In this study, we have investigated the global regulatory network required for Salmonella virulence under specific in vitro (infection-mimicking) conditions by employing regulatory protein perturbations and high throughput sample-matched omics measurements coupled with computational network analysis. This systems-level analysis of the Salmonella virulence program inferred 168 proteins which were clustered close to SPI-2 virulence proteins in the regulatory network and thus were likely to be involved in pathogenicity (Additional file 4, Table S2). A set of these predicted virulence candidate proteins were tested to verify the network prediction, including SrfN, PagD, PagC, STM1548, PdgL, STM1633, PagK2 and STM3595. These eight proteins were expressed more highly within macrophages than under LB log phase condition (Figure 2 and Additional file 7, Figure S3), which is characteristic of the SPI-2-encoded virulence proteins and suggests that their functions may be required for intracellular growth. Interestingly, SrfN, PagC, PagD, PagK2 (and its close homologues PagK1 and PagJ) were translocated to the macrophage cytoplasm (Figure 3), suggesting that these six proteins are secreted virulence effectors interacting with host cellular components to promote bacterial proliferation. In fact, SrfN and PagK homologues were required for Salmonella systemic infection in mice (Figure 4).
We showed that the integration of transcriptomics and proteomics data at the network level could provide enhanced predictions of components important for virulence. This is a novel way of integrating these two disparate data types, and should be applicable to other systems. Three of the proteins chosen for investigation had very high betweenness measures in the network (SrfN, PdgL, and STM2585A/PagK2; see Additional file 4, Table S2). It is interesting that the two translocated proteins from this later group, SrfN and PagK2, were also shown to have an effect on virulence in mice. Though a limited validation, this supports our hypothesis that the topology of the network can be informative about the importance of genes/proteins. PdgL was not translocated, but possesses a PhoP box as other PhoP-regulated virulence proteins (e.g., MgtC, PagK, VirK, PipD) responding to the intracellular environment [47, 48] and its inactivation rendered Salmonella hypervirulent in mice infection, suggesting its crucial role concerned with Salmonella intracellular fitness , a possibility we are currently examining. Our approach, combining phenotype-specific regulatory mutants with computational analysis of resultant multi-omics data, provides a novel and useful method for prediction of virulence-related genes.
Osborne et al.  recently reported that srfN, an ancestral PhoP-regulated gene, acquired an SsrB-regulatory module during Salmonella's evolution to a pathogenic bacterium. We also verified SsrB- and PhoP-dependent transcription of srfN measuring mRNA levels during growth in AMM1 medium (Additional file 12, Figure S8). However, they did not observe SrfN translocation to the host cytoplasm . The most probable explanation for this apparent discrepancy is due to the different cell types used, i.e., epithelial cells used in that study versus macrophages used in our study. Secretion of effector proteins may be specific to the intracellular environment, which has been reported for SseJ . SseJ, a SPI-2 effector was not translocated into the cytosol of HeLa cells at least at 2 h post-infection . Macrophage-biased expression was also reported for a number of SsrB-regulated SPI-2 genes .
PagK was first identified as a PhoP-activated virulence gene, where TnphoA insertion decreased Salmonella virulence significantly in mice . However, a deletion strain lacking pagK exhibited a wild-type phenotype in virulence . Although Gunn et al. identified a homologue, pagJ, in a 1.6 kb duplicated DNA region; they could not reproduce virulence attenuation with deletions of these homologue genes individually or in combination, suggesting that the original transposon insertion may have affected expression of additional genes . This virulence trait is explained by the results described here because the presence of a third related gene was not detected in that study and all three must be deleted to see the largest virulence defect. PagK (or PagK1) has a high amino acid identity with PagK2 (95%) as well as PagJ (83%). Furthermore, all of these three homologues were translocated into the host cytoplasm and negatively regulated in a similar way by SPI-2 type III secretion, suggesting a close correlation and shared properties among three PagK-homologous proteins. Supporting the possibility of a partially redundant function among the three homologues, the attenuated survival of bacteria lacking all three genes was able to be complemented only in part by PagK2 expression in trans (Figure 4).
Comparing mRNA levels in ΔssrA/ssrB and ΔphoP/phoQ strains, transcription of srfN and pag genes (pagC, pagD, and pagK1/pagK2/pagJ) was strongly dependent on PhoP, as noted previously [27, 38], and to a lesser extent on the SPI-2-encoded two-component regulator SsrA/SsrB, suggesting a positive role of the SPI-2 system in their regulation (Additional file 11, Figure S7). However, the absence of SPI-2 T3SS increased the levels of SrfN and Pag proteins inside macrophages, even though bacterial growth was restrained due to the lack of SPI-2 T3SS (Figure 5). Considering the result that the lack of SPI-2 T3SS did not increase mRNA levels of srfN and pagK1/pagK2/pagJ inside macrophages (Additional file 10, Figure S6), the increase in SrfN and Pag proteins must be a consequence of post-transcriptional regulation. The negative regulation of the translation of srfN and pagK homologues by SPI-2 T3SS raises the possibility that an activator such as a small non-coding RNA differentially regulates their translation in response to the lack of SPI-2 T3SS secretion following macrophage infection. The unusually long 5'-untranslated region of srfN (654-bp upstream of the translation start site; ) may form a complex hairpin stem-loop secondary structure that would block the ribosomal binding site and thereby interfere with translation initiation. We observed that deleting N-terminal amino acids in SrfN abolished the negative regulation of SPI-2 T3SS on srfN translation and, furthermore, Hfq overexpression increased srfN translation, but not its transcription (unpublished data). These observations support the hypothesis that a small RNA, as yet unidentified, might bind to the srfN mRNA 5'-region covering the N-terminus and alleviate an inhibitory secondary structure at the translation initiation site, sensing the absence of SPI-2 type III secretion, as is being investigated further.
The T3SS is a typical mechanism for pathogenic bacteria to deliver virulence factors into extracellular environments. However, SrfN and PagK homologues were translocated into the host cytoplasm independently from any T3SSs tested in this study. As a vehicle system to deliver bacterial components to the extracellular milieu, outer membrane vesicles (OMV) have been studied over the past few decades [[51–53]]. Deatherage et al. recently proposed a mechanism for OMV biogenesis wherein envelope protein interconnections modulate OMV release . OMV, observed in a variety of Gram-negative pathogens, are composed of outer membrane proteins, periplasmic proteins, lipopolysaccharide (LPS), and phospholipids and transfer bacterial DNAs and virulence factors to adjacent bacterial or host cells [[51, 52, 55]]. PagC was recently found to be secreted extracellularly via OMV as a major component of the outer membrane . The result that PagC was regulated and translocated in a similar manner with SrfN and PagK homologues in our study suggests the possibility of OMV-mediated transfer of SrfN and Pag proteins. In fact, PagK homologues were distributed as punctate compartments apart from intracellular bacteria inside macrophages in microscopic observation, supporting the possibility .