Computer aided analysis of the regulatory network 
The fact that all of the 14 regulators in this study have biological functions during infection suggests they could be part of a coordinated network.
Acidic minimal medium is a well-established in vitro condition for SPI-2 expression; accordingly, we focused on examining regulation of genes expressed under these conditions [47],[48],[58].
To identify coordinated regulation we compared expression of each regulator in each mutant background when grown in minimal acidic medium (Figure 6A).
RNA samples were prepared from three separate cultures and used as a template in separate qRT-PCR experiments.
A matrix with 14 mutants in columns and 14 regulatory genes and SPI-2 genes (6 genes except for ssrB as used in Figure 4) in rows was constructed based on qRT-PCR data and z-scores were calculated based on average and standard deviations from columns and rows.
A network among regulators and SPI-2 was mapped as described in Lee et al. by sorting out values that changed in a specific mutant background [59].
We visualized the resulting relationships using Cytoscape [57].
Nodes indicate regulators or SPI-2 and red and blue arrows indicate activation and repression respectively (see Figure 6B).
In the computed network multiple regulators act both directly and indirectly to control SPI-2 expression.
However, direct or indirect regulation cannot be distinguished without additional experimental verification.
So, in Figure 6B all regulatory effects on SPI-2 have been removed except for those mediated directly by slyA and ssrB based on additional genetic data as described below.
The network suggests that both slyA and ssrA/ssrB could coordinate regulation of SPI-2 and other virulence factors by integrating signals from multiple regulators as was tested next.
An overall network was also generated by integrating 4 data sets; two CLR algorithm data from the complete microarray results and GSE2456 public microarray database and two matrix analysis data from the transcription profiles and qRT-PCR results (Figure S3; the limitation is that no distinction is made between positive and negative regulation in CLR algorithm data).
The consensus network combining all data reported in this study includes the network computed from qRT-PCR (Figure 6B) in part and suggests a predictive regulatory cascade that merits a test.
