Salmonella Synthesizing 1-Monophosphorylated Lipopolysaccharide Exhibits Low Endotoxic Activity while Retaining Its Immunogenicity

Recombinant attenuated Salmonella vaccines (RASVs) can deliver Ags from a variety of pathogens, generating a range of immune responses, including serum Abs, mucosal IgA, and a panoply of cell-mediated immune responses, at local and distal sites (1–4). However, one problematic issue in the field has been that although candidate RASVs are adequately attenuated in animal models, when administered to humans, these vaccines can produce unwanted side effects, including fever and intestinal distress (5, 6). One possible cause of this fever is the lipid A component of LPS, also known as endotoxin (7). This could be of particular concern when using live strains exhibiting regulated delayed lysis in vivo to deliver a bolus of rAg and/or to confer complete biological containment (8).

LPS, the major surface membrane component present in almost all Gram-negative bacteria, consists of lipid A, a core oligosaccharide, and a highly variable and immunogenic O-Ag polysaccharide. Lipid A (Fig. 1A) is responsible for the toxicity of LPS (9, 10). Lipid A is detected by the TLR4/myeloid differentiation factor 2 (MD-2) receptor complex of the mammalian innate immune system (11–17). The structure–activity relationship of lipid A has been extensively studied, and factors governing its immunological activity have been identified. The total number and length of the acyl chains and two phosphate groups at the 1 and 4′ positions are critical factors for full lipid A activation of human TLR4/MD-2 (18–20). Hexa-acylated Escherichia coli lipid A with both 1- and 4′-phosphate moieties and acyl chains from 12–14 carbons in length has optimal proinflammatory activity, whereas altering the number or length of the attached fatty acids or altering the charge of lipid A can reduce the magnitude of the signal (18, 19, 21). The recent TLR4–MD-2–LPS crystal structure shows that the 1- and 4′-phosphate groups interact with a cluster of positively charged residues from dimeric TLR4 and MD-2 (20). Removal of the 1- or 4′-phosphate weakens the ligand affinity and may induce structural rearrangement of the TLR4/MD-2 multimer (20).

Monophosphoryl lipid A (MPL) is used clinically as a vaccine adjuvant in Europe and Australia (22) and was recently approved for use in the United States. As an adjuvant, MPL improves vaccine efficacy, induces dendritic cell maturation, induces primarily a Th1 response, indirectly reduces the threshold for activation of Th1 cells, and upregulates MHC class II molecules (CD80 and CD86) (23–26). Lipid A activates both the TLR4–TRIF and TLR4–MyD88 pathways, whereas MPL selectively activates the TLR4–TRIF signaling pathway, leading to significantly lower secretion of proinflammatory cytokines, such as IL-6, IL-1β, and IFN-γ, than does wild-type lipid A but robust induction of G-CSF, MCP-1, and IP-10 (15). Therefore, the MPL vaccine adjuvant maintains or enhances immunostimulatory benefits but possesses reduced toxicity (21).

LpxE, an inner membrane phosphatase from Francisella tularensis subspecies novicida, strain Utah 112, can selectively remove the 1-phosphate group of lipid A in living cells of E. coli and Salmonella (Fig. 1B) (27), generating a close analog of MPL that remains covalently linked to LPS. In previous studies, lpxE was expressed from a multicopy plasmid, a method not ideal for use in vaccine strains because of potential stability issues and because target Ag genes are typically expressed from multicopy plasmids. In this study, we identified a unique chromosomal location for lpxE insertion that supports levels of lpxE transcription to provide levels of LpxE adequate for nearly complete 1-dephosphorylation of the lipid A in Salmonella. In addition, we demonstrated that lpxE expression from this chromosomal location in live Salmonella is attenuating. Our ultimate goal is to modify live attenuated Salmonella vaccine strains to produce dephosphorylated lipid A as an additional safety feature. Therefore, we introduced our optimal lpxE construct into an attenuated Salmonella strain. We found that chromosomal lpxE expression in a vaccine strain carrying additional attenuating deletions does not result in loss of efficacy when used to deliver a heterologous Ag.

Salmonella typhimurium cultures were routinely grown at 37°C in LB broth (28) or LB agar with or without 0.1% arabinose, nutrient broth (Difco), or N minimal medium (29) (pH 5.8) supplemented with 0.1% casamino acids, 38 mM glycerol, and 10 μM MgCl₂. Diaminopimelic acid was added to a concentration of 50 μg/ml for the growth of Δasd strains (30). LB agar containing 5% sucrose was used for sacB gene-based counterselection in allelic-exchange experiments. Streptococcus pneumoniae WU2 was cultured on brain heart infusion agar containing 5% sheep blood or in Todd-Hewitt broth plus 0.5% yeast extract. MOPS minimal medium (31), with or without 10 μg/ml p-aminobenzoic acid, was used to confirm the phenotype of ΔpabA ΔpabB mutants.

DNA manipulations were carried out as described (32). Transformation of E. coli and Salmonella enterica was performed by electroporation. Transformants were selected on LB agar plates containing appropriate antibiotics. Selection for Asd⁺ plasmids were done on LB agar plates. The primers used in this study are listed in Supplemental Table I. For construction of the ΔpagL7 mutation, which deletes the entire pagL open reading frame, the S. typhimurium χ3761 (33) genome was used as template for cloning. A 350-bp DNA fragment containing the region upstream of the pagL gene (from ATG start codon, but not including ATG) was PCR amplified using primers PagL-Del1 and PagL-SbfI2 (Supplemental Table I); another 350-bp DNA fragment containing the region downstream of pagL gene (from TAA stop codon, but not including TAA) was PCR amplified using primers PagL-SbfI3 and pagL-del4. The two PCR fragments were purified in agarose gels, combined at a 1:1 molar ratio, and joined by PCR using primers pagL-Del1 and pagL-Del4. The resulting PCR product was digested with KpnI and XmaI and ligated to KpnI-XmaI–digested pRE112 (34), resulting in plasmid pYA4284. The same strategy was used to construct pYA4288 and pYA4287, which were used to delete the entire pagP and lpxR open reading frames, respectively. There is an SbfI site at the point of each deletion in all three plasmids. The P_lpp promoter from E. coli was amplified by Plpp-FSbfI and Plpp-R. The lpxE gene from pXYW-1 (pACYC184 carrying a 2.5-kb F. tularensis genomic DNA fragment containing lpxE) (27) was amplified using primers LpxE-F1 and LpxE-RsbfI. The two PCR fragments were purified from agarose gels, combined at a 1:1 molar ratio, and joined by PCR using primers Plpp-FSbfI and LpxE-RSbfI. The resulting PCR product was digested with SbfI and ligated with SbfI-digested, shrimp alkaline phosphatase-treated pYA4284 (ΔpagL7), pYA4288 (ΔpagP8), or pYA4287 (ΔlpxR9). Candidate plasmids were screened by PCR to verify the orientation of the insert. The resulting plasmids were pYA4291 (ΔpagL72::P_lpp lpxE), pYA4473 (ΔpagP82::P_lpp lpxE), and pYA4474 (ΔlpxR92::P_lpp lpxE). The same strategy was used to construct suicide plasmids pYA4294, pYA4295, and pYA4296 for insertion of P_lpp lpxE (codon optimized) (Table II).

The mutations were introduced into S. typhimurium by allelic exchange by conjugation with E. coli strain χ7213 (35) harboring suicide vectors pYA4284, pYA4288, and pYA4287 to generate the ΔpagL7, ΔpagP8, and ΔlpxR9 mutations, respectively. Plasmids pYA4291, pYA4473, and pYA4474 were used to generate strains χ9437 (ΔpagL72::P_lpp lpxE), χ9701 (ΔpagP82::P_lpp lpxE), and χ9703 (ΔlpxR92::P_lpp lpxE), respectively. Plasmids pYA4294, pYA4295, and pYA4296 were used to generate χ9440 (ΔpagL71::P_lpp lpxE), χ9732 (ΔpagP81::P_lpp lpxE), and χ11092 (ΔlpxR91::P_lpp lpxE), respectively.

Some mutations were also introduced into the attenuated S. typhimurium vaccine strain χ9241 using the same methods. The presence of the ΔpabA1516 and ΔpabB232 mutations in strain χ9241 were verified by the inability of the strains to grow in MOPS minimal medium without p-aminobenzoate. The presence of the ΔasdA16 mutation was confirmed by the inability to grow in media without diaminopimelic acid, as well as by PCR. The ΔaraBAD23 mutation was verified by a white colony phenotype on MacConkey agar supplemented with 1% arabinose, as well as by PCR. LPS profiles of Salmonella strains were examined by previously described methods (36).

Strains were grown in LB or N media modified to 10 μM Mg²⁺ (pH 5.8) (37). Typically, 200-ml cultures were inoculated from overnight cultures of like media to A₆₀₀ 0.02 and grown to A₆₀₀ 0.8–0.9 for LB and 0.4–0.6 for N medium. The cells were pelleted by centrifugation at 3500 × g for 20 min at 4°C, washed with PBS, and pelleted again. The cell pellets were resuspended in 20 ml PBS and transferred to chloroform-safe Nalgene bottles. Chloroform and methanol were added to a final ratio of 1:2:0.8 (chloroform/methanol/PBS; v/v/v), forming a single-phase Bligh/Dyer mixture (38). The cell suspensions were agitated by stirring for 1 h at room temperature, followed by centrifugation at 2500 × g for 20 min. The supernatants containing phospholipids were discarded, and the pellets were resuspended by vortexing in 40 ml a single-phase Bligh/Dyer mixture and transferred to a 50-ml Pyrex tube with a Teflon-lined cap. The tubes were centrifuged again, and the supernatants were discarded. The air-dried pellets were resuspended in 25 ml 50 mM sodium acetate (pH 4.5) by vortexing and probe sonication (39). After sonication, the pH was adjusted to 4.5 with acetic acid if necessary. The tubes were placed in boiling water for 30 min to release the lipid A from the LPS. After cooling and vortexing, the solutions were transferred to chloroform-safe Nalgene bottles. Chloroform and methanol were added to a final ratio of 2:2:1.8 (chloroform/methanol/aqueous sodium acetate; v/v/v), forming a two-phase Bligh/Dyer mixture (38). The mixture was vigorously shaken to extract the lipid A, and the bottles were centrifuged at 2500 × g for 20 min at 20°C to separate the two phases. The lower (chloroform) phase from each bottle was carefully removed to a round-bottom flask and dried down on a rotary evaporator. The remaining upper phase was extracted again with fresh pre-equilibrated lower phase from a two-phase Bligh/Dyer mixture (2:2:1.8, chloroform/methanol/PBS; v/v/v). After centrifugation, the lower phase was added to the same flask and dried down again. The dried lipid A species were redissolved in chloroform/methanol (2:1; v/v) by vortexing and sonication, transferred to a glass tube, dried down, sealed with a Teflon-lined cap, and stored at −80°C.

Spectra were acquired on an ABI QSTAR XL quadropole time-of-flight tandem mass spectrometer (ABI/MDS-Sciex, Toronto, ON, Canada), equipped with an electrospray ionization (ESI) source. Spectra were acquired in the negative-ion mode and typically were the accumulation of 60 scans collected from m/z 200–2000. The dried lipid A species were prepared for mass spectrometry (MS) by dissolving them in 1.5 ml chloroform/methanol (2:1, v/v). An aliquot of 10–20 μl was diluted 10-fold in the same solvent and supplemented with 2 μl piperidine (final concentration 1%, v/v). Immediately after preparation, each sample was directly infused into the ion source at 5–6 μl/min. The negative-ion ESI was carried out at −4200 V. Data acquisition and analysis were performed using Analyst QS software (ABI/MDS-Sciex).

For tissue-culture experiments, LPS was prepared from 20 ml bacterial culture with Trireagent (Sigma, St. Louis, MO), as described previously (40). The samples were repurified using the deoxycholate-phenol method to remove trace protein (41). LPS preparations were quantitated using the 3-deoxy-d-manno-octulosonic acid (Kdo) method, according to published procedures (42). Briefly, Kdo is released from LPS after 0.5 M H₂SO₄ hydrolysis. Each LPS molecule contains three Kdo moieties. After determining the Kdo concentration, we calculated the LPS concentration and diluted the samples to the concentration appropriate for each experiment.

Seven-week-old female BALB/c mice were obtained from Charles River Laboratories (Wilmington, MA). All animal procedures were approved by the Arizona State University Animal Care and Use Committees. Mice were acclimated for 7 d after arrival before starting the experiments. For determination of LD₅₀, bacteria were grown statically overnight at 37°C in LB, diluted 1:50 into fresh LB media, and grown with aeration (180 rpm) at 37°C. When the cultures reached OD₆₀₀ = 0.8–0.9, they were harvested by room temperature centrifugation at 4000 rpm, washed once, and normalized to the required inoculums density in buffered saline with gelatin (BSG) by adjusting the suspension to the appropriate OD₆₀₀ value. Groups of five mice were infected orally with 20 μl containing various doses of S. typhimurium χ3761 or its derivatives, ranging from 1 × 10³ to 1 × 10⁹ CFU. Oral infections were performed using a 20-μl pipette. Animals were observed for 4 wk postinfection, and deaths were recorded daily. Surviving mice in some groups were challenged with 1 × 10⁹ CFU wild-type strain χ3761. Animals were observed for 4 wk postinfection, and deaths were recorded daily.

To evaluate colonization, mice were orally inoculated with 20 μl BSG containing 1 × 10⁹ CFU each strain. At days 3 and 6 after inoculation, three animals per group were euthanized; Peyer’s patches (PP) and spleen and liver samples were collected. Each sample was homogenized in BSG at a final volume of 1 ml. Dilutions of 10⁻¹ to 10⁻⁶ (depending on the tissue) were plated onto MacConkey and LB agar to determine the number of viable bacteria. Twenty colonies from each animal were randomly selected to confirm genotypic markers by PCR. Each experiment was performed twice.

The murine macrophage cell line RAW264.7 (American Type Culture Collection, Rockville, MD) was maintained in DMEM (Invitrogen, San Diego, CA) supplemented with 10% FBS and 100 μg/ml each gentamicin and penicillin. The human monocytic leukemia cell line Mono Mac 6 (MM6) (Lonza, Braunschweig, Germany) was cultured in RPMI 1640 containing sodium bicarbonate (2 g/l), insulin (10 μg/ml), oxalacetic acid (1 mM), 100 μg/ml each gentamicin and penicillin, glutamine (2 mM), nonessential amino acids for MEM (1% v/v), sodium pyruvate (1 mM), folic acid (40 μg/ml), and 15% FBS. Cells were seeded in 96-well microtiter plates (2.5 × 10⁵/well) in 150 μl the above medium and cultured at 37°C with 5% CO₂. After 6 h, various dilutions of LPS preparations were added in 16 μl (10× dilution). Twenty-four hours later, culture supernatants were collected, centrifuged to remove contaminating cells, and stored at −80°C until determination of cytokine content. All experiments were performed at least twice.

Cytokine concentrations (human IL-6, mouse TNF-α, or mouse multiple cytokines) were determined using the Bio-Plex Protein Array System (Bio-Rad), according to the manufacturer’s recommendations. Serum samples were diluted with specific serum dilution buffer (Bio-Rad). Peritoneal washes were centrifuged at 4000 rpm for 10 min, and the supernatants were assayed for cytokines. Samples were incubated with Ab-coupled beads for 30 min with continuous shaking. The beads were washed three times with 100 μl wash buffer to remove unbound protein and then incubated with biotinylated detection cytokine-specific Ab for 30 min with continuous shaking. The beads were washed three times and incubated with streptavidin-PE for 10 min after incubation. The beads were washed three times in washing buffer and resuspended in 125 μl assay buffer, and the constituents of each well of microtiter plate were drawn up into the flow-based Bio-Plex suspension Array System. Cytokine concentrations were automatically calculated by Bio-Plex Manager software by using a standard curve derived from a recombinant cytokine standard; two readings were made on each bead set.

New Zealand White rabbits were fasted overnight and then anesthetized with isoflurane through an endotracheal tube; the ileum was exposed and ligated into several loops 3–5 cm long using 1-cm spacers. S. typhimurium strains were injected into separate loops in a volume of 1 ml at a titer of 1 × 10⁹ CFU. LB broth was injected into one of the loops as a control. The abdominal musculature was closed using 3-0 chromic gut sutures, and the skin was closed with 3-0 Ethilon sutures. Rabbits were maintained in a thermal blanket at 37°C. After 8 h, each rabbit was euthanized with an overdose of sodium pentobarbital. The abdomen was reopened, and the fluid within the ligated loops was collected, the volume was measured, and the bacterial content was enumerated. The loops were fixed in 10% formalin and subjected to H&E staining histopathological examination.

rPspA protein was purified as described (43). S. typhimurium LPS was obtained from Sigma. The rPspA clone was a kind gift from Dr. Susan Hollingshead (University of Alabama at Birmingham).

Protein samples were boiled for 5 min in loading buffer and separated by SDS-PAGE. For Western blotting, proteins separated by SDS-PAGE were transferred electrophoretically to nitrocellulose membranes. The membranes were blocked with 3% skim milk in 10 mM Tris-0.9% NaCl (pH 7.4) and incubated with rabbit polyclonal Ab specific for PspA (44) or anti-GroEL (Sigma) Abs. The secondary Ab was an AP-conjugated goat anti-rabbit IgG (Sigma). Immunoreactive bands were detected by the addition of BCIP/NBT solution (Sigma). The reaction was stopped after 2 min by washing with large volumes of deionized water several times.

RASV strains were grown statically overnight in LB broth with 0.1% arabinose at 37°C. The following day, 2 ml the overnight culture was inoculated into 100 ml LB broth with 0.1% arabinose and grown with aeration at 37°C to an OD₆₀₀ of 0.8–0.9. Cells were harvested by room temperature centrifugation at 4000 rpm for 15 min, and the pellet was resuspended in 1 ml BSG. Mice (n = 13/group) were orally inoculated with 20 μl BSG containing 1 × 10⁹ CFU each strain on day 0 and boosted at 5 wk with the same dose of the same strain. Blood was obtained by mandibular vein puncture at biweekly intervals. Following centrifugation, the serum was removed from the whole blood and stored at −20°C.

ELISA was used to assay serum Abs against S. typhimurium LPS, rPspA, and whole-cell bacterial suspensions (1 × 10⁹ CFU/ml), as previously described (45). Color development (absorbance) was recorded at 405 nm using an automated ELISA plate (model SpectraMax M2e; Molecular Devices, Sunnyvale, CA). Absorbance readings 0.1 higher than PBS control values were considered positive reactions.

We assessed the protective efficacy of the attenuated Salmonella expressing pspA at week 8 by i.p. challenge of immunized mice with 4 × 10⁴ CFU S. pneumoniae WU2 (46) in 200 μl BSG (47) The LD₅₀ of S. pneumoniae WU2 in BALB/c mice (n = 13/group) was 2 × 10² CFU by i.p. administration (data not shown). Challenged mice were monitored daily for 30 d.

Numerical data are expressed as means ± SEM. A two-way ANOVA analysis, followed by the Bonferroni multiple-comparison test, was used to evaluate differences in Ab titer data. One-way ANOVA analysis, followed by the Dunnett multiple-comparison test, was used to evaluate cytokine level and colonization for multiple comparisons among groups. The median lethal dose (LD₅₀) was estimated using a probit analysis based on the XLSTAT. The Kaplan–Meier method was used for survival, and differences were analyzed by the log-rank sum test. All analyses were performed using GraphPad Prism 5.0; p < 0.05 was considered statistically significant.

To evaluate the effect of chromosomally expressed lpxE in Salmonella, we inserted F. tularensis lpxE under constitutive transcriptional control of the strong E. coli promoter P_lpp into several positions of the Salmonella chromosome. Lipid A was isolated from each strain after mild acid hydrolysis at pH 4.5 to cleave the Kdo–lipid A linkage (48) and subjected to ESI-MS. Consistent with previous reports, the lipid A of the wild-type strain contains predominantly hexa-acylated lipid A (Fig. 1A, Supplemental Fig. 1A, [M-2H]²⁻ species near m/z 897.6). Our initial results indicated that the amount of LpxE synthesized in each strain was not adequate to 1-dephosphorylate >50% of the lipid A in the cell (Fig. 1B, Supplemental Fig. 1A, [M-2H]²⁻ species with m/z near 857.6). To reduce the levels of inflammatory lipid A further and increase the levels of 1-dephospho–lipid A, the lpxE sequence was codon optimized for high-level expression in Salmonella. Three mutant strains expressing codon-optimized P_lpp lpxE were constructed and designated χ9440 (ΔpagL71::P_lpp lpxE), χ9732 (ΔpagP81::P_lpp lpxE), and χ11092 (ΔlpxR91::P_lpp lpxE). ESI-MS of the lipid A for the mutant strains demonstrated that codon optimization led to more efficient lipid A 1-dephosphorylation than in isogenic strains expressing the native lpxE sequence (Supplemental Fig. 1B). The level of P_lpp lpxE expression was dependent on the chromosomal location, because only strain χ9732 (ΔpagP81::P_lpp lpxE) produced lipid A that was essentially completely 1-dephosphorylated (Fig. 1B, Supplemental Fig. 1B, [M-2H]²⁻ species near m/z 857.6). Based on these results, the pagP81::P_lpp lpxE cassette was chosen for further evaluation.

The structure of Salmonella lipid A is nearly identical to that of E. coli (9, 10). However, Salmonella lipid A can be covalently modified in response to environmental conditions. For example, the 4′-phosphate group can be covalently modified with a 4-amino-4-deoxy-l-arabinose (l-Ara4N) residue, and the 1-phosphate can be modified with a phosphoethanolamine moiety in response to specific environmental signals (9, 10). In addition, the fatty acyl chains can be modified by hydroxylation of the 3′ secondary myristate chain by LpxO, the addition of a secondary palmitate chain at the 2-position by PagP, or the removal of the 3 or 3′ acyl chain(s) by PagL or LpxR, respectively (Fig. 1A) (10, 49). To evaluate the effect of deleting pagL, pagP, and lpxR genes in vitro and in the presence or absence of lpxE, we constructed mutant strains χ9434 (ΔpagP8) and χ9732 (ΔpagP81::P_lpp lpxE) and triple-mutant strains χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9). As expected, the ΔpagP strains χ9434 (ΔpagP8) and χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) lack the palmitate-containing peak seen in the wild-type strain χ3761 (Fig. 1C, [M-H]⁻ species near m/z 1016.66; data not shown). Because LpxR and PagL are latent in normal laboratory growth conditions, no other differences were seen for χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) compared with χ3761. In χ9732 (ΔpagP81::P_lpp lpxE), the four major peaks detected are consistent with MPL (m/z 857.6), LpxO-modified MPL (m/z 865.5), and their acetate adducts (m/z 887.6 and 895.6, respectively) (Fig. 1C). The lipid A structures in strains χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) are similar to those in strains χ9434 (ΔpagP8) and χ9732 (ΔpagP81::P_lpp lpxE), respectively. The small MPL peak (m/z 857.6) seen in strains χ3761 (wild-type), χ9434 (ΔpagP8), and χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) is due to minor chemical 1-dephosphorylation as a result of the mild acid hydrolysis step used to liberate lipid A from LPS by cleavage of the Kdo–lipid A linkage.

The phoPQ and pmrAB two-component systems play a key role in controlling the remodeling of lipid A (10). The phoPQ regulon (which includes pmrAB) is turned on during invasion of host tissues in response to conditions that can include low pH and/or a low concentration of Mg²⁺ (10) or sublethal concentrations of cationic peptides (50, 51). Among other modifications induced, the addition of phosphoethanolamine (pEtN) to the 1-phosphate by EptA is upregulated (10). It is possible that these PhoPQ-regulated modifications may interfere with the action of LpxE.

To evaluate whether LpxE can efficiently remove the 1-phosphate group from lipid A when the phoPQ regulon is induced, strains were grown in media at low pH and low Mg²⁺ concentration, conditions that upregulate phoPQ (52), and lipid A was analyzed. ESI-MS data for χ9732 (ΔpagP81::P_lpp lpxE) showed that LpxE efficiently removed the 1-phosphate from lipid A when the phoPQ regulon was activated (Fig. 1D). In addition to the major peaks seen in cells grown on LB medium, the peaks observed in the low-pH, low-Mg²⁺ medium are consistent with 1-dephospho–lipid A, modified by l-Ara4N (Fig. 1D). Lipid A modification by ArnT (adding l-Ara4N to the 4′ phosphate) and by EptA (adding phosphoethanolamine to the 1-phosphate) each take place on the periplasmic surface of the inner membrane, the same location as 1-dephosphorylation by LpxE (10). The prevalence of l-Ara4N–modified 1-dephospho–lipid A species suggested that either LpxE acts prior to lipid A modification by ArnT or efficiently uses modified lipid A as a substrate. More importantly, despite the upregulation of EptA, LpxE clearly acts prior to addition of pEtN to the 1-position, suggesting that LpxE will function under the range of conditions experienced in the host.

We next determined the role of the individual mutations in mouse virulence (Table I). The oral LD₅₀ of wild-type strain χ3761 (1.0 × 10⁴ CFU) was similar to that previously observed (53). The ΔpagP8 mutant strain χ9434 (ΔpagP8) had the same oral LD₅₀ as did χ3761 (wild-type), extending the previous observation that a pagP mutant is unaltered for virulence in mice when introduced by the i.p. route (54). The LD₅₀ of χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) was increased 10-fold compared with the wild-type strain. However, the LD₅₀ of χ9732 (ΔpagP81::P_lpp lpxE) was ∼10⁵-fold greater than χ3761 (wild-type), although at the highest doses we observed mild to severe clinical manifestations of disease (scruffy coat, lethargy) from which some mice recovered (Table I). Strain χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) was completely avirulent, and no disease symptoms were observed even at the highest dose, indicating an LD₅₀ value ≥10⁵-fold greater than the wild-type.

Although expression of lpxE in strains χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) was attenuating, this phenotype was not due to a general growth defect, because each mutant strain had growth characteristics nearly identical to wild-type strain χ3761 when grown in LB medium (data not shown). Strains χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) exhibited an LPS profile in silver-stained SDS-PAGE gels similar to that of χ3761 (Supplemental Fig. 2A). In addition, χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) were 2-fold more sensitive to the bile salt deoxycholate and SDS than were other strains (data not shown), which may affect their survival in the intestinal tract, but the strains remained resistant to ox bile (data not shown).

Because the ability to colonize lymphoid tissues (e.g., PP, spleen, and liver) is an important attribute for live attenuated Salmonella vaccines (47, 55–57), we evaluated the effect of these mutations on tissue colonization. After oral dosing, each mutant strain was able to colonize mouse lymphoid tissues, although there were significant differences among some of the strains (Fig. 2). Notably, strain χ9434 (ΔpagP8) achieved higher titers than did the other mutants in the spleen and liver at 3 d postinfection. The titer of strains χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) in spleen and liver did not increase at 6 d postinfection, resulting in significantly lower numbers than for strains χ9434 (ΔpagP8) and χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9).

Mice that survived inoculation with 1 × 10⁶ to 1 × 10⁹ CFU of the lpxE-expressing strains χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) were orally challenged with 1 × 10⁹ CFU (1 × 10⁵ LD₅₀) of the wild-type strain χ3761 30 d after inoculation with the attenuated strains. All immunized mice survived challenge (Table I). Thus, the presence of LpxE attenuated the strains without compromising their ability to transiently colonize lymphoid tissues or their immunogenicity.

To evaluate the stability of the P_lpp lpxE gene in vivo, mice were orally inoculated with ∼1 × 10⁹ CFU of either χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) or χ9732 (ΔpagP81::P_lpp lpxE). After 3 d, spleens were harvested from the mice, and colonizing Salmonella was isolated. Spleen isolates were grown in LB and used to inoculate more mice. Our results indicated that the P_lpp lpxE gene was stable in these strains after three mouse passages, as determined by PCR analysis of 38 colonies followed by DNA- sequence analysis of three PCR⁺ colonies isolated after the third passage (data not shown).

We determined the effect of modified lipid A on cytokine stimulation in tissue culture (Fig. 3A, 3B) and in animals (Fig. 3C). In tissue culture, 0.1 pmol/ml of LPS purified from strains expressing lpxE (χ9732 and χ9705) induced significantly lower levels of the proinflammatory cytokines IL-6 in the human cell line MM6 (Fig. 3A) and TNF-α in the mouse macrophage cell line RAW264.7 (Fig. 3B) than did other strains. However, there were no differences in induction of either cytokine when cells were stimulated with 10 pmol/ml (data not shown). When sera from immunized mice were analyzed for cytokine production, we found that immunization with χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) induced lower levels of all cytokines tested, with the exception of IL-17 (Fig. 3C).

These mutant strains were also evaluated in rabbit ileal loops. Cells were introduced into ligated loops and incubated for 8 h. Histological samples from loops infected with the wild-type strain χ3761 or mutant strains were stained with H&E and examined microscopically (Fig. 4). The wild-type strain induced severe destruction of the mucosa, including large areas of the necrotic epithelia, flattened villi, and heavy infiltration by polymorphonuclear leukocytes (PMNs) (Fig. 4A, data not shown). In contrast, mutant strains χ9434 (ΔpagP8) and χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) induced some tissue destruction, although it was much milder than the wild-type strain (Fig. 4B, 4C). Loops injected with mutant strains χ9732 (ΔpagP81::P_lpp lpxE) and χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) expressing lpxE were similar to the LB control loops, with no tissue destruction or PMN infiltration apparent (Fig. 4D, 4E). These data are consistent with the mouse cytokine data that indicated that the 1-monophosphorylated LPS in Salmonella leads to a reduction in inflammatory responses postinfection.

S. typhimurium strain χ9241 (ΔpabA1516 ΔpabB232 ΔasdA16 ΔaraBAD23 ΔrelA198::araC P_BAD lacI TT) is an attenuated vaccine strain that has been successfully used to deliver the pneumococcal surface protein PspA and induce protective immunity against S. pneumoniae challenge (Table II) (44, 53). Strain χ9241 carries the ΔrelA198::araC P_BAD lacI TT deletion/insertion, which provides arabinose-regulated delayed Ag synthesis (44). To evaluate the effect of lipid A modification on the efficacy of this strain, the single or triple mutations were introduced into χ9241 to yield strains χ9844 (ΔpagP8), χ9845 (ΔpagP81::P_lpp lpxE), χ9846 (ΔpagL7 ΔpagP8 ΔlpxR9), and χ9881 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) (Table II), and the expected lipid A modifications were confirmed by ESI-MS (data not shown). Subsequently, the Asd⁺ recombinant plasmid pYA3493 (vector control) or pYA4088, which encodes a recombinant pspA gene fused to DNA encoding the β-lactamase signal sequence (44), was introduced into each strain. In plasmid pYA4088, pspA expression is driven by the P_trc promoter, and the bla signal sequence directs periplasmic secretion of PspA (Supplemental Fig. 2B).

Groups of mice were orally inoculated with 1–2 × 10⁹ CFU of each strain and boosted 5 wk later with the same dose of the same strain. Consistent with the attenuation of χ9241, there were no significant adverse health effects on the inoculated mice. Blood was taken at various times, and serum IgG titers against PspA were measured (Fig. 5A). Serum IgG titers against PspA and S. typhimurium LPS were below the limits of detection (<50) prior to immunization. Each strain synthesizing PspA triggered the production of significant levels of anti-PspA Abs compared with the vector control strain χ9241 (pYA3493) (p < 0.001). Importantly, the triple-mutant strain expressing lpxE, χ9881 (pYA4088), induced anti-PspA titers comparable to the parent strain χ9241 (pYA4088). All of the strains triggered the production of serum IgG Abs against Salmonella LPS (Fig. 5B). The anti-LPS responses were similar for all strains except χ9844(pYA4088) (ΔpagP8) and χ9881 (pYA4088) (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9), which induced significantly lower titers than did the others at week 4 (Fig. 5B). By week 8, the anti-LPS IgG responses in all immunized mice were similar. Analysis of the ratios of IgG2a/IgG1 titers against PspA (Fig. 5C) indicated that all strains induced a Th1 response typical of Salmonella (58). All strains expressing pspA elicited anti-PspA mucosal IgA responses significantly greater than those of the negative control strain χ9241 (pYA3493) (Fig. 5D).

To evaluate whether immunization elicited protective immunity, 9 wk after the primary immunization, mice were challenged i.p. with ∼4 × 10⁴ CFU virulent S. pneumoniae strain WU2 (200 × LD₅₀). Some mice in all groups immunized with strains expressing pspA survived challenge (Fig. 5E), whereas no mice inoculated with the control strain χ9241 (pYA3493) survived challenge (p < 0.01). There was no statistical difference in the level of protection elicited by any of the strains expressing pspA.

The potential usefulness of Salmonella as an Ag-delivery vector is predicated on its ability to invade and transiently persist in host tissues, thereby stimulating robust mucosal, cellular, and humoral immune responses. A virtual plethora of attenuating mutations and strategies have been described and shown to be effective at immunizing animals, particularly mice in the case of S. typhimurium (1, 3). However, results in human trials using Salmonella typhi or S. typhimurium as vectors have yet to demonstrate similar results, with vaccine strains being reactogenic and/or overattenuated, eliciting poor immune responses against vectored Ags. Rendering lipid A nontoxic may provide a means to safely increase the dose and/or reduce the level of attenuation to facilitate induction of a robust response to the vectored Ag. This may also provide an adequate level of safety to allow the use of live attenuated bacterial vaccines in infants and children, a target group that can benefit the most from this technology and for strains exhibiting regulated delayed lysis (8).

Salmonella lipid A is a mixture of closely related species that contain five to seven fatty acid moieties, and it may be modified with other polar substituents (Fig. 1A). About 15% of Salmonella lipid A is hepta-acylated, whereas the most abundant species is hexa-acylated, as in E. coli (59). Dephosphorylation of the 1-phosphate group of lipid A is sufficient to reduce or eliminate its toxic effects (15, 60). The MPL used clinically consists primarily of 3-O-deacylated-4′-MPL (3D-MPL), which is obtained by sequential mild acid and mild base hydrolysis of S. enterica serovar Minnesota R595 LPS (61). It is a mixture of penta- and hexa-acylated species. The hexa-acylated component of 3D-MPL lacks the hydroxymyristic acid moiety at position 3 and possesses a secondary palmitate chain on the 2 position R-3-hydroxymyristate chain (Fig. 1A). Initially, we attempted to generate 3D-MPL in S. typhimurium by the combined overexpression of pagL, pagP, and lpxE. However, we were unable to achieve PagP- and PagL-mediated lipid A modification >50% of the total lipid A content under our standard laboratory growth conditions (data not shown). We then opted to eliminate pagL and pagP to engineer Salmonella synthesizing the 4′-monophosphoryl-hexa-acylated lipid A variant shown in Fig. 1B, which is also an excellent adjuvant.

In this report, we demonstrated that insertion of lpxE into the Salmonella chromosome resulted in dephosphorylation of the lipid A, although LpxE activity was dependent on the position of lpxE in the chromosome. These results are consistent with the observations of Beckwith et al. (62), described >40 y ago in E. coli and subsequently confirmed for Salmonella (63), demonstrating that the precise location in the bacterial chromosome can affect the level of gene transcription. Expression of codon-optimized lpxE inserted into chromosome at pagP resulted in complete removal of the lipid A 1-phosphate group in living S. typhimurium cells (Fig. 1) under a variety of growth conditions (Supplemental Fig. 1B).

Expression of lpxE was attenuating for virulence (Table I). This virulence reduction seems to be directly related to the production of detoxified lipid A, because a ΔpagP mutant was not attenuated. Evidence supporting this correlation comes from the finding that the mutants expressing lpxE induced lower levels of cytokine secretion compared with the parent strain in in vitro and in vivo models (Figs. 3, 4). Inoculation of mice with strain χ9705, in which 1-dephosphorylation was essentially complete, resulted in lower induction levels of a number of proinflammatory cytokines, including TNF-α and IL-6 (Fig. 3), while retaining overall immunogenicity (Table I). It is difficult to compare these results directly to what occurs when humans are vaccinated with a live Salmonella vaccine for two reasons. First, because S. typhi does not infect mice, it is not possible to generate meaningful cytokine data with S. typhi vaccines in mice. Second, cytokine measurements in circulating human blood are not typically reported in trials with attenuated S. typhi vaccines. However, in one study, peripheral blood monocytes were isolated from persons vaccinated with one of several S. typhi vaccine strains, and the levels of cytokines produced in response to S. typhi flagella and LPS were measured (64). In typhoid patients, infected with S. typhi, TNF-α levels ranging from 2609–6338 pg/ml and IL-6 levels ranging from 2416–7713 pg/ml of serum in the acute and convalescent phases, respectively, were reported (65). These cytokine levels were substantially higher than were those we observed in mice with any of our S. typhimurium strains, including the wild-type strain χ3761 (Fig. 3).

Although the cytokine levels in the tissue-culture assays were higher than expected, the IL-6 levels that we obtained for χ9705 LPS stimulation of MM6 cells were similar to those reported previously for MM6 cells exposed to MPL (66). In another study using MM6 cells, there was ∼3-fold difference in TNF-α production by wild-type lipid A and MPL (67). The lipid A preparation from parent strain χ3761 elicited nearly 2-fold greater levels of TNF-α than did the MPL from χ9705 (Fig. 3). The difference in results may be due to effects caused by other structural differences between MPL and the lipid A from χ9705.

Interestingly, triple-mutant strain χ9705 was slightly more attenuated than was χ9732 (ΔpagP81::P_lpp lpxE), indicating that deletion of pagL and/or lpxR, in the presence of ΔpagP81, reduced virulence. We observed a similar correlation when comparing strains χ9434 (ΔpagP8) and χ9485 (ΔpagL7 ΔpagP8 ΔlpxR9) that did not express lpxE (Table I). Although it is not clear whether deletion of pagL, lpxR, or both is required for this small decrease in virulence, we speculate that the pagL deletion is responsible. The actions of PagL and PagP serve to decrease the interaction of lipid A with TLR4, which was suggested to enhance the ability of Salmonella to adapt to the host environment (68). Importantly, 1-dephosphorylated lipid A did not alter the immunogenicity of attenuated S. typhimurium strain χ9241 expressing the pneumococcal Ag gene pspA (Fig. 5), indicating that this modification may be used in conjunction with other attenuating mutations to reduce reactogenicity and enhance the safety profile of the vaccine.

Our finding that the production of MPL, which induces a limited immune response in the host, reduced the virulence of S. typhimurium is in contrast to what happens with Yersinia pestis, in which the production of strongly stimulatory LPS results in a significant reduction in virulence (69, 70). The different outcomes are due to the different pathogenesis mechanisms used by the two pathogens. Y. pestis uses a weakly stimulatory LPS synthesized at 37°C to escape host detection and killing by local inflammatory responses when it infects the mammalian hosts via the bite of infected fleas. For Salmonella, induction of a local inflammatory response early in infection seems to be part of its survival strategy, providing one mechanism (among many) to facilitate its invasion of the gut mucosa (71). It is clear from previous research that Salmonella msbB mutants, which produce penta-acylated lipid A, have a reduced affinity to TLR4, resulting in attenuation (72, 73).

Our results demonstrated that production of 4′-MPL by an attenuated Salmonella strain does not diminish its ability to elicit an immune response against its own LPS (Fig. 5B) or a heterologous Ag, PspA (Fig. 5A). Attenuated Salmonella strain χ9241 (pYA4088) and its derivative, χ9881 (pYA4088), which produces 1′-MPL, both induced a strong Th1 response, as determined by the high ratio of IgG2/IgG1 (Fig. 5C). This is consistent with observations by our group and other investigators that Salmonella typically induces a Th1-biased response (43, 53, 74), although a more balanced Th1/Th2 response can be obtained by altering the Salmonella genotype or immunization strategy (43, 45).

Although strain χ9705 (ΔpagL7 ΔpagP81::P_lpp lpxE ΔlpxR9) produces nearly all of its lipid A lacking the 1-phosphate group (Fig. 1) and is completely attenuated for virulence, production of MPL is not an acceptable attenuation strategy, because a single-point mutation inactivating lpxE would render the strain as virulent as the wild-type parent (compare strains χ9485 and χ9706, Table I). We based our approach on the fact that commercial MPL has been proven safe and immunogenic for use as an adjuvant in humans. Thus, lpxE expression may enhance the safety of a Salmonella vaccine strain attenuated by other means (e.g., when using the regulated delayed lysis in vivo attenuation to deliver a bolus of protective Ag and confer complete biological containment) (8, 75). We also wanted to prevent unwanted side effects that can be caused by lipid A, such as the mild diarrhea or fevers experienced by some vaccinees that have been observed in clinical studies (5, 76–79) using attenuated S. typhi vaccines. This type of modification, although helpful and necessary, is likely to be incremental, particularly when evaluated in a mouse model, in which these types of side reactions are typically not seen with attenuated Salmonella strains.

There have been a number of recent improvements in the design of S. typhi strains for Ag delivery, including the use of promoters that allow regulated Ag expression (80, 81), refinement of attenuating mutations (3, 82), and regulated delayed attenuation (83). Now, in addition, lipid A detoxification by lpxE provides a new and useful tool for improving live Salmonella vaccines to prevent typhoid and paratyphoid fever (84) and for Ag delivery, thus improving the safety of the vaccine without compromising immunogenicity. This may permit the use of strains that are more aggressive for colonizing mucosal tissues and, thereby, enhancing immune responses without compromising safety.

We thank Jacquelyn A. Kilbourne, Dale DeNardo, Bronwyn Gunn, and Heather Matthies for expert technical assistance.

The authors have no financial conflicts of interest.