Virulence from vesicles: Novel mechanisms of host cell injury by Escherichia coli O104:H4 outbreak strain

The highly virulent Escherichia coli O104:H4 that caused the large 2011 outbreak of diarrhoea and haemolytic uraemic syndrome secretes blended virulence factors of enterohaemorrhagic and enteroaggregative E. coli, but their secretion pathways are unknown. We demonstrate that the outbreak strain releases a cocktail of virulence factors via outer membrane vesicles (OMVs) shed during growth. The OMVs contain Shiga toxin (Stx) 2a, the major virulence factor of the strain, Shigella enterotoxin 1, H4 flagellin, and O104 lipopolysaccharide. The OMVs bind to and are internalised by human intestinal epithelial cells via dynamin-dependent and Stx2a-independent endocytosis, deliver the OMV-associated virulence factors intracellularly and induce caspase-9-mediated apoptosis and interleukin-8 secretion. Stx2a is the key OMV component responsible for the cytotoxicity, whereas flagellin and lipopolysaccharide are the major interleukin-8 inducers. The OMVs represent novel ways for the E. coli O104:H4 outbreak strain to deliver pathogenic cargoes and injure host cells.

Protein composition of E. coli O104:H4 OMVs. Protein profiles of LB226692 and C227-11Φ cu OMVs determined by gel electrophoresis followed by silver staining were similar albeit not identical (Fig. 1f). Nano-LC-MS/MS analysis identified 77 proteins (67 in both OMV preparations), including outer membrane, periplasmic, inner membrane, cytoplasmic, extracellular, and proteins with unknown localisation (Supplementary Table S3). The percentual distribution of OMV-associated proteins according to their subcellular localisation is shown in Fig. 1g. Stx2a and H4 flagellin were the major virulence factors identified in LB226692 OMVs, whereas only flagellin was found in C227-11Φ cu OMVs (Supplementary Table S3).
OMVs contain a subset of virulence factors of E. coli O104:H4 outbreak strain. To identify additional E. coli O104:H4 virulence factors within OMVs, we analysed OMVs and OMV-free supernatants from strains LB226692 and C227-11Φ cu for outer membrane protein A (OmpA; an OMV marker), Stx2a, ShET1, Pic, SigA, SepA, AAF/I A subunit, and H4 flagellin by immunoblot. Stx2a was identified both in OMVs (30% of the total Stx2a produced by strain LB226692) and in OMV-free supernatant (70%) (Fig. 2a). H4 flagellin was almost equally distributed between OMVs and OMV-free supernatants of strains LB226692 (44% and 56%, respectively) and C227-11Φ cu (41% and 59%, respectively) (Fig. 2a). In contrast, ShET1 was solely associated with OMVs, whereas Pic, SigA, SepA, and AAF/I A subunit were found only in OMV-free supernatants (Fig. 2a). To gain deeper insight into the association of Stx2a, ShET1 and H4 flagellin with OMVs, we performed OptiPrep density gradient fractionation of LB226692 OMVs and analysed the resulting fractions for OmpA and the virulence proteins by immunoblot. OMVs were identified in fractions 1 to 7 (Fig. 2b). The first six fractions also contained Stx2a and ShET1, which were absent from OmpA-free fractions 8 to 12 (Fig. 2b). Flagellin was identified in fractions 4 to 12; almost half (45%) of flagellin was present in OMV-containing fractions 4 to 7 (Fig. 2b), and was thus, like Stx2a and ShET1, presumably tightly associated with OMVs. The tight association of these virulence factors with OMVs was confirmed by dissociation assays. Treatment with salt (1 M NaCl), alkaline (0.1 M Na 2 CO 3 ) or chaotropic reagent (0.8 M urea) did not release any of the virulence proteins from OMVs (Fig. 2c). This was only possible by treatment with 1% SDS, which completely disrupts OMV membranes (Fig. 2c). Altogether, these experiments demonstrated that LB226692 OMVs consist of several subpopulations, which differ by their virulence factors cargoes, in particular by the presence of flagellin (Fig. 2b,d). The distribution of OMVs, ShET1, and flagellin in OptiPrep gradient fractions of OMVs C227-11Φ cu was similar to that in OMVs LB226692, but Stx2a was absent. To investigate biological effects of the OMV-associated virulence factor cocktail on human IECs, we used in further experiments pools of the fractions 1 to 7 as purified OMVs. The amounts of the total protein, Stx2a, H4 flagellin and O104 LPS in the purified OMVs are shown in Supplementary Table S4. Localisation of virulence factors within OMVs. Electron microscopy of purified LB226692 OMVs using anti-Stx2a and anti-H4 immunogold staining visualised Stx2a mostly inside OMVs and occasionally in association with OMV membranes (Fig. 3a). Flagellin was mostly located on the exterior of OMVs (Fig. 3b). To verify these results, we subjected the OMVs to proteinase K (PK) digestion, in which proteins inside OMVs are protected from degradation 12 . This approach allowed us to determine also the localisation of ShET1 (the antibody against which was not suitable for electron microscopy). The inability (a) Distribution of virulence factors of the outbreak strain in OMVs and OMV-free supernatants. OMVs and OMV-free supernatants from strains LB226692 and C227-11Φ cu were analysed by immunoblot with antibodies against OmpA (an OMV marker) and the indicated virulence proteins (signals obtained from OMVs and OMV-free supernatants tested on the same membrane are separated by a vertical line). (b) Distribution of Stx2a, ShET1 and H4 flagellin in OptiPrep density gradient fractions of OMVs LB226692 demonstrated by immunoblot. The numbers above the blots (from left to right) indicate the order of the fractions in which they were collected from top to bottom of ultracentrifugation tubes. The lanes designated OMV contain non-fractionated OMVs (positive control). (c) Dissociation assay. Pools of LB226692 OMV OptiPrep fractions 1 to 6 and 4 to 7, respectively, were incubated in HEPES buffer alone (control), or in HEPES buffer with the indicated chemicals. After ultracentrifugation, pellets (P; containing OMVs) and supernatants (S; containing proteins released from OMVs) were immunoblotted with anti-Stx2a or anti-ShET1 (pool of fractions 1 to 6) or anti-H4 (pool of fractions 4 to 7) antibodies. Signals in panels (a-c) were visualised with Chemi Doc XRS imager, and quantified (in panels (a,b)) using Quantity One ® software. of PK to degrade Stx2a and ShET1 in intact OMVs indicated that these proteins are located intravesicularly (Fig. 3c). In contrast, flagellin was degraded by PK, indicating that it is exposed on the exterior of OMVs (Fig. 3c). The PK protection assay thus corroborates the electron microscopic observations. E. coli O104:H4 OMVs bind to human IECs and are internalised via dynamin-dependent endocytosis. To determine whether LB226692 and C227-11Φ cu OMVs bind to and are internalised by IECs, the OMVs labelled with the fluorescent membrane dye 3,3 dioctadecyloxacarbocyanine perchlorate (DiO) were incubated with cells and the kinetics of their uptake was monitored by flow cytometry. Internalised OMVs were distinguished from cell-bound, but non-internalised OMVs by quenching extracellular DiO-OMV fluorescence with trypan blue (TB) 14 . DiO-OMVs from each strain bound to and were internalised by each cell line in a time-dependent manner (Fig. 4a-c). Notably, the abilities of OMVs to bind and enter the cells substantially varied between the lines as indicated by the different mean fluorescence intensities after 24 h on Caco-2 (without TB ~ 65, with TB ~ 20), HCT-8 (without TB ~ 40, with TB ~ 12) and HT-29 cells (without TB ~ 20, with TB ~ 10) (Fig. 4a-c). However, despite these cell-specific variations, there were no significant differences in cellular binding and internalisation  (a-c) Kinetics of OMV binding and internalisation. Caco-2 (a) HCT-8 (b) and HT-29 cells (c) were incubated with DiO-labelled OMVs from strains LB226692 or C227-11Φ cu for 24 h and fluorescence was measured at the times indicated using FACScan flow cytometer before (total cellassociated OMVs) and after trypan blue (TB) quenching (internalised OMVs). The data were analysed using CellQuest TM Pro software, expressed as geometric means of fluorescence intensities from 10,000 cells after subtraction of background fluorescence of cells without OMVs, and are presented as means ± standard deviations from three experiments. (d-g) CLSM of OMV uptake. Caco-2 (d) HCT-8 (e) and HT-29 (f) cells were incubated with OMVs from strains LB226692 or C227-11Φ cu for the times indicated. OMVs (green) were detected with anti-E. coli O104 LPS antibody and Alexa Fluor 488-conjugated goat anti-rabbit IgG, actin (red) with phalloidin-TRITC and nuclei (blue) with DRAQ5. (g) Control cells incubated for 24 h with OMV buffer instead of OMVs and stained as above. Pictures were taken using a laser-scanning microscope (LSM 510 META microscope, equipped with a Plan-Apochromat 63x/1.4 oil immersion objective). All three images were merged and confocal Z-stack projections are included in panels (d-f). The cross hairs show the position of the xy and yz planes. Scale bars are 10 μ m. (h-j) Effects of inhibitors of endocytosis on OMV uptake. LB226692 and C227-11Φ cu OMVs labelled with rhodamine isothiocyanate B-R18 were incubated for 6 h with Caco-2 (h), HCT-8 (i) and HT-29 (j) cells which had been pretreated (30 min) with the indicated inhibitors or remained untreated. After solubilisation of cells with Triton-X-100 fluorescence was measured (FLUOstar OPTIMA) and OMV uptake (reflected by the fluorescence intensity) in the presence of each inhibitor was expressed as the percentage of OMV uptake by control, inhibitor-untreated cells (100%). **P < 0.01, and ***P < 0.001 compared to inhibitor-untreated cells (one-way ANOVA). Data are means ± standard deviations from three independent experiments.
The time-dependent cellular uptake of OMVs was confirmed by confocal laser scanning microscopy (CLSM). After 15 min, single OMVs were associated with cell membranes (Fig. 4d-f, panels 15 min). After 6 h and 24 h, the amounts of cell-associated OMVs increased; OMVs were internalised and accumulated in perinuclear regions ( Fig. 4d-f, panels 6 h and 24 h). No OMV signals were observed in control cells incubated for 24 h with OMV buffer instead of OMVs (Fig. 4g).
To gain insight into the mechanism of the OMV uptake, we analysed effects of inhibitors of endocytosis on this process. Dynasore, an inhibitor of dynamin-1 15 , inhibited the OMV uptake to ≤ 25% of their uptake by control, inhibitor-untreated cells (P < 0.001) (Fig. 4h-j). Significant inhibition (to less than 70% of control in each cell line) (P < 0.01) was also caused by chlorpromazine and hypertonic (0.45 M) sucrose, respectively ( Fig. 4h-j), which both inhibit clathrin-mediated endocytosis 16,17 . In contrast, filipin and nystatin, which disrupt lipid rafts and caveolae 18,19 , and inhibitors of macropinocytosis including amiloride and cytochalasin D 20 had no effects on OMV uptake in any cell line ( Fig. 4h-j). These experiments indicated that LB226692 and C227-11Φ cu OMVs are internalised via dynamin-dependent endocytosis, which might be clathrin-mediated.
OMVs deliver Stx2a, ShET1 and flagellin intracellularly and are toxic to human IECs. Presence of OMVs and the OMV-associated virulence factors (Stx2a, ShET1 and H4 flagellin) in cell lysates after 6 h of incubation of cells with OMVs was demonstrated using immunoblot (Fig. 5a). The intracellular localisation of the virulence factors was confirmed by CLSM which demonstrated that all of them colocalised with OMVs ( Fig. 5b-d). Neither OMVs nor any of the virulence factors were detected in control cells exposed to OMV buffer instead of OMVs ( Supplementary Fig. S1).
To determine if the OMV-delivered Stx2a is biologically active, we tested LB226692 and C227-11Φ cu OMVs for their toxicities towards IECs and highly Stx-sensitive Vero cells 21 which also internalised the OMVs (Fig. 5a). LB226692 OMVs were toxic to each cell line (Fig. 5e). Their cytotoxicity titres were similar to those of the supernatant of a prototypic EHEC O157:H7 strain EDL933 used as a positive control, and, in accordance with low sensitivity of human IECs to Stx 22,23 , they were significantly lower on IECs than on Vero cells (Supplementary Table S5). Preincubation of OMVs with Stx2a-neutralising antibody 24 did not reduce their cytotoxicity titres (Supplementary Table S5). This confirmed that intravesicular Stx2a (which is not accessible to the antibody) and not free toxin (neutralisable by the antibody 24 ), which might have accidentally contaminated OMV surface during their handling, is responsible for the cytotoxicity. Stx2a-negative C227-11Φ cu OMVs were non-toxic (Fig. 5e, Supplementary Table S5), confirming that the cytotoxicity elicited by OMVs LB226692 was mediated by Stx2a.
OMV-associated Stx2a causes apoptosis of human IECs via caspase-9 and caspase-3 activation. Next, we investigated the mechanism of cell death caused by LB226692 OMVs by analysing IECs exposed to purified OMVs (containing 580 ng/ml of Stx2a) for apoptosis and necrosis by Cell Death Detection ELISA. The LB226692 OMVs induced apoptosis in each cell line, which was evident after 24 h (HCT-8 and HT-29) or 48 h (Caco-2) (Fig. 6a). They caused no considerable necrosis up to 48 h (Fig. 6b). OMVs C227-11Φ cu caused neither apoptosis nor necrosis (Fig. 6a,b) indicating that Stx2a is the major OMV component responsible for the apoptotic cell death. Accordingly, purified free Stx2a (580 ng/ml; used as a control) caused apoptosis in a similar extent as did the same dose of LB226692 OMV-associated Stx2a (Fig. 6a).
To confirm the apoptotic effect of LB226692 OMVs and to determine the minimum amounts of OMV-associated and free Stx2a capable of triggering apoptosis, we exposed IECs to decreasing doses of the OMVs and purified Stx2a and quantified cells with hypodiploid nuclei by flow cytometry. Both LB226692 OMVs and free Stx2a elicited a significant and dose-dependent formation of cells with hypodiploid nuclei in the range of toxin doses between 580 ng/ml and 29 ng/ml in Caco-2 cells, and 580 ng/ ml and 14.5 ng/ml in HCT-8 and HT-29 cells (Fig. 6c). Altogether, these experiments demonstrated that OMV-associated Stx2a causes apoptosis of IECs as efficiently as free Stx2a.
Since the formation of hypodiploid nuclei induced by OMVs LB226692 was significantly reduced by preincubation of cells with a pan-caspase inhibitor z-VAD-fmk (Fig. 6d), we next investigated which caspases are involved in apoptosis caused by these OMVs. Caspase-9 and caspase-3 (Fig. 6e), but not caspases-2, -6 and -8 ( Supplementary Fig. S2), were activated in each cell line after 48 h of incubation with LB226692 OMVs. The caspase-9 and caspase-3 activation was inhibited by cell pretreatment with the specific caspase inhibitor (z-LEHD-fmk and z-DEVD-fmk, respectively) (Fig. 6e). No activation of caspase-9 or caspase-3 was detected in cells treated with C227-11Φ cu OMVs (Fig. 6e). Thus, Stx2a is the key OMV component triggering the caspase activation and subsequent apoptosis.

Time-lapse digital holographic microscopy (DHM). Time-lapse DHM of living HCT-8 cells was
used to monitor in detail cellular changes caused by LB226692 OMVs. Control cells displayed regular cell divisions (Supplementary Video 1) resulting in an increase of the cell-covered area, cell density and cell layer thickness between 0 and 48 h (Fig. 7a, left panel). Dry mass of the control cell patch increased 3-fold within 48 h, whereas cells treated with LB226692 OMVs showed no divisions (Supplementary Video 1) and no gain of dry mass (Fig. 7b). Moreover, single OMV-treated cells moved out of the cell  Apoptotic cells were quantified as in c; **P < 0.01, and *P < 0.05 for comparison between non-pretreated and z-VAD-pretreated cells (unpaired Student´s t test). (e) Cells were incubated (48 h) with OMVs LB226692 (containing 580 ng/ml of Stx2a) or OMVs C227-11Φ cu (Stx2a-negative) or remained untreated. Caspase-9 and caspase-3 activities in cell lysates were determined using colorimetric substrates (LEHD-pNA and DEVD-pNA, respectively); the colour intensity, which is proportional to the level of caspase enzymatic activity, was measured spectrophotometrically at 405 nm (FLUOstar OPTIMA reader). The caspase activity in OMV-treated cells was expressed as a fold-increase of that in untreated control cells (set up as 1). Inhibitors of caspase-9 (z-LEHD-fmk) or caspase-3 (z-DEVD-fmk) were added to cells 30 min before OMVs; *P < 0.05 compared to untreated cells (one-way ANOVA). Data in all panels are shown as means ± standard deviations from three independent experiments.  Video 1). This led to the disintegration of the cell patch (Fig. 7a, right panel, 48 h). DHM thus confirmed the OMV LB226692-mediated apoptosis of HCT-8 cells.

E. coli O104:H4 OMVs induce interleukin (IL)-8 secretion by human IECs.
We next investigated purified OMVs from strains LB226692 and C227-11Φ cu for their abilities to induce production of proinflammatory cytokines by human IECs and identified the OMV components involved in this effect. Of 12 cytokines tested (see Methods), IL-8 was produced by all cell lines after exposure to each OMV preparation. The IL-8 secretion rapidly increased between 15 min and 1 to 3 h, when it either reached a plateau (HCT-8) or further increased up to 24 h (Caco-2, HT-29 cells) (Fig. 8a). The IL-8 amounts elicited by OMVs C227-11Φ cu (containing 80 ng/ml of flagellin and 950 ng/ml of LPS) were usually higher, though not always significantly, than those induced by LB226692 OMVs (containing the same amounts of flagellin and LPS plus 58 ng/ml of Stx2a) (Fig. 8a). This suggested that Stx2a is dispensable for OMV-mediated IL-8 secretion. This was further supported by the very low IL-8 amounts elicited after 24 h by purified Stx2a (58 ng/ml) from Caco-2 (5 pg/ml), HCT-8 (69 pg/ml), and HT-29 (42 pg/ml) cells.
To explore the roles of H4 flagellin and O104 LPS in the OMV-induced IL-8 secretion, we first compared the IL-8 responses elicited after 24 h by OMVs and by the same amounts of isolated flagellin and O104 LPS (80 ng/ml and 950 ng/ml, respectively). The IL-8 response elicited by isolated flagellin was similar to (HCT-8 and HT-29) or greater than (Caco-2) those induced by LB226692 and C227-11Φ cu OMVs (Fig. 8b, columns without antibodies). The IL-8 response induced by isolated O104 LPS was significantly lower than that elicited by each OMV preparation in each cell line (Fig. 8c, columns without polymyxin B). Second, we analysed the impact of inhibiting the flagellin-mediated and LPS-mediated signalling, respectively, which underlies IL-8 secretion 25,26 on the OMV abilities to induce IL-8 response. To this end, we preincubated OMVs (or isolated H4 flagellin as a control) with anti-H4 antibody and/or cells with anti-Toll-like receptor (TLR) 5 antibody to inhibit the flagellin-TLR5 signalling. Alternatively, we preincubated OMVs (or isolated O104 LPS as a control) with polymyxin B, which inhibits LPS recognition by the TLR4/MD-2 complex 27,28 . TLR5, as well as TLR4 and MD-2, were expressed by all cell lines (Fig. 8e). Both anti-H4 and anti-TLR5 antibody significantly reduced the OMV-induced and flagellin-induced IL-8 responses; a combined use of both antibodies almost completely inhibited IL-8 secretion (Fig. 8b). Preimmune sera had no inhibitory effects (Supplementary Fig. S3). Moreover, polymyxin B also significantly reduced the IL-8 amounts induced by OMVs and O104 LPS (Fig. 8c). Altogether, these data indicate that both H4 flagellin and O104 LPS contribute to the OMV O104:H4-induced IL-8 secretion by human IECs, and that TLR5-and TLR4/MD-2-mediated signalling, respectively, is involved in this process. The OMV-induced IL-8 response was dose-dependent in each cell line; the lowest OMV doses which elicited a measurable IL-8 secretion after 24 h contained 1 ng/ml of H4 flagellin and 11.88 ng/ml of O104 LPS (Fig. 8d).

Discussion
The highly virulent E. coli O104:H4 hybrid, which caused the largest and most devastating HUS outbreak in history 1 , releases a cocktail of its virulence factors via OMVs. The OMVs bind to and are internalised by human IECs, deliver the OMV-associated virulence factors intracellularly, and induce caspase-9-mediated apoptosis and an inflammatory response, in particular secretion of IL-8. Stx2a is the key OMV component responsible for the cytotoxicity, whereas flagellin and LPS are the major factors involved in IL-8 secretion.
The Stx2a secretion via OMVs described here and reported previously for Stx2a of EHEC O157:H7 29 demonstrates that besides release of free Stx2a from bacteria via stx-phage-mediated lysis 30 , the OMVs represent a novel efficient mechanism for bacterial toxin release. Importantly, we show for the first time that OMV-associated Stx2a is taken up by the host cells and causes cytotoxicity as efficiently as free Stx2a. In contrast to free Stx2a, which requires the presence of globotriaosylceramide (Gb3) receptor for its cellular binding, internalisation and exerting cytotoxicity 31 , the Stx2a-containing OMVs bind to and are internalised by IECs independently of Stx2a and thus, plausibly, independently of Gb3. This is supported by: (i) similar rates of cellular binding and internalisation of Stx2a-containing and Stx2a-lacking OMVs; (ii) intravesicular localisation of Stx2a, which makes the OMV-associated toxin inaccessible for the receptor binding; and (iii) the inability of filipin and nystatin, which disrupt lipid rafts 18,19 , the site of Gb3 clustering in Stx-sensitive cells 31,32 including IECs used in our study 32,33 , to inhibit OMV cellular uptake.
By its dispensability for OMV cellular binding and internalisation, the OMV-associated Stx2a essentially differs from other AB 5 toxins, namely OMV-associated heat-labile enterotoxin (LT) of enterotoxigenic E. coli 34 and cholera toxin 35 , each of which mediates cellular binding and internalisation of the toxin-containing OMVs via interaction with its specific receptor GM1 34,35 . In contrast, the Stx2a-carrying OMVs serve as vehicles for toxin-independent Stx2a intracellular delivery. Thus, the association of a subset of Stx2a secreted by the E. coli O104:H4 outbreak strain and by classical EHEC 29 with OMVs suggests a new, hitherto unappreciated, mechanism for a putative toxin´s interaction with cells which do not express Gb3. The roles of Stx2a-containing OMVs in interactions of the toxin with Gb3-negative human colon epithelium 33,36 , specifically in the toxin delivery into colonocytes where it was observed in EHEC-infected patients 37 , and in translocation of the toxin across the intestinal barrier 37,38 , which is Scientific RepoRts | 5:13252 | DOi: 10.1038/srep13252 critical for its systemic spread into the kidneys during HUS 39 , warrant further investigations. Interestingly, the Stx2a delivery into human intestinal epithelial cells via bacterial OMVs parallels the recently reported systemic transfer of Stx2a and its delivery into glomerular endothelial cells, the major toxin targets during HUS, via microvesicles derived from human blood cells 40 .
In contrast to its dispensability for OMV cellular binding and internalisation, Stx2a is the key component of E. coli O104:H4 OMVs responsible for their apoptotic potential. Induction of apoptosis by and C227-11Φ cu. (b) Cells were exposed for 24 h to: OMVs or isolated H4 flagellin (80 ng/ml) in the absence of antibodies; OMVs or flagellin which had been preincubated with anti-H4 antibody; OMVs or flagellin after cell preincubation with anti-TLR5 antibody; OMVs or flagellin which had been preincubated with anti-H4 antibody after cell preincubation with anti-TLR5 antibody. IL-8 secretion was quantified as above; **P < 0.01 or ***P < 0.001 (one-way ANOVA) for comparison between each OMV preparation or flagellin without and after preincubation with each respective antibody combination. (c) IL-8 secretion was quantified in cells incubated for 24 h with OMVs or isolated O104 LPS (950 ng/ml) without polymyxin B (polyB-) or with OMVs or LPS which had been preincubated with polymyxin B (polyB+ ); *P < 0.05 (one-way ANOVA) for comparison between each OMV preparation or LPS without and after polymyxin B preincubation; ** P < 0.01 (one-way ANOVA) for comparison between each OMV preparation and isolated O104 LPS. (d) Cells were incubated for 24 h with 10-fold dilutions of OMVs LB226692 or C227-11Φ cu containing the indicated amounts of flagellin and LPS; IL-8 secretion was quantified as above. Data in panels (a-d) are means ± standard deviations from three independent experiments. (e) Detection of TLR5, TLR4 and MD-2 in lysates of Caco-2, HCT-8, and HT-29 cells using immunoblot. Signals were visualised with Chemi Doc XRS imager. Sizes of immunoreactive bands are shown on the right side. Crops of representative immunoblots are shown. Full immunoblots are shown in Supplementary Fig. S10.
OMV-associated Stx2a is in accordance with the ability of free, purified Stx2a to cause apoptosis of IECs (Fig. 6a,c) and other cell types (reviewed in 41 ), as well as with reports of Stx-attributable apoptosis of the renal tubular and endothelial cells in patients with HUS caused by EHEC 42 and E. coli O104:H4 outbreak strain 43 . Contributions of OMV-associated Stx2a to the apoptosis of the colonic and caecal epithelium and subsequent diarrhoea reported in infant rabbits infected with the outbreak strain 7 need to be determined.
The role of flagellin in the virulence of E. coli O104:H4 has not been investigated, but flagellin contributes to the virulence of both EHEC and EAEC by upregulating secretion of proinflammatory cytokines by intestinal epithelial cells 25,36,[44][45][46][47] . In EHEC, this process facilitates Stx translocation across the intestinal epithelium, which is the prerequisite for the toxin´s entry into the blood stream and reaching its target organs 39 . We demonstrate that OMVs released by the E. coli O104:H4 outbreak strain induce IL-8 secretion by IECs and that H4 flagellin and O104 LPS are essential OMV components involved in this process. This is in agreement with observations reported for OMVs from Pseudomonas aeruginosa 28 . We also show that the TLR5-and TLR4/MD-2-mediated signalling underlies the IL-8 response elicited by OMV-associated H4 flagellin and O104 LPS, respectively.
Notably, Stx2a plays, based on our data, little or no role in OMV-mediated inflammatory responses by IECs as indicated by similar IL-8 amounts elicited by OMVs with and without Stx2a (Fig. 8a) and by minimal IL-8 amounts elicited by purified Stx2a. Our findings for OMVs are in agreement with studies using EHEC bacteria, in which flagellin, and not Stx, is the major bacterial factor that upregulates proinflammatory cytokine production by cultured IECs (HCT-8, Caco-2) 44,45 . In other studies, however, purified Stx1a and Stx2a induced IL-8 production by IECs either alone 48 or in synergy with flagellin 46 . These variant observations may result from different experimental conditions. We could not confirm, using OMVs from strain C227-11Φ cu and nonpolarised IECs, the previous observation on polarized T84 cells 8 that the loss of the Stx2a-encoding phage by the outbreak strain significantly reduced the IL-8 response. However, a study using human colonic xenografts indicates that flagellin and not Stx2a is the main inducer of proinflammatory cytokine production by human colonic epithelium in vivo 36 .
In summary, OMVs represent yet unrecognized powerful tools of the E. coli O104:H4 outbreak strain for the delivery of its virulence factors into IECs and causing cell injury and inflammatory responses. Our data have also implication for the use of OMVs as potential vaccine candidates to prevent disease caused by E. coli O104:H4. The mechanisms of intracellular trafficking of OMVs and the associated virulence factors, as well as the signalling pathways underlying apoptosis and IL-8 secretion induced by OMVs clearly warrant further investigations.

Methods
Ethics statement. This study was performed in accordance with guidelines approved by the Ethical Committee of the Medical Faculty of the University of Muenster and of the Aerztekammer Westfalen-Lippe. Our institutional review board waived the need for written informed consent from the participants.
Kinetics of OMV production. Kinetics of OMV production was determined as described 14 . Briefly, samples of overnight LB broth cultures were taken each hour between 1 h and 12 h, after 16 h and after 24 h, and OMVs and OMV-free supernatants were prepared as above. Aliquots (~5 μ g of protein/lane) were separated by sodium dodecylsulfate polyacrylamide gel electrophoresis (SDS-PAGE) and imunoblotted 14 with anti-OmpA antibody (an OMV marker). Signals were visualised with Chemi Doc XRS imager (BioRad) and quantified densitometrically (Quantity One ® ). Bacterial growth was monitored by measuring optical density at 600 nm. Protein composition of OMVs. OptiPrep gradient purified OMVs (5 μ g of protein/lane) were separated by SDS-PAGE and proteins were visualised with ProteoSilver ™ Plus Silver Stain Kit (Sigma).
Proteins were identified in collaboration with Alphalyse (Odense, Denmark) by tryptic digestion of total proteins from the gel followed by Q-TOF nano-LC-MS/MS. The MS/MS spectra were used for Mascot (www.matrixscience.com/) searching in the NBCI database. Protein subcellular localisations were determined with PsortB (www.psort.org/psortb/).

Detection of OMV-associated DNA and virulence genes. Presence of DNA in OMVs (intact,
either DNase-untreated or DNase-treated, or lysed after the DNase treatment 14 ) was analysed with Quant-iT PicoGreen dsDNA Assay (Molecular Probes) 14 . PCRs for virulence genes detection (listed in Supplementary Table S2) were performed with published primers 2,52,53 in 25 μ l volume containing 2.5 μ l of OMVs (either non-fractionated, DNase-untreated or DNase-treated 14 , or pools of OptiPrep fractions 1 to 7). DNA from strains LB226692 and C227-11Φ cu was a positive, and DNA from E. coli K-12 C600 a negative control.

Analyses of OMVs and OMV-free supernatants for virulence factors.
OMVs and OMV-free supernatants (~5 μ g of protein/lane) were separated by SDS-PAGE and immunoblotted with antibodies against OmpA, Stx2a, ShET1, Pic, SigA, SepA, AAF/I A subunit and H4 flagellin. After densitometric quantification of signals, the percentages of Stx2a and H4 flagellin associated with OMVs and present in OMV-free supernatants were calculated.

Concentrations of total protein, virulence proteins and LPS in purified
OMVs. Protein concentration was determined with Roti-Nanoquant. Stx2a and flagellin concentrations were determined by densitometric comparison of signals produced by 10 μ l of OMVs and 10 μ l of serial dilutions of purified Stx2a (protein concentration 2.8 mg/ml) and isolated H4 flagellin (protein concentration 640 μ g/ml) used to generate calibration curves. OMV LPS content was determined with LAL Chromogenic Endotoxin Quantitation Kit (Thermo Fisher Scientific). OMV association with intestinal epithelial cells. Flow cytometric analysis of OMV cellular binding and internalisation was performed with DiO (Molecular Probes)-labelled OMVs as described 14 . For confocal laser scanning microscopy (CLSM), cells were incubated with OMVs (~50 μ g of protein) or OMV buffer for 15 min to 24 h and stained with anti-E. coli O104 LPS antibody and Alexa Fluor 488-conjugated goat anti-rabbit IgG. Actin was counterstained with phalloidin tetramethyl rhodamine (phalloidin-TRITC) (Sigma) and nuclei with DRAQ5 (Cell Signalling). Preparations were analysed with a confocal laser-scanning microscope (LSM 510 META microscope, equipped with a Plan-Apochromat 63x/1.4 oil immersion objective; Carl Zeiss).