O-Antigen-Dependent Colicin Insensitivity of Uropathogenic Escherichia coli

Escherichia coli infections can be a major health burden, especially with the organism becoming increasingly resistant to “last-resort” antibiotics such as carbapenems. Although colicins are potent narrow-spectrum antimicrobials with potential as future antibiotics, high levels of naturally occurring colicin insensitivity have been documented which could limit their efficacy. We identify O-antigen-dependent colicin insensitivity in a clinically relevant uropathogenic E. coli strain and show that this insensitivity can be circumvented by minor changes to growth conditions. The results of our study suggest that colicin insensitivity among E. coli organisms has been greatly overestimated, and as a consequence, colicins could in fact be effective species-specific antimicrobials targeting pathogenic E. coli such as uropathogenic E. coli (UPEC).


RESULTS AND DISCUSSION
UPEC ST131 is insensitive to multiple colicins. UPEC ST131 exposed to colicin E9 showed a Ͼ6,000-fold difference in MIC compared to that of the colicin-sensitive E. coli K-12 strain JM83 when tested on solid medium and a 1,000-fold higher MIC in liquid culture (LB medium) (Fig. 1). Resistance to colicins is typically associated with a cognate immunity protein or mutations in receptor/translocator proteins (1,27). To test whether these mechanisms were responsible for the observed resistance, UPEC ST131 was exposed to different colicins. The chosen colicins target different receptors/translocation proteins (ColIa, Cir; ColD, FepA; ColE9, BtuB/OmpF), encompassing both the Tol (group A colicins) and the Ton (group B colicins) complexes, and kill the target cell via different mechanisms, including via the depolarization of the cytoplasmic membrane and nucleolytic activity in the cytoplasm (ColIa, pore forming; ColD, tRNase; ColE9, DNase). UPEC ST131 had a 10,000-fold higher MIC of ColIa compared to that of E. coli JM83 and was unaffected by colicin D at the highest concentration tested (10 M) (see Fig. S1 in the supplemental material). A genome analysis of UPEC ST131 did not identify significant similarity to colicin immunity protein genes for any of the colicins tested, suggesting that mutations in receptors or the presence of immunity proteins is not responsible for the resistance and that UPEC ST131 has a nonspecific mechanism which causes insensitivity to colicin.
To determine if insensitivity toward colicins is a widely observed phenomenon across UPEC strains, 10 UPEC strains were exposed to ColE9 to determine the MICs. Six of the ten strains were not affected by ColE9 at concentrations up to 10 M (see Tables   Time (minutes) OD 600 0 nM FIG 1 Solid and liquid medium killing assays of UPEC ST131 and colicin-sensitive E. coli JM83 exposed to purified ColE9. (Top) Solid medium killing assays on LB agar using serial dilutions of ColE9. UPEC ST131 shows slight sensitivity to the highest concentrations of ColE9, whereas E. coli JM83 is sensitive at picomolar concentrations. (Bottom) Liquid medium killing assay using LB. UPEC ST131 had a 1,000-fold higher MIC than JM83. S1 and S2). The four susceptible isolates had MICs for ColE9 between 30-fold and 2,000-fold higher than the sensitive E. coli MG1655 strain.
LPS and O-antigen biosynthesis genes are associated with colicin insensitivity. To elucidate the genetic basis for colicin insensitivity in UPEC ST131, a transposondirected insertion sequencing (TraDIS) approach was used. TraDIS is a powerful technique to assay the fitness contribution of genes across an entire genome (28). A TraDIS transposon library for UPEC ST131, NCTC 13441, was established using the method described previously (29,30). The library had a high insertion density, with 450,000 unique insertion sites across the 5.4-Mbp genome, giving an average density of 1 insertion per 12 bp, allowing for an accurate and sensitive evaluation of the role of each gene in colicin insensitivity (barring those which appear essential under these conditions). To test for a role in colicin insensitivity, two samples of the transposon library were prepared: one sample was exposed to a sub-MIC of ColE9 (0.5 nM) and the other was treated with buffer. Genes which provide insensitivity will have a reduced number of transposon insertions compared to that of a control sample, as insertions would disrupt that gene and stop the production of a protective protein. One hundred fourteen genes had a significant decrease in insertions (genes normally involved in insensitivity) compared to that in the non-colicin-treated control, and 34 genes had an increase in insertions (genes normally involved in rendering bacteria susceptible to the colicin) (see Data Set S1). Thirty-three of the genes with decreased insertions (28%) were mapped to five operons responsible for LPS biosynthesis, O-antigen biosynthesis, enterobacterial common antigen (ECA) biosynthesis, membrane lipid asymmetry, and lipid A modification.
The presence of the O-antigen has been shown to provide insensitivity to colicins in some E. coli backgrounds but not for UPEC ST131 (16). Nine of the genes with a role in nonspecific colicin insensitivity are annotated with functions associated with O-antigen production and have been shown to decrease sensitivity toward SDS and serum (31). We also identified seven core LPS genes with roles in colicin insensitivity, in keeping with the O-antigen being attached to core LPS. O-antigen is ligated to core LPS via the O-antigen ligase RfaL, which we also found to be associated with colicin insensitivity. Additionally, rfaH, the LPS biosynthesis regulator, was also identified as providing an increase in fitness in the presence of colicin. Four genes of the arn operon, involved in the modification of lipid A, were also found to be required for colicin insensitivity. The arn operon is involved in resistance to cationic antimicrobial peptides (CAMPs) by increasing the net positive charge of lipid A in the OM through the addition of a 4-amino-4-deoxy-L-arabinose (L-Ara4N) moiety (32). The PhoPQ two-component system is required for activation of the arn operon, and both of these components were highlighted as being involved in colicin insensitivity. Increases in L-Ara4N have been shown to increase resistance to polymyxins in Salmonella enterica serovar Typhimurium and Klebsiella pneumoniae (33). The mla operon encodes an ATP transporter that is responsible for maintaining the asymmetry of the OM, and mutation of the mla genes has been demonstrated to cause sensitivity to SDS and EDTA (34). In addition to the overrepresentation of genes involved in five OM-related operons, we identified 12 genes which had no obvious role in OM stability from their annotation but were shown to have a role in serum resistance in E. coli ST131 (31,35). Though seven genes from the ECA biosynthesis operon were highlighted by TraDIS, these genes may have roles in O-antigen biosynthesis, and the role of ECA in colicin insensitivity is still to be determined (36). Genes which increased colicin E9 susceptibility included, as expected, those for its receptor and translocator proteins, btuB and ompF, respectively. Two genes involved in formation of the core LPS, rfaQ and rfaY, were also identified as increasing colicin susceptibility. Though these genes were not previously identified as necessary for sensitivity to colicins, a positive role of the core LPS in colicin activity has been identified for colicins N and A (37).
Destabilizing the LPS decreases nonspecific insensitivity toward colicin. The TraDIS analysis highlighted a significant role for the O-antigen in colicin insensitivity. To test this, we used EDTA to destabilize the LPS packing and assayed for an increase in colicin sensitivity. EDTA destabilizes LPS packing by chelating divalent cations which stabilize adjacent lipid A molecules (38). The introduction of EDTA into the growth medium for UPEC ST131 reduced the MIC for three bacteriocins by 10,000-fold for ColIa and 100-fold for ColE9 (see Fig. S2). UPEC ST131 also became sensitive toward ColD at 10 M, whereas it previously showed no activity. Repeating the TraDIS analysis in the presence of EDTA (in the media for both the colicin-treated and control pools) revealed 21 genes associated with colicin insensitivity. These genes were all involved in LPS biosynthesis, O-antigen biosynthesis, ECA biosynthesis, and lipid A modification (Table 1).
O-antigen-restored E. coli K-12 is colicin insensitive. To test directly the effect of the O-antigen on colicin sensitivity, we used strains of E. coli K-12 MG1665 which have a restored O-antigen (39) ( Table 2). Henderson and coworkers restored the wbbL gene in E. coli K-12 MG1655 to give rise to two strains: L9 with a wild-type (WT) gene order and L5 with incorrect synteny and a significant (ϳ10-fold) decrease in O-antigen (39). Both MG1655 and the L5 derivative were sensitive to multiple colicins, whereas the L9 strain, with its fully restored O-antigen, was insensitive to 10 M ColE9 (Ͼ100,000-fold increase in MIC) on solid medium and insensitive up to 250 nM ColE9 in liquid medium (Ͼ5,000-fold increase in MIC) (Fig. 2). L5 and L9 strains differ only in the densities of O-antigen, demonstrating that the presence of O-antigen is not sufficient to generate colicin insensitivity but that the density of O-antigen is the critical factor for the 100,000-fold increase in MIC. The LPS core has previously been reported as aiding the translocation of colicins N and A (37). These colicins still had a greatly reduced activity toward the L9 strain compared to that toward the L5 strain (see Fig. S3).  Colicin insensitivity in UPEC ST131 is dependent on growth conditions. The L5 and L9 strains showed that changes in the densities of O-antigen can render bacteria sensitive to colicin. Conversely, destabilization of the LPS packing in the outer membrane by EDTA treatment causes an increase in colicin sensitivity. Given these impacts on colicin sensitivity, we set out to determine if growth conditions influenced colicin sensitivity. Bacteria occupy a wide variety of environmental niches and, as a result, are exposed to different conditions and stresses which could influence the density of O-antigen and therefore colicin sensitivity. Biolog phenotypic microarrays were used to study respiration in the presence and absence of sub-MIC ColE9 across hundreds of different growth conditions (40). Four plates were investigated: two screened various carbon sources, one explored a range of pHs with different additives, and a final plate varied osmolytes and osmolarity (see Data Set S2).
No difference in colicin sensitivity was observed for any of the altered carbon source conditions (191 different conditions). However, colicin sensitivity was found to be dependent on osmolarity and pH. Concentrations of NaCl of Ͼ0.68 M caused sensitivity to ColE9 at 41 nM (one-third MIC of ColE9 in Biolog inoculation medium), and sodium lactate (0.62 M) and KCl (Ͼ0.4 M) had similar effects. Interestingly, UPEC ST131 was sensitive to sub-MIC ColE9 in the presence of urea. UPEC ST131 was also sensitive to ColE9 at alkaline pH (pH Ͼ8). We therefore decided to validate two conditions: Tris and urea. Tris is a commonly used buffer, but it is also known to trigger the release of LPS from the OM (41). Urea was chosen as it is present in the urinary tract as the major constituent of urine (ϳ150 to 500 mM) (42). UPEC ST131 and the O-antigen-presenting In contrast, the L9 strain was only sensitive in the presence of Tris. This may indicate that the L9 strain is producing a denser O-antigen than UPEC ST131 and Tris is a better agent for LPS removal or that UPEC ST131 O-antigen is particularly affected by urea (Fig. 3).
Dense O-antigen diminishes colicin receptor binding. Translocation of ColE9 across the outer membrane involves binding to an outer membrane receptor (BtuB) before threading an unstructured domain through OmpF to contact the Tol system in the periplasm (6,43). O-antigen might provide insensitivity by blocking the binding of ColE9 to its receptor or by preventing the unstructured domain from threading through OmpF. To distinguish between these possibilities, ColE9 binding to BtuB on the surfaces of target bacteria in the presence and absence of O-antigen was investigated by fluorescence microscopy, using Alexa Fluor 647-labeled ColE9 (44).
A statistically significant difference was observed between the L5 and L9 strains: L5 displayed clear colicin binding, whereas no colicin binding was observed for L9 (Student's t test, P Ͻ 0.01) (Fig. 4). Tris, which decreases the MIC of colicins for both the L9 strain and UPEC ST131, was evaluated to test if a decrease in MIC correlated with an increase in receptor binding. No differences were observed in colicin binding for the L5 strain in the presence or absence of Tris, suggesting that receptor binding is saturated and the presence of low-density O-antigen is not sufficient to block colicin binding. The addition of Tris caused a significant increase in receptor binding for both the L9 and UPEC ST131 strains. No significant difference was observed in the intensities between the L5 strain and the L9 strain with addition of Tris, suggesting that 100 mM Tris is sufficient to remove any protection afforded by the O-antigen. Urea was not used as a condition for microscopy, as it greatly increased background fluorescence.
Previous studies using fluorescently labeled colicins have shown that outer membrane proteins (OMPs) cluster into supramolecular assemblies termed OMP islands in E.   (44,45). We therefore set out to determine if O-antigen affected this clustering behavior. Using total internal-reflection fluorescence (TIRF) microscopy, we observed OMP islands in the MG1655, L5, and UPEC strains (Fig. S3). The L9 strain could not be validated for the presence of OMP islands, as colicin did not bind BtuB under the conditions tested. Concluding remarks. The presence of O-antigen in the bacterial outer membrane has been linked to nonspecific insensitivity toward colicin, phages, and complement (17,46). Moreover, the length of O-antigen affects colicin insensitivity in Shigella (17). The difference in colicin sensitivity between the E. coli L5 and L9 strains in the present study, coupled with the observation that Tris and EDTA render cells more sensitive to colicin, suggests that subtle changes in the density of O-antigen have significant effects on colicin sensitivity. High levels of colicin resistance across E. coli collections have been observed under lab conditions. Bacteria in the environment will be subjected to a range of stresses and insults to the OM and are unlikely to have an O-antigen as dense as that under laboratory conditions, meaning the scale of colicin insensitivity is likely overes- O-antigen-producing L5 and L9 mutants were compared to UPEC ST131 by exposure to fluorescently labeled ColE9 in LB medium (green) or LB medium supplemented with 100 mM Tris (orange). Each point represents the average fluorescence intensity over a single cell. Statistically significant differences in labeling were observed between the isogenic L5 and L9 strains (t test). Addition of 100 mM Tris-HCl caused significant increases in cell labeling for both the L9 strain and UPEC ST131.
timated. This also suggests that such nonspecific colicin insensitivity could change depending on the environment, host infection site, and stage of infection. For example, during a chronic infection, Pseudomonas aeruginosa loses its O-antigen, while Burkholderia dolosa gains the ability to produce O-antigen (47). For UPEC ST131, the sensitivity toward colicin in the presence of urea presents an intriguing possibility. Uropathogenic E. coli is transmitted to a new host by fecal-oral contamination before mechanical transmission to the urinary tract from the anus (48). UPEC bacteria could therefore be resistant to colicin in the gut, where they will be surrounded by other E. coli, but become sensitive to colicin in the urinary tract, a relatively sterile environment where bacteria would face less competition. It therefore becomes possible that colicins could be an effective treatment for UTI while not affecting commensal E. coli.
Colicin-like proteins are found throughout Gammaproteobacteria (24). To date, O-antigen dependent insensitivity has only been studied using colicins. As the O-antigen varies between strains, so may its ability to protect against receptor binding. The O-antigen is not the only structure present at the OM, many bacteria produce a capsule, a thick layer of polysaccharide which can protect against antimicrobial proteins and has been associated with virulence (49). The thickness of this capsule varies greatly between species (E. coli, 10 nm; K. pneumoniae, 160 nm) and could be a significant barrier for colicin receptor binding (50).
Many studies, both theoretical and experimental, have indicated roles for colicins in bacterial competition and in maintaining a diverse microbial community (51,52). However, it is possible that colicins also have another purpose. In a mixed microbial community of colicin-producing and colicin-insensitive (due to O-antigen) bacteria, the basal level of colicin production would act as a selective pressure to ensure bacteria maintain their O-antigen, increasing the fitness of the overall population.
In summary, we solve the problem of why smooth E. coli harbor colicin plasmids yet appear insensitive toward colicins when assayed under laboratory conditions. O-antigen-dependent colicin insensitivity was observed for UPEC ST131, a clinically important pathogen. Using isogenic O-antigen mutants, we found that the density of the O-antigen is important for colicin sensitivity. Additives that cause shedding of LPS decreased the MIC of colicins for insensitive strains, suggesting that environmental factors influence the density of O-antigen and hence sensitivity toward colicin.
Solid medium killing assay. Soft LB agar (0.7% [wt/vol)] was inoculated (1:25) with a culture of the indicator strain (optical density at 600 nm [OD 600 ] of ϳ0.4) and overlaid onto LB agar (1.5% [wt/vol]). Serial dilutions of colicins were prepared using 20 mM Tris-HCl (pH 7), and 2 l was spotted onto the agar. Plates were incubated overnight at 37°C, and the MIC was defined as the lowest concentration at which a zone of clearance in the soft agar was observed.
Liquid medium killing assay. Overnight cultures of the strains E. coli MG1655, E. coli L5, E. coli L9, and UPEC ST131 were used to inoculate (1:100) 10 ml fresh medium supplemented as appropriate and cultured at 37°C until cultures reached an OD 600 of ϳ0.4. Cultures were then diluted (1:3, 200 l) in a flat-bottomed 96-well plate in triplicates containing medium supplemented with the appropriate additive with or without the addition of colicin. Growth was monitored by measuring the OD 600 using the Clariostar high-performance microplate reader, and MIC was defined as the concentration at which Ͻ50% of growth was inhibited).
TraDIS library generation, testing, and analysis. The UPEC ST131 (NCTC 13441) TraDIS library was generated using the method previously described (29 6 or without colicin. Overnight cultures were used to inoculate fresh identical media 1:10 and cultured overnight at 37°C. DNA was extracted from the passaged culture and sequenced using the Illumina HiSeq 2500 platform. Insertions were analyzed using the BioTraDIS pipeline, a pearl library with a host of tools for TraDIS analysis (30). Only reads which contained the transposon tag were considered for analysis. Twenty reads per insertion was considered significant, and reads within 10% of the 3= end of the open reading frame were ignored, as these insertions may not disrupt protein function (54). A log fold change of Ͼ2 with a false-discovery rate (q value) of Ͻ0.00005 was used as a significant threshold.
Biolog. Biolog phenotypic microarrays (PM) monitor the respiration of bacteria across different growth conditions by measuring the reduction of a tetrazolium-based dye (40). Four PM plates were selected for testing: PM1 and PM2, which test different carbon sources; PM9, which tests the effects of osmolytes; and PM10, which tests a range of pH values. UPEC ST131 was cultured overnight in MMTE medium at 37°C overnight with shaking and was used to inoculate the inoculation fluid used in Biolog. Colicin E9 Im9(His) 6 was added to the PM plates, and as a control, an inactive colicin, which lacks the unstructured region at the N terminus and contains a disulfide bond which does not allow for the conformational change needed for translocation, was used to test for UPEC ST131 respiration without the effect of a cytotoxic colicin (55). Inoculated PM plates were observed by using an OmniLog incubator and reader for 48 h, measuring absorbance every 15 min. Biolog absorbance data were analyzed using the DuctApe PM analysis suite (56). DuctApe measures nine parameters and performs K-means clustering to assign each curve an activity index. Differences between the active-and inactive-colicin-treated clusters were calculated.
Microscopy and image analysis. For fluorophore labeling, site-directed mutagenesis was used to incorporate a cysteine into ColE9 (H551A K469C) or ColE9 K469C with an N-terminal deletion, Δ 2-62 , which is known to block import into cells (44). Purified protein was labeled with Alexa Fluor 647 (AF647). AF647 with an exposed maleimide group was covalently linked to the sulfhydryl of the engineered cysteine to form a stable conjugate. Excess dye was removed by passing the protein through a Superdex 75 10/300 column (GE Healthcare).
E. coli L5, E. coli L9, and UPEC ST131 were incubated overnight at 37°C with shaking in LB medium. Overnight cultures were used to inoculate (1:100) LB medium with additives and were incubated at 37°C until an OD 600 of ϳ0.4 was reached. From each sample, 1,600 l was pelleted by centrifugation at 7,000 ϫ g for 3 min, and the pellet was resuspended in 200 l of 1.5 M ColE9 and incubated for 30 min with rotary inversion at room temperature. Cells were washed three times in 400 l of LB with the additive of interest. Cells were resuspended in approximately 200 l of medium, and 10 l of cells was placed onto a 1% agar pad. Fluorescence was visualized by widefield microscopy using the DeltaVision OMX V3 system. Images were acquired following excitation at 642 nm for 50 ms and processed using Fiji image processing software (57). Student's t tests were used to identify statistically significant differences between treatment and strains.
Accession number(s). TraDIS sequencing data are available under the accessions ERS939266 to ERS939273 and ERS441440 to ERS441443. Other data supporting this study can be found within the article and the supplemental material or requested from the corresponding author.

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at https://doi.org/10.1128/JB .00545-18.