DegP Chaperone Suppresses Toxic Inner Membrane Translocation Intermediates

The periplasm of Gram-negative bacteria includes a variety of molecular chaperones that shepherd the folding and targeting of secreted proteins. A central player of this quality control network is DegP, a protease also suggested to have a chaperone function. We serendipitously discovered that production of the Bordetella pertussis autotransporter virulence protein pertactin is lethal in Escherichia coli ΔdegP strains. We investigated specific contributions of DegP to secretion of pertactin as a model system to test the functions of DegP in vivo. The DegP chaperone activity was sufficient to restore growth during pertactin production. This chaperone dependency could be relieved by changing the pertactin signal sequence: an E. coli signal sequence leading to co-translational inner membrane (IM) translocation was sufficient to suppress lethality in the absence of DegP, whereas an E. coli post-translational signal sequence was sufficient to recapitulate the lethal phenotype. These results identify a novel connection between the DegP chaperone and the mechanism used to translocate a protein across the IM. Lethality coincided with loss of periplasmic proteins, soluble σE, and proteins regulated by this essential stress response. These results suggest post-translational IM translocation can lead to the formation of toxic periplasmic folding intermediates, which DegP can suppress.


Introduction
Although all bacterial proteins are synthesized in the cytoplasm, many are subsequently secreted to the cell surface, where they function to sense and respond to the cellular environment. In Escherichia coli, approximately 10% of proteins contain an N-terminal signal sequence for translocation across the inner membrane (IM) to the periplasm [1]. Most bacterial secreted proteins, including those used for virulence, enter the periplasm through the primary export channel SecYEG [2][3][4], either co-or post-translationally. While the rate of co-translational translocation is limited by the rate of protein synthesis [3,4], post-translational described previously as pPERPLC01; the protein product accumulates as inclusion bodies in the E. coli cytosol [39]. Plasmid pCEFN1 was used to express pet under the control of its endogenous promoter [40]. All plasmids were constructed using standard molecular biology techniques. All primers were purchased from IDT. E. coli DH5α (Invitrogen) was used for cloning procedures.
Plasmid pPRN-ELP, to express prn from a constitutive tac promoter, was generated from its parents pP.93WT [29] and pPERPLC01 [39]. Three point mutations present in pP.93WT were removed by ligating a portion of the gene encoding the passenger region from pPERPLC01 into pP.93WT. To do this, both plasmids were digested using at the unique KpnI and RsrII sites and the encompassing region from pPERPLC01 with the correct sequence was ligated in pP.93WT. As a result, prn differs from the gene encoded by pP.93WT in three positions (resulting in three amino acid substitutions: D262E, P264L, S332P). All prn gene variants used in this study were in this background (D262E, P264L, S332P).
Construct T5p-prn was made by cloning the prn sequence in pQE-2 (Qiagen), resulting in plasmid pQE-2-PRN. First, a BamHI site was introduced in pQE-2 by site-directed mutagenesis using primers pQE2 BamHI SDM FW and pQE2 BamHI SDM RV (S1 Fig), resulting in the plasmid pQE-2-BamHI. The BamHI site was located downstream of the region encoding an N-terminal His-tag in pQE-2 to ensure that the pertactin gene would not have a His-tag. The coding sequence of pertactin was amplified from prn using primers PCR amp P. 93 BamHI FW and PCR amp P. 93 HindIII RV (S1 Fig) and ligated in pQE-2-BamHI via the BamHI and Hin-dIII sites, resulting in plasmid pQE-2-PRN.

Growth curves
All cultures were grown at 30°C. To compare growth of different strains in parallel, cultures were first normalized during inoculation from overnight cultures as follows: The overnight culture with the highest optical density (OD 600 ) was inoculated at a 1:50 dilution and the remaining cultures were normalized to have the same starting OD 600 . After growth to log phase (~OD 600 = 0.4), fresh cultures were once again started to ensure that all cultures started at a similar growth phase. For the second inoculation, the equivalent of a 1:50 dilution from a culture corresponding to OD 600 = 2.5 was used. The growth curve after this second inoculation was recorded.

Cell viability assay
To quantify the phenotypic effects of ATs on E. coli growth in different genetic backgrounds, the viability of E. coli (KS303 and KS476) was quantified after plasmid transformations. In each transformation reaction, 1 μL of plasmid DNA from a 50 ng μL -1 stock was added to 50 μL of chemically competent E. coli. In some cases, a plasmid encoding a variant of degP (degP S210A) expressed from a trc promoter (pCS10) [16] was co-transformed. The gene product from this degP variant is a protease deficient version of DegP. After heat shock transformation for 45 s at 42°C, 200 μL of SOC medium (2% tryptone, 0.5% yeast extract, 10 mM NaCl, 2.5 mM KCl, 10 mM MgCl 2 , 10 mM MgSO 4 , 20 mM glucose) was added and cells were allowed to recover for 1 h at 30°C. Transformation reactions were diluted in SOC medium and plated on LB-agar supplemented with appropriate antibiotics (100 μg mL -1 ampicillin (Amp) for single transformations with plasmids encoding AT variants, plus 100 μg mL -1 and 35 μg mL -1 chloramphenicol (Cam) for co-transformations with pCS10).
Mass spectrometry E. coli KS303 or E. coli KS476 transformed with pQE-2-PRN ( Fig 1B) were grown at 30°C. For each sample, 200 mL LB medium supplemented with 100 μg mL -1 Amp was inoculated with an overnight culture at a 1:50 dilution. Expression of prn was induced with 0.5 mM IPTG at the indicated OD 600 , and cells were pelleted at the indicated times. Cell pellets were lysed using a bead beater (BioSpec) in 50 mM Tris pH 8.0, 1% TritonX-100, 150 mM KCl, 1mM EDTA, 1mM PMSF, and clarified by centrifugation at 12,000 x g for 12 min.
Total protein were quantified using Micro BCA (Pierce) according to the manufacturer's instructions, and 100 μg of each lysate was precipitated and digested with trypsin using the FASP protocol as described previously [44][45][46]. All reagents were purchased from Sigma unless otherwise indicated. Briefly, Tris(2-carboxyethyl)phosphine-reduced and iodoacetamide alkylated samples were buffer-exchanged using filtration concentrators (Amicon; 10,000 MW cutoff) with five washes of 8 M urea in 100 mM Tris pH 8.0, then exchanged with two washes of 50 mM ammonium bicarbonate pH 8.0. One μg of trypsin (Promega) was added to each sample and digested at 37°C for 8 h. Peptides were recovered from the filtration concentrator by spinning and a second spin with addition of 100 μl of 500 mM NaCl and quenched with addition of 5% trifluoroacetic acid. Peptide mixtures were desalted using C18 SPE spin columns (Protea Sciences), dried and resuspended in 0.2% formic acid for LC/MS/MS analysis.

Proteomic data analysis
MS/MS files were searched using the Andromeda search engine [47] against the current E. coli FASTA database supplemented with laboratory contaminants and the sequence of pertactin (B. pertussis) produced in these strains. Deamidation of NQ, oxidation of M, carboxyamidomethyl-C, and N-terminal acetylation were considered as modifications. The peptide-spectral matches false-discovery rate (FDR) was set to 0.01%. Protein FDR was set to 0.02% using target-decoy false discovery calculation [48].
Protein quantification was performed using label-free-quantification based on extracted ion chromatograms. MaxLFQ within MaxQuant v 1.5.2.8 software was used for peak integration and normalization [49]. Suggested settings were used for integration. Two or greater quantified peptides per protein were required and normalization against total protein and corrected mass spectrometry response was performed. Deletion of DegP was readily verified within these quantifications (panel A in S2 Fig). Quantitative data is given as the log2 transformed fold-change versus the E. coli KS303 strain after 2 h of pertactin induction. Error bars are plotted as standard error (∑/mean) for each label-free measurement. RAW data files, search results, search parameters and databases were deposited into MassIVE and are available for download at massive.ucsd. edu with the identifier MSV000080095 or ftp://MSV000080095@massive.ucsd.edu.
To estimate abundance of periplasmic proteins, the 1000 most abundant proteins across all conditions were analyzed further. Out of this set, 39 proteins have 'periplasm' in their name. These proteins were used as a proxy for all periplasmic proteins. The quantity of these periplasmic proteins during death (5 h post induction, KS476) was compared to the control condition (2 h time point, KS303, no induction). Out of all 39 periplasmic proteins identified in this way, six did not change significantly or increased and one had ambiguous quantification. The remaining 32 proteins (82% of total) decreased by at least two-fold during death.

Measuring accumulation and secretion of AT variants
Accumulation and secretion of AT variants in vivo was verified by measuring the appearance and proteolytic processing of the AT precursor in cell lysates using western blotting, as described previously [24,25,29,32,50]. E. coli were transformed with plasmids encoding the indicated genes and grown overnight. The equivalent of 1 mL of cells at OD 600 = 0.5 were pelleted for 5 min at 8,000 x g and resuspended in 50 μL SDS loading buffer. For cells where variants of the Pet passenger were produced, a sample of the growth medium was collected to quantify the secreted passenger. After removing bacteria by centrifugation, the spent growth medium was filtered through a 0.22 μm filter and diluted 15-fold with SDS loading buffer. WCL and spent media samples were boiled for 5 min, separated by SDS-PAGE and analyzed by western blotting as described below.

Western blotting
Samples for western blot analysis were separated by SDS-PAGE and transferred to a polyvinylidene fluoride (PVDF) (Bio-Rad) membrane as follows. The membrane was wet in methanol, and then both the membrane and the SDS-PAGE gel were equilibrated in western transfer buffer (WTB, 25 mM Tris base, 192 mM glycine, 20% (v/v) methanol) for 10-15 min. The gel was transferred to the PVDF membrane at 4°C for 2 h at 70 V. After transfer, the membrane was equilibrated for 5 min in 20 mL Tris buffered saline Tween-20 (TBST) consisting of 20 mM Tris HCl pH 7.5, 150 mM NaCl and 0.5% (v/v) Tween-20 supplemented with 5% (w/v) powdered dry milk (PDM). The primary antibody was added to the TBST/5% PDM solution and incubated while shaking. The membrane was washed with 20 mL TBST three times to remove unbound antibody followed by incubation with the secondary antibody in 20 mL TBST/5% PDM. Then, the membrane was incubated for 30 min with an alkaline phosphatase (AP)-conjugated secondary antibody and washed three times with 20 mL TBST to remove unbound antibody. The membrane was rinsed with ddH 2 O five times and developed by incubation with 10 mL alkaline phosphatase (AP) buffer (100 mM Tris HCl, pH 9.5, 100 mM NaCl, 5 mM MgCl 2 ) containing 66 μL of nitro blue tetrazolium and 33 μL of 5-bromo-4-chloro-3-indoyl-phosphate (Promega). The blot was rinsed with water to stop the reaction.

Constitutive expression of prn, but not pet, is lethal in the absence of DegP
To systematically investigate the effects of deleting degP during AT expression, we compared E. coli growth during the synthesis of two dissimilar ATs, Bordetella pertussis pertactin (gene name: prn) [24,29,39] and Pet, an AT from the serine protease autotransporters of Enterobacteriaceae (SPATE) subfamily [24,40,52] (Fig 1B). Surprisingly, expression of prn in the absence of DegP was lethal: transformation of a degP null strain with a plasmid to constitutively express prn yielded no colonies (Fig 2A). Likewise, induction of prn expression in the degP deletion background resulted in a dramatic decrease in cell density (OD 600 ; Fig 2B), a characteristic phenotype of cell death. Such conditional synthetic lethality has not previously been observed during AT expression in the absence of DegP. Past studies have indicated DegP deletion has no significant effect on E. coli growth, with or without AT expression [38]. Consistent with these studies, we observed that transforming a plasmid to constitutively express pet did not affect viability of E. coli in the absence of DegP (Fig 2A).
The DegP chaperone function is required for E. coli survival during pertactin production and secretion To dissect the specific contributions of the chaperone vs. protease function of DegP to the observed lethality, we first determined whether the DegP chaperone activity alone was sufficient to rescue the lethal phenotype. We transformed a plasmid to constitutively express prn in the E. coli degP deletion strain as before, but also co-transformed a plasmid encoding a protease-deficient version of DegP that retains the DegP chaperone activity (DegP S210A) [16]. Supplying the DegP chaperone function was sufficient to restore E. coli growth during prn expression ( Fig 3A).
We next asked whether supplying the DegP chaperone rescues pertactin secretion, in addition to enabling E. coli growth. We and others have previously shown that pertactin processing is a reliable assay for secretion across the OM; secretion-impaired mutants of ATs remain unprocessed in the periplasm [24,25,28,29,53,54]. We observed that the processed pertactin passenger accumulated to similar levels when the DegP chaperone variant was produced ( Fig  3B), demonstrating that the DegP chaperone is sufficient to restore growth and allow pertactin secretion in E. coli.
Next, we tested the role of the DegP protease during AT secretion in our model systems. We asked whether the pertactin precursor is also a substrate for the DegP protease while critically requiring the DegP chaperone. We constitutively expressed prn in E. coli strains with or without degP on the chromosome. To avoid the lethal phenotype described above, we also supplied the DegP chaperone activity by producing the protease deficient DegP variant from a plasmid in both strains ( Fig 3B). We observed that the processed pertactin passenger accumulated to similar levels with or without the DegP protease activity (Fig 3B), indicating that the DegP protease activity was not crucial for pertactin secretion or cell viability. However, the pertactin precursor accumulated to higher levels when the protease activity of DegP was omitted (Fig 3B,  compare lanes 1 and 2), indicating that DegP degraded a portion of the pertactin precursor pool. We tested the DegP protease activity during expression of pet as well ( Fig 3C). As seen for pertactin, absence of the DegP protease resulted in an increase in the Pet precursor pool, but no detectable changes for passenger secretion efficiency. Taken together, these results demonstrate that DegP, when present, degrades a portion of both the pertactin and Pet precursors. This result is consistent with previous reports of DegP degradation of AT secretion intermediates in the periplasm [38,54,55].
We next tested whether prn expression was dependent on other E. coli chaperones besides DegP. To do this, we turned to an existing series of E. coli chaperone deletion strains [43]. We transformed the degP deletion strain from this series (JB8) with a plasmid constitutively expressing prn and confirmed that, as for the degP deletion strain KS476, transformation with These results demonstrate that despite the known overlapping substrate specificity amongst periplasmic chaperones [13], lethality upon prn expression in E. coli is related to a specific requirement for the DegP chaperone activity.

Lethality in the absence of DegP occurs prior to pertactin OM translocation
Previous reports suggested that DegP contributes a quality control function during AT OM translocation [37,56,57]. To test whether the lethality of prn expression in the absence of DegP To test if the DegP chaperone alone is sufficient to restore viability, a protease deficient version of DegP [16] was produced by co-transforming a plasmid expressing degP S210A. (B) Expression of prn results in production of the precursor and processing of the passenger (samples shown here were collected from cells described in (A). The role of the DegP protease for pertactin accumulation and secretion was assessed by expressing prn when only the DegP chaperone was present, versus the DegP chaperone plus the chromosomal wild type DegP. In the presence of the DegP protease activity, less pertactin precursor accumulates (compare lanes 1 vs. 2). (C) Wild type pet was expressed overnight in E. coli with or without a chromosomal copy of degP (same samples as grey bars in Fig 2A). In the absence of DegP, more full-length precursor accumulates in whole cell lysates (WCL). The yield of secreted Pet in the media after overnight production was not significantly altered with or without DegP. (D) After transforming prnΔβ-barrel in the E. coli degP null strain, very small colonies appeared after prolonged incubation (black triangle). (A, D) Error bars are standard deviations from at least three biological repetitions. is related to pertactin OM translocation, versus an earlier step in secretion, we constructed prnΔβ-barrel (Fig 1B). Because the product of this construct lacks the C-terminal OM β-barrel, it will neither integrate into the OM nor will the passenger translocate across the OM [58]. If lethality was related to pertactin β-barrel OM assembly or OM translocation, deleting this domain should reverse the lethal phenotype. However, expression of prnΔβ-barrel in the degP deletion strain resulted in a severely toxic phenotype (Fig 3D), yielding only extremely small colonies after prolonged incubation (see Table 1 for a comparison of relevant phenotypes). The severe growth defect of prnΔβ-barrel in the absence of DegP indicates that DegP plays a crucial role in early steps of pertactin secretion, prior to OM translocation.
Fusion of pertactin to a co-translational signal sequence is sufficient to suppress DegP chaperone dependency We next sought to identify what features of pertactin, but not Pet, trigger lethality in the degP deletion background during early secretion steps. A notable difference lies within the signal sequences: the Pet signal sequence has an N-terminal extension specific to the SPATE sub-family of AT proteins [59,60]. Although the precise function of this N-terminal signal sequence extension remains unclear, it has been proposed to slow down the rate of IM translocation [60][61][62][63][64][65].
To systematically test the effects of altering the IM translocation mechanism on DegP-dependent prn expression, we replaced the pertactin signal sequence with a well-characterized signal sequence from dsbA or phoA, to generate dsbA(ss):prn and phoA(ss):prn (Fig 1B). DsbA is known to be translocated co-translationally across the IM [66], whereas PhoA IM translocation is mainly post-translational [5,67]. Both of these E. coli signal sequences are routinely used to target proteins to the periplasm [67][68][69]. As expected, neither of these signal sequences altered pertactin production or secretion in the parent E. coli strain (Fig 4A). Likewise, we observed no significant changes in growth rates (Fig 4B). However, we found that replacing the pertactin signal sequence with the co-translational signal sequence of dsbA was sufficient to bypass the lethality of expressing prn in the degP deletion strain and restore pertactin secretion (Fig 4). In contrast, fusion to the PhoA signal sequence was lethal (Fig 4C). The lethality of phoA(ss):prn in in the degP deletion background indicates that post-translational translocation of pertactin across the IM is sufficient to recapitulate the conditional synthetic lethality observed for pertactin in the absence of DegP. Importantly, both dsbA and phoA are endogenous to E. coli. Recapitulation of both the lethal and viable phenotypes when these signal sequences were fused to pertactin confirms that the phenotype is directly linked to co-vs. post-translational IM translocation in E. coli.
The extended Pet signal sequence can partially rescue lethality of pertactin in the absence of degP The dsbA(ss):prn and phoA(ss):prn results indicate that the nature of the signal sequence or the resulting differences in IM translocation mechanism largely determined the DegP dependence, leading us to closely examine the differences in the Pet and pertactin signal sequences. Our current understanding of AT IM targeting is based on studies of the SPATE sub-family of AT proteins, all of which share the N-terminal signal sequence extension described above [59,60]. A survey of the literature revealed that DegP is dispensable during production of ATs with extended signal sequences, including IcsA [56,57,70], Hbp [34,38,54,55,64], EspP, Pet and other SPATEs [37], as well as the two-partner secretion (TPS) protein FHA [71,72]. Furthermore, the SPATE signal sequence extension appears to reduce the rate of post-translational IM translocation and/or release of the AT from the IM [60][61][62][63][64][65]. We hypothesized that, like the cotranslational DsbA signal sequence, the extended SPATE signal sequence could similarly reduce DegP dependence.
To test whether the extended signal sequence from Pet can rescue lethality in pertactin, we replaced the region of prn encoding the signal sequence with the region encoding the Pet signal sequence (resulting in construct pet(ss):prn; Fig 1B) and transformed the plasmid expressing this chimera into the degP deletion strain to assess viability. We observed small colonies after prolonged incubation, but these did not appear within the timeframe required to qualify as CFUs in our assay (Fig 5A). Note that, in contrast, we did not detect any colonies, even very small ones, when prn was transformed into the degP deletion strain (Fig 2A). We hypothesized that the colonies resulting from transformations with the plasmid harboring pet(ss):prn might be viable, albeit with a strong growth defect. We tested growth of these small colonies in liquid culture. As a control, we transformed the E. coli parent strain with the plasmid to express pet(ss): prn and no growth defect was observed in the E. coli parent (Fig 5B). As hypothesized, we were able to grow the degP deletion strain transformed with a plasmid to express pet(ss):prn in culture (Fig 5C), albeit at a slow growth rate (see also Table 1). These results suggest that fusing the Pet signal sequence to pertactin improves viability of the degP deletion strain, albeit to a limited extent. To test directly if pertactin secretion was also rescued in this chimera context, we measured pertactin accumulation and secretion as described above. Indeed, mature pertactin accumulated in the degP deletion strain, although the secretion yield for the chimera was reduced (Fig 5D). The overall slower growth rate (Fig 5C) likely contributes to this reduced secretion yield. Taken together, these results indicate that the extended Pet signal sequence can partially rescue the lethal phenotype of pertactin in E. coli, although significant toxicity remains.
We next tested the reverse hypothesis: whether fusion of the pertactin signal sequence to Pet was sufficient to induce toxicity in the degP null strain. We transformed a plasmid encoding prn(ss):pet (Fig 1B) in the degP deletion strain and observed cell growth similar to cells transformed with wild type pet (Fig 5A and 5C). In contrast to wild type Pet, there was a dramatic reduction of Pet precursor accumulation in cell lysates and mature Pet passenger in the media (Fig 5E). A similar reduction was also observed when the chimera was produced in the E. coli parent strain. Taken together, these results indicate that E. coli growth is not DegP-sensitive when Pet is produced with a pertactin signal sequence. Reduced accumulation of the chimera may permit viability.

Loss of periplasmic proteins during prn-induced cell death
The results above demonstrate that in the absence of DegP, post-translational IM-translocation of pertactin is lethal. This finding was unexpected: There is no precedent in the literature for a link between DegP dependence, the IM translocation mechanism and cell death. To broadly explore the underlying cell death mechanism, we used quantitative mass spectrometry to measure changes in protein abundance across the proteome upon induction of this lethal phenotype. Defects in growth upon induction of prn in the degP deletion strain were first detectable 2 h post-induction, indicating that this represents an early stage of cell death (Fig 2B). At 5 h post induction, the OD 600 had decreased significantly, indicating lethality. To capture both early and advanced stages of cell death, we analyzed changes to the proteome at both 2 h and 5 h after prn induction. Cells were grown as described previously (Fig 2B), and we collected samples for proteomic analysis at the chosen time points (Fig 6A and 6B). We compared prn induction in the degP deletion strain versus its parent strain and also included a control condition where prn expression was not induced in the degP deletion strain. Together, we collected samples in these three different conditions at 2 h and 5 h post induction, resulting in six samples overall (see Fig 6A and 6B for details).
Quantification of the entire proteome across our experimental conditions was performed using normalized label-free-quantification (LFQ), where measured peak areas from peptides specific to each protein were integrated, averaged and normalized [45,49]. Overall, 1897 proteins were quantitatively described across all mass spectrometry runs, representing approximately 50% of the total E. coli proteome (see Fig 6C and Methods for details). Our results above indicated that lethality was caused by a periplasmic defect, motivating us to investigate changes to the periplasmic proteome during cell death. Of the most abundant periplasmic proteins across our experimental conditions, 82% (32 proteins) were 2-fold less abundant (or absent entirely) in the degP deletion strain at 5 h post prn induction. Representative examples are shown in Fig 6D. The striking reduction in the abundance of most periplasmic proteins suggests severe stress within the cell envelope. In contrast, we detected no loss of outer membrane porins (including OmpC, LamB and BamA) during prn expression in the absence of degP, and similarly stable levels for cytosolic markers including σ 70 and DnaK

Collapse of the σ E stress response during prn-induced cell death
The loss of periplasmic proteins as cells approached death led us to hypothesize that dysfunction of one or both of the two major periplasmic stress responses, σ E and/or Cpx [73], might trigger lethality. We first investigated the σ E stress response, in which the alternative sigma factor σ E regulates expression of downstream genes [74][75][76][77]. Surprisingly, the σ E protein itself was undetectable in the degP deletion strain 5 h after the induction of prn, but was detected in all other control conditions (Fig 6E). From this we conclude that the σ E pathway is inactivated during death. Loss of σ E could arise due to its degradation and/or reduced production. Alternatively, the σ E protein could be sequestered at the IM where it becomes more difficult to detect using mass spectrometry. Indeed, when the σ E pathway is inactivated the σ E protein is sequestered at the IM [78]. Either of these scenarios is consistent with our data and would result in inactivation of the σ E pathway. Since cytosolic σ E is an essential component of E. coli [79], its absence is sufficient to explain the lethality of prn expression in the absence of DegP [80].
To test further for the activity of the σ E response in our experimental conditions, we quantified 25 σ E -dependent gene products [74] detected in our dataset. We found that at least 13 σ Edependent gene products are less abundant (2-fold) or undetectable in the degP deletion strain upon induction of prn (representative examples are shown in Fig 6E). These results indicate that collapse of the σ E response contributes to lethality when pertactin accumulates in the periplasm in the absence of degP.
Next we investigated the Cpx pathway, which leads to degP upregulation upon cell envelope stress [81]. Although we identified 26 of the known Cpx-regulated proteins [82] in our proteomics dataset, most were of low abundance and therefore unsuitable for reliable quantification. Nevertheless, six Cpx substrates could be quantified to high reliability (panel C in S2 Fig).
Although we observed some changes in these Cpx substrates during induction of prn (panel C in S2 Fig), these changes were less severe than the complete abrogation of σ E and some of its substrates. Although we cannot rule out contributions from Cpx towards lethality, the observed changes in Cpx alone are insufficient to explain the lethal phenotype in E. coli. In contrast, the complete loss of σ E , which is essential, and loss of several σ E -dependent essential genes is sufficient to explain lethality.

Discussion
It has previously been suggested that periplasmic chaperones participate in late steps of AT secretion, ensuring that the AT passenger remains in a high-energy, secretion competent conformation required for efficient transfer to BAM or TAM [25,31,53,[83][84][85][86]. In particular, previous studies revealed a variety of effects that hint at an interplay between DegP and AT secretion, including increased degP expression upon AT expression [34,38,55], defects in cell growth and AT secretion in degP null strains [37,56,57] and binding of DegP to an AT passenger in vitro [37]. However, the results presented here represent the first identification of a lethal phenotype during AT expression in a degP deletion background. Moreover, to our knowledge this is the first observation of a link between the function of a periplasmic chaperone and the IM translocation mechanism of a secreted protein in Gram-negative bacteria The mechanistic connection between substrate IM import mechanism and DegP dependency may arise due to differences in IM translocation kinetics. Co-translational IM translocation is reported to be slower than classical post-translational IM translocation [5,8]. Hence changing the IM translocation rate could alter the conformation of the passenger as it enters the periplasm, akin to altered co-translational folding under different translation rates [87][88][89] (Fig 7). We propose a model where fast IM translocation via the PhoA signal sequence results in the rapid appearance of the substrate in the periplasm and subsequent accumulation of toxic conformations. These toxic conformations can be masked or avoided by DegP. Alternatively, reducing the rate of appearance of the passenger in the periplasm can avoid toxicity, for example during the slower, co-translational IM translocation of proteins bearing the DsbA signal sequence (Fig 7). This model predicts that the IM translocation mechanism and DegP work together to maintain E. coli secreted proteins in a non-native, non-toxic unfolded conformation. Consistent with this model, we found that swapping the pertactin N-terminal signal sequence with the co-translational DsbA signal sequence was sufficient to restore E. coli growth in the absence of DegP, whereas the mainly post-translational PhoA signal sequence recapitulated the lethal phenotype. It is important to point out that both the DsbA and PhoA signal sequences are endogenous to E. coli. Hence while we cannot exclude that expressing the heterologous B. pertussis prn AT in E. coli may exacerbate lethality, the observation that these E. coli signal sequences were sufficient to recapitulate or reverse the phenotype demonstrates that the IM translocation mechanism is sufficient to alter the fate of a secreted protein in the E. coli periplasm.
Independent of the AT signal sequence, we found that properties of the AT passenger also contribute to lethality to some extent. Fusing the Pet signal sequence to the pertactin passenger only partially rescued the lethal phenotype, suggesting that the sequence and/or conformational properties of the AT passenger contribute to the viability phenotype (Fig 5). Although the pertactin signal sequence alone was not sufficient to render the Pet AT passenger lethal in the absence of DegP, these findings are ambiguous due to the dramatically reduced level of Pet accumulation (Fig 5E). Taken together, we found that ATs can avoid DegP dependence when at least one of two criteria are met: They are either translocated across the IM co-translationally, and/or have passenger folding properties capable of avoiding conformations that lead to lethality in the absence of DegP (Fig 7).
We characterized the lethality caused by pertactin synthesis in the degP deletion strain in an unbiased manner by measuring broad changes in the proteome. Although many periplasmic proteins were depleted during lethality in the absence of DegP (Fig 6D), we did not detect significant changes in the abundance of cytosolic or OM components (panel B in S2 Fig). Prior to pertactin production, we observed measurable levels of σ E and σ E -induced proteins, which appears constitutive under our growth conditions. In contrast, we observed nearly complete inactivation of the σ E stress response in the degP deletion strain upon prn induction. These results are consistent with lethality propagated from the periplasm, triggered by the periplasmic conformation of the pertactin passenger in the absence of DegP, which leads to collapse of the σ E periplasmic stress response (Figs 6E and 7). Inactivation of σ E could either be cause or consequence of leakage of soluble periplasmic material. The σ E response is induced when DegS binds unfolded substrate proteins [75]. Therefore, its inactivation may start by release of DegS from its substrates, perhaps due to their dilution during periplasm leakage (Fig 7). In this scenario, unbound and inactive DegS prevents degradation of RseA, resulting in sequestration of the σ E protein at the IM to RseA [75,80]. Membrane sequestration of the σ E protein would also make it less likely to be detected by mass spectrometry, which could also lead to its absence in our proteomics assay. Consistent with this model, we also did not detect significant levels of other IM components of the σ E machinery (DegS, RseA, RseP).
We were initially surprised that outer membrane proteins were largely unaffected by σ E loss, since BamA and other OMPs are generally positively and negatively regulated by σ E [74] and OMP levels overall were stable even after 5 h of prn expression. However, our results mirror those of Ades and coworkers, who found that an induced suppressor of σ E function produced an analogous lethal phenotype with no apparent defects in outer membrane proteins [80]. Presumably, the unusually high stability of native OMPs [31,90] reduces their turnover rate, making them less susceptible to σ E loss on our experimental time scale. Together, our unbiased characterization of changes across the proteome as cells approach death provides a unique view on time-resolved, global responses from the E. coli proteome when the periplasmic chaperone network collapses.
It is important to note that because the lethal DegP-dependent phenotype reported here may arise due to high-level production of a heterologous AT protein in E. coli, these results may have important implications for exploiting the AT secretion mechanism to produce heterologous proteins via autodisplay [91][92][93]. Furthermore, some ATs, including pet, are naturally encoded on plasmids. Differential reliance upon DegP could therefore prevent a Gram-negative pathogen from acquiring a plasmid via horizontal gene transfer. It will be important to investigate whether other Gram-negative species have a similar dependence upon DegP.  [43] were transformed with plasmids that constitutively express prn or pet, as described in Fig 2A. Error bars in bold are standard deviations from at least three biological replicates. The vertical lines for pet transformed degP::kan indicates the spread of two biological replicates. (B) prn is expressed and the passenger is secreted in fkpA:: cm and skp::kan. As a control the parent strain (BW30270) was used. Secretion is routinely assayed by the production of the processed passenger in whole cell lysate samples. (TIF)