Outer membrane vesicles displaying engineered glycotopes elicit protective antibodies

Significance

Conjugate vaccines have proven to be an effective and safe strategy for reducing the incidence of disease caused by bacterial pathogens. However, the manufacture of these vaccines is technically demanding, inefficient, and expensive, thereby limiting their widespread use. Here, we describe an alternative methodology for generating glycoconjugate vaccines whereby recombinant polysaccharide biosynthesis is coordinated with vesicle formation in nonpathogenic Escherichia coli, resulting in glycosylated outer membrane vesicles (glycOMVs) that can effectively deliver pathogen-mimetic glycotopes to the immune system. An attractive feature of our approach is the fact that different plasmid-encoded polysaccharide biosynthetic pathways can be readily transformed into E. coli, enabling a “plug-and-play” platform for the on-demand creation of glycOMV vaccine candidates that carry heterologous glycotopes from numerous pathogenic bacteria.

Abstract

The O-antigen polysaccharide (O-PS) component of lipopolysaccharides on the surface of gram-negative bacteria is both a virulence factor and a B-cell antigen. Antibodies elicited by O-PS often confer protection against infection; therefore, O-PS glycoconjugate vaccines have proven useful against a number of different pathogenic bacteria. However, conventional methods for natural extraction or chemical synthesis of O-PS are technically demanding, inefficient, and expensive. Here, we describe an alternative methodology for producing glycoconjugate vaccines whereby recombinant O-PS biosynthesis is coordinated with vesiculation in laboratory strains of Escherichia coli to yield glycosylated outer membrane vesicles (glycOMVs) decorated with pathogen-mimetic glycotopes. Using this approach, glycOMVs corresponding to eight different pathogenic bacteria were generated. For example, expression of a 17-kb O-PS gene cluster from the highly virulent Francisella tularensis subsp. tularensis (type A) strain Schu S4 in hypervesiculating E. coli cells yielded glycOMVs that displayed F. tularensis O-PS. Immunization of BALB/c mice with glycOMVs elicited significant titers of O-PS–specific serum IgG antibodies as well as vaginal and bronchoalveolar IgA antibodies. Importantly, glycOMVs significantly prolonged survival upon subsequent challenge with F. tularensis Schu S4 and provided complete protection against challenge with two different F. tularensis subsp. holarctica (type B) live vaccine strains, thereby demonstrating the vaccine potential of glycOMVs. Given the ease with which recombinant glycotopes can be expressed on OMVs, the strategy described here could be readily adapted for developing vaccines against many other bacterial pathogens.

For decades, vaccines have served as an important pillar in preventative medicine, providing protection against a wide array of disease-causing pathogens by inducing humoral and/or cellular immunity. In the context of humoral immunity, carbohydrates are appealing vaccine candidates owing to their ubiquitous presence on the surface of diverse pathogens and malignant cells. For example, most pathogenic bacteria are prominently coated with carbohydrate moieties in the form of capsular polysaccharides (CPSs) (1) and lipopolysaccharides (LPSs) (2), which are often the first epitopes perceived by the immune system. However, a major impediment to the development of polysaccharide-based vaccines is the fact that pure carbohydrates typically stimulate T cell-independent immune responses (3–5), which are characterized by lack of IgM-to-IgG class switching (6), failure to induce a secondary antibody response after recall immunization, and no sustained T-cell memory (7).

A common strategy for enhancing the immunogenicity of carbohydrates and evoking carbohydrate-specific immunological memory is to covalently couple a carbohydrate epitope to a CD4⁺ T cell-dependent antigen such as an immunogenic protein carrier. Indeed, conjugate vaccines composed of bacterial CPS- or LPS-derived glycans chemically bound to a carrier protein induce glycan-specific IgM-to-IgG switching, memory B-cell development, and long-lived T-cell memory (5, 8–11). Such glycoconjugates have proven to be a highly efficacious and safe strategy for protecting against virulent pathogens, including Haemophilus influenzae, Neisseria meningitidis, and Streptococcus pneumoniae (10, 12, 13), with several already licensed and many others in clinical development (9, 12).

Despite their effectiveness, traditional conjugate vaccines are not without their drawbacks. Most notable among them is the complex, multistep process required for the purification, isolation, and conjugation of bacterial polysaccharides, which is expensive, time consuming, and low yielding (14). A greatly simplified and cost-effective alternative, known as protein glycan coupling technology (PGCT), has been described recently (15). This approach is based on engineered protein glycosylation in living Escherichia coli (16), wherein an O-antigen polysaccharide (O-PS), the outermost component of bacterial LPS (2), is conjugated to a coexpressed carrier protein by the Campylobacter jejuni oligosaccharyltransferase PglB (CjPglB). However, whereas PGCT has been used to make several novel protein/glycan combinations (15, 17, 18), it currently has a limited substrate specificity because the natural substrate specificity of the conjugating enzyme, CjPglB, restricts the diversity of glycans that can be transferred (19) and causes the conjugation efficiency between certain nonnative glycan and protein substrates to be very low (18). Additionally, it remains to be determined whether the carrier proteins used in licensed glycoconjugate vaccines, such as the toxins from Clostridium tetani and Corynebacterium diphtheriae, are compatible with expression and CjPglB-mediated glycosylation in E. coli.

Here, we sought to create a new approach for the production of glycoconjugate vaccines that circumvents these problems by combining recombinant O-PS biosynthesis with outer membrane vesicle (OMV) formation in laboratory strains of E. coli. OMVs are naturally occurring spherical nanostructures (∼20–250 nm) produced by all gram-negative bacteria. They are composed of proteins, lipids, and glycans, including LPS, derived primarily from the bacterial periplasm and outer membrane (20). In recent years, OMVs have garnered attention as a vaccine platform because they are nonreplicating, immunogenic mimics of their parental bacteria that stimulate both innate and adaptive immunity and possess intrinsic adjuvant properties (21–23). These characteristics are exemplified by OMVs isolated directly from N. meningitidis, which induce potent protective immune responses and have been incorporated successfully into several commercial vaccine formulations for use in humans (22, 24, 25). To expand the vaccine potential of OMVs, several groups have used genetic engineering techniques to load OMVs with foreign protein antigens by targeting expression either to the outer membrane or the periplasm of an OMV-producing host strain (26–32). These OMV-associated recombinant proteins were internalized by eukaryotic cells (26, 27) and stimulated strong and specific immune responses in mice (28–32). However, although efforts to load OMVs with recombinant protein antigens are well documented (33), an analogous strategy to engineer the polysaccharide component of OMVs for specific vaccine applications has yet to be demonstrated.

We sought to engineer OMVs that efficiently deliver surface-associated glycotopes to the immune system in a manner that induces protective immunity. Toward this goal, heterologous O-PS structures were expressed in hypervesiculating E. coli cells, resulting in glycosylated OMVs (glycOMVs) whose surfaces were remodeled with pathogen-mimetic polysaccharides. A major advantage of this approach is that designer carbohydrates are directly conjugated to lipid A, which is a powerful adjuvant whose bioactivity and toxicity can be genetically modulated (34). One of these candidate glycOMVs was subsequently evaluated for its ability to confer protection against highly virulent Francisella tularensis subsp. tularensis (type A) strain Schu S4, a gram-negative, facultative coccobacillus and the causative agent of tularemia. This bacterium is one of the most infectious agents known to man and is categorized as a class A bioterrorism agent due to its high fatality rate, low dose of infection, and ability to be aerosolized (35). Although there is currently no available licensed vaccine, several studies have confirmed the important role of antibodies directed against F. tularensis LPS, specifically the O-PS repeat unit, in providing protection against the highly virulent Schu S4 strain (36–38). More recently, a purified recombinant vaccine comprising the F. tularensis Schu S4 O-PS conjugated to the Pseudomonas aeruginosa exotoxin A carrier protein was produced using PGCT (17). This glycoconjugate boosted IgG levels and significantly increased the time to death upon subsequent pathogen challenge, albeit with the less virulent F. tularensis subsp. holarctica (type B) strain HN63. Here, we show that immunization of mice with glycOMVs displaying F. tularensis Schu S4 O-PS induced high titers of functional serum IgG antibodies against Schu S4 LPS as well as vaginal and bronchoalveolar IgA antibodies. Importantly, glycOMVs significantly extended time to death upon subsequent challenge with F. tularensis Schu S4 and provided complete protection against challenge with F. tularensis subsp. holarctica live vaccine strain (LVS) Iowa and LVS Rocky Mountain Laboratories (RML) that display the same O-PS structure on their outer membrane (38). Overall, OMVs displaying designer glycotopes on lipid A, itself a strong adjuvant, represent a potent glycoconjugate vaccine design that, given the generality of the approach, could be developed for numerous other bacterial pathogens.

Results

Glycosylation of OMVs with Heterologous O-PS.

LPS is found exclusively in the outer leaflet of the gram-negative outer membrane and consists of three distinct regions: a hydrophobic domain known as “lipid A,” a core oligosaccharide, and an O-PS (2) (Fig. 1). Laboratory E. coli strains usually lack O-PS structures but do produce a complete lipid A-core that serves as an acceptor for O-PS if the genes for its synthesis are supplied in trans (39). Because LPS is a major component of released OMVs (20), we postulated that expression of heterologous O-PS pathways in hypervesiculating E. coli would result in OMVs whose lipid A-core was glycosylated with desired O-PS structures (Fig. 1). To test this notion, we introduced the gene cluster for the synthesis of F. tularensis Schu S4 O-PS into O-PS–deficient E. coli strain JC8031. This strain was chosen for its ability to hypervesiculate, due to genetic knockout of tolRA (40), and has been used extensively in OMV engineering applications (27, 28, 33). OMVs were isolated from JC8031 cells expressing the Schu S4 O-PS gene cluster from pGAB2 (17) and subjected to Western blot analysis using an F. tularensis O-PS–specific antibody named “FB11” (37). We observed a classical ladder-like pattern typical of LPS (Fig. 2), which results from O-PS chain length variability generated by the Wzy polymerase (2, 16). F. tularensis O-PS was absent in OMVs derived from JC8031 cells carrying empty plasmid (Fig. 2). Likewise, when the Schu S4 O-PS antigen genes were expressed in strain CE8032, which lacks the waaL gene encoding the ligase that transfers O-PS to lipid A-core (Fig. 1) (2, 16), the resulting OMVs were no longer detected with the FB11 antibody (Fig. 2).

Fig. 1.

Assembly and display of pathogen-specific O-PS structures on OMVs. Schematic of a pathway for biosynthesis of heterologous O-PS structures and their incorporation into OMVs in E. coli. Lipid-linked O-PS repeating units are assembled on the cytoplasmic face of the inner membrane (IM) by glycosyltransferases encoded in plasmid pO-PS, after which translocation to the periplasmic face occurs by the action of endogenous flippase Wzx. Polymerization of O-PS repeating units on the periplasmic face of the inner membrane is then catalyzed by endogenous Wzy polymerase in a block transfer mechanism that is regulated by endogenous Wzz. The resulting O-PS is then transferred to the lipid A-core polysaccharide by endogenous O-antigen ligase, WaaL. The resulting lipopolysaccharide is shuttled to the outer membrane (OM), where it becomes incorporated in budding vesicles to produce pathogen-specific glycOMVs.

Fig. 2.

Incorporation of pathogen-specific O-PS in OMVs. Western blot analysis of OMV fractions isolated from E. coli JC8031 (WaaL, +) or CE8032 (WaaL, −) cells carrying an empty plasmid (pO-PS, −) or a heterologous O-PS pathway plasmid (pO-PS, +) corresponding to the pathogenic strain indicated below each panel. The O-PS pathway plasmids and empty control plasmids are provided in SI Appendix, Table S3. Antibodies specific to each O-PS (SI Appendix, Table S4) were used to detect heterologous glycan structures displayed on the glycOMVs. Molecular mass markers are labeled on the right.

To demonstrate the generality of the approach, an expanded repertoire of pathogen-mimetic O-PS structures was expressed in OMVs. Specifically, plasmids containing the O-PS gene clusters from a variety of gram-negative pathogenic bacteria, including uropathogenic E. coli (UPEC) strain VW187 (O7:K1), enterotoxigenic E. coli (ETEC) strains O78 and O148, P. aeruginosa strain PA103, Shigella dysenteriae 1 strain W30864, Shigella flexneri serotype 2a, and Yersinia enterocolitica strain 6471/76, were transformed in strain JC8031. OMVs prepared from these cells were all cross-reactive with antibodies specific for the respective O-PS structures (Fig. 2). In contrast, control OMVs prepared from either CE8032 cells expressing the same O-PS pathway genes or JC8031 cells carrying empty plasmids were not detected by the cognate O-PS–specific antibodies (Fig. 2). In the case of OMVs displaying Y. enterocolitica O-PS, we observed a smear rather than a clear ladder; however, this smear is typically produced after electrophoresis of LPS preparations from these bacteria (41). Taken together, these results suggest that lipid A-core in OMVs was glycosylated with heterologous, strain-specific O-PS structures.

The Outer Surface of Intact OMVs Is Remodeled with F. tularensis O-PS.

Given our interest in creating a vaccine candidate against F. tularensis, OMVs generated by strain JC8031 carrying pGAB2, termed “Ft-glycOMVs,” were further characterized. To confirm that F. tularensis O-PS was on the outer surface of vesicles, dot blots were performed by spotting fractions containing Ft-glycOMVs directly onto nitrocellulose membranes without any denaturation steps. Only the OMV fractions derived from JC8031 cells carrying plasmid pGAB2 were detected by the FB11 antibody (SI Appendix, Fig. S1A), suggesting that nondenatured, intact vesicles carried F. tularensis O-PS on their surface. As expected, nondenatured vesicles derived from CE8032 carrying pGAB2 did not give a strong signal using FB11 (SI Appendix, Fig. S1A). The size and shape of Ft-glycOMVs appeared indistinguishable from control OMVs (SI Appendix, Fig. S1B), indicating that incorporation of foreign O-PS into E. coli LPS structures had no visible effect on vesicle nanostructure. Next, we determined whether F. tularensis O-PS detected in the pelleted supernatant was associated with intact vesicles or with released outer membrane fragments. To this end, the OMV-containing fraction isolated from JC8031 cells carrying pGAB2 was separated by density gradient ultracentrifugation. Western blotting and Coomassie staining of the resulting fractions revealed that O-PS glycans and total proteins comigrated to denser fractions (SI Appendix, Fig. S1 C and D), reminiscent of the gradient profiles seen previously for intact OMVs and OMV-associated proteins (27, 42). After nondenaturing dot blotting, the FB11 antibody was observed to cross-react with these same, denser fractions, confirming that O-PS glycans were on the exterior surface of intact vesicles (SI Appendix, Fig. S1C). It is particularly noteworthy that OMVs generated from JC8031 cells carrying pGAB2 were observed to cross-react with the mouse IgG2a antibody FB11. We conclude that the O-PS structure generated on OMVs is immunologically relevant given that FB11 targets a unique terminal F. tularensis O-PS epitope, confers survival to BALB/c mice infected intranasally with the F. tularensis type B LVS, and prolongs survival of BALB/c mice infected intranasally with highly virulent F. tularensis type A strain Schu S4 (37).

Structural Characterization of Heterologous F. tularensis O-PS.

To shed light on the identity of the heterologous O-PS, Ft-glycOMVs were structurally characterized. Western blot analysis using FB11 revealed nearly identical laddering for Ft-glycOMVs compared with hybrid E. coli LPS capped with the F. tularensis O-PS (Ft-glycLPS) extracted directly from intact JC8031 cells carrying pGAB2 (Fig. 3A), indicating that the engineered LPS molecules in the outer membrane of JC8031 are structurally similar to those loaded in OMVs. Compared with the native F. tularensis Schu S4 LPS (FtLPS), the height of these ladders (i.e., the chain length of O-PS) was notably shorter (Fig. 3A).

Fig. 3.

Structural analysis of heterologous F. tularensis O-PS. (A) Western blot analysis of Ft-glycOMVs, Ft-glycLPS, and FtLPS. Blots were probed with the FB11 antibody. Molecular mass (MW) markers are labeled on the left. (B and C) MALDI-TOF MS spectra: MS¹ of isolated O-PS tetrasaccharide m/z 849.2 [M+Na]⁺ and 865.2 [M+K]⁺ (B); and product-ion MS/MS of m/z 849.2 [M+Na]⁺ (C). The fragment ions were reported using the Domon and Castello nomenclature (62).

The O-PS repeating unit in native FtLPS is the tetrasaccharide [2)-β-Qui4NFm-(1→4)-α-GalNAcAN-(1→4)-α-GalNAcAN-(1→3)-β-QuiNAc-(1→] (43). To determine the structure of the carbohydrate moiety in Ft-glycOMVs, we performed NMR analysis on LPS derived from JC8031 carrying pGAB2. Ft-glycLPS extracted from these cells was delipidated by mild acid hydrolysis and purified by size-exclusion chromatography (SEC). SEC yielded carbohydrate fractions that were subjected to structural analysis by 1D and 2D NMR. The 1D proton spectrum of the isolated product revealed the presence of over 15 signals with different intensities in the anomeric region (δ 5.5–4.4), and 2D NMR spectra revealed the presence of many spin systems. By study of the 2D COSY, heteronuclear single-quantum coherence (HSQC), and heteronuclear multiple bond correlation (HMBC) spectra (SI Appendix, Fig. S2), four residues belonging to 2-acetamido-2-deoxy-galacturonamide (GalNAcAN), 4,6-dideoxy-4-formamidoglucose (Qui4NFm), and N-acetylglucosamine (GlcNAc), in a 2:1:1 ratio, could be discriminated (SI Appendix, Table S1). In the 1D NMR spectrum, the signals at 5.41 and 5.03 ppm corresponded to the anomeric protons of two GalNAcAN residues, designated as residues B and C. The HSQC spectrum showed downfield signals for C-4 of residue B at 81.3 ppm and of residue C at δ 78.0, in accordance with a 4-substituted α-d-GalNAcAN residue (43). The anomeric signal at 4.50 ppm was attributed to terminal Qui4NFm (residue A). The methyl group of this residue resonates at 1.18 ppm (44). The assignment of residues Dα and Dβ was complicated by extensive overlap of signals stemming from other carbohydrate material. The broad signals at ∼5.15 ppm (Dα) and ∼4.57 ppm (Dβ) belong to reducing end N-acetylglucosamine (GlcNAc) (SI Appendix, Table S1). The ¹³C chemical shifts, deduced from the HSQC spectrum, showed the downfield position of C-3 of Dβ at 82.8 ppm, indicating a 3-substituted residue (45). The linkage sequence of the monosaccharide was determined by the HMBC spectrum. The 4-substitution of residues B and C was supported by correlations between H-1 of B and C-4 of C and between H-1 of A and C-4 of B, respectively. Furthermore, the correlation between H-1 of C and C-3 of Dβ was clearly observed. On the basis of these ¹H and ¹³C NMR data, it can be concluded that the isolated compound is a tetrasaccharide of the following structure:

To confirm this conclusion from NMR, we analyzed the isolated oligosaccharide using MALDI-TOF MS in positive ion mode. The peaks at m/z 849.2 and at m/z 865.2 correspond to the sodium and potassium adducts of the tetrasaccharide, respectively (Fig. 3B). To further characterize the topology of this oligosaccharide, the ion at 849.2 [M+Na]⁺ was subjected to tandem MS. In the resulting MS² spectrum (Fig. 3C), the Y₃ ion and Y₂ ion at m/z 676.5 and at m/z 460.4, respectively, showed a loss of Qui4NFm and GalNAcAN from the nonreducing end. The presence of fragment ions at m/z 646.5 (C₃) and m/z 442.4 (Z₂) indicated a loss of GlcNAc and GalNAcAN from the reducing end. This result confirmed the Qui4NFm-GalNAcAN-GalNAcAn-GlcNAc sequence.

Structural Diversification of Lipid A Yields Less Toxic Ft-glycOMVs.

LPS is a main contributing factor in triggering host immune response during infection through recognition of lipid A, also known as endotoxin, by toll-like receptor 4 (TLR4). Immune recognition of lipid A results in production of proinflammatory cytokines that are crucial to fight infection but may also contribute to lethal septic shock at high levels (46). Thus, for glycOMVs to be a viable vaccine platform, it is necessary to reduce the toxicity of lipid A while also maintaining its adjuvanticity. One such LPS derivative, monophosphorylated lipid A (MPL) from Salmonella minnesota R595, is an approved adjuvant with reduced toxicity (47). MPL is a mixture of monophosphorylated lipids, with the primary species being pentaacylated, monophosphorylated lipid A. In contrast, native E. coli lipid A is characterized by the presence of six acyl chains and two phosphate groups. Indeed, MS analysis of isolated lipid A from selected E. coli strains, including JC8031 or JC8031 carrying plasmid pGAB2, revealed a prototypical hexaacylated, bis-phosphorylated lipid A (SI Appendix, Fig. S3 A–C). To mimic the MPL structure in our glycOMVs, we adopted a lipid A remodeling strategy described by Trent and coworkers (34). Specifically, we deleted the acyltransferase-encoding lpxM gene in JC8031, resulting in strain JH8033 that synthesized pentaacylated lipid A in the absence or presence of pGAB2 (SI Appendix, Fig. S3 D and E). Expression of the F. tularensis phosphatase LpxE from plasmid pE resulted in a strain that produced nearly homogenous pentaacylated, monophosphorylated lipid A (SI Appendix, Fig. S3F). When combinations of lipid A-modifying enzymes (e.g., LpxE and E. coli lipid A palmitoyltransferase PagP coexpressed from plasmid pEP; Salmonella typhimurium lipid A 3′-O-deacylases PagL and LpxR, and PagP coexpressed from plasmid pLPR) were coexpressed, a more heterogeneous mixture was observed as a consequence of the substrate specificity and limited expression level of the transmembrane lipid A-modifying enzymes (SI Appendix, Fig. S4 A and B), as seen previously (34). Importantly, JH8033 cells carrying pGAB2, or pGAB2 along with any of the plasmids encoding lipid A-modifying enzymes, produced F. tularensis O-PS on the cell surface on par with that produced by JC8031 carrying pGAB2 (SI Appendix, Fig. S3G). Likewise, glycOMVs harvested from these strains also displayed the F. tularensis O-PS at levels comparable with glycOMVs from JC8031 (SI Appendix, Fig. S3G).

Next, toxicity of whole cells and OMVs was evaluated by measuring human TLR4 activation in HEK-Blue hTLR4 reporter cells. These cells express hTLR4 and respond to activation of the receptor by the production of secreted embryonic alkaline phosphatase (SEAP) (34). Upon incubating this reporter cell line with whole bacterial cells, significantly lower TLR4 activation (P < 0.01) was observed for all JH8033 strains compared with their JC8031 counterparts (SI Appendix, Fig. S5A), with signals that were comparable with a previously detoxified strain: namely BN2 (as W3110 ΔlpxM) (34). Notably, the presence of the F. tularensis O-PS did not have any impact on TLR4 activation. The TLR4 activation assay was also run by treating the HEK-Blue reporter cell line with purified OMVs. As with whole cells, OMVs derived from all JH8033 strains showed significantly reduced activation of TLR4 compared with OMVs derived from the corresponding JC8031 strain (SI Appendix, Fig. S5B).

Ft-glycOMVs Protect Against Lethal F. tularensis Challenge.

An effective vaccine for tularemia will likely require multiple antigens, but, as an initial step in determining whether Ft-glycOMVs might be a candidate vaccine component for a multivalent vaccine, we evaluated their protective efficacy in mice infected with the highly virulent F. tularensis type A strain Schu S4, which has an LD₅₀ of <10 colony-forming units (cfus) in mice (48). BALB/c mice were immunized by an i.p. route with Ft-glycOMVs derived from either JC8031 cells, JH8033 cells, or JH8033 cells carrying a plasmid with the lipid A-modifying enzymes, purified FtLPS, empty OMVs from JC8031 cells, empty OMVs from JH8033 cells, or PBS. At 56 d after the initial dose, immunized mice were challenged with 25 cfus of F. tularensis Schu S4 via i.p. injection, and survival of the mice was monitored. All mice receiving one of the Ft-glycOMV preparations survived until day 6 or 7, which corresponded to a significantly (P < 0.05) delayed time to death (mean increase of 2.0–2.4 d) compared with PBS-treated control mice (Fig. 4 and SI Appendix, Table S2). In contrast, mice immunized with either purified FtLPS or either of the empty OMV preparations all succumbed to infection within 5 d (Fig. 4 and SI Appendix, Table S2), which was the same time as the PBS control group. Importantly, there was no significant difference in survival between any of the various Ft-glycOMV–treated groups, suggesting that detoxified Ft-glycOMVs afforded the same level of protection against pathogen challenge as their unmodified counterpart.

Fig. 4.

Ft-glycOMVs delay onset of lethal disease with Schu S4. Kaplan–Meier survival analysis of nine groups of BALB/c mice, five mice per group, immunized i.p. with the following: PBS, FtLPS, empty OMVs derived from JC8031 or JH8033, Ft-glycOMVs derived from JC8031 pGAB2 or JH8033 pGAB2 (Upper); and Ft-glycOMVs derived from JH8033 pE pGAB2, JH8033 pEP pGAB2, or JH8033 pLPR pGAB2 (Lower). To ensure that an equivalent amount of LPS was used in each case, the LPS content of OMVs and purified LPS was normalized based on reactivity to FB11 antibody. Mice were boosted 28 d after the original immunization with the same antigen and amount as the original dose. At 56 d after the primary injection, all mice were challenged i.p. with 25 cfus of F. tularensis Schu S4. Survival of mice in the Ft-glycOMV groups compared with those in the PBS or empty OMV control groups was found to be significant (P < 0.05; log-rank test).

To confirm the reproducibility of these results, a second nearly identical challenge was performed, except with two additional control groups: Ft-glycLPS alone and “sham” glycOMVs from JC8031 cells expressing S. dysenteriae O-PS genes (Sd-glycOMVs). At 56 d after the initial dose, immunized mice were challenged with 22 cfus of F. tularensis Schu S4 via i.p. injection, and survival of the mice was monitored. As above, mice immunized with Ft-glycOMVs demonstrated a significantly (P < 0.05) delayed time to death (mean increase of 3 d) compared with PBS-treated control mice, with all mice in this group surviving until day 6 and three of the mice surviving until day 7 (SI Appendix, Fig. S6 and Table S2). This increase in time to death was specific to the F. tularensis O-PS on OMVs because mice immunized with Sd-glycOMVs experienced no such increase in protection (P > 0.1) (SI Appendix, Fig. S6 and Table S2). Immunization with polysaccharides alone also afforded no protection against challenge because mice that received either FtLPS or Ft-glycLPS perished at the same rate as control mice that had received PBS (P > 0.1), with all mice except one dying within 4 d (one mouse receiving Ft-glycLPS succumbed to infection on day 5) (SI Appendix, Fig. S6 and Table S2).

To determine whether Ft-glycOMVs can provide cross-strain protection against other bacteria having structurally similar O-PS structures, immunized mice were challenged with two different F. tularensis subsp. holarctica (type B) LVS isolates, both significantly less virulent than Schu S4. Specifically, LVS Iowa, which has an intranasal LD₅₀ of ∼3,000–4,000 cfus, and the significantly more virulent LVS RML (intranasal LD₅₀ is ∼175 cfus) were used. Mice were immunized identically as above with Ft-glycOMVs derived from JC8031 or JH8033 cells, or PBS. At 56 d after the initial dose, immunized mice were challenged with 4–400 cfus of LVS RML or LVS Iowa. All of the PBS-treated control mice when infected i.p. with only 4 cfus of either LVS isolate died within a week. The PBS-treated control mice infected with LVS RML all died on day 5 whereas those infected with LVS Iowa all died by day 7 (four mice died on day 6 and the last one on day 7) (SI Appendix, Fig. S7 A and B and Table S2). In contrast, mice immunized with Ft-glycOMVs were completely protected against challenge by either strain up to 400 cfus, which was the highest dose tested (SI Appendix, Fig. S7 A and B and Table S2). Not only did we observe excellent protection against both of these strains, but none of the Ft-glycOMV–vaccinated mice even seemed sick, suggesting that protection was probably higher than we were able to see in this experiment. Taken together, our results demonstrate that Ft-glycOMVs offer protection against both type A and type B infection.

Ft-glycOMVs Induce a Mixed Th1/Th2 Response.

To confirm that the protective effects seen with Ft-glycOMVs correlated with increased antibody titers, the levels of FtLPS-specific IgGs were assessed in mice before challenge. Using ELISA with native FtLPS as the antigen, the total IgG titers for mice receiving Ft-glycOMVs were significantly increased (P < 0.01) compared with all other groups as early as 14 d after immunization (SI Appendix, Fig. S8). At this time, the mean IgG titer was two orders of magnitude greater than the mean titer of PBS control group mice. This differential became further amplified after the booster injection, reaching a maximum difference of three orders of magnitude at 56 d (Fig. 5A and SI Appendix, Fig. S9A). Likewise, immunization with Ft-glycOMVs derived from JH8033 cells carrying pGAB2, or pGAB2 along with any of the plasmids encoding lipid A-modifying enzymes, was significantly higher than the PBS and FtLPS control groups (P < 0.01) (Fig. 5A). Importantly, there was no significant difference in IgG titers after immunization with Ft-glycOMVs harboring WT lipid A versus remodeled lipid A (P > 0.2) (Fig. 5A), suggesting that detoxification of OMVs had no measurable effect on their immunogenicity or adjuvanticity. IgG antibody titers were further broken down by analysis of IgG1 and IgG2a titers, wherein mean IgG1-to-IgG2a antibody ratios served as an indicator of a Th1- or Th2-biased immune response. Mice immunized with Ft-glycOMVs showed a significant (P < 0.01) increase in mean titers of both FtLPS-specific IgG1 and IgG2a (Fig. 5B and SI Appendix, Fig. S9B). The higher IgG1 versus IgG2a titers suggested a slight bias toward a Th2 response. The relative titers of IgG1 and IgG2a subtypes from groups immunized with JH8033-derived Ft-glycOMVs were comparable with the titers observed for JC8031-derived Ft-glycOMVs. Classically, a Th1-biased immune response is important for intracellular pathogens such as F. tularensis; however, several groups have shown the importance of both Th1 and Th2 immune responses for this particular pathogen (36).

Fig. 5.

Ft-glycOMVs boost production of FtLPS-specific IgG antibodies. (A) FtLPS-specific IgG titers in endpoint (day 56) serum of individual mice (black dots) and median titers of each group (red lines). Nine groups of BALB/c mice, five mice per group, immunized i.p. with the following: PBS, FtLPS, empty OMVs derived from JC8031 or JH8033, Ft-glycOMVs derived from JC8031 pGAB2 or JH8033 pGAB2, and Ft-glycOMVs derived from JH8033 pE pGAB2, JH8033 pEP pGAB2, or JH8033 pLPR pGAB2. To ensure that an equivalent amount of LPS was used in each case, the LPS content of OMVs and purified LPS was normalized based on reactivity to FB11 antibody. Mice were boosted on day 28 with the same doses. (B) Median IgG subtype titers measured from endpoint serum with IgG1 titers in gray and IgG2a in black. An asterisk (*) indicates statistical significance (P < 0.01; Tukey–Kramer HSD) of antibody titers against PBS control group. A double asterisk (**) indicates statistically significant difference (P < 0.05; unpaired t test) in IgG1 and IgG2a titers within the group.

Ft-glycOMVs Induce Mucosal IgA Production.

Recent studies indicate that IgA antibodies also play a significant role in protection against F. tularensis infection (49). As the predominant antibody found at mucosal sites, increased IgA production provides one possible means of enhancing protection against mucosal infection. Consistent with the latter, Ft-glycOMVs derived from JH8033 cells, which significantly delayed time to death against F. tularensis Schu S4 challenge and generated 100% protection against F. tularensis LVS challenge, significantly (P < 0.01) enhanced FtLPS-specific IgA production in the bronchoalveolar lavage (BAL) and vaginal lavage (VL) fluids, as well as the sera of immunized mice above that of empty OMVs, FtLPS, or PBS (shown for BAL and VL fluids in SI Appendix, Fig. S10 A and B, respectively). The increased IgA titers correlated with a similarly significant (P < 0.01) increase in total FtLPS-specific blood sera IgG titers (SI Appendix, Fig. S10C).

Discussion

In the present study, we describe a glycoconjugate vaccine platform that leverages the immunological potential of recombinant OMVs. This platform is founded in part on our previous finding that remodeling the surface of OMVs with weakly immunogenic protein antigens yielded OMV-based vaccine candidates that boosted antigen-specific IgG levels (28). In fact, the response elicited by the engineered OMVs rivaled that obtained when the same protein antigen was adsorbed to the FDA-approved adjuvant alum (28), suggesting that OMVs function not only as nanoparticulate vaccine carriers but also as vaccine adjuvants (50). The basis for this adjuvanticity is likely due to the fact that OMVs (i) are readily phagocytosed by professional antigen-presenting cells; (ii) carry pathogen-associated molecular patterns (PAMPs) within their structure that can stimulate both innate and adaptive immunity; and (iii) possess strong proinflammatory properties (21–23).

Because carbohydrates are also commonly known to be weak antigens (3–5), we hypothesized that delivery of specific polysaccharide structures by engineered OMVs would enhance the immune response to these weakly immunogenic epitopes due to the natural adjuvanticity of the OMV carriers. To test this hypothesis, we created glycoengineered OMVs by combining the vesicle formation process with heterologous glycan biosynthesis machinery in laboratory strains of E. coli. An attractive feature of our approach is the fact that different plasmid-encoded O-PS biosynthetic pathways can be readily transformed into E. coli, enabling a “plug-and-play” platform for the creation of glycOMVs that surface display heterologous glycotopes from pathogenic bacteria. In the most notable example, hypervesiculating E. coli strain JC8031 harboring the F. tularensis Schu S4 O-PS pathway genes yielded OMVs that were glycosylated with a structural mimetic of F. tularensis O-PS. Although glycan analysis revealed subtle chemical structural differences between the tetrasaccharide unit found in native FtLPS and the heterologous O-PS on Ft-glycOMVs, these structural differences did not seem to significantly alter the properties of the O-PS. Indeed, the heterologous glycan exhibited a laddering pattern that was still recognized by FB11 antibodies generated against the native FtLPS structure (37).

The best evidence for authentic glycomimicry, however, was the fact that vaccination with the resulting Ft-glycOMVs significantly boosted the production of FtLPS-specific IgG antibodies as early as 2 wk after the initial immunization and by as much as 2–3 orders of magnitude above all controls, including native FtLPS. This finding is particularly noteworthy in light of the generally observed phenomenon that the immune response generated against purified LPS is T cell-independent and does not result in the production of antigen-specific IgG antibodies (4), as was confirmed here with native FtLPS and engineered Ft-glycLPS. The high IgG titers and broad response produced as a result of vaccination with Ft-glycOMVs suggests stimulation of immunological responses that are otherwise nonexistent against classical T cell-independent antigens. Indeed, T cells and the presence of toll-like receptor signals on OMVs may serve secondary roles during immune responses (51). Importantly, the robust immune response elicited by glycOMVs provided protection against lethal challenge by F. tularensis Schu S4, as demonstrated by an increased time to death compared with vaccination with different controls, including FtLPS alone. The extended protection was attributed to the presence of the O-PS because no protection was afforded to mice vaccinated with sham OMVs containing a nonspecific O-PS structure (Sd-glycOMV). In addition to protection against this type A strain of F. tularensis, glycOMVs also provided complete protection against two different F. tularensis type B strains that displayed similar O-PS structures in the outer leaflet of their outer membranes. Specifically, mice vaccinated with any of the different Ft-glycOMV preparations were able to clear the infections caused by F. tularensis subsp. holarctica LVS RML and LVS Iowa, and survive. At present, the underlying reason or reasons for why glycOMVs were more effective against the LVS isolates than the more virulent Schu S4 remain unknown. However, one possibility might be related to the observation that highly virulent Schu S4, but not the closely related LVS, is able to bind the host serine protease plasmin, which allows evasion of opsonization by antibodies and thus dampens the protective effects of these host molecules (52).

One major route of F. tularensis infection is through inhalation and other mucosal routes, and thus the presence of mucosal IgA antibodies is important in protection (49). To stimulate a protective mucosal immune response, vaccines must often be introduced through mucosal routes, such as intranasal administration. Here, we have shown that glycOMVs can generate an antigen-specific mucosal IgA response through s.c. administration, and this response correlates with both high antigen-specific IgG titers in sera as well as protection against challenge.

Antibodies alone do not provide hosts with protection against tularemia. Indeed, studies have shown that adaptive immunity against F. tularensis also requires a robust cell-mediated response (53). Specifically, a T cell-dependent response is required to control infection and is likely to hinge on the activation of macrophages. Incidentally, F. tularensis is known to target macrophages and is able to suppress the early inflammatory responses necessary in containing the pathogen (54). Thus, early IFNγ activation of macrophages is vital to control infection (55). Intracellular cytokine staining of splenocytes from mice vaccinated with Ft-glycOMVs revealed a population of CD3⁺ T cells that responded to restimulation with FtLPS in vitro with increased production of IFNγ (SI Appendix, Fig. S11), suggesting the generation of a small T cell-dependent response. However, this response was not limited to the Ft-glycOMV–vaccinated group; similar shifts in T-cell population were seen in the FtLPS group as well as the PBS control group (SI Appendix, Fig. S11). This finding, coupled with the fact that the shift in IFNγ-producing T cells was small, suggests that the observed T-cell activation may be nonclassical because the antigen is not one classically associated with MHC presentation. Indeed, the small shift in IFNγ-producing cells may be the result of stimulating γ/δ T cells, a rare subset of T cells capable of MHC-independent activation (56).

One concern with the use of bacterial OMVs as a vaccination platform is toxicity as a result of the presence of LPS on the membrane surface. This concern may be addressed by chemically stripping away LPS from OMVs through the use of polymyxin B columns (23, 28). However, because our strategy involves assembling O-PS directly upon lipid A, stripping away LPS would remove the desired polysaccharide epitope as well. To circumvent this issue, the lipid A structure of JC8031 E. coli was remodeled at the genetic level to yield a variant that is significantly less toxic, as measured by hTLR4 activation, while still retaining desirable immunomodulatory qualities. Previous combinatorial engineering of E. coli lipid A demonstrated that removal of an acyl chain by LpxM yields a pentaacylated lipid A structure with significantly reduced toxicity (34). Further detoxification was achieved by removal of the 1-phosphate group by expression of F. tularensis LpxE (34). Here, we showed that strain JH8033, which was engineered to produce pentaacylated lipid A, induced significantly lower TLR4 activation compared with the parental JC8031 strain producing a prototypical hexaacylated, bis-phophorylated lipid A structure. In our hands, the further expression of LpxE (or any other lipid A-modifying enzymes including PagP, PagL, and LpxR) did not further reduce activation in either the JC8031 or JH8033 strain background, suggesting that LpxM-mediated deacylation is responsible for the observed detoxification of E. coli lipid A. The resulting detoxified Ft-glycOMVs stimulated FtLPS-specific IgG antibody titers that were nearly indistinguishable from those elicited by Ft-glycOMVs bearing native E. coli lipid A, suggesting that there was no loss in adjuvant activity for glycOMVs with remodeled lipid A, including pentaacylated, monophosphorylated structures resembling MPL.

Overall, the results from this study represent a promising proof of concept for the use of engineered OMVs as a platform for the delivery of carbohydrate-based vaccines. This system combines the benefits of natural and synthetic vaccines into a singular platform that overcomes many of the production and formulation hurdles that have plagued other glycoconjugate-based vaccines. Moreover, the vesicle architecture helps surface-exposed membrane antigens (e.g., proteins and polysaccharides) maintain their physico-chemical stability (24). Our results clearly show that glycOMVs are an all-in-one antigen, adjuvant, and delivery platform that is able to generate a robust antibody response, induce a T-cell response, and confer protection against lethal challenge, whereas the polysaccharide antigen alone failed in all three criteria. Compared with conventional approaches for producing glycoconjugate vaccines, glycOMV vaccine production is significantly less complicated, less time consuming, less expensive, and more scalable. It requires only one cultivation step to generate the final product, which can be easily and economically isolated by a single ultracentrifugation step (28, 50). Another advantage is that, by combining the polysaccharide biosynthesis and conjugation steps in a single, nonpathogenic strain of E. coli, final products are well-defined and can be flexibly tailored for specific diseases simply by rewiring the polysaccharide biosynthesis pathway. Best of all, this result can be accomplished without ever having to handle or cultivate pathogenic bacteria. Finally, it should be pointed out that other biomolecular features of OMVs (e.g., proteins and lipids) can also be engineered (33), in harmony with OMV vaccine designs that contain a high density of specific carbohydrate structures, making it possible in the future to create designer OMVs against a wide variety of targets with tunable immunomodulatory effects.

Materials and Methods

Bacterial Strains and Plasmids.

The bacterial strains and plasmids used in this study are described in SI Appendix, Table S3. Briefly, E. coli strain JC8031, a tolRA mutant strain that is known to hypervesiculate (40), was used for preparation of OMVs. Strain CE8032, a waaL::Kan mutant derived from JC8031, was used as a control (57). Strain JH8033 was generated from JC8031 using P1 transduction of the lpxM::kan allele from the Keio collection as described in previous work (58). Plasmids pE, pEP, and pLPR were constructed previously (34). F. tularensis Schu S4 was provided by Jeannine Peterson (Centers for Disease Control and Prevention, Fort Collins, CO). F. tularensis subsp. holarctica LVS Iowa is the original ATCC 29684 strain that has been passaged for several years in the B.D.J. laboratory and was originally provided by Karen Elkins (US Food and Drug Administration, Rockville, MD). F. tularensis subsp. holarctica LVS RML was provided by Katy Bosio [Rocky Mountain Laboratories (RML), NIAID, NIH, Hamilton, MT]. LVS RML was originally acquired by Fran Nano (University of Victoria, Victoria, BC, Canada). The strain designations of both LVS isolates have been confirmed by the absence of pdpD, absence of pilA, and deletion in the C terminus of FTT0918 (59).

Cell Growth and Preparation of OMVs.

OMVs were prepared as described previously (28). Briefly, a plasmid containing a specific O-PS pathway (SI Appendix, Table S3) was transformed into the hypervesiculating E. coli strain JC8031, or related strain, and selected on medium supplemented with the appropriate antibiotic. An overnight culture of a single colony was subcultured into 100–200 mL of Luria–Bertani (LB) medium. The culture was grown to mid-log phase, at which time protein expression was induced with l-arabinose (0.2%) or isopropyl β-d-1-thiogalactopyranoside (IPTG) (0.1 mM), if necessary. Cell-free culture supernatants were collected 16–20 h postinduction and filtered through a 0.2-µm filter. Vesicles were isolated by ultracentrifugation (Beckman-Coulter; TiSW28 rotor; 141,000 × g; 3 h; 4 °C) and resuspended in PBS. OMVs were quantified by the bicinchoninic-acid assay (BCA Protein Assay; Pierce) using BSA as the protein standard.

Preparation of LPS.

Purified F. tularensis subsp. holarctica LPS (FtLPS), whose O-PS repeats are identical in structure to the O-PS repeats in F. tularensis subsp. tularensis Schu S4 LPS (38), was obtained from BEI Resources. LPS derived from E. coli carrying pGAB2 (Ft-glycLPS) was prepared using a modification of a previously published protocol (60). Briefly, an overnight culture of a single colony was subcultured into 500 mL of LB medium. The culture was grown overnight (16–20 h), and the cell pellet was collected by centrifugation. The pellet was resuspended in 10 mL of lysis buffer [2% (wt/vol) SDS, 4% (vol/vol) β-mercaptoethanol, and 100 mM Tris⋅HCl, pH 7.5] and heated in a boiling water bath for 10 min. Proteinase K was added to a final concentration of 2 mg/mL and incubated at 50 °C overnight. The next morning, phenol was added, and the mixture was incubated at 70 °C for 15 min, with vortexing every 5 min. The mixture was cooled on ice and then centrifuged for 10 min at 13,000 × g. The aqueous phase was collected and extracted with ether and then centrifuged for 5 min at 13,000 × g. The aqueous phase was collected, containing the LPS. This solution was dried on a glass plate to remove any residual organic phase and determine the mass of the purified LPS.

Fractionation of OMVs.

Prepared OMVs were separated by density-gradient ultracentrifugation as previously described (27). Briefly, OMVs were prepared as above but resuspended in a 50-mM Hepes, pH 6.8 solution. This solution was adjusted to 45% (vol/vol) Optiprep (Sigma) in 1.5 mL. All other Optiprep layers were prepared using the same 50-mM Hepes, pH 6.8 solution. Optiprep/Hepes gradient layers were added to a 12-mL ultracentrifuge tube as follows: 0.33 mL of 10%, 0.33 mL of 15%, 0.66 mL of 20%, 0.66 mL of 25%, 0.9 mL of 30%, 0.9 mL of 35%, and 1.5 mL of 45% containing the prepared OMVs, and enough 60% to nearly fill the tube. Gradients were centrifuged (Beckman-Coulter; TiSW41 rotor; 180,000 × g; 3 h; 4 °C), and then a total of 10 fractions of 0.5 mL each were removed sequentially from the top of the gradient. These fractions were analyzed by Western blot and dot blot analyses as described below.

Western Blot Analysis.

OMV and LPS samples were prepared for SDS/PAGE analysis by boiling for 15 min and cooling to room temperature in the presence of loading buffer containing β-mercaptoethanol. Samples were run on 12% (wt/vol) polyacrylamide gels (Mini-PROTEAN TGX; Bio-Rad) and transferred to a PVDF membrane. After blocking with a 5% (wt/vol) milk solution, membranes were probed first with a primary antibody against the specified O-PS and then with the corresponding HRP-conjugated secondary antibody (SI Appendix, Table S4). Signal was visualized using HRP substrate and either an X-ray film developer or a ChemiDoc Imaging System (Bio-Rad).

Dot Blot Analysis.

OMV samples were prepared by making the appropriate dilutions and spotting directly onto a nitrocellulose membrane, or by boiling for 10 min and cooling to room temperature before spotting on the membrane. After blocking with a 5% (wt/vol) milk solution, membranes were probed first with the mouse mAb FB11 to F. tularensis LPS and then with HRP-conjugated anti-mouse IgG. Signal was visualized using HRP substrate and either an X-ray film developer or a ChemiDoc Imaging System (Bio-Rad).

Electron Microscopy.

Structural analysis of vesicles was performed via transmission electron microscopy as previously described (28). Briefly, vesicles were negatively stained with 2% (wt/vol) uranyl acetate and deposited on 400-mesh Formvar carbon-coated copper grids. Imaging was performed using an FEI Tecnai F20 transmission electron microscope.

Preparation of Ft-glycLPS for Structure Determination.

An overnight culture of E. coli JC8031 carrying pGAB2 was subcultured in 4 L of LB medium. The culture was grown overnight, and the cell pellet was collected via centrifugation. The cell pellet was suspended in 2 mL of water, to which nine volumes of ethanol were added, and agitated for 1 h at room temperature. After centrifugation at 3,000 × g for 15 min, the supernatant was removed, and the cells were resuspended in 10 mL of 90% (vol/vol) ethanol and extracted again with nine volumes of ethanol for 15 min at room temperature with agitation. The cells were then pelleted via centrifugation and resuspended in 20 mM Tris⋅HCl, pH 8.0 containing 2 mM CaCl₂, and digested overnight with 2–5 mg/mL Proteinase K at room temperature. Digestion was followed by ultracentrifugation at 100,000 × g for 16 h. The pellet was then subjected to phenol/water extraction. All of the samples were transferred into glass tubes, freeze-dried, and resuspended in 5 mL of water. The cell suspension was heated to 65 °C with stirring and extracted with 5 mL of preheated 90% (wt/vol) phenol for 1 h. The suspension was cooled on ice, and the mixture was centrifuged at 3,000 × g. The phenol phase was reheated and reextracted with 5 mL of hot water. This process was repeated one more time. The combined aqueous phases were dialyzed (1000-Da MWCO), freeze-dried, and resuspended in 1.8 mL of 20 mM Tris⋅HCl, pH 8.0 containing 2 mM MgCl₂. A 100-μL aliquot of 7 mg/mL DNase I in 20 mM Tris⋅HCl, pH 8.0 and 2 mM MgCl₂ was added to the sample. After incubation for 3 h at 37 °C, 100 μL of 17 mg/mL RNase A was added, which was followed by another 3 h incubation at 37 °C. Finally, CaCl₂ was added to a final concentration of 2 mM, and the sample was digested with 400 µg of Proteinase K overnight at room temperature. The Proteinase K was inactivated at 100 °C for 5 min, and the samples were ultracentrifuged at 100,000 × g overnight. The pellets containing isolated LPS were lyophilized and subjected to a chromatographic separation. The crude LPS was dissolved in 50 mM ammonium acetate buffer and injected into an Agilent 1200 HPLC equipped with a refractive index (RI) detector. The separation was performed on a Superose 12 10/300 GL column, equilibrated, and eluted by 50 mM ammonium acetate, pH 5.5 at a flow rate of 0.5 mL/min. The fraction eluted at void volume was collected and freeze-dried. The dried fraction was resuspended into DOC buffer (200 mM NaCl, 10 mM Tris⋅HCl, 0.25% deoxycholate sodium, 1 mM EDTA, pH 9.2) and injected into an Agilent 1200 HPLC equipped with RI detector and a Superdex 75 10/300 GL column equilibrated previously with DOC buffer. DOC buffer was used as eluent with a flow of 0.5 mL/min. The major fractions were collected and dialyzed using a 2,000 molecular weight cut-off (MWCO) membrane against three changes of buffer containing 9% EtOH, 4 mM Tris, and 40 mM NaCl, followed by three changes of deionized (DI) water. The retentate containing the LPS was lyophilized. Next, the isolated LPS was dissolved in 500 μL of 1% acetic acid and incubated at 100 °C overnight. The supernatant was taken and freeze-dried after centrifugation at 6,000 × g for 30 min and subjected to further analysis.

NMR Spectroscopy and MALDI-TOF MS.

For NMR spectroscopy, the sample was dissolved in D₂O (99.8% D; Aldrich), freeze-dried, and again dissolved in 280 μL of D₂O (99.96% D; Cambridge Isotope Laboratories) containing 0.5 μL of acetone as internal reference. The sample was placed into a 5-mm Shigemi NMR tube with magnetic susceptibility plugs matched to D₂O; 1D proton, 2D gradient correlation spectroscopy (gCOSY), zero quantum-filtered total correlation spectroscopy (zTOCSY), adiabatic rotating frame nuclear Overhauser effect spectroscopy (ROESYad), and multiplicity-edited gradient HSQC (gHSQC) spectra were acquired on a Varian 600-MHz instrument at 30 °C. The mixing times for zTOCSY and ROESY were 80 and 200 ms, respectively. The spectra were referenced relative to the acetone signal (δ_H = 2.218 ppm; δ_C = 33.0 ppm). For MALDI-TOF analysis, the experiments were performed in reflector-positive ion mode using an AB SCIEX TOF/TOF 5800 (Applied Biosystems). The acquisition mass range was 200–6,000 Da. Samples were prepared by mixing on the target 1-μL sample solutions with 1 μL of 2,5-dihydroxybenzoic acid in 50% (vol/vol) methanol as matrix solution.

Mouse Immunizations and F. tularensis Challenge.

Six groups of 10 BALB/c female mice aged 6 to 8 wk old [National Cancer Institute (NCI)] were each immunized i.p. with 100 µL of PBS containing LPS or OMVs, prepared as described. All PBS used was at pH 7.4. The six groups were immunized with either PBS alone (control), 2 µg of native F. tularensis LPS (FtLPS), 10 μg of OMVs from JC8031 or JH8033 cells carrying no plasmid (empty OMVs), 10 μg of OMVs from JC8031 cells harboring pGAB2 (Ft-glycOMVs), or 10 µg of OMVs from JC8033 cells harboring pGAB2 and plasmids encoding for lipid A modifying enzymes. Before immunization, the LPS content of OMVs and purified LPS was quantified based on reactivity to FB11 antibody in ELISA format and by a standard colorimetric assay to detect 2-keto-3-deoxyoctonate (KDO), a core sugar component of LPS, as described previously (28). The amount of LPS in each preparation was then normalized to ensure that an equivalent amount of LPS was administered in each case. Each group of mice was boosted with an identical dosage of antigen 28 d after the priming dose. Blood was collected from five mice of each group from the mandibular sinus immediately before and 14 d after the first immunization, immediately before the boosting dose, and at 14 and 28 d after the boosting dose. Terminal splenectomies were performed on one-half (n = 5) of all six groups at 56 d after the priming dose. The remaining five mice in each of the six groups were challenged i.p. with 25 cfus of F. tularensis Schu S4, and the health of the mice was examined daily for signs of disease. A separate challenge was performed as described above with the following changes. Mice groups were immunized with either PBS alone (control), 2 µg of native F. tularensis LPS (FtLPS), 2 µg of LPS derived from JC8031 cells producing heterologous F. tularensis O-PS from pGAB2 (Ft-glycLPS), 10 μg of OMVs from JC8031 cells harboring pGAB2 (Ft-glycOMVs), or 10 µg of sham OMVs from JC8031 cells producing heterologous S. dysenteriae O-PS from pSS37 (Sd-glycOMVs). The five mice in each of the groups were challenged i.p. with 22 cfus of F. tularensis Schu S4, and the health of the mice was examined daily for signs of disease.

Challenge against F. tularensis subsp. holarctica LVS was performed as follows. Groups of five 6- to 8-wk-old BALB/c female mice (NCI) were each immunized i.p. with 100 µL of PBS containing LPS or OMVs, prepared as described. All PBS used was at pH 7.4. The groups were immunized with either PBS alone (control) or with 10 µg of OMVs from JC8031 or JH8033 cells harboring pGAB2 (Ft-glycOMVs). Mice were boosted with the same dosages 28 d after the initial immunization. At 56 d after the initial immunization, each group was challenged i.p. with either F. tularensis LVS RML or F. tularensis LVS Iowa at 4, 40, or 400 cfus. The PBS control groups were challenged with 4 cfus of either strain. The health of the mice was examined daily for signs of disease.

For all of the above experiments, when an animal became moribund, it was killed according to the procedure in the approved protocol. Mice were monitored until 14 d, at which time a Kaplan–Meier plot was generated. Statistical significance was determined using a log-rank test compared with survival of the PBS control group. The protocol number for the animal studies was no. 1305086 approved by the University of Iowa Animal Care and Use Committee.

Mucosal Response Immunizations.

Groups of 10 BALB/c female mice aged 6 to 8 wk old (The Jackson Laboratory) were each immunized s.c. with either PBS (pH 7.4) alone (control) or 100 µL of PBS (pH 7.4) containing 10 µg of Ft-glycOMVs from JH8033, 10 µg of empty OMVs from JH8033, or 2 µg of FtLPS. Each group of mice was boosted 28 d after the initial immunization with the same dosage. Blood was collected from each mouse from the mandibular sinus immediately before the initial and boost immunizations. Forty-two days after the initial immunization, the mice were killed, and blood was collected via cardiac puncture. Mucosal samples were also collected via bronchoalveolar lavage and vaginal lavage. The protocol number for the animal studies (2009-0096) was approved by the Institutional Animal Care and Use Committee at Cornell University.

Enzyme-Linked Immunosorbent Assay.

FtLPS-specific antibodies produced in immunized mice were measured via indirect ELISA using a modification of a previously described protocol (28). Briefly, sera were isolated from the collected blood draws after centrifugation at 2,200 × g for 10 min. Ninety-six–well plates (Maxisorp; Nunc Nalgene) were coated with FtLPS (5 µg/mL in PBS, pH 7.4) and incubated overnight at 4 °C. All PBS used was at pH 7.4. The next day, plates were washed three times with PBST (PBS, 0.05% Tween 20, 0.3% BSA) and blocked overnight at 4 °C with 5% (wt/vol) nonfat dry milk (Carnation) in PBS. Samples were serially diluted, in triplicate, between 1:100 and 1:12,800,000 in blocking buffer and added to the plate for 2 h at 37 °C. Plates were washed three times with PBST and incubated for 1 h at 37 °C in the presence of one of the following horseradish peroxidase-conjugated antibodies: goat anti-mouse IgG (1:25,000; Abcam), anti-mouse IgG1 (1:25,000; Abcam), anti-mouse IgG2a (1:25,000; Abcam), or anti-mouse IgA (1:5,000; Abcam). After three additional washes with PBST, 3,3′-5,5′-tetramethylbenzidine substrate (1-Step Ultra TMB-ELISA; Thermo Scientific) was added, and the plate was incubated at room temperature for 30 min. The reaction was halted with 2 M H₂SO₄. Absorbance was quantified via microplate spectrophotometer (Molecular Devices) at a wavelength of 450 nm. Serum antibody titers were determined by measuring the lowest dilution that resulted in signal three SDs above background. Statistical significance was determined using Tukey’s post hoc honest significant difference (HSD) test and compared against the PBS control case.

Intracellular Cytokine Staining.

Splenocytes were seeded in 96-well plates at a density of 1 × 10⁶ cells per well in complete RPMI 1640 and supplemented with 10% (vol/vol) FBS, 100 U/mL penicillin, 100 µg/mL streptomycin, and 50 U/mL IL-2 (eBioscience). To each well, 100 µg/mL FtLPS was added and incubated at 37 °C for 24 h. Brefeldin A (eBioscience) was added 4 h before harvesting. Cells were then harvested, blocked with anti-CD16/32, and stained with Alexa 488-conjugated anti-CD3e. Cells were washed and fixed using 2% (wt/vol) paraformaldehyde (eBioscience). Cells were then permeabilized with 0.1% saponin (eBioscience) and incubated with anti-INFγ, anti-TNFα, or anti–IL-4, all PE-Cy7.5–conjugated. Data were collected on a FACScalibur flow cytometer (Becton Dickinson) and analyzed using FlowJo (Treestar). All antibodies used in this section were sourced from eBioscience unless noted otherwise.

Characterization of Mutant Lipid A.

Lipid A was prepared from 15-mL cultures and analyzed using a MALDI-TOF/TOF (ABI 4700 Proteomics Analyzer) mass spectrometer in the negative ion linear mode as previously described (61).

TLR4 Activation Assay.

HEK-Blue hTLR4 cell lines were purchased from Invivogen and maintained according to the manufacturer’s specifications. Cells were plated into 96-well plates at a density of 1.4 × 10⁵ cells per mL in HEK-Blue detection media (Invivogen). Antigens were added at the following concentrations: 10⁴ cells per mL for whole cells; and 10 ng/mL for OMVs. Purified E. coli O55:B5 LPS (Sigma-Aldrich) and detoxified E. coli O55:B5 (Sigma-Aldrich) were added at 100 ng/mL and served as positive and negative controls, respectively. Plates were incubated at 37 °C, 5% CO₂ for 10–16 h, after which time the plates were analyzed using a microplate reader at 620 nm. Statistical significance was determined via unpaired t test.