A Molecular Mechanism for Lipopolysaccharide Protection of Gram-negative Bacteria from Antimicrobial Peptides*

Cationic antimicrobial peptides serve as the first chemical barrier between all organisms and microbes. One of their main targets is the cytoplasmic membrane of the microorganisms. However, it is not yet clear why some peptides are active against one particular bacterial strain but not against others. Recent studies have suggested that the lipopolysaccharide (LPS) outer membrane is the first protective layer that actually controls peptide binding and insertion into Gram-negative bacteria. In order to shed light on these interactions, we synthesized and investigated a 12-mer amphipathic α-helical antimicrobial peptide (K5L7) and its diastereomer (4D-K5L7) (containing four d-amino acids). Interestingly, although both peptides strongly bind LPS bilayers and depolarize bacterial cytoplasmic membranes, only the diastereomer kills Gram-negative bacteria. Attenuated total reflectance Fourier transform infrared, CD, and surface plasmon resonance spectroscopies revealed that only the diastereomer penetrates the LPS layer. In contrast, K5L7 binds cooperatively to the polysaccharide chain andthe outer phosphate groups. As a result, the self-associated K5L7 is unable to traverse through the tightly packed LPS molecules, revealed by epifluorescence studies with LPS giant unilamellar vesicles. The difference in the peptides' modes of binding is further demonstrated by the ability of the diastereomer to induce LPS miscellization, as shown by transmission electron microscopy. In addition to increasing our understanding of the molecular basis of the protection of bacteria by LPS, this study presents a potential strategy to overcome resistance by LPS, and it should help in the design of antimicrobial peptides for future therapeutic purposes.

Cationic antimicrobial peptides are gene-encoded polypeptides, which serve as essential components of the innate immune system of all species, including humans (1)(2)(3)(4). These peptides are mobilized shortly after microbial infection and act rapidly to neutralize a broad range of microbes. This quick response is crucial, because activation of pathogen-specific immune responses occurs slowly relative to the potential kinetics of microbial proliferation (1, 5) (for a recent review, see Ref. 6). At present, hundreds of these peptides have been isolated from different biological sources (see list on the World Wide Web at www.bbcm.univ. trieste.it/ϳtossi/pag1.htm), some of which were subjected to mode of action studies. These studies indicate that the bacterial cytoplasmic membrane is one of the main targets for many of these peptides (7)(8)(9)(10)(11)(12). Interestingly, however, some antimicrobial peptides, which bind, permeate, and depolarize efficiently bacterial membranes, have limited activity on intact bacteria (11,(13)(14)(15). For example, insect cecropins are in practice active only on Gram-positive bacteria (16), compared with magainins (17), dermaseptins (18), and many others, which are active on both Gram-negative and Gram-positive bacteria. Other examples include cathelicidins, protegrins, defensins, and temporins (16 -21).
For most antimicrobial peptides that are not active against Gram-negative bacteria, the outer membrane is believed to be a major barrier (10,16,(22)(23)(24)(25)(26)(27)(28)(29). However, the details of the interactions of the peptides with these membranes are still not fully known (23, 30 -32). The outer membrane of Gram-negative bacteria mostly contains (Ͼ90%) the glycolipid lipopolysaccharides (LPS) 1 on its surface, the structure of which is not well known (23,(33)(34)(35)(36)(37)(38). LPS is generally thought to be a protective wall, rendering bacteria resistant to a variety of host defense molecules (24). Indeed, LPS allows the traverse of hydrophobic molecules at about 1-2% of the rates observed with typical phospholipid bilayers (36,39). LPS has two potential barriers; one is hydrophilic (provided by the oligosaccharide core, which is densely packed (33)), and the other is hydrophobic (provided by the hydrocarbon chain region, which contains six fatty acids per molecule (32,40)).
In order to address the question as to whether bacterial LPS is responsible for the different susceptibilities of Gram-negative bacteria toward a certain peptide, we synthesized and investigated a 12-mer amphipathic peptide (K 5 L 7 ) and its diastereomeric counterpart (containing 33% D-residues). These peptides were selected because they have the same amino acid composition, but the incorporation of D-amino acids should affect the structure and organization of the diastereomer. Importantly, we found that the diastereomer is highly active on Gram-negative bacteria, whereas the parental peptide is not.
The peptides were investigated, by using various biophysical methods, for their interaction with LPS in the form of bilayers, small (SUVs), large (LUVs) and giant unillamelar vesicles (GUVs). These include fluorescence spectroscopy, Fourier transform infrared spectroscopy (FTIR), surface plas-mon resonance (SPR) spectroscopy, epifluorescence, and transmission electron microscopies. The observed biophysical data were complemented by biological assays, both providing information about the dependence of the antimicrobial activity on the peptides' hydrophobicity, structure, and oligomeric state. Based on the data collected, we propose a molecular mechanism for the peptide-LPS interactions that allows the peptides to traverse or prevents them from transversing the LPS layer and the target inner membrane.

Peptide Synthesis and Purification
Peptides were synthesized by an Fmoc solid phase method on Rink amide 4-methyl benzhydrylamine resin, using an ABI 433A automatic peptide synthesizer. Cleavage of the peptides from the MBHA resin resulted in C terminus-amidated peptides. The peptides were purified by RP-HPLC on a C 18 reverse phase Bio-Rad semipreparative column (250 ϫ 10 mm, 300-Å pore size, 5-cm particle size). The column was eluted in 40 min, using a linear gradient of 20 -60% acetonitrile in water, containing 0.05% trifluoroacetic acid (v/v), at a flow rate of 1.8 ml/min. The purified peptides were further subjected to amino acid analysis and electrospray mass spectroscopy to confirm their compositions and molecular weights.

Preparation of SUVs, LUVs, and GUVs
Multilamellar vesicles were first made using the following procedure. PE/PG (7:3, w/w) and LPS were dissolved in chloroform/methanol (2:1), and after the solvent was removed they were suspended in buffer (10 mg/ml), heated to 60°C (only in LPS), and vortexed. The different types of vesicles were made as follows. (i) For BIAcore experiments, SUVs were prepared by sonication of LPS multilamellar vesicles as described previously (37,41). (ii) For membrane depolarization and electron microscopy experiments, LUVs were prepared using a procedure described elsewhere (36). Briefly, PE/PG multilamellar vesicles were extruded 15 times through two stacked 100-nm polycarbonate filters, whereas for LPS, the extrusions were performed at 60°C (34). (iii) For epifluorescence experiments, PE/PG/PE-Rho (7/3/0.1, w/w/w) or LPS/PE-Rho (10/ 0.1, w/w) GUVs were prepared using a procedure described previously (42). Briefly, the lipid films were dissolved in chloroform (0.1 M), and 20 l of this solution was added to a 50-ml round-bottomed flask containing 980 l of chloroform and 150 l of methanol. The aqueous phase (7 ml of PBS) was then cautiously added along the flask walls. The organic solvent was removed in a rotary evaporator under reduced pressure at 40°C and 40 rpm. The resulting aqueous solution contained GUVs in a high concentration.
Both SUVs and LUVs were visualized using a JEOL JEM 100B electron microscope (Japan Electron Optics Laboratory Co., Tokyo, Japan) as follows. A drop of the vesicles was deposited on a carboncoated grid and negatively stained with uranyl acetate. Examination of the grids showed that the vesicles were unilamellar with an average diameter of 20 -50 nm with SUV and 100 nm with LUV. GUVs were visualized using an Olympus FV500 confocal laser-scanning microscope. Examination of the samples showed that the vesicles were unilamellar with an average diameter of 10 -20 m. Phospholipid concentrations in SUVs, LUVs, and GUVs were measured by phosphate analysis (34).

Antibacterial Activity of the Peptides and Their Stability toward Enzymatic Degradation
The antibacterial activity of the peptides was examined in sterile 96-well plates (Nunc F96 microtiter plates) in a final volume of 100 l as follows. Aliquots (50 l) of a suspension containing bacteria (midlog phase) at a concentration of 10 6 colony-forming units/ml in culture medium (LB medium) were added to 50 l of water containing the peptide in serial 2-fold dilutions in water. Inhibition of growth was determined by measuring the absorbance at 600 nm with a Microplate autoreader El309 (Bio-tek Instruments) after an incubation of 18 -20 h at 37°C. Antibacterial activities were expressed as the minimal inhibitory concentration (MIC), the concentration at which 100% inhibition of growth was observed after 18 -20 h of incubation. In order to test the antibacterial activity of the peptide in serum, a similar protocol was used but with the peptides dissolved in 30% human blood serum. The Gram-negative bacterial strains used were E. coli ATCC 25922, Shigella sonnei ATCC 25931 and Salmonella choleraesuis ATCC 14028.
For stability studies, trypsin or proteinase-K (10 g/ml in PBS) was added to a solution of the peptides in PBS (1 M), and the reaction was monitored by using RP-HPLC and electrospray mass spectroscopy, as described before.

Kinetics of Killing
The in vitro bactericidal potency (time-kill curve) of the peptides toward the Gram-negative bacteria Escherichia coli ATCC 25922 was also measured. In this assay, about 10 6 colony-forming units/ml of bacteria in culture medium were separately incubated in the presence of the peptides (in their MIC values) for 1, 2, 4, 6, 8, 12, and 18 h. Aliquots of bacterial culture were plated on agar medium, and colonies were counted after incubation for 16 h at 37°C.

Hemolysis of Human Red Blood Cells (hRBCs)
Fresh hRBCs were rinsed three times with PBS (35 mM phosphate buffer, 0.15 M NaCl, pH 7.3) by centrifugation for 10 min at 800 ϫ g and resuspended in PBS. Peptides dissolved in PBS were then added to 50 l of the stock hRBC solution in PBS to make a final volume of 100 l (final erythrocyte concentration, 4% (v/v)). The resulting suspension was incubated with agitation for 60 min at 37°C. The samples were then centrifuged at 800 ϫ g for 10 min. By measuring the absorbance of the supernatant at 540 nm, we monitored the release of hemoglobin. Controls for zero hemolysis (blank) and 100% hemolysis consisted of hRBCs suspended in PBS and 1% Triton, respectively.

Increase in Membrane Permeability Induced by the Peptides
Membrane destabilization, which results in the collapse of the diffusion potential, was detected fluorimetrically as described previously (43)(44)(45). Briefly, a PE/PG or LPS LUV suspension, prepared in "K ϩ buffer" (50 mM K 2 SO 4 /25 mM HEPES-sulfate, pH 6.8), was added to an isotonic K ϩ -free buffer (50 mM Na 2 SO 4 , 25 mM HEPES-sulfate, pH 6.8), and the dye diS-C 3 -5 was then added. The subsequent addition of valinomycin created a negative diffusion potential inside the vesicles by the selective efflux of K ϩ ions, which resulted in a quenching of the dye's fluorescence (⌬⌿ ϭ Ϫ142 mV). Peptide-induced membrane permeation for all of the ions in the solution caused a dissipation of the diffusion potential, manifested by an increase in fluorescence. Fluorescence was monitored using excitation and emission wavelengths at 620 and 670 nm, respectively. The percentage of fluorescence recovery, F t , was defined by the equation, where I t represents the fluorescence observed after the addition of a peptide at time t, I o is the fluorescence after the addition of valinomycin, and I f is total fluorescence before adding valinomycin.

Transmembrane Potential Depolarization Assay with Bacteria
The depolarization assay was conducted with intact Gram-negative bacteria and bacterial spheroplasts using a protocol similar to what was reported before (25) and for mutated bacteria (46).
Intact Bacteria-E. coli ATCC 25922 bacteria were grown at 37°C with agitation until reaching a midlog phase (A 600 ϭ 0.4). The bacteria were centrifuged and washed once with buffer (20 mM glucose, 5 mM HEPES, pH 7.3), and resuspended to an A 600 of 0.05 in a similar buffer containing 0.1 M KCl. The bacteria (in the same concentrations as in the biological activity assay) were then incubated with 1 M diS-C 3 -5 until a stable reduction of fluorescence was achieved (approximately after 60 min), indicating that the dye was incorporated into the bacterial membrane. The peptides were then added from a water stock solution (1 mg/ml) and dissolved in the KCl buffer to achieve the desired concentration (0.1-5-fold the MIC). Membrane depolarization was monitored by observing the change in the intensity of the fluorescence emission of the membrane potential-sensitive dye diS-C 3 -5 (excitation wavelength ex ϭ 622 nm, emission wavelength em ϭ 670 nm), after the addition of different concentrations of the peptide.
Bacterial Spheroplasts-Spheroplasts of E. coli ATCC 25922 (LPS and peptidoglycan-free bacteria (47)) were prepared by the osmotic shock procedure as follows. First, the cells from cultures grown to A 600 ϭ 0.8 were harvested by centrifugation and washed twice with 10 mM Tris/H 2 SO 4 , 25% sucrose, pH 7.5. Next, the cells were resuspended in a washing buffer containing 1 mM EDTA. After a 10-min incubation at 20°C with rotary mixing, the cells were collected by centrifugation and resuspended immediately in water (4°C). After a 10-min incubation at 4°C with rotary mixing, the spheroplasts were collected by centrifugation. The spheroplasts were then resuspended to A 600 ϭ 0.05 in a buffer containing 20 mM glucose, 5 mM HEPES, 1 M KCl, pH 7.3. Further treatment was done exactly as described for the intact bacteria.

ATR-FTIR Measurements
This technique was used to obtain information on the influence of the peptides on the phosphate groups and on the acyl chains of LPS as well as the influence of LPS on the secondary structure of the peptides. Spectra were obtained with a Bruker equinox 55 FTIR spectrometer equipped with a deuterated triglyceride sulfate detector and coupled to an ATR device as described previously (48). Briefly, a mixture of LPS (0.5 mg) alone or with peptide (ϳ20 g) was deposited on a ZnSe horizontal ATR prism (80 ϫ 7 mm). Lipid/peptide mixtures were prepared by dissolving them together in a 1:2 MeOH/CH 2 Cl 2 mixture and drying under a stream of dry nitrogen while moving a Teflon bar back and forth along the ZnSe prism. Spectra were recorded, and the respective pure phospholipid spectra were subtracted to yield the difference spectra. The background for each spectrum was a clean ZnSe prism. Hydration of the sample was achieved by introducing excess deuterium oxide ( 2 H 2 O) into a chamber placed on top of the ZnSe prism in the ATR casting and incubating for 2 h before acquisition of the spectra. H/D exchange was considered complete after total shift of the amide II band. Any contribution of 2 H 2 O vapor to the absorbance spectra near the amide I peak region was eliminated by subtracting the spectra of pure lipids equilibrated with 2 H 2 O under the same conditions.

ATR-FTIR Data Analysis
Prior to curve fitting, a straight base line passing through the ordinates at 1700 and 1600 cm Ϫ1 was subtracted. To resolve overlapping bands, the spectra were processed using PEAKFIT TM (Jandel Scientific, San Rafael, CA) software. Second derivative spectra were calculated to identify the positions of the component bands in the spectra. These wave numbers were used as initial parameters for curve fitting with Gaussian component peaks. Position, band widths, and amplitudes of the peaks were varied until (i) the resulting bands shifted by no more than 2 cm Ϫ1 from the initial parameters, (ii) all of the peaks had reasonable half-widths (Ͻ20 -25 cm Ϫ1 ), and (iii) good agreement between the calculated sum of all of the components and the experimental spectra was achieved (r 2 Ͼ 0.99). The relative contents of the different secondary structure elements were estimated by dividing the areas of individual peaks, assigned to a specific secondary structure, by the whole area of the resulting amide I band. The results of three independent experiments were averaged.
The interactions of the peptides with LPS head groups were studied by monitoring the antisymmetric stretching vibration of the negatively charged phosphate groups, as (PO 2 ), ranging from 1220 to 1260 cm Ϫ1 (38,49). The interaction of the peptides with LPS acyl chains was studied by analyzing the polarized ATR-FTIR spectra as described previously (50,51).

CD Spectroscopy
The CD spectra of the peptides were measured with a Jasco J-500A spectropolarimeter after calibrating the instrument with (ϩ)-10-camphorsulfonic acid. The spectra were scanned at 25°C in a capped, quartz optical cell with a 0.5-mm path length. Spectra were obtained at wavelengths of 190 -250 nm. Eight scans were taken for each peptide at a scan rate of 20 nm/min. Mean residue ellipticities were expressed as [] (degrees⅐cm 2 /dmol). The peptides were scanned at 25 M in the presence of 0.1% (ϳ0.22 mM) LPS. Fractional helicity was calculated from the dichroic minimum at 222 nm, as described previously (52,53). The 222 / 208 ratio was calculated in order to estimate helix-helix interactions (54).

LPS Binding Obtained by SPR
Binding of the peptides to LPS bilayers (using L1 chip) was assessed with a BIAcore 3000 analytical system (Biacore, Uppsala, Sweden).
This technology uses SPR to study biomolecular interaction in real time. A similar approach was used to study LPS interaction with the antimicrobial peptide protegrin-1 (using an HPA sensor chip, which creates LPS monolayers) (55). The L1 sensor chip is composed of hydrophobic aliphatic chains that contain exposed polar head groups. Thus, when in contact with vesicles, a lipid bilayer is formed. We performed the protocol described by Mozsolits et al. (56). Briefly, LPS SUVs (80 l, 0.5 mM) were applied to the L1 chip surface at a low flow rate (2 l/min). To remove any multilamellar structures from the lipid surface, we increased the flow rate to 50 l/min, which resulted in a stable base line corresponding to the lipid bilayer linked to the chip surface. The negative control bovine serum albumin was injected (25 l; 0.1 mg/l in PBS) to confirm complete coverage of the nonspecific binding sites. The bilayer linked to the chip surface was then used as a model cell membrane surface to study the peptide-membrane binding. Peptide solutions (40-l solution of 0.39 -12.5 M peptide) were injected on the lipid surface at a flow rate of 5 l/min. Buffer (pH 7.0) alone was then replaced by the peptide solution for 30 min to allow peptide dissociation. Peptide solutions were either prepared in 20 mM Hepes, 10 mM EDTA, 90 mM NaCl, pH 7.0, or in 20 mM Hepes, 0.5 mM MgCl 2 , 90 mM NaCl, pH 7.0 (34). The SPR, which is expressed in resonance units (RU), detects changes in the reflective index of the surface layer of peptides and lipids in contact with the sensor chip. A sensogram was obtained by plotting the RU against time. The peptide-lipid binding event was analyzed from a series of sensograms collected at seven different peptide concentrations.
We used a kinetic model to describe the peptide-lipid interactions. More specifically, the sensograms for each peptide-lipid bilayer interaction were analyzed by curve fitting using numerical integration analysis (57). The BIAevaluation software offers different reaction models to perform complete kinetic analyses of the peptide sensograms. One curve-fitting algorithm (the two-state reaction model) was chosen on the basis of what was known about the possible binding mechanisms of lytic peptides (58). The data were fitted globally by simultaneously fitting the peptide sensograms obtained at seven different concentrations. The two-state reaction model was applied to each data set. This model describes two reaction steps (56), which, in terms of peptide-lipid interaction, corresponds to the following, where, in the first step, peptide (P) binds to lipids (L) to give PL, which is changed to PL* in the second step. PL* cannot dissociate directly to P ϩ L and may correspond to partial insertion of the peptide into the lipid bilayer. The corresponding differential rate equations for this reaction model are represented by the following, where RU 1 and RU 2 are the response units for the first and second steps, respectively, C A is the peptide concentration, RU max is the maximal response unit, and k a1 , k d1 , k a2 , and k d2 are the association and dissociation rate constants for the first and second steps, respectively. Whereas k a1 has 1/(Mϫ s) units, k d1 , k a2 , and k a2 have 1/s units. Thus, the combined affinity constant for all of the process, K, has M Ϫ1 units. We also described the relationship between the maximal binding response for each peptide concentration (RU max ) and the peptide concentration (C) to evaluate cooperativity in binding.

Epifluorescence Microscopy
For epifluorescence microscopy studies (59,60), the fluorescent dye Rho-PE was dissolved with PE/PG (7:3, w/w) and LPS in chloroform/ methanol in molar ratios [ (38). GUVs were prepared (as described above), and images were obtained using an Olympus FV500 confocal laser-scanning microscope. Liquid-condensed (LC) domains and liquid-expanded (LE) domains were distinguished. Using the fluorescent dye Rho-PE, the LE domains appear red, and the LC domains dark (38). The setting of the photomultipliers (gain and black level) was constant for the series of images. Care was taken that any photobleaching did not compromise the interpretation, and laser irradiation and other illumination were prevented between acquisitions. The confocal images were obtained at a 12-bit resolution.

Examination of LUV Damage by Electron Microscopy
Samples containing LPS LUVs were incubated with (i) 10 M K 5 L 7 and (ii) 3 and 5 M of 4D-K 5 L 7 (at which the diastereomer exhibited 50 and 80% of its maximal activity in the membrane permeability study). A drop containing the vesicles was deposited onto a carbon-coated grid and negatively stained with phosphotungstic acid (2%), pH 6.8. The grids were examined using a JEOL JEM 100B electron microscope (Japan Electron Optics Laboratory Co., Tokyo, Japan).

RESULTS
In order to correlate antimicrobial activity with binding and permeating by the bacterial LPS, we used two antimicrobial peptides; one is highly active against Gram-negative bacteria but is not hemolytic, and the other is highly hemolytic but inactive against the same bacteria. The sequence of the peptides, their biological activity, and RP-HPLC retention times are given in Table I. Both peptides have the same amino acid composition, but one contains four Leu as D-enantiomers, which rendered it significantly less hydrophobic compared with the parental peptides (shorter retention time). Because of their similar composition, the difference in their retention times is probably due to differences in their structures upon their interaction with the hydrophobic C 18 chain of the RP-HPLC column, which can be considered as a simple model for lipid environment (61). The peptides were studied for their structure and mode of action with intact bacteria, bacteria spheroplasts, phospholipid membranes, and LPS membrane systems (including SUVs, LUVs, GUVs, bilayers, and multibilayers), by using fluorescence, CD, and FTIR spectroscopy as well as SPR and confocal and electron microscopy.
Antimicrobial Activities of the Peptides in Buffer and Serum and Their Hemolytic Activity-Antibacterial activity was assayed against a representative set of Gram-negative bacteria including E. coli ATCC 25922, S. sonnei ATCC 25931, and S. choleraesuis ATCC 14028. The parental peptide, K 5 L 7 , was designed to have an amphipathic ␣-helix structure with one face composed of charged aa and the opposite face composed of hydrophobic aa. Unexpectedly, K 5 L 7 was more than 10-fold less active than its diastereomer against all Gram-negative bacteria tested (Table I). Similar results were obtained in the presence of 30% human serum. Furthermore, incorporating D-aa into the peptide's sequence rendered it less susceptible to enzymatic degradation ( Table  I). Note that, whereas the antimicrobial assay proceeds for 18 -20 h at 37°C, all other assays took place for significantly shorter times (0.5-2 h). We therefore performed time-killing experiments in which we monitored the killing of bacteria as a function of time. The data revealed similar kinetics for the two peptides at their MICs; 90% suppression of bacteria growth was observed 1 h following the addition of 10 M 4D-K 5 L 7 and 100 M K 5 L 7 . These results enabled us to correlate between the biological and biophysical assays.
Membrane Depolarization Induced by the Peptides-The ability of the peptides to depolarize the bacterial inner and outer membranes was determined in PE/PG LUVs (62), LPS LUVs, E. coli spheroplasts, and intact bacteria, as described previously (25,46). Briefly, the peptides were mixed at various concentrations with vesicles (or bacterial cells) pretreated with the fluorescent dye, diS-C 3 -5, and valinomycin (for vesicles only). The fluorescent dye exhibits decreased fluorescence when it partitions into the membrane. An increase in fluorescence in the presence of a peptide indicates its ability to depolarize the membrane. Fig. 1 shows the dose-dependent dissipation of the transmembrane potential of PE/PG LUVs (A), LPS LUVs (B), E. coli spheroplasts (C), and intact cells (D). The data revealed that both K 5 L 7 and 4D-K 5 L 7 induced the depolarization of PE/PG LUVs and E. coli spheroplasts to a similar extent, despite the significant difference in their antimicrobial activities (Table I). However, a direct correlation was found between the antimicrobial activity of the peptides and their ability to depolarize LPS LUVs and intact E. coli cells. This evidence strongly suggests that the major target of the peptides is the inner membrane, and that their different MIC values are the result of their different abilities to traverse the outer membrane of the LPS, a process described as a self-promoted uptake (63). The incorporation of D-aa into the peptide affected its structure and oligomeric state (see below).
Secondary Structure of the Peptides in LPS Membranes, as Determined by FTIR Spectroscopy-Fourier-transformed infrared spectroscopy (FTIR) is a powerful tool available to obtain information on the mode of interaction of peptides with membranes. The secondary structure components of a peptide in a membrane-bound state can be determined from the wave numbers of the amine I vibration (followed by deconvolution) either before or after deuteration. Here we determined the secondary structure of the peptides bound to LPS after complete deuteration. The amide I region spectra as well as the fitted band components (obtained by deconvolution) of K 5 L 7 and 4D-K 5 L 7 that are bound to LPS multibilayers are shown in Fig. 2, A and  B, respectively. Assignment of the different secondary structures to the various amide I regions was calculated according to the values taken from others (64, 65) and our previous results (66 -68). The assignments and the relative areas of the component peaks are summarized in Table II. The results indicate strong bands typical for ␣-helical structures at 1650 cm Ϫ1 and the 3 10 -helix at 1658 cm Ϫ1 , for K 5 L 7 and 4D-K 5 L 7 , respectively. In support of this, the amide I region from 1656 to 1670 cm Ϫ1 is characteristic of a 3 10 -helix or dynamic/distorted ␣-helix, as was previously suggested in a study that examined the structural changes in phospholipase A 2 (69). The assignment of the amide I region from 1656 to 1670 cm Ϫ1 to the dynamic helix is further supported by studies that examined distorted ␣-helical structures (a II -helices) in bacteriorhodopsin and other proteins (70,71). Similar structures were reported previously for a model amphipathic ␣-helical peptide and its diastereomer (25). a Underlined and boldface amino acids are D-enantiomers. All the peptides are amidated at their C terminus. b A C 18 reverse phase analytical column was used. The peptides were eluted in 40 min, using a linear gradient of 20 -60% acetonitrile in water, both containing 0.05% trifluoroacetic acid (v/v). c Peptides were treated for 2 h with trypsin or proteinase-K. d The activities of the peptides in the presence of serum are shown in parentheses.
The structure of each peptide did not change when 1:60 and 1:120 peptide/lipid molar ratios were used.

Interaction of the Peptides with Phosphate Groups and Acyl Chains in LPS-
The antisymmetric stretching vibration of the negatively charged phosphate as (PO 2 Ϫ ) was used to indicate an interaction of the peptides with the backbone region of LPS. In the range of this vibration, namely 1280 -1230 cm Ϫ1 , a drastic change in the band shape can be seen. More specifically, the band is split into two vibrational bands, indicating different hydration states (Fig. 2C, bands II and III), with the band at the higher wave number (band III) representing a lower hydration (38,40). The addition of both K 5 L 7 and 4D-K 5 L 7 led to a strong reduction of the band intensities, indicating immobilization of the phosphate groups. Interestingly, the binding of K 5 L 7 led to a reduction of only band II, whereas the binding of 4D-K 5 L 7 reduced both bands (II and III). This is probably due to the inability of the aggregated all-L-peptide (see below) to reach the less exposed inner phosphate (band III).
The ability of the peptides to disrupt lipid acyl chains was evaluated by using polarized ATR-FTIR spectroscopy. The symmetric ( sym (CH 2 ) Ϸ 2853 cm Ϫ1 ) and the antisymmetric ( sym (CH 2 ) Ϸ 2922 cm Ϫ1 ) vibrations of lipid methylene C-H bonds are perpendicular to the molecular axis of a fully extended hydrocarbon chain. Thus, measuring the dichroism of infrared light absorbance can reveal the order of the membrane sample relative to the prism surface. The effect of the peptides on the acyl chain order was estimated by comparing the CH 2 stretching dichroic ratio of LPS phospholipid multibilayers alone with that obtained with LPS-bound peptides. The calculated values (R), based on the antisymmetric vibrations, were higher for 4D-K 5 L 7 compared with K 5 L 7 , indicating that the diastereomer had a stronger effect on the acyl chain order of LPS (Table II). Similar results were obtained based on the symmetric vibrations. These results are in agreement with the notion that the diastereomer penetrates better than K 5 L 7 into the LPS acyl chains.
Secondary Structure of the Peptides in LPS Dispersions as Examined by Circular Dichroism Spectroscopy-The secondary structure of the peptides (25 M) was determined in LPS dispersions (0.1%, ϳ0.22 mM). The data revealed a 78% ␣-helical structure for K 5 L 7 , characterized by double minima at 208 and 222 nm (Fig. 3). As expected, the estimated helicity was significantly decreased (20%) for 4D-K 5 L 7 . This is in agreement with

FIG. 1. Maximal dissipation of the diffusion potential in lipid vesicles (A and B), membranes of E. coli spheroplasts (C) and intact E. coli cells (D).
In the case of vesicles, the peptides were added to isotonic K ϩ -free buffer containing LUVs composed of PE/PG (A) or LPS (B), pre-equilibrated with the fluorescent dye diS-C 2 -5 and valinomycin. In the case of the spheroplasts (C) and intact bacteria (D), the peptides were added to bacterial cells that were pre-equilibrated with the fluorescent dye diS-C 2 -5 for 60 min. Fluorescence recovery was measured 1-120 min (at 5-min intervals) after the peptides were added and its maxima were recorded. Designations are as follows: K 5 L 7 (•) and 4D-K 5 L 7 (E). the FTIR experiments. Note that a 222 / 208 ratio larger than 1.0 for all L-amino acid peptides characterizes interacting ␣-helices (54,72). This ratio equals 1.10 for K 5 L 7 , indicating its assembly in LPS. In contrast, no helix-helix interaction was observed for the diastereomer as revealed by FTIR (Table II) and SPR (see below).
Peptide Partitioning and Insertion into LPS Bilayers as Determined by SPR-We used SPR to determine the ability of the peptides to bind bacterial LPS in the presence or absence of Mg 2ϩ . Despite the complex structure of the LPS molecule, it forms bilayer structures similar to phospholipids (35). The binding sensorgrams of K 5 L 7 and 4D-K 5 L 7 to LPS bilayers (in Mg 2ϩ -free buffer) are shown in Fig. 4, A and B, respectively. The relationships between the maximal binding response obtained (R max ) and the peptide's concentration are also given. Both K 5 L 7 (A) and 4D-K 5 L 7 (B) interact with LPS bilayers, with the former exhibiting a cooperative binding. This supports our conclusion that K 5 L 7 but not 4D-K 5 L 7 aggregates in LPS membranes. The cationic antimicrobial peptide SNAP-29 and the human cathelicidin LL-37, which both contain two LPS binding sites, were also found to bind LPS cooperatively (49). However, this cooperativity was primarily intramolecular (i.e. involving the N-and C-terminal LPS-binding sites from the same peptide molecule).
A Two-state Model for the Peptide's Interaction with Membranes-We employed a numerical integration analysis that uses nonlinear analysis to fit an integrated rate equation directly to the sensorgrams (56). When fitting the peptide's sensorgrams globally (using different concentrations of peptide) with the simplest 1:1 Langmuir binding model, a poor fit was obtained ( 2 Ͼ 100), confirming that this model does not represent the lipid binding mechanism of the peptides. However, a significantly improved fit ( 2 Ͻ 1) was obtained by applying numerical integration of the two-state reaction model with the binding sensorgrams. This model describes two steps in the binding process between the peptides and lipids; the first step is the binding of the peptide to the surface, followed by a second step in which the peptide is inserted into the membrane's core (58). A set of peptide sensorgrams with different peptide concentrations was used to estimate the kinetic parameters from which the binding constants were derived. The average values for the binding constants are listed in Table III. The results can be summarized as follows. (i) The active 4D-K 5 L 7 has the highest affinity (ϳ18 ϫ 10 4 M Ϫ1 ) toward LPS bilayers in the absence of Mg 2ϩ . However, similar affinities were obtained with E. coli LPS bound to the ␣-helical cationic antimicrobial peptides polyphemusin I and polymyxin B, using the dansyl polymyxin displacement assay (23,73). Note that the affinity of SMAP-29, which contains two LPS binding sites (49), to E. coli LPS is similar to that of 4D-K 5 L 7 , although the latter lacks any known LPS binding domains. Furthermore, deleting one LPS binding site in SMAP-29 completely abolished LPS binding (49). (ii) Consistent with the observation that peptides generally interact with divalent cation binding sites on LPS, the addition of Mg 2ϩ resulted in a marked reduction in the affinity of both peptides to LPS. Furthermore, as expected, this reduction was mainly noted in the first binding step. This is in agreement with previous results obtained with polyphemusins binding to E. coli LPS (23). (iii) The higher affinity of 4D-K 5 L 7, compared with K 5 L 7 , is due to the second step, insertion into the membrane core. This is demonstrated by the similar K 1 values obtained for the two peptides but higher K 2 values obtained for 4D-K 5 L 7 compared with K 5 L 7 (Table III). (iv) In all cases, the contribution of the second step to the overall binding process (K 2 , insertion into the oligosaccharide core) was much smaller than the contribution of the first step (K 1 , initial binding to the surface). The experiments were repeated three times with an S.D. of 10%.
Epifluorescence Microscopy-Membranes can be divided into co-existing liquid domains with different compositions. These lipid moieties include unordered and organized bilayer domains. This division process, which is similar to raft formation in cell membranes, has already been observed in GUVs (74,75). To observe changes in the liquid expanded/liquid condensed domain structures of PE/PG and LPS GUVs, we doped the lipids with 1% Rho-PE and performed epifluorescence microscopy studies (38,59,60). For PE/PG GUVs, LE domains were formed (red areas in Fig. 5A). On the other hand, in the LPS GUVs, there was a marked reduction in the LE domain and the appearance of other structurally different areas, which are LC domains (dark areas in Fig. 5C). This demonstrates that, compared with PE/PG bilayers, LPS bilayers are generally more condensed and packed and therefore display a large impermeable barrier for the peptides, especially if they aggregate.
The Effect of the Peptides on the Morphology of LPS LUVs Examined by Electron Microscopy-The active 4D-K 5 L 7 was added to LPS LUVs at a molar concentration corresponding to an ϳ50% (3 M) and ϳ80% (5 M) increase in membrane depolarization (based on Fig. 1B). The inactive K 5 L 7 was added to LPS LUVs at 10 M, which is 2-fold higher than 4D-K 5 L 7 . After incubation for 30 min, a drop from a sample was removed, fixed, and examined using transmission electron microscopy. Interestingly, 4D-K 5 L 7 caused micellization and a decrease in  the size of the vesicles (Fig. 6, C and D). In contrast, K 5 L 7 did not damage the vesicles at all (Fig. 6B) in comparison with the untreated control (Fig. 6A).

DISCUSSION
The Activity of the Peptides on LPS Reflects Their Potency against Intact Bacteria-This study increases our knowledge regarding the molecular mechanism by which LPS, although it is highly negatively charged and considered as the first target for cationic antimicrobial peptides, can also serve as a barrier to prevent the insertion of peptides into the inner phospholipid membrane. In line with this, we found that the all-L-aa peptide, K 5 L 7 , was practically inactive against bacteria, although it is highly active in depolarizing phospholipid membranes. The increased ability of this peptide and its 4D-diastereomer to depolarize PE/PG can partially explain why these peptides are similarly active toward bacterial spheroplasts (that lack outer membrane) (Fig. 1, A and C, respectively) but cannot explain their different activities toward intact bacteria. However, we found a direct correlation between biological activity and the ability to permeate LPS (Fig. 1, B and D, respectively). Differences in activities against PE/PG, compared with LPS, may result from structural and packing differences between these two lipids. Indeed, epifluorescence micrographs using lipid GUVs showed a significant difference in the domain formation between PE/PG and LPS, particularly changes in the size of the LC and the LE domains. This change could result from intermolecular saccharide-saccharide interactions, not found in typical phospholipids, making LPS molecules tightly associated (32,33,40). This unusual tight packing is probably the reason for the decreased ability of the aggregated K 5 L 7 (based on FTIR, CD, and SPR experiments) to traverse into the cytoplasmic membrane. A similar argument was proposed for the reduced activity of melittin toward LPS and PC/cholesterol, based on membrane binding and leakage experiments (36).
LPS Binding and Permeation Are Governed by the Amphipathicity and Aggregational States of the Peptides-The structural data can further explain the different biological activities of K 5 L 7 and its diastereomer. Both peptides are unstructured in solution (data not shown) but adopt different structures when bound to LPS dispersions (Figs. 2 and 3). Nevertheless, both bind strongly to LPS bilayers, as revealed by using SPR, indicating that structural alterations within the same peptide do not interfere with LPS binding. Note that their affinities for LPS (ϳ18 ϫ 10 4 M Ϫ1 ) are similar to those of other ␣-helical cationic antimicrobial peptides (76). However, K 5 L 7 and its diastereomer have different modes of interactions; K 5 L 7 aggregates and binds in a positive cooperative manner, whereas 4D-K 5 L 7 does not (Fig. 4, C and D).
The affinities of the K 5 L 7 and its diastereomer for LPS also approach the values shown for polymyxin B (66 ϫ 10 4 M Ϫ1 ), an  antiseptic cyclic cationic acylated decapeptide (73,77). Using fluorescence studies, it has been shown that polymyxin B also recognizes LPS in a biphasic manner, corresponding to a rapid initial association followed by a slower insertion (77). The findings that K 5 L 7 has significantly reduced LPS-permeating activity but has relatively high LPS binding capacity demonstrate that binding and permeation, although consecutive, are governed by two different molecular mechanisms and are in-fluenced by different peptide parameters. Whereas a nonaggregated conformation is preferred for membrane-permeating activity, LPS binding seems to be independent of the oligomeric state of the molecule.
The Diastereomer but Not the All-L-peptide Traverses the Polysaccharide Core of LPS-The results of the ATR-FTIR spectroscopy, showing surface localization for K 5 L 7 further support the notion that the peptide cannot traverse through  7. A, schematic structure of Gramnegative outer membrane. Kdo, 3-deoxy-D-manno-oct-2-ulopyranosonic acid; P, phosphate. B, schematic representation of mode of binding of the K 5 L 7 peptide with LPS. The peptide binds first predominantly by electrostatic interactions to the LPS, self-associates, and cannot traverse into the lipid core. C, schematic representation of the possible mechanism of membrane lysis by the 4D-K 5 L 7 peptide. The peptide binds first predominantly by electrostatic interactions to the LPS and stays as monomers. The peptide then accumulates on the surface of the lipidic core until a threshold concentration is reached, followed by LPS micellization. the hydrophobic core of LPS. Regarding 4D-K 5 L 7 , it perturbs both the hydrophilic polysaccharide (head group) region and the hydrophobic lipid core of LPS (Table II). This conclusion is further supported by the different effects of the peptides on the two phosphate groups of the LPS molecules (Fig. 2C). On the basis of their different degrees of hydration (78), the two phosphate vibrational bands, at wave number 1224 cm Ϫ1 and at wave number 1260 cm Ϫ1 , originate mainly from the more hydrated 1Ј-phosphate (facing the aqueous phase) and the less hydrated 4Ј-phosphate (facing the lipidic phase), respectively (40). We found that the peptides interact differently with these two phosphates; 4D-K 5 L 7 interacts similarly with the two phosphates, but K 5 L 7 interacts only with the more hydrated 1Јphosphate (Fig. 2C). These observations indicate that 4D-K 5 L 7 penetrates deeper than K 5 L 7 into LPS. Thus, it seems reasonable to assume that the LPS polysaccharide core (including its 3-deoxy-D-manno-oct-2-ulopyranosonic acid (Kdo) moiety) (Fig.  7) prevents the oligomeric (and therefore bulky) K 5 L 7 from reaching the lipid core.
The Diastereomeric Peptide Depolarizes and Micellizes the Bacterial Outer Membrane as Part of Its Lytic Mechanism-LPS molecules normally occupy up to 90% of the outer leaflet of the outer membrane and are connected by the Mg 2ϩ ions to form an oriented and highly ordered structure (see epifluorescence results (Fig. 5)) (30,32). We further explored the process of peptide permeation into LPS, which is still controversial (23, 36 -38). The initial binding of the hydrophilic face of the diastereomer to the LPS molecules probably involves the displacement of the Mg 2ϩ ions (SPR results), followed by increased mobility and disordered packing of the polysaccharide region and acyl chains in LPS. Because the diastereomer molecules are significantly larger than the Mg 2ϩ ions that they replace, their insertion into the outer LPS leaflet (Fig. 7) would cause its expansion (FTIR results) as well as "loosening" and depolarization. This might be the reason for the micellization of the LPS vesicles (Fig. 6D, TEM images). This type of activity, which results in transient "cracks" in the outer membrane, permits the passage of a variety of molecules, including the uptake of the peptide itself. This process, which is referred to as the "self-promoted uptake" pathway (79), may also be explained by the "carpet-like" model of action proposed for many antimicrobial peptides (7, 11, 22, 41, 80 -83). This mechanism was recently visualized for the action of the antimicrobial peptide PGLa on intact bacteria by using atomic force microscopy, showing a loosening of the bacterial outer membrane and the formation of micelles originating from disrupted LPS (84).
In summary, this study shed light on several steps regarding the mechanism by which LPS protects Gram-negative bacteria from the action of potent membrane-active antimicrobial peptides (Fig. 7). The investigation of two peptides with the same amino acid composition but with different structures and hydrophobicities enabled better understanding of the specific roles of the helicity, hydrophobicity, and aggregation states of the peptide in LPS binding and permeation. Binding of the peptides to LPS triggers their oligomerization, consequently forming a bulky compound unable to insert into the target's cytoplasmic membrane (Fig. 7B). Importantly, the incorporation of D-amino acids can overcome this mechanism by preventing the formation of oligomers and therefore allows the peptide to traverse through the lipid core of LPS (Fig. 7C). An interesting example suggesting that native antimicrobial peptides use a similar mechanism to overcome resistance by LPS is presented by bombinins H2 and H4, 20-residue naturally occurring cationic antimicrobial peptides (85). They differ from each other only by the D-configuration of the second amino acid in H4, as a consequence of post-translational modifications (85). Although the reason is not yet known, bombinin H4, similar to D-K 5 L 7 , is more potent than bombinin H2 toward Gram-negative bacteria (86).
Overall, in addition to increasing our understanding of the molecular basis of the protection of bacteria by LPS, this study presents a potential strategy to overcome resistance by LPS, and it should help in the design of antimicrobial peptides for future therapeutic purposes.