We previously discovered a dominant-negative isoform of ObgE that causes cell death in E. coli. This mutant isoform contains a single amino acid substitution, K268I, and is called ObgE^∗ (Dewachter et al., 2015, 2016a,b). When ObgE^∗ is expressed, it causes a drastic reduction in the number of viable cells, as measured by colony forming units (CFUs). In contrast, expression of wild-type ObgE does not influence viability in comparison to the vector control (Figure 1A). The drop in viability caused by ObgE^∗ is accompanied by a loss of membrane integrity, which can be detected by staining with propidium iodide (PI), a red fluorescent dye that can only enter cells with compromised membranes. To reconstruct in detail the order of events occurring upon ObgE^∗ expression, we carried out time lapse experiments of E. coli expressing ObgE^∗ in the presence of PI (Figure 1B). First, ObgE^∗ very rapidly causes a defect in cell separation; newly formed daughter cells fail to separate and instead remain together in a cell chain. After one or two rounds of defective cell division, cells cease to grow and divide and start staining PI-positive, indicating that their membrane integrity is lost. Remarkably, not all cells in one chain stain PI-positive at the same time, indicating that at least in some cases constriction has proceeded normally and has separated the cytoplasm of the daughter cells. Cells that stain PI-positive are able to maintain this PI staining over several hours. However, over a time course of approximately 10–12 h, cytoplasmic content together with PI is lost from the cell, indicating that ObgE^∗ causes stepwise, slowly progressing cell lysis. Since all PI-positive cells eventually lyse and PI-negative cells remain intact, we can quantify lysis by PI staining, as was done previously (Packard et al., 2013). Because individual cells in a chain were never able to remain intact when parts of the chain stained PI-positive, the entire chain was considered to be compromised if at least one cell lost its membrane integrity. This analysis shows that ObgE^∗ triggers lysis in the majority of the population, while virtually all cells remain intact upon expression of wild-type ObgE (Figure 1C).

A detailed study of E. coli morphology by scanning electron microscopy revealed that ObgE^∗ expression leads to the formation of membrane protrusions, termed blebs (Figure 2A). Similar membrane structures were previously associated with cell lysis (Yao et al., 2012; Sutterlin et al., 2016). The excess amount of membrane that forms blebs points to disturbance of membrane homeostasis by ObgE^∗. To gain further structural insight into the nature of these blebs, the cytoplasm, membranes and peptidoglycan of E. coli expressing ObgE^∗ were simultaneously labeled (Figure 2B). Cytoplasm was visualized by the expression of a cytoplasmic GFP label, membranes were stained with the red lipophilic dye FM4-64, and peptidoglycan was visualized using HADA [HCC-amino-D-alanine, a fluorescently labeled D-amino acid that is readily incorporated into the peptides of peptidoglycan (Kuru et al., 2015)]. No membrane blebs were found in the vector control or E. coli expressing wild-type ObgE, although the latter did influence cell morphology by increasing cell length, in accordance with literature (Kobayashi et al., 2001; Dutkiewicz et al., 2002). Expression of ObgE^∗ leads to the formation of membrane blebs that contain the cytoplasmic GFP label. The lumen of these blebs is therefore in direct contact with the cytoplasm. Because of this continuum between cytoplasm and blebs, it is likely that they are lined by inner as well as outer membrane. The presence of inner membrane inside blebs was confirmed by construction of a 3D-image of blebs by focused ion beam-scanning electron microscopy (FIB-SEM), a technique that allows for high resolution imaging of a desired volume in three dimensions by electron microcopy (Kizilyaprak et al., 2014) (Figure 2C). However, although blebs contain inner membrane, there is no clear defect in the underlying peptidoglycan layer since HADA labeling is uniform and uninterrupted at the site of bleb formation. Any potential peptidoglycan defect allowing for the protrusion of inner membrane should therefore be rather small. Additionally, the composition of peptidoglycan remains unaltered in the presence of ObgE^∗, arguing against major rearrangements or disturbance of peptidoglycan structure (Supplementary Figure S1). FM4-64 and HADA staining revealed that blebs are membrane structures that lack the rigidity and protection of the peptidoglycan layer. Membrane blebs are therefore highly fragile structures that are prone to rupturing (Yao et al., 2012). Indeed, blebs usually have short life spans and, importantly, their rupturing coincides with loss of the cytoplasmic GFP label not only from the lumen of the bleb but from the entire cell (Figure 2D). Taken together, these data indicate that lysis proceeds through the formation of membrane blebs that contain both outer and inner membrane and thus are connected to the cytoplasm. When these blebs rupture, cytoplasmic content is released and lysis occurs.

Quantification of the localization of blebs showed that in 57% of all cases, blebs were situated on top of a fully formed septum (Figure 3A). These defective septa are not resolved to form two entirely separate and unlinked daughter cells but instead remain part of a cell chain. The predominant occurrence of blebs at these aberrant division sites indicates that an additional defect in these septa allows for the formation of blebs. 22% of blebs were found at cell poles and the remaining 21% were located elsewhere in the cell (Figure 3A). Interestingly, blebs belonging to the latter category are not randomly located along the cell length but instead are centered around midcell position (Figure 3B). Since cell division occurs at this location, these blebs likely arise through a division defect. Together with blebs that are formed on top of a septum, this makes for a total of 78% of blebs that are formed at division sites, either before or after constriction. Since cell lysis proceeds through the formation and rupture of blebs, this observation hints at an important role for cell division in the progression of lysis.

Since ObgE^∗ causes cell chaining and thus clearly interferes with normal progression of cell division, and because the majority of blebs are found at division sites and thus likely occur through a septum defect, we investigated the role of cell division in the progression of lysis. To do this, several conditions were selected in which division is either inhibited or severely slowed down. First, ObgE^∗ was expressed at two different time points in stationary phase and its effect on lysis and loss of viability was determined by PI staining and plate counting, respectively. Likewise, we examined the effect of ObgE^∗ during growth in M9 minimal medium, where the doubling time of E. coli is approximately two times higher than in lysogeny broth (LB) and cells thus only divide half as fast (Supplementary Figure S2). However, since both conditions also have a large effect on general metabolism and other aspects of cell growth, the role of cell division was also investigated more directly by expressing ObgE^∗ in the presence of either aztreonam or cephalexin, two antibiotics that specifically inhibit division. They do so by mainly targeting PBP3 and thus inhibit the synthesis of septal peptidoglycan while having no or very limited effects on cell elongation (Satta et al., 1995; Yao et al., 2012). It is likely that some of these conditions affect the expression level of ObgE^∗ inside the cell. To account for this, lysis and loss of viability were first quantified under standard conditions at 25 different ObgE^∗ concentrations. These data were used to construct a correlation curve that reflects the extent of lysis and loss of viability at every possible concentration of ObgE^∗ within a certain range (Supplementary Figure S3). In conditions that slow down or inhibit cell division, the percentage of intact cells is greatly increased and is much higher than what is expected based on the expression level (Figure 4A). In fact, in all conditions where cell division is inhibited (i.e., ObgE^∗ expression in stationary phase or in the presence of aztreonam or cephalexin), the percentage of intact cells approaches 100%, indicating that virtually no lysis occurs anymore. Cell division therefore appears to be essential for the progression of ObgE^∗-mediated lysis, more specifically we postulate that a defect in septal peptidoglycan metabolism is responsible for lysis. When ObgE^∗ is expressed in M9 minimal medium there is still a detectable amount of lysis. This is to be expected since cell division and septal peptidoglycan synthesis still occur in minimal medium, albeit with a lower frequency. To validate that it is division and/or septal peptidoglycan synthesis that causes lysis and not merely peptidoglycan synthesis in general, the amount of lysis was also measured in the presence of mecillinam. Mecillinam is an antibiotic that targets PBP2 and thus specifically inhibits peptidoglycan synthesis for cell elongation while having no effect on septal peptidoglycan synthesis (Satta et al., 1995). In the presence of this antibiotic, the percentage of intact cells is exactly what is expected based on the expression level, confirming that it is indeed cell division that is necessary for lysis to occur (Supplementary Figure S4).

In addition to estimating lysis by PI staining, ObgE^∗-mediated toxicity was also quantified by CFU counting. Since the latter assay is based on cell growth, it determines the viability of cells instead of the integrity of the membrane as PI staining does. As shown in Figure 4B, expression of ObgE^∗ in stationary phase, in M9 minimal medium or in the presence of either aztreonam or cephalexin has no effect on ObgE^∗-mediated loss of viability. Measured values are located within the 99% prediction bands and thus do not significantly differ from the expected value. We therefore conclude that, while cell division is essential for the progression of lysis, it does not affect ObgE^∗-mediated loss of viability.

The cell chaining phenotype caused by ObgE^∗ points to a defect in the separation of newly formed daughter cells. Moreover, at these unresolved division sites, the cytoplasmic membrane is able to penetrate the peptidoglycan layer to form blebs that cause lysis. Both phenotypes could be caused by a faulty regulation of the amidases AmiA, AmiB, and AmiC. Although many enzymes cooperate to achieve cell separation, they are seen as the main executioners of this final stage of cell division (Heidrich et al., 2002; Vollmer et al., 2008; Typas et al., 2011). In their absence, E. coli forms chains of cells up to 20 cells long that are unable to separate (Heidrich et al., 2002). A lack of amidase activity thus causes cell chaining. Moreover, since amidases are peptidoglycan hydrolases, uncontrolled activity could also cause excessive degradation of peptidoglycan, leading to cell lysis. Another system of interest is the Tol-Pal protein complex that helps maintain a fixed distance between all three envelope layers (Gerding et al., 2007; Gray et al., 2015). Disruption of this system leads to a moderate cell chaining phenotype in medium of low osmolarity. Moreover, in the absence of a functional Tol-Pal complex, cells form outer membrane vesicles at division sites and cell poles (Gerding et al., 2007). Because of the obvious similarities with phenotypes associated with ObgE^∗-mediated lysis, we quantified lysis in an E. coli triple amidase knock-out strain (E. coli ΔamiA ΔamiB ΔamiC) and in single-gene knock-outs of the tolQ or tolR genes (Figures 5A,B). In all three strains, membrane integrity is decreased under control conditions; whereas 99.6% of a wild-type population is intact, this value drops to 63.6% by deletion of amiA, amiB, and amiC, and is lowered to 93.8 and 97.1% in the ΔtolQ and ΔtolR strain, respectively (Figure 5A, Vector). Surprisingly, overexpression of the essential ObgE protein further decreases membrane integrity of these knock-out strains. Upon expression of ObgE, integrity decreases by 5.1% in the triple amidase knock-out strain (p = 0.0322), 5.7% in the ΔtolQ strain (p = 0.0471), and 6.2% in the ΔtolR strain (p = 0.0305). Since ObgE has no effect on the integrity of a wild-type strain (p = 0.999), these remarkable results hint at a functional link between ObgE, the amidases and the Tol-Pal complex.

ObgE^∗ expression still has a very strong negative effect on cell integrity in all three deletion mutants. The amount of ObgE^∗-mediated lysis varies slightly in the different strains, owing to the fact that the baseline of membrane integrity is affected by the genomic deletions. When the amount of lysis caused by ObgE^∗ is normalized to the amount of lysis detected when wild-type ObgE is expressed, no significant differences in the percentage of intact cells were found among the three deletion strains (Figure 5B). Since lysis upon ObgE^∗ expression does not decrease in the absence of AmiA, AmiB, and AmiC or in the absence of a functional Tol-Pal complex, we conclude that none of these proteins directly cause ObgE^∗-mediated lysis.

When lysis is prevented, ObgE^∗ expression still decreases cell viability (Figure 4). To determine how ObgE^∗ does so, the resumption of growth after ObgE^∗ expression was monitored by time lapse microscopy. To separate the lysis and loss of viability phenotypes, ObgE^∗ was expressed in stationary phase where lysis is prevented. Afterward, cells were seeded on an LB agarose pad lacking the inducer of ObgE^∗ expression and incubated to allow cell growth. These conditions thus mimic the plate counting experiments in which a loss of viability is detected. In the control experiment where wild-type ObgE was expressed during stationary phase, cells rapidly resume growth after they are transferred to fresh medium. They quickly start to divide and form colonies (Figure 6A). When ObgE^∗ was expressed during stationary phase, plate counting experiments demonstrated that there is a drop of 6 log units in colony formation (Figure 1A). This time lapse experiment showed that colonies fail to develop because cell division cannot be restored when cells are allowed to recover on fresh medium (Figure 6A). Even after 48 h, cells have not divided (data not shown). However, bacteria do appear to resume cell growth briefly (Figure 6A, inset). Quantification of cell length at the start of the time lapse and 6 h later revealed that, during this time, cell length increases to a limited extent (Figure 6B). ObgE^∗ thus causes loss of viability by blocking progression through the cell cycle after allowing a limited amount of cell elongation but before cell division occurs. Importantly, this cell cycle arrest is not specific to exit from stationary phase and/or elevated (p)ppGpp levels. When ObgE^∗ was expressed in exponential phase in the presence of aztreonam to prevent cell lysis, cell division could not be restored once cells were washed and resuspended in fresh medium (Supplementary Figure S5). Eventually, the arrested cells lose membrane integrity, as evidenced by gradually progressive PI staining. No membrane blebs were observed (data not shown).

Because of the role of ObgE in licensing chromosome segregation and subsequent cell division (Kobayashi et al., 2001; Foti et al., 2007), we checked whether ObgE^∗ causes cell cycle arrest at the stage of chromosome segregation. To this end, the distribution of DNA inside the cell was monitored by a fluorescent fusion of venus to the hupA subunit of the general nucleoid-associated HU protein. Results indicate that chromosome segregation is not prevented by ObgE^∗ (Figure 6C). One hour after transfer from stationary phase to fresh medium, chromosome segregation has occurred in approximately 42% of all cells. After this point, segregated nucleoids slowly start to reunite and spread throughout the cell (Figure 6D). However, even cells with segregated nucleoids fail to divide. This experiment demonstrates that it is not a defect in chromosome segregation per se that inhibits further progression through the cell cycle. We thus conclude that ObgE^∗ interferes with another stage of the cell cycle. This point of interference is located after completion of chromosome segregation and inhibits the onset of cell division.

Experiments were performed with E. coli K-12 BW25113, except for data shown in Supplementary Figure S3. Single-gene knock-out strains were obtained from the Keio collection (Baba et al., 2006). The E. coli BW25113 ΔamiABC strain was constructed by sequentially replacing the amiA, amiB, and amiC genes with a kanamycin resistance cassette by phage transduction from HSC085, HSC078, and HSC071, respectively (Chung et al., 2009). After replacement of the amiA and amiB genes, the resistance cassette was removed by FLP-mediated recombination prior to the subsequent round of gene deletion. Knock-out of the correct genes was confirmed by PCR. E. coli BW25113 hupA-venus was constructed by phage transduction starting from E. coli MG1655 hupA-venus-Cm^R (Abram Aertsen, KU Leuven, personal communication). The chloramphenicol resistance cassette was excised by FLP-mediated recombination.

For all tests, unless indicated otherwise, overnight cultures were diluted 100 times in LB containing the appropriate antibiotics (ampicillin 100 μg/ml, chloramphenicol 35 μg/ml, gentamicin 25 μg/ml, and kanamycin 40 μg/ml) and incubated at 37°C with continuous shaking at 200 rpm. When the OD_{595 nm} reached 0.4, expression from pBAD/His A (Figure 2A), pBAD33Gm (Figures 6C,D, for construction see supplemental experimental procedures) or pBAD33 (all other figures) was induced with arabinose (0.2% w/v). Expression of GFP from pQE80L-gfp (Orman and Brynildsen, 2015) was induced by IPTG (1 mM) at the same time. An alternative protocol was followed to investigate the effect of cell division on toxicity; induction was either postponed until one of two possible points in stationary phase (10 or 16 h after dilution of the overnight culture), carried out in M9 medium with casamino acids (1% w/v) as carbon source, or simultaneously aztreonam (0.2 μg/ml), cephalexin (50 μg/ml) or mecillinam (0.64 μg/ml) were added to the cultures. The concentration of these antibiotics is situated between the minimal inhibitory concentration (MIC) and minimal bactericidal concentration (MBC) (Supplementary Table S1), so as to inhibit cell growth without causing cell death.

Serial dilutions were prepared in 10 mM MgSO₄ and plated on medium containing 1.5 % agar. After overnight incubation at 37°C, CFUs were determined and, if needed, the percentage viability was calculated by dividing the number of CFUs per ml obtained after ObgE^∗ expression by the number of CFUs per ml after ObgE expression.

At indicated time points or – when no time point was specified – 6 h after induction with arabinose, cultures were diluted 10 (exponential-phase cultures) or 100 (stationary-phase cultures) times in PBS containing 0.03 μM PI and incubated in the dark at room temperature for 15 min. The fraction of the population that stained PI-positive was measured by flow cytometry with a BD Influx cell sorter equipped with 488 and 561 nm lasers and standard filter sets. Flow cytometry experiments were repeated n times. In each repeat, 100,000 cells were collected.

To determine the intracellular ObgE^∗ concentration in several conditions, a translational ObgE^∗-Venus fusion (Dewachter et al., 2016a) was expressed and Venus fluorescence was measured by flow cytometry. Presence of the Venus tag did not influence toxicity in any way (data not shown).

To simultaneously visualize blebs, cytoplasm, membranes and peptidoglycan (Figure 2B), 0.5 mM HADA was added to the cultures at the time of induction with arabinose. At this time GFP expression was also induced by adding 1 mM IPTG. Two hours later, cultures were washed twice with PBS to remove unincorporated HADA and pellets were dissolved in LB medium containing the appropriate antibiotics, 0.2% arabinose, 1 mM IPTG, and 100 mM MgSO₄ to stabilize blebs. Cultures were incubated for another 2 h to allow bleb formation and were then stained with 10 μg/ml FM4-64. After 10 min incubation in the dark at room temperature cells were visualized on a poly-L-lysine coated glass slide.

Visualization of membrane blebs in E. coli pBAD33-obgE^∗ was performed by adding 100 mM MgSO₄ at the time of induction and staining cells with 10 μg/ml FM4-64 4 h later (Figure 3A). MgSO₄ was added to stabilize blebs and thus increase their lifetime, as was done previously (Yao et al., 2012).

For the time lapse microscopy experiment showing PI staining (Figure 1B), an overnight culture of E. coli pBAD33-obgE^∗ was seeded on an LB agarose pad (2% w/v) containing the appropriate antibiotic, 0.2% arabinose and 0.06 μM PI. Cells were incubated at 37°C and growth was monitored for 12 h.

The time lapse microscopy experiment with a cytoplasmic GFP label (Figure 2D) was performed with E. coli pBAD33-obgE^∗ pQE80L-gfp. An overnight culture of this strain was seeded on an LB agarose pad containing the appropriate antibiotics, 0.2% arabinose and 1 mM IPTG. Cells were incubated at 37°C and growth was monitored for 12 h.

Resumption of growth and chromosome segregation after ObgE or ObgE^∗ expression in stationary phase (Figures 6A,C) was monitored in a time lapse experiment where overnight cultures of E. coli pBAD33-obgE and E. coli pBAD33-obgE^∗ or E. coli hupA-venus pBAD33Gm-obgE^∗ were diluted and grown for 16 h into stationary phase. Expression was then induced by adding 0.2% arabinose for 2 h after which cells were seeded on LB agarose pads containing the appropriate antibiotics without arabinose. Cells were incubated at 37°C and growth was monitored for 12–48 h. Cell length or segregation of fluorescent foci were quantified by MicrobeJ (Ducret et al., 2016).

For all microscopy experiments, cells were imaged by a Nikon Ti-E inverted microscope with Qi2 CMOS camera and temperature controlled cage incubator.

Scanning electron microscopy of E. coli WM2949 ΔrecA pBAD/His A-obgE^∗ was performed as described previously (Dewachter et al., 2015).

Focused ion beam-scanning electron microscopy was performed as described previously (Vanwalleghem et al., 2015).

All statistical analyses were performed with GraphPad Prism 6. Normality of representative data from CFU counts (n = 30) and PI staining (n = 15) was verified by D’Agostino and Pearson omnibus normality test. The different samples were compared using one-way ANOVA with p-values obtained from Bonferroni’s multiple comparisons test.