Origin of Low Mammalian Cell Toxicity in a Class of Highly Active Antimicrobial Amphipathic Helical Peptides*

We recently described a novel antimicrobial peptide, RTA3, derived from the commensal organism Streptococcus mitis, with strong anti-Gram-negative activity, low salt sensitivity, and minimal mammalian cell toxicity in vitro and in vivo. This peptide conforms to the positively charged, amphipathic helical peptide motif, but has a positively charged amino acid (Arg-5) on the nonpolar face of the helical structure that is induced upon membrane binding. We surmised that disruption of the hydrophobic face with a positively charged residue plays a role in minimizing eukaryotic cell toxicity, and we tested this using a mutant with an R5L substitution. The greatly enhanced toxicity in the mutant peptide correlated with its ability to bind and adopt helical conformations upon interacting with neutral membranes; the wild type peptide RTA3 did not bind to neutral membranes (binding constant reduced by at least 1000-fold). Spectroscopic analysis indicates that disruption of the hydrophobic face of the parent peptide is accommodated in negatively charged membranes without partial peptide unfolding. These observations apply generally to amphipathic helical peptides of this class as we obtained similar results with a peptide and mutant pair (Chen, Y., Mant, C. T., Farmer, S. W., Hancock, R. E., Vasil, M. L., and Hodges, R. S. (2005) J. Biol. Chem. 280, 12316–12329) having similar structural properties. In contrast to previous interpretations, we demonstrate that these peptides simply do not bind well to membranes (like those of eukaryotes) with exclusively neutral lipids in their external bilayer leaflet. We highlight a significant role for tryptophan in promoting binding of amphipathic helical peptides to neutral bilayers, augmenting the arsenal of strategies to reduce mammalian toxicity in antimicrobial peptides.

During the last 15-20 years, the growing problem of resistance to classical antibiotics has focused attention on novel classes of antimicrobial molecules (1)(2)(3). One of these is the broad group of antimicrobial peptides that constitute the first line of defense against invading organisms in higher animals (4 -8). Two general features of these peptides are that they are amphipathic (adopting conformations that separate polar and nonpolar surface to match the polar-nonpolar interfacial regions of cell membranes) and are positively charged (promoting interaction with the negatively charged membranes of prokaryotic cells). Antimicrobial peptides are effective at low micromolar concentrations against a broad range of microorganisms, including in many cases those resistant to traditional antibiotics (4 -8). In general, however, therapeutic applications of these peptides have been hindered by several problems, perhaps the most important being toxicity, cost of production, and bioavailability. Continuing advances in peptide synthesis and purification, combined with economies of scale in large volume syntheses, indicate that the cost of therapeutics based on peptides of relevant size (12-20 amino acid residues) is less of an issue than in the past (9). Bioavailability problems are likely to be overcome with peptides showing genuine promise (for example by using D-amino acid versions of effective peptides (10)). Toxicity, however, remains a limiting factor in peptide antimicrobial use. This is illustrated by the observation that antimicrobial peptides are rarely effective in animal studies at doses below ϳ10 -20 mg kg Ϫ1 . Because peptides of this class so far cannot be administered orally, this therapeutic dose requires injections of significant volumes of peptide at concentrations of 1 or 2 mg ml Ϫ1 , corresponding to concentrations in the millimolar range. Thus, although the effective dose as determined by MICs 2 may be in the low micromolar range, the dosing methods require that peptides have low eukaryotic cell toxicity at rather high concentrations. Partly as a result of these considerations, first generation antimicrobials based on peptides derived from animal or bacterial sources have been limited to topical use (e.g. pexaganin based on magainin from frog skin (11)) or are chemically modified to reduce in vivo toxicity (colistin methanosulfonate in which the active form of the peptide is probably the unmethanosulfonated form resulting from loss of side chain protection in vivo (12)).
A promising approach to the development of safer peptide antimicrobials is the identification of peptides secreted internally in animals (13)(14)(15) or from commensal organisms in animals. In these cases the natural evolutionary systems will be expected to have generated effective sequences with low toxicity. Although the direct therapeutic use of endogenous peptides might itself be problematic because of the potential promotion of resistance against parts of the endogenous immune system (16), the development of peptide antimicrobials incorporating structural themes from endogenous peptides is expected to be useful.
We have recently described such a peptide, RTA3, based on the identification of antimicrobial activity in the commensal organism, Streptococcus mitis. 3 This peptide (see Fig. 1) was based on a gene sequence discovered while investigating anti-Pseudomonas activity in sputa from cystic fibrosis patients. Although the putative parent peptide, upon synthesis, was highly salt-sensitive, a modification to RTA3 based on systematic alanine scan mutagenesis and sequence changes based on rational design principles proved to retain high anti-Pseudomonas activity, low salt sensitivity, high activity against multidrugresistant Gram-negative bacterial strains, and extremely low eukaryotic cell toxicity.
The low toxicity of RTA3 has prompted further investigation, because understanding its molecular basis will enhance the prospects for developments in the antimicrobial peptide field. We surmised that the very low toxicity might be related to the observation that the amphipathic nature of RTA3 is disrupted by a charged and polar residue (Arg-5) that falls on the nonpolar face of the amphipathic helix (see Fig. 1), which is formed when the peptide encounters the target cell membrane. Note that although residue 1 is also an Arg, and appears to fall on the nonpolar face, this residue, in a membrane surface-localized helical peptide (17), can re-position itself in the polar regions of the membrane interface by helix "fraying" that is often encountered in the interaction of interfacially localized helical peptide with membranes (18,19). To address the role of the Arg-5 residue and the effects of disrupting the nonpolar face of the helical peptide, we have prepared tryptophan-substituted versions of RTA3 (RTA3-F4W) and a variant (RTA3-F4W,R5L) in which the strict amphipathic helical amino acid distribution is recovered, and we tested the membrane binding and biological activities of this peptide (the F4W substitution was introduced to facilitate membrane binding studies). To determine whether the observations on this peptide pair are generally applicable, we have also synthesized a previously described (20) pair of peptides, V 681 (an antimicrobial amphipathic helical peptide with high eukaryotic cell toxicity), and a mutant, V 681 -V13K ( Fig. 1), that had suppressed eukaryotic cell toxicity upon disruption of the nonpolar helix face, and we tested their interactions with membranes of varying composition. In contrast to previous explanations involving the subtle dispositions of the peptides in membranes of different compositions, we find that disruption of the hydrophobic face of the amphipathic helical peptides in wild type RTA3 and the mutant V 681 -V13K reduces eukaryotic cell toxicity because of a greatly reduced affinity for neutral membranes, which is particularly marked for RTA3. Surprisingly, each of the peptides with a disrupted hydrophobic surface appears to bind to negatively charged membranes without major disruption of helical conformations to move the positively charged side chain (Arg or Lys) from its apparent loca-tion on the nonpolar surface directed toward the membrane interior. The results also highlight a significant role for tryptophan in promoting helical peptide binding to neutral phospholipid membranes, and thus promoting eukaryotic cell toxicity.

EXPERIMENTAL PROCEDURES
Peptide Synthesis, Purification, and Characterization-The peptides listed in Table 1 were synthesized by Dr. G. Bloomberg of the Bristol Centre for Molecular Recognition using standard Fmoc (N-(9-fluorenyl)methoxycarbonyl) solid phase synthesis. The peptides were purified by high pressure liquid chromatography and were confirmed to be at least 97% pure by analytical high pressure liquid chromatography and to have the predicted m/e ratio by mass spectrometry. Phospholipids were from Lipid Products (Nutfield, UK); carboxyfluorescein (CF) was from Sigma, and fluorescein-phosphatidylethanolamine (FPE) was from Avanti Polar Lipids.
Biological Activities-MICs of the peptides were determined by broth microdilution according to the Clinical and Laboratory Standards Institute (21). 100 l of 0.5-1 ϫ 10 6 colonyforming units per ml of Pseudomonas aeruginosa in Mueller Hilton media (plus cations) broth (BD Biosciences) were incubated in 96-well microtiter plates with serial 2-fold dilutions of the peptides. MICs were defined as the lowest peptide concentration with no visible growth of bacteria from the MIC microtiter plates after 24 h at 37°C. Hemolytic activity was determined by incubating 10% (v/v) suspension of horse erythrocytes with peptides. Erythrocytes were rinsed in 10 mM phosphate-buffered saline, pH 7.2, by repeated centrifugation and resuspension (3 min at 3000 ϫ g). Erythrocytes were incubated at room temperature for 1 h in either deionized water (fully hemolyzed control), phosphate-buffered saline, or with peptide in phosphate-buffered saline. After centrifugation at 10,000 ϫ g for 5 min, the supernatant was separated from the pellet and the absorbance measured at 570 nm. Absorbance of the suspension treated with deionized water defined complete hemolysis.
Preparation of Lipid Vesicles-All experiments were performed at room temperature. Small unilamellar vesicles (100 nm diameter) were used for all spectroscopic measurements except for CD spectroscopy for which smaller (50 nm) vesicles were used to minimize light scattering effects. Lipids were dried from chloroform:methanol solution and pumped under high vacuum overnight to remove traces of solvent. Dried lipids were hydrated at a concentration of 10 mg ml Ϫ1 in 10 mM Tris-HCl, pH 7.4, containing either 107 mM NaCl (buffer A) or, for the CF dye-release experiments, 50 mM CF. Vesicles doped with FPE were prepared similarly except that 0.5 mol % of FPE in methanol was added to the lipids in organic solvent before drying. Hydrated lipids were extruded 10 times through two 100 or 50 nm pore membranes, using a Lipex Biomolecular extruder (Vancouver, Canada). Vesicles for CD and peptide binding, monitored using either tryptophan fluorescence or FPE fluorescence, were used directly. Vesicles for CF dye-release measurements were used after gel filtration on a Sephadex G-15 column with buffer A as the mobile phase, to remove nontrapped CF. Thus in all experiments, interaction of the peptide with vesicles was determined in the same buffer (buffer A).
Fluorescence Spectroscopy-Fluorescence measurements were made using a SPEX Fluoromax fluorimeter. Peptide solutions were made in plastic tubes or cuvettes to minimize loss of peptide at low concentrations because of binding to glass surfaces. For the measurement of vesicle-induced changes in the emission spectra of tryptophan in Trp-containing peptides, a 2 M peptide solution was incubated in buffer A, and aliquots of vesicles suspension were added to give total lipid concentrations in the range 0 to 300 M total lipid. Tryptophan fluorescence was excited at 280 nm, and the emission spectrum was measured between 300 and 450 nm in 1-nm increments with 1-s signal averaging. Binding data were fitted to a simple hyperbolic function to obtain estimates of the maximum fluorescence emission blue shift (⌬ max ) and the concentration of lipid at which the lipid-induced blue shift was half-maximal.
Peptide binding to FPE-labeled vesicles was measured by adding aliquots of peptide to a suspension of vesicles at 65 M total lipid concentration in buffer A. FPE emission was measured at 520 nm (excitation at 490 nm). The experiments were made by adding successive aliquots of peptide to a single vesicle sample. Control experiments showed that the same FPE fluorescence enhancement was obtained by adding a single large aliquot of peptide or the same amount of peptide in successive small aliquots; the latter method facilitates analysis of peptide binding at low peptide concentrations allowing an assessment of the extent to which binding is cooperative (22).
Peptide-induced dye release from vesicles loaded with CF was measured from the loss of CF self-quenching as the dye dilutes into the extravesicular medium. Experiments were done with the same lipid concentration (65 M) as the FPE binding measurements and in buffer A so that data from the different experiments can be interpreted in a consistent manner. CF emission was measured at 520 nm (excitation at 490 nm). The fluorescence resulting from 100% release of encapsulated CF was determined by adding 10 l of 20% Triton X-100. To ensure rapid mixing of peptide and vesicles, and to avoid high local concentrations of peptide, 1 ml of a peptide solution at double the post-mix concentration was rapidly ejected from an Eppendorf pipette into 1 ml of a vesicle suspension at 130 M concentration (65 M post mix) to initiate binding and dye release. The fluorescence emission intensity was measured 3 min after mixing CF-loaded vesicles with peptide.
Circular Dichroism Spectroscopy-CD spectra were obtained at 20°C using a Jobin-Yvon CD6 spectropolarimeter. All samples were made in buffer A. Spectra of peptides in solution were measured in 1-or 2-mm quartz cuvettes. Spectra in the presence of vesicles were measured in 0.1-mm path length cuvettes to minimize light scattering contributions. All spectra are averages of 5 (vesicle-free solutions) or 9 -11 scans (peptides plus vesicles) with appropriate peptide-free blank spectra subtracted and were zeroed at 260 nm before plotting without smoothing. Peptide helix content was calculated from the ellipticity at 222 nm ( 222 ) (23) using parameters determined by Luo and Baldwin (24).

Peptide Nomenclature and Biological Activities of Peptides
The experiments described below were designed to compare the membrane binding properties of two sets of peptides, each set comprising a peptide(s) having the nonpolar helix face disrupted by a positively charged Arg or Lys residue (RTA3-dis; WRTA3-dis and V 681 -dis), and the other of the set having a nondisrupted helix face (WRTA3-non-dis and V 681 -non-dis). We have employed the "dis"/"non-dis" designation here to simplify the terminology of these peptides, particularly because the "wild type" RTA3 peptide (RTA3-dis) has a disrupted (dis) nonpolar helix face, whereas it is the "mutant" V 681 peptide (V 681 -V13K or V 681 -dis) that has the disrupted (dis) nonpolar helix face. We synthesized F4W variants of the RTA peptides (WRTA3-dis and WRTA3-non-dis), in which the Phe-4 residue was replaced with a tryptophan, for two reasons. First, to obtain accurate peptide concentrations that are important for quantitative interpretation of CD data. Second, the presence of a tryptophan residue allows quantitative analysis of peptide binding to lipid vesicles through the effects of binding on tryptophan fluorescence. V 681 already contains a tryptophan (Trp-2) in its "native" sequence ( Fig. 1).
The minimum inhibitory concentrations against P. aeruginosa and the hemolytic activities of the peptides are given in  (20), are listed with mutated residues italicized. We employ a simplified peptide nomenclature in which peptides having the nonpolar helix face disrupted with an Arg or Lys residue are designated dis (e.g. RTA3-dis) and those with intact nonpolar helix faces are designated non-dis. These designations are given for each peptide in parentheses in the figure. Table 1 and Fig. 2. We present hemolysis data up to very high concentrations that are relevant for in vivo dosing considerations as described in the Introduction.
Peptides with disrupted nonpolar helical faces have greatly reduced hemolytic activity while retaining anti-Pseudomonas activity similar to that of the equivalent peptide having a nondisrupted nonpolar helix face. The latter peptides have very high hemolytic activity. These observations are generally consistent with the results on V 681 and V 681 -V13K previously described by Chen et al. (20). Two important observations that we explore below are the extremely low hemolytic activity of wild type RTA3 at very high concentrations, and the significantly enhanced hemolytic activity of the analog of RTA3 in which the Phe-4 residue is replaced with a tryptophan (Table 1; Fig. 2).

Conformational Transitions upon Membrane Binding
Each of the peptides in Fig. 1 is unstructured in buffer A as indicated by a "random coil" CD spectrum (not shown). Fig.  3A shows that both WRTA3-dis and WRTA3-non-dis adopt helical structure upon binding to membrane vesicles composed of 50% PC and 50% PG in buffer A. The helical content was around 78% in each case, which is very similar to the maximum helical content induced by these peptides in 40% trifluoroethanol (in buffer A; data not shown), which generally induces a maximum helix content in peptides with moderate to high helix-forming potential (24). The high concentrations of peptide (150 M) and lipid (10 mM) required for adequate spectroscopic CD data are expected to promote peptide binding to lipid, and for the strongly binding peptides the CD signal should be representative of the conformation of the peptide in its membrane-bound state. The equivalence of the helix content in membrane-bound forms of WRTA3-dis and WRTA3-non-dis indicates that the presence of an Arg residue on the hydrophobic face of the otherwise amphipathic helix does not detectably disrupt helical conformations when WRTA3-dis binds to a negatively charged membrane (50% PC:PG).
WRTA3-dis adopts minimal helical structure (around 10%) when incubated at very high concentrations with vesicles composed exclusively of PC (Fig. 3A). However, if the Arg-5 residue in WRTA3-dis is replaced with a leucine (in WRTA3-non-dis), a helical conformation (82%) is induced, presumably as a result of membrane binding, that is essentially equivalent to that induced by interaction of both WRTA3-dis and WRTA3-non-dis with negatively charged membrane vesicles (50% PC, 50% PG). The small but measurable helical content of WRTA3-dis is interesting in view of the complete absence of helical content in RTA3-dis (wild   Low Toxicity in the Antimicrobial Peptide RTA3 type RTA3) when incubated with high concentrations of vesicles composed of neutral lipid (Fig. 3A). This observation indicates that the apparently conservative replacement of the Phe-4 residue of RTA3-dis with a Trp results in measurable differences in the interaction of RTA3-dis and WRTA3dis with neutral membranes.
Similar observations were made with the wild typeV 681 -nondis and V 681 -dis mutant (V 681 -V13K) that has the disrupted nonpolar helix face (Fig. 3B). Each of the peptides has a high helical content induced upon incubation with negatively charged membranes, but only the V 681 -non-dis wild type peptide, having an intact nonpolar helix face (Fig. 1), has a high helix content induced upon interaction with membranes composed of neutral lipids. The V 681 -dis peptide has slightly reduced helical content when bound to negatively charged membranes (68%) compared with the wild type V 681 -non-dis (81%) indicating that helical conformations are somewhat disrupted upon binding at the membrane interface. However, because 81% helix corresponds to around 21 amino acid residues in V 681 and 68% corresponds to 17-18 amino acid residues, the V13K mutation does not result in large scale unfolding of helical conformations to release the Lys-13 residue from the nonpolar helix face.

Membrane Binding
Trp Fluorescence-The sample restrictions in CD spectroscopy require that conformational analyses that relate to mem-brane binding can only be done under conditions of very high lipid and peptide concentrations. In addition, the possibility that the absence or limited amount of helical conformations in the presence of lipid vesicles results from binding without structure formation cannot be formally ruled out from CD data alone. We therefore performed two separate series of experiments to assess peptide binding directly, and under conditions that more closely match the peptide concentrations used in antimicrobial assays. These experiments allow an assessment of relative membrane binding affinities of the peptides, and additionally can establish whether binding is cooperative or exhibits "hyperbolic" behavior. These are each of interest with respect to the mechanisms of the membrane properties of the peptides. Because RTA3 does not contain a Trp residue, we used F4W "mutants" to assess binding to phospholipid vesicles based on binding-induced perturbation of Trp fluorescence.
The sequestering of the Trp indole group in an environment of reduced polarity upon membrane binding results in a blue shift in the fluorescence excitation maximum, as illustrated by the emission spectrum of the Trp-2 residue of V 681 -dis when titrated with increasing concentrations of lipid in the form of vesicles composed of 50% PC:PG (Fig. 4A). Titrating the same peptide with neutral lipid vesicles (100% PC) results in very small lipid-dependent blue shifts (Fig. 4B). Fig. 5 illustrates the lipid concentration-dependent fluorescence blue shifts of each of the tryptophan-containing peptides. Very similar behavior was observed for the RTA3 (Fig. 5A) and V 681 (Fig. 5B) series of peptides. In each case disruption of the nonpolar helix face (with an Arg (WRTA3-dis) or Lys (V 681 -dis)) had only a minor effect on peptide binding to negatively charged vesicles, whereas binding to neutral vesicles was greatly suppressed. An intact nonpolar helix face (WRTA3-non-dis, V 681 -nondis) resulted in strong binding to neutral lipid vesicles, although the reduced maximum blue shift upon saturation indicates that the tryptophan indole in each peptide may adopt a shallower membrane insertion compared with the binding to negatively charged vesicles. Likewise the reduced Trp-4 blue shift of WRTA3-dis, compared with WRTA3non-dis, on binding to negatively charged membranes may also indicate a location of Trp-4 closer to the membrane surface for the former peptide compared with the latter.
Fluorescein-Phosphatidylethanolamine (FPE) Fluorescence-The very small structure-formation in WRTA3-dis (Fig. 3A) and V 681 -dis (Fig. 3B) upon incubation with vesicles composed of neutral lipids, and the very small perturbation of Trp fluorescence induced by neutral lipid vesicles (Figs. 5, A and B) support the conclusion that these peptides bind very poorly to membranes lacking a negative surface charge. However these experiments do not formally rule out the possibility that the peptides might bind with neither structure formation nor burial of the respective Trp res-  idues in an apolar environment. In addition, the tryptophan fluorescence data do not address the difference in structuring of RTA3-dis and WRTA3-dis upon incubation with neutral vesicles (Fig. 3A). To address these points we made a series of experiments in which vesicles "doped" with a small concentration (0.5 mol %) of FPE were used to assess membrane binding upon titrating with increasing concentrations of peptide. FPE is very sensitive to the surface charge of the membrane, through charge effects on the pK a of the carboxylate of the fluorescein moiety localized in the head group region of the bilayer. For reduction of the negative membrane surface charge resulting from positively charged peptide binding to negatively charged vesicles (50% PC, 50% PG) or enhancement of the positive surface charge by positively charged peptide binding to neutral membranes, each results in small pK a shifts promoting deprotonation of the fluorescein carboxylate and fluorescence enhancement (25). These effects do not require burial of any region of the peptide into the membrane and can therefore assess binding in any form, even if this involves superficial interaction of positively charged side chains with negatively charged membrane lipids. This technique is particularly suited to the highly positively charged antimicrobial peptides, and it is a useful way of assessing cooperativity in peptide binding under conditions that do not involve extremely high peptide:lipid ratios at the initial parts (low lipid concentration) of the binding curves determined by titrating a fixed concentration of peptide with increasing lipid concentrations. Fig. 6, A and B, illustrates the fluorescence enhancement of FPE resulting from titrating negatively charged vesicles or neutral vesicles, respectively, with increasing concentrations of RTA3 peptides. All of the peptides bound with high affinity to negatively charged vesicles, with the binding data for RTA3-dis and WRTA3-dis being virtually superimposed. WRTA3-non-dis binds marginally less strongly and saturates at a maximum fluorescence enhancement that is around 15% smaller than that of RTA3-dis and WRTA3-dis. Each of these is consistent with the reduced positive charge of the latter peptide (ϩ5) compared with RTA3-dis and WRTA3-dis (ϩ6), because enhanced positive charge should both promote binding to negatively charged vesicles, and will make a stronger contribution to reduction of the negative surface charge density of vesicles composed of 50% negatively charged lipids.
Consistent with the tryptophan fluorescence binding data, neutral vesicles strongly distinguish between peptides having intact and disrupted nonpolar helix faces (Fig. 6B). The binding of WRTA3-non-dis to PC vesicles is nearly as strong as binding to PC:PG vesicles, whereas WRTA3-dis binds very weakly. Reverting WRTA3-dis back to the wild type peptide (RTA3dis) virtually abolishes binding to neutral vesicles. In fact, the wild type peptide with a perturbed nonpolar helix face and lacking a tryptophan residue has a binding constant for neutral membranes reduced more than 1000-fold compared with its interaction with negatively charged membranes. This observation explains the difference in structuring between RTA3-dis and WRTA3-dis on incubating with very high concentrations of neutral vesicles (Fig. 3A).
The FPE vesicle binding data for the V 681 peptides (Fig. 7) are generally consistent with those described for the RTA3 peptides (Fig. 6) and for V 681 peptides binding measured using tryptophan fluorescence (Fig. 5B). We were only able to determine binding of wild type V 681 (V 681 -non-dis) to PC:PG vesicles at moderate peptide concentrations because interaction of the peptide with these vesicles at higher concentrations induced light scattering because of peptide-induced vesicle aggregation or fusion. However, the data demonstrate rather similar binding of V 681 -non-dis and V 681 -dis to negatively charged vesicles (Fig. 6A). Likewise the FPE binding data confirm that the minimal structuring of V 681 -dis upon incubation with neutral lipid vesicles (Fig. 3B) and the limited effect of neutral vesicles on the fluorescence emission of the Trp2 residue (Fig. 5B) is because of low binding affinity of this peptide to neutral membranes.  For all of the peptides studied, no evidence of cooperative binding (either positive or negative cooperativity) was apparent, because in each case linear variations in fluorescence enhancement with increasing peptide concentrations were observed at low peptide concentrations.

Membrane Perturbation, Vesicle Dye Release
How do the membrane binding properties of the peptides relate to their effects on the structural integrity of phospholipid membranes? We tested the ability of the peptides to release CF trapped at high concentrations within 100 nm SUV composed of either 50% PC:PG or 100% PC under conditions (65 M total lipid, high salt buffer) equivalent to those used for the FPE binding experiments of Fig. 7. All of the data are compiled in Fig. 8.
Each of the five peptides releases internal CF from negatively charged vesicles at low peptide concentrations, either through pore formation or generalized disruption of the lipid membrane bilayer organization ( Fig. 8A; notice that the designation of the peptides is different in Fig. 8, with open and closed symbols referring to RTA3 and V 681 peptides, respectively). The membrane lytic activity generally relates to the positive charge density of the peptides with the highly charged (5 or 6 positive charges per 16 amino acids) RTA3 peptides being particularly active with half-maximal dye release occurring at concentrations near 0.1 M. The less highly charged V 681 peptides were half-maximally active at concentrations of around 0.3 M (V 681 -dis; 7 positive charges per 26 residues) and 0.5 M (V 681non-dis; 6 positive charges), the relative activity again corresponding to the positive charge density. The relationship between activity and peptide concentration was sigmoidal in negatively charged vesicles with apparent "cooperativity" (see Table 2 and below) in the range of 1.8 -2.3.
Peptides having nondisrupted nonpolar helix faces (WRTA3-non-dis and V 681 -non-dis) had very high activity against neutral lipid vesicles with half-maximal activities in the range 0.3-0.5 M (Fig. 8B). Consistent with the membrane binding data of Figs. 5-7, peptides with nonpolar helix faces disrupted with a positively charged residue had very low activity against neutral vesicles. Wild type RTA3 (RTA3-dis) was particularly ineffective against neutral vesicles with barely detectable dye release (1-2%) at a concentration of 50 M. This corresponds to a ratio of effectiveness in dye release from negatively charged vesicles over neutral vesicles of around 10,000-fold, a factor that results from a combination of the greatly reduced binding of RTA3 to neutral membranes (near 1000-fold) and the apparent cooperativity in peptide-induced membrane perturbation.

Summary of Peptide Membrane Interactions
The membrane-induced helix formation, membrane binding, and membrane perturbation data for the five peptides studies is summarized in Table 2. In all experiments, peptides having nonpolar helix surface disrupted with a positively charged Lys (V 681 -V13K) or Arg residue (RTA3, RTA3-F4W) showed limited binding to neutral lipid vesicles coupled with very low activity in disrupting vesicles composed of neutral lipids. On the other hand, these In this figure open symbols represent RTA3 peptides, and filled symbols represent V 681 peptides. A is CF-release data from PC:PG (50:50, mol/mol) vesicles, and B is dye release from 100% PC vesicles. Squares represent peptides with nondisrupted nonpolar helical faces (WRTA3-non-dis and V 681 -non-dis); circles represent peptides with disrupted nonpolar helical faces (WRTA3-dis and V 681 -dis), and triangles are data from RTA3 (RTA3-dis). Dotted lines are fits to the equation p ϭ P max c n /(a ϩ c n ), where P is the percentage of dye release; P max is the maximum dye release; c is the peptide concentration, and a is a constant. The values of n are listed in Table 2.  peptides retain strong binding to negatively charged vesicles and perturb these membranes (either via nonspecific membrane perturbation or "pore" formation) at low concentrations. This general observation is entirely consistent with the relative activities of the peptides on eukaryotic cells (erythrocyte hemolysis) and bacterial cells (MICs), respectively ( Fig. 2 and Table 1).
We have tabulated the data in Table 2 in terms of the lipid or peptide concentration required to elicit half-maximal binding or half-maximal dye release. This is appropriate for the dye-release data because the dose-response curves (Fig. 8) in PC:PG are sigmoidal with near second-order dependence on peptide concentration for each of the peptides studied ( Table 2). The binding data fitted simple hyperbolic binding isotherms in all cases, indicating that any cooperativity in peptide-induced dye release in PC:PG vesicles must arise from the properties of the membrane-bound peptides, rather than in the binding process itself.
Although the membrane binding measured by Trp fluorescence (titrating peptide with increasing vesicular lipid concentration) and using FPE-doped membranes (titrating vesicular lipid with increasing peptide concentration) probes a single binding equilibrium for any particular peptide:lipid composition pair (e.g. RTA3 binding to pure PC membranes), the FPE and Trp fluorescence data cannot be combined and fitted to a single equilibrium constant, because the peptide-membrane interaction corresponds to a partitioning rather than a discrete site-binding phenomenon (26). At the very least, one would need to know the number of lipids that constitute a "peptidebinding site" to make this analysis; however, the membrane is essentially a continuous "homogeneous" surface that cannot formally be dissected into "binding sites." On the other hand, the Trp fluorescence data can be used to determine a mole fraction partition coefficient (K x ), according to Equation 1, that quantitates the partitioning of peptide between the aqueous and membrane phases (26), and these values are useful for comparison with previous studies of peptidemembrane partitioning.
In Equation 1, ⌬ and ⌬ max are the lipid-induced shift in the peptide Trp fluorescence emission wavelength, and its maximum value, respectively; [L] is the lipid concentration, and [W] is the concentration of water (55.3 M). For the strongly partitioning peptides, K x is in the range of 2.4 ϫ 10 6 to 1.5 ϫ 10 7 , whereas for the weakly binding systems these are 2 orders of magnitude or more smaller (estimated to be less than 5 ϫ 10 4 and less than 2 ϫ 10 4 for RTA3-F4W/PC; V 681 -V13K/PC, respectively). These values can be compared, for example, to the binding of magainin to PC:PG (50:50) (K x ϭ 5.4 ϫ 10 6 , in a similar high salt buffer to that used here (22)) and of melittin binding to neutral vesicles (K x ϳ 10 5 (27)). K x , in this analysis, cannot be measured for peptides lacking a Trp residue but must be well below 10 4 for RTA3 binding to neutral membranes, because this peptide binds at least 10 times more weakly than RTA3-F4W and V 681 -V13K as measured by FPE fluorescence ( Table 2).

Lipid Composition Dependence of RTA3 Peptide Activities
The binding and membrane-disrupting effects of V681 and RTA3 peptides on PC and PC:PC (50:50) membrane vesicles broadly correlate with their effects on mammalian (erythrocytes) and bacterial (P. aeruginosa) cells, respectively. To determine whether these observations apply to other lipid compositions, we measured RTA3-induced membrane binding and trapped dye release using 100 nm SUV having lipid compositions more similar to bacterial and mammalian membranes. These lipid compositions were PE:PG: cardiolipin (80:10:15) as an analog of bacterial cell membranes (28), and PC:PE:sphingomyelin:cholesterol (35:30:25:10), as an analog of erythrocyte membranes (28,29). These data are compiled in Table  3, and binding and dye-release data with the erythrocyte-mimetic membrane vesicles are shown in Fig. 9. The RTA3 peptides show strong binding to membranes having lipid compositions similar to those of  bacterial membranes, although the binding is reduced by 3-5fold compared with binding to PC:PG (50:50) membranes (Table 3), probably because of the reduced negative surface charge in membranes containing 20 -25%, rather than 50%, negatively charged lipids. Likewise, the binding and membrane perturbation of RTA3 peptides containing disrupted nonpolar helix surfaces (RTA3 and RTA3-F4W) is negligible when tested against PC:PE:sphingomyelin:cholesterol (35:30:25:10). Indeed, these RTA3 peptides bind even less strongly and are less effective in perturbing the bilayer membrane integrity, when tested against vesicles having the latter lipid composition, compared with the effects on pure PC membranes. This reduced binding affinity may be due to the effect of cholesterol in reducing the fluidity of the bilayer membrane and a suppression of peptide partitioning into the membrane interfacial region (30). The replacement of Arg-5 of RTA3 with Leu-5 in RTA3-R5L restores strong binding and dye release ( Fig. 9) consistent with the observations described in Figs. 2, 3, and 5-8 and Table 2.

DISCUSSION
This study illustrates that vesicle bilayer membranes composed either of neutral lipids or mixed PG:PC membranes can be surprisingly good analogs for eukaryotic and bacterial cell membranes, respectively, consistent with a body of work that highlights membrane surface charge as a dominant feature in the selectivity of positively charged amphipathic peptides for bacterial membranes (5, 30 -33). These observations also apply to more complex lipid mixtures designed to match the lipid compositions of bacterial and mammalian cell cytoplasmic membranes. The general correspondence between the effects of the peptides on vesicles (Figs. 8 and 9; Tables 2 and 3) and on bacteria or erythrocytes ( Fig. 2 and Table 1) provides strong evidence that interaction with the membranes of target cells plays an important role in their mechanisms of antimicrobial action. A detailed analysis of vesicle binding, membrane-induced structure formation, vesicle membrane perturbation, and activity measurements against target bacteria and erythrocytes yields a consistent interpretation of the effects of disruption of the amphipathic structure of positively charged antimicrobial helical peptides by inserting a positively charged amino acid onto the nonpolar helix face. As described previously (20), this modification can result in greatly reduced eukaryotic cell toxicity while retaining high antimicrobial activity, a conclusion that is reinforced by similar observations with RTA3 and the designed sequence variants ( Fig. 2; Table 1). It was previously suggested that the reduction of activity against eukaryotic membranes upon disruption of the nonpolar helix face (with a Lys in V 681 -V13K) resulted from the inability of the peptide to insert deeply into the cell membrane to form peptide "pores" (20). Our results demonstrate that the loss of membrane disruption of neutral membranes, including those of eukaryotes, has a simpler explanation; the peptides simply do not bind to membranes without negative surface charge.
This conclusion is generally consistent with our understanding of the interaction of amphipathic helical peptide with the interfacial region of bilayer membranes. For neutral membranes, binding is dominated by the interaction of the amphipathic peptide structure with the complementary membrane interfacial region. Amphipathic helix formation is crucial for this process because it allows the polar peptide backbone to substitute intramolecular (helical) hydrogen bonding for the loss of solvation energy as the peptide buries within the interfacial region of the bilayer (34). If the amphipathic structure is perturbed, this essential contribution to binding is lost. If the peptide helix partially unwinds to remove the positively charged amino acid from the nonpolar helix face, then the amphipathic nature of the peptide that dominates the favorable binding energy is again attenuated. Thus nonamphipathic peptides cannot easily bind in the interfacial region of neutral bilayer membranes. The results shown here demonstrate that a single positively charged amino acid (Lys or Arg) on the nonpolar face of an interfacially bound amphipathic helical peptide is sufficient for very large attenuation of binding to neutral phospholipid bilayer membranes, a result that is consistent with the very high energetic barriers to the burial of a positively charged residue in the nonpolar regions of a membrane (35).
On the other hand, binding to negatively charged membranes has contributions both from the complementary nature of helical amphipathic peptides and the interfacial region of the bilayer, and complementary electrostatics. As indicated by the CD data of Fig.  3, the negatively charged membrane surface may also provide complementary negative charges for the positively charged amino acids disrupting the nonpolar helix face. This seems to be required to explain the observation of the retention of virtually unperturbed helical content of RTA3 and RTA3-F4W, and the relatively small perturbation of helical structure inV 681 -V13K, on binding to negatively charged membranes. Observations of RTA3-induced vesicle fusion, together with molecular dynamics simulations, suggest that the accommodation of the Arg residue on the nonpolar helix surface of RTA3 may be achieved by tilting of the peptide helix at the membrane interface that allows the Arg-5 side chain to reach the well solvated regions of the interface. 4 However, the accommodation of the Lys-13 residue on the nonpolar helix face of V 681 -V13K upon binding to negatively charged membranes is unlikely to be achieved by helix tilting because the positively charged residue lies near the center of the relatively long peptide helix, and an unrealistically high degree of helix tilt would be required for the Lys residue to "reach" the highly hydrated regions of the membrane interface. In the case of V 681 -V13K, lipid head group phosphates might provide "neutralizing" charges for the positive Lys-13 charge.
An unexpected finding is the observation that substitution of Phe-4 of RTA3 with Trp significantly enhances binding to neutral membranes. In hindsight, this is understandable in terms of the very high interfacial propensity of Trp compared with all the other amino acids, even though Phe is a more "hydrophobic" amino acid side chain (36). This observation indicates that the removal of Trp residues from amphipathic helical antimicrobial peptides might be an additional general strategy for reducing eukaryotic cell toxicity. The absence of a tryptophan, combined with the presence of an arginine residue on the non-polar helix face, results in a peptide with extraordinarily low binding to neutral lipid membranes, and this seems to underlie the extremely low eukaryotic cell toxicity in RTA3, despite the retention of very high affinity for binding to negatively charged membranes (Table 2) and strong antimicrobial activity (Table  1). These specific sequence characteristics are a product of evolutionary design within a commensal organism (S. mitis) and highlight the potential of commensals as a source of novel antimicrobials having low eukaryotic cell toxicity.