Characterization of the Citrobacter rodentium Cpx regulon and its role in host infection

Envelope‐localized proteins, such as adhesins and secretion systems, play critical roles in host infection by Gram‐negative pathogens. As such, their folding is monitored by envelope stress response systems. Previous studies demonstrated that the Cpx envelope stress response is required for virulence of Citrobacter rodentium, a murine pathogen used to model infections by the human pathogens enteropathogenic and enterohemorrhagic Escherichia coli; however, the mechanisms by which the Cpx response promotes host infection were previously unknown. Here, we characterized the C. rodentium Cpx regulon in order to identify genes required for host infection. Using transcriptomic and proteomic approaches, we found that the Cpx response upregulates envelope‐localized protein folding and degrading factors but downregulates pilus genes and type III secretion effectors. Mouse infections with C. rodentium strains lacking individual Cpx‐regulated genes showed that the chaperone/protease DegP and the disulfide bond oxidoreductase DsbA were essential for infection, but Cpx regulation of these genes did not fully account for attenuation of C. rodentium ΔcpxRA. Both deletion of dsbA and treatment with the reducing agent dithiothreitol activated the C. rodentium Cpx response, suggesting that it may sense disruption of disulfide bonding. Our results highlight the importance of envelope protein folding in host infection by Gram‐negative pathogens.


Introduction
The Gram-negative envelope, which consists of the inner and outer membranes (IM and OM) and intervening periplasmic space, plays a critical role in the cell's interactions with its environment. Envelope-localized proteins perform essential functions, including nutrient uptake, extrusion of waste and toxic molecules, electron transport, adherence to surfaces, motility and signal transduction, among others. As such, bacteria require a means for sensing and correcting problems with protein folding in the envelope. Among the numerous envelope stress responses present in enterobacteria, the Cpx envelope stress response plays a particularly important role in monitoring the folding of periplasmic and IM proteins (Vogt and Raivio, 2012;Raivio, 2014;Guest and Raivio, 2016). The Cpx response is mediated by a two-component system consisting of the IM-localized histidine kinase CpxA and the cytoplasmic response regulator CpxR. Several activating cues for Escherichia coli CpxA have been identified, including alkaline pH, alterations to the phospholipid composition of the IM, and expression of exogenous pilin proteins in the absence of their cognate chaperones (Jones et al., 1997;Mileykovskaya and Dowhan, 1997;Danese and Silhavy, 1998;Nevesinjac and Raivio, 2005). Although the molecular nature of the CpxA activating cue is still unknown, all of the known inducing conditions are expected to generate misfolded proteins in the envelope. Activation of CpxA causes it to autophosphorylate at a conserved histidine residue and subsequently act as a CpxR kinase (Raivio and Silhavy, 1997). Phosphorylated CpxR then acts as a transcription factor to activate or repress transcription of dozens of genes. In E. coli, in which the Cpx regulon has been best characterized, CpxR activates expression of a suite of periplasmic chaperones and proteases (such as degP, dsbA, ppiA and spy) and represses expression of envelope-localized protein complexes, such as flagella Danese and Silhavy, 1997;Pogliano et al., 1997;De Wulf et al., 1999;Raivio et al., 2000;. By increasing the cell's capacity to degrade or refold envelope proteins while also reducing the flux of proteins entering the envelope, the Cpx response reduces the burden of misfolded proteins in the envelope compartment.
Given the many envelope-localized proteins that play a crucial role in pathogens' ability to infect their hosts, such as fimbrial and non-fimbrial adhesins and secretion systems, it is unsurprising that envelope stress responses are essential for the virulence of many pathogens (Raivio, 2005;Vogt and Raivio, 2012). The effect of the Cpx response on expression of virulence determinants has been particularly well studied in attaching and effacing (A/E) pathogens. A/E organisms are a group of non-invasive diarrheal pathogens, including the human pathogens enteropathogenic E. coli (EPEC) and enterohemorrhagic E. coli (EHEC) (Moon et al., 1983;Sherman et al., 1988;reviewed in Croxen et al., 2013). A/E pathogens initially adhere to the host intestinal epithelium using pili -typical EPEC strains use the bundle-forming pilus (BFP) for this purpose (Girón et al., 1991;Cleary et al., 2004). Subsequently, the pathogen uses a type III secretion system (T3SS) encoded by the locus of enterocyte effacement (LEE) pathogenicity island to inject a number of effector proteins into the host cytoplasm.
The Cpx response is known to affect expression of both pili and T3SSs that are essential for A/E pathogen virulence. Studies in EPEC and EHEC have examined the effect of two opposing changes to Cpx activity: the Cpx response can be inactivated through mutation of cpxR, or it can be constitutively activated -for example, by introducing signal-blind mutations into CpxA that cause it to constitutively phosphorylate CpxR. When the Cpx response is inactivated in EPEC, abundance of the BFP proteins is reduced, as is EPEC's ability to adhere to cultured human cells (Nevesinjac and Raivio, 2005). Decreased pilus elaboration in the absence of Cpx pathway activity has been attributed to reduced expression of envelope-localized protein folding factors, including DsbA, DegP and CpxP, which have been shown to promote stability of pilus component proteins (Zhang and Donnenberg, 1996;Vogt et al., 2010;Humphries et al., 2010). However, inactivation of cpxR has little effect on the transcription of the bfp genes or the expression and in vitro functionality of the T3SS in EPEC and EHEC (MacRitchie et al., 2008;De la Cruz et al., 2016). Conversely, constitutive Cpx activation dramatically reduces expression of the BFP in EPEC and the T3SS in both EPEC and EHEC (MacRitchie et al., 2008;Vogt et al., 2010;De la Cruz et al., 2016). Repression of the BFP when the Cpx response is activated occurs primarily at the level of transcription of the bfp genes, while repression of the T3SS involves both transcriptional and post-translational effects (MacRitchie et al., 2008;Vogt et al., 2010). Together, these results suggest that, at basal levels of pathway activity, the Cpx response promotes biogenesis of pili by enhancing expression of envelope protein chaperones and proteases; however, when the Cpx pathway is strongly activated, expression of pili and the T3SS is repressed, which may help to reduce protein traffic to the envelope. These studies provided a great deal of insight into the molecular mechanisms by which the Cpx response regulates expression of EPEC and EHEC virulence factors; however, they could not answer the question of whether the Cpx response would be beneficial to these A/E pathogens in vivo.
Both EPEC and EHEC are human-specific pathogens that have a limited ability to colonize the mouse intestine and do not produce symptoms of disease in mice that reflect those observed in human infections (Mundy et al., 2006). For this reason, Citrobacter rodentium, a natural pathogen of mice that also carries the LEE, is frequently used to model A/E infections (Collins et al., 2014). In many strains of mice, including C57BL/6, C. rodentium infection leads to self-limiting colitis that resolves approximately three weeks post-infection (Simmons et al., 2002). However, in susceptible strains such as C3H/HeJ, infection with wild-type C. rodentium is lethal within 6-10 days .
The effect of a ΔcpxRA loss-of-function mutation on the ability of C. rodentium to infect its host was previously examined (Thomassin et al., 2015;2017). The ΔcpxRA mutant is severely attenuated in its ability to infect both C57BL/6 and C3H/HeJ mice. Compared to wild-type C. rodentium, the ΔcpxRA mutant produces fewer histopathological changes in C57BL/6 mice and causes no mortality in C3H/HeJ mice; both of these changes may be related to the significantly lower level of gut colonization by the ΔcpxRA mutant (Thomassin et al., 2015). However, since the ΔcpxRA mutant has no defect in growth or T3S in vitro (Thomassin et al., 2015), the reason for the severe attenuation of ΔcpxRA in vivo was unclear.
In this study, we characterized the C. rodentium Cpx regulon in order to identify the mechanisms by which the Cpx response promotes host infection by C. rodentium. We used RNA-Seq and stable isotope labeling by amino acids in cell culture (SILAC) approaches in parallel to identify transcripts and proteins, respectively, whose abundance is altered in C. rodentium ΔcpxRA. To our knowledge, this is the first direct comparison of the Cpx two-component system's effects on the cellular transcriptome and proteome in the same genetic background in any organism. Follow-up studies with mutants lacking a subset of Cpx-regulated genes identified degP and dsbA as being particularly important for C. rodentium's ability to infect mice, although Cpx regulation of these genes could not account for the severe attenuation of C. rodentium ΔcpxRA. Disruption of disulfide bonding through deletion of dsbA or treatment with a chemical reducing agent activated the C. rodentium Cpx envelope stress response, suggesting that sensing and correcting envelope protein misfolding may promote A/E pathogenesis.

Results
Characterization of the C. rodentium Cpx regulon To characterize the C. rodentium Cpx regulon, we used a dual approach: RNA-Seq was used to identify transcripts whose abundance differs between C. rodentium DBS100 and ΔcpxRA (Dataset S1), and SILAC was used to characterize changes in whole-cell protein abundance between strains (Dataset S2). SILAC enables the detection of post-transcriptional and post-translational effects that would be missed by RNA-Seq alone, such as changes in protein abundance mediated by altered protease activity. For both RNA-Seq and SILAC, C. rodentium was grown statically in DMEM in 5% CO 2 , a condition which is known to activate expression of virulence genes, such as those encoding the T3SS . Using a cutoff of a twofold change in abundance between strains with a false-discovery rate < 0.05, we found 338 transcripts that were differentially expressed by RNA-Seq (207 transcripts more abundant in wild-type than in ΔcpxRA and 131 more abundant in ΔcpxRA than in wild-type; Dataset S3); by SILAC, we found 19 proteins that were differentially expressed (8 proteins more abundant in wild-type than in ΔcpxRA, and 11 more abundant in ΔcpxRA than in wild-type; Dataset S3). A comparison of the RNA-Seq and SILAC hits revealed that, while five out of eight proteins positively regulated at least twofold by CpxAR (i.e. more abundant in wild-type than in ΔcpxRA) were also upregulated at least twofold at the transcript level in the RNA-Seq results (Fig. 1A), only 3 of 11 proteins negatively regulated at least twofold by CpxAR (i.e. more abundant in ΔcpxRA than in wild-type) were also downregulated at least twofold at the transcript level (Fig. 1B). These results suggest that the post-transcriptional/post-translational effects mediated by CpxAR may primarily act to decrease protein abundance.
Many of the transcripts/proteins that are more abundant in DBS100 than in ΔcpxRA represent envelope-localized protein folding and turnover factors (Table 1). These include genes known to be Cpx-activated in E. coli, including those encoding the chaperone and regulator of the Cpx response CpxP, the modulator of proteolysis YccA, the endoprotease HtpX, the chaperone Spy, the peptidyl-prolyl cis-trans isomerase PpiA, and the disulfide bond isomerase DsbA (Vogt and Raivio, 2012). In addition, a number of genes/proteins not previously linked to the Cpx response in E. coli were found to be positively regulated by CpxAR in C. rodentium, including the protease PtrA and the serine protease inhibitor ecotin. The protease/chaperone DegP, which is a well-characterized Cpx regulon member in E. coli, did not pass the twofold cutoff in either the RNA-Seq or the SILAC screen (Table 1; Datasets S1 and S2); however, both the transcript and the protein were present at significantly higher levels in DBS100 than ΔcpxRA (FDR < 0.05). By RT-qPCR, we confirmed that transcripts for cpxP, yccA, eco, spy, and ppiA were significantly more abundant in DBS100 than in ΔcpxRA (P < 0.05, Fig. 2A); transcripts for degP and dsbA were also more abundant in wild-type C. rodentium, but the difference between strains was not significant after correction for multiple comparisons ( Fig. 2A). All of these genes were expressed at a similar level in ΔcpxR and ΔcpxA single mutants as in the ΔcpxRA double mutant (Fig. S1A-G), suggesting that mutation of either cpxR or cpxA is sufficient to inactivate the pathway. Single-copy chromosomal complementation of the ΔcpxRA, ΔcpxR and ΔcpxA mutations restored expression of these genes to their wild-type levels ( Fig. S1A-G).
RNA-Seq showed that genes encoding several pilus components were also differentially expressed in the proteins differentially expressed at a cutoff of twofold in DBS100 and ΔcpxRA with a statistical cutoff of FDR < 0.05, as determined by RNA-Seq and SILAC respectively. A. Transcripts/proteins positively regulated by CpxAR (more abundant in wild-type than in ΔcpxRA). B. Transcripts/proteins negatively regulated by CpxAR (more abundant in ΔcpxRA than in wild-type).

A B
ΔcpxRA mutant. The kfcCDEFG transcript encoding a K99-type chaperone-usher pilus was less than half as abundant in DBS100 as in ΔcpxRA (Table 1). Although not every protein encoded in this operon was detected by SILAC, KfcC was also less abundant in DBS100 at the protein level (Table 1; Dataset S2). Downregulation of kfcC by the Cpx response was confirmed by RT-qPCR ( Fig. 2B; expression in ΔcpxR and ΔcpxA and complemented strains shown in Fig. S1H). Numerous T3S-related genes and proteins were found to be differentially expressed in ΔcpxRA by both RNA-Seq and SILAC (Table 1). Several non-LEE encoded effectors were found to be expressed at higher levels in the ΔcpxRA mutant by nleG1,nleE,espK,nleC,nleG8,nleG7,espX7,espM3 and espS); although not all of these proteins were detectable by SILAC, those that were detectable (NleB1, NleG1, NleE, NleC, EspX7 and EspS) were generally also found to have higher protein abundance in ΔcpxRA. Expression of nleB1, the most downregulated non-LEE encoded effector gene in wild-type C. rodentium when compared with the ΔcpxRA mutant according to RNA-Seq, was further analyzed by RT-qPCR. The latter confirmed that the nleB1 transcript is approximately twofold less abundant in wild-type C. rodentium than in the ΔcpxRA mutant (Figs 2C and S1I). Interestingly, no LEE-encoded transcripts were found to be differentially expressed by RNA-Seq; however, the LEE-encoded T3SS translocator protein EspB was found to be greater than twofold more abundant in ΔcpxRA by SILAC (Table 1). Using RT-qPCR, we confirmed that the espB transcript was equally abundant in both strains (Figs 2C and S1J). Western blotting showed that the EspB protein was more abundant in whole cell lysates of the ΔcpxR and ΔcpxA mutants (2.3-fold and 1.4-fold higher than the wild-type level respectively), although increased cellular abundance of EspB was not consistently observed in the ΔcpxRA mutant itself (Fig. 2D). The abundance of EspB secreted into the culture supernatant did not differ between wild-type and cpx mutant strains, nor did abundance of the other major secreted proteins, EspA and EspD (Fig. S2). These results suggest that the C. rodentium Cpx response negatively regulates expression of T3S-related genes in two separate ways: the non-LEE encoded effectors appear to be repressed primarily at the transcriptional level, while the LEE-encoded protein EspB may be repressed at the post-transcriptional or post-translational level.
RT-qPCR was also performed to confirm differential expression of several additional genes belonging to other functional categories. By RNA-Seq and SILAC, expression of rdoA was found to be significantly higher in wild-type C. rodentium (Table 1). RdoA is a serine/threonine protein kinase involved in regulating programmed cell death (Dorsey-Oresto et al., 2013); rdoA is encoded immediately upstream of dsbA and the two genes are known to be co-transcribed from a Cpx-activated promoter in E. coli (Pogliano et al., 1997). RT-qPCR analysis confirmed that rdoA is modestly but reproducibly positively regulated by CpxAR in C. rodentium as well (Figs 2E and S1K). Four genes comprising a putative operon (ROD_17451, ROD_17461, ROD_17471 and ROD_17481) encode several of the transcripts and proteins that were most strongly activated by CpxAR in both the RNA-Seq and SILAC datasets (Table 1). ROD_17451 is homologous to E. coli gene of unknown function yciG; ROD_17461 and ROD_17471 both contain putative ruberythrin/ferritin-like metal-binding domains, and ROD_17481 is a putative Mn-containing catalase. RT-qPCR confirmed that all four genes are positively regulated by CpxAR in C. rodentium (Figs 2E and S1L). Finally, several proteins involved in 1,2-propanediol utilization (PduE, PduB and PduJ) were or SILAC were validated by RT-qPCR (A-C, E) or Western blotting (D) respectively. cDNA templates for RT-qPCR were prepared from strains grown in DMEM in biological triplicate cultures. Western blots were performed using whole-cell lysates from strains grown in DMEM. Significant differences in transcript abundance in DBS100 and ΔcpxRA were assessed using multiple t-tests with the Holm-Sidak correction for multiple comparisons: *, P < 0.05; **, P < 0.01; ***, P < 0.001.

A B D E
C found by SILAC to be repressed by CpxAR in C. rodentium (~3-fold less abundant in wild-type than ΔcpxRA; Table  1), yet the RNA-Seq results indicate that the transcripts encoding these proteins were present at similar levels in the two strains. Three additional Pdu proteins (PduK, PduA and PduD) were also found to be downregulated more than twofold by SILAC but not by RNA-Seq; however, since these proteins were identified in only one of three SILAC replicates, they were not considered statistically significant (Dataset S2). By RT-qPCR, we confirmed that the pduJ transcript is not differentially expressed in ΔcpxRA compared to DBS100 (Figs 2E and S1M), suggesting that the Pdu proteins are repressed by CpxAR at the post-transcriptional or post-translational level.

Identification of Cpx-regulated genes required for host infection
Having characterized the C. rodentium Cpx regulon, we next aimed to determine whether any of the Cpxregulated genes play a role in host infection. To this end, we generated C. rodentium mutants carrying deletions in individual genes or operons positively regulated by CpxAR. Since the large number of Cpx-regulated genes precluded analysis of every Cpx regulon member, we prioritized genes that are known to be directly regulated by CpxR in E. coli (cpxP, degP, dsbA, ppiA, spy and yccA) or that were strongly upregulated by CpxAR in the RNA-Seq or SILAC experiments (ROD_17451-81 putative operon, eco).
Prior to performing infection studies, we performed preliminary characterization of the mutants in vitro. All of the mutants grew at a rate comparable to wild-type C. rodentium in both rich medium (LB) and conditions known to stimulate expression of virulence genes (DMEM with 5% CO 2 ) (Fig. S3). Most of the Cpx regulon mutants had secreted protein profiles that were similar to that of wildtype strain DBS100; however, the supernatant of the ΔdegP mutant contained large amounts of non-type III-secreted proteins, while the ΔdsbA mutant supernatant contained reduced amounts of EspA, EspB and EspD (Fig. 3). Western blotting of secreted proteins with an α-EspB antibody confirmed reduced levels of EspB in the ΔdsbA culture supernatant and also revealed that, in spite of the increased quantity of protein in the ΔdegP supernatant, the amount of EspB secreted by this strain was actually less than the wildtype level (Fig. 3). Similar phenotypes have been observed in EPEC ΔdegP and ΔdsbA mutants (Miki et al., 2008;MacRitchie et al., 2012), suggesting that these proteins play a conserved role in envelope integrity and assembly or function of the T3S machinery respectively. Single-copy chromosomal complementation of the ΔdegP and ΔdsbA mutants completely restored the aberrant secreted protein profiles to match the wild-type phenotype (Fig. S4A).
We next assessed the ability of the C. rodentium mutants to colonize and cause lethal infection in C3H/HeJ mice. In agreement with previous studies, we found that the ΔcpxRA mutant was attenuated in this mouse model (Fig. 4A). However, in our experiments, the ΔcpxRA mutant was still able to cause lethal infection in about half of the mice, in contrast to the 100% survival previously reported (Thomassin et al., 2015;2017). The reason for the differing results is unknown, since all C3H/HeJ mice originated from the same supplier. However, C. rodentium infection outcomes are known to be sensitive to alterations to the gut microbiota Wlodarska et al., 2011;Kamada et al., 2012), and therefore, differences in mouse chow or other environmental parameters that affect gut microbiota composition between animal facilities might affect the virulence of C. rodentium mutants. The ΔcpxR and ΔcpxA single mutants were attenuated to a similar degree as the ΔcpxRA double mutant (Fig.  S5), as expected based on the similarity in gene expression between these three strains (Fig. S1). The majority of the Cpx regulon mutants were indistinguishable from wild-type C. rodentium in their ability to cause lethal infection (Fig. 4A); however, the ΔdegP and ΔdsbA mutants did not kill any of the infected animals. Virulence was fully restored to the ΔdegP and ΔdsbA mutants by single-copy chromosomal complementation ( Fig. S4B and C). All of the attenuated mutants (ΔcpxRA, ΔdegP and ΔdsbA) were able to colonize the mouse gut, as assessed by fecal shedding of C. rodentium, albeit at a level 1 to 2 logs lower than the wild-type strain (Fig. 4B). Thus, among the Cpx regulon members tested, DegP and DsbA are most important for the ability of C. rodentium to colonize and cause infection in C3H/HeJ mice.

Cpx regulation of degP and dsbA does not fully account for CpxAR's role in virulence
Since the ΔdegP and ΔdsbA mutants were essentially avirulent in C3H/HeJ mice (Fig. 4), we hypothesized that the decreased colonization and virulence of the ΔcpxRA mutant could result from its reduced expression of degP and dsbA. To test this hypothesis, we sought to identify and mutate the CpxR binding sites upstream of degP and dsbA, leaving the genes and promoters otherwise intact.
The location of the CpxR binding site in the E. coli degP promoter has been experimentally identified (Pogliano et al., 1997), and a sequence alignment shows that the CpxR binding motifs are conserved in the C. rodentium degP promoter (Fig. 5A). In order to verify that this region is responsible for CpxAR regulation of the degP promoter in C. rodentium, we cloned the degP promoter into the luxCDABE transcriptional reporter plasmid pNLP10. We then measured activity of the wild-type degP reporter as well as a reporter with a 25-bp deletion encompassing the CpxR boxes (as shown in Fig. 5A) in both wild-type and ΔcpxRA strains of C. rodentium. Activity of the wildtype degP promoter was approximately four-fold higher in wild-type C. rodentium than in ΔcpxRA (Fig. 5B); however, activity of the P degPΔCpxR reporter did not differ between DBS100 and ΔcpxRA, suggesting that the deleted region contained the CpxR binding sites. To confirm that no other region of the degP promoter is required for CpxR regulation, we generated a chromosomal deletion of the same 25 bp in the degP promoter and measured degP transcript abundance by RT-qPCR. While introduction of the ΔcpxRA deletion into C. rodentium with a wild-type degP promoter caused a significant decrease in degP transcript abundance, no change in degP transcript abundance was seen when the ΔcpxRA allele was introduced into the P degPΔCpxR background (Fig. 5C). These results indicate that the same region of the degP promoter is responsible for CpxR activation in both E. coli and C. rodentium.
In E. coli, dsbA is expressed both from a proximal promoter directly upstream of dsbA and from a distal promoter located upstream of rdoA, the gene immediately upstream of dsbA (Belin and Boquet, 1994); only the distal promoter is subject to CpxR regulation (Pogliano et al., 1997). To assess the effect of CpxAR on the rdoA and dsbA promoters in C. rodentium, the two promoter

A B
regions were cloned into medium-copy luminescence reporter vector pJW15. Activity of the rdoA reporter was significantly higher in wild-type C. rodentium than in ΔcpxRA, while activity of the dsbA reporter was actually slightly higher in the ΔcpxRA mutant (Fig. 6A), indicating that CpxR activates dsbA expression via the rdoA promoter in C. rodentium, similarly to E. coli. Next, we set out to mutate the rdoA promoter in order to abolish Cpx regulation of the rdoA-dsbA operon. However, we could not delete the entire CpxR binding region as we did with the degP promoter, because: (i) the CpxR binding region in the rdoA promoter partially overlaps the coding sequence of the upstream gene ROD_39011 (underlined in Fig. 6B) -we therefore needed to ensure that any mutations were synonymous with respect to the ROD_39011 coding sequence; and (ii) deletion of the 3′ CpxR box completely abolished rdoA promoter activity (data not shown), perhaps due to its proximity to the −35 element (Fig. 6B). For this reason, we introduced three point mutations into the first CpxR box as shown in Fig. 6B. These point mutations abolished CpxAR regulation of the mutant P rdoAmut -lux reporter (Fig. 6C), although the lower activity of this reporter relative to the wild-type P rdoA -lux reporter in the ΔcpxRA background suggested that the mutations may have also affected basal promoter activity. However, when the P rdoAmut mutations were introduced into the chromosome, abundance of the rdoA and dsbA transcripts was not reduced below their levels in the ΔcpxRA mutant ( Fig. 6D and E), indicating that the low expression of P rdoAmut -lux may have been a multicopy artifact. In order to examine whether Cpx regulation of degP and dsbA is important for C. rodentium host infection, we infected C3H/HeJ mice with C. rodentium strains carrying one or both of the P degPΔCpxR and P rdoAmut mutations. It is important to note that, although rdoA expression is also reduced in the P rdoAmut strain, we assumed that any decrease in virulence would be due to misregulation of dsbA, since a ΔrdoA mutant was not attenuated in this model (Fig. S6). There was no A. Alignment of the degP promoter sequences from E. coli MG1655 (Eco) and C. rodentium ICC168 (Crod). The CpxR binding region identified by DNase footprinting in E. coli (Pogliano et al., 1997) is identified with a red arrow, and two inverted CpxR binding motifs are identified with red boxes. The 25 bp region deleted in the C. rodentium promoter to give rise to P degPΔCpxR is shown with a blue box. Coordinates are relative to the translational start site. B. Activity of P degP -lux and P degPΔCpxR -lux reporters in wild-type and ΔcpxRA strains of C. rodentium. Data represent the mean and standard deviation of three biological replicate cultures. C. Transcript abundance of degP in wild-type and ΔcpxRA strains of C. rodentium, with chromosomal wild-type or P degPΔCpxR promoters, as measured by RT-qPCR. * denotes P < 0.05 and **** denotes P < 0.0001, one-way ANOVA with Sidak's multiple comparison test; n.s., no significant difference.

A B C
statistically significant difference in mortality between mice infected with wild-type C. rodentium and any of the single or double promoter mutants (Fig. 7A, P > 0.05, Mantel-Cox test with Bonferroni's correction for multiple comparisons). Fecal shedding of the promoter mutants was also indistinguishable from that of wild-type C. rodentium, whereas fecal shedding of the attenuated ΔcpxRA, ΔdegP and ΔdsbA mutants tended to be lower than wild-type throughout the course of infection (Fig.  7B). Together, these results indicate that, although degP and dsbA are essential for C. rodentium infection of mice, CpxAR regulation of only these genes does not fully account for the impaired virulence of the ΔcpxRA mutant in this infection model.

Disulfide bond disruption is an inducing signal for the C. rodentium Cpx response
To further examine the physiological role of the C. rodentium Cpx response, we next addressed the question of which signal activates the response. Although it has previously been reported that the C. rodentium Cpx response is activated at alkaline pH (Thomassin et al., 2015), the molecular nature of the inducing signal remains unknown.
Since the Cpx response is believed to sense protein misfolding in the inner membrane and/or periplasm (Vogt and Raivio, 2012), we examined whether any of our Cpx regulon mutants -many of which lack important periplasmic protein folding and degrading factors -had altered A. Activity of P rdoA -lux and P dsbA -lux reporters in wild-type and ΔcpxRA strains of C. rodentium. B. Alignment of the rdoA promoter sequences from E. coli MG1655 (Eco) and C. rodentium ICC168 (Crod). The CpxR binding region identified by DNase footprinting in E. coli (Pogliano et al., 1997) is identified with a red arrow, and two CpxR binding motifs are identified with red boxes. The three substitution mutations introduced into the C. rodentium promoter to give rise to P rdoAmut are shown in blue. The coding sequence of the gene upstream of rdoA (ROD_39011) is underlined. Coordinates are relative to the translational start site. C. Activity of P rdoA -lux and P rdoAmut -lux reporters in wild-type and ΔcpxRA strains of C. rodentium. D. and E. Transcript abundance of rdoA (D) and dsbA (E) in wild-type and ΔcpxRA strains of C. rodentium, with chromosomal wild-type or P rdoAmut promoters, as measured by RT-qPCR. Data from luminescence assays represent the mean and standard deviation of three biological replicate cultures. * denotes P < 0.05, ** denotes P < 0.01, *** denotes P < 0.001 and **** denotes P < 0.0001, one-way ANOVA with Sidak's multiple comparison test; n.s., no significant difference.

A B C D E
Cpx pathway activity using a cpxP-lux transcriptional reporter. Since cpxP is the gene that was most strongly activated by CpxAR in C. rodentium according to RNA-Seq and RT-qPCR (Table 1 and Fig. 2A) and is not known to be regulated at the transcriptional level by any regulators other than CpxR in either C. rodentium or E. coli, activity of the cpxP-lux transcriptional reporter is a good proxy for Cpx pathway activity. As expected, we found that the ΔcpxRA mutant had dramatically reduced cpxPlux activity compared to the wild-type strain (Fig. 8A). In addition, two regulon mutants had cpxP-lux activity that was significantly higher than DBS100 (P < 0.0001). The C. rodentium ΔcpxP mutant had an approximate fourfold increase in cpxP-lux activity compared to the wild-type strain (Fig. 8A), which is in keeping with the finding that CpxP acts as a negative regulator of Cpx pathway activity in E. coli (Raivio et al., 1999). Interestingly, the ΔdsbA mutant had even higher cpxP-lux activity than the ΔcpxP mutant ( Fig. 8A), suggesting that disruption of disulfide bonding in the envelope could act as an inducing cue for the C. rodentium Cpx response. To further examine this idea, we measured the activity of the Cpx pathway in the presence of dithiothreitol (DTT), a chemical reducing agent known to disrupt disulfide bonds in proteins. We found that activity of the cpxP-lux reporter increased in the presence of DTT in a dose-dependent manner (Fig.  8B), while activity of several lux reporters not regulated by the Cpx response was not increased by DTT (data not shown). These data suggest that disruption of disulfide bonding in envelope proteins may represent a physiological activating cue for the C. rodentium Cpx response.

Discussion
Citrobacter rodentium harbours 26 two-component systems (2CSs); mutants in only six of these 2CSs are attenuated in the mouse infection model, with ΔcpxRA having the largest virulence defect (Thomassin et al., 2017). The main question we set out to answer in this study is why CpxAR is so important for host infection.
Using both transcriptomic and proteomic approaches, we found that CpxAR regulates expression of several hundred genes and proteins at both the transcriptional and post-transcriptional/post-translational level (Dataset S3). Numerous envelope-localized chaperones and proteases were positively regulated by CpxAR, while the Kfc pilus, several T3S effectors and several proteins comprising the propanediol utilization microcompartment were negatively regulated by CpxAR (Table 1 and Fig. 1). Among these Cpx-regulated genes, we were able to identify two genes, encoding the major periplasmic protease DegP and the disulfide bond oxidoreductase DsbA, that are essential for C. rodentium virulence in C3H/HeJ mice (Fig. 4). Therefore, we propose that ensuring correct folding of envelope proteins is a major physiological role of the Cpx response in C. rodentium that is likely important during host infection. Our results are consistent with a previous study that found that a C. rodentium degP mutant had a reduced ability to colonize C57BL/6 mice (Cheng et al., 2012). One possible reason for the virulence defect of the ΔdegP and ΔdsbA mutants could be related to problems with assembly of the T3SS. We noticed that the ΔdegP and ΔdsbA mutants both had aberrant T3S profiles (Fig. 3), similar to previous findings in EPEC (MacRitchie et al., 2012). Miki et al. (2008) showed that DsbA catalyzes disulfide bond formation in EPEC EscC, the outer membrane secretin component of the T3SS, and that this disulfide bond is likely formed between cysteine residues 136 and 155. Given that both of these cysteines are conserved in C. rodentium EscC (data not shown), it is likely that DsbA performs a similar role in C. rodentium. In addition A. Survival of mice infected with wild-type and mutant strains of C. rodentium. Mice were monitored daily and euthanized upon reaching the humane endpoint described in Experimental procedures. * denotes P < 0.05, Mantel-Cox test with Bonferroni's correction for multiple comparisons. B. Fecal shedding of C. rodentium throughout the course of infection. The connecting lines denote the mean and error bars denote standard deviation. LOD, limit of detection. Both panels show the combined results of six separate experiments, with a total of N = 30 mice per strain for DBS100 and ΔcpxRA, N = 20 mice per strain for ΔdegP and ΔdsbA, N = 10 mice per strain for P degPΔCpxR P rdoAmut , and N = 5 mice per strain for P degPΔCpxR and P rdoAmut .

A B
to facilitating proper biogenesis of the T3SS, DegP and DsbA also promote biogenesis of the bundle-forming pilus in EPEC (Zhang and Donnenberg, 1996;Vogt et al., 2010;Humphries et al., 2010). Since several pili reportedly contribute to C. rodentium colonization of the mouse gut (Mundy et al., 2003;Hart et al., 2008;Caballero-Flores et al., 2015), proper folding of these proteins could be another important role for DegP and DsbA in vivo.
Despite the fact that the ΔdegP and ΔdsbA deletion mutants were avirulent in C3H/HeJ mice (Fig. 4), mutation of the CpxR boxes located upstream of the two genes did not significantly reduce virulence compared to wildtype C. rodentium (Fig. 7). This discrepancy can likely be attributed to the relatively weak activation of these two genes by CpxAR. Although expression of degP and dsbA is reduced by deletion of cpxRA (Table 1, Figs 2A,5,and 6) or by mutation of the CpxR box in their promoter regions (Figs 5 and 6), the decrease in expression is around twofold or less. Therefore, even in the absence of Cpx activation, basal expression of degP and dsbA may be sufficient for proper protein folding in vivo. Alternatively, or in addition, other signaling pathways may upregulate these genes in vivo. This finding leaves open the question of why the ΔcpxRA mutant is attenuated. One possibility is that there remains one or more Cpxregulated genes that are essential for host infection, but whose contribution to virulence was not examined in this study. Although we did delete several of the most strongly Cpx-regulated genes (cpxP, yccA, eco, ROD_17451-81) and found the mutants to be fully virulent, there remain a number of Cpx-regulated genes whose contribution has not yet been examined. Another possible explanation for the attenuation of the ΔcpxRA mutant could be the cumulative effect of misregulation of numerous envelope protein folding factors. Although the decreased expression of degP and dsbA in the ΔcpxRA mutant is not sufficient to cause attenuation (Fig. 7), expression of several additional proteases, protease regulators and chaperones is also reduced in this strain (Table 1 and Fig. 1A). Perhaps these proteases and chaperones can compensate for reduced expression of degP and dsbA in an otherwise wild-type background, but not at their reduced levels in the ΔcpxRA mutant.
Another intriguing possibility is that repression of target genes by CpxR is important in vivo. We found that CpxAR downregulates expression of the kfc pilus gene cluster, numerous T3S effectors, and the Pdu propanediol utilization microcompartment proteins (Table 1 and Fig. 1). All of these proteins are structural components of, or require secretion by, large macromolecular complexes that require substantial cellular resources to produce. Thus, overexpression of all of these proteins might confer a growth disadvantage in vivo. In addition, overexpression of virulence-related proteins can have other detrimental effects. For example, deletion of cpxR in UPEC strain UTI89 causes overexpression of the hemolysin HlyA (Nagamatsu et al., 2015). UTI89 ΔcpxR has a reduced ability to colonize the bladder of C3H/HeN mice, which is likely the result of this strain's increased ability to induce exfoliation of infected urothelial cells (Nagamatsu et al., 2015). Importantly, the virulence defect of the ΔcpxR mutant can be attributed to its increased expression of hlyA, since deletion of hlyA in the ΔcpxR background reduces urothelial cell exfoliation and restores wild-type colonization ability (Nagamatsu et al., 2015). Thus, it is possible that increased expression of virulence proteins, such as T3S effectors, in C. rodentium ΔcpxRA is detrimental to host colonization, perhaps by more strongly inducing host defense pathways.
We found that Cpx repression of several protein complexes in C. rodentium appears to happen partially or entirely post-transcriptionally. Although A. Activity of P cpxP -lux reporter in wild-type and mutant strains of C. rodentium. B. Activity of P cpxP -lux reporter in wild-type C. rodentium in the presence of DTT. Data from luminescence assays represent the mean and standard deviation of three biological replicate cultures. **** denotes P < 0.0001, one-way ANOVA with Dunnett's multiple comparison test.

A B
numerous non-LEE-encoded T3S effectors appeared to be repressed by CpxAR at the transcriptional level (e.g. nleB1; Table 1 and Fig. 2C), none of the genes encoded in the LEE were differentially expressed between DBS100 and ΔcpxRA at the transcript level according to our RNA-Seq data (Table 1), and we confirmed by RT-qPCR that espB transcripts were equally abundant in DBS100 and ΔcpxRA (Fig. 2C). However, we observed by SILAC and confirmed by Western blotting that the T3S translocator protein EspB was more abundant in Δcpx cell pellets than in the wild-type strain (Table 1 and Fig. 1D). In addition, several Pdu proteins involved in formation of the 1,2-propanediol microcompartment were found to be more abundant in ΔcpxRA by SILAC (Table 1), even though their transcripts were not differentially expressed according to RNA-Seq and RT-qPCR (Fig. 2E). Several other examples of post-transcriptional effects mediated by CpxAR have been described in other organisms. For example, deletion of cpxA in EHEC (which activates the Cpx pathway due to loss of CpxA's phosphatase activity) causes decreased expression of LEE-encoded T3SS proteins such as EspA, EspB and EspD (De la Cruz et al., 2016). The repression of T3SS expression in EHEC ΔcpxA is dependent on the presence of Lon protease, suggesting that activation of the Cpx response causes Lon to degrade a regulator required for T3SS gene expression (De la Cruz et al., 2016). Lon is also required for Cpx-mediated repression of T3SS-1 expression in Salmonella Typhimurium; in this case, activation of the Cpx response causes Lon to degrade the regulatory protein HilD (De la Cruz et al., 2015). It is currently unknown whether Lon might be responsible for the post-transcriptional repression of EspB and the Pdu proteins in C. rodentium. However, if Lon is involved, the mechanism likely differs from that in EHEC, since Cpx activation in EHEC causes reduced transcription of LEEencoded genes including espA, ler and tir (De la Cruz et al., 2016), whereas we observed differences in EspB protein abundance but not espB transcript abundance in C. rodentium ΔcpxRA (Table 1 and Fig. 1). In any case, these findings together point to a previously underappreciated ability of the Cpx response to alter the cellular proteome through post-transcriptional mechanisms.
A major outstanding question about the physiological role of the Cpx response pertains to the nature of its inducing cue(s). Here, we found that deletion of dsbA caused activation of the Cpx response in C. rodentium (Fig. 8A). Since treating cells with the chemical reductant DTT also activated CpxAR (Fig. 8B), reduction of disulfide bonds in envelope proteins may represent a cue for Cpx activation in C. rodentium. It is currently unknown how CpxAR might sense problems with disulfide bond formation in envelope proteins. Neither CpxA (the sensor kinase) nor CpxP (the periplasmic inhibitory protein) contain any cysteine residues in C. rodentium (data not shown); thus, they cannot directly sense disruption of disulfide bonding. However, the lipoprotein NlpE, which acts as an accessory regulator capable of activating the Cpx response in E. coli , is capable of forming an intramolecular disulfide bond (Hirano et al., 2007). Since the redox-active cysteine residues in E. coli NlpE are conserved in C. rodentium (data not shown), this could represent a potential mechanism for Cpx sensing of disulfide bond disruption. It is also unclear whether disulfide bond disruption is a physiologically relevant inducer of the Cpx response during C. rodentium growth in the mouse gut, where Thomassin and colleagues (2015) previously showed that the Cpx-activated genes cpxR, cpxA and cpxP are expressed. Several host-derived and microbiota-derived reducing agents are known to be present in the gut; for example, human thioredoxin is expressed in the gut mucosa and is responsible for reducing disulfide bonds in the antimicrobial peptide β-defensin 1 (Schroeder et al., 2011). Further research will be required to determine whether reduction of disulfide bonds contributes to activity of the C. rodentium Cpx response in vivo.
In summary, our analysis of the C. rodentium Cpx regulon demonstrates the conserved role of the Cpx envelope stress response in enterobacteria, with envelope-localized chaperones and proteases being Cpxactivated and envelope protein complexes, such as pili and secretion systems being Cpx-repressed in numerous species. Interestingly, the Cpx response appears to be important for gut colonization by pathogens and commensals alike, since a ΔcpxR mutant of mouse commensal E. coli strain MP1 has a severe colonization defect (Lasaro et al., 2014). Together, these results highlight the importance of envelope protein biogenesis for the ability of Gram-negative bacteria to interact with their hosts.

Bacterial strains and growth conditions
Bacterial strains and plasmids used in this study are listed in Table S1. Unless otherwise stated, all strains were grown in lysogeny broth (LB; 10 g l -1 tryptone, 5 g l -1 yeast extract, 10 g l -1 NaCl) at 37°C with aeration at 225 rpm or on LB agar at 37°C. Antibiotics and supplements were used when necessary at the following concentrations: ampicillin, 100 µg ml -1 ; chloramphenicol, 30 µg ml -1 ; kanamycin, 50 µg ml -1 ; diaminopimelic acid (DAP), 0.3 mM.

RNA-Seq
Strains were grown overnight in biological triplicate cultures in 5 ml of LB at 37°C, aerated at 220 rpm, and then diluted in LB to an OD 600 of 1.0. One ml of each dilution was pelleted at 2300×g for 5 min at room temperature and resuspended in 20 ml of prewarmed DMEM (HG) (Caisson Labs,Cat. Number: DML07), to which arginine and lysine were added to final concentrations of 0.2 and 0.8 mM respectively. Cultures were incubated for 4.5 h at 37°C in a 5% CO 2 incubator, statically. RNA was isolated from these cultures using the MasterPure RNA purification kit (Epicentre) following the manufacturer's instructions and including a further 2 Units of DNase I (Invitrogen) treatment at 37°C for 30 min. Final RNA samples were resuspended in 50 µl of nuclease-free water and reverse transcription and real time qPCR were done to verify DNA depletion using primers specific for the dnaQ gene (Table S2). 20 µg of RNA were submitted for sample preparation and RNA-Seq analysis by GENEWIZ (Plainfield, NJ). Single read sequencing was done on an Illumina HiSeq 2500 platform. Reads were mapped to the C. rodentium ICC168 reference genome (NC_013716, NC_013717, NC_013718 and NC_013719) using EDGE-pro (Magoc et al., 2013) and differentially expressed genes were identified using DESeq2 (Love et al., 2014), based on a > 2 fold or < 0.5 fold change in the wild-type DBS100 strain as compared to the ΔcpxRA mutant with a Benjamini-Hochbergadjusted P value < 0.05.

Sample preparation for whole proteome analysis
Antibiotic-, serum-, arginine-and lysine-free Dulbecco's modified Eagle medium (DMEM, Caisson Laboratories Inc.) was supplemented with l-arginine and l-lysine or l-[ 13 C 6 ] arginine and l-[ 2 H 4 ]lysine and prewarmed at 5% (v/v) CO 2 and 37°C overnight. SILAC-labeled bacteria corresponding to a bacterial load of a 1 ml culture with OD 600 of 1 were centrifuged at 2300×g for 5 min at room temperature and resuspended in 20 ml prewarmed DMEM in biological triplicate cultures. The cultures were incubated standing in a 10 cm petri dish at 5% (v/v) CO 2 and 37°C for 4.5 h. Bacteria were pelleted at 3000×g and 4°C for 10 min, washed once in ice-cold phosphate-buffered saline (PBS), resuspended in 50 mM ammonium bicarbonate and 150 mM sodium deoxycholate and incubated at 99°C under constant agitation for 15 min. MgCl 2 was added to a final concentration of 1.5 mM and DNA digestion was achieved by Benzonase endonuclease (Santa Cruz Biotechnology) at room temperature for 30 min. Subsequent to centrifugation at 16000×g and room temperature for 1 min, the protein concentration was determined by bicinchoninic acid assay (Thermo Scientific Pierce) and the soluble lysate of light and heavy labeled bacteria were combined at a ratio of 1:1. Proteins were reduced with 10 mM dithiothreitol (DTT) at room temperature for 30 min. Samples were then alkylated with 55 mM iodoacetamide in the dark at room temperature for 20 min, sequence grade trypsin (Promega) was added, and protein digestion was achieved under shaking conditions at 37°C for 16 h. Prior to basic reverse-phase fractionation, peptides were desalted using C18 StAGE Tips (Rappsilber et al., 2007).

Basic reverse-phase fractionation
Basic reverse-phase fractionation was undertaken according to the protocol of Udeshi et al. with minor modifications (2013). Briefly, peptides were separated using an 1100 series HPLC instrument with a Zorbax Extend C 18 column (1.0 by 50 mm, 3.5 µm; Agilent) at a flow rate of 100 µl/min. The following gradient was run: initial 5 min from 100% buffer A (5 mM ammonium formate, 2% acetonitrile, pH 10) to 6% buffer B (5 mM ammonium formate, 90% acetonitrile, pH 10), then in 2 min to 8% buffer B, followed by an increase upto 27% buffer B in 38 min, to 31% B in 4 min, to 39% B in 4 min, to 60% B in 7 min, and completion with a 4-min run at 100% buffer B and a 26-min gradient back to 100% buffer A. Fractions of 100 µl were collected in a 96-well plate with every eighth fraction combined to generate a total of eight fractions that were concentrated by vacuum centrifugation and subjected to MS analysis.

MS data analysis
MS data were processed with MaxQuant (v1.5.2.8) (Cox and Mann, 2008). Database searching was carried out against the reference C. rodentium ICC168 proteome (downloaded from UniProt on 22 March 2015; 4775 proteins) with the following search parameters: carbamidomethylation of cysteine as a fixed modification, oxidation of methionine, acetylation of protein N-terminal trypsin/P cleavage with a maximum of two missed cleavages. A multiplicity of two was used, with each multiplicity denoting one of the SILAC amino acid combinations (light and heavy respectively). The precursor mass tolerance was set to six parts-permillion (ppm) and MS/MS tolerance 0.5 Da for LTQ-velos data with a maximum false discovery rate of 1.0% set for protein identifications. To enhance the identification of peptides between fractions and replicates, the Match between Runs option was enabled with a precursor match window set to 2 min and an alignment window of 10 min. The resulting protein group output was processed within the Perseus (v1.5.0.9) (Tyanova et al., 2015) analysis environment to remove reverse matches and common protein contaminants. Normalized, log2 transformed H/L SILAC ratios (wildtype Citrobacter rodentium versus Citrobacter rodentium ΔcpxRA) were calculated. Proteins were considered regulated if they showed an average fold change in abundance of at least twofold from at least two out of three biological replicates and passed the statistical analysis with multiple hypothesis corrections using a Benjamini-Hochberg procedure with FDR of 0.05. The calculation of a protein SILAC ratio by the MaxQuant software required a minimum of two unique peptides to be identified. Some proteins were differentially expressed according to RNA-Seq data, detected in mass spectrometry but no SILAC ratios were determined by the MaxQuant software since less than two unique peptides were identified for these particular proteins. Thus, for reasons of better comparability between both data sets, not protein-but peptide-based SILAC ratios were considered for this particular set of proteins. All mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD009049 (Vizcaíno et al., 2016).

Reverse transcriptase quantitative PCR (RT-qPCR)
For RNA extraction, strains were first cultured overnight in LB, then subcultured into 20 ml of prewarmed Dulbecco's Modified Eagles Medium (DMEM) (HyClone cat. no. SH30243.01) in a 100 mm Petri dish at a starting OD 600 of 0.05. Cultures were incubated for 4.5 h at 37°C in a static incubator with 5% CO 2 . RNA was extracted from 500 µl of culture using RNAprotect Bacteria Reagent (Qiagen) followed by the GeneJET RNA Purification Kit (Thermo Fisher Scientific). Contaminating genomic DNA was removed from 2-µg aliquots of purified RNA using the TURBO DNA-free Kit (Thermo Fisher Scientific), followed by reverse transcription with the QuantiTect Reverse Transcription Kit (Qiagen). RT-qPCR was performed using the QuantiTect SYBR Green PCR Kit (Qiagen) on an Applied Biosystems 7500 Fast Real-Time PCR System, using the ΔΔC T relative quantitation method with dnaQ (which was experimentally verified to be expressed at equal levels in the wild-type and ΔcpxRA strains; data not shown) as the endogenous control. Primers used for RT-qPCR are listed in Table S2; the efficiency of all primer pairs was verified to be within the range of 90-110% (data not shown). No-template and no-reverse transcriptase controls were included in each RT-qPCR plate to confirm the absence of primer dimer and contaminating genomic DNA respectively. All RT-qPCR data represent biological triplicate cultures.

Western blot analysis
Citrobacter rodentium strains were cultured as described above for RT-qPCR analysis. After 4.5 h growth in DMEM, 1 ml of each culture was pelleted and resuspended in 75 µl 2 × sample buffer [125 mM Tris (pH 6.8), 20% glycerol, 10% β-mercaptoethanol, 6% sodium dodecyl sulfate, 0.2% bromophenol blue]. Electrophoresis and Western blotting to detect EspB and DnaK were performed as previously described (Deng et al., 2004). Quantification of proteins in Western blots was performed using the Image Lab software on a ChemiDoc gel imager (Bio-Rad).

Strain and plasmid construction
All C. rodentium deletion mutants were generated by allelic exchange. Briefly, in-frame deletion constructs for each gene were generated by overlap-extension PCR (Ho et al., 1989) using the UpF-UpR and DnF-DnR primers listed in Table S2. Overlap PCR products were restriction digested and ligated into pUC18. All inserts were confirmed by Sanger sequencing, then subcloned into suicide vector pRE112 (Edwards et al., 1998). Suicide plasmids were transferred into C. rodentium by biparental mating using MFDPir as the donor (Ferrières et al., 2010) with transconjugants selected on LB chloramphenicol plates. Loss of the pRE112 plasmid from the C. rodentium chromosome was subsequently selected for by growth on LB agar with 5% sucrose. Sucrose-resistant, chloramphenicol-sensitive colonies were screened for presence of the intended deletion by PCR.
Single-copy, chromosomally complemented strains were constructed using the mini-Tn7 system (Choi et al., 2005). Briefly, the chloramphenicol resistance cassette from pKD3 (Datsenko and Wanner, 2000) was amplified using primers P1_cat and P2_cat (Table S2) and cloned into the mini-Tn7 transposon in pUC18R6KT-mini-Tn7T (Choi et al., 2005). Genes to be complemented, driven by their native promoters, were amplified using primers listed in Table S2 and cloned into the KpnI and XhoI sites in pUC18R6KT-mini-Tn7T-Cm. The ΔcpxA mutant was complemented using cpxRA since cpxA expression is driven by the promoter upstream of cpxR. Transposon-containing plasmids were transferred into target strains by triparental mating using MFDPir as the donor strain and the pTNS2 transposase-encoding plasmid (Choi et al., 2005). Insertion of the mini-Tn7 element into the correct chromosomal location was verified by PCR using the primers PTn7R and PglmS-down_Citro (Table S2).
In order to construct DBS100 P degPΔCpxR , P rdoAmut , and the related lux reporters pNLP10P degPΔCpxR and pJW15P rdoAmut , we used overlap-extension PCR (with primers listed in Table  S2) to generate PCR products containing the desired mutation with ~1 kb of flanking DNA on each side. The overlap PCR products were then cloned into pUC18 and sequenced. To generate chromosomal mutations, the inserts were subcloned into pRE112 and transferred into C. rodentium as described above. To generate lux reporters, the mutated promoter regions were amplified from the pUC18 plasmids using primers PdegPF/PdegPR and PrdoAF/PrdoAR (Table  S2); PCR products were then restriction digested and ligated into pNLP10 and pJW15 respectively.

T3S assay
Proteins secreted by C. rodentium during growth in DMEM were precipitated using trichloroacetic acid (TCA) and analyzed by SDS-PAGE and Coomassie staining as previously described . Two micrograms of purified bovine serum albumin (BSA) were added to each collected supernatant prior to addition of TCA to aid in protein precipitation.

Mouse infections
All animal experiments were performed in accordance with the guidelines of the Canadian Council on Animal Care and the University of British Columbia (UBC) Animal Care Committee (certificate A12-0238). Mice were ordered from Jackson Laboratory (Bar Harbor, ME) and maintained in a specific pathogen-free facility at UBC. Seven-week-old female C3H/HeJ mice were orally gavaged with 100 µl of overnight culture of C. rodentium grown in LB (containing ~3 × 10 8 CFU of bacteria as confirmed by retrospective plating). Mice were monitored daily for weight loss and clinical symptoms. C. rodentium shedding was monitored by plating dilutions of fecal samples on MacConkey agar every two days throughout the 30-day infection. Upon reaching the humane endpoint (weight loss of 20%; or any one of: bloody diarrhea, severe hunching, severe piloerection, slow or no response to stimuli, labored breathing, or rectal prolapse; or any three of: moderate hunching, moderate piloerection, some lethargy, some change in breathing rate, effort or pattern), mice were euthanized by isoflurane anesthesia followed by carbon dioxide inhalation.

Luminescence assay
Strains harboring lux reporters were cultured in triplicate overnight in LB with kanamycin, then subcultured 1:100 into the same medium. After 3 h growth at 37°C with aeration, 100 µl of each culture was transferred to a black/clear bottom 96-well plate and dithiothreitol was added where indicated. Luminescence and OD 600 were measured every 5 min for 2 h using a Tecan Infinite 200 plate reader. Normalized luminescence was calculated by dividing raw luminescence by the OD 600 of the same well.