Abiotic stressors impact outer membrane vesicle composition in a beneficial rhizobacterium: Raman spectroscopy characterization

Outer membrane vesicles (OMVs) produced by Gram-negative bacteria have roles in cell-to-cell signaling, biofilm formation, and stress responses. Here, the effects of abiotic stressors on OMV contents and composition from biofilm cells of the plant health-promoting bacterium Pseudomonas chlororaphis O6 (PcO6) are examined. Two stressors relevant to this root-colonizing bacterium were examined: CuO nanoparticles (NPs)-a potential fertilizer and fungicide- and H2O2-released from roots during plant stress responses. Atomic force microscopy revealed 40–300 nm diameter OMVs from control and stressed biofilm cells. Raman spectroscopy with linear discriminant analysis (LDA) was used to identify changes in chemical profiles of PcO6 cells and resultant OMVs according to the cellular stressor with 84.7% and 83.3% accuracies, respectively. All OMVs had higher relative concentrations of proteins, lipids, and nucleic acids than PcO6 cells. The nucleic acid concentration in OMVs exhibited a cellular stressor-dependent increase: CuO NP-induced OMVs > H2O2-induced OMVs > control OMVs. Biochemical assays confirmed the presence of lipopolysaccharides, nucleic acids, and protein in OMVs; however, these assays did not discriminate OMV composition according to the cellular stressor. These results demonstrate the sensitivity of Raman spectroscopy using LDA to characterize and distinguish cellular stress effects on OMVs composition and contents.

Virulence and signaling through OMVs released by human pathogens 11,30 and plant pathogens 31,32 have been investigated for managing infections. In contrast, factors stimulating OMV release and OMV roles for plant health-promoting bacteria are generally unknown. The beneficial relationship between the plant and its associated microbiome requires cross-kingdom signaling that is mediated through many small metabolites, including hormones and quorum sensing molecules 33,34 . OMV release presents one potential method for the secretion and delivery of these signals.
The potential packaging of nucleic acids and LPS into OMVs may be important as signals to the plant which then activates plant defenses. LPS is a microbe-associated molecular pattern (MAMP) and DNA fragments are damage-associated molecular patterns (DAMPs) recognized by specific plant cell receptors to trigger innate immunity 35 . OMVs contain large quantities of LPS 36 , likely due to the high surface area of these nano-vesicles. OMVs released by P. aeruginosa may contain plasmid DNA 12 , coding chromosomal DNA 5 , or noncoding extracellular DNA (eDNA) 37 .
Here, OMVs of a plant health-promoting bacterium, Pseudomonas chlororaphis O6 (PcO6), were studied. PcO6 is a Gram-negative, rod-shaped bacterium originally isolated from roots of commercial dryland wheat grown in Cache Valley, UT, USA 38 . PcO6 is representative of many beneficial soil bacteria, boosting plant health through multiple protective pathways. PcO6 is an aggressive root colonizer and forms robust biofilms on plant roots 39,40 and abiotic surfaces 41 . Cell growth within the biofilm is nurtured through catabolism of the metabolites in the root exudates 41 . In return, the bacterium protects the host by producing phenazines and other antibacterial and antifungal compounds as well as triggering systemic resistance 42 . PcO6 protects plants from drought, in part by producing a volatile, butanediol, that triggers partial stomatal closure 43 . Biofilm formation may also promote crop drought tolerance because the biofilm matrix maintains moisture around the plant roots 44,45 .
Potential OMVs are observed in atomic force microscopy images of PcO6 cells 46,47 , but OMV composition and their role in PcO6 signaling, biofilm architecture, and stress responses are currently unknown. Two relevant stressors to the rhizosphere, the space around plant roots, and their effects on PcO6 and subsequent OMV production were examined in this study. The first stressor was CuO nanoparticles (NPs). Numerous engineered metal and metal oxide NPs are explored for agricultural applications including delivering macronutrient and micronutrient to crops 48,49 , controlling pests and pathogens 49 , and protecting crops against abiotic stresses 50 such as drought 40,49,51,52 . In some cases, NPs enhance microbial synthesis of products that improve plant health 53 .
In PcO6-colonized wheat, CuO NPs upregulate genes associated with drought tolerance 52 and increase lignification in wheat sclerenchyma, a strengthening tissue of the plant 40 . Thus NP-induced gene expression can contribute to crop stress tolerance 52,54 . NPs also change gene expression in various beneficial and antagonistic microorganisms 54 . Sublethal doses of CuO NPs increase PcO6 cell size without impeding biofilm formation 41 , decrease the production of the Fe-scavenging siderophore pyoverdine 55 , and increase the production of the plant growth regulator indole-3-acetic acid 56 . Hydrogen peroxide (H 2 O 2 ) was examined as a second bacterial stressor as ROS are generated as a stress response by plant root cells 57 , including roots exposed to sublethal CuO NP challenges 58 .
To characterize OMVs, as well as any compositional changes due to these abiotic stressors, Raman spectroscopy was chosen as the primary analytical technique. Compared to other spectroscopy methods, Raman spectroscopy is well suited for biological samples as it requires minimal sample preparation, is non-destructive, and yields a linear correlation of compound concentration to signal strength 59 . Raman spectroscopy is regularly used for cellular identification and characterization, even at a single-cell level [60][61][62] . Changes in biomolecular profiles according to the growth stage can be detected by Raman spectroscopy in mycobacterial cells 63 . Raman spectroscopy has been used to characterize extracellular vesicles from eukaryotic cells [64][65][66] , but no current literature reports Raman spectroscopy characterization of OMVs.
Raman spectroscopy was used to explore the chemical signatures of biofilm PcO6 cells under baseline conditions and after exposure to sublethal doses of CuO NP or H 2 O 2 . OMVs were isolated and characterized from both control and stressed cells. Spectra were compared using linear discriminant analysis (LDA). This machine learning technique creates an algorithm to sort input data sets according to treatments, then uses the algorithm to predict the treatments of the input data sets. The actual treatment and predicted treatment are then compared to determine consistencies and variations between spectra. Biochemical assays detecting protein, LPS, and DNA concentrations were performed to measure their presence in OMVs and to further examine changes in OMV composition.

Results
Physical characterization of outer membrane vesicles. OMV production by PcO6 biofilm cells was confirmed in situ with imaging by atomic force microscopy (AFM) and scanning electron microscopy (SEM). AFM images of PcO6 biofilm cells, transferred from minimal medium agar plates, without stressors show OMVs as aggregates between PcO6 cells (Fig. 1a) and as linear assemblies apart from the cells (Fig. 1b). Purified control and stress-induced OMVs appear as clustered and linear aggregates in AFM images (Fig. 2). SEM analysis also reveals a propensity of purified OMVs to aggregate (Fig. 3). SEM images of PcO6 biofilms formed on hollow fiber membranes draped in minimal medium, a substrate that allows imaging of intact biofilms encased in extracellular polymeric substances, also showed OMVs as part of the biofilm matrix as well as budding from and/or adhered to cells (Fig. 3b).
Size analysis of the OMVs in the AFM images (n = 100 from four images per treatment) showed average diameters of individual OMVs in the range of 40-190 nm, 45-295 nm, and 45-140 nm for control, CuO NPinduced, and H 2 O 2 -induced OMVs (Supplemental Fig. S2). Two-way ANOVA of the data, accounting for the treatment and the image used, showed statistically significant differences (p-value < 0.05) between several images within each treatment (data not shown) leaving uncertainty for the results comparing OMVs from one treatment.  www.nature.com/scientificreports/ Peak assignments were consistent with proteins containing amino acids such as phenylalanine and tyrosine as well as unsaturated lipids and polysaccharides. The signatures of bases adenine, uracil, and guanine indicate the presence of nucleic acids. A full list of peak assignments is shown in Supplemental Table S1. There were no peaks unique for any treatment nor any peaks that disappeared in any one treatment. However, peak intensities differed between treatments with the spectra of control PcO6 giving the highest relative intensities, followed by H 2 O 2 -treated PcO6, then CuO NP-treated PcO6. The following peaks showed the greatest decreases in peak intensity for stressed cells relative to the control: Examining the majority of the Raman spectra (750-1700 and 2670-3100 cm −1 ) with LDA, it is apparent that PcO6 cells exhibit distinct chemical profiles according to the applied stressor (LDA plot shown in Figure 5 and LDA confusion matrix are shown in Table 1). Though several spectra are misclassified from each treatment, the algorithm had an 84.7% accuracy. LDA was also performed on narrower regions of the Raman spectra (Supplemental Fig. S4) to correlate how restricted input data sets influenced LDA grouping and accuracy. LDA results of spectra subsets trended towards increased distance between data groupings. Accuracy of the LDA assignment varied between spectral regions.
The LDA plot of the 2670-3100 cm −1 showed loose grouping of the treatments and the lowest accuracy of the examined regions was 82% (Supplemental Fig. S4b). This region of the spectra corresponds with several C-H and C-H 2 signatures which would be unlikely to change across experimental groups. This region also contains the 2935 cm −1 peak (C-H bonds) which was the highest peak in all spectra and used to normalize the spectra. By contrast, LDA of lower wavenumbers tended to show tighter and more distinct groupings and higher accuracy. This is especially notable in the 1400-1500 cm −1 region (Supplemental Fig. S4f) and the 1500-1700 cm −1 region (Supplemental Fig. S4g) which both gave 94.4% accuracies. All treatments are distinctly grouped in these LDA plots, especially the H 2 O 2 treatment in the 1400-1500 cm −1 region and the CuO NP treatment in the 1500-1700 cm −1 region. The 1400-1500 cm −1 region contains several protein signatures. The 1500-1700 cm −1 www.nature.com/scientificreports/ region contains various amino acid, nucleic acid, amide, and lipid signatures (see Supplemental Table S1 for a list of all peak assignments).

Comparing Raman spectra of PcO6 cells and isolated OMVs. Raman spectroscopy was used to
compare the chemical profiles of PcO6 cells and isolated OMVs. Unique peaks were found in comparisons between Raman spectra of control biofilm PcO6 cells and purified OMVs from control cells ( Fig. 6): 27 peaks are unique to PcO6 cells, 14 peaks are unique to OMVs, and 10 peaks are shared (a full list of peak assignments for PcO6 cells and OMVs is shown in Supplemental Table S1). Figure 6 shows these shared peak assignments, and In this region, differences are seen in relative peak intensity indicating differences in relative concentrations of various compounds. In both graphs, notable peak assignments are shown with an arrow and labeled with both the peak number and peak assignment. See Supplemental Table S1 for a list of all peak assignments and respective sources.
Scientific Reports | (2020) 10:21289 | https://doi.org/10.1038/s41598-020-78357-4 www.nature.com/scientificreports/ whether relative peak intensity is higher for control PcO6 cells or resultant purified OMVs. The same trends were seen in H 2 O 2 and CuO NP-treated PcO6 cells and purified OMVs of stressed cells (data not shown). The relative intensity differed between shared peaks. All but two of these shared peaks had higher relative intensity for the OMV spectra than spectra of intact PcO6 cells. Though there are few peaks shared by intact PcO6 cells and purified OMVs, several unique peaks for PcO6 and OMV Raman spectra correspond to the same chemical compounds: i.e. PcO6 peak 644 cm −1 and OMV peak 870 cm −1 both correspond with tyrosine. Raman peaks assigned to the amino acid tryptophan and the nucleic acid bases adenine, cytosine, and guanine are present in spectra PcO6 cells and purified OMVs. However, the peaks assigned to these chemistries differ between the spectra for intact biofilm cells and the isolated OMVs. Both PcO6 and OMV spectra show various lipid, carbohydrate, protein, and nucleic acid signatures although different signatures of these macromolecules appear in each: i.e. PcO6 peak 810 cm −1 indicates lipid O-P-O bonds while OMV peak 1092 cm −1 indicates lipid C-C bond (Supplemental Table S1).
Especially notable in Raman spectra of isolated OMVs was the increased intensity of nucleic acid peaks which implies that DNA and/or RNA are selectively packaged by PcO6 into and/or onto OMVs. This conclusion was confirmed by measuring OMV 260/280 UV-absorption ratios, which compares the concentration of nucleic acids to the concentration of protein at their respective absorption peaks of 260 and 280 nm. OMV 260/280 ratios did not vary according to cell stressor (Supplemental Table S2) with an overall mean of 5.91 ± 0.09. Generally, a 260/280 ratio above 1.8 is obtained with a pure DNA sample and 2.0 is obtained with a pure RNA sample. The micro-BCA assay results confirmed varying amounts of protein in OMV suspensions (Supplemental Fig. S5b). These high 260/280 ratios for OMVs could be due to other compounds within OMVs that absorb light at similar wavelengths. PcO6 is known to secrete several secondary metabolites including phenazines and siderophores 42 and these results suggest that some metabolites may be packaged into OMVs.
Chemical characterization of stress-induced outer membrane vesicles with Raman spectroscopy. The chemical profiles of control OMVs were compared with H 2 O 2 -and CuO NP-induced OMVs with  www.nature.com/scientificreports/  Table S1 for a list of all peak assignments and respective sources.

Figure 7.
Averaged Raman spectra of OMVs harvested from PcO6 colonies exposed to H 2 O 2 , CuO NPs, or no stressor at all (3 replicates with 4 spectra each for a total of n = 12) from 600 to 3200 cm -1 . Differences are seen in normalized peak intensities indicating large differences in relative concentrations of compounds in the OMVs. Notable peak assignments are marked with an arrow and labeled with their respective peak number and peak assignment. For a list of all peak assignments and their sources, see Supplemental Table S1.  Table 2). Narrower portions of the spectra were also examined with LDA (Supplemental Fig. S6). Much like the LDA examination of PcO6 spectra, LDA of different portions of the OMV spectra showed differences in the grouping on LDA plots. However, the accuracy of the algorithms did not vary greatly between the spectra portions (83-89% accuracy). Five of the six spectra that were misclassified in LDA of the broader spectra were misclassified in most or all of the narrower portions. The LDA plots of these narrower portions control spectra were far removed from the spectra of the other treatments. In contrast, the spectra of OMVs harvested from H 2 O 2 -treated and CuO NP-treated cells tended to be grouped apart as well though with a large intermingled section between them that contained many of the misclassified spectra.
Biochemical assays were also used to measure the LPS and protein content of OMVs. LPS content (average of 15.6 ± 2.7 endotoxin units (EU)/mL solution), which should correlate with OMV number, increased slightly in H 2 O 2 -induced OMVs (Supplemental Fig. S5a). This result matches literature sources that OMV production is increased in response to stress, including ROS stress 21,27,28 . Protein content (average of 23.3 ± 13.4 μg/mL solution) of OMVs was uncorrelated with cellular stressor and trial (Supplemental Fig. S5b) even when normalized to the LPS content (Supplemental Fig. S5c).

Discussion
DLS and AFM measurements showed that OMV sizes were highly polydisperse in size regardless of the cellular stressor. Differing OMV diameter according to the analysis method has been reported in the literature 67 , likely due to differences in OMV aggregation and/or hydration. Aggregation of OMVs observed in AFM and SEM images may be due to outer membrane surface features similar to those involved in adhesion for intact cells 69 . Extracellular biomolecules within the biofilm matrix, which include proteins, polysaccharides, and eDNA 70 , may also bind to OMVs and cause this aggregation. Similar OMV aggregation was observed in studies of OMVs produced by Neisseria lactamica by Gorringe et al. 71 . The researchers observed that numerous large OMV aggregates occurred at pH 7.0 with fewer aggregates occurring at pH 8.0, implying that many OMV surface proteins reached an isoelectric equilibrium at this higher pH, thus reducing the number of OMV aggregates. LPS charge may also be altered by environmental pH, influencing electrostatic and steric barriers to agglomeration. OMV aggregation may be counterproductive to signaling and similar OMV roles, however, OMV assembly may contribute to the assembly of scaffolds to create and support the biofilm matrix. OMVs produced by biofilm Myxococcus xanthus cells form chains that tether biofilm cells together 72 . Xyllela fastidiosa uses OMV aggregates to mediate surface adhesion, either promoting or preventing biofilm formation by the bacterium 17 . PcO6 is a strong biofilm former 41 and OMV release by this bacterium may promote its aggressive colonization of plant roots during adhesion and biofilm formation as patches on the plant root 40 .
This work established that Raman spectroscopy coupled with LDA was a reliable method for characterization of OMV content; the creation of Raman peak libraries (such as Supplemental Table S1 and references [73][74][75][76] ) is valuable in assigning structural changes to cells under different exposures. In this study, LDA was particularly important as Raman spectra of stressed PcO6 cells did not have different peaks from those of control cells, only peak intensities differed. This same pattern occurred for OMVs isolated from stressed and control cells.
To examine what Raman bands were most heavily weighted during LDA grouping calculations, LDA was also performed on smaller spectral regions (Supplemental Fig. S4). For the spectra of PcO6 cells, it appears that LDA is primarily grouped based on peak differences in the range of 750-1700 cm −1 . The 1400-1500 cm −1 and 1500-1700 cm −1 regions were especially unique for H 2 O 2 and CuO NP treatments, respectively. LDA of these regions yielded distinct groupings with few miscalculations. For the OMV spectra, the LDA algorithm primarily converged on the 750-1700 cm −1 range as well. However, LDA of narrower spectral portions within this range did not improve LDA results, indicating that LDA groupings of OMV spectra are less reliant on single peaks and regions than LDA groupings of PcO6 cells.
Comparing Raman spectra of intact PcO6 cells and isolated OMVs, it is clear that certain PcO6 cell components are enriched while others are excluded in OMVs, which agrees with studies of other Gram-negative bacteria 8,77 . The decreased peak intensity of H 2 O 2 -treated PcO6 cells is likely because unsaturated lipids react with H 2 O 2 to form lipid peroxides and reactive aldehydes 78 . The fact that the same peaks are affected in CuO NP-stressed PcO6 cells likely indicates that ROS are formed in cells as a response to CuO NP stress, a cell response that contributes to the dose-dependent bactericidal activity of CuO NPs 79 . The relative greater content of unsaturated lipids in OMVs harvested from these stressed cells compared to OMVs produced by control cells could relate to changes in membrane integrity of cells under stress.
Also notable is the high amount of nucleic acids in OMVs compared to whole biofilm PcO6 cells, implicating the potential role of OMVs to transport DNA and/or RNA in PcO6 biofilms. RNA in OMVs has been reported in the forms of ribosomal RNA, mRNA, and small RNAs 80 . OMVs released by P. aeruginosa may contain plasmid DNA 12 , coding chromosomal DNA 5 , or noncoding eDNA 37 . These nucleic acids extend proposed OMV roles in cell-to-cell communication to both gene and transcription levels assuming OMV contents are taken up by live cells. OMVs of P. aeruginosa contain chromosomal DNA encoding genes related to bacterial survival under stress conditions 5 . The eDNA is an important part of the biofilm matrix 45 and is involved in adhesion to surfaces, aggregation of bacterial cells, and the exchange of genetic information 70 . In H. pylori, eDNA was observed on OMV surfaces, with suggested roles in OMV aggregation and cell-to-cell binding 6 including within the biofilm matrix 72 .
The biochemical assays confirmed the presence of protein, LPS, and nucleic acids in the OMVs from PcO6 that are indicated from the Raman spectral peaks. The inability of these assays to discriminate among OMVs according to the cellular stressor is unsurprising: OMVs are chemically heterogeneous in nature 6,16,68 due to the multiple potential mechanisms and pathways that lead to OMV formation 11,81,82 . The presence of LPS, lipoproteins, and DNA in OMVs is significant because these are among the structures known as MAMPs and DAMPs that induce plant innate resistance 35 . Thus, the contents of the OMVs released from PcO6 cells may be implicated in the induction of systemic resistance observed in plants with roots colonized with PcO6.
The findings of OMV production from the plant-beneficial microbe, PcO6, and the changes in their composition with NP and ROS exposure revealed features that influence plant and bacterial responses in the rhizosphere. The nature of the charge on the OMV surfaces and the eDNA content might affect biofilm formation. Proteins such as catalase found in OMVs from other bacteria 28 potentially protect the bacterial cell from oxidative damage. DNA and RNA release through OMVs could affect plant gene expression. These bacterial nucleic acids, along with other confirmed MAMPs such as LPS, carried by OMVs, could induce plant resilience to stress.
In summary, research into the heterogeneous nature and multiple roles of OMVs is supported by applying sensitive techniques such as Raman spectroscopy supported by appropriate analyzation algorithms such as LDA. In this study, these methods were able to chemically characterize purified OMVs and categorize these vesicles based on two cellular stressors relevant to the soil environment of the plant health-promoting bacterium, PcO6. The ability to differentiate H 2 O 2 -induced and CuO NP-induced OMVs reveals that these stressors do not merely induce OMV production. Rather, PcO6 responds to these stressors with changes in OMV contents and composition. The construction of an OMV Raman peak assignment library presented here provides a baseline for future analyses by our group and others. www.nature.com/scientificreports/ Methods Bacterial growth conditions. PcO6 stocks were kept at − 80 °C in 15% glycerol and thawed before use. PcO6 biofilms were grown at 22 °C on minimal medium (K 2 HPO 4 -10.5 g/L, KH 2 PO 4 -4.5 g, Na*citrate*2H 2 O-0.5 g/L, (NH 4 ) 2 SO 4 -1 g/L, sucrose-2 g/L, anhydrous MgSO 4 -0.125 g/L) 2% agar plates (15 × 100 mm) for 48 h to a confluent lawn.
PcO6 biofilms were also grown on hollow fiber membranes for SEM imaging 41 . Hollow fiber membranes were inoculated with 2 μL PcO6 suspended in sterile double distilled H 2 O (ddH 2 O) (resistance > 18 MΩ cm) at a concentration of 10 6 colony forming units (CFUs)/mL, draped across wells with liquid minimal medium, and allowed to grow at 22 °C for 28 h.
Abiotic stressor preparations. Sterile ddH 2 O was used as the control treatment. Commercial CuO NPs (nominal size < 100 nm, 99.95% purity) were obtained from American Elements as a nanopowder and stored protected from light. NP size distribution and agglomeration profiles were confirmed with scanning electron microscopy (FEI Quanta FEG 650) (Supplemental Fig. S7). NP elemental composition was determined by scanning electron microscopy with energy-dispersive X-ray spectroscopy using an X-Max Detector (Oxford Instruments). NP stress used CuO NPs suspended in sterile ddH 2 O (30 mg Cu from CuO NPs/L) through sonication (Q500, QSonica LLC) for 10 min with an alternating 10 s on/off cycle at 25% amplitude. The ROS stress was 3% Isolating and purifying outer membrane vesicles. OMVs were isolated from PcO6 cells using a method adapted from Zhou et al. 83 . Twenty minimal medium plates with confluent PcO6 lawns (a total surface area of 0.628 m 2 ) were used per treatment to maximize OMV yields for analysis. The cell suspension (10 mL/ plate for a total of 200 mL/treatment) was poured from the petri dishes into 50 mL polypropylene conical tubes and centrifuged (10,000×g, 20 min) to generate a cell pellet leaving OMVs and other secreted materials in the supernatant. This step also removed CuO NPs from the solution when this stressor was used. The OMVs were concentrated from the supernatant by ammonium sulfate precipitation. Ammonium sulfate (Mallinckrodt chemicals) was added over the course of two hours (240 g/L total) at 15 min intervals to the supernatant, which was left undisturbed at 22 °C until precipitates formed (anywhere from 2-8 h). The precipitates were pelleted by centrifugation (10,000×g, 10 min), resuspended in 1 mL sterile deionized H 2 O, and dialyzed against ddH 2 O for at least 16 h. The solution was sterile filtered (0.45 μm, Ultrafree PVDF centrifugal filter units, Beckman Coulter Inc.) to remove any remaining cells or contaminants.

Treatments of
To purify OMVs from flagella, pili, and secondary metabolites (Supplemental Fig. S8), density gradient ultracentrifugation was performed using methods adapted from Chutkan et al. 84 . The crude OMV suspension was mixed with OptiPrep iodixanol gradient medium (Sigma Aldrich) to create a 45% OptiPrep solution (vol/ vol). A 2 mL aliquot was loaded into the base of each ultracentrifuge tube (Ultraclear 12.5 mL centrifuge tube, Beckman Coulter) and covered sequentially with 2 mL layers of 40,35,30,25, and 20% OptiPrep before centrifugation (212,000×g, 3 h, 4 °C) (Optima LE-80 K Centrifuge, Beckman). Aliquots of 1 mL were collected from the top of the gradient in a cold room to minimize diffusion. OMV-containing fractions, the top 1 mL of each tube, were pooled from each treatment, diluted at least 10 times with sterile deionized H 2 O, loaded into centrifuge tubes (26.3 mL Polycarbonate Bottle with Cap, Beckman Coulter), and centrifuged (40,000×g, 3 h) to pellet the OMVs. The OMV pellet was resuspended in 750 μL sterile ddH 2 O and sterile filtered (0.45 μm, Ultrafree PVDF centrifugal filter units). The presence of OMVs in the filtrate was confirmed with AFM imaging. Pure OMV preps were stored at 4 °C until used for Raman spectroscopy, which took place within 48 h of OMV isolation and purification. Samples were frozen (− 20 °C) until used for biochemical assays.
Atomic force microscopy (AFM). AFM was performed on a Nanoscope III Bioscope (Digital Instruments, Inc.) in tapping mode. Budget Sensors-Tap 300AL-G cantilevers with a tip radius of curvature < 10 nm, length of 125 μm, width of 30 μm, thickness of 4 μm, and a 40 N/m force constant were employed. Images were collected at 256 × 256 resolution and 1 Hz over a range of scan sizes and scan angles. For intact biofilm PcO6 cells, a sample of the bacterial lawn on a sterile inoculation loop was smeared on a clean glass slide and immediately imaged. For OMVs, 20 μL purified OMV suspension was pipetted onto a clean glass slide, allowed to dry, and immediately imaged. OMV diameter was determined using the horizontal distance and the vertical distance in the line cut feature in the Nanoscope software (Supplemental Fig. S9). Four images were used for each treatment with a total of 100 OMVs measured per treatment. Raman spectroscopy and linear discriminant analysis (LDA). To examine the intact cells, confluent PcO6 lawns from two minimal medium plates with or without the stress treatments were centrifuged from solution (14,000×g, 10 min), and resuspended in 1 mL sterile, deionized H 2 O. This process was repeated two additional times to remove stressors. After the third centrifugation, PcO6 cells were resuspended in 200 μL sterile, deionized H 2 O. A total of 6 PcO6 replicates were performed per treatment. The majority of OMVs, cell secretions, and other cell debris remained in the supernatant which was discarded after each centrifugation. A 10 μL aliquot of the PcO6 suspension was pipetted onto aluminum tape affixed to a clean glass microscope slide and allowed to dry. For OMVs, no additional preparation was needed for Raman spectroscopy after isolation and purification steps. A total of 3 isolations were performed with each treatment. Twenty μL purified OMV suspension was pipetted onto aluminum tape affixed to a clean glass microscope slide and allowed to dry. Raman spectra were obtained using a Renishaw inVia Raman microscope with a 633 nm laser, 1200 g/ mm grating, and 14 mW laser power. Spectra were obtained over a 30 s acquisition time with a wavenumber range of 200-3200 cm −1 , with 3 accumulations per spectrum. Four spectra were captured per treatment during each replicate for a total of n = 24 for PcO6 spectra and n = 12 for OMV spectra. Spectrogryph 85 was used to visualize spectra, obtain peak numbers, and average spectra. LDA of spectra was performed with R following background subtraction (WiRE 4.1), removal of cosmic rays (WiRE 4.1), and normalization of spectra (R). Data were truncated to 750-1700 cm −1 and 2670-3100 cm −1 for broad-spectrum analysis. PcO6 cell spectra were further truncated to 750-1700 cm −1 , 2670-3100 cm −1 , 700-950 cm −1 , 950-1200 cm −1 , 1200-1400 cm −1 , 1400-1500 cm −1 , or 1500-1700 cm −1 for additional examination with LDA. OMV spectra were further truncated to 750-1700 cm −1 , 2670-3100 cm −1 , 1000-1200 cm −1 , 1200-1500 cm −1 , 1500-1600 cm −1 , or 1500-1600 cm −1 for additional examination with LDA.

Scanning electron microscopy (SEM
Micro-bicinchoninic acid (micro-BCA) assay. OMV protein content was quantified with a Pierce micro-BCA assay kit (Thermo Scientific). Frozen OMV samples were thawed and prepared according to the manufacturer's instructions and absorbance was read at 562 nm (Synergy HT, BioTek Instruments Inc.) with appropriate bovine serum albumin standards from the manufacturer with concentrations of 0 to 200 μg protein/ mL. Duplicates were run for each standard and sample.

Lipopolysaccharide (LPS) quantification. OMV LPS content was quantified with a Pierce Endotoxin
Quantification Kit (Thermo Scientific). Frozen samples were thawed and prepared according to the manufacturer's instructions and absorbance was read at 405 nm (Synergy HT, BioTek Instruments Inc.) with LPS standards from Escherichia coli supplied by the manufacturer in the range of 0.1 to 1 Endotoxin Units (EU)/mL. Duplicates were run for each standard and sample. 260/280 absorption ratio. The nucleic acid to protein ratio was determined based on the 260/280 UVabsorption ratio. Frozen samples were thawed and 2 μL of each sample was pipetted into wells of a Take3 Micro-Volume plate (BioTek Instruments Inc.) and absorbance was read at 260 and 280 nm (Synergy HT, BioTek Instruments Inc.). Standards of bovine serum albumin (40 μg/mL) from a micro-BCA assay kit (Thermo Scientific) and double-stranded DNA (100 μg/mL) from a PicoGreen DNA quantification kit (Thermo Scientific) were also run. Duplicates were run for each sample and standard.

Data availability
The datasets generated during this study are available from the corresponding authors on appropriate request.