Modulation of Endotoxicity of Shigella Generalized Modules for Membrane Antigens (GMMA) by Genetic Lipid A Modifications

Background: GMMA from Gram-negative bacteria are an attractive vaccine technology, but lipopolysaccharide (LPS) reactogenicity limits use. Results: Genetic LPS modification resulting in penta-acylation reduced Shigella GMMA reactogenicity to a TLR2-mediated limit. Modifications resulting in palmitoleoylated hexa-acylated LPS triggered higher TLR4-mediated reactogenicity. Conclusion: Use of GMMA as vaccines will likely require LPS penta-acylation. Significance: Understanding the relative contribution of TLR activation guides GMMA vaccine development.

Gram-negative bacteria naturally shed particles that consist of outer membrane lipids, outer membrane proteins, and soluble periplasmic components. These particles, called native outer membrane vesicles, have been proposed for use as vaccines (1). However, the yield is usually too low for a practical vaccine production. We have developed genetic modification of bacteria to induce high level shedding of particles called generalized modules for membrane antigens (GMMA) 2 (2,3) and the corresponding industrial processes required for a practical vaccine platform (2). In the case of Shigella, the required genetic modification is a deletion of the tolR gene whose corresponding protein is involved with linking the inner and outer membranes. This development is part of a program to develop an effective and affordable vaccine for the causative agents of shigellosis, a global human health problem, especially in developing countries and in children younger than 5 years (4), with more than 125 million cases (5) and 100,000 deaths per year (6). Shigella are Gram-negative bacteria divided into 50 different serotypes based on the carbohydrate composition of the O antigen of their lipopolysaccharide (LPS) (7). A limited number of serotypes contributes to the global burden of shigellosis, but the leading disease-causing serotypes vary between regions (8). The current globally dominant serotypes are Shigella sonnei and Shigella flexneri 2a, which account for more than 20% of shigellosis cases each (9,10).
GMMA are highly immunogenic (2), in part, probably because of strong self-adjuvanticity. Because they are derived from the outer membrane of Gram-negative bacteria, they have high levels of LPS and lipoproteins, molecules that are strong activators of the innate immune response through recognition by different pattern recognition receptors, including Toll-like receptors (TLRs), a widely expressed sets of molecules present in mammalian cells (11). The receptors of particular impor-* This work was supported by European Union Seventh Framework Pro-tance for recognition of Gram-negative bacteria are TLR2 and TLR4. TLR2 is involved in the recognition of a wide range of pathogen-associated molecular patterns that include lipoproteins (di-or tri-acetylated) (11). TLR4 is the receptor involved in the recognition of LPS with MD-2, CD14, and LPS-binding protein (11,12).
Depending on the dose of GMMA to be used in a vaccine, this strong activation of innate immunity may lead to unacceptable reactions in human subjects, e.g. a febrile response or, in extreme cases, septic shock (13), especially if parenterally administered. Further genetic manipulations of the bacteria could be used to reduce this risk. The single most abundant and highly potent immunostimulatory component in GMMA is LPS (1). Its structure is amenable to genetic modification to reduce the activation of the TLR4 pathway. To assess the effect of such genetic modifications, it is important to determine the relative contribution of TLR2 and TLR4 pathways to the reactogenicity of GMMA. This needs to be tested in human cells or cell lines containing the human TLR recognition system because the cytokine response induced varies considerably between mammalian species. For example, penta-acylated lipid A is a poor inducer of cytokines from human cells but is a strong stimulator of mouse cells (14,15).
LPS consists of three main regions as follows: the glycolipid, lipid A, a core oligosaccharide, and an oligosaccharide chain (O antigen, OAg) usually consisting of 20 -40 repeating units comprising 2-8 sugar molecules (16). Lipid A, which anchors the LPS to the outer membrane of bacteria, is the endotoxic part of LPS. The most endotoxic form of lipid A consists of a hexaacylated glucosamine disaccharide phosphorylated at the 1 and 4Ј position with acyl chains from 12 to 14 carbons in length and an asymmetric (4/2) distribution (14,16,17). This structure is common to Escherichia coli and Shigella (18,19). During the synthesis of the lipid A, first a tetra-acylated core structure carrying R-3-hydroxymyristate at the 2-, 3-, 2Ј-, and 3Ј-positions, called lipid IVA, is generated, which is conserved in most of the Gram-negative bacteria (18,19). Subsequently, the late acyltransferases HtrB (also called LpxL) and MsbB (also called LpxM) transfer a lauroyl fatty acid to the 3Ј-position (HtrB (20)) and a myristoyl fatty acid at position 2Ј (MsbB (21)). In some bacteria there are alternative pathways leading to a modified structure of lipid A, and these may depend on environmental conditions, e.g. a palmitoleic acid is incorporated at the 3Ј-position (palmitoleoylation) instead of the lauroyl chain, catalyzed by LpxP in E. coli under cold shock conditions (22,23).
The total number and the length of acyl chains plus the presence of the two phosphates in positions 1 and 4Ј are critical factors for full lipid A activation of human TLR4/MD-2 (14,16,17,19). Changes to the structure of the lipid A, either by removal of critical components or by replacement of one of the acyl chains by a different acyl residue, affects the binding and recognition by TLR4 and results in a lower endotoxicity in vitro (14,16). A primary focus for reducing the reactogenicity of lipid A has been the modification of the acylation by inactivation of the genes encoding the late acyltransferases HtrB (20) and MsbB (21). The resulting predominantly penta-acylated lipid A showed a strong decrease of reactogenicity in human cells. For example, an htrB mutant of Haemophilus influenzae elicited a 40-fold lower expression of tumor necrosis factor-␣ (TNF-␣) than the wild type (24), and purified LPS from an E. coli ⌬msbB mutant had a 10,000-fold decreased ability to induce TNF-␣ production (25). Shigella carries two copies of the msbB gene, one encoded on the chromosome (msbB1) and the other on the virulence plasmid (msbB2) (26). Mutant strains of S. flexneri 5 lacking both copies showed an 11-20-fold lower capability to induce TNF-␣ from adherent human monocytes (26). An msbB double mutant strain of S. flexneri 2a was shown to elicit decreased cytokine responses from a murine macrophage line (27), but it has not been reported to be tested with human cells.
Previously, we showed that genetic lipid A modification is compatible with the high yield production process using a S. sonnei ⌬msbB GMMA-producing strain (2). However, as that study focused on production, reactogenicity tests with the resulting GMMA were not performed. In this study, we investigate the effects of msbB and htrB deletions on the lipid A structure in GMMA-producing strains of S. sonnei and S. flexneri 2a and the relative activation by the resulting GMMA of the TLR4 and TLR2 pathways in human peripheral blood mononuclear cells (PBMC) and in cells transfected with human TLR4 and TLR2. We show that the penta-acylated lipid A generated by the msbB mutation has a 600-fold lower ability to stimulate TLR4 than wild-type lipid A, whereas the penta-acylated lipid A from the S. sonnei ⌬htrB mutant is 60,000-fold less active. We further demonstrate that the cytokine production in PBMC stimulated by GMMA with penta-acylated lipid A is substantially reduced and similar for GMMA from S. sonnei ⌬htrB (800-fold) and S. sonnei and S. flexneri 2a ⌬msbB mutants (300fold) and that the residual cytokine production is largely due to TLR2 activation. We also show that, in contrast to S. sonnei ⌬htrB, a compensatory palmitoleoylation occurs in the S. flexneri 2a ⌬htrB mutant resulting in a high proportion of hexaacylated lipid A with substantial ability to activate TLR4, albeit requiring approximately an order of magnitude higher concentration than the isogenic cell line.

EXPERIMENTAL PROCEDURES
Strains and Generation of Mutations-S. sonnei 53G (28) and S. flexneri 2a 2457T (29) were chosen as parent strains. The list of Shigella mutant strains used in this study and their abbreviated identifications are listed in Table 1. As all strains used in this study are GMMA-producing strains, the abbreviated names only refer to their additional mutations and characteristics. For generation of mutants from S. flexneri 2a without virulence plasmid, a white colony was selected by white appearance on Congo red agar before the start of the genetic modification. The curing of the virulence plasmid (pINV) was confirmed by the absence of the origin of replication (ori) and the plasmid-encoded genes, virG and ospD3, using PCR. The primers are listed in Table 2. To generate the tolR deletion in S. flexneri 2a and plasmid-cured S. flexneri 2a-pINV, the same strategy and primers as described previously for the generation of the S. sonnei ⌬tolR mutant (2) were used.
The null mutation of msbB1 (26), htrB (20), and rfbG (essential for OAg biosynthesis in S. flexneri 2a (30)) was obtained by replacing the gene of interest ("gene") with an antibiotic resistance cassette, using the following strategy. The upstream and downstream regions of gene were amplified using the primer pairs gene-U and gene-D or rfbF and rfc (for the rfG knock out). The resistance cassette used to replace gene was amplified using primer pairs EcoRV.Ery.F/EcoRV.Ery.R or EcoRV.Cm.F/EcoRV. Cm.R. The fragments were inserted into pBluescript (Stratagene) so that the antibiotic resistance gene interposed the flanking regions of gene. The replacement construct (upstream region-resistance cassette-downstream region) was amplified using the primers binding to the 5Ј end of the upstream flanking region and the 3Ј end of the downstream flanking region of gene (see Table 2) and used to transform recombination-prone tolR deletion strains of S. sonnei or S. flexneri as described previously (2). In S. sonnei, the htrB gene was replaced by the chloramphenicol resistance gene cat (31). In S. flexneri 2a, msbB1 and htrB were replaced by cat, and rfbG was replaced by the erythromycin resistance gene erm (32). In the rfbG knock out, also the flanking genes rfbF and rfc (30) were partially deleted. The rfbG deletion was introduced before the msbB or htrB deletion. The msbB1 mutation was only introduced into the plasmidcured strain as the plasmid carries a second copy of msbB (msbB2 (26)). For simplicity, the mutant is referred to as ⌬msbB.
To complement strains carrying the htrB deletion, the htrB gene was amplified from S. sonnei 53G, including 239 bp upstream and 172 bp downstream using primers P1.htrBcompl-EcoRI and P2.htrBcompl-NcoI and inserted into low copy vector pACYC184 (New England Biolabs). The resulting plasmid pACYChtrB was introduced into electrocompetent S. sonnei or S. flexneri 2a ⌬htrB cells.
For GMMA production, overnight cultures were grown in the presence of the specific selective antibiotics and used to inoculate the production medium to an OD of 0.03-0.05. Production cultures were incubated at 30°C and 200 rpm overnight. Culture supernatants were collected by a 10-min centrifugation at 5,000 ϫ g followed by a 0.22-m filtration. GMMA were concentrated using an Amicon stirrer cell with a regenerated cellulose filter with a 100-kDa nominal molecular mass limit (Amicon Ultracell) under nitrogen flow. The retentate was collected in a 70-ml ultracentrifuge propylene tube (Beckman Coulter) and ultracentrifuged at 186,000 ϫ g using 45Ti rotor (Beckman Coulter) for 2 h at 4°C. Pellets were resuspended in 4 ml of PBS followed by 0.22-m filtration. GMMA were stored at 4°C.
GMMA Protein and KDO Quantification-GMMA quantities were expressed as total protein present in GMMA. The protein quantity was determined using the DC protein assay (Bio-Rad) kit (Lowry assay) according to the manufacturer's instructions. Bovine serum albumin (Pierce) was used for the standard curve in the range 1-50 g/assay. Measurements of GMMA were performed in two different dilutions, each in duplicate. Core-reducing end 2-keto-3-deoxyoctonate (KDO) after lipid A cleavage was quantified using the semicarbazide/HPLCsize exclusion chromatography method as we reported previously (33). To apply the method to GMMA, 150 g of GMMA were hydrolyzed in 1% acetic acid for 3 h at 100°C and subsequently centrifuged for 15 min at 14,000 ϫ g. Supernatants were collected and dried in a SpeedVac, and the pellets were dissolved in water. Samples and a standard of 4 -40 g/ml KDO ammonium solution (Sigma, K2755) were derivatized using semicarbazide and analyzed by HPLC-size exclusion chromatography using a TSKgel G3000 PW-XL column (TOSOH, 808021). The amount of core reducing end KDO was calculated using the calibration curve built with the peak areas of derivatized KDO standard at 252 nm.
Negative Staining Transmission Electron Microscopy-A drop of 5 l of GMMA suspension at a concentration of 100 g/ml in PBS was adsorbed onto 300 mesh copper Formvar/ carbon-coated grids for 5 min. Grids were then washed with a few drops of distilled water and dried by blotting with Whatman filter paper. For negative staining, grids were treated with 2% uranyl acetate in double distilled H 2 O for 1 min, blotted with Whatman filter paper, air-dried, and observed with a Tecnai 2 Spirit transmission electron microscope (FEI, Eindoven, The Netherlands) operating at 80 kV. Electron micrographs were recorded at a nominal magnification of 105,000ϫ. GMMA diameters were measured manually on printed copies of the electron micrographs in comparison with the scale bar. The nonparametric Kruskal-Wallis test was used to statistically compare the sizes of GMMA obtained from different strains.

SDS-PAGE and Peptide Mass
Fingerprinting-SDS-PAGE of GMMA was performed using 12% (w/v) polyacrylamide gels (Bio-Rad) as described previously (2). The gels were stained using Coomassie Blue stain (Sigma) according to the manufacturer's instructions for proteins.
Protein spots were excised from the gel and processed as described previously (34). Briefly, mass spectra were acquired on an Ultraflex MALDI TOF-TOF mass spectrometer (Bruker Daltonics) in reflectron, positive mode, in the mass range of 900 to 3,500 Da. Spectra were externally calibrated by using a combination of standards pre-spotted on the target (Bruker Daltonics). MS spectra were analyzed by peptide mass fingerprinting with FlexAnalysis (FlexAnalysis version 2.4, Bruker Daltonics). Peaks were annotated with FlexAnalysis default parameters and manually revised. Protein identification was carried from the generated peak list using the Mascot program (Mascot server version 2.2.01, Matrix Science). Mascot was run on a database containing protein sequences deduced from sequenced Shigella genomes (2).

MALDI-TOF and MS/MS Analysis on Lipid A-Lipid
A was precipitated from GMMA using mild acid hydrolysis with 1% acetic acid for 2 h at 100°C (35). Samples were centrifuged at 14,000 ϫ g for 15 min; the pellets were resuspended in water and washed twice with water. The pellets were dried overnight using a SpeedVac and resuspended in chloroform/methanol 4:1 and mixed with an equal volume of Super DHB solution (Sigma). 2 l of the mixture were loaded to the target plate (MTP 384 target plate ground steel BC, Bruker Daltonics) and analyzed by Ultraflex MALDI-TOF (Bruker Daltonics) in reflectron ion-negative mode. A peptide calibration standard (Bruker Daltonics), mixed with the Super DHB solution, was included in each analysis. For MS/MS analysis of lipid A, main peaks from the linear mode analysis were selected for collisioninduced dissociation, and the resulting fragments were detected by MALDI TOF-TOF in ion negative mode. For each sample, spectra represent the integration of the analysis of 20 different areas of the spot by 50 single laser shots. The m/z rations were determined by Flex Analysis software in comparison with the peptide standard.
Quantitative Real Time PCR and RNA Isolation-RNA was purified using RNeasy Plus mini kit (Qiagen) from 2 ml of bacteria grown at 30°C to an optical density of 1. Reverse transcription (2 g of RNA/reaction) was performed using Superscript II, and the product was purified by QIAquick PCR purification kit (Qiagen). Quantitative RT-PCR for genes msbB and lpxP and was performed using 10 ng of cDNA, SYBR Green kit (Invitrogen), and 0.4 M primers msbB.F/msbB.R and lpxP.F/ lpxP.R in thermocycler MX3005P (Stratagene) with 40 cycles (95°C, 15 s; 60°C, 60 s). Fold induction was calculated as 2 Ϫ⌬(⌬Ct) , were ⌬Ct is the difference between the numbers of cycle of amplification needed to reach the threshold (fluorescence dRn ϭ 0.018) for lpxP versus the cycle of amplification of the late acyltransferase gene msbB of the constitutive lipid A biosynthesis pathway (36) in same strain. ⌬(⌬Ct) represents the difference in the ⌬Ct of lpxP in Sf2a Ϫp Ϫ OAg ⌬htrB compared with Ss Ϫp Ϫ OAg ⌬htrB.
For the NF-B luciferase assay, 25,000 cells/well were seeded in 90 l of complete DMEM without antibiotics in 96-well Clear luciferase plates (PBI International) and incubated for 24 h at 37°C. 10 l of serial 5-fold dilutions of GMMA in PBS (0.0001-1,000 ng/ml final concentration in the assay) were added. All GMMA concentrations were tested in duplicate. After incubation for 5 h at 37°C, supernatants were aspired from each well, and cells were lysed for 20 min at room temperature using 20 l/well of 1:5 diluted "passive lysis buffer" (Promega). Produced luciferase was detected using 100 l/well luciferase assay reagent (Promega), and emitted light was immediately quantified using a luminometer Lmax II 384 (Molecular Devices). NF-B activation of cells stimulated with GMMA is expressed as fold-increase of emitted light over the average result of PBS-stimulated control cells. GMMA concen-trations needed to obtain a 3-fold (TLR4 experiments) or 10-fold (TLR2 experiments) induction of NF-B were arbitrarily used for comparing relative activity because that was in the middle of the linear part of the sigmoidal curves. They were determined as x axis intercepts from the generated curves. The nonparametric Mann-Whitney test was used to statistically evaluate the results obtained with different GMMA.
PBMC Isolation, TLR Blocking, and Stimulation (Monocyte Activation Test)-Buffy coats from three different donors were used to isolate PMBC using Ficoll density centrifugation as reported previously (38). PBMC were cultured at a density of 2 ϫ 10 5 cells/well with 180 l of RPMI 1640 medium complemented with 25 mM HEPES, 2 mM glutamine, 10% FBS, 1% penicillin/streptomycin (Invitrogen) in 96-well round bottom plates. After incubation at 37°C for 30 min, 20 l of 10-fold serial dilutions of GMMA in PBS (0.0001-1,000 ng/ml final concentration in the assay) were added. Each GMMA concentration was tested in duplicate. During experiment setup, PBMC were exposed to PBS as control to assess the baseline cytokine release and to determine the concentration of GMMA that did not trigger a significant activation of IL-6 release over baseline (0.0001 ng/ml). In further experiments, this GMMA concentration was used as background control. In blocking experiments, 25 g/ml TLR4 blocking antibody (eBioscience) and/or 15 g/ml TLR2 blocking antibody (eBioscience) in RPMI 1640 medium were added for 30 min before the addition of GMMA. Cells were incubated for 4 h at 37°C, and supernatants were recovered after centrifugation of the plates at 400 ϫ g and stored at Ϫ70°C until analysis.
Cytokine Analysis by ELISA and 7-Plex Mesoscale-Nunc MaxiSorp 96-well plates were coated overnight at 4°C with 2 g/ml human IL-6 capture antibody (eBioscience 14-7069) in PBS, subsequently washed three times with PBS with 0.05% Tween 20 (PBST), blocked for 1 h with PBS with 1% BSA at room temperature, and washed three times with PBST. 50 l of supernatants from PBMC experiments, diluted 1:4 with PBS, were incubated for 2 h at room temperature. A 2-fold dilution series of recombinant human IL-6 (eBioscience 39-8069) of 31.24 to 4,000 pg/ml in RPMI 1640 medium with 10% FBS was included as standard curve on each plate. Plates were washed three times with PBST. Bound IL-6 was detected using 2 g/ml biotin-conjugated anti-human IL-6 (eBioscience 13-7068) in PBST with 0.1% BSA for 2 h at room temperature, followed by three washes with PBST, 20 min of incubation at room temperature with streptavidin/horseradish peroxidase (R&D Systems, DY998) diluted 1:200 in PBST with 0.1% BSA, three washes with PBST, and a color reaction with 100 l/well substrate (R&D Systems, DY999) for 8 min at room temperature in the dark. The reaction was stopped by adding 50 l/well of 12.5% sulfuric acid. The plates were read at 450 and 630 nm and the A 450 -630 nm was determined. IL-6 concentrations in the samples were calculated in comparison with the standard. Results below the detection limit were assigned half of the detection limit.
Mesoscale 7-spot (MSD Technology) analysis for cytokines IL-6, IL-8, IL-1␤, TNF-␣, IL-10, IL-12, and IFN-␥ was performed with 25 l of supernatants from PBMC according to the manufacturer's instructions. Concentrations of the different cytokines in the samples were determined in comparison with the preloaded standard in the plates.
For the analysis of the cytokine release by PBMC, the average cytokine levels in the duplicate assays were plotted against the GMMA concentration. GMMA concentrations needed to obtain a 10-fold increase of IL-6 release over the average level obtained at the lowest GMMA concentration (background) were determined as x axis intercepts from the generated curves and used to compare the stimulatory activity of the different GMMA. The nonparametric Mann-Whitney test was used to statistically evaluate the results obtained with different GMMA.
For statistical analysis of the results of the TLR blocking experiments, the ratio of IL-6 produced by PBMC treated with anti-TLR4 or anti-TLR2 and the IL-6 produced by PBMC not treated with blocking antibodies, stimulated with the same concentration of GMMA, were calculated to normalize between the different experiments using PBMC from different donors. The ratio was determined for each replicate in the experiments. The nonparametric Wilcoxon signed rank test was used to assess whether the obtained ratios were significantly different from 1 (no effect by blocking).

Mutant Production and Conditions for GMMA Production-
We previously reported the generation of a S. sonnei-pSS ⌬tolR ⌬msbB mutant (Ss Ϫp Ϫ OAg ⌬msbB) that was able to grown to high optical densities in chemically defined medium developed for fermentation at 30°C. The Ss Ϫp Ϫ OAg ⌬msbB strain was cured of the virulence plasmid (2) to remove a second copy of the msbB gene (msbB2 (26)) and the OAg biosynthesis genes (39) encoded on the plasmid. Thus, we chose a plasmid-deficient background to generate the htrB mutant of S. sonnei (Ss Ϫp Ϫ OAg ⌬htrB) and also used a plasmid-and OAg-deficient background in S. flexneri 2a to compare GMMA from the respective ⌬msbB and ⌬htrB mutants to the GMMA from the S. sonnei strains.
Similar to the Ss Ϫp Ϫ OAg ⌬msbB mutant (2), the Sf2a Ϫp Ϫ OAg ⌬msbB mutant retained the ability to grow in complex media at 37°C but at a slower rate (2-fold duplication time) than a single ⌬tolR mutant. In contrast, ⌬htrB strains only grew in chemically defined or minimal media and only at 30°C. The duplication time of the ⌬htrB or ⌬msbB strains increased from 30 min to ϳ2 h, but all the strains were able to reach high ODs (OD 10) after overnight incubation in chemically defined medium at 30°C. Thus 30°C and chemically defined medium were chosen as the standard growth conditions. All strains with lipid A modifications yielded more than 50 mg/liter GMMA protein.
GMMA from Shigella strains carrying different mutations showed similar morphology by electron microscopy (Fig. 1A) with average sizes of 30 -32 nm in all six strains and a size distribution of 17-53 nm, measured with 30 GMMA per strain. A comparison of the GMMA sizes from all strains gave no significant difference (p ϭ 0.90). To characterize whether the genetic lipid A modifications might alter the protein composition of GMMA, the protein pattern of GMMA from the different mutants was evaluated by SDS-PAGE (Fig. 1B). Although the overall pattern remained similar, four protein bands, iden-tified as pyruvate dehydrogenase, glutamine synthetase, ketolacid recutoisomerase, and D-3-phosphoglycerate dehydrogenase (Fig. 1B) by peptide mass fingerprinting, were found to be up-regulated in GMMA from Sf 2a Ϫp Ϫ OAg ⌬htrB. As these proteins are cytoplasmic proteins, no effect on the reactogenicity studies was expected.

Characterization of Lipid A by MALDI-TOF and MALDI-TOF/ TOF-The lipid A of LPS of the mutants was extracted and
analyzed by MALDI-TOF. The spectra are reported in Fig. 2, and the structures of lipid A corresponding to the main peaks were assigned on the basis of mass and by comparison of results with similar mutants of E. coli (Fig. 2L) (40). The main peaks in the mass spectra obtained by MALDI-TOF from lipid A purified from GMMA from S. sonnei and S. flexneri 2a strains with wild-type (WT) LPS (Fig. 2, Ss Ϫp Ϫ OAg (A) and Sf 2a Ϫp Ϫ OAg (B)) had an m/z corresponding to the theoretical mass of the hexa-acylated lipid A of 1,798 Da. The main peaks obtained by mass spectrometry from the Ss Ϫp Ϫ OAg ⌬msbB GMMA (Fig.  2C) and Sf 2a Ϫp Ϫ OAg ⌬msbB (Fig. 2D) GMMA corresponded, in both strains, to a penta-acylated lipid A lacking a myristoyl chain (theoretical mass 1,588 Da, 210 m/z shift to WT lipid A due to the absence of a C 14 fatty acid chain) consistent with msbB knock outs.
The mass spectrum of lipid A from Ss Ϫp Ϫ OAg ⌬htrB GMMA (Fig. 2E) showed a main peak corresponding to a penta-acylated lipid A lacking a lauroyl chain (theoretical mass 1,616 Da, corresponding to the absence of a C 12 fatty acid chain giving a m/z shift of 182), consistent with an htrB knock out. The spectra obtained from GMMA of Sf 2a Ϫp Ϫ OAg ⌬htrB (Fig. 2F) also had the penta-acylated lipid A species with the deletion of a lauroyl chain but showed a new peak at m/z 1,850 (Fig. 2F), most likely corresponding to a hexa-acylated lipid A species with an m/z different from WT lipid A. The mass of this lipid A species corresponds to acylation by a palmitoleoyl chain (a C 16 (Fig. 2, H and I) showed in both cases a hexa-acylated WT lipid A as the main peak (observed, m/z 1,797; theoretical, m/z 1,797), and no heptaacylated peak (theoretical m/z 2,034 for a wild-type with extra palmitoleoyl chain) was observed.
The palmitoleoylated hexa-acylated form (m/z 1,852) was also the main form in MALDI-TOF spectra of lipid A purified from GMMA from Sf 2a Ϫp ϩ OAg (Fig. 2K), Sf 2a ϩp ϩ OAg , and S. flexneri 3a and 6 ⌬tolR ⌬htrB strains. 3 To test whether the production of the palmitoleoylated hexaacylated lipid A species (m/z 1,851) could be induced in S. sonnei under stress conditions, Ss Ϫp Ϫ OAg ⌬htrB was grown at 12°C to induce a cold stress response. In the corresponding lipid A analysis by MALDI-TOF, a small amount (signal intensity less than 5% of the main species) of the palmitoleoylated lipid A species was identified (Fig. 2J).
Palmitoleoylation of lipid A in the absence of the lauroyl chain is consistent with the activity by the late acyltransferase LpxP. Thus, the expression level of lpxP was quantified in 3 F. Citiulo, unpublished results.  TLR4-NF-B Luciferase Reporter Assay-We stimulated HEK293 cells stably transfected to express only human TLR4 recognition complex and an NF-B-inducible luciferase reporter gene (41) to characterize TLR4 stimulation by GMMA with the different lipid A species. The results obtained with different concentrations of GMMA are displayed in Fig. 3 and Table 3.
GMMA from Sf 2a Ϫp Ϫ OAg ⌬msbB and Ss Ϫp Ϫ OAg ⌬msbB stimulated similar levels of NF-B expression in the HEK293 TLR4 transfectant cells (Fig. 3) and required ϳ600-fold more GMMA than the parent GMMA (Table 3) to give a 3-fold increase in NF-B activity. In contrast, GMMA from Ss Ϫp Ϫ OAg ⌬htrB and Sf2a Ϫp Ϫ OAg ⌬htrB gave very different results. NF-B induction by Ss Ϫp Ϫ OAg ⌬htrB GMMA was only detectable at a high concentration, requiring ϳ60,000-fold more GMMA than the GMMA with wild-type lipid A and 100fold more than ⌬msbB GMMA to stimulate the same level of NF-B activity (Fig. 3 and Table 3). In contrast, Sf 2a Ϫp Ϫ OAg ⌬htrB GMMA (Fig. 3 and Table 3) retained higher TLR4 stimulation and required 10-fold less GMMA (p ϭ 0.0286) than ⌬msbB GMMA to result in a similar induction of NF-B. Accordingly, the decrease of TLR4 stimulation compared with GMMA with wild-type lipid A was the smallest (50-fold) of all tested GMMA with lipid A modifications.
To ensure that the observed differences are the results of differences of the stimulatory activity of the lipid A and not caused by different amounts of lipid A present in GMMA, the molar amount of lipid A per mg of protein was determined by quantifying the LPS core sugar KDO. The amounts of lipid A were similar in all GMMA. In comparison with GMMA from Ss Ϫp Ϫ OAg ⌬htrB with the lowest activity, GMMA from Ss Ϫp Ϫ OAg contained 1.1-fold, Sf2a Ϫp Ϫ OAg 2.0-fold, Sf2a Ϫp Ϫ OAg ⌬htrB 0.3-fold, Ss Ϫp Ϫ OAg ⌬msbB 0.8-fold, and Sf2a Ϫp Ϫ OAg ⌬msbB 1.1-fold the amount of lipid A.
Cytokine Release from Human PBMC-In order measure the endotoxin level of GMMA in a more natural context, and in particular to examine whether the GMMA could stimulate additional pattern recognition receptors to TLR4, GMMA purified from different mutants were used to stimulate human PBMC in the monocyte activation test (Fig. 4). GMMA purified from Shigella strains without lipid A modification (both Ss Ϫp Ϫ OAg and Sf 2a Ϫp Ϫ OAg ) induced high levels of the proinflammatory cytokines interleukin 6 (IL-6), TNF-␣, IL-1␤, and

TABLE 3 TLR4 NF-B-specific assay results
Concentrations of each type of GMMA resulting in 3-fold increased activation of NF-B were determined in four independent experiments (wild-type and ⌬htrB GMMA) or two independent experiments (⌬msbB GMMA) using TLR4/MD2/CD14 HEK293 transfectant cells as shown in Fig. 3. Results from GMMA from Ss Ϫp Ϫ OAg and Sf2a Ϫp Ϫ OAg with wild-type lipid A and GMMA from Ss Ϫp Ϫ OAg ⌬msbB and Sf2a Ϫp Ϫ OAg ⌬msbB (⌬msbB GMMA) were combined for further analysis. Fold differences of the mean concentration of the respective GMMA to stimulate 3-fold NF-B induction were calculated in comparison with GMMA with wild-type lipid A and ⌬msbB GMMA.  (colored dashed lines). Cytokine release by PBMC exposed to PBS was used as control.
The GMMA concentration required to give a 10-fold increase in IL-6 release over background (Table 4) was used for comparing relative activity (42). The same analysis performed at other points within the linear part of the curve (3-and 30-fold over background, respectively) gave similar results (data not shown).
GMMA purified from Shigella strains carrying lipid A modification gave a similar rank order in reduction of IL-6 release upon stimulation in comparison with the parent strains with wild-type lipid A ( Fig. 4 and Table 4), as observed in the TLR4specific assay ( Fig. 3 and Table 3) but with smaller differences than in the TLR4-specific assay. The amount of GMMA required to give a 10-fold increase in IL-6 release was as follows: Ϫp Ϫ OAg ⌬htrB (50ϫ) the amount of GMMA wild-type lipid A (Table 4).
TLR Blocking-With the objective of identifying the TLRs that contribute to the residual activation by GMMA with lipid A modification, PBMC were incubated with TLR2 and/or TLR4 blocking antibodies before stimulation with 1 and 10 ng/ml of GMMA, concentrations chosen to give a significant but not saturating increase of IL-6 (Fig. 5). The three GMMA with penta-acylated lipid A, Ss Ϫp Ϫ OAg ⌬htrB (Fig. 5D), Ss Ϫp Ϫ OAg ⌬msbB (Fig. 5B), and Sf 2a Ϫp Ϫ OAg ⌬msbB (Fig. 5C), gave similar results as follows: IL-6 production was substantially reduced following incubation with TLR2 blocking antibody (70 -90%) but either no reduction or minimal reduction was observed with TLR4 blocking antibody (10 -30%) suggesting residual activity was principally due to TLR2 activation. With the Ss Ϫp Ϫ OAg ⌬msbB and Sf Ϫp Ϫ OAg ⌬msbB GMMA, a small reduction was obtained with the TLR4 blocking antibody alone (Ss Ϫp Ϫ OAg ⌬msbB, p ϭ 0.0234; Sf 2a Ϫp Ϫ OAg ⌬msbB, p ϭ 0.0078 in four experiments) suggesting that there was still residual TLR4 activation. Further reduction was achieved with the combination of TLR2 and TLR4 blocking antibodies compared with TLR2 blocking antibody alone. With the Ss Ϫp Ϫ OAg ⌬htrB GMMA, no effect from TLR4 blocking was observed.
The GMMA with mostly hexa-acylated lipid A (Sf 2a Ϫp Ϫ OAg ⌬htrB, Fig. 5E) gave a substantial reduction with a TLR4 blocking antibody, i.e. 80% reduction at 1 ng/ml and 40% at 10 ng/ml GMMA concentration. TLR2 blocking showed no effect on the IL-6 release at 1 ng/ml GMMA concentration and resulted in ϳ35% reduction (average of three independent experiments) at the 10 ng/ml GMMA concentration. Incubation with both TLR2 and TLR4 blocking antibodies gave lower IL-6 production at 1 and 10 ng/ml GMMA suggesting that both TLR were still active but that the TLR4 activation was dominant especially at lower GMMA concentrations.
To confirm that the differences in the relative contribution of TLR4 and TLR2 to activation observed in the blocking experiments were primarily dependent on the differential TLR4 activation by the different GMMA, the ability of the GMMA to activate TLR2 was tested by stimulating HEK293-TLR2 transfectant cells. All four ⌬msbB or ⌬htrB GMMA required similar GMMA concentrations (2.6 -4.9 ng/ml) to give a 10-fold NF-B induction (Fig. 6).

DISCUSSION
GMMA are attractive candidates for vaccines as they present surface antigens in their natural environment and conformation. For use as vaccines, dependent on the dose, the reactogenicity needs to be reduced, as GMMA contain LPS and other TLR stimulatory components, e.g. lipoproteins. Previously, we described a high yield production process for GMMA (2). The goal of this study was to demonstrate the impact of reducing the endotoxin potential of Shigella GMMA by genetic modification of the lipid A component of the GMMA LPS. The results of deleting the msbB gene from both S. sonnei ⌬tolR and S. flexneri 2a ⌬tolR and by deleting the htrB gene from S. sonnei ⌬tolR were as expected: conversion of a hexaacylated lipid A to a penta-acylated lipid A through loss of a myristic acid (⌬msbB) or lauric acid (⌬htrB). However, we have demonstrated the presence of a palmitoleoyl chain in lipid A purified from Sf Ϫp Ϫ OAg ⌬htrB GMMA.
Palmitoleoylation in lpxL (htrB) mutants has previously been reported in E. coli. It was shown to be a moderately abundant species at 30°C and as the dominant species at 37°C (43). In S. flexneri 2a ⌬htrB, the signal of palmitoleoylated lipid A was the dominant species at 30°C, with very little signal due to the penta-acylated species present. Similar results were obtained for the S. flexneri 3 and S. flexneri 6 strains tested.
The inability to detect a hepta-acetylated lipid A in either the Sf 2a Ϫp Ϫ OAg GMMA or the Sf 2a Ϫp Ϫ OAg ⌬htrB GMMA complemented with htrB expression from pACYChtrB suggests that the palmitoleoylation is on the same site occupied by lauroyl acid in the wild-type lipid A, although further studies would be required to prove this. This would also be consistent with palmitoleoylation catalyzed by the action of LpxP, a late acyltransferase that acts at the HtrB site, as part of what has been described as a cold response in E. coli (23). Thus, the palmitoleoylation could be part of a stress response. Similarly, in the E. coli ⌬lpxL mutant, the abundance of the palmitoleoylated lipid A increased with stress (growth at 37°C (43)). In ⌬tolR mutant strains, the tolR mutation could provide stress and thus result in higher abundance of palmitoleoylated lipid A at 30°C, although why this was not found in the Ss Ϫp Ϫ OAg ⌬htrB strain examined is not clear. We were able to observe a small signal due to palmitoleoylated lipid A in this S. sonnei strain grown at 12°C indicating that there was a functional LpxP present. However, using quantitative real time PCR, we found a 7-fold higher gene expression of lpxP in Sf 2a Ϫp Ϫ OAg ⌬htrB compared with in Ss Ϫp Ϫ OAg ⌬htrB grown at 30°C, possibly resulting in a higher level of LpxP protein present at 30°C in the S. flexneri 2a ⌬htrB strain.
All ⌬htrB S. flexneri isolates examined had the palmitoleoylation of the lipid A regardless of the presence or type of OAg on the LPS (S. flexneri 2a, 3, and 6 have different O antigens) or the presence of the virulence plasmid (Sf2a Ϫp ϩ OAg ⌬htrB, Sf2a ϩp ϩ OAg ⌬htrB). For these, palmitoleoylation might be a strongly selected compensation after the htrB knock out. The reason for this is unclear. However, in line with this hypothesis, all of our attempts to introduce an lpxP knock out into S. flexneri 2a ⌬htrB failed.
Although the impact of lipid A modifications on TLR4 activation has been widely studied (24 -27, 44), the availability of GMMA from isogenic Shigella lines, all with similar concentrations of lipid A, enabled us to compare in detail the residual ability of modified lipid A to activate human TLR4 receptors either in HEK293-TLR4 transfectant cells or in human PBMC. The TLR4 reporter cell line was used to specifically assess the stimulatory activities of the different lipid A structures (45), whereas in vitro cytokine release from PBMC tests the stimulatory potential of a substance in the context of all TLRs and thus has been used to estimate human responses and to predict the safety of vaccines (38). In the TLR4 reporter cell line, we could detect no residual TLR4 activity of the penta-acylated lipid A in S. sonnei ⌬htrB, indicated by a 60,000ϫ increase in the amount of GMMA (500 ng/ml), required to give similar activation compared with GMMA with WT lipid A (0.008 ng/ml). GMMA from both S. sonnei and S. flexneri 2a ⌬msbB mutants gave similar TLR4 activation at comparable GMMA concentrations (4.7 and 4.6 ng/ml). Although this required 600ϫ more GMMA compared with GMMA with WT lipid A, the amount was 100ϫ less compared with GMMA from the S. sonnei ⌬htrB with penta-acyl lipid A, showing that the ⌬msbB penta-acyl lipid A retains some residual ability to activate human TLR4. Although the magnitude of difference in TLR4 activation between the Ss Ϫp Ϫ OAg ⌬htrB and the ⌬msbB GMMA with penta-acylated lipid A was unexpected, minor modifications in penta-acylated lipid A have been shown to affect TLR4 stimulation (46). In Sf 2a Ϫp Ϫ OAg ⌬htrB GMMA, the substitution of a lauroyl chain with the longer palmitoleoyl chain in hexa-acylated lipid A also led to approximately a 50ϫ decrease in the ability to simulate human TLR4 (0.4 ng/ml) compared with GMMA with WT lipid A. The mass spectrum showed a mixture of hexa-, penta-, and tetra-acylated lipid A, of which the palmitoleoylated hexaacylated peak was by far the dominant peak present. Although the relative height of the peaks in this experiment is not necessarily strictly proportional to the abundance in the GMMA, it seems highly unlikely that a 50ϫ reduction in activation could be due to the decrease in percentage of the hexa-acylated by the presence of these lower sized peaks but that the reduction of TLR4 stimulation is rather linked to the difference of the acyl chain composition. In PBMC, GMMA without lipid A modification predominantly stimulated the pro-inflammatory cytokines IL-6, TNF-␣, IL-1␤, and IL-8 as observed previously for Neisseria meningitidis native outer membrane vesicles (47). All GMMA with lipid A modification showed a marked decrease of cytokine release compared with GMMA with WT lipid A. IL-6 release was chosen for detailed comparison of the relative ability of GMMA to elicit cytokine release due to its role in fever pathogenesis (48). In contrast to the TLR4-specific assay, the difference in IL-6 stimulation by S. sonnei ⌬htrB GMMA and S. sonnei or S. flexneri 2a ⌬msbB GMMA was very small, only ϳ3-fold (S. sonnei ⌬htrB 800-fold and ⌬msbB 300-fold), compared with the 100-fold difference in TLR4 activation (S. sonnei ⌬htrB 60,000-fold and ⌬msbB 600-fold). This small difference is in accordance with the presence of additional TLR being stimulated by GMMA. It indicates that the decrease of GMMA reactogenicity obtained by either form of penta-acylated lipid A is close to the maximum decrease of GMMA reactogenicity achievable by lipid A modification, when assayed in a complex system as PBMC that is more relevant for estimating responses in humans (38). Based on the large reduction of in vitro cytokine release from PBMC, an S. sonnei ⌬htrB GMMA was chosen as a first candidate for vaccine development. Phase 1 clinical trials are currently underway and will give an important indication of the tolerability of GMMA from these constructs.
To characterize which TLR receptors contribute to the residual activity of GMMA, TLR blocking experiments in PBMC were performed. We demonstrated that the remaining activity of GMMA with penta-acylated lipid A was predominantly due to TLR2 activation. With GMMA from Ss ϪpϪOAg ⌬htrB, no effect from TLR4 blocking was observed, whereas GMMA resulting from a msbB deletion retained detectable TLR4 activity in line with the results of the TLR4-specific assay. However, the induction of IL-6 via lipid A by the ⌬msbB GMMA that was dependent on TLR4 was smaller than the induction via TLR2. In contrast, GMMA from S. flexneri 2a ⌬htrB with palmitoleoylated hexa-acylated lipid A primarily stimulated TLR4. Further reduction of reactogenicity of GMMA from Ss Ϫp Ϫ OAg ⌬htrB and the ⌬msbB strains would require measures to reduce TLR2 activation. However, native outer membrane vesicles from N. meningitidis group B ⌬lpxL1 strain with penta-acylated lipid A (49) have been shown to be safe in Phase I clinical trials (50). Thus, although N. meningitidis and Shigella are not closely related and e.g. have different LPS structures (49) modification of TLR2 activators is not expected to be required.
The broad aim of this study was to examine ways of reducing the reactogenicity of GMMA to make them suitable for use as a human vaccine, but it resulted in a surprisingly complex outcome. Deleting the htrB gene from S. sonnei or the msbB gene from S. sonnei and S. flexneri 2a resulted in GMMA with a penta-acyl lipid A with a marked reduction in induction of inflammatory cytokines from human PMBC for which the residual activity is probably due to non-lipid A-related TLR2 activation. In contrast, compensatory palmitoleoylation in the S. flexneri 2a ⌬htrB GMMA results in retained TLR4 activation. Although the level of reduction of GMMA reactogenicity required for an acceptable vaccine will depend on the dose required to give a strong immune response, which can only be determined in clinical trials, the data suggest that the use of GMMA as vaccines will likely require lipid A penta-acylation and that GMMA with palmitoleoylated hexa-acylated LPS are less likely to result in a useful vaccine.