The strains and plasmids used in this study were listed (Table 1). The WT E. coli K-12 strain BW25113 and its derivatives were obtained from the National BioResource Project [National Institute of Genetics (NIG), Mishima, Japan; Baba et al., 2006]. The double-gene knockout mutants were constructed by P1 transduction using the P1kc phage (Ojima et al., 2020). Each strain was transformed with pCA24N-gfp (Ojima et al., 2018) to express the His-tagged GFP in the cytosol.

The E. coli cells were cultured in lysogeny broth (LB; 10 g/L Bacto™ Tryptone, 5 g/L yeast extract, and 10 g/L NaCl). The culture medium for the strains harboring pCA24N-gfp was supplemented with 50 mg/L chloramphenicol and 1 mM IPTG. All the test cultures were precultured in LB for 18 h at 37°C and inoculated into 100 ml of fresh LB in a 500 ml baffled conical flask to achieve the optical density at 660 nm (OD₆₆₀) of 0.01. The cultures were incubated on an NR-30 rotary shaker (Taitec, Osaka, Japan) at 140 strokes/min. Cell growth was measured at OD₆₆₀.

Cells of each strain were harvested at 3 and 24 h after inoculation by centrifugation at 10,000 × g and 4°C for 10 min. Cells were washed by filtrated PBS (pH7.5) and resuspended in PBS. Cell size distribution analysis was conducted using the qNano system (qNano, IZON Science Ltd., Christchurch, New Zealand) equipped with Nanopore NP-1000. Scanning ion occlusion sensing allows single-particle measurements since colloids or biomolecular analytes or both are driven through pores one at a time. Particles crossing the nanopore are detected as a transient change in the ionic current flow, also denoted as a blockade event whose amplitude is the blockade magnitude. As blockade magnitude is proportional to particle size, accurate particle sizing can be achieved after calibration with a known standard. Here, size calibration was conducted using CPC1000 standard particles.

For QFDE-EM analysis, bacterial cells were first washed with PBS (pH 7.5) twice, resuspended in HEPES-NaCl buffer or 15% (v/v) glycerol solution, and centrifuged (Tulum et al., 2019; Takaki et al., 2020). Then, a rabbit lung slab, mica flakes, and bacterial cell pellets were placed in this order onto a paper disk attached to an aluminum disk, and the samples were quickly frozen using liquid helium with a CryoPress (Valiant Instruments, St. Louis, MO, United States). The rabbit lung slab and the mica flakes function as shock absorber in quick freezing and flat background in observation, respectively. The specimens were stored in a chamber at −180°C using a JFDV freeze-etching device (JEOL, Tokyo, Japan). After the samples’ temperature was increased to −120°C, they were freeze-fractured with a knife and freeze-etched at −104°C for 15 min. The freeze-etched step was omitted when 15% (v/v) glycerol was used as the solvent. Subsequently, the samples were coated with platinum at a thickness of 2 nm and a rotary shadowing angle of 20° and then coated with carbon at a rotary shadowing angle of 80°. Next, the replicas were floated on full-strength hydrofluoric acid, rinsed in water, cleansed with commercial bleach containing sodium hypochlorite, rinsed in water, and finally placed onto 400-mesh Cu grids. The replica specimens were observed with transmission electron microscopy (TEM) using a JEM-1010 (JEOL).

Outer membrane vesicles were isolated as previously described (Gujrati et al., 2014) with some modifications. After incubation for 24 h, 100 ml of the E. coli culture was centrifuged at 10,000 × g for 10 min at 4°C to remove the cells. Then, the supernatant was passed through a 0.45 μm filter. Ammonium sulfate was added at the final concentration of 400 g/L to incubate for 1 h at room temperature to precipitate the contents. Crude OMVs were obtained via centrifugation at 11,000 × g for 30 min at 4°C. The crude extracts were dissolved in 500 μl of 15% (v/v) glycerol and concentrated using a CS100FNX ultracentrifuge (Hitachi Koki Co., Tokyo, Japan) at 109,000 × g for 1 h. The OMV pellets were resuspended in 100 μl of 15% (v/v) glycerol. The resulting OMV samples were 1,000 times more concentrated than that in the original culture because of the volume decreasing from 100 ml to 100 μl. The OMV samples were placed onto a 200-mesh copper grid and negatively stained with 4% uranyl acetate for TEM observation under a JEM-2100 (JEOL).

Ten microliters of the isolated OMVs or a culture of E. coli cells was analyzed via sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and Coomassie Blue staining. OMV production was quantified as previously described (Schwechheimer and Kuehn, 2013) with some modifications. The OMV concentration was measured by quantifying the band at approximately 37 kDa in the SDS-PAGE gel using the Image J software (National Institutes of Health, Bethesda, MD, United States). The OMV production of each mutant was shown as relative value to WT.

Outer membrane vesicles were also quantified with a lipophilic dye FM4-64 according to a previously described method with minor modifications (McBroom et al., 2006; Manning and Kuehn, 2011; Klimentová and Stulík, 2015). Isolated OMVs were incubated with 5 μg/ml FM4-64 (Molecular Probes/Thermo Fisher, Waltham, MA, United States) in PBS (pH 7.5) for 20 min. Then, OMVs were measured at the excitation and emission wavelengths of 558 and 734 nm, respectively, using an INFINITE 200 PRO spectrofluorophotometer (TECAN, Switzerland). The OMV sample without staining by a lipophilic dye (FM4-64) or the lipophilic dye alone was used as the negative controls.

The His-tagged GFP included in OMV samples was analyzed via western blotting by first transferring the protein from a gel to an Immobilon-P membrane sheet (Merck Millipore, Billerica, MA, United States) using the semidry transfer method. The blot was hybridized with an anti-histidine-tag primary monoclonal antibody (Medical & Biological Laboratories Co., Nagoya, Japan) and then a secondary antibody, the anti-mouse immunoglobulin G (whole molecule)–alkaline phosphatase (Sigma-Aldrich, St. Louis, MO, United States). The hybridization signals were detected using a BCIP-NBT Solution Kit for Alkaline Phosphatase Stain (Nacalai Tesque, Kyoto, Japan). The level of the target protein was quantified by analyzing the bands on the western blot band using densitometry. GFP secretion was quantified on the basis of the intensity of a GFP standard sample (50 μg/L) with a His-tag (Sino Biological Inc., Beijing, China). The GFP produced in each E. coli cell was also analyzed via western blotting. In the case of cell analysis, a 10 μl sample at OD₆₆₀ = 0.3 was loaded into each well of gel to include the same amount of cells.

The GFP-derived fluorescence of the OMV samples was observed at an excitation wavelength of 488 nm using a confocal laser scanning microscope (CLSM; DM6000B, Leica, Wetzlar, Germany) with the TCS SP8 software.

Outer membrane vesicles isolated by ultracentrifugation as described above were resuspended in PBS buffer (pH 7.5). After dilution of OMV samples with pure water, dynamic light scattering (DLS) measurements were conducted at 25°C using a ZetaSizer NanoSeries equipped with a HeNe laser source (λ = 633 nm; Malvern Instruments Ltd., Worcestershire, United Kingdom) and analyzed using the Dispersion Technology Software (Malvern Instruments Ltd.). For each sample, the autocorrelation function was the average of five 10 s runs and then repeated approximately 3–6 times. CONTIN analysis was subsequently used for the number vs. hydrodynamic size profiles to study the dispersions.

After 24 h in culture, the E. coli cells were collected via centrifugation at 10,000 × g and 4°C for 10 min and suspended in PBS buffer (pH 7.5). These steps were repeated once. Then, the cells were resuspended in 10% SDS (w/v) and incubated at 95°C for 12 h. The sacculi were harvested via centrifugation at 200,000 × g and 20°C for 40 min, washed three times in Milli-Q water, and observed using TEM with negative staining or QFDE-EM.

The cell growth, cell volume, relative OMV production, and average OMV size of each strain of hypervesiculating E. coli, including the single-deletion mutants ΔmlaE and ΔnlpI and the double-deletion mutant ΔmlaEΔnlpI at the end of a 24 h culture, were measured (Table 2). The OD₆₆₀ and relative OMV production based on the SDS-PAGE analysis were taken from our previous study (Ojima et al., 2020).

In this study, OMV production was quantified with a lipophilic dye, FM4-64. The relative OMV production by ΔnlpI and ΔmlaE strains assayed with FM4-64 was 5.8 and 8.7 times, respectively, that of the WT strain (Table 2), consistent with the results from the SDS-PAGE analysis in our previous study. Meanwhile, the OMV production by ΔmlaEΔnlpI strain was determined to be approximately 14 times higher than that of the WT strain, in contrast to the 30-fold difference revealed via the SDS-PAGE in the previous study. Although there was a deviation depending on the measurement method, it was again confirmed that the OMV production was significantly increased in ΔmlaEΔnlpI.

Next, the diameter of the OMVs in each strain was measured via DLS, and the size distribution of the OMVs was plotted (Figure 1). The WT strain had a normal size distribution, with a peak value of approximately 8% at approximately 90 nm; this distribution agreed with previous research (Alves et al., 2016). The average OMV size was 83.9 ± 0.3 nm (Table 2). The values are the means of average values from three independent experiments. The distribution of the OMVs of ΔnlpI strain almost overlapped with that of WT strain; the average size of ΔnlpI OMVs was 81.7 ± 0.3 nm, comparable with that of the WT strain. Meanwhile, the distribution of the ΔmlaE OMVs shifted to a smaller value, with a peak value of 8% at approximately 60 nm. The average OMV size of ΔmlaE was 65.9 ± 0.6 nm. These data suggested that ΔmlaE produced smaller OMVs. By contrast, the distribution of ΔmlaEΔnlpI shifted to a larger value, with the peak value of 10% at approximately 100 nm. Furthermore, the plot of ΔmlaEΔnlpI strain was not a normal distribution; it had a shoulder at approximately 70 nm. The average OMV size was 100.4 ± 0.5 nm, significantly larger than that of WT strain. Therefore, the size distributions of the OMVs differ between the knockout mutant strains.

The cell volume of each strain was measured at 3 and 24 h postinoculation using qNano and plotted in histograms (Supplementary Figure S1). At 3 h, the plot of ΔnlpI shifted to a smaller value than the others. The cell volume of the WT strain was 1.08 ± 0.27 fL, consistent with that observed in E. coli cells, at 1.16 fL, in a previous study (Yamada et al., 2017). The cell volumes of ΔmlaE and ΔmlaEΔnlpI strains were 1.04 ± 0.27 and 1.08 ± 0.28 fL, respectively. The deletion of these genes likely did not affect the cell volume in the growth phase. By contrast, the cell size of ΔnlpI was slightly smaller than the others at 0.85 ± 0.26 fL.

At 24 h, all the cell volume plots also followed a normal distribution; however, the plots of the mutant strains shifted to a smaller value compared with that of the WT strain. The cell volume of WT was 0.52 ± 0.18 fL, smaller than its volume at 3 h, consistent with a previous report that the cell volume of E. coli decreased to approximately 0.5 fL in a medium with poor nutrients (Kubitschek, 1990). Thus, it is reasonable that the cell size decreases in the stationary phase. The cell volumes of ΔnlpI, ΔmlaE, and ΔmlaEΔnlpI were 0.39 ± 0.15, 0.41 ± 0.14, and 0.39 ± 0.20 fL, respectively. Therefore, the cell volume was significantly smaller in the mutant strains at the end of the culture.

Since the cell size changed in the hypervesiculating strains, there might be a major change in the appearance of the cells. QFDE-EM is a powerful tool for investigating the spatial structure of bacterial envelopment (Tulum et al., 2019); it has been applied to analyze the biogenesis of OMV in B. agrestis (Takaki et al., 2020). Here, the cell structure of each E. coli strain was visualized using QFDE-EM. The cells at 24 h were collected, centrifuged, placed on glass, frozen, fractured, deeply etched, and shadowed with platinum. Then, the platinum replicas were recovered and observed.

The shape and dimensions of the WT cells were consistent with the images of living WT cells obtained by optical microscopy (Figures 2A,C). The surface of the WT cells was observed using QFDE-EM (Figures 2C,D). The ΔmlaEΔnlpI cells appeared longer than the WT cells (Figures 2B,E,F); furthermore, several mutant cells were observed to form vesicles (Figure 2E). The magnified images demonstrated that the ΔmlaEΔnlpI cells generated OMVs from the tip of their long axis; the diameters of the OMVs were approximately 400 (Figure 2G) and 200 nm (Figure 2H). The size of these OMVs is relatively larger but in the range of the size distribution (Figure 1).

The intracellular compartments, i.e., outer membrane, inner membrane, and cytoplasm (CP), of the WT cells were visualized using the freeze-fractured sections without the freeze-etched step (Figures 3A,B). The ΔmlaEΔnlpI cells were elongated, and large periplasmic spaces were observed in most cells (Figures 3C,D). The magnified images revealed that plasmolysis was induced in these mutant cells (Figures 3E,F). Plasmolysis is the shrinking of protoplasm away from the cell wall of a plant or bacterium (Woldringh, 1994; Lang et al., 2014). The protoplasmic shrinking is often due to water loss via exosmosis, thereby resulting in gaps between the cell wall and the plasma membrane. Clear plasmolysis was observed in approximately 50% or more of the ΔmlaEΔnlpI cells. Meanwhile, plasmolysis was also observed in the single-deletion mutants ΔnlpI and ΔmlaE, although less frequently than in the double-deletion mutant (Figures 3G,H). These results indicate that plasmolysis is a key factor for OMV production.

Further observation revealed that the tip of the plasmolyzed ΔmlaEΔnlpI cells was elongated, and OMVs were generated (Figures 4A,B). Since this phenomenon was frequently confirmed in many other cells (Figures 4C,D), it may be the main route of OMV production in the ΔmlaEΔnlpI cells. Interestingly, intracellular vesicles in the enlarged periplasmic space were observed in the ΔmlaEΔnlpI cells (Figure 4D). In the left cell of Figure 4D; the intracellular vesicle appeared to be produced by the destruction of the inner membrane. By contrast, these intracellular vesicles were observed at an exceptionally low frequency in the ΔnlpI cells (Figure 4E), and never observed in the WT and ΔmlaE cells. Even more interesting, multilamellar OMVs were observed in the ΔmlaEΔnlpI cells (Figure 4F). To further confirm the existence of multilamellar OMVs, the OMVs were isolated by ultracentrifugation and negatively stained with uranyl acetate (Figure 5). The multilamellar OMVs were observed in the ΔmlaEΔnlpI sample (Figure 5B). In addition, the multilamellar OMVs were observed in ΔnlpI samples at an extremely low frequency (Figure 5C). By contrast, multilamellar OMVs were not observed in the WT and ΔmlaE cells (Figures 5A,D). These results are consistent with those from QFDE-EM that intracellular vesicles were observed in the ΔmlaEΔnlpI cells, at a low frequency in the ΔnlpI, and never observed in the WT and ΔmlaE cells.

To investigate the possibility that the multilamellar OMVs in ΔmlaEΔnlpI are OIMVs, the presence of cytosolic components in the multilamellar OMVs was examined by expressing recombinant GFP protein in the cytosol of each E. coli strain. In this case, recombinant GFP is not normally incorporated into OMVs because of its lack of a signal peptide. The OD₆₆₀ values of the WT, ΔnlpI, ΔmlaE, and ΔmlaEΔnlpI strains expressing the pCA24N-gfp plasmid were measured (Figure 6A). The OD₆₆₀ values of ΔnlpI and ΔmlaE were 4.6 ± 0.2 and 4.5 ± 0.1, respectively, lower than that of the WT strain at 6.0 ± 0.2. ΔmlaEΔnlpI mutant had the lowest OD₆₆₀ at 2.7 ± 0.2.

We also estimated OMV production on the basis of the densitometric analysis of the protein band at approximately 37 kDa that included OmpC, F combined with OmpA (Figures 6B,C). Although the OMV production by ΔnlpI/gfp and ΔmlaE/gfp strains was 5.7 ± 0.6 and 5.1 ± 1.0 times, respectively, that of the WT/gfp strain, the OMV production by ΔmlaEΔnlpI/gfp strain was found to be 55.8 ± 8.3 times that of the WT strain. These results suggested that introducing the pCA24N-gfp plasmid did not substantially affect the OMV production.

Figure 6D shows the results of western blots of GFP in the OMV fraction. Here, the OMV fraction was first collected as ammonium sulfate precipitate before ultracentrifugation step, which could have also precipitated soluble GFP from the culture supernatants if there were damaged cells and may not necessarily mean that the GFP is located inside the vesicles. However, in the preliminary experiments, the viability for each strain was similar; therefore, the E. coli cells were considered to be intact. Additionally, the OMV fraction was washed at ultracentrifugation step. These facts together indicate that the possibility of coprecipitation of soluble GFP is low. As the result, the bands appeared at the expected molecular mass of GFP protein at approximately 27 kDa, except for the WT strain (Figure 6D), suggesting that the OMV fraction from the WT strain did not contain detectable GFP. In the ΔnlpI/gfp and ΔmlaE/gfp strains, the same band appeared. This band was much more intense in ΔmlaEΔnlpI/gfp strain, even after a 1:100 dilution. The amount of GFP secreted through OMVs in the culture medium was quantified on the basis of the densitometry of the band in the western blot against the commercial GFP with a known concentration. Here, the commercial GFP band migrated at approximately 34 kDa. GFP was not detectable in the OMV fraction of the WT/gfp strain (Figure 6E). Since the recombinant GFP was produced in the cytosol, it is reasonable that the WT did not secrete GFP via OMV. Conversely, GFP secretion by ΔnlpI and ΔmlaE strains was 14.3 ± 6.8 and 36.5 ± 21.3 μg/L, respectively. The double-deletion mutant ΔmlaEΔnlpI/gfp secreted GFP at 1.8 ± 0.1 mg/L in the medium, significantly more than other strains. Considering that the secretion of GFP by WT was not detectable and less than that of ΔnlpI (14.3 ± 6.8 μg/L), the GFP secretion of ΔmlaEΔnlpI is more than 100 times higher than that of WT or ΔnlpI, and about 50 times higher than that of ΔmlaE. The OMV samples of WT/gfp and ΔmlaEΔnlpI/gfp cultures were observed using CLSM (Supplementary Figure S2). The OMV sample of ΔmlaEΔnlpI/gfp culture was much more fluorescent, consistent with the results from the western blots.

Because of the extremely high secretion of GFP by the ΔmlaEΔnlpI/gfp strain, we hypothesized that besides enhancing the OMV production, the deletion of both mlaE and nlpI genes increased the level of GFP in the E. coli cell. The same amount of whole-cell lysates was analyzed in western blotting to compare the GFP production in different cells (Figure 6F). The bands were similarly intense among the strains, suggesting that GFP production per cell unit was comparable. These results indicate that the increased amount of GFP in the OMV fraction of the ΔmlaEΔnlpI mutant was due to a promoted secretion of GFP rather than a high expression level in the cells.

The existence of intracellular and multilamellar vesicles in the ΔmlaEΔnlpI cells was confirmed. Also, the secretion of an unusual amount of cytosolic GFP into the medium by the ΔmlaEΔnlpI cells was verified. These results suggested the generation of a hole in the PG layer might allow the release of the inner membrane with cytosolic components, followed by the formation of intracellular and multilamellar vesicles.

Thus, the sacculi of each E. coli strain were prepared according to previous research (Vollmer et al., 2008; Jeske et al., 2015). After negative staining with uranyl acetate, the thin sacculi of WT appeared as flat, empty cell envelopes (Figure 7A), as reported previously (Vollmer et al., 2008). The WT sacculi were further observed using QFDE-EM (Figures 7B–F). The surface of the typical WT sacculi appeared remarkably smooth (Figures 7B,C,E). The magnified images confirmed the roughness of the surface of the PG network (Figures 7D,F); however, no obvious holes were observed.

Sacculi were also successfully prepared from the ΔmlaEΔnlpI cells (Figures 8A,B). The sacculi of the double-deletion mutant were longer on the long axis. The magnified QFDE-EM images showed that the ΔmlaEΔnlpI sacculi had several holes at the tip of the long axis (Figure 8C). The diameter of the largest hole was more than 50 nm (Figure 8D), substantially larger than the mean radius of the pores for E. coli PG, at 2.06 nm, in a previous report (Demchick and Koch, 1996). The holes at the tip of the long axis in the sacculi were confirmed in many other cells (Figure 8E). A magnified image revealed that although the sacculi appeared intact at lower magnifications, they had smaller holes that were noticeably larger than the pores for E. coli PG (Figure 8F). Such smaller holes were not observed in the WT sacculi. These results suggested that the generation of the holes in the PG layer of the ΔmlaEΔnlpI cells allowed the release of the inner membrane, followed by the intracellular and multilamellar vesicles.