Cluster analysis identifies additional genes that are co-regulated with known virulence factors 
The results from the microarray analysis identified many genes that appeared to be coordinately regulated including all SPI-2 encoded genes that are part of the type III secretion system (Table S2).
We searched for additional previously unidentified virulence factors encoded elsewhere on the chromosome that show the same pattern of expression as those located within SPI-2.
A systematic way of identifying co-regulation is with one of several computer algorithms that group genes together if their expression is similarly changed under like conditions.
The diverse growth conditions used in this study, as well as differences between isogenic strains containing mutations in virulence regulators, make our data set ideal for this type of analysis.
In addition, we also performed cluster analysis using public data sets acquired from the NIH sponsored public repository of microarray data, gene expression omnibus (GEO).
We chose 120 transcriptional profiles from GEO that were derived by extracting RNA from the same parent Salmonella strain grown under a wide variety of culture conditions (GSE2456; G. Yun et al., unpublished data).
Expression profiles for all genes were uploaded to SEBINI (Software Environment for Biological Network Inference [18]).
SEBINI incorporates statistical and network algorithms into a framework for network inference [55],[56].
To detect dependencies among genes over different conditions we employed the context likelihood of relatedness algorithm (CLR; [19]), which is a plugin for SEBINI.
The results were visualized as a network of similarity relationships using Cytoscape [57].
A recent study with E. coli shows that the CLR algorithm was the most accurate in computing the correct network for experimentally verified regulatory interactions [19].
The results shown in Figure 5 use a force-directed network layout algorithm where genes (shown as small colored circles) are generally closer together when their statistical association, and thus degree of predicted co-regulation, is stronger (the cutoff for the genes shown is 5 standard deviations from the mean or greater; p<.0001).
All SPI-2 secretion apparatus genes as well as associated effectors and chaperones were found to very tightly cluster in the network.
In support of this conclusion, the cluster analyses based on our data and based on publicly available data sets are consistent.
The cluster analysis identified 92 genes that are co-regulated with the SPI-2 encoded type III secretion system but not located within SPI-2.
This includes known SPI-2 secreted effectors encoded elsewhere on the chromosome as well as many genes for which no function has been assigned (all total 123 genes; see Table 2).
Several of these genes are A+T rich (>60% compared to 48% for the Salmonella chromosome) and located within sequences that are not present in close relatives of Salmonella suggesting that they may be unidentified secreted effectors or other virulence determinants that have been acquired from other pathogens.
All in all, the implication of the cluster data is that the genes shown in Table 2 are co-regulated with SPI-2.
One interesting point is that the algorithm does not distinguish between positive and negative regulation only that it is coincident.
We analyzed expression of each of the 123 genes and found that only two genes (pepA and mopB) are negatively regulated; the remaining 121 are positively regulated.
