacteria (see Additional data file 1), a weight matrix [32] was made that generated the consensus sequence ATGWAARCGTTTWCA (where W represents A or T, and R represents A or G) (see Additional data file 2). Subsequent screening (<Search>, 'Search Trained Set') for this consensus sequence, with an arbitrary cutoff of 8 (a perfect match would give a score of 10.8 with our weight matrix), identified 1,807 putative cre-boxes in the genome of L. lactis IL1403. Around 43% of these boxes are located in intergenic regions. As CcpA can act as a repressor or activator depending on the position of a cre-box relative to the RNA polymerase binding site [33], consensus -35 and -10 promoter element positions [34] of genes were predicted in the genome of L. lactis IL1403 using the MEME motif search routine [35] and Genome2D (A.L. Zomer, G. Buist, J.K. and O.P.K, unpublished data). The prediction was performed on intergenic regions from the L. lactis IL1403 genome, the primary location for promoter elements, which were extracted using Genome2D. Finally, the datasets from the cre-box and promoter element predictions were visualized onto the linear genome map of L. lactis IL1403 (see Figure 1 and Additional data file 3). Visual inspection confirmed the presence of operons previously described as regulated by CcpA [36–38]. Thirteen L. lactis genes (out of 116 putative CcpA-regulated genes) have counterparts in B. subtilis, on the basis of protein sequence comparisons using BLASTP (e-values were lower than 10-49) ([39] and see Additional data file 4). The 13 B. subtilis genes were among those that were recently shown to be CcpA-regulated in B. subtilis using DNA macroarray analyses [40], indicating that Genome2D can be used to generate relevant predictions on gene regulation. However, we would like to emphasize that the in silico prediction of gene regulation has to be corroborated by 'real' biological experiments, such as genome-wide transcriptome analysis [24, 41, 42].

Conclusions

Our analyses of transcriptome data in relation to the activity of specific transcription factors and their operator sites required a more flexible genome visualization program than is currently publicly available. We therefore developed Genome2D, a software tool that enables visualization of transcriptome data onto a linear map of an annotated bacterial genome and at the same time highlights additional features, such as putative regulatory sequences and terminators. The combination of information extraction and visualization facilitates rapid, easy and intuitive analysis of genomics data, and in our research group Genome2D proved to be of great assistance in the study of transcriptome data. New algorithms can be rapidly implemented in the Genome2D program menu structure. Regular updates of Genome2D will be available via the Internet [43]. Because of the exponential increase of publicly available bacterial genome sequences and large-scale experiments, tools like Genome2D will become indispensable for the interpretation of complex datasets, such as those from transcriptome and proteome studies.

Additional data files

The following additional data are available with the online version of this article: a list of cre-box sequences found in Gram-positive bacteria (Additional data file 1); a screen dump from Genome2D showing the cre-box weight matrix (Additional data file 2); a Genome2D input (tab-delimited text) file with the coordinates of the identified cre-boxes and promoter elements in the genome of L. lactis IL1403 (color file) (Additional data file 3); a table of cre-boxes identified in promoters of genes in the L. lactis IL1403 genome (Additional data file 4). All additional data files can also be obtained from [44].