Cyclic Enterobacterial Common Antigen Maintains the Outer Membrane Permeability Barrier of Escherichia coli in a Manner Controlled by YhdP

ABSTRACT Gram-negative bacteria have an outer membrane (OM) impermeable to many toxic compounds that can be further strengthened during stress. In Enterobacteriaceae, the envelope contains enterobacterial common antigen (ECA), a carbohydrate-derived moiety conserved throughout Enterobacteriaceae, the function of which is poorly understood. Previously, we identified several genes in Escherichia coli K-12 responsible for an RpoS-dependent decrease in envelope permeability during carbon-limited stationary phase. For one of these, yhdP, a gene of unknown function, deletion causes high levels of both vancomycin and detergent sensitivity, independent of growth phase. We isolated spontaneous suppressor mutants of yhdP with loss-of-function mutations in the ECA biosynthesis operon. ECA biosynthesis gene deletions suppressed envelope permeability from yhdP deletion independently of envelope stress responses and interactions with other biosynthesis pathways, demonstrating suppression is caused directly by removing ECA. Furthermore, yhdP deletion changed cellular ECA levels and yhdP was found to co-occur phylogenetically with the ECA biosynthesis operon. Cells make three forms of ECA: ECA lipopolysaccharide (LPS), an ECA chain linked to LPS core; ECA phosphatidylglycerol, a surface-exposed ECA chain linked to phosphatidylglycerol; and cyclic ECA, a cyclized soluble ECA molecule found in the periplasm. We determined that the suppression of envelope permeability with yhdP deletion is caused specifically by the loss of cyclic ECA, despite lowered levels of this molecule found with yhdP deletion. Furthermore, removing cyclic ECA from wild-type cells also caused changes to OM permeability. Our data demonstrate cyclic ECA acts to maintain the OM permeability barrier in a manner controlled by YhdP.

T he cellular envelope of Gram-negative bacteria consists of an inner membrane (IM) surrounding the cytoplasm, an asymmetrical outer membrane (OM), and a thin layer of peptidoglycan found in the periplasm separating the two membranes (1). While the inner leaflet of the OM is composed of phospholipids, the outer leaflet is mainly composed of lipopolysaccharide (LPS). LPS possesses a number of negatively charged residues that are bridged by divalent cations to form a strong network of interactions between neighboring LPS molecules (2). Due to these interactions and the amphiphilic nature of LPS, the OM provides the cell a robust permeability barrier, resistant to both large and hydrophobic molecules (3). For this reason, the OM has proven an impediment for the design of new antibiotics to treat Gram-negative bacterial infections.
Enterobacterial common antigen (ECA) is an invariant carbohydrate-derived molecule that is present in the OM and periplasm of members of Enterobacteriaceae (4). Although ECA is restricted to one family of bacteria, four of the seven species identified by the World Health Organization as being of high concern due to frequent antibioticresistant infections are members of this family (Klebsiella pneumoniae, Escherichia coli, nontyphoidal Salmonella, and Shigella species) (5). Despite the conserved nature of this molecule within Enterobacteriaceae (6), its function is largely unknown. In part, this is because the biosynthesis pathways for ECA, O antigen, and peptidoglycan overlap in such a way that gene deletions preventing ECA biosynthesis often also prevent O-antigen production (7)(8)(9) or perturb peptidoglycan biosynthesis, causing envelope stress responses to be activated (i.e., Cpx, Rcs, E ) (10)(11)(12). Thus, in interpreting the results of high-throughput screens (13)(14)(15)(16), it is difficult to determine whether phenotypes are directly related to the presence or absence of ECA or are instead related to changes to other aspects of the cell envelope. Nevertheless, it is thought that ECA plays a small role in bile salt resistance and in organic acid resistance (17,18). It is generally assumed that the surface-exposed forms of ECA are responsible for these phenotypes. In addition, in Salmonella enterica serovar Typhimurium, O-antigen and ECA biosynthesis are not genetically connected, and the first gene in ECA biosynthesis, wecA, can be deleted without activating stress responses, affecting O-antigen biosynthesis, or impairing peptidoglycan biosynthesis (19). Studies in this strain have demonstrated that cells without ECA are deficient in pathogenesis (19), suggesting that ECA plays an important role in the host.
The structure of ECA is conserved throughout Enterobacteriaceae, with each unit of ECA consisting of GlcNAc (N-acetylglucosamine), ManNAcA (N-acetyl-D-mannosaminuronic acid), and Fuc4NAc (4-acetamido-4,6-dideoxy-D-galactose) (20,21). The pathway of ECA biosynthesis is analogous to that of O-antigen biosynthesis (see Fig. S1 in the supplemental material). GlcNAc-1-phosphate is linked to undecaprenyl-phosphate (Und-P), a lipid carrier in the IM also used for the biosynthesis of O antigen, peptidoglycan, and capsule carbohydrates, and then ManNAcA and Fuc4NAc are attached (22,23). Many genes in the ECA biosynthesis operon are responsible for synthesizing these sugars and linking them to Und-P (22,24,25). The ECA unit linked to Und-P is then flipped across the IM by WzxE (26). The ECA chains are polymerized by WzyE (27), and the chain length is controlled by WzzE (28). Three forms of ECA are made from polymerized ECA chains. In the first, LPS-linked ECA (ECA LPS ), the ECA chain is transferred to the core sugar moiety of LPS by WaaL (29), the same gene responsible for attaching O antigen to core, and the molecule is transferred to the cell surface, presumably by the Lpt system. In the second, phosphatidylglycerol-linked ECA (ECA PG ), the ECA chain is attached to phosphoglyceride by a phosphodiester linkage (30) and the molecule is surface exposed through an unknown pathway (31,32). In the third form, cyclic ECA (ECA CYC ), an ECA chain of a precise chain length (4 to 6, depending on species) is cyclized in a reaction dependent on WzzE (33)(34)(35). This molecule remains in the periplasm (34).
Previously, we investigated changes to the Escherichia coli K-12 OM that occur during growth under different nutrient conditions (36) and determined that an RpoSdependent mechanism strengthens the envelope permeability barrier under carbonlimiting conditions in a manner that depends on the presence of the genes for several proteins. Of these, YhdP is a large protein of unknown function that is predicted to be located in the IM with the majority of the protein exposed to the periplasm (37). Unlike our other hits, yhdP deletion (⌬yhdP) has strong phenotypes regardless of growth phase. In fact, in a large-scale study on the effects of gene deletions on chemical sensitivity (38), yhdP scored second highest for SDS EDTA (sodium dodecyl sulfate, ethylenediaminetetraacetic acid) sensitivity as well as in the top 15 hits for vancomycin sensitivity. Despite the high level of envelope permeability caused with deletion of yhdP, the function of YhdP is completely unknown. In addition, yhdP appears to be restricted phylogenetically (39), suggesting that it may play a role that is specific to a subset of species.
Here, we demonstrate that mutations that block ECA biosynthesis restore the envelope permeability barrier of ΔyhdP strains. Furthermore, we demonstrate that yhdP, which phylogenetically co-occurs with ECA biosynthesis genes, directly or indirectly controls ECA levels. We were able to trace the suppression specifically to the removal of ECA CYC and demonstrate that, even in a wild-type background, removing ECA CYC changes the OM permeability barrier. Therefore, ECA CYC plays a role in maintaining the OM permeability barrier, and its activity is regulated by YhdP.

RESULTS
Deletion of yhdP causes OM permeability. Our initial screen identifying yhdP was based on sensitivity to SDS treatment during stationary phase, and we also found that ⌬yhdP caused sensitivity to 2% SDS in actively growing cells (36); therefore, we treated yhdP ϩ and ⌬yhdP cells with SDS and increasing concentrations of EDTA and measured their growth. EDTA disrupts the bridging of LPS molecules by divalent cations, sensitizing the OM to the presence of detergents and allowing for the detection OM defects (40). Both yhdP ϩ and ⌬yhdP cells grew to stationary phase with 0.05% SDS alone; however, a low concentration of EDTA (0.25 mM) caused a large growth defect in ⌬yhdP cells, while causing a minimal effect on yhdP ϩ cells (Fig. 1A). In addition, a higher concentration of EDTA (0.5 mM) completely impaired the growth of ⌬yhdP cells, while still allowing for growth of yhdP ϩ cells. These data suggest that there is a change in outer membrane structure when YhdP is removed, and so we investigated whether ⌬yhdP causes permeability to other toxic agents.
Disk assay results suggested that ⌬yhdP cells might be sensitive to vancomycin. Vancomycin is a glycopeptide antibiotic that targets peptidoglycan biosynthesis, which is commonly used to treat antibiotic-resistant Gram-positive infections but is largely incapable of traversing the Gram-negative OM (41). Permeability of the OM to vancomycin is thought to be caused by "cracks" between patches of phospholipids and LPS (42). We analyzed growth curves of yhdP ϩ and ⌬yhdP cells with increasing dosages of vancomycin. Similarly to ⌬yhdP's SDS EDTA sensitivity, a low dose of vancomycin (50 g/ml) caused lysis of ⌬yhdP cells while not affecting yhdP ϩ growth (Fig. 1B). A higher dose of vancomycin (100 g/ml) completely inhibited growth of ⌬yhdP cells while only minimally affecting yhdP ϩ cells. In fact, the increase in vancomycin sensitivity with ⌬yhdP can also be observed by a lowering of the vancomycin MIC (MIC) for this strain (Fig. S2). Both the increased vancomycin and SDS EDTA sensitivity of ⌬yhdP cells suggest that there is a change in OM structure in this mutant that leads to increased permeability.
Disrupting ECA biosynthesis suppresses envelope permeability in ⌬yhdP strains. As no functions are known for YhdP, we then sought to isolate mutations suppressing strain ⌬yhdP's envelope permeability defects in order to determine in what pathway YhdP might be working. The slight growth we observed late after SDS EDTA and vancomycin treatment ( Fig. 1) suggested that spontaneous suppressor mutants are common within ⌬yhdP strain cultures. Therefore, we plated ⌬yhdP cells on a concentration of vancomycin at which growth of these cells is inhibited but yhdP ϩ cells could grow, and we isolated spontaneous suppressor mutants that were capable of growth on this medium. Then, we conducted a secondary screen of these suppressors to identify those that restored both vancomycin and SDS EDTA resistance. We isolated seven spontaneous suppressors that restored both phenotypes, all of which mapped to the ECA biosynthesis (wec) operon. All seven appeared to be loss-of-function alleles ( Fig. 2A and B).
We then asked whether suppression of the ⌬yhdP strain envelope permeability by disruption of the wec operon is specific to yhdP or is a general mechanism of vancomycin and SDS EDTA resistance. To answer this question, we utilized a deletion allele of bamE, a component of the ␤-barrel assembly machine responsible for folding outer membrane proteins (OMPs) into the outer membrane (43). Removal of this nonessential lipoprotein from the complex leads to a similar level of vancomycin sensitivity as ⌬yhdP and causes slight SDS EDTA sensitivity (44,45). We combined this deletion with a deletion of wecA, which is responsible for the addition of the first sugar in ECA to Und-P (46). Some vancomycin resistance was caused by deletion of wecA alone (Fig. 2C). When combined with a deletion of bamE, wecA deletion caused only minimal suppression of vancomycin sensitivity and worsened the bamE SDS EDTA sensitivity. In contrast, wecA deletion fully suppressed both the vancomycin and SDS EDTA sensitivities of a ⌬yhdP strain. These data demonstrate that disruption of ECA biosynthesis is not a universal suppressor of vancomycin and SDS EDTA sensitivity and suggest that this suppression is specific to the ⌬yhdP strain.
We conducted a transposon mutagenesis screen in an effort to identify additional suppressors of the ⌬yhdP mutant strain envelope permeability. Briefly, we identified vancomycin-resistant clones from a pool of 10,000 transposon mutants in a ⌬yhdP strain, mapped the transposon insertion sites in these mutants, and conducted a secondary screen for SDS EDTA resistance (Fig. S3A). The only mutations we identified that suppressed both the vancomycin and SDS EDTA phenotypes were in the wec operon (Table S1); furthermore, we identified mutations in every gene in the wec operon except those that are redundant with O-antigen biosynthesis genes (rffG, wzxE) (34,47) and wzyE, for which disruption is toxic (34) (Fig. S3B). Given that the wec genes form an operon, it is likely that some of these insertions may be polar. With more than 2ϫ genome coverage, wec operon mutations were the only mutations to suppress both of ⌬yhdP's phenotypes, suggesting that our suppressor screen may be saturated. Loss of ECA is directly responsible for the suppression of the ⌬yhdP strain. Because the ECA biosynthetic pathway interacts with the biosynthesis pathway for peptidoglycan and other extracytoplasmic glycans, disruption of ECA biosynthesis can cause cellular changes that are more wide-ranging than simple removal of ECA. Specifically, O-antigen biosynthesis, peptidoglycan biosynthesis, and ECA biosynthesis all compete for both precursor sugar molecules and for the lipid carrier on which the molecules are assembled, Und-P. Although our strains are O-antigen negative, when intermediate steps in the ECA biosynthesis pathway are disrupted, ECA intermediates (aminoglycans linked to Und-P-P) accumulate and sequester Und-P, stressing the peptidoglycan biosynthesis pathway (10). In contrast, when the first step in ECA biosynthesis (catalyzed by WecA) ( Vancomycin 90 mg/L Cyclic ECA Maintains the Outer Membrane Barrier ® Und-P available for peptidoglycan synthesis is increased. Therefore, we sought to determine whether (i) the phenotypes of ⌬yhdP mutant cells were caused by Und-P stress and (ii) whether relieving Und-P stress suppresses the ⌬yhdP strain. Stress on the pool of Und-P has previously been detected using linkage disruption with a marker linked to mrcB::kan (48). As mrcB (PBP1B) is important for transglycosylation and transpeptidation of peptidoglycan precursors (49), deletion of mrcB in a strain with stress on the pool of available Und-P causes significant toxicity and can be synthetically lethal. This causes disruption of the linkage between the Tn10 marker and the mrcB deletion (i.e., fewer colonies with the Tn10 marker have received the mrcB deletion). However, we detected no linkage disruption with the ⌬yhdP strain in the presence or absence of wecA (Fig. S4A), suggesting that the ⌬yhdP mutant does not cause lipid carrier stress. To relieve possible Und-P stress, we overexpressed uppS, responsible for synthesizing Und-P, and murA, responsible for the first committed step in peptidoglycan synthesis (50,51). Overexpression of these genes has previously been shown to relieve peptidoglycan stress caused by Und-P availability (48,52). Overexpression of these genes had no effect on envelope permeability in a ⌬yhdP background (Fig. 3A). Overexpression of uppP, a gene responsible for recycling Und-P (53), and mrcB, the gene encoding PBP1B, also had no effect on the ⌬yhdP strain's envelope permeability (Fig. S4B). These data show that the phenotypes of the ⌬yhdP deletion mutant are not caused by Und-P stress.
To verify further that the mechanism of ⌬yhdP strain suppression by disruption of the ECA biosynthesis operon was not through effects on Und-P, we asked whether To determine whether the ⌬yhdP strain is suppressed by relieving Und-P stress, EOPs were performed on strains carrying the indicated overexpression constructs. Overexpression of uppS and murA did not suppress the ⌬yhdP strain, demonstrating that the ⌬yhdP strain's phenotypes are not caused by effects of Und-P stress on peptidoglycan. (B) EOPs were performed to determine whether disruptions in ECA biosynthesis that increased Und-P availability for peptidoglycan synthesis (⌬wecA) and that decreased Und-P availability for peptidoglycan synthesis (⌬wecE) both suppress the ⌬yhdP strain phenotypes. Both of these mutations suppressed the ⌬yhdP strain phenotypes to an equal extent, suggesting that suppression is unrelated to Und-P availability. (C) EOPs were performed to determine whether the Cpx response was responsible for suppression of the ⌬yhdP strain's phenotypes by disruptions of ECA biosynthesis. Suppression was observed with wecA deletion, even in the presence of cpxR deletion, demonstrating that the Cpx response is not necessary for suppression. (D) EOPs were performed to determine whether the Rcs response was required for suppression. Suppression was observed in the presence of rcsB deletion, demonstrating that the Rcs response is not necessary for the suppression of the ⌬yhdP strain's phenotypes. EOPs images are representative of three independent experiments. (E) Activity of a E reporter was assayed to determine whether suppression of the ⌬yhdP strain correlated with E activation. Suppression of the ⌬yhdP strain's phenotypes did not correlate with E activation, suggesting this is not the mechanism of suppression. Data shown are the average results of three independent biological replicates Ϯ the SEM. Significance was calculated using the Mann-Whitney test. *, P Ͻ 0.05 compared to the appropriate parent strain (MG1655 or ⌬yhdP).
deletion of the first gene in ECA synthesis, wecA, and deletion of a gene in an intermediate step of ECA synthesis, wecE, would have to the same effects on envelope permeability in a ⌬yhdP strain. Deletions of both wecA and wecE caused slight vancomycin resistance and SDS EDTA sensitivity in a yhdP ϩ background; however, when they were combined in the ⌬yhdP deletion strain, the vancomycin and SDS EDTA resistance were both restored to the level of the ECA mutants alone (Fig. 3B). As both ⌬wecA and ⌬wecE mutant strains fully suppress the envelope permeability defects of a ⌬yhdP strain despite having opposite effects on availability of Und-P and precursors, these data demonstrate that the suppression for the ⌬yhdP strain is not due to modification of the peptidoglycan biosynthesis pathway. Because of peptidoglycan defects and the accumulation of Und-P-linked ECA precursors, disruption of the ECA operon can also activate the Cpx, Rcs, and E stress responses (10,11). In fact, in Serratia marcescens, even disruption of wecA can activate the Rcs response (12). Therefore, we tested whether activation of stress responses was responsible for suppression of the ⌬yhdP strain's envelope permeability by disruption of the ECA biosynthesis operon. The Cpx and Rcs stress responses are nonessential and their activity can be prevented by removal of their response regulators, CpxR and RcsB, respectively (54,55). Disruption of cpxR has no effect on the suppression of the ⌬yhdP strain's vancomycin sensitivity by ⌬wecA, although synthetic SDS EDTA sensitivity in ΔwecA ΔcpxR double mutants prevents assessment of the role of Cpx on SDS EDTA sensitivity caused by yhdP deletion (Fig. 3C). Disruption of rcsB has no effect on the suppression of either the ⌬yhdP strain's vancomycin or SDS EDTA sensitivity by the wecA deletion (Fig. 3D). These data demonstrate that neither the Cpx nor the Rcs stress response is necessary for the suppression of the ⌬yhdP strain's phenotypes by the disruption of ECA biosynthesis. Although the E response is essential in E. coli (56), the activation of the E response can be monitored using reporters linked to E -responsive promoters. One such reporter consists of the micA promoter, driving expression of green fluorescent protein (GFP) (57). Using this reporter, we found that activation of the E response by ECA operon disruptions was not necessary for these disruptions to suppress the ⌬yhdP strain's envelope permeability (Fig. 3E). These data together with the rarity of other suppressing mutations for ⌬yhdP strongly suggest that YhdP is functionally connected with ECA.
The genes for YhdP and ECA occur in the same genomes. As our data suggested that ECA and YhdP may interact and ECA is restricted to Enterobacteriaceae, we examined the phylogenetic distribution of yhdP. We used STRING-DB (39) to search for possible homologues of yhdP and to score the homology of the detected genes. The vast majority of the homologues for yhdP were found to be in Enterobacteriaceae. In fact, homologues of YhdP outside of Enterobacteriaceae are only detected in some other Gammaprotobacteria and some Betaprotobacteria; however, none of the YhdP homologues detected for YhdP outside of Enterobacteriaceae had a higher homology with E. coli K-12 YhdP than a possible homologue in Indian rice (Oryza sativa Indica) (Fig. S5A). As YhdP is part of a family of proteins, the AsmA family (58), the yhdP homologues detected outside Enterobacteriaceae possibly represent other members of the AsmA family.
To examine further the distribution of the ECA biosynthetic genes and yhdP, we used STRING-DB (39) to calculate phylogenetic co-occurrence scores based on genes with homology found across genomes. The three genes within the wec operon whose products form a complex to flip ECA to the outer leaflet of the IM (wzxE), polymerize ECA (wzyE), and control ECA chain length (wzzE) have phylogenetic co-occurrence scores with each other of 0.70 to 0.78 (Fig. S5B). The co-occurrence scores for these genes with yhdP range from 0.40 to 0.76, which are within the range of the cooccurrence scores for pairs of genes within the wec operon (0.15 to 0.78). The highest co-occurrence pair for yhdP was found with wzzE. These data demonstrate that genomes containing the machinery to make ECA also contain yhdP.
Cyclic ECA Maintains the Outer Membrane Barrier ® YhdP changes ECA levels. Given that YhdP and ECA are functionally related, we investigated whether deletion of ⌬yhdP causes changes to ECA abundance or chain length. There is no apparent change in the surface exposure of ECA PG or ECA LPS in the absence of yhdP (Fig. S6). By immunoblotting, we detected ECA LPS and ECA PG and compared the levels and chain length with and without yhdP (Fig. 4B). We observed a range of bands with the lowest molecular weight band likely indicating molecules with one repeat unit of ECA and higher molecular weight bands indicating molecules with more repeat units of ECA. Lanes 2 and 7 show a combination of ECA LPS and ECA PG due to the wild-type genetic background, while lanes 4 and 9 (⌬waaL strains) show ECA PG alone, as ECA LPS cannot be produced in these strains (Fig. 4A). Given the lack of bands with ΔwecA samples where there is no ECA, all bands other than the band designated with an asterisk were taken to be ECA. Due to the use of a polyclonal antibody, the levels of ECA LPS and ECA PG cannot be directly compared.
There was no apparent difference in chain length between yhdP ϩ and ⌬yhdP strains (lanes 2 and 7); however, levels of ECA LPS and ECA PG together (lanes 2 and 7) were higher in a ⌬yhdP strain than in a yhdP ϩ strain. This is also true for ECA PG alone (lanes 4 and 9). Interestingly, this 2-to 3-fold increase in ECA levels only occurred in the presence of wzzE (Fig. 4C). The reason for the large increase in ECA PG levels between the ⌬waaL and ⌬wzzE ⌬waaL strains remains an interesting question for further investigation.
Unlike the lipid-linked forms of ECA, ECA CYC is not charged and cannot be detected by immunoblotting (Fig. 4A). Instead, we utilized a quantitative MALDI-TOF (matrixassisted laser desorption ionization-time of flight) approach to detect and quantitate ECA CYC in purified samples. By examining the m/z ratios of ECA CYC peaks, which are present in wzzE ϩ strains and absent in ⌬wzzE strains (Fig. 4A), we determined that the cyclization and nonstoichiometric acetylation of ECA CYC were not changed in a ⌬yhdP strain (Fig. S7A). Therefore, to quantitate ECA CYC levels, we utilized ⌬wecH strains that do not acetylate ECA and that do not affect ⌬yhdP phenotypes to minimize the number of ECA CYC peaks. We then grew yhdP ϩ strains with a nitrogen source containing 15 N and ⌬yhdP cells with a nitrogen source containing 14 N. This shifted the m/z ratio of the ECA CYC by 12, as ECA CYC contains 12 nitrogen atoms, and allowed the comparison of the yhdP ϩ and ⌬yhdP strains' ECA CYC on the same spectra (Fig. S7B).
To quantitate relative ECA CYC levels, we combined yhdP ϩ cells grown with 15 N with an equal number of ⌬yhdP cells grown with 14 N as one sample before purification of ECA CYC , allowing direct comparison of the peaks generated from each strain. This approach indicated that levels of ECA CYC are decreased in the ⌬yhdP strain (Fig. 4D). Over several biological replicates, we found the decrease in the ⌬yhdP strain to be almost 8-fold (Fig. 4E). The phylogenetic co-occurrence of yhdP with ECA biosynthesis genes and the changes to ECA in the absence of YhdP provides further evidence that YhdP plays a role related to ECA.
YhdP prevents ECA CYC from damaging the OM permeability barrier. Given that the different forms of ECA are present in different cellular compartments and presumably play different roles, we then asked which form of ECA is responsible for the ⌬yhdP strain's phenotypes. We hypothesized that one of the membrane-associated forms of ECA would be responsible for the permeability defects, as these molecules are part of the OM and the levels of these molecules are increased when YhdP is removed. We tested this hypothesis by removing specific forms of ECA and determining whether the ⌬yhdP strain's envelope permeability was suppressed.
We compared wecA deletion, which removes all forms of ECA (46) and suppresses, with waaL deletion, which specifically prevents the formation of ECA LPS (29), and with wzzE deletion, which prevents the formation of ECA CYC but allows the formation of ECA LPS and ECA PG , albeit with random chain length (28,34). Currently, there is no way to remove ECA PG without removing the other forms of ECA. We observed full suppression of the ⌬yhdP strain's envelope permeability with both the wecA and wzzE deletions; however, waaL deletion had no effect on the ⌬yhdP strain's phenotypes (Fig. 5A). (Continued on next page) Cyclic ECA Maintains the Outer Membrane Barrier ® These data demonstrate that ECA LPS does not contribute to the ⌬yhdP strain's phenotypes. The suppression of the ⌬yhdP strain by wzzE deletion did not rely on stress response activation (Fig. S8). WzzE is the chain length regulator for ECA (28) and its removal causes several changes to ECA, some of which can be observed in Fig. 4B (compare lanes 2 and 3 and To eliminate the possibility that suppression in a ⌬wzzE deletion strain occurs through changes in ECA chain length, EOPs were performed at 30°C and at 37°C to assay the ⌬yhdP strain's phenotypes. Immunoblot analysis was used to determine ECA chain length at these temperatures. The model chain length of ECA was found to be four at 30°C and six at 37°C. However, the ⌬yhdP strain has strong phenotypes at both temperatures, suggesting that changes in chain length are not responsible for suppression. All images are representative of three independent experiments. lanes 4 and 5): (i) the amount of ECA LPS and ECA PG is increased; (ii) more short (less than 6 copies) and long (more than 7 copies) ECA chains are made; (iii) no ECA CYC is produced (see Fig. S7). Given that completely removing ECA suppresses the ⌬yhdP strain's phenotypes and yhdP deletion increases levels of ECA LPS and ECA PG , we find it unlikely that the mechanism through which wzzE deletion suppresses the ⌬yhdP strain's phenotypes is through further increasing these levels. To investigate the possibility that increasing or decreasing ECA chain length suppresses the ⌬yhdP strain, we tested the ⌬yhdP strain's envelope permeability and suppression at 30°C, where the modal chain length of ECA appears to be four copies; at 37°C the modal chain length of ECA is six to seven copies (Fig. 5B). Although intrinsic sensitivities to SDS EDTA and vancomycin differ between these temperatures, neither temperature suppresses the ⌬yhdP strain's phenotypes, nor does the suppression of the ⌬yhdP strain occur by preventing ECA synthesis change. Thus, despite the fact that yhdP deletion lowered ECA CYC levels, preventing ECA CYC synthesis suppressed the envelope permeability of ⌬yhdP strains, demonstrating a specific, functional interaction between YhdP and ECA CYC . ECA CYC maintains the OM permeability barrier. As the presence of ECA CYC leads to damage to the OM permeability barrier in the absence of YhdP, we then asked whether, in wild-type cells, ECA CYC plays a broader role in maintaining the OM permeability barrier. With high-throughput studies, the specificity of effects caused by insertions in the wec operon can be unclear, as these mutations, including those in wzzE, can be polar, causing loss of all ECA species (28). Thus, we investigated changes to envelope permeability with clean wzzE deletion. Cells with ⌬wzzE showed a level of vancomycin resistance that was higher than that in wzzE ϩ cells and was, in fact, equal to that of ⌬wecA cells (Fig. 6). The vancomycin resistance was similar to that observed with deletion of mlaA. MlaA is the first protein in the Mla pathway, which facilitates retrograde phospholipid transport and is responsible for preventing phospholipids from accumulating in the outer leaflet of the OM (59). In addition to vancomycin resistance, ⌬wzzE cells also show sensitivity to deoxycholate, a detergent derived from bile salts, although less than that observed in ⌬wecA cells. Deletion of waaL, which prevents formation of ECA LPS , had no effect on vancomycin resistance or DOC sensitivity. As WaaL does not influence deoxycholate resistance, these data suggest that both ECA PG and ECA CYC contribute to the deoxycholate phenotype, while ECA CYC is responsible for vancomycin phenotype. Interestingly, combining strain ⌬wzzE and mlaA deletion leads to an increase in SDS EDTA, deoxycholate, and bile salt sensitivity over that observed with either parent strain. This increase in detergent sensitivity suggests that combining these deletions causes larger changes to the OM than result from the individual mutations. Overall, these data demonstrate removal of ECA CYC causes clear changes to the OM permeability barrier.

DISCUSSION
In this work, we have established that ECA CYC helps to maintain the OM permeability barrier and that YhdP controls this activity of ECA CYC in such a way as to prevent damage to the OM. ECA CYC , not the membrane-associated form of ECA, is responsible Cyclic ECA Maintains the Outer Membrane Barrier ® for some of the OM permeability phenotypes caused by removal of ECA. Furthermore, in a ⌬yhdP background, uncontrolled aberrant activity of ECA CYC causes envelope permeability despite the fact that removing YhdP greatly lowered levels of ECA CYC . This role in maintenance of the OM permeability barrier is the first phenotype described for ECA CYC . ECA is conserved throughout Enterobacteriaceae despite one of the OM forms acting as a surface-exposed common antigen that can lead to antibody production (29). Therefore, ECA must perform cellular functions that justify not only the risk of expressing a common antigen but also the potential for damage caused by ECA CYC . For the surface-exposed forms of ECA, one can imagine roles relating to direct host interactions, such as receptor binding interactions, interactions with other members of Enterobacteriaceae, or roles directly influencing the penetration of toxic substances into the cell; however, it is very difficult to imagine that ECA CYC , from its location in the periplasm, is responsible for interacting with the environment. Instead, ECA CYC must play a role intrinsic to the cell. The changes in OM permeability that occur with removal of ECA CYC demonstrate that ECA CYC plays a role in maintaining the barrier function of the OM.
ECA CYC , as a cyclic soluble molecule made of aminosugars, has some resemblance to cyclodextrins. Cyclodextrins are cyclic carbohydrates made of glucose monomers that have a hydrophilic exterior and a hydrophobic cavity that allows them to bind to hydrophobic guests to increase their solubility and decrease their volubility (60). In fact, some cyclodextrins can pull specific molecules, such as cholesterol, out of membranes without binding to or disrupting the membranes (61)(62)(63). These properties have led to their use in drug formulations, as food additives, in cosmetics, as air deodorizers, and in many other applications (64). It is tempting to speculate that ECA CYC may have similar properties allowing it to bind to specific target molecules in the periplasm and transfer them to or from the OM. In this case, it may be that YhdP is responsible for controlling what molecules are bound or where and how the molecules are unloaded.
Despite the large size of YhdP and strong phenotypes caused by its removal (36,38), it is not apparent that YhdP has any unique role independent of ECA CYC , emphasizing the importance to the cell of controlling ECA CYC activity. This also makes YhdP an important tool allowing for investigation of the effects of uncontrolled ECA CYC in order to elucidate its normal function. The decrease in ECA CYC in the ⌬yhdP strain likely reflects a cellular mechanism to decrease the OM damage due to ECA CYC in the absence of YhdP. The cell compensates for the loss of YhdP either through decreased synthesis or increased degradation of ECA CYC to minimize OM damage. It is also possible that some ECA CYC may leak out of ⌬yhdP cells due to the OM damage; however, the methods for detecting ECA CYC make this very difficult to determine. Nevertheless, even the low levels of ECA CYC remaining are capable of damaging the OM barrier when its activity is uncontrolled. In the future, it will be of interest to investigate the mechanisms through which ECA levels are regulated. Nevertheless, our data suggest that, in the absence of ECA CYC downregulation, the phenotypes of the ⌬yhdP strain would be extremely severe.
Although the structure and topology of YhdP have not been experimentally determined, it is predicted to be an inner membrane protein with an N-terminal and possibly a C-terminal transmembrane helix, with the remainder of the protein exposed in the periplasm (37). YhdP is classified as a member of the AsmA family of proteins due to the presence of a C-terminal AsmA_2 domain and an N-terminal DUF3971 domain, which contain shared sequence motifs with those found in AsmA (58). In E. coli, there are six members of this family, AsmA, YhdP, TamB (YtfN), YhjG, YicH, and YdbH (65). Although the function of these proteins is largely unknown, mutations in asmA have been found to suppress assembly-defective mutations in OMPs (66,67). In addition, TamB has been suggested to interact with the OMP TamA to allow secretion of autotransporters (68). Half of the DUF490 domain of TamA has been crystallized and found to adopt a "taco-shaped" ␤-sheet with a hydrophobic cavity (69). The remainder of the protein is thought to adopt a similar conformation, perhaps allowing amphipathic OMP segments to be transferred to TamA to avoid the aqueous periplasm. This structure is similar to that of the ␤-jellyroll conformation found in the LptA protein responsible for LPS transport across the periplasm (69,70).
Although TamA and YhdP share only 25% identity, it is possible that YhdP adopts a similar conformation allowing it to bind to hydrophobic molecules. If this is the case, then YhdP may bind to hydrophobic molecules and pass them to ECA CYC or unload molecules from ECA CYC . We are currently investigating the specific mechanisms and pathways through which YhdP and ECA CYC act, including the possibility that YhdP and ECA CYC may interact physically. Nevertheless, the functional interaction between YhdP and ECA CYC and the strong phenotype caused by ECA CYC in the absence of YhdP represent an important aspect to envelope biology that has yet to be explored.
The difference in cellular location between ECA CYC and the other forms of ECA, the ability of the cell to make three forms of ECA, and the differing antibiotic sensitivities with removal between these forms, suggests that the function of ECA CYC and YhdP are likely not the same as the functions of ECA PG and ECA LPS . In addition, while ECA PG has a role, direct or indirect, in excluding toxic substances, ECA LPS appears to have no role in maintaining the OM permeability barrier and may instead have a role in interacting with the environment or be a by-product of the reaction that attaches O antigen to LPS. Interestingly, when investigating changes to the various forms of ECA in the presence and absence of yhdP, we observed that the modal chain length of the membranebound forms of ECA varied based on temperature. Knowledge on changes in ECA chain length in response to temperature has not been reported. This change in chain length may be due to specific regulation of ECA length by temperature or by a temperaturedependent change in the activity of the ECA polymerase, WzyE. However, in Yersinia enterocolitica, expression of ECA has been found to be modulated by temperature changes, with high levels of ECA at 22°C and almost undetectable levels at 37°C (71,72). These data suggest that the functional requirements for the membrane-bound forms of ECA may depend on temperature. Furthermore, the differences in regulation of ECA expression between genera in Enterobacteriaceae suggest that the role of ECA may be adapted or modified for the lifestyles of different species. Investigation of these differences may lead to interesting insights into the biology of these species.

MATERIALS AND METHODS
Strains and growth conditions. The strains used in this work are listed in Table S2. Cultures were grown at 37°C in LB medium unless otherwise noted. When necessary, cultures were supplemented with 20 mg/liter chloramphenicol, 25 mg/liter kanamycin, or 25 mg/liter tetracycline. To quantitate ECA CYC levels, cells were grown in M63 medium without nitrogen and supplemented with 0.2% glucose, 0.2% (NH 4 ) 2 SO 4 , 1 mM MgSO 4 , and 100 g/ml thiamine. Deletion alleles originated from the Keio collection (73), unless otherwise noted, and were moved into our strains by P1vir transduction (74). Unless otherwise indicated, resistance cassettes were flipped out as has been described previously (75).
Antibiotic sensitivity assays. For growth curves, overnight cultures were diluted 1:1,000 into 2 ml fresh LB containing the compounds indicated in a 24-well format, sealed with breathable film, and grown shaking at 37°C in a BioTek Synergy H1 plate reader. The optical density at 600 nm (OD 600 ) was assayed every 10 min. MICs were determined as has been reported elsewhere (57). The MIC was taken to be the minimum concentration of antibiotic at which no growth was observed. For efficiency of plating (EOP) assay, 10-fold dilutions of overnight cultures were made and replicate plated onto LB plates supplemented with the indicated chemicals. Plates were incubated at 30°C (unless otherwise noted) overnight and plates were imaged.
Generation and mapping of suppressor mutations. To generate spontaneous suppressor mutants, we plated 10 7 ⌬yhdP strain cells on LB supplemented with 70 mg/liter vancomycin and incubated the plates overnight at 30°C. Colonies were picked and subjected to secondary screening for vancomycin and SDS EDTA resistance. The suppressor mutations were mapped as has been described elsewhere (76). We generated a Tn5 mutant library in a ⌬yhdP strain as has been described elsewhere (36). Our selection and screening strategy for isolating suppressing mutations is outlined in Fig. S3A. The transposon insertion site were determined by arbitrary PCR as has been described previously (59), except that the TetA-out and TetA-seq primers were replaced with Tn5-out (5= GGTTGTAACACTGGCAGAGC 3=) and Tn5-seq (5= TCCGTGGCAAAGCAAAAGTT 3=).
Phylogenetic co-occurrence and homologies. To determine whether yhdP and the genes of the wec operon tend to occur in the same genomes across organisms, we utilized the co-occurrence channel of STRING-DB (39,77). We searched the database in multiple-protein mode for yhdP and ECA biosynthesis genes and took the phylogenetic co-occurrence scores from the generated table. The derivation of these scores from homology tables has been described elsewhere (77). To examine the level of homology for Cyclic ECA Maintains the Outer Membrane Barrier ® possible YhdP homologues, we used the homology scores generated via the STRING database to find whether the indicated classification of organisms was predicted to have an YhdP homology and what the highest and lowest homology scores were for the organisms included in STRING-DB within that classification. These scores were then plotted. E reporter assay. To determine the level of activation of the E system, we utilized a plasmid reporter with the promoter from micA driving expression of GFP (57) as has been reported elsewhere (78). Each of the three independent experiments was conducted in technical triplicate. The significance of the differences observed was calculated using the nonparametric Mann-Whitney test.
Quantification of ECA levels. Membrane-associated forms of ECA were analyzed by immunoblot analysis. Cells from an overnight culture were resuspended in BugBuster protein extraction reagent (Millipore Sigma) at an equivalent OD 600 of 40 and then combined with an equal volume of Laemmli sample buffer (Bio-Rad) with 4% ␤-mercaptoethanol. Samples were boiled 5 min and then cooled and loaded on 12% TGX gels (Bio-Rad). The samples were transferred to nitrocellulose and were probed with a 1:10,000 dilution of anti-ECA antibody. Rabbit polyclonal anti-ECA antibody was a kind gift from Renato Morona (University of Adelaide). Donkey anti-rabbit secondary antibody conjugated to horseradish peroxidase was utilized at a 1:20,000 dilution and detected using a Crescendo ECL system (Millipore Sigma). The specificity of the ECA antibody could be observed based on the lack of signal with the ⌬wecA strain (Fig. 4B, lanes 1 and 6). Levels of ECA were quantitated using ImageJ. Densitometry was performed on blots with the lowest exposure at which the ECA bands for the indicated samples could be detected. Densitometry was performed on the whole lane and manually baselined. Similar results were found when each ECA band was measured individually. For each of three to five biological replicates, fold values to the yhdP ϩ sample were calculated. Then, the biological replicates were averaged and the standard errors of the means (SEM) were calculated. Significance was calculated using the Mann-Whitney test.
ECA CYC was purified as has been described before, with minor modifications (34). Cells were grown in LB medium for determination of the ECA CYC structure. For determination of ECA CYC levels, cells were grown in M63 medium with either a normal or heavy ( 15 N) nitrogen source, and cultures for comparison were combined at the beginning of purification. After ethanol precipitation, supernatants were lyophilized and subsequently resuspended with 0.1% formic acid. Acidified samples were loaded on C 18 StageTips (79), washed twice with 0.1% formic acid, and eluted with 20% acetonitrile with 0.1% formic acid. Eluates were then dried in a Speedvac before reconstitution with 20% acetonitrile. Samples were analyzed by MALDI-TOF/mass spectroscopy as has been previously described (34). Spectra were obtained with a Bruker UltrafleXtreme instrument calibrated with Red phosphorous. For relative quantification, the ratio of the areas of the heavy and normal ECA CYC peaks was calculated for three biological replicates. Significance was calculated using the Mann-Whitney test.