Biophysical and proteomic analyses of Pseudomonas syringae pv. tomato DC3000 extracellular vesicles suggest adaptive functions during plant infection

ABSTRACT Vesiculation is a process employed by Gram-negative bacteria to release extracellular vesicles (EVs) into the environment. EVs from pathogenic bacteria play functions in host immune modulation, elimination of host defenses, and acquisition of nutrients from the host. Here, we observed EV production of the bacterial speck disease causal agent, Pseudomonas syringae pv. tomato (Pto) DC3000, as outer membrane vesicle release. Mass spectrometry identified 369 proteins enriched in Pto DC3000 EVs. The EV samples contained known immunomodulatory proteins and could induce plant immune responses mediated by bacterial flagellin. Having identified two biomarkers for EV detection, we provide evidence for Pto DC3000 releasing EVs during plant infection. Bioinformatic analysis of the EV-enriched proteins suggests a role for EVs in antibiotic defense and iron acquisition. Thus, our data provide insights into the strategies this pathogen may use to develop in a plant environment. IMPORTANCE The release of extracellular vesicles (EVs) into the environment is ubiquitous among bacteria. Vesiculation has been recognized as an important mechanism of bacterial pathogenesis and human disease but is poorly understood in phytopathogenic bacteria. Our research addresses the role of bacterial EVs in plant infection. In this work, we show that the causal agent of bacterial speck disease, Pseudomonas syringae pv. tomato, produces EVs during plant infection. Our data suggest that EVs may help the bacteria to adapt to environments, e.g., when iron could be limiting such as the plant apoplast, laying the foundation for studying the factors that phytopathogenic bacteria use to thrive in the plant environment.

S uccessful colonization of hosts depends on the ability of microbes to defend themselves against host immune responses and acquire nutrients. Bacterial pathogens use macromolecular translocation systems and deliver virulence proteins, so-called effectors, to circumvent host immunity (1). Pseudomonas syringae pv. tomato (Pto) DC3000 is the causal agent of bacterial speck, a common disease that affects tomato production worldwide (2,3). Pto DC3000 is a Gram-negative bacterium that invades through openings in the plant surface and propagates in the apoplast, where it takes up nutrients and proliferates (4)(5)(6). Plants respond rapidly to colonization by microbes, activating interlinked innate defense strategies (7), which can broadly be categorized into pattern-triggered immunity (PTI) activated by pathogen-associated molecular patterns (PAMPs) and effector-triggered immunity induced upon recognition Research Article mBio median sizes of 100 and 115 nm for fluid samples and gradient-collected samples, respectively ( Fig. 1E and F). It is possible that conditions used for SEM and NTA differ in their capacity to hydrate the vesicles and/or that NTA underestimates smaller particles (32).
To determine whether EV production is an active process, EVs were quantified from culture supernatants of Pto DC3000 over cultivation time, with increasing particle numbers observed with bacterial density (Fig. S2A and B). Calculation of the amount of EVs produced per bacterium showed that numbers were similar between growth stages ( Fig. S2A and B). The median diameter and ζ-potential of EVs were mostly comparable across growth stages, yet differed slightly between the sample types ( Fig.  S2C and D). Albeit we cannot exclude the possibility that vesicles could be derived, e.g., from exploding cells (20), the vesicles recovered from culture samples appeared to be predominantly produced by bacteria as an active process since the number of dead cells from planktonic cultures was little compared with heat killing (Fig. S2E), which also caused higher particle numbers (Fig. S2F). This is consistent with the observations from TEM ( Fig. 1C and D).

EVs from cultured Pto DC3000 are enriched in 369 proteins
To gain insights into the biogenesis and functions of Pto DC3000 EVs, we characterized the proteome of EVs using liquid chromatography-based tandem mass spectrometry (LC-MS/MS). To this end, we cultivated Pto DC3000 in a rich, yet iron-limited medium, allowing for high bacterial growth and thus EV yield as well as considering iron limitation in the leaf apoplast during pathogen infection (25,33). The Pto DC3000 EV-associated proteins were isolated from Pto DC3000 cultures by gradient enrichment. In parallel, we determined the proteomes of whole cells (WC) and OM preparations. We detected the highest number of proteins from the WC sample (n = 1,587), followed by the EV sample (n = 890) and 212 proteins in OM samples ( Fig. 2A; Table S1). In total, 2,898 proteins were identified over all samples, of which 1,899 proteins were identified at least in three of the four samples per sample type (WC, EV, or OM). These proteins were taken forward for further analysis (Table S1). Similar protein intensity distributions were obtained for all samples [label-free quantification (LFQ) values were generated by MaxQuant, Fig. S3], and the four replicate measurements per sample type fell into sample clusters on the first and second principal components, suggesting a systematic difference in the proteomes of these three sample types (Fig. 2B). By comparing the proteomes of EV and WC, we identified 369 EV-enriched proteins, consisting of 162 proteins exclusively identified in at least 3 replicates of EV sample (EV unique detected; Fig. 2C) and 207 proteins significantly higher in the EV compared with WC ( Fig. 2C; Table S1). Next, we analyzed the proteomics data (i) using computational approaches (bioinformatics and database searches) and (ii) building on current knowledge (EV biogenesis and immunomodulatory activities).

Pto DC3000 EVs are enriched in proteins with predicted roles in transport and antimicrobial peptide resistance
We performed a gene set analysis on the 369 EV-enriched proteins to examine the biological processes, cellular component, and molecular function in which these proteins are involved [from gene ontology (GO)] (34,35). In total, 20 GO terms were significantly enriched [ Fig. 3A through C; false discovery rate (FDR) <0.05; DAVID bioinformatics resources] (36,37). Eight terms were found in the category "biological process, " out of which, four terms were associated with "cellular processes" related to cell division, shape, and cell wall remodeling. An increasing release of EVs was observed in cells that grow at exponential phase, likely due to an increased turnover of peptidoglycan during cell division (38). Three terms were connected to the general process of "transport, " including transmembrane transport, intracellular transmembrane transport, and protein transport by the Sec complex ( Fig. 3A; Table S2). This suggests a specific and/or selective mecha nism for the delivery of protein cargoes into EVs. Indeed, classification of the proteins enriched in EVs by putative subcellular localization revealed distinct localization profiles compared with WC and OM proteins. While 66% of WC proteins were cytoplasmic, about half (51%) of the EV-enriched proteins were cytoplasmic membrane associated, with the next largest known class being OM associated (11%) (Fig. S3C). This is consistent with the four terms found in the category "cellular component, " connected to the general compartment "membrane, " including OM and plasma membrane (Fig. 3B). Combining the data from proteomics and TEM, it may suggest that Pto DC3000 produces EVs in the form of both OMVs and OIMVs, as previously described for the closely related species P. aeruginosa (20). The category "molecular function" was enriched in eight terms, includ ing processes associated with peptidoglycan synthesis also found in the "biological process" category, and siderophore transport (Fig. 3C). Siderophores are secondary metabolites that can sequester iron. Bacteria secrete siderophores under iron-limiting environments, improving iron uptake and thereby contributing to bacterial survival (39). The enrichment of siderophore transport proteins in Pto DC3000 EVs suggests that the release of EVs may contribute to the acquisition of iron.
EV yield and likely cargo composition are affected by the environment, in which bacteria grow (25). Having identified proteins enriched in EVs collected from cultured Pto DC3000 bacteria, this could limit evidence on the role of EV-enriched proteins during plant infection. If EV-enriched proteins would be involved in infection, we assumed that (i) proteins enriched in EVs from cultured bacteria would be present in bacteria in planta and, thus, (ii) genes coding for EV-enriched proteins would be expressed in bacteria in planta, and (ii) genes coding for EV-enriched proteins could respond to the plant's immune status. We, therefore, inspected available Pto DC3000 transcriptome data (33).
The expression patterns of genes coding for EV-enriched proteins differed mostly between Pto DC3000 cultured in vitro (in both minimal and rich media), present in planta (in both untreated and mock-treated plants), and present in flg22 immune-induced plants (Fig. 3D). We found five clusters of gene expression patterns across these condi tions. Of note, genes in cluster II were upregulated in bacteria in response to in planta conditions but downregulated in bacteria from flg22-induced plants. It is thus possible that the proteins encoded by the genes in cluster II are also present at EVs produced by Pto DC3000 in untreated and mock-treated plants. Since cluster II is enriched in the GO term "siderophore uptake transmembrane transporter activity" (Fig. 3E; Table 1; Table  S2), EVs may play roles in iron acquisition (Fig. 2C, orange labeling).
Cluster III contains genes that were similarly expressed in bacteria grown in vitro and bacteria from flg22-induced plants but differed in their expression in response to in planta conditions (Fig. 3D). This cluster is associated, e.g., with the GO term "bacterialtype flagellum hook" (Fig. 3E), which has been described in the biogenesis of EVs (40). Also, cluster III is associated with the GO term "serine-type D-Ala-D-Ala carboxypeptidase activity, " which has a cross-reference with the term "Penicillin-binding protein 2" in the InterPro database of protein families. It is worth mentioning that genes PSPTO_3987 and PSPTO_4977 both annotated with the "β-lactam resistance" function are also present in cluster III, although not assigned in the above-mentioned GO term.

Purified Pto DC3000 EV samples have immunomodulatory activities
Given the presence of flagellin in our Pto DC3000 EV samples, we next examined the ability of the Pto DC3000 EVs to modulate the outcome of bacterial infection. We pretreated Arabidopsis thaliana leaves with Pto DC3000 EVs, which limited the growth of subsequently infected Pto DC3000 bacteria in planta (Fig. 4A). Thus, the immunogenic activity of Pto DC3000 EVs is sufficient to restrict bacterial colonization, consistent with recent observations (30). In agreement, seedlings treated with purified EVs showed induction of pFRK1::GUS expression, albeit lower when compared with treatments with flg22 ( Fig. 4B; Fig. S4A). We also tested whether treatment with Pto DC3000 EVs could arrest seedling growth, a prototypic PTI response of plants to continual PAMP stimulation (41). We observed no significant growth reduction in this experiment (Fig. 4C).
Six flagella-associated proteins were enriched in Pto DC3000 EVs, of which flagellin was more than twofold enriched relative to the WC proteome (Table S1). Therefore, to determine the pathway by which the Pto DC3000 EVs trigger immune responses, we treated A. thaliana mutants of the FLAGELLIN SENSING 2 (FLS2) and EF-Tu receptor (EFR) (E) GO terms associated with clusters.
Research Article mBio immune receptors responsible for recognition of flg22 and elf18, respectively (42). The Pto DC3000 EVs triggered FRK1 gene expression in wild-type (WT) and efr-1 mutants to similar levels (Fig. 4D). No FRK1 induction was observed in fls2 mutants. Thus, the EV samples isolated from Pto DC3000 cultures must contain bacterial flagellin as the immunogenic molecule.
Notably, SEM analysis of gradient-collected EV samples showed the co-purification of filament-like structures (Fig. 1B), which could represent detached bacterial flagellar or pili. Since flagellin could not be detected in proteinase K-treated EVs and proteinase K-treated EVs did not significantly induce pFRK1::GUS expression ( Fig. 4E and F; Fig.  S4B), taken together, it is possible that flagellin is a co-purifying immunogenic molecule present in Pto DC3000 EV samples and recognized in A. thaliana. This is consistent with the observation that EVs purified from the Pto DC3000 ΔfliC mutant did not significantly induce pFRK1::GUS expression ( Fig. 4G; Fig. S4C). Considering co-purifying flagellin as the major immunogenic molecule in Pto DC3000 EV samples, its amount might be insufficient to repress seedling growth over time.
The EV-enriched proteome included proteins related to virulence ( Fig. 2C; Table  S1), such as MucD (PSPTO_4221) (43), HopAJ2 (PSPTO_4817) (44), and HopAH2-2 (PSPTO_3293) (45,46). A major function of virulence proteins is the suppression of PTI (47). Recently, the integration of X. campestris pv. campestris OMVs into plant plasma membranes was observed, which might suggest that vesicle cargoes such as virulence proteins could be discharged into plant cells (48). To test whether Pto DC3000 EVs could modulate a prototypic PTI response, we pretreated leaves with EVs from cultured bacteria 24 hours before eliciting an ROS burst with the immunogenic peptides flg22 from bacterial flagellin and elf18 from EF-Tu. EV pretreatments neither significantly reduced nor increased the PAMP-induced ROS production (Fig. 4H). This suggests that under the tested conditions, Pto DC3000 EVs are not predominantly involved in inhibiting and/or further enhancing the PAMP-induced ROS responses.

Pto DC3000 bacteria produce EVs in planta
The observation that EVs collected from Pto DC3000 cultures may not play major roles in host immune modulation raises the question whether Pto DC3000 releases EVs during plant infection. To address this, apoplastic fluids were recovered from Pto DC3000-infec  The size of apoplastic fluid vesicle-like particles determined by both NTA and SEM did not significantly differ between control and infected leaves ( Fig. 5A; Fig. S5C). Particle abundance increased upon infection with Pto DC3000 (Fig. 5B), consistent with previous findings (49). Increased particle abundance correlated with both bacterial infection time and titers ( Fig. 5B; Fig. S6A and B). We also analyzed EVs from the apoplastic fluids of plants that were co-treated with 100 nM flg22 and Pto DC3000. Particle numbers were lower than those recovered from Pto DC3000 infection only, consistent with induced plant resistance and not significantly different from plants only stimulated with flg22 ( Fig. 5C; Fig. S6C). Taken together, comparing the particle profiles of fluids isolated from Pto DC3000-infected plants with flg22 immune-stimulated plants, both the higher particle number and the polydisperse particle size hint at bacterial-derived EVs present in the apoplast of infected A. thaliana.
Since Pto DC3000 (fluid sample) and A. thaliana (apoplastic fluid samples) EVs did not significantly differ in diameter ( Fig. 1E and 5A), we focused on the charge of EVs, reflecting the different surface composition of bacterial (prokaryotic) and plantderived (eukaryotic) EVs. Evaluation of the mean ζ-potential identified significantly less negatively charged EVs recovered from apoplastic fluids of Pto DC3000-infected plants at 3 days post-infection (dpi) compared with control treatments and earlier time points (Fig. 5C). This time point correlated with in planta bacterial proliferation and depended on bacterial inoculum ( Fig. S6A and B). Plotting the relative particle abundance over particle charge, the ζ-potential profiles of EVs recovered from apoplastic fluids of untreated, control-treated, and flg22-treated A. thaliana identified major peaks around −32 mV ( Fig. 5D; Fig. S6J). By contrast, the ζ-potential profile of EVs recovered from apoplastic fluids of Pto DC3000-infected A. thaliana had a broader distribution with a similar major peak around −32 mV and an additional shoulder around −10 mV ( Fig. 5E; Fig. S6K). Comparison of the different ζ-potential profiles revealed similarities of the major −32 mV peak across all plant samples, likely representing a plant-derived EV pool. Notably, the shoulder around −10 mV detected from apoplastic fluids of Pto DC3000-infected plant samples showed an overlay with the ζ-potential profile of EV recovered from Pto DC3000 cultures (fluid samples), with a peak from −20 mV to 0 mV ( Fig. 5F and G). This could, therefore, represent a bacterial-derived EV pool. Since the ζ-potential profiles of EVs recovered from Research Article mBio apoplastic fluids of flg22-treated A. thaliana did not differ between untreated and control-treated leaves (Fig. S6J), we found no evidence that plant EVs modulate their surface charge during infection.

Pto DC3000 EV-associated proteins are detected during plant infection
We next aimed to identify EV-associated proteins that could be used as markers for Pto DC3000 EVs in planta. To this end, we addressed whether the protein composition of EVs from Pto DC3000 and EVs from related bacteria shares similarities. We focused on three published P. aeruginosa PAO1 EV proteomes since a number of EV proteomes have been reported from P. aeruginosa (50)(51)(52). We found that 103 proteins were identified in the EV proteomes across the three reports (50)(51)(52). Of the 103 shared EV proteins from PAO1, we could identify 100 orthologous proteins encoded in the Pto DC3000 genome, and 44 proteins were enriched in Pto DC3000 EVs ( Table 2; Table S1). We refer to these as the EV "core. " These proteins were highly enriched in localization to the OM (44%) and cytoplasmic membrane (26%) (Fig. S7A), consistent with EVs released in the form of OMVs. From these 44 proteins, 20 were putative OM-localized proteins and thus represent good candidate biomarkers for the detection of EVs (Table S1; Fig. 2C, blue labeling). To gain insights into the expression of the EV "core" encoding genes during infection, we searched available transcriptome data (33), which revealed two clusters of generally lower or higher expression levels across the conditions (Fig. S7B). Of the genes coding for EV "core" proteins, a majority encoding OM-localized and cytoplasmic proteins were upregulated in planta (Table S1), which suggests their presence during bacterial infection.
One of the predicted EV markers is OprF (Fig. 2C, black labeling), which we used for immunodetection of Pto DC3000 EVs in planta and purified from Pto DC3000 cultures. OprF is a porin integral to the OM and found in EVs of P. aeruginosa (53). Using anti-OprF antibodies, we identified specific bands in filtered apoplastic fluids of A. thaliana leaves infected with Pto DC3000 at 2 and 3 dpi but not in control-treated plants (Fig. 6A). In combination with using anti-TET8 (anti-TETRASPANIN 8) antibodies, we could confirm that the apoplastic fluids of control and infected leaves contain both plant-derived and  Only in rare cases, gold particles were found at membranous debris. In heat-treated samples, a large amount of membraneous structures (gray areas, white arrowheads) were observed, likely due to the disruption of EVs by the heat treatment. Here, immunogold labeling was observed at the membranous debris (some gold particles are indicated with white arrows). Two representative images are shown for each condition.
Research Article mBio bacterial EVs (Fig. 6B). OprF-positive EVs were also detected from the Pto DC3000 ΔfliC mutant (Fig. 6C), confirming successful purification of EVs from this strain. However, OprF could not be detected in proteinase K-treated EVs (Fig. 6C). In P. aeruginosa, OprF is a general porin of the OM (54) and, consistently, we detected OprF also in the proteome of the OM samples (Table S1). It is thus possible that the immunogenic epitope recognized by the anti-OprF antibodies is located at the external side of the EVs, making it sensitive to proteinase K treatment. It is currently emerging that the EV surface corona plays roles in EV functions (55), and it is possible that OprF might be part of the Pto DC3000 EV corona. We next aimed to identify an EV-enriched protein localized to the EV lumen and explored available antibodies. For example, EVs of Stenotrophomonas maltophilia contained β-lactamase (56). Given that two genes annotated with the "β-lactam resistance" function were present in cluster III and although not detected in our proteome samples, we reasoned that β-lactamase could be present in Pto DC3000 EVs. Using anti-β-lactamase antibodies, we identified a specific band in gradient-collected EV samples of cultured Pto DC3000, which was also detected after proteinase K treatment ( Fig. 6D; Fig. S7C). In addition, a band could be revealed in apoplastic fluids collected from Pto DC3000-infected leaves, suggesting the presence of β-lactamase-positive Pto DC3000 EVs in these samples ( Fig. 6E; Fig. S7C). We explored immuno-negative staining to gain further insights into the localization of β-lactamase in Pto DC3000 EVs (Fig. 6F). In mock samples, several intact vesicles were found yet without significant immunogold labeling. Only in rare cases, gold particles were observed at membranous debris. By contrast, after breaking up the vesicles by heat treatment, the membraneous structures showed clear immunogold labeling as several gold particles were observed at these structures. This suggests that β-lactamase was more accessible to ampC immunogold labeling when EVs were disrupted, consistent with β-lactamase present in the lumen of Pto DC3000 EVs. Having identified two possible markers for Pto DC3000 EVs, membranebound OprF and soluble β-lactamase located at the external side and the lumen of EVs, respectively, these data provide additional evidence that Pto DC3000 releases EVs in planta during infection.

DISCUSSION
A range of activities has been associated with bacterial EVs during infection. This includes modulation of host immunity (PAMPs, effectors, and bacterial cell wall remodeling), elimination of host defenses (antibiotic tolerance and decoys), and acquisition of nutrients from the host (metal ions) (57). In this study, we used a proteo mics approach aiming to gain insights into the presence and role of Pto DC3000 EVs during plant infection.
PTI responses to Pto DC3000 involve recognition by FLS2, EFR, and LIPO-OLIGO SACCHARIDE-SPECIFIC REDUCED ELICITATION (LORE), which detect immunogenic flg22, elf18, and 3-OH-FAs, a fatty acid co-purifying with LPS (58). EVs from bacterial phytopath ogens are enriched in EF-Tu and LPS (25,26,28), suggesting the presence of elf18 and 3-OH-FAs. Bahar et al. demonstrated that BRI1-ASSOCIATED KINASE 1 (BAK1) and SUPPRESSOR OF BIR 1 (SOBIR1), interacting co-receptors of PRRs (pattern recognition receptors), mediate the immunogenic perception of EVs from X. campestris pv. campestris (25). We show that vesicle samples from Pto DC3000 elicit immune responses that are dependent on FLS2 and the presence of flagellin in EV samples (Fig. 4 thorugh G). Since flagella proteins such as FliC have a specific affinity for EVs and are involved in EV production in Escherichia coli (40), they could be considered external EV cargoes. However, filamentous structures were observed in SEM analysis (Fig. 1B). It is, therefore, possible that flagella co-purify with the Pto DC3000 EV samples despite depleting extracellular components from EV samples by density gradient centrifugation. Contami nation of flagella in EVs was reported to contribute to the detection of FliC in EVs from P. aeruginosa (59).
In contrast to our results, McMillan et al. reported significant seedling growth repression in response to Pto DC3000 EVs (30). This disparity in results may be due to several factors, including differences in the growth conditions of both the bacterial cultures and the A. thaliana seedlings, the type of biochemical isolation of EVs and vesicle dose, e.g., resulting in different amounts of co-purifying flagella. The stronger protective immune response observed by McMillan et al. may also be due to such differences in experimental procedures (30). Importantly, EVs purified from Pto DC3000 fliC mutant bacteria did not significantly induce defense gene expression (Fig. 4G). This non-significant immunogenicity of Pto DC3000 EVs would be congruous with Pto DC3000 releasing EVs in favor of plant infection.
The plant's apoplast, which is the niche colonized by Pto DC3000, represents an environment where bacteria are challenged with plant defense molecules, competition with members of the microbiome, and acquisition of nutrients (33). Bacteria respond to environmental stress with the production of EVs, which allows for cell surface remodel ing, secretion of degraded and damaged cargo, and uptake of nutrients in bacterial communities, e.g., by packaging transporters in EVs (14,20,60). The hypothesis that Pto DC3000 may use EVs to adapt to the host environment is evidenced by our finding that Pto DC3000 EVs contain β-lactamase (Fig. 6D). Several studies demonstrated that EVs can improve bacterial survival during antibiotic exposure. S. maltophilia produced more EVs upon treatment with the β-lactam antibiotic imipenem (56,61). Its EVs contained β-lactamase and increased S. maltophilia survival in the presence of antibiotics (56). Plants defend infection by upregulation of many defense-related genes, including genes coding for antimicrobial peptides (62). Likewise, the host microbiome protects plants against infectious pathogens (63). It is possible that Pto DC3000 produces EVs to counter the action of plant-and microbe-derived antimicrobial peptides, i.e., its EVs could improve antimicrobial resistance of Pto DC3000 (64,65).
Proteins involved in siderophore transport were enriched in the EVs from cultured bacteria (Fig. 2C). Siderophores are important virulence factors of bacterial pathogens, directly competing for iron with the host (66). Since the ability of Pto DC3000 for iron acquisition is correlated with its growth in planta (33), siderophores and their transport are likely to play roles in Pto DC3000 infection success. Interestingly, the expression of genes coding for all siderophore transport proteins enriched in EVs was upregulated in planta compared with in vitro conditions as well as downregulated upon induction of PTI (Fig. 3D). Thus, regulation of siderophore transport proteins can be considered as an adaptive response of Pto DC3000 to iron/metal ion availability, and secretion into EVs may allow improved acquisition of iron, analogous to EV secretion of the siderophore mycobactin in Mycobacterium tuberculosis (67). In addition, the ability of siderophore uptake into EVs may prevent activation of immunity as siderophores can trigger immune responses (66). Taken together, we propose that Pto DC3000 produces EVs to improve its growth ability both in culture and in planta. Having identified two markers for Pto DC3000 EVs, membrane-bound OprF detected as external EV cargo and β-lactamase as luminal EV cargo, future studies to determine the composition of EVs in planta and track their interaction with plant cells are now becoming possible.

Bacterial strains and growth
Pto DC3000 used in this study was routinely cultured at 28°C in King's B (KB) medium containing 50 µg/mL rifampicin at 180 rpm and on plates with 1% agar without agitation. Planktonic growth was performed in 500 mL cultures, and growth rates were measured over time as OD 600 . Pto DC3000 ΔfliC was cultivated at 28°C in KB medium containing 50 µg/mL rifampicin and 5 µg/mL chloramphenicol at 180 rpm (68).

Extraction and purification of bacterial EVs
EVs were routinely isolated from planktonic cultures across growth phases (Fig. S8A). The starting inoculum was OD 600 = 0.01. Samples were taken according to incubation time, evidently slightly differing in the then measured OD 600 , and therefore, we indicate a range of colony-forming units per milliliter bacterial density, from which EVs were collected. One hundred milliliters of planktonic bacteria (grown in liquid cultures) and 10 mL of biofilm bacteria (grown on plates, collected, and resuspended), respectively, were pelleted at 4,500 × g for 2 × 20 minutes, the supernatant was decanted and passed through a 0.22-µm membrane (fluid samples; Fig. S1C). Particles were pelleted from the cell-free supernatant at 100,000 × g for 1.5 hours. The pellet was resuspended in 1.7 mL 1 mM EDTA and loaded on sucrose density step-gradient (1.7 mL of sucrose 25%, 35%, 45%, 50%, and 55%) and centrifuged at 160,000 × g for 18 hours. Two milliliter samples were collected from each of the sucrose density steps and diluted with 1 mM EDTA to 30 mL. Particles were pelleted at 100,000 × g for 2 hours, and the pellets were each resuspended in 0.16 mL 1 mM EDTA (gradient-collected samples; Fig. S1C). EV samples were immediately frozen in liquid nitrogen. Since most EVs migrated to the 55% density fraction (Fig. S8B), we then collected EVs across fractions 3-5, which were less variable in ζ-potential, and size compared to fractions 1 and 2 ( Fig. S8C and D).

Extraction of leaf apoplastic fluids
Apoplastic fluids were collected from leaves of 6-7-week-old plants (Col-0 WT, mock treated, and infected with WT Pto DC3000). Rosettes of 22-29 plants were vacuum infiltrated with particle-free 1 mM EDTA. After removing the excess buffer, infiltrated leaves were placed into 20 mL syringes and centrifuged in 50 mL conical tubes at 900 × g for 20 minutes at 4°C. The resulting apoplastic wash was passed through a 0.22-µm membrane (apoplastic fluid samples). To confirm the successful filtering of apoplastic fluids, 5 µL was incubated on LB plates containing 50 µg/mL rifampicin. No Pto DC3000 colonies were detected after 3 days. For SEM pictures and immunoblots detecting ß-lactamase and TET8, apoplastic fluids were additionally ultracentrifuged for 1 hour at 100,000× g. The pellet was resuspended in 100 µL particle-free 1 mM EDTA.

EV quantification, size, charge measurements, and proteinase K treatment
EVs were quantified and had their size and charge measured by NTA using ZetaView BASIC PMX-120 (Particle Metrix, Germany) at room temperature. To detect EVs, we used the manufacturer's default settings for liposomes. Particle quantification and size measurements were performed by scanning 11 cell positions each and capturing 30 frames per position with the following settings: focus: autofocus; camera sensitivity for all samples: 85; shutter: 100; scattering intensity: detected automatically. After capture, the videos were analyzed by the built-in ZetaView Software 8.05.11 (ZNTA) with the following specific analysis parameters: maximum area: 1,000; minimum area: 5; minimum brightness: 25; trace length: 15 ms; hardware: embedded laser: 40 mW at 488 nm; camera: CMOS. For particle charge measurements, the same settings were used except minimum brightness: 30. Statistical analysis was performed using either one-way analysis of variance (ANOVA) with Tukey post hoc test or Welsch's ANOVA with Dunnett's T3 multiple comparisons post hoc test.
For proteinase K treatment, EVs were incubated with 10 µg/mL proteinase K (NEB, P8107S) or mock treated for 30 minutes at 37°C before boiling in Lämmli buffer at 95°C for 5 minutes.

Propidium iodide staining
The viability assay was done with some modifications according to reference (73). In brief, Pto DC3000 cultures were grown until OD 600 = 1-2 and OD 600 = 3-4. Propidium iodide (PI) (Sigma-Aldrich) was added to a final concentration of 20 µM. After 10 minutes incubation time, 5 µL of the stained cultures was transferred to a microscopy slide, and pictures were obtained with a Leica 3D Assay THUNDER Imager (Leica, Wetzlar) using an HC PL Fluotar L 40×/0.60 dry objective. PI was excited at 642 nm, and the emission range was 100%. As a negative control, bacteria were boiled in a microwave for several minutes before PI staining. Two technical replicates were performed for OD 600 = 1-2 and OD 600 = 3-4, respectively.

Scanning electron microscopy
Planktonic-grown bacteria at OD 600 = 3-4 (1.5-2 × 10 9 cfu/mL), gradient-collected EVs (0.5-1.5 × 10 10 particles), and apoplastic fluids passed through 0.2 µm filters were used for SEM. The cells were chemically fixed using 2.5% glutaraldehyde in 50 mM cacodylate buffer (pH 7.0) containing 2 mM MgCl 2 . Then, the cells were applied to a glass slide, covered with a cover slip, and plunged frozen in liquid nitrogen. After this, the cover slip was removed, and the cells were placed in a fixation buffer again. After washing four times with buffer, post-fixation was carried out with 1% OsO 4 for 15 minutes. Two additional washing steps with buffer were followed by three times washing with double distilled water. The samples were dehydrated in a graded acetone series, critical point dried, and mounted on an aluminium stub. To enhance conductivity, the samples were sputter coated with platinum. Microscopy was carried out using a Zeiss Auriga Crossbeam workstation at 2 kV (Zeiss, Oberkochen, Germany). The vesicle size was manually measured across five randomly selected SEM micrographs using Fiji software (74).

Transmission electron microscopy
Planktonic-grown Pto DC3000 at OD 600 = 3-4 (1.5-2 × 10 9 cfu/mL) was used for ultrathin sectioning and subsequent TEM. The cells were concentrated by centrifugation, and the cells were high-pressure frozen using a Leica HPM100 (Leica Microsystems, Wetzlar, Germany). This was followed by freeze substitution with 0.2% osmium tetroxide, 0.1% uranyl acetate, and 9.3% water in water-free acetone in a Leica AFS 2 (Leica Microsys tems, Wetzlar, Germany) as described previously (75). After embedding in Epon 812 substitute resin (Fluka Chemie AG, Buchs Switzerland), the cells were ultrathin sectioned (50-100 nm thickness) and post-stained for 1 minute with lead citrate. TEM of ultrathin sections was carried out with a JEOL F200 cryo-S(TEM), which was operated at 200 kV and at room temperature in the TEM mode. Images were acquired using a bottommounted XAROSA 20 mega-pixel CMOS camera (EMSIS, Münster, Germany).
For immuno-negative staining, freshly purified EVs from Pto DC3000 were used. Herein, 15 µL sample was applied to 175 mesh nickel grids, which had been covered with collodium plastic foil and coated with carbon in advance. After incubation for 5 minutes, the grids were blocked for 25 and 30 minutes with 0.1% BSA (bovine serum albumin) in 1× PBS. After this, the grids were incubated with the primary antibody (rabbit anti-P. aeruginosa ampC polyclonal antibody, dilution 1:1,000 or 1:5,000 in 1× PBS) for 30 minutes. This was followed by washing six times for 5 minutes with 0.1% BSA in 1× PBS before adding the secondary antibody (goat-anti-rabbit, coupled to 10 nm colloidal gold, dilution: 1:20) for 30 minutes. After this, the grids were washed 2 × 5 minutes with 0.1% BSA in 1× PBS, 2 × 5 minutes with 1× PBS, and 2 × 5 minutes with sterile water. After blocking with a filter paper, the samples were negatively stained with 1% uranyl acetate for 2 minutes, blotted again on a filter paper, and air dried.
As it is expected that the epitope for immunodetection is on the inside of the EVs, we carried out a heat treatment in parallel to break the EVs open. For this, 15 µL of the sample was applied to 400 mesh carbon-coated copper grids and incubated for 20 minutes at 120°C. From this point on, the treatment of the grids was identical to the protocol above.

Pto DC3000 infection assay
Overnight plate-grown Pto DC3000 cells were resuspended in 10 mM MgCl 2 and diluted to OD 600 = 0.0006. Using a needle-less syringe, the bacterial suspension was infiltrated into mature leaves of 5-6-week-old plants, three leaves per plant. For pretreatments, gradient-collected EVs from planktonic Pto DC3000 (concentration ≈1.10 10 ) and 0.02 mM EDTA as a negative control and 100 nM flg22 (EZbiolabs) as a positive control were syringe infiltrated into leaves 24 hours prior to Pto DC3000 inoculation. Discs of the infected leaves (one disc per leaf, 0.6 cm diameter) were excised at 1, 2, or 3 dpi. The three leaf discs from each plant were pooled and ground in 1 mL 10 mM MgCl 2 . Serial dilutions were plated on LB medium with rifampicin (50 µg/mL), and bacterial colonies were counted 1 day after incubation at 28°C. Statistical analysis was performed using a two-tailed Welsch's t-test.

Fluorimetric GUS assay
For fluorimetric GUS assays, 11-12-day-old seedlings were treated with gradient-collec ted Pto DC3000 EVs (concentration ≈1.10 10 ) or with 100 nM flg22 (EZbiolabs) or as a control with 0.02 mM EDTA for 18 hours. Treated seedlings were frozen in liquid nitrogen in 2 mL conical tubes containing two clean sterile glass beads and liquid nitrogen. The frozen samples were dry homogenized using a Retch mixer mill (Retch). Homogenized samples were kept on ice and cold (4°C). For total protein extraction, GUS extraction buffer was added as described (78) [50 mM sodium phosphate (pH 7); 10 mM 2-mer captoethanol; 10mM Na 2 EDTA; 0.1% Triton X-100; 0.1% sodium lauryl-sarcosine and PPIC (plant protease inhibitor cocktail)]. GUS activities were measured fluorimetrically in reaction buffer (see below) using methylumbelliferyl-β-D-glucuronic acid dihydrate (MUG) (Biosynth) as a substrate. Reaction buffer was the same solution as extraction buffer with one modification: PPIC was replaced by 1 mM MUG. The fluorescence was measured using TECAN fluorimeter at excitation 360 nm and emission 465 nm. The enzymatic activity of the sample was calculated to protein concentration measured by Bradford protein assay. The absorbance was measured using TECAN spectrometer absorbance at 595 nm. Statistical analysis was performed using one-way ANOVA with Tukey post hoc test.

RNA extraction and RT-qPCR analysis
Gene transcription analysis was performed with 12-day-old seedlings. The seedlings were treated with gradient-enriched EVs (concentration 1.10 10 ) and 0.02 mM EDTA as control for 3 hours, frozen in liquid nitrogen, and ground with 2.5-mm-diameter silica beads using a homogenizer (Retch, Germany). Total RNA was isolated using a TRIzol reagent (Invitrogen, USA) according to the manufacturer's protocol. The extracted RNA was treated with a DNA-free kit (Ambion, USA). Subsequently, 1 µg of RNA was converted into cDNA with M-MLV RNase H-Point Mutant reverse transcriptase (Promega Corp., USA) and an anchored oligo dT21 primer (Metabion, Germany). Gene transcription was quantified by qPCR using a LightCycler 480 SYBR Green I Master kit and LightCycler 480 (Roche, Switzerland). The PCR conditions were 95°C for 10 minutes followed by 45 cycles of 95°C for 10 seconds, 55°C for 20 seconds, and 72°C for 20 seconds. Melting curve analyses were then carried out. Relative transcription was normalized to the housekeep ing gene AtTIP41 (79). Primers were designed using PerlPrimer v1.1.21 (80). The primers used are AtFRK1_FP, GCCAACGGAGACATTAGAG and AtFRK1_RP, CCATAACGACCTGACT CATC. Statistical analysis was performed using one-way ANOVA with Tukey post hoc test.

Seedling growth analysis
Four-day-old seedlings were transferred from MS solid media into the liquid MS media in transparent 96-well microplates. Each well contained 100 µL of media either containing 0.02 mM EDTA as a control or gradient-collected Pto DC3000 EVs (concentration ≈1.10 10 ) or with 100 nM flg22 (EZbiolabs) as a positive control. After 8 days, the treated seedlings were dried using a paper towel and then the fresh weight was measured. Based on the weight of each seedling, relative seedling growth (%) to control seedlings was calculated. Statistical analysis was performed using Welsch's ANOVA with Dunnett's T3 multiple comparisons post hoc test two-tailed Student t-test.

ROS measurements
ROS production was determined using the luminol-based assay as previously described (81). Briefly, leaves of 5-6-week-old A. thaliana plants were infiltrated with gradient-col lected EVs (concentration ≈1.10 10 ) or particles isolated from KB. After 2 hours, discs were excised from the infiltrated leaves and 24 hours incubated in ddH2O at 22°C. Then, the leaf discs were treated with water as mock and with 100 nM flg22 or 100 nM elf18 (EZbiolabs) to induce the production of ROS. The total photon count was collected for 45 min using a TECAN luminometer. Statistical analysis was performed using a two-tailed Student t-test.
Briefly, for WC, the pellet was resuspended in 1 mL of 20 mM Tris-HCl (pH 8.0), frozen in liquid nitrogen, three times thawing-freezing, and three times sonicated for 10 minutes at 4°C. The samples were centrifuged at 6,000 × g for 10 minutes at 4°C, and supernatants were collected and frozen in liquid nitrogen.
For OM preparations, the pellet was resuspended in 1 mL 20 mM Tris-HCl (pH 8.0), sucrose (20%), followed by adding 5 µL lysozyme (15 mg/mL) and 10 µL 0.5 M EDTA, incubation for 40 minutes on ice, and adding 20 µL 0.5 M MgCl 2 . After centrifugation at 9,500 × g for 20 minutes at 4°C, the pellet was resuspended in 1 mL ice-cold 10 mM Tris-HCl (pH 8.0) followed by sonication three times for 10 minutes on ice. The samples were then centrifuged at 8,000 × g for 5 minutes at 4°C, washed with cold 10 mM Tris-HCl (pH 8.0), resuspended in cold, sterile MilliQ water followed by three times freezing thawing in liquid nitrogen, incubation for 20 minutes at 25°C, and adding the sarcosyl to final concentration 0.5%. The samples were then centrifuged at 40,000 × g for 90 minutes at 4°C, the pellet was resuspended in ice-cold 10 mM Tris-HCl (pH 8.0), and frozen in liquid nitrogen.
Gradient-collected EVs were isolated from the bacteria cultures as described above (Fig. S1C). The protein concentration in the samples was measured using Bio-Rad Protein Assay which is based on Bradford method (83).
For proteomics, the samples were denatured by addition of 1× SDS loading buffer. In-gel trypsin digestion was performed according to standard procedures (84). Briefly, 2 µg of EV and OM samples and 20 µg of WC samples were loaded on a NuPAGE 4%-12% Bis-Tris Protein gels (Thermofisher Scientific, USA), and the gels were run for 3 minutes only. Subsequently, the still not size-separated single protein band per sample was cut, reduced (50 mM DTT), alkylated (55 mm CAA, chloroacetamide), and digested overnight with trypsin (trypsin-gold, Promega).
The Fusion Lumos Tribrid mass spectrometer was operated in data-dependent acquisition and positive ionization mode. MS1 spectra (360-1,300 m/z) were recorded at a resolution of 60,000 using an automatic gain control (AGC) target value of 4e 5 and maximum injection time (maxIT) of 50 ms. After peptide fragmentation using higherenergy collision-induced dissociation, MS2 spectra of up to 20 precursor peptides were acquired at a resolution of 15,000 with an AGC target value of 5e 4 and maxIT of 22 ms. The precursor isolation window width was set to 1.3 m/z and normalized collision energy to 30%. Dynamic exclusion was enabled with 20-second exclusion time (mass tolerance ±10 ppm).

Computational analysis of proteomes
LFQ values were used in the statistical analysis of proteome data. To select EV-enriched proteins, Welch t-test was used to compare protein intensities between EV and WC samples. The resulting P-values were corrected using the Benjamini-Hochberg (BH) method to control the FDR. The proteins with FDR <0.05 and with the intensity in EV at least twice higher than in WC were selected as EV-enriched proteins (n = 207). In addition, we selected proteins that were exclusively identified in at least three (out of four) replicates of EV (n = 162). A complete list of EV-enriched proteins is given in Table S1. The functional enrichment analysis of the EV proteins was performed using the DAVID functional annotation tool (36,37). Cluster maps were generated using the SEABORN python library (https://seaborn.pydata.org/), with small cosmetic changes based on its documentation. Gene clusters were generated by SEABORN default method of hierarchical clustering. These gene clusters were also subjected to functional enrichment analysis using the DAVID functional annotation tool (36,37).

Database searches
Peptide identification and quantification were performed using MaxQuant (version 1.6.3.4) with its built-in search engine Andromeda (85,86). MS2 spectra were searched against a Pto protein database (UP000002515, downloaded from Uniprot 04.05.2020) supplemented with common contaminants (built-in option in MaxQuant). For all MaxQuant searches, default parameters were employed. Those included carbamidome thylation of cysteine as a fixed modification and oxidation of methionine and N-terminal protein acetylation as variable modifications. Trypsin/P was specified as a proteolytic enzyme. Precursor tolerance was set to 4.5 ppm, and fragment ion tolerance to 20 ppm. Results were adjusted to 1% FDR on peptide spectrum match and protein level, employing a target-decoy approach using reversed protein sequences. LFQ algorithm was enabled. The minimal peptide length was defined as seven amino acids, and the "match-between-run" function was not enabled. Each sample type (EV, OM, WC) was analyzed in biological quadruplicates (Table S1).
We used available localization prediction data from the Pseudomonas genome database (pseudomonas.com) (87). Predicted protein localizations are presented as stacked bar charts (made in MS Excel) as a percentage of the total number of proteins in the analyzed sample. We used the available software DAVID bioinformatic resource 6.8 (https://david.ncifcrf.gov/) for GO term and KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis, and the adjusted P-value cutoff was set to 0.05 (36,37). We compared the EV-enriched proteins from Pto DC3000 with EV proteomes from planktonic-grown P. aeruginosa PAO1 (51,52,82). We focused on the proteins that were identified in OMVs from P. aeruginosa PAO1 across all three studies and identified their gene orthologs in Pto DC3000 using the Pseudomonas genome database (pseudo monas.com) (87). This set of proteins was compared with the Pto DC3000 EV-enriched proteins to predict EV biomarkers. The EV-enriched proteins were also compared with available in planta Pto DC3000 transcriptome and proteome datasets (11,33).

Coomassie brilliant blue and silver staining
Proteins were separated on 10% SDS-PAGE gels using Hoefer's vertical electrophoresis system (SE250, Hoefer). The gels were subsequently either incubated with Coomassie brilliant blue G-250 staining buffer at room temperature, or the silver staining was performed using ROTI Black P kit (L533, Carl Roth) following the protocol provided by the manufacturer.

Statistical analysis
Student t-test, Welsch's t-test, one-way ANOVA followed by Tukey multiple comparisons test, and Welsch's ANOVA with Dunnett's T3 multiple comparisons post hoc test were performed using GraphPad Prism version 8.3 for Windows, GraphPad Software, San Diego, CA, USA, www.graphpad.com.

ACKNOWLEDGMENTS
We'd like to thank members of the Robatzek laboratory for fruitful discussions and Eliana Mor for support with sampling. We acknowledge Youssef Belkhadir (GMI) for sharing the anti-flagellin antibody and the Pto DC3000 ΔfliC mutant, Lucia Grenga and Catriona Thompson (JIC), Franziska Hackbarth and Hermine Kienberger (BayBioMS) for their laboratory assistance, Miriam Abele for her mass spectrometric support at the BayBioMS as well as Jennifer Grünert and Cornelia Niemann (LMU Biocenter) for technical assistance in electron microscopy, and Oksana Iakovenko (USB) for help with taking seedling pictures.

DATA AVAILABILITY
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (89) partner repository with the dataset identifier PXD023971.

ADDITIONAL FILES
The following material is available online.