Antibiotics modulate Escherichia coli-derived bacterial extracellular vesicle production and their upregulation of ICAM-1 in human endothelial cells

ABSTRACT

Antibiotic treatment is often necessary to eliminate life-threatening bacterial infections. However, these treatments can alter production of bacterial extracellular vesicles (BEVs), which often contain pro-inflammatory biomolecules. In this study, we examined how the clinically relevant antibiotics meropenem, tobramycin, and ciprofloxacin impacted BEV production from a urinary tract infection-associated Escherichia coli strain (CFT073 [WAM2267]) and a meningitis-associated strain (K1 RS218). BEVs from both strains caused a dose-dependent increase in expression of intercellular adhesion molecule-1 (ICAM-1) in human umbilical vein endothelial cells, priming the endothelium for interactions with immune cells. Blockade of toll-like receptor 4 revealed that this receptor was responsible for BEV–endothelial interactions. Treatment with meropenem, a β-lactam antibiotic, increased production of BEVs from strain K1 RS218. Furthermore, meropenem treatment caused strain CFT073 [WAM2267] to produce BEVs with heightened stimulatory capacity, possibly by amplifying the content of lipoprotein Lpp in these BEVs as measured by mass spectrometry. To our knowledge, this is the first study examining the interplay between antibiotic treatment and the effects of the resulting BEVs on endothelial ICAM-1 expression. Our results indicate treatment risks of certain antibiotics against specific strains of E. coli and could help identify therapeutic targets to reduce BEV-mediated endothelial stimulation during infection.

Summary: Escherichia coli production of LPS-carrying vesicles changed after exposure to different classes of antibiotics. Treating endothelial cells revealed that one subset of BEVs became more stimulatory due to meropenem.

INTRODUCTION

Sepsis arises from an exaggerated immune response to infection, leading to widespread inflammation and tissue damage (Minasyan, 2019). It remains a global health crisis, with approximately 48.9 million cases and 11 million deaths reported annually, underscoring the urgent need for effective prevention and treatment strategies (Jing et al., 2022). A critical component of this inflammatory response is intercellular adhesion molecule-1 (ICAM-1), a 90 kDa member of the immunoglobulin superfamily whose expression is upregulated in response to pro-inflammatory cytokines (Lawson and Wolf, 2009). ICAM-1 is essential for leukocyte arrest and transmigration and is constitutively present on endothelial cells (Lawson and Wolf, 2009). In sepsis, these processes can lead to immune cell infiltration into the brain, exacerbating disease outcomes by increasing the likelihood of developing cognitive impairment (Andonegui et al., 2018; Mostel et al., 2019; Trzeciak et al., 2019). ICAM-1 upregulation in the endothelium may also occur due to exposure to bacterial products such as lipopolysaccharide (LPS), which is a major outer membrane component of gram-negative bacteria (Morise et al., 1999). In other words, bacteria and their secretions can prime the endothelium for interactions with leukocytes that can contribute to sepsis-related inflammatory pathologies.

Bacterial extracellular vesicles (BEVs) are small, 40–400 nm lipid-membrane-bound particles that are secreted by both gram-negative and gram-positive bacteria (Toyofuku et al., 2023). Their cargo includes various nucleic acids, signaling factors, and toxins (Brown et al., 2015). In gram-negative bacteria such as Escherichia coli (E. coli), many of these vesicles are derived from the outer membrane and are alternatively referred to as outer membrane vesicles (Guerrero-Mandujano et al., 2017). As such, they can contain membrane components such as LPS and outer membrane proteins, and these factors can promote an inflammatory response (Beveridge, 1999; Hosseini-Giv et al., 2022). One interesting possibility is that blood-borne BEVs may persist longer in the circulation compared to smaller free bacterial factors. For example, the half-life of free LPS in mice is only 2–4 min whereas mammalian extracellular vesicles have a half-life of 30–360 min in blood (Yao et al., 2016; Karaman et al., 2024). The stability of E. coli-derived BEVs may also be aided by expression of anti-phagocytic proteins that could inhibit clearance by immune cells, so there is potential for BEVs to travel longer, penetrate deeper into tissues, and thus have more opportunities to induce both acute and sustained pro-inflammatory stimulation than smaller bacterial factors (David et al., 2022; Sun et al., 2024).

Although antibiotics are critical for infection control, they can paradoxically enhance BEV release, which may deliver pathogenic factors that alter host cell responses (Michel et al., 2020; Ye et al., 2021; Torabian et al., 2025). Investigating the role of these antibiotic-induced BEVs in driving ICAM-1 upregulation is vital, as this process can exacerbate inflammation and contribute to endothelial dysfunction, a hallmark of sepsis pathogenesis (Lawson and Wolf, 2009). Notably, prior research has demonstrated that BEVs can significantly elevate endothelial ICAM-1 expression (Laakmann et al., 2023). However, to our knowledge, it has not yet been determined how antibiotic treatment modulates the ability of BEVs to activate endothelial cells and upregulate ICAM-1 expression.

We investigated two E. coli strains to determine if the effects of antibiotic treatment on BEV production could vary in a strain-dependent manner. The first was uropathogenic E. coli strain CFT073 [WAM2267], responsible for causing urinary tract infections (UTIs), which are the second most common infectious diseases in humans after respiratory infections (Noonin et al., 2022). UTIs occur when pathogenic bacteria, predominantly E. coli, invade urothelial cells, triggering inflammatory responses in the urethra that result in symptoms such as dysuria, perineal discomfort, and increased urinary frequency and urgency (Xue et al., 2021). The CFT073 strain is particularly virulent, capable of colonizing the bladder and ascending to the kidneys, which can lead to conditions such as urosepsis. Uropathogenic E. coli employ various pathogenic mechanisms including fimbrial adherence, toxin production, and evasion of host defenses (Anfora et al., 2008). The second strain, K1 RS218, was isolated from the cerebrospinal fluid of a neonate with meningitis. This strain is particularly significant due to its ability to cross the blood-brain barrier, a critical step in causing neonatal meningitis. K1 RS218 achieves this through specialized virulence factors, such as the K1 capsule, which provides resistance to host immune defenses, facilitating the development of severe inflammation and neurological damage in newborns (Teng et al., 2005; Arredondo-Alonso et al., 2023). K1 E. coli can also initiate sepsis (Yousuf et al., 2014). Like other gram-negative bacteria, both of these E. coli strains produce LPS, which can induce pro-inflammatory signaling by binding to the extracellular domain of toll-like receptor-4 (TLR4) expressed on endothelial cells, immune cells, and various other cell types (Tam et al., 2021). Multiple TLR4 inhibitors have been investigated as a means to improve outcomes from severe sepsis, but so far these approaches have been unsuccessful in clinical trials (Rice et al., 2010; Opal et al., 2013). This may be due to the complexity of sepsis and/or the involvement of various other bacterial products during infection.

To examine the impact of antibiotics on E. coli BEV production and stimulatory capacity, we selected meropenem, tobramycin, and ciprofloxacin because of their varying modes of action as well as their clinical relevance and efficacy against UTIs, urosepsis, and bacterial meningitis (Walker et al., 2022; Holubar and Meng, 2024; Deresinski, 2025; Pacifici, 2023; Cao et al., 2021). Meropenem is a broad-spectrum β-lactam antibiotic effective against severe infections. It exerts its bactericidal action by binding to penicillin-binding proteins in the bacterial cell wall, inhibiting peptidoglycan crosslinking and cell wall synthesis, which ultimately leads to cell death. Its utility in treating resistant infections is emphasized in clinical guidelines (Dhillon, 2018). Tobramycin, an aminoglycoside antibiotic, is particularly effective against gram-negative bacteria, including E. coli, making it a valuable treatment option for complicated infections (Pacifici, 2023). In clinical studies, 21 out of 30 patients with severe or complicated gram-negative UTIs were cured after a 5-day course of tobramycin, with no reported side effects (Bailey and Peddie, 1976). It has also been used to successfully treat child and adult bacterial meningitis (Pacifici, 2023). Ciprofloxacin, a third-generation fluoroquinolone, is a key treatment for UTIs, particularly those caused by E. coli (Xue et al., 2021). Its effectiveness is due to its ability to inhibit bacterial DNA gyrase, preventing DNA synthesis and arresting bacterial growth. Additionally, ciprofloxacin's high drug permeability allows it to quickly reach therapeutic concentrations in prostatic fluid and bladder urine, resulting in a rapid bactericidal effect (Xue et al., 2021).

The present study sought to understand whether these three different classes of antibiotics would have varying effects on E. coli BEV production and the capacity of the secreted BEVs to promote endothelial activation. These interactions were studied with both CFT073 [WAM2267] and K1 RS218 to determine whether there might be strain-specific differences in E. coli BEV release in response to antibiotic treatment. We found that there were indeed antibiotic- and strain-dependent differences in the BEVs produced by these E. coli that resulted in altered effects on endothelial expression of ICAM-1. The results from this study may one day translate to clinical knowledge about the risks of treating infections with certain antibiotics depending on which bacterial strains are present.

RESULTS

BEVs cause increased endothelial ICAM-1 expression in a dose-dependent manner

To investigate the potential of BEVs to promote endothelial activation, we examined expression of inflammatory marker ICAM-1 on the surface of human umbilical vein endothelial cells (HUVECs) following 16–17 h exposure to BEVs derived from two E. coli strains: CFT073 [WAM2267] and K1 RS218. Immunofluorescence staining showed a dose-dependent increase in ICAM-1 expression in response to BEVs from both strains, with significant increases above baseline at concentrations of 1E6 BEVs ml⁻¹ (approximately 3.30 BEVs per HUVEC) for the CFT073 strain and at 1E7 BEVs ml⁻¹ (approximately 37.8 BEVs per HUVEC) for the RS218 strain (Fig. 1). This indicated that the CFT073-derived BEVs were more potent than the RS218-derived BEVs, despite the fact that they were obtained from two strains of the same bacterial species. It is therefore possible that strain-specific variations exist in the composition of BEVs that can influence their stimulatory capacity. Additionally, the immunofluorescence images indicated some heterogeneous upregulation of ICAM-1 even at lower BEV concentrations, which could suggest HUVEC sensitivity to individual BEVs. For both strains, ICAM-1 upregulation following BEV exposure was comparable to that induced by 10 ng ml⁻¹ LPS, which was used as a positive control for increased ICAM-1 expression. This amount of LPS is over twenty times the concentration that has been observed in patient serum during sepsis and should therefore elicit a consistently high ICAM-1 response (Marshall et al., 2002). The results from these dosing experiments support the idea that BEVs play an active role in modulating endothelial cell function and could contribute to vascular inflammation in the context of bacterial infections. Because treatment with 1E8 BEVs ml⁻¹ from both E. coli strains routinely promoted high ICAM-1 expression by HUVECs, this concentration was selected for all of the following cell stimulation experiments except where otherwise indicated. Note that this concentration corresponded to approximately 322.6–385.4 BEVs per HUVEC in these dosing experiments, but we have decided to express concentrations in terms of BEVs ml⁻¹ throughout this manuscript to avoid the assumption that 100% of the BEVs settled to the bottom of the solution to interact with the HUVEC monolayer during the treatment period.

Fig. 1.

HUVEC ICAM-1 expression increases due to exposure to E. coli-derived BEVs in a dose-dependent manner. (A) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with increasing concentrations of BEVs derived from E. coli strain CFT073 [WAM2267] or with 10 ng ml⁻¹ LPS as a positive control for ICAM-1 expression for 16–17 h. (B) Bar plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain CFT073 [WAM2267]. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. These values were then normalized to the average for the wells that received no stimulatory treatment. (C) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with increasing concentrations of BEVs derived from E. coli strain K1 RS218 or with 10 ng ml⁻¹ LPS as a positive control for ICAM-1 expression for 16–17 h. (D) Bar plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain K1 RS218. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. These values were then normalized to the average for the wells that received no stimulatory treatment. n=3. Error bars=s.d. Performed one-way ANOVA with Dunnett post-hoc test: **P<0.01, ***P<0.001, ****P<0.0001.

E. coli-derived BEVs cause a stimulatory response by binding to TLR4 on HUVECs

To investigate whether LPS was a primary stimulatory component of these E. coli-derived BEVs, we blocked TLR4 by treating HUVECs with inhibitor TAK-242 (CAS 243984-11-4). Since TLR4 is the receptor responsible for binding and sensing LPS, this inhibition would hypothetically prevent BEVs from causing increased ICAM-1 expression in HUVECs. First, we confirmed TAK-242's specificity for TLR4 by stimulating HUVECs with either LPS or with a pro-inflammatory cytomix consisting of TNF-α, IL-1β, and IFN-γ. The cytomix components are known to bind to the receptors TNFR1, TNFR2, IL-1R1, and IFNGR rather than TLR4 and can promote ICAM-1 upregulation (Zhou et al., 2007; Kaneko et al., 2019; Ng et al., 2023). When we treated HUVECs with LPS and TAK-242 there was no increase in ICAM-1 expression, but treating HUVECs with cytomix caused significantly higher ICAM-1 expression with and without TAK-242 (Fig. S1). This demonstrated that HUVECs were still capable of upregulating ICAM-1 during TAK-242 treatment as long as they sensed non-TLR4-mediated stimuli. When we dosed HUVECs with BEVs derived from E. coli strains CFT073 [WAM2267] or K1 RS218 for 16–17 h, co-treatment with TAK-242 brought ICAM-1 expression back to baseline levels (Fig. 2A,B). Therefore, we concluded that TLR4 is the primary receptor mediating the stimulatory interactions between E. coli-derived BEVs and HUVECs, and it is likely that LPS is a major stimulatory component of these BEVs, consistent with findings reported in previous studies (Kim et al., 2013; Soult et al., 2013; Laakmann et al., 2023). To our knowledge, this is the first demonstration that blockade of TLR4 can protect HUVECs from BEV-induced upregulation of ICAM-1.

Fig. 2.

BEVs stimulate HUVECs by binding to TLR4. (A) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were pre-treated with 10 µM of TLR4 inhibitor TAK-242 or a vehicle control (0.04% DMSO) for 4–5 h. They were subsequently exposed to 10 ng ml⁻¹ LPS or 1E8 BEVs ml⁻¹ derived from E. coli strains CFT073 [WAM2267] or K1 RS218 with either the TAK-242 or vehicle control for 16–17 h. (B) Bar plot of ICAM-1 fluorescence intensity. The average fluorescence intensity of ICAM-1 in each image was background-subtracted and then divided by the number of cell nuclei in the corresponding DAPI channel. These values were then normalized to the average for the wells that received the vehicle control with no stimulatory treatment. n=3. Error bars=s.d. Performed two-way ANOVA with Tukey post-hoc test. (C) Bar plot of the LPS concentration per individual BEV as measured by a chromogenic endotoxin quantification kit. n=4 for CFT073, n=3 for RS218. Error bars=s.d. Performed Welch's t-test: **P<0.01, ****P<0.0001.

We confirmed the presence of LPS in the BEVs by performing a chromogenic quantification assay (Fig. 2C). Strain CFT073 [WAM2267] was found to have 1.80±0.37 fg LPS per BEV, which was significantly higher than the 0.84±0.17 fg LPS per BEV found for strain K1 RS218. In our prior dosing experiments, the concentrations of 1E6, 1E7, and 1E8 BEVs ml⁻¹ respectively equated to 1.8, 18, and 180 ng ml⁻¹ LPS for CFT073 [WAM2267] and 0.84, 8.4, and 84 ng ml⁻¹ for K1 RS218, which both appeared roughly comparable to the 10 ng ml⁻¹ LPS positive control (Fig. 1).

Estimating minimum inhibitory concentration values

The minimum inhibitory concentration (MIC) is the lowest concentration of an antimicrobial agent that inhibits visible bacterial growth. In this study, the MICs of meropenem, tobramycin, and ciprofloxacin for E. coli strain CFT073 [WAM2267] were determined via the broth-dilution assay, adhering to EUCAST guidelines (Table 1) (Park et al., 2023). The MICs for strain K1 RS218 had previously been determined by our group (Torabian et al., 2025).

Table 1.

Comparison of MIC values between E. coli strains CFT073 [WAM2267] and K1 RS218 for meropenem, tobramycin, and ciprofloxacin

E. coli strain	Meropenem MIC (µg ml−1)	Tobramycin MIC (µg ml−1)	Ciprofloxacin MIC (µg ml−1)
CFT073 [WAM2267]	0.025	10	0.1
K1 RS218	0.1	10	0.1

While RS218 and CFT073 showed similar MICs for tobramycin and ciprofloxacin, the MIC for meropenem was fourfold higher in RS218.

Characterization of BEVs derived from antibiotic-treated E. coli

E. coli were incubated for 3.5 h with twice the MIC (2MIC) of meropenem, tobramycin, ciprofloxacin, or no treatment (control) before BEVs were collected via differential centrifugation. After resuspending the BEVs in PBS, they were quantified by nanoparticle tracking analysis (NTA). Three biological replicates were collected for each examined condition. We observed that the average concentration of recovered BEVs from the control conditions was 3.91E10±1.81E9 particles ml⁻¹ for strain CFT073 and 1.71E11±8.69E9 particles ml⁻¹ for strain RS218 (Fig. 3A). In other words, there was nearly an order of magnitude more BEVs recovered from the RS218 cultures compared to the other strain. This trend remained true for all of the antibiotic-treated conditions: meropenem treatment resulted in recovery of 1.48E10±3.29E8 CFT073 BEVs ml⁻¹ and 2.52E11±1.05E10 RS218 BEVs ml⁻¹, tobramycin treatment resulted in recovery of 4.53E10±1.58E9 CFT073 BEVs ml⁻¹ and 1.98E11±1.02E10 RS218 BEVs ml⁻¹, and ciprofloxacin treatment resulted in recovery of 2.73E10±1.08E9 CFT073 BEVs ml⁻¹ and 1.89E11±1.06E10 RS218 BEVs ml⁻¹ (Fig. 3A). Notably, meropenem treatment resulted in a significant increase in production of BEVs by strain RS218 over the control condition, suggesting a potential link between cell wall-targeting antibiotics and increased vesiculation. However, there was no significant change in BEV production by strain CFT073 when exposed to meropenem and, in fact, there was a non-significant trend towards decreased BEV production under these conditions. This may highlight another strain-specific difference in behavior or it could be related to the different meropenem MICs that were identified for the two strains.

Fig. 3.

E. coli demonstrate strain-specific changes in BEV production after antibiotic treatment. (A) Bar plot of BEV concentrations as measured by nanoparticle tracking analysis (NTA). Performed two-way ANOVA with Tukey post-hoc test: *P<0.05, ****P<0.0001. n=3. Error bars=s.d. (B) Bar plot of BEV diameters as measured by NTA. Performed two-way ANOVA with Tukey post-hoc test and found a significant effect due to the different E. coli strains: *P<0.05. n=3. Error bars=s.d. (C) Transmission electron microscopy images of BEVs derived from E. coli strains CFT073 [WAM2267] and K1 RS218 treated with various antibiotics or no antibiotic (control). (D) Western blot of Pal in equal volumes of the BEV samples. All of the CFT073 blots have been rearranged from a single gel image, and all of the RS218 blots have been rearranged from a single gel image. For the original full gel images, see Fig. S2. (E) Bar plot of Pal band densities from western blot images. n=3 for CFT073, n=6 for RS218. Error bars=s.e.m.

BEV size measurements were also obtained from NTA (Fig. 3B). There were no significant changes in particle size due to antibiotic treatment. However, there was a significant difference in the sizes of BEVs produced by the two E. coli strains; the average diameter of BEVs derived from CFT073 was 181.8±2.063 nm while the average for RS218 was 175.0±1.346 nm. This size discrepancy was not necessarily apparent in all of the transmission electron microscopy (TEM) images that were captured for the various BEV preparations (Fig. 3C). The TEM images did, however, agree with the NTA data concerning the increase in vesicle production when strain RS218 was treated with meropenem. These TEM images also confirmed that the recovered particles featured the typical morphology of BEVs (Wen et al., 2023).

Western blot of equal volumes of the BEV preparations confirmed the presence of peptidoglycan-associated lipoprotein (Pal) in all samples (Fig. 3D). Pal was selected as a representative, stably expressed outer membrane/BEV-associated marker, allowing us to qualitatively verify BEV integrity and compare relative protein packaging across conditions. For strain CFT073, there was a non-significant decrease in Pal content due to meropenem treatment (Fig. 3E). This corresponded with the non-significant drop in the concentration of BEVs when the CFT073 E. coli were treated with meropenem as measured by NTA. However, the meropenem-induced increase in production of BEVs by strain RS218 did not cause a noticeable change in Pal band density in the western blots. It is possible that less Pal was packaged in RS218-derived BEVs due to meropenem treatment, which could be another strain-specific behavioral difference.

Some, but not all, antibiotic treatments at 2MIC reduced the bulk stimulatory capacity of E. coli-derived BEVs.

To explore the influence of antibiotic-induced BEVs on endothelial activation, we measured ICAM-1 expression in HUVECs treated with BEVs derived from E. coli strains CFT073 [WAM2267] and K1 RS218 that had been exposed to different antibiotics. HUVECs were incubated with BEVs for 16–17 h, after which ICAM-1 expression was assessed using immunofluorescence staining (Fig. 4A,C). For each strain, BEVs derived from the control condition (no antibiotic treatment) were diluted to 1E8 particles ml⁻¹. The same dilution factor was then applied to all of the BEVs obtained from the antibiotic-treated conditions such that the HUVECs were treated with equal volumes of each BEV preparation. Quantitative analysis of the fluorescence intensity revealed that BEVs from E. coli strain CFT073 [WAM2267] that had been treated with meropenem or tobramycin resulted in significantly lower ICAM-1 expression compared to the control, indicating a lower stimulatory potential (Fig. 4B). CFT073 BEVs from the ciprofloxacin conditions also demonstrated a non-significant trend towards decreased stimulatory capacity. BEVs from strain K1 RS218 that had been treated with tobramycin exhibited significantly lower ICAM-1 expression than the control, meropenem, and ciprofloxacin conditions (Fig. 4D). However, there were no significant differences between these other conditions. These findings suggest that meropenem exposure alters the stimulatory capacity of BEVs in a strain-dependent manner, potentially due to changes in BEV composition or because of the differences in BEV production demonstrated by the two strains during meropenem treatment (Fig. 3A). Conversely, tobramycin seems to consistently reduce the stimulatory potential of BEVs regardless of E. coli strain. The reduced ability of BEVs from antibiotic-treated bacteria to upregulate ICAM-1 may have implications for modulating endothelial inflammation during infections and antibiotic therapy. Furthermore, it is notable that ciprofloxacin treatment did not significantly lower the stimulatory capacity of BEVs from either strain. This indicates that, while ciprofloxacin may be effective at treating bacterial infections, the BEVs produced by the inhibited or dying bacteria may still demonstrate equal stimulatory effects to BEVs secreted by an untreated E. coli infection.

Fig. 4.

Treatment of E. coli with antibiotics can lower the stimulatory capacity of equal volume BEV preparations. (A) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with BEVs derived from E. coli CFT073 [WAM2267] that had been treated without antibiotics (control) or with 2MIC of meropenem, tobramycin, or ciprofloxacin for 16–17 h. The control BEVs were diluted to 1E8 particles ml⁻¹, and the same dilution factor was applied to the other experimental conditions to ensure that cells were treated with equal volumes of BEVs. (B) Violin plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain CFT073 [WAM2267]. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. The baseline ICAM-1 fluorescence intensity of untreated HUVECs was then subtracted, and the resulting values were normalized to the mean of the control condition. n=9, outliers removed. Performed Kruskal–Wallis with Dunn's correction: **P<0.01. (C) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with BEVs derived from E. coli K1 RS218 that had been treated without antibiotics (control) or with 2MIC of meropenem, tobramycin, or ciprofloxacin for 16–17 h. The control BEVs were diluted to 1E8 particles ml⁻¹, and the same dilution factor was applied to the other experimental conditions to ensure that cells were treated with equal volumes of BEVs. (D) Violin plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain K1 RS218. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. The baseline ICAM-1 fluorescence intensity of untreated HUVECs was then subtracted, and the resulting values were normalized to the mean of the control condition. n=9, outliers removed. Performed Kruskal–Wallis with Dunn's correction: ***P<0.001.

Treating CFT073 E. coli with meropenem increases the stimulatory capacity of individual BEVs

To investigate the impact of antibiotic treatment on the stimulatory potential of individual BEVs, we standardized the BEV concentrations to 1E8 BEVs ml⁻¹ for all conditions before exposing them to HUVECs. Immunofluorescence staining for ICAM-1 revealed strain-dependent differences in endothelial stimulation (Fig. 5A,C).

Fig. 5.

Meropenem treatment increases the potency of individual BEVs derived from E. coli strain CFT073 [WAM2267]. (A) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with BEVs derived from E. coli CFT073 [WAM2267] that had been treated without antibiotics (control) or with 2MIC of meropenem, tobramycin, or ciprofloxacin for 16–17 h. All treatment concentrations were standardized to 1E8 BEVs ml⁻¹. (B) Violin plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain CFT073 [WAM2267]. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. The baseline ICAM-1 fluorescence intensity of untreated HUVECs was then subtracted, and the resulting values were normalized to the mean of the control condition. n=9, outliers removed. Performed Kruskal–Wallis with Dunn's correction: *P<0.05, **P<0.01. (C) Immunofluorescence images of HUVECs stained with DAPI (blue) and anti-ICAM-1 antibody (yellow). Cells were treated with BEVs derived from E. coli K1 RS218 that had been treated without antibiotics (control) or with 2MIC of meropenem, tobramycin, or ciprofloxacin for 16–17 h. All treatment concentrations were standardized to 1E8 BEVs ml⁻¹. (D) Violin plot of ICAM-1 fluorescence intensity from HUVECs treated with BEVs derived from E. coli strain K1 RS218. The average fluorescence intensity of ICAM-1 in each image was divided by the number of cell nuclei in the corresponding DAPI channel. The baseline ICAM-1 fluorescence intensity of untreated HUVECs was then subtracted, and the resulting values were normalized to the mean of the control condition. n=9, outliers removed. Performed Kruskal–Wallis with Dunn's correction and found no significant differences.

For CFT073 [WAM2267]-derived BEVs, ICAM-1 expression was higher when BEVs were obtained from bacteria treated with meropenem, while BEVs from tobramycin- and ciprofloxacin-treated bacteria induced ICAM-1 expression at levels similar to the control (Fig. 5B). This suggests that meropenem exposure may enhance the stimulatory potential of BEVs, whereas tobramycin and ciprofloxacin do not substantially alter the BEVs' effect on endothelial stimulation.

In contrast, for K1 RS218-derived BEVs, ICAM-1 expression levels remained comparable across treatment groups, indicating that the stimulatory potential of BEVs from this strain was not substantially affected by antibiotic exposure (Fig. 5D). These findings highlight potential strain-specific differences in how antibiotics influence BEV cargo and their downstream effects on endothelial cells.

Antibiotic treatment modulates LPS content in E. coli-derived BEVs

To examine why meropenem treatment of strain CFT073 [WAM2267] yielded BEVs with an increased stimulatory capacity, we used a chromogenic quantification assay to measure LPS content (Fig. 6). Across the different antibiotic conditions, CFT073 [WAM2267] had significantly more LPS per BEV than K1 RS218, consistent with our previous findings. Interestingly, antibiotic treatment tended to increase the amount of LPS per BEV, with tobramycin inducing the largest increase for both strains. The amount of LPS in meropenem-associated BEVs derived from strain CFT073 [WAM2267] was significantly higher than the amount of LPS in BEVs from the untreated control, which made sense given the increased stimulatory effect of BEVs from the meropenem condition. However, these BEVs had significantly less LPS than BEVs from the tobramycin and ciprofloxacin conditions even though these latter conditions had equal or reduced stimulatory capacity compared to the meropenem-associated BEVs. This indicated that antibiotics might also modulate other BEV cargo that could be responsible for the observed alterations in HUVEC stimulation.

Fig. 6.

Antibiotic treatment can increase the amount of LPS per BEV. Bar plot of the LPS concentration per individual BEV as measured by a chromogenic endotoxin quantification kit. n=4, except RS218 Control (n=7). Error bars=s.d. Performed two-way ANOVA with Tukey post-hoc test and found a significant effect due to the different E. coli strains: *P<0.05, **P<0.01, ****P<0.0001.

Protein profiling

To further investigate why meropenem treatment impacted the stimulatory capacity of BEVs derived from E. coli strain CFT073 [WAM2267], we performed mass spectrometry and looked for changes in the composition of non-LPS BEV cargo. BEVs isolated from E. coli treated with meropenem, tobramycin, ciprofloxacin, or the untreated control were all included for comparison (Fig. 7A). This analysis allowed us to identify changes in the protein cargo of BEVs in response to different antibiotic classes, revealing potential alterations in bacterial stress responses, virulence factor secretion, and membrane-associated proteins. By comparing the proteomic profiles across conditions, we aimed to determine how antibiotic-induced changes in BEV content could influence host-pathogen interactions, particularly endothelial activation and inflammation.

Fig. 7.

Mass spectrometry demonstrates differences in BEV protein content. (A) Heatmap of the top ten most abundant BEV proteins across conditions. Relative protein abundance (% of total protein) in BEVs from E. coli exposed to meropenem, tobramycin, ciprofloxacin, or no antibiotic (control). Data reflect the top ten most abundant proteins identified by mass spectrometry. Outer membrane proteins dominated across all conditions, with antibiotic-specific shifts observed. (B) Bar plot of protein content per BEV normalized to the control condition. The reads of individual peptides were summed and divided by the BEV concentrations obtained from NTA prior to normalization. n=1.

Among the most abundant proteins detected across all conditions were outer membrane proteins (OMPs), including OmpA, NmpC, and major outer membrane lipoprotein Lpp, which are known to play critical roles in vesicle structure, host interaction, and immune activation (Rodrigues et al., 2022). OmpA remained consistently enriched across all conditions. In contrast, NmpC, a porin involved in nutrient uptake and stress adaptation (Hindahl et al., 1984), was highly enriched in control conditions but markedly reduced following antibiotic treatment. Interestingly, lipoprotein Lpp, which tethers the outer membrane to the peptidoglycan layer, was notably increased in BEVs following meropenem treatment and to a lesser degree by ciprofloxacin treatment (Schwechheimer et al., 2013, 2014). These findings indicate antibiotic-dependent modulation of the proteins present in E. coli-derived BEVs.

To investigate differences in protein packaging per BEV across the various conditions, we divided the total number of peptide reads from mass spectrometry by the BEV counts obtained from NTA (Fig. 7B). Relative to the control condition, meropenem treatment resulted in an 8.8% decrease in protein per BEV, tobramycin treatment resulted in a 40.5% decrease in protein per BEV, and ciprofloxacin treatment resulted in a 29.7% decrease in protein per BEV.

DISCUSSION

In this work, we examined how treatment of two strains of E. coli with various antibiotics affected the production and stimulatory characteristics of BEVs. The results showcased differences between two strains of the same species, which could possibly help inform strategies to mitigate sepsis stemming from either UTIs or meningitis. By modeling the interactions between BEVs and human endothelial cells, we recreated a potentially important component of sepsis that can promote endothelial inflammatory activation and thereby attract immune cells to the vascular wall. While bacteria are often responsible for initiating sepsis, the negative outcomes associated with this disease state are more commonly attributed to the host immune response, including increased production of pro-inflammatory cytokines (Jing et al., 2022). However, the interaction between bacterial products and the endothelium should not be ignored as it may play a role in destabilizing vital regions of the vasculature such as the blood-brain barrier, leading to neurological changes that could continue to impact sepsis survivors, even following clearance of the bacterial infection (Al-Obaidi and Desa, 2018; Mostel et al., 2019).

We found that HUVECs exhibited a dose-dependent increase in ICAM-1 expression following exposure to E. coli-derived BEVs. This underscores the potential risks of treating bacteria with antibiotics that could promote increased BEV production. However, we also found that the increase in BEV release by meropenem-treated E. coli from strain K1 RS218, which likely occurred due to meropenem's disruption of tethers between the outer membrane and the bacterial cell wall, did not exhibit increased stimulatory effects on the HUVECs (Michel et al., 2020). In this instance, the relationship between higher doses of BEVs and ICAM-1 expression was not one-to-one, indicating that there was either a decrease in the stimulatory capacity of the BEVs from the meropenem condition or that differences in BEV potency may only be apparent when concentrations differ by at least an order of magnitude. Similarly, ICAM-1 expression appeared to be inversely related to the dose of BEVs derived from tobramycin-treated E. coli. This directly contradicts the dose-response data, but a potential explanation stems from the fact that tobramycin itself has been demonstrated to have anti-inflammatory effects (Gziut et al., 2013). This is also true for ciprofloxacin (Lahat et al., 2007; Sachse et al., 2008). While we suspect that the majority of free antibiotics were removed from the BEV preparations following ultracentrifugation, there is a possibility that a small amount remained behind or that it had been encapsulated within the BEVs. In fact, antibiotic sequestration within BEVs can lower the concentration of free antibiotics and has been implicated as a bacterial survival strategy (Federica et al., 2025). This could account for the paradoxically lower ICAM-1 response when the HUVECs were treated by a larger number of tobramycin-associated BEVs. Furthermore, it would explain why some of the antibiotic-associated conditions failed to demonstrate increased stimulatory capacity despite incorporating more LPS per individual BEV. While a limitation of this study was that we did not quantify or remove co-isolated antibiotics from our BEV preparations, we propose that the contributions of any residual antibiotics may actually lend these experiments greater physiological relevancy, since both BEVs and antibiotics would be circulating in a patient undergoing treatment for sepsis.

One other important factor that significantly influenced the potency of the BEVs was the source strain. Despite the fact that CFT073 [WAM2267] and K1 RS218 are the same bacterial species, BEVs from the former strain were capable of increasing HUVEC ICAM-1 expression at a tenfold lower concentration than BEVs from the latter strain. While our LPS quantification data did not demonstrate a tenfold difference between BEVs from the two strains, it did confirm that more LPS was present in BEVs derived from CFT073 [WAM2267] than in BEVs from K1 RS218. This could reflect the ability of K1 RS218 to evade the host inflammatory response, for example by producing the K1 capsule, which resembles structures present on neurons and immune cells and helps to disguise the invading pathogen (Arredondo-Alonso et al., 2023). It is conceivable that BEVs produced by this strain would also possess surface motifs that could act as camouflage when the particles remain below a certain concentration threshold. There are also a number of gram-positive and gram-negative bacteria that express anti-phagocytic proteins, and packaging of anti-autophagic cargo into BEVs of E. coli strain SP15 has also been observed (David et al., 2022; Sun et al., 2024). Such species- and strain-specific differences in defenses and host evasion tactics could certainly account for the different levels of ICAM-1 upregulation observed in the present study. With that said, it is curious that K1 RS218 produced an order of magnitude more BEVs than CFT073 [WAM2267]. Similar differences have been observed between E. coli strains ED1284 and ED1374, with the latter producing 100 times more BEVs than the former (Barbieri et al., 2025). Genetic differences can result in altered BEV production by E. coli, with nlpI and degP mutants demonstrating hypervesiculation and nlpA mutants demonstrating hypovesiculation (Schwechheimer and Kuehn, 2013; Schwechheimer et al., 2015). It is likely that the pronounced disparity in BEV production observed in this study is due to the strain-specific genomes, though additional work may be required to determine which particular mutations are directly responsible. Additionally, differences in BEV production might occur if the E. coli were co-cultured directly with the human cells. It is clear that environmental stressors such as iron deficiency, which occurs in the host environment, can impact vesiculation from bacteria (Mozaheb and Mingeot-Leclercq, 2020). Interestingly, the meropenem MIC was found to be higher for K1 RS218 than for CFT073 [WAM2267]. This difference may be due to RS218's unique virulence features, such as the aforementioned K1 capsule and enhanced ability to persist within host cells, which could confer increased tolerance to cell wall-targeting antibiotics like meropenem, even in the absence of classic resistance mechanisms (Kim, 2003; Teng et al., 2005; Alkeskas et al., 2015).

Another strain-specific difference was the size of the secreted BEVs as detected by NTA. CFT073 [WAM2267] produced BEVs with an average diameter greater than the K1 RS218-derived BEVs. This could also result in a larger interaction area between individual vesicles and the endothelium, potentially leading to increased presentation of surface LPS and other inflammatory factors to the endothelial cells. This mechanism could also explain the measured differences in BEV potency.

We recognize that ultracentrifugation, while a widely used and accepted method (Saint-Pol and Culot, 2025), can allow some co-purification of cellular debris. A subset of the TEM images included flagella that were co-isolated with BEVs. Flagella can impact inflammation by stimulating cells through TLR5, and TAK-242 treatment does not interfere with this interaction (Hayashi et al., 2001; Kawamoto et al., 2008). Based on our TLR inhibition studies, the stimulation of HUVECs was due to TLR4, suggesting that the contribution of the flagellin-TLR5 pathway was negligible. An ongoing study in our laboratory comparing ultracentrifugation, ion-exchange chromatography, density gradient purification, and tangential flow filtration will ideally enable more rigorous separation of intact BEVs from other bacterial components and provide a clearer assessment of antibiotic-induced vesicle production.

TLRs are type 1 transmembrane proteins that play a key role in the immune system by recognizing various molecular signals, including pathogen-associated signals, damage-associated molecular patterns, such as calprotectin, and pathogen-associated molecular patterns, such as the aforementioned flagellin and LPS (Molteni et al., 2018). These receptors are found on immune cells, such as dendritic cells and macrophages, and non-immune cells, such as fibroblasts and endothelial cells. Upon ligand binding, TLRs initiate a signaling cascade primarily through adaptor molecules such as MyD88 or TRIF, leading to the activation of transcription factors like NF-κB (Piras and Selvarajoo, 2014). This results in the upregulation of pro-inflammatory cytokines and type I interferons, which coordinate an immune response to combat infection or cellular damage (Tam et al., 2021). TLR4 recognizes LPS from gram-negative bacteria, leading to the activation of both anti-inflammatory and pro-inflammatory pathways (Effah et al., 2024). The engagement of TLR4 by BEVs results in the upregulation of inflammatory markers such as ICAM-1, CXCL10, and IL-6, which are crucial in mediating endothelial cell activation and the subsequent inflammatory response (Soult et al., 2013; Laakmann et al., 2023).

TAK-242 is a specific inhibitor of TLR4 and has previously been demonstrated to have no functional effect on other TLRs (Kawamoto et al., 2008). It functions by binding to TLR4's intracellular domain to disrupt the interactions with the MyD88 and TRIF adaptor molecules and thereby prevent the pro-inflammatory signaling cascade (Kawamoto et al., 2008; Ono et al., 2020). In the present study, we validated TAK-242's specificity for inhibiting TLR4 by examining its ability to prevent endothelial ICAM-1 upregulation during LPS treatment but not during treatment with a pro-inflammatory cytomix consisting of equal parts TNF-α, IL-1β, and IFN-γ. These three cytokines engage receptors other than TLR4 and thus upregulate endothelial ICAM-1 expression via alternate pathways. Interestingly, there was a significant decrease in ICAM-1 expression when TAK-242 was co-administered with the cytomix compared to the cytomix alone, though the ICAM-1 signal remained far above basal levels. Even though the cytomix components promote endothelial activation via non-TLR4 receptors, it is possible for all three of these cytokines together to induce upregulation of TLR4 (Tewari et al., 2012; Liu et al., 2014; Faure et al., 2001). TLR4 overexpression, in turn, can cause an increase in ICAM-1 expression by endothelial cells (Shinohara et al., 2007). It is therefore conceivable that the change in ICAM-1 signal during TAK-242 cotreatment with cytomix was due to inhibition of excess TLR4, but the fact that ICAM-1 expression remained elevated indicated that the other pro-inflammatory pathways were still engaged.

In our experiments with BEVs, TAK-242 treatment prevented increases in ICAM-1. This indicated that the primary component of E. coli BEVs responsible for stimulating endothelial cells was likely LPS since the inhibited receptor, TLR4, is responsible for sensing LPS. Laakmann et al. similarly demonstrated that lung endothelial cells could be protected from the pro-inflammatory effects of BEVs derived from gram negative bacteria by neutralizing LPS with polymyxin B (Laakmann et al., 2023). Ho et al. showed that BEVs derived from Porphyromonas gingivalis, another gram-negative species, could promote TLR4 upregulation in HUVECs while Li et al. demonstrated that the same was true for E. coli-derived BEVs (Ho et al., 2016; Li et al., 2025). However, the results in the present study are, to our knowledge, the first to demonstrate that TLR4 blockade could protect HUVECs from the stimulatory effects of BEVs. Additional findings across the literature have made inhibition of LPS-TLR4 signaling an attractive goal for sepsis mitigation. In animal models, treatment with TAK-242 prevented LPS-induced damage to kidneys, lungs, and skeletal muscle (Seki et al., 2010; Fenhammar et al., 2011; Ono et al., 2020). Unfortunately, TAK-242 was not an effective sepsis treatment in human phase III clinical trials (Rice et al., 2010). This may be due to the inherent complexity of sepsis, though it is worth noting that these trials did not include TAK-242 pre-treatment since the patients were already suffering from sepsis prior to TAK-242 administration. Additionally, the main readouts from these trials were measurements of serum IL-6 levels and mortality rates rather than endothelial ICAM-1 upregulation, so it is possible that TLR4 inhibition could still prove useful in combination with other therapies (Rice et al., 2010). Significantly, BEVs possess other cargo that is known to have pro-inflammatory properties, including OmpA and Pal (Michel et al., 2020; Skerniškytė et al., 2021). The former was found to be among the most highly incorporated proteins in CFT073-derived BEVs regardless of antibiotic treatment, underscoring its structural and immunogenic significance (Zhao et al., 2023). Pal was also shown to be a stable presence in the recovered BEVs, as seen in both western blot and mass spectrometry data. Even if these factors did not contribute significantly to ICAM-1 expression on endothelial cells in the present study, it is possible that they would have a substantial impact on immune cells and other cell types even if TLR4 were inhibited. Regarding Pal, it is conceivable that ultracentrifugation could lead to co-isolation of non-BEV-associated Pal. However, given Pal's low molecular weight of 18–19 kDa, we expect that most free Pal was removed in the ultracentrifugation supernatant (Hellman et al., 2002). Any remaining additional Pal could be considered physiologically relevant in the same way as other cellular debris or residual antibiotics since it would be present in the circulation following antibiotic treatment.

One other example of BEV cargo that could serve as a stimulant is major outer membrane lipoprotein Lpp, also referred to as Braun's lipoprotein. Mass spectrometry revealed a large increase in the proportion of Lpp in BEVs derived from CFT073 [WAM2267] E. coli due to treatment with meropenem. This makes sense since β-lactam antibiotics such as meropenem act by disrupting the bacterial cell wall, which could untether the Lpp connecting the peptidoglycan layer with the outer membrane (Schwechheimer and Kuehn, 2015; Dhillon, 2018). The altered abundance of this protein suggests antibiotic-specific modulation of BEV cargo, likely linked to membrane stress responses and envelope integrity maintenance mechanisms (Torabian et al., 2025). The greater Lpp content in BEVs derived from meropenem-treated CFT073 E. coli could hypothetically explain the significant increase in ICAM-1 expression that occurred when HUVECs were exposed to these BEVs since Lpp has previously been observed to promote endothelial pro-inflammatory activation (Neilsen et al., 2001). Lpp has also been shown to cause inflammation in vitro and in vivo via interactions with TLR2 (Lakshmikanth et al., 2016). A future study could test the hypothesized contribution of the increased Lpp content by examining whether TLR2 blockade eliminates the enhanced stimulatory effects of the BEVs derived from meropenem-treated CFT073 E. coli. Although these meropenem-associated BEVs were more stimulatory than the other CFT073 BEV preparations, this effect was only apparent when HUVECs were treated with equal concentrations of the BEV preparations. Overall, meropenem treatment significantly reduced the ability of secreted BEVs to cause ICAM-1 upregulation, probably because of a non-significant reduction in the production of BEVs by CFT073 E. coli. This once again emphasizes the importance of the size of the BEV dose experienced by the endothelium.

Another interesting change in BEV protein cargo that was observed in our mass spectrometry results was the reduction of putative outer membrane porin protein NmpC in all antibiotic-treated conditions compared to the untreated control. Porins serve as channels for passive diffusion through the outer membrane and thus could act as conduits for antibiotics to access the interior of the bacteria. Mutations in porin-encoding genes as well as changes in porin expression have been observed as a means for bacteria to limit antibiotic infiltration (Delcour, 2009). Since many of our BEVs may have originated from the outer membrane, the reduced abundance of NmpC in these vesicles probably stems from changes in the parent bacteria outer membrane that were made to prevent antibiotics from entering the E. coli. These results align with prior findings of outer membrane remodeling and vesicle content regulation across gram-negative pathogens. For example, in Klebsiella pneumoniae, alterations in outer membrane porin expression are tightly linked to antibiotic resistance and have been shown to influence vesicle composition and downstream immune signaling (Doménech-Sánchez et al., 1999; Hernandez-Alles, 2000; Turner et al., 2016). Similar cross-species observations suggest that antibiotic pressure broadly shapes vesicle cargo, supporting the idea that our results in E. coli reflect a conserved bacterial strategy (Kim et al., 2020; Federica et al., 2025).

In general, antibiotic treatment appeared to cause less protein packaging per BEV. The least amount of protein was found in BEVs derived from tobramycin-treated E. coli, which is logical since aminoglycosides such as tobramycin function by inhibiting protein synthesis (Becker and Cooper, 2013). However, this reduction in protein did not seem to result in a loss of stimulatory capability, as all of the examined conditions were capable of upregulating ICAM-1 at least to the same degree as the control condition when HUVECs were treated with equal concentrations of the BEVs. Perhaps the apparent inverse correlation between LPS incorporation and protein packaging could help explain this phenomenon. This again indicates that the size of the BEV dose may be the most important factor for predicting the endothelial response. But changes due to dosing may not purely be a function of the amount of LPS per BEV, at least in the antibiotic-associated conditions. As previously noted, other cargo or sequestered antibiotics could influence the reaction of endothelial cells. It is also worth noting that, despite the measured differences in LPS content, all of the assessed BEVs contained the same order of magnitude of LPS per particle. The possibility exists that these differences may be insignificant from a cellular perspective, though further study may be necessary to determine the sensitivity of HUVECs to small variations in LPS concentration.

This work was largely focused on endothelial expression of ICAM-1, which primes the endothelial cells for interactions with passing immune cells. In the future, it may be interesting to see whether BEVs derived from bacteria treated with varying antibiotics may also have differing effects on these downstream endothelial-immune interactions. In addition to dosing the endothelium, the effect of antibiotic-induced BEVs on immune cell morphology, migratory response, cytokine production, and expression of inflammatory markers could help inform researchers and clinicians about how antibiotic treatment modulates key processes during sepsis. Regarding clinical translation, it will likely be necessary to explore the interactions between antibiotic-induced BEVs and the endothelium in vivo or in advanced human tissue-on-chip models of sepsis (McCloskey et al., 2024) prior to drawing conclusions about physiological effects. A limitation of the present study is that we only included one representative from each of three antibiotic classes: β-lactams, aminoglycosides, and fluoroquinolones. In order to generalize these results, similar experiments should be performed with other examples of these antibiotic classes. Furthermore, only a single bacterial species was examined. It is possible that strain-specific differences, such as the ones determined in this study, may occur across many sepsis-associated bacterial species. Additionally, future experiments examining a broader range of antibiotic concentrations, including sub-MIC and higher levels, would provide greater insight into how dosing influences BEV release and composition. Consistently treating with 2MIC for the antibiotics enabled clear comparisons but does not necessarily reflect clinical concentrations (Torabian et al., 2025). Much additional work can be performed in order to create a more complete understanding of the relationships between antibiotics, bacterial strains, BEV production and pro-inflammatory capacity. By linking antibiotic class- and strain-dependent BEV changes to vascular activation, this study advances beyond descriptive BEV work and provides a functional framework for considering how antimicrobial therapy may influence host inflammatory outcomes.

In summary, our findings highlight that antibiotic treatment not only influences the quantity of BEVs released by different E. coli strains but also alters their capacity to stimulate endothelial cells. These antibiotic-driven changes in BEV production and composition significantly impact endothelial activation, including ICAM-1 upregulation, suggesting a previously underappreciated link between antimicrobial therapy and host vascular responses.

MATERIALS AND METHODS

Cell culture

Pooled primary HUVECs were obtained from ATCC (PCS-100-013), Lonza (C2519A), and Thermo Fisher (C01510C). Cells were maintained in Human Large Vessel Endothelial Cell Medium (Gibco, M200500) supplemented with 2% Large Vessel Endothelial Supplement (Gibco, A1460801) and 1% Penicillin/Streptomycin, which was exchanged every 2–3 days. HUVECs were incubated at 37°C, 5% CO₂. For experiments, cells were seeded on 96-well plates at a density of approximately 40,000 cells cm⁻² and were used up to passage 9.

BEV isolation

E. coli were cultured on Luria-Bertani (LB) agar at 37°C overnight; colonies from the plate were used to inoculate a 50 ml growth in LB broth, which was cultured overnight at 37°C, shaking at 160 rpm. This small culture was divided in half to inoculate two larger cultures of 200 ml each, which were grown to log phase (optical density at 600 nm ∼0.8). These cultures were split into 49.5 ml aliquots in sterilized 125 ml flasks. All antibiotic solutions were prepared at 2MIC and then added into their corresponding flasks (500 µl of antibiotic solution+49.5 ml of media). The E. coli cultures were incubated with antibiotics (or no antibiotic, for control) for 3.5 h (37°C, 180 rpm). Afterward the samples were first centrifuged at 5000×g for 15 min to pellet the E. coli. Then the supernatant was passed through a 0.45 µm syringe filter and ultracentrifuged at 100,000×g for 2 h to pellet the BEVs. Pellets were resuspended in 400 µl of PBS so that the final BEV preparations were brought to 125 times their original LB broth concentration and were either kept at 4°C for less than 1 month or at −20°C until needed for experiments. To confirm that no live E. coli were co-isolated with the BEVs, samples were mixed in cell culture media and inspected for signs of growth after overnight incubation at 37°C, 5% CO₂. As a control, sterile LB broth alone was subjected to nanoparticle tracking analysis (see below), revealing that negligible vesicle-sized particles were present in the absence of bacteria.

Endothelial stimulation and immunofluorescent labeling

BEVs or purified LPS (Sigma-Aldrich, L3012-5MG) were diluted in cell culture media and then added to confluent HUVECs in a 96-well plate. Treatment lasted 16–17 h at 37°C, 5% CO₂. Cells were treated with anti-ICAM-1 antibody (BioLegend, 353102) diluted 1:100 in cell culture media for 15 min, then washed with PBS, and fixed with 4% formaldehyde for 15 min. The fixed cells were subsequently washed three times with PBS, blocked with 10% normal goat serum (Thermo Fisher, 50062Z) for 30 min at room temperature, then stained with a secondary goat anti-mouse IgG antibody conjugated to Alexa Fluor 488 (Thermo Fisher, A-11001) diluted 1:200 in 10% normal goat serum for 2 h in the dark at room temperature. Following this incubation, cells were washed with PBS and their nuclei were labeled with DAPI (Thermo Fisher, D1306) diluted 1:400 in deionized water for 3 min at room temperature prior to three final washes with PBS. Samples were imaged on a Leica DMI6000 B fluorescence microscope with a Rolera EM-C2™ EMCCD Camera (QImaging) using MetaMorph software (Molecular Devices). Images were captured with a 10× objective lens using fluorescence filter cubes for DAPI and GFP. Image analysis was performed using Fiji software (Schindelin et al., 2012). To quantify ICAM-1 labeling, background signal was subtracted from the average fluorescence measured from GFP channel images. This background-subtracted average was then divided by the number of nuclei counted in DAPI channel images to yield an estimate of the amount of ICAM-1 fluorescence signal per individual cell. To determine the BEV concentration in terms of BEVs per HUVEC, the total number of BEVs per well was calculated based on the volume added to each well of the 96-well plate (100 µl). This was divided by the estimated number of HUVECs in each well, which was found by taking the number of nuclei counted in DAPI channel images, calculating the average number of cells per unit area, and then multiplying this by the known area of the well.

Blockade of TLR4

To investigate the role of TLR4 in BEV-induced inflammation, we used a specific antagonist, TAK-242, to block the TLR4 receptor. HUVECs were pre-treated with 10 µM TAK-242 or a 0.04% DMSO vehicle control for 4–5 h prior to introducing 1E8 BEVs ml⁻¹ or 10 ng ml⁻¹ LPS with additional TAK-242 or vehicle control for 16–17 h. The cells were subsequently stained for ICAM-1 (see above). To confirm that TAK-242 did not inhibit the ability of other receptors to cause increased ICAM-1 expression, HUVECs were treated with 0.5 ng ml⁻¹ pro-inflammatory Cytomix consisting of equal parts TNF-α, IL-1β, and IFN-γ.

LPS quantification

To quantify LPS in our BEV samples, we used the Pierce™ Chromogenic Endotoxin Quant Kit (Thermo Fisher, A39552) per the manufacturer's instructions. Briefly, the kit's lyophilized endotoxin standard was reconstituted in water and vortexed for 15 min. This was used to create a series of comparison standards by serial dilution. BEV samples were also serially diluted in endotoxin-free water. A multiwell plate was pre-equilibrated at 37°C on a heating block for 10 min prior to adding 50 µl of the standards and samples. Amebocyte lysate reagent was reconstituted in water immediately before use and 50 µl was added to the wells. After mixing the solutions by tapping the plate, the reaction was allowed to run for 11 min at 37°C on the heating block. Chromogenic substrate was reconstituted in water, 100 µl was added to the plate, and mixing was performed by gentle tapping before returning the plate to the heating block for 6 min. 50 µl 25% acetic acid was added to stop the reaction and the plate was mixed by gentle tapping. The optical density was measured at 405 nm. All measurements were subtracted by the average absorbance of blank wells that were loaded with endotoxin-free water. The blank-subtracted standards were used to plot the optical density at 405 nm as a function of endotoxin units (EU) per ml, and this curve was used to calculate the LPS concentration in the BEV samples. Samples that fell beyond the assay detection range or measured below the blank wells were omitted from analysis. A conversion factor of 3,000,000 EU mg⁻¹, which was reported on the specification sheet for our purified LPS, was used to calculate the final LPS concentrations.

Determination of MIC

The MIC of each antibiotic was determined using the broth-dilution method according to EUCAST guidelines, with a modification to use LB media to match the conditions of all subsequent E. coli culture experiments (Clinical Microbiology and Infection, 2003). E. coli cultures were grown overnight in LB broth and inoculated into test tubes containing fresh LB broth with serially diluted antibiotics. Eight serial 1:2 dilutions were prepared for each antibiotic to create a concentration range. An antibiotic-free control was included to ensure optimal bacterial growth conditions. The cultures were incubated overnight at 37°C with shaking at 160 rpm, and bacterial growth was assessed the next day by examining the tubes for turbidity, which indicated visible growth. The MIC was defined as the lowest antibiotic concentration that inhibited visible growth, keeping the broth clear. For subsequent experiments, E. coli cultures were treated with antibiotics standardized at 2MIC to enable comparative analysis across different antibiotics.

NTA

Nanoparticle size distributions and concentrations were assessed using a NanoSight NS300 (Malvern Panalytical, Malvern, UK) equipped with a sCMOS camera, 532 nm green laser, and a 565 nm long pass filter. Multiple sample dilutions were created in PBS to ensure that at least one dilution would fall into the accurate concentration detection range between 1E8–1E9 particles ml⁻¹. Three videos of each sample were collected for 30 s each, with a camera level of 15 and detection threshold of 5. Diffusion coefficients were calculated based on the particle tracks, which allowed determination of the particle size distribution. Average particle concentrations were reported after taking the sample dilution factors into account.

TEM

The BEVs were isolated from cultures grown with 2MIC of meropenem, tobramycin, ciprofloxacin, and no antibiotic (control). 200 mesh copper grids coated with formvar/carbon film were glow discharged for 30 s at 30 mA in a PELCO Easiglow prior to 3 μl of the liquid BEV sample being applied for 30 s. Excess sample was wicked away and grids were exposed to three 15 μl washes with molecular grade water prior to negative staining with two applications of 10 μl filtered 0.75% uranyl formate, with wicking of excess fluid using hardened Whatman 50 filter paper, between steps. The grids were allowed to dry prior to examination on a Talos 120C transmission electron microscope equipped with a CETA 16 megapixel camera (Thermo Fisher) for image capture using TIA (Thermo Fisher). For all of the samples, seven to eight images at 28k magnification were taken in at least three distinct grid squares.

Western blot

Samples were analyzed using standard SDS-PAGE (4–16% bis acrylamide, Precast gels, VWR) and immunoblotting techniques. Equal volumes of all samples were loaded in each well. For immunoblotting, we used anti-Pal antisera from mice inoculated with purified recombinant non-lipidated Pal protein (∼21 kDa; contains a 6xHis-tag for purification; Rochester General Hospital) (Torabian et al., 2025). Pal was chosen as a representative outer membrane structural protein that is consistently incorporated into BEVs, providing a reference marker to assess vesicle integrity and changes in cargo loading across conditions. Immunoblots were developed using SuperSignal^TM West Femto Maximum Sensitivity Substrate (Thermo Fisher) and imaged using the Bio-Rad ChemiDoc XRS+ Imaging System. The automatic exposure setting was used to avoid over-saturating the blot images. Band volumes were quantified using Bio-Rad's Image Lab Software Version 6.1 and normalized to the control condition for each respective bacteria strain.

Mass spectrometry

Protein samples were loaded onto a 4–12% SDS-PAGE gel and resolved for approximately 5 min until a ∼10 mm band formed. After staining with SimplyBlue SafeStain (Invitrogen) and washing overnight in water, the gel band was excised, diced into 1 mm cubes, de-stained, reduced with dithiothreitol (Sigma), alkylated with iodoacetamide (Sigma), and dehydrated with acetonitrile. Trypsin (Promega) was reconstituted to 10 ng µl⁻¹ in 50 mM ammonium bicarbonate and added to the gel pieces. After 30 min at room temperature, additional ammonium bicarbonate was added to fully submerge the gel, and the samples were incubated at 37°C overnight. Peptides were extracted the following day using 50% acetonitrile and 0.1% trifluoroacetic acid (TFA), dried using a CentriVap concentrator (Labconco), desalted with homemade C18 spin columns, dried again, and reconstituted in 0.1% TFA. Peptide concentrations were measured using a fluorometric peptide assay (Thermo Fisher). Peptides were then separated by liquid chromatography using a Vanquish Neo UHPLC system (Thermo Fisher) and analyzed on an Orbitrap Astral mass spectrometer (Thermo Fisher) equipped with an Easy-Spray ion source operating at 2 kV. Samples were first trapped on a 75 µm×2 cm trap column (Thermo Fisher), followed by separation on a 75 µm×15 cm Aurora Elite C18 analytical column (IonOpticks). The mobile phases consisted of 0.1% formic acid in water (Solvent A) and 0.1% formic acid in 80% acetonitrile (Solvent B). The gradient began at 1% B, increased to 5% in 0.1 min, then to 30% in 12.1 min, 40% in 0.7 min, and 99% in 0.1 min, which was held for 2 min before re-equilibration, resulting in a total run time of 15 min. The mass spectrometer was operated in data-independent acquisition (DIA) mode. MS1 scans were acquired in the Orbitrap at a resolution of 240,000, with a maximum injection time of 5 ms and an m z⁻¹ range of 380–980. MS2 scans were collected in the Astral analyzer using a variable windowing scheme (4 Da from 380–750 m z⁻¹ and 6 Da from 750–980 m z⁻¹), with a 6 ms injection time, 28% HCD collision energy, AGC target of 500%, and fragment ion detection from 150–2000 m z⁻¹. The total cycle time was 0.6 s. Raw files were analyzed using DIA-NN (v1.9.2) in library-free mode, using the E. coli UniProt reference proteome (UP000000625_83333) with deep learning-based spectral and RT prediction enabled (Demichev et al., 2020). DIA-NN settings included one missed cleavage, a single variable modification (oxidation of methionine), peptide length range of 7–30, precursor charge range of 2–4, precursor m z⁻¹ range of 380–980, and fragment m z⁻¹ range of 150–2000. Quantification was performed in “Robust LC (high precision)” mode with RT-dependent normalization, match-between-runs (MBR) enabled, and protein inference set to “Genes” (heuristic inference disabled). MS1/MS2 tolerances and scan windows were automatically determined by the software. Precursors were filtered at a 1% library precursor q-value, 1% protein group q-value, and 50% posterior error probability. Protein quantification was performed using the MaxLFQ algorithm implemented in the DIA-NN R package, and peptide counts per protein group were determined with the DiannReportGenerator package (Cox et al., 2014; Demichev et al., 2020).

Statistical analysis

Analysis was performed using GraphPad Prism software (GraphPad, La Jolla, CA, USA). In experiments where HUVECs were dosed with BEVs derived from E. coli treated with different antibiotics, outlying data points were removed using the ROUT method with Q=5. For comparison between two conditions with a single independent variable and normally distributed data, Welch's t-test was used. For comparisons between groups with a single independent variable and non-normally distributed data, Kruskal–Wallis tests with Dunn's test for multiple comparisons were used. For comparisons between groups with a single independent variable and normally distributed data, one-way ANOVA with Dunnett's test for multiple comparisons was used. For comparisons between groups with two independent variables and normally distributed data, two-way ANOVA with Tukey's test for multiple comparisons was used.