The microbial strains, plasmids, and primers used in this study are listed in Table 1. Microbial cells were grown in tryptic soy broth (TSB: Becton, Dickinson and Company, Franklin Lakes, NJ, USA) medium shaken at 200 rpm. Buttiauxella spp., Corynebacterium glutamicum, Micrococcus luteus, Flavobacterium johnsoniae, Rhizobium spp., and Pseudomonas alcaligenes were grown at 30°C. Bacillus subtilis, Hydrogenophaga pseudoflava, E. coli, Erwinia persicina, and P. aeruginosa were grown at 37°C. For genetic manipulations, LB (Luria-Bertani Lennox: 1% w/v tryptone, 0.5% yeast extract and 0.5% NaCl) was used. When necessary, chloramphenicol was used at a concentration of 20 μg/mL for E. coli and B. agrestis. E. coli β2163 was used for conjugation to transfer the plasmid pBBR1MCS-1 to B. agrestis. DAPA (2,6-diaminopimelic acid) was used at a concentration of 60 μg/mL in LB medium for the growth of E. coli β2163.

Membrane vesicle extraction and purification were carried out as previously described (Tashiro et al., 2009). To obtain a sample of MVs without bacterial cells, 100 mL of overnight batch culture was centrifuged for 15 min at 6,000 × g, 4°C. The supernatant was filtered through 0.45 and 0.20 μm membrane filters and ultracentrifuged for 2 h at 100,000 × g and 4°C using an angle rotor (P45AT, Hitachi, Tokyo, Japan). The pellets were resuspended in 200 μL of 50 mM HEPES-0.85% NaCl (HEPES-NaCl buffer). For MV purification, MVs were labeled with 100 μg/mL FM4-64 in HEPES-NaCl buffer and washed using ultracentrifugation. MV samples were adjusted to 1 mL of 45% (w/v) iodixanol (OptiPrep; Axis-Shield Diagnostics Ltd., Dundee, UK) in HEPES-NaCl, transferred to the bottom of ultracentrifuge tubes, and layered with iodixanol-HEPES-NaCl (2 mL of 40, 35, 30, 25, and 20%). The samples were ultracentrifuged for 3 h at 100,000 × g and 4°C using a swing rotor (P40ST, Hitachi). Then, 500 μL fractions were collected from each gradient. To confirm the fraction containing MVs, the phospholipid and protein concentrations in each fraction were measured using the Stewart assay and the bicinchoninic acid (BCA) protein assay. The fraction containing MVs was ultracentrifuged and resuspended in HEPES-NaCl.

Membrane vesicle production was quantified by determining the phospholipid concentration in the bacterial culture supernatant. The phospholipid concentration was measured using a previously reported method with some modifications (Stewart, 1980). Briefly, 1 mL of the supernatant filtered through a 0.20-μm membrane, 200 μL of concentrated ammonium ferrothiocyanate solution (135 g/L iron (III) chloride hexahydrate and 152 g/L ammonium thiocyanate), and 200 μL of chloroform were vortexed. The absorption of the chloroform layer (lower layer) was measured at 488 nm. To construct a calibration curve, L-α-phosphatidylethanolamine was used as a reference standard. MV production was normalized to the quantity of cell protein. Bacterial cells were lysed by 5% SDS, and the protein concentration was determined by a BCA protein assay (Pierce BCA Protein Assay Kit; Thermo, Rockford, IL, USA).

The MV association degree was determined by the relative fluorescence units (RFUs) of fluorescently labeled MVs per concentration of cell protein. The extracted MVs (200 μg phospholipid) were incubated with 5 μg/mL FM4-64 in 1 mL of phosphate-buffered saline (PBS) for 30 min at 30°C and washed three times in PBS. In some cases, 100 μg/mL fluorescein isothiocyanate (FITC) or 5 μM DiO was used instead of FM4-64. Bacterial cells grown to the late exponential phase were washed in PBS and diluted to an OD₆₀₀ of 1.0. FM4-64-labeled MVs (20 μg of phospholipid) were incubated with 1 mL of bacterial cells (OD₆₀₀ of 1.0) for 30 min at 30°C. Bacterial cells with FM4-64-labeled MVs were washed in PBS. The fluorescence of FM4-64 was measured using a microplate reader (Infinite M200; TECAN, Männedorf, Switzerland), and the cell protein concentration was measured by a BCA protein assay.

Flow cytometry analysis using an EPICS Altra AJ49030 system (Beckman Coulter, Brea, CA, USA) was performed to detect cells associated with fluorescence-labeled MVs. MVs (200 μg of phospholipid) were labeled with FM4-64 and washed in PBS by ultracentrifugation. Bacterial cells were adjusted to OD₆₀₀ = 1.0 in PBS, labeled with 100 μg/mL FITC for 30 min at 30°C and washed three times in PBS. FM4-64-labeled MVs and FITC-labeled bacterial cells were incubated for 30 min at 30°C and washed with PBS three times. The particle number and fluorescence intensity of bacterial cells associated with MVs were analyzed by flow cytometry.

For fluorescence microscopy analysis, the cells were examined using the fluorescence microscope OLYMPUS BX53 (OLYMPUS, Tokyo, Japan), and images were taken with the CCD camera DP72 (OLYMPUS) and the imaging software CellSens (OLYMPUS). Bacterial cells were adjusted to OD₆₀₀ = 1.0 in PBS, labeled with 100 μg/mL FITC for 30 min at 30°C and washed three times in PBS. FM4-64-labeled MVs and FITC-labeled bacterial cells were incubated for 30 min at 30°C and washed with PBS three times.

For transmission electron microscopy observations of purified MVs, MVs were placed on Cu400 mesh grids, stained with 2% uranyl acetate and visualized by JEM100EX (JEOL, Tokyo, Japan) by the Hanaichi Ultrastructure Research Institute (Okazaki, Aichi, Japan). For other TEM analyses, the samples were placed on Cu200 mesh grids that had been treated with 0.01% poly-L-lysine in PBS. The grids were rinsed with distilled water and stained with 2% phosphotungstic acid neutralized with KOH. After rinsing with distilled water and dried, the samples were visualized by JEM 2000FX-II (JEOL).

The extracted MVs were labeled in 100 μg/mL FITC and washed with PBS by ultracentrifugation. Bacterial cells were adjusted to OD₆₀₀ = 1.0 in PBS, incubated with FITC-labeled MVs for 30 min at 30°C and washed with PBS three times. The samples were placed on Cu200 mesh grids that had been treated with 0.01% poly-L-lysine in PBS. The bacteria were fixed with 4% paraformaldehyde in PBS for 30 min and washed with PBS. The specimens were incubated with 1% bovine serum albumin (BSA) in PBS for 5 min and washed again with PBS. The specimens were then incubated with anti-FITC antibody (Sigma–Aldrich) at a dilution of 1:300 in PBS containing 1% BSA for 60 min. The specimens were washed with PBS, incubated with colloidal gold-labeled anti-rabbit immunoglobulin (diameter of colloidal gold particles, 15 nm) at a dilution of 1:1000 in PBS containing 1% BSA for 60 min, and then washed. The grids were fixed with 2% glutaraldehyde in PBS for 15 min and washed with PBS. After being stained with 2% phosphotungstic acid and washed, the samples were observed by JEM 2000FX-II.

The particle size distribution and zeta potential were analyzed with a Zetasizer Nano ZS particle analyzer (Malvern Instruments, Malvern, UK) at 25°C. Purified MVs were adjusted to a concentration of 20 μg of phospholipid/mL in PBS. Cells grown to the stationary phase in TSB were washed with PBS and adjusted to OD₆₀₀ = 0.1. The hydrodynamic zeta average diameter was calculated by the light scattering method, and the zeta potential was determined by applying the Smoluchowski approximation. The electrostatic charge on the surfaces of MVs and cells was characterized as the zeta potential.

The potential interaction energy between cells and vesicles was calculated based on the Derjaguin–Landau–Verwey–Overbeek (DLVO) theory. The interaction energy V_total was defined as:

where V_A is an attractive London-van der Waals interaction, and V_R is an electric repulsive interaction energy. Based on previous studies (Ohki and Ohshima, 1999), V_A was defined as follows:

where

where A is the Hamaker constant; a and a′ are the radii of the cells and vesicles, respectively; R is the separation distance between the center of the particles of cells and vesicles; and r is their separation distance. Hamaker constant A is described as follows:

where q is the volume density, and λ is London-van der Waals constant. A of the phospholipid bilayer was expressed as 4 × 10^-19 N (Ohki et al., 1982).

According to a previous report (Ohki and Ohshima, 1999), V_R was defined as follows:

where φ and φ′ are the Stern potentials of cells and vesicles, respectively, and κ is the Debye constant. The values of the Stern potentials were used as zeta potentials (ζ) obtained by the experiments. When the absolute values of φ and φ′ were less than 60 mV (κr > 1), V_R was approximated as follows:

where 𝜀_r is the relative permittivity for the medium, and 𝜀₀ is the permittivity of a vacuum. The Debye constant κ was expressed by the following:

where e is the elementary charge, N_A is Avogadro’s number, c is the ion concentration, z is the charge of ions and k is the Boltzmann constant.

Experiments for vesicle-mediated gene transfer were based on a previously published method (Yaron et al., 2000). The donor strain, B. agrestis CUETM77-167/pBBR1MCS-1, was grown in LB medium containing chloramphenicol at 200 rpm for 16 h at 30°C. MVs were extracted using sterile ultracentrifuge tubes after filtering the supernatant. Extracted MVs were reacted with DNase I in reaction buffer [40 mM Tris-HCl (pH 7.9), 10 mM NaCl, 6 mM MgCl₂, and 1 mM CaCl₂], and the extracellular DNA around the MVs was degraded. Then, the MVs were washed with PBS by ultracentrifugation. MVs (20 μg of phospholipids) and approximately 1,000 recipient bacterial cells of B. agrestis were mixed in 1 mL of PBS, and the cells were washed with PBS at each time point. The transfer of plasmid DNA was examined by counting the colony forming units (CFUs) on agar medium containing chloramphenicol. As a control experiment, naked plasmid DNA, which was extracted from B. agrestis CUETM77-167/pBBR1MCS-1, was added to the cell suspension (final DNA concentration was 10 ng/mL), and the possibility of natural transformation was examined by CFU counting. PCR analysis with a primer pair (CamR-f and CamR-r, listed in Table 1), and plasmid extraction were conducted to confirm whether the obtained colonies harbored the plasmid pBBR1MCS-1.

The copy number of the plasmid was determined using real-time PCR analysis. The chloramphenicol acyl transferase gene in pBBR1MCS-1 was amplified and quantified using LightCycler FastStart DNA Master SYBR Green I (Roche, Basel, Switzerland) with a primer pair (CamR-f and CamR-r) on a LightCycler 2.0 (Roche) according to the instruction manual.

The concentration of MVs was measured by nanoparticle tracking analysis using a NanoSight LM10 instrument (Malvern) equipped with sCMOS camera (Andor, Belfast, UK). The samples were diluted to an average concentration of 10⁶–10⁹ particles per milliliter in 50 mM HEPES buffer. Five replicates videos were collected from each sample, and particle movement was analyzed by NTA software (Version 3.1, Malvern). The velocity of particle movement was used to calculate the particle size using the two-dimensional Stokes-Einstein equation.

Gentamicin-containing MVs (g-MVs) were prepared as previously described (Kadurugamuwa and Beveridge, 1995). Briefly, B. agrestis CUETM77-167 was grown in TSB medium until the late stationary phase, and gentamicin was added to a final concentration of 32 μg/mL [four times the minimal inhibitory concentration (MIC) of gentamicin for this strain]. After a further 30 min of cultivation, MVs were extracted from the bacterial culture. The amount of gentamicin in the MV suspension was determined by examining the bacterial killing effect of homogenized MVs. The concentration of gentamicin in MVs was calculated using the number of MVs in the MV suspension. To analyze the killing effect on a single bacterial species, MVs (20 μg of phospholipids) and approximately 1,000 bacterial cells were mixed in 1 mL of PBS, and the cells were washed with PBS at each time point. The surviving cells were counted by CFU. To analyze the killing effect in mixed bacterial culture, MVs (20 μg of phospholipids) and approximately 1,000 cells of each bacterial species (B. agrestis CUETM77-167 and E. coli DH5α) were mixed in 1 mL of PBS, and the cells were washed with PBS at each time point. The surviving cells were counted by CFU on an agar plate containing 200 μg/mL 5-bromo-4-chloro-indolyl β-D-galactopyranoside (X-gal) and 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). B. agrestis and E. coli DH5α showed blue and white colonies, respectively.

Vesicles associated with gentamicin were homogenized by an ultrasonic disintegrator. The suspended lysed vesicles containing gentamicin were diluted and added to LB agar on a 24-well plate before the agar had gelatinized. A cell suspension of E. coli MG1655 was grown on the plate, and the result was compared to that on an agar plate containing gentamicin. Using the MIC, the gentamicin concentration in the sample was determined. The vesicle particle number and average particle size were measured by nanoparticle tracking analysis, and the gentamicin concentration in the vesicle particle was calculated.

Sequence data were accessed from NCBI (GenBank accession numbers are AJ233400 to AJ233406 and NC_000913). 16S rRNA gene sequences were aligned using the ClustalW program, and distance matrix trees were constructed by the neighbor-joining method. The topology of the trees was evaluated by bootstrapping with 1,000 replicates, and the percentages of values are shown on branching points.

Error bars throughout the figures refer to the standard error for all of the experiments. The statistical significance of the differences between groups was examined using unpaired and one-way Student’s t-tests. A P-value of <0.005 was considered to indicate statistical significance.

To identify putative high MV-producing bacteria, we examined the phospholipid concentrations in the supernatant of 76 stationary cultures of microbial strains in laboratory stocks. We arbitrarily selected eleven bacterial strains, including C. glutamicum AJ2247, M. luteus JCM 1464, B. subtilis C1, F. johnsoniae JCM 8514, R. halotolerans JCM 17536, R. soli DS-42, H. pseudoflava GA3, B. agrestis CUETM77-167, E. persicina HK204, P. aeruginosa PAO1, and P. alcaligenes JCM 20561, which are supposed to release MVs in the culture at high levels compared to E. coli MG1655 (Supplementary Figure S1), for use in further analyses. Each supernatant from the bacterial culture was ultracentrifuged, and the pellet was labeled with the styryl dye FM4-64, a lipophilic styryl compound that stains the vacuolar membrane. To determine whether each putative MV selectively interacted with bacterial cells, MVs associated with bacterial cells were quantified by fluorescence detection (Supplementary Table S1). The interaction of MVs derived from P. aeruginosa with bacterial cells has been studied well (Kadurugamuwa and Beveridge, 1996, 1997; Li et al., 1998; Tashiro et al., 2010) and showed different levels of association based on bacterial species in this study (Figure 1A). Notably, MVs derived from B. agrestis CUETM77-167 interacted with the parent strain at a significant level but not with other cells (Figure 1B). This specific interaction was not due to the labeling of FM4-64 because similar results were obtained when MVs from B. agrestis were labeled with amino group-reactive FITC or carbocyanine membrane dye DiO (Supplementary Figure S2).

To further corroborate the specific interaction of MVs derived from B. agrestis with the same species, the association between MVs and bacterial cells was analyzed at the single-cell level. B. agrestis cells labeled with FITC were incubated with FM4-64-labeled MVs for 30 min, and washed cells were analyzed by flow cytometry. As a control, FITC-labeled cells, which were not treated with MVs, were also assessed by flow cytometry. The fluorescence intensities of both FM4-64 and FITC were detected in MV-reacted cells (approximately 1 × 10² RFU), while only the intensity of FITC was detected in the control sample (Figure 2A). Then, the same experiment was conducted using E. coli and MVs derived from B. agrestis. The fluorescence of FM4-64 was not detected in E. coli either with or without B. agrestis MVs (Figure 2B). Similar results were also obtained by fluorescence microscopy analysis; the fluorescence of both FM4-64 and FITC was observed in MV-treated B. agrestis (Figure 2C), but the fluorescence of FM4-64 was not observed in E. coli, which was treated identically (Figure 2D). These data confirm the hypothesis that MVs derived from B. agrestis specifically interact with bacterial cells of the same species.

To determine if bacterial viability is essential for MV-cell interactions, the associations of MVs with viable and dead cells were examined. Ultraviolet (UV)-treated cells showed a high association with B. agrestis MVs as well as non-UV treated cells (Supplementary Figure S3), suggesting that the bacterial viability of recipients is not required for the specific MV-cell interaction in B. agrestis.

We confirmed that the extract from the supernatant of B. agrestis culture from ultracentrifugation contained spherical MVs by TEM observation (Figure 3A). Next, we investigated if supplemented MVs interacted with bacterial cells using TEM observation. FM dyes have been used to determine the localization of supplemented vesicles because their photoactivation creates reactive oxygen species, which form an electron-dense precipitate in the vacuoles that is readily visualized under electron microscope (Harata et al., 2001). Then, FM4-64-labeled MVs were incubated with non-labeled cells of B. agrestis, and TEM analysis was conducted to observe the interaction of added MVs with B. agrestis cells. The result showed that MVs attached to the cellular surface were darkly stained (Figure 3B), suggesting that additional MVs were definitely associated with cells of B. agrestis. To further analyze MV-cell interaction, FITC-labeled MVs were incubated with non-labeled cells, and FITC was labeled by gold particles using a gold-conjugate antibody against FITC. The fusion of MVs onto bacterial cells was visualized, and MVs were labeled with gold particles (Figure 3C), suggesting that the MVs in the image were not those being released from cells but rather those being incorporated into cells. When MVs were not added to bacterial suspension before immunolabeling as the control, dense gold particles were not observed around cells (Figure 3D). Thus, the interaction of MVs with bacterial cells is not only the surface attachment but also the membranous fusion.

Because the specific interaction between MVs and cells occurred regardless of the viability of recipient cells, we hypothesized that the physicochemical properties of surfaces affect this association. When MVs and bacterial cells are assumed to be colloid particles, the interaction between two particles is explained by the DLVO theory (Hermansson, 1999; Ohki and Ohshima, 1999; Strevett and Chen, 2003). When the primary maximum energy is positive, there is an energy barrier between MVs and cells (Figure 4A). The interaction energy is the total of the van der Waals’ force and electric repulsion energy, and these depend on the particle size and zeta potential of the particles, respectively. To evaluate the interaction energy between MVs and bacterial cells, their surface thermodynamic properties were measured using a Zeta Sizer (Figure 4B). The interaction energy between B. agrestis MVs and each bacterial cell was calculated based on the DLVO theory (Supplementary Figure S4). Interestingly, the primary energy in the interaction of B. agrestis MVs with the same species was significantly lower than that of other bacterial cells (Figure 4C). These results suggest that one factor in the specific interaction of MVs with the same species in B. agrestis is the low interaction energy based on the DLVO theory.

Although a specific interaction of MVs with the same species was observed in B. agrestis, it was unknown whether this characteristic was confined to this bacterium. We prepared 6 other Buttiauxella strains (B. brennerae S1/6-571, B. ferragutiae CDC1180-81, B. noackiae NSW11, B. izardii S3/2-161, B. warmboldiae NSW326, and B. gaviniae S1/1-984), and the phylogenetic relationship among them and E. coli is shown in Figure 5A. Interestingly, the MVs derived from B. agrestis CUETM77-167 showed a much higher association with cells of Buttiauxella spp. than with E. coli (Figure 5B), suggesting that the specific interaction of MVs with cells is conserved in Buttiauxella spp. To determine whether the high interaction between MVs derived from CUETM77-167 and Buttiauxella strains is due to the physicochemical interaction energy, the zeta potentials and hydrodynamic diameters of each cell were examined. The calculated interaction energy based on the DLVO theory showed that the primary maximum energies of several Buttiauxella cells with MVs derived from CUETM77-167 were significantly lower than that for E. coli (Figure 5C and Supplementary Figure S5). We next evaluated the relationship between the primary maximum energies calculated by DLVO theory and the association values of MVs with bacterial cells. Figure 5D shows that plots of Buttiauxella strains are localized at different positions from those of other strains. Statistical analyses indicated significant differences between Buttiauxella strains and other strains in both association with B. agrestis MVs and primary maximum energies (Supplementary Figure S6), suggesting that interaction energy based on the DLVO theory is one factor explaining why MVs specifically interact with bacterial cells of Buttiauxella spp. However, the correlation between the two parameters is not linear in Figure 5D (the coefficient of determination R² is 0.59), and other factors besides low interaction energy based on the DLVO theory may affect the specific interaction of MVs in Buttiauxella strains. When MVs and/or cells were treated with proteinase K, the association between MVs and cells in B. agrestis was decreased more than 50% (Supplementary Figure S7), suggesting that proteins localized on the surface of cells and MVs also affect the MV-cell interaction in B. agrestis.

To determine whether the interaction of MVs with bacterial cells contributes to the delivery of the MV contents to bacterial cells, we evaluated vesicle-mediated plasmid DNA transfer. B. agrestis harboring pBBR1MCS-1 was grown in TSB medium to the stationary phase, and plasmid-containing MVs (p-MVs) were extracted from the supernatant. An examination of the pBBR1MCS-1 concentration by quantitative PCR showed 3.11 × 10⁶ and 2.27 × 10⁶ copies/mL in the supernatant before and after the removal of MVs through ultracentrifugation, respectively, suggesting that at least approximately one-third of the plasmid localized in the extracellular milieu was associated with p-MVs in the B. agrestis supernatant. The external DNA surrounding p-MVs was degraded by DNase I treatment, and the plasmid concentration in the p-MVs was calculated by counting the number of MVs using nano tracking analysis. The results showed that p-MVs contain 1.03 × 10⁹ copies/mL of pBBR1MCS-1, indicating that p-MVs maintain a high concentration of plasmid and that the DNA in p-MVs was stable against DNase I treatment. When B. agrestis cells (approximately 1.0 × 10³ cells/mL) were treated with an excessive amount of p-MVs, more than 30% B. agrestis transformants were obtained after 3 h of incubation with DNase I-treated p-MVs and non-treated p-MVs (Figure 6), suggesting that the plasmid contained in MVs was transferred to bacterial cells. When naked plasmid DNA instead of p-MVs was added to directly bacterial cell suspension in this experiment (data not shown), indicating that natural transformation was not occurred in the condition. Notably, other methods for DNA transformation into this strain have not yet been established in our experiments, suggesting that DNA transfer via p-MVs is a useful tool to obtain transformants in this strain.

We expected that the specific interaction of MVs with bacterial cells could be used for the selective control of bacterial cells. Exposure to the antibiotic gentamicin has been shown to increase MV formation, and gentamicin was retained within the MVs (Kadurugamuwa and Beveridge, 1995; Fulsundar et al., 2014). We evaluated the potential of MVs as tools for the delivery of gentamicin to specific bacterial cells. B. agrestis was grown to the stationary phase, and gentamicin was added to the culture at a final concentration of 32 μg/mL (four times the MIC). MVs were extracted, and gentamicin was concentrated in MVs (146 μg/mL) based on the calculated gentamicin concentration and the number of MVs. We prepared gentamicin-associated MVs (g-MVs) and homogenized g-MVs, in which the gentamicin concentration was likely the same as that in the sample of g-MVs. As a negative control, MVs derived from B. agrestis not associated with gentamicin (n-MVs) were also prepared. These MVs were incubated with cells of B. agrestis, E. coli, and P. aeruginosa, and the survival rates were examined. The killing effect of g-MVs on B. agrestis was much greater than that on E. coli and P. aeruginosa, while homogenized g-MVs had a high killing effect on all of the bacteria that were tested in this experiment (Figures 7A–C). Under these experimental conditions, the gentamicin concentration of each sample ranged from 1.3 to 2.3 μg/mL, and this concentration was lower than the MIC of each strain. Subsequently, to determine whether selective antibiotic delivery via MVs occurred in the microbial complexes, the effect of g-MVs in the mixed samples containing B. agrestis and E. coli cells was examined. The results showed that the killing effect of g-MVs on B. agrestis was higher than that on E. coli, even in mixed samples (Figure 7D). These data indicate that the specific interaction of B. agrestis MVs is useful for killing target species in heterogeneous samples.