Peptide–membrane interactions are essential for many processes of physiological or pathological relevance, such as the effects of hormones or toxins on host cells or the effects of host defence peptides on microbial cells [1–3]. The mode of action of cytotoxic peptides can indicate new design strategies for the development of therapeutic agents, such as immunotoxins [4], antitumour drugs [5] or anti-infective agents capable of overcoming the currently serious problem of resistance to conventional antibiotics [6]. For AMPs (antimicrobial peptides), activity depends principally on membrane permeabilization or lysis, and several mechanisms of action have been proposed, ranging from the formation of highly organized transmembrane pores, to more dynamic supramolecular peptide–lipid complexes leading to toroidal or wormhole-type pores, to a relatively disorganized, detergent-like membrane disaggregation due to hydrophobic and/or electrostatic interactions with the membrane surface [7–11]. However, the molecular details underlying these processes, which have important implications for the selectivity of the peptides' activity, are still not entirely understood.

One important class of membrane-interacting peptides assume an amphipathic α-helical conformation that permits the insertion of a well-defined hydrophobic sector into the lipid bilayer [8]. We have undertaken a systematic study of how the structural and physical characteristics of these peptides influence the mode and selectivity of their membranolytic action by rationally and independently varying properties such as charge, hydrophobicity, amphipathicity and helix-forming propensity [8,12–14]. To do this, we have used a self-consistent set of peptides of fixed size, whose design is based on a sequence template obtained from the comparative analysis of the sequences from numerous natural antimicrobial and/or cytotoxic peptides of different origins [8,13]. Thus these peptides are based on a scaffold that has been selected by evolution in many different organisms, whereas the precise sequence has been rationally adapted to probe the role of specific aspects of the molecules' shape and stability. Previous studies have indicated that the helix-forming propensity is a key factor in determining selectivity, i.e. the differential cytotoxicity towards bacterial or host cell membranes, whereas the properties of the hydrophobic and polar sector can influence both host cell cytotoxicity and antimicrobial potency [13,15].

The roles of helical structuring and hydrophobic face properties in determining membrane interaction are intimately related. Helix formation permits an optimal spatial arrangement of aliphatic side chains for membrane insertion. The strong hydrophobic interaction of these side chains with the lipid layer, in turn, stabilizes the helical conformation, reducing main-chain hydrophobicity and thus allowing a deeper insertion into the bilayer. To probe these interrelationships, we have designed a new set of peptides in which the size, cationicity and the amphipathic residue arrangement were not varied, whereas (i) the size of aliphatic side chains for residues forming the hydrophobic surface of the helix was systematically varied to determine the minimum requirements for interaction with different types of cells, leading to membrane lysis; (ii) the shape of the hydrophobic surface was modulated to test whether an oblique hydrophobic gradient could affect membrane insertion and selectivity; (iii) the size of side chains for residues forming the polar surface of the helix was varied to determine how the degree of insertion of the peptide via the so-called snorkel effect affected activity; and (iv) the helix-forming propensity was independently varied to probe the role of helix stability in modulating selectivity. Although the literature on structure–activity relationships in α-helical cell-lytic peptides is vast [8,16–20], our method, which is based on the rational and systematic modulation of macroscopic structural characteristics on a template originating from a large number of cell-lytic natural peptides, is unique.

The capacity of the peptides to compromise microbial cells was assessed in terms of MIC (minimum inhibitory concentration) values towards a number of laboratory microbial strains and of the kinetics for bacterial killing and membrane permeabilization in reference Gram-positive and Gram-negative microorganisms. Flow cytometric techniques were used to correlate permeabilization of circulating blood cells with morphological changes, which could also be visualized by SEM (scanning electron microscopy). Furthermore, a plasmon resonance technique was used to assess the mode of interaction with both anionic- and zwitterionic-model lipid bilayers. Finally, the activity of the most active peptides was assessed for a wide panel of indicator microbial strains, including several Gram-positive, Gram-negative and fungal clinical isolates, some of which are multi-resistant to conventional antibiotics. The wide-spectrum and potent activity of some of the peptides indicate that they may be suitable lead compounds for the development of novel anti-infective agents.

Table 1 Sequences of synthesized model peptides, based on a template derived from natural α-helical AMPs

Figure 1 Helical wheel projections for model peptides with altered hydrophobic face

Figure 2 Helical wheel projections for model peptides with altered polar face

The amino acid hydrophobicity index scale used for determining the mean hydrophobicity and amphipathicity per residue of these peptides (Table 2) was developed ad hoc as a normalized and filtered consensus of 163 published scales [21], and is arbitrarily ranged between values of +10 for Phe and −10 for Arg. For more details, see the Supplementary Materials and methods section at http://www.BiochemJ.org/bj/390/bj3900177add.htm.

Table 2 Correlation between peptide physicochemical properties and antimicrobial or haemolytic activity

The peptide–lipid binding event was analysed using a series of sensograms collected at six different peptide concentrations. As the system did not reach steady state during the injection of the sample, the affinity constant was determined using kinetic models suitable for peptide–lipid interactions, by curve fitting using numerical integration analysis (see the Supplementary Materials and methods section at http://www.BiochemJ.org/bj/390/bj3900177add.htm). The BIA evaluation 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. The data were treated globally by simultaneously fitting the peptide sensograms obtained at the six different concentrations. The two-state reaction model [23] was applied to each dataset corresponding to the following model:

The permeabilization of the cytoplasmic membranes of E. coli ML-35 and of Staphylococcus aureus 710A, by synthetic peptides, was evaluated by following the unmasking of cytoplasmic β-galactosidase and phosphogalactosidase activities respectively to extracellular ONPG (Sigma, St. Louis, MO, U.S.A.) or ONPG-6-phosphate, as described previously [24] and in more details in the Supplementary Materials and methods section at http://www.BiochemJ.org/bj/390/bj3900177add.htm. Unmasking of periplasmic β-lactamase activity to CENTA was used to follow the Gram-negative outer-membrane permeabilization.

The bactericidal activity was tested against E. coli ML-35 or Staph. aureus 710A in the exponential phase (10⁷ c.f.u./ml in PBS incubated at 37 °C). At different times, 50 μl of the bacteria/peptide suspension was diluted severalfold in ice-cold PBS, and the solution was plated on nutrient agar and incubated for 16–18 h to allow colony counts.

In the series P2–P6, the depth of the hydrophobic face was systematically varied by reducing the size of the aliphatic side chain from 4 to 1 carbon atoms one at a time, by replacing Nle with Nva (norvaline), Abu (2-aminobutyric acid) or Ala (see Table 1 and Figure 1). Furthermore, achiral, α-branched residues with side chains varying from 3 to 1 carbon atoms [Dpg, its cyclic analogue Acp (aminocylcopentanecarboxylic acid), Deg (diethylglycine) or Aib (2-aminoisobutyric acid)] were placed in positions 7, 10 and 13 to modulate side-chain density and helix stability [25]. Aib and Acp in particular are reported to stabilize the helical structure efficiently, whereas the capacity of Dpg and Deg to do this is less certain [25,26].

In the series P2, P7–P9 (Table 1, Figure 2), the length of the charged or polar residue side chains was varied by interchanging the cationic residues Lys, Orn, Dab (2,4-diaminobutyric acid) and Dap (2,3-diaminopropionic acid) and the neutral polar residues Gln, Hse (homoserine) and Ser while keeping the hydrophobic sector constant.

A longitudinal hydrophobicity gradient was introduced by systematically decreasing linear and branched aliphatic residue side-chain sizes along the helical axis, in both N- to C-terminal (P10) and C- to N-terminal (P11) directions (see Table 1 and Figure 1).

It should be stressed that an attempt was made to vary the parameters as far as possible independent of each other. Thus the charge and size were kept constant, as were the residue distribution (in terms of polar/hydrophobic residues), thus suggesting that some parameters that can markedly affect membrane interaction [19], such as the angle subtended by the hydrophobic and polar sectors and therefore their surface area, are also fairly constant. However, these to some extent do depend on the spread resulting from side-chain lengths and also assume complete helical structuring. In some cases, and in particular for peptides with a lower helix-forming propensity in TFE (trifluoroethanol) or SDS, a more significant deviation may occur [27]. This helix-forming propensity should however be distinguished from the helical content of the peptide when it is membrane-bound. It is well known that the membrane environment very strongly induces the helical conformation, imposing it even on diastereomeric peptides [28].

The propensity of the peptides to assume a helical structure in the presence of the helix-stabilizing solvent TFE or membrane mimicking SDS micelles, as determined by CD (Table 2, and Figure S1 at http://www.BiochemJ.org/bj/390/bj3900177add.htm), depended on both the overall properties of the hydrophobic sector and the presence of α-branched residues. A shallow hydrophobic face correlated with a reduced helix-forming propensity (P4 and P5 in Table 2), and this was also the case for the peptides with a depth gradient in this face (P10 and P11 in Table 2, and Figure S1 at http://www.BiochemJ.org/bj/390/bj3900177add.htm), i.e. where one or other of the extremities is shallow. P3 showed the most marked increase in conformational stability in all environments, due to the presence of the cyclic, α-branched Acp residues, being the only peptide to structure significantly also in aqueous solution. Aib also aided structuring, but to a lesser extent (e.g. P2 versus P1), whereas Dpg (P6) did not, in agreement with literature reports [25], probably because of steric reasons. The characteristics of the polar sector also affected structuring (Table 2), but in a manner that did not correlate with its depth, as P2 showed a reproducibly lower helix content (~60%) compared with the peptides with a deeper (P7) or shallower (P8 and P9) polar face. The reason for this is not clear. The reduced capacity of P9 to structure in the presence of SDS micelles with respect to TFE can be interpreted as being due to its reduced capacity to penetrate into the lipid layer via the snorkel effect.

A clear correlation was observed between RP-HPLC retention times and both the hydrophobic and polar face depths (P1>P2>P4>P5 and P7>P2>P8>P9). In this case, the C₁₈ layer can also be considered as a simple model for the membrane lipid environment [29], and the first series confirms the importance of hydrophobic interactions, whereas in the second series, retention times decrease with decreasing polar side-chain size, confirming the snorkel effect. In this respect, the optimized structuring and hydrophobicity of P3 resulted in the highest retention time, but the latter property seems to be the key parameter, as shown by the high retention time of P6 also, which is almost as hydrophobic, but has a lower propensity for helical structuring. In fact, this experiment exemplifies the interconnection between structuring and the effective hydrophobicity. The global hydrophobicity determines the initial interaction with the lipid layer, inducing structuring and thus the formation of a hydrophobic surface. This hydrophobic surface area then sinks into the lipid layer, in a manner that depends on the side-chain size, resulting in a further stabilization of the structure and increased interaction.

Unexpectedly, P3, which has optimized hydrophobicity, amphipathicity and helix stability, showed a significantly reduced potency. This was however found to be quite medium-dependent, so that activity compared favourably with that of the peptides P1 or P2 in reduced medium [e.g. 5% (v/v) MH broth in PBS; Table 2, values in parentheses], and in some batches of full strength MH broth, but not in others. An explanation for this is that this peptide probably aggregates and/or interacts more strongly with components of the medium, so as to reduce its activity in these assays. Synthetic peptides based on a hybrid cecropin-melittin sequence, in which a stabilized helical structure was imposed by the formation of lactam bridges between Gluⁱ and Lysⁱ⁺⁴ side chains in the polar sector, also showed reduced activity [30].

The killing kinetics for some of the peptides against reference Staph. aureus and E. coli strains is shown in Figure 3. Trends are similar for both microorganisms, although for any given peptide, the killing of the Gram-positive microorganisms was considerably slower. P3, with its optimized physicochemical properties and structural stability, was the most efficient at killing both microorganisms, under the conditions of this assay. Furthermore, the significantly faster killing kinetics of P7 with respect to peptides with a shallower polar sector indicates that insertion depth, via the snorkel effect, affects the killing mechanism. P3 and P7 retained a significant killing capacity towards E. coli also at reduced concentrations (Figure 3A).

Figure 3 Kinetics of bacterial killing for selected model peptides

The haemolytic activity of the peptides, a measure of their selectivity for microbial cells, follows the same trend (P6>P1=P2>P4>P5 and P7>P2=P8>P9) as antimicrobial activity, consistent with a membranolytic mechanism for both types of cells, but is generally quite moderate at 10 μM, a concentration at which most peptides are antimicrobially active in vitro. Thus host cell disruption is also affected by both the hydrophobic face depth and the snorkel effect. In contrast with the antimicrobial activity, however, the presence of an oblique hydrophobicity gradient (P10 and P11) seems to favour haemolytic activity. The crucial factor appears to be the conformational stability, with P3 showing a particularly high haemolytic activity already at 10 μM, confirming previously reported observations [15].

Permeabilization of mouse lymphocytes to PI was determined for P1, P3 and P7, using flow cytometric methods (Figure 4). In the absence of peptides, a homogeneous population of non-fluorescing, undamaged cells dominates. In the presence of 10 μM P1 or P7, a second population of damaged and permeable cells [high fluorescence, decreased FS (forward scattering)] becomes dominant, in line with a previous observation, probably indicating a necrotic inactivation of cells [15]. P3 acts via a different mechanism, the massive permeabilization being accompanied by increased side scatter, indicating a more granular morphology. As the highly haemolytic activity of this peptide was reminiscent of that of the α-helical bee venom toxin melittin, this was also tested under the same conditions, with an intermediate result, both decreased-FS and increased-SS (side scattering) populations being observed.

Figure 4 Three-dimensional flow cytometric contour-plot analysis of mouse lymphocytes treated with selected model peptides and the bee toxin melittin

The behaviour of P3 was studied in more detail, comparing flow cytometric and SEM results (Figure 5). At a low concentration (≤1 μM), it appears to behave like the other peptides (P1 or P7), inducing a decreased-FS population (Figures 5B and 5F), so that cells appear to shrink and become highly permeable to PI, although little apparent membrane damage is evident by SEM (Figure 5J). At a higher concentration (10 μM), only the highly permeable, increased-SS population is evident (Figures 5C and 5G), corresponding to a highly granular morphology as observed by SEM (Figure 5K). Melittin (Figures 5D, 5H and 5L) again showed an intermediate behaviour.

Figure 5 Effect of P3 on mouse lymphocytes determined by flow cytometry and SEM

The different behaviours of P3 towards both host cells and microbial cells with respect to that of the other peptides from this series, prompted us to investigate interactions with model membranes using SPR. The sensograms for P3 and P1, interacting with anionic PE/PG or zwitterionic PC/cholesterol model membranes, are compared in Figure 6, and it is immediately apparent that they behave differently. P1 binds more efficiently to PE/PG than PC/cholesterol, but also washes out more easily. P3 binds with a similar efficiency to PE/PG and PC/cholesterol but dissociates significantly more slowly than P1, with a significant portion of the peptide remaining bound to the PC/cholesterol membrane on washing, in the timescale of the experiment.

Figure 6 SPR sensograms for the binding of P1 and P3 with model membranes

The average values for the rate constants obtained after fitting an integrated rate equation for a two-state reaction model directly to the observed sensograms [23] are listed in Table 3, along with the estimated affinity constant values for the two steps (K₁ represents the binding to the surface and K₂ further insertion) implicit in this model, and lead to several significant observations. First, P3 has higher affinity than P1 towards PE/PG bilayers as a result of the first binding step (K₁=2×10⁵ and 5.5×10⁴ M⁻¹ respectively), with an approx. 20-fold faster binding (k_a1) and approx. 2-fold slower dissociation (k_d1) than P1. The binding rate constants to PE/PG for the two peptides in the second (insertion) step are instead similar (Table 3). Conversely, P3 and P1 have similar affinities towards PC/cholesterol in the first binding step (K₁~1.7×10⁵ M⁻¹), but P3 inserts into PC/cholesterol approx. 10-fold better in the second step. Thus P3, which is more active than P1 against bacterial cells under appropriate conditions, binds better to PE/PG bilayers principally due to the first surface binding process. On the other hand, the increased cytotoxic activity of P3 towards host cells may be derived from the second membrane insertion step. SPR experiments performed with melittin under the same conditions indicate that it behaves similar to P3 with regard to PC/cholesterol in the second insertion step (Table 3), although showing a higher affinity in the first binding step. Thus both SPR experiments on model membranes and flow cytometric experiments on host cells indicate that P3 behaves more like the bee venom, melittin, a model of a non-cell-selective peptide, whereas P1 behaves more like canonical helical AMPs such as magainin that are bacteria-selective [31].

Table 3 Association (ka1 and ka2) and dissociation (kd1 and kd2) rate constants in bilayers determined by SPR with numerical integration using the two-state reaction model

Figure 7 Spectrum of antimicrobial activity of selected model peptides

The initial electrostatic interaction of AMPs with the negative cell surface of bacteria is widely believed to be the first step in their mechanism of action and may not require structuring. However, the efficiency of the second step, which is the transition through the cell wall and partitioning into the membrane, depends on size, charge and the structural properties of the peptides. Here, by using a set of model peptides with rationally designed systematic variations in three key structural characteristics: hydrophobic sector depth, polar sector depth and propensity for helical structuring, while keeping other parameters such as size, charge and hydrophobic/polar residue distribution constant, we have been able to probe the interaction of helical AMPs with membranes in both model membranes and cellular systems.

Permeabilization and killing kinetics are consistently slower for Gram-positive with respect to Gram-negative bacteria (Figure 3, and Figure S2 at http://www.BiochemJ.org/bj/390/bj3900177add.htm.) [12–14], so that the thick, polar peptidoglycan layer of the former may be of greater impedance than the outer membrane of the latter, which is an anisotropic layer with a lipid core. Breaching of the outer membrane in E. coli is in fact quite rapid, as subsequent permeabilization of the cytoplasmic membrane is hardly delayed (Figure S2 at http://www.BiochemJ.org/bj/390/bj3900177add.htm.), and requires hydrophobic interaction of peptides with the lipid core as indicated by the reduced ability of P4 and P5 to effect its permeabilization (see also [33,34]).

Once the peptides reach the negatively charged cytoplasmic membrane, they lie parallel to the surface of the outer leaflet, according to what has been termed the ‘carpet model’, where they accumulate until a threshold concentration is reached [7,9,35,36]. These processes have been variously proposed to result in a purely detergent-like disruption of the membrane [9], in more well-defined membrane apertures usually described as ‘toroidal’, ‘wormhole’ or ‘dynamic peptide–lipid supramolecular’ pores [35,37,38], or from amorphous areas of increased permeability termed ‘aggregate channels’ [11]. The formation of membrane spanning classical ‘barrel stave’ pores instead is unlikely [35].

Membrane insertion stabilizes the helix throughout its length and decreases main-chain polarity by promoting more extensive H-bonding [39]. This allows a deeper penetration into the bilayer and it is then the length of side chains on the polar face, which are aliphatic except for the head-groups, that determines the final degree of insertion by what is termed the snorkel effect. We could observe how this modulates the strength of interaction even in a very simple model for the lipid layer provided by the C₁₈ phase in RP-HPLC (see above, Table 2), and how it affects antimicrobial potency. This snorkel effect furthermore seems to be a common requirement for permeabilization of both bacterial and eukaryotic membranes.

Structuring is required for, but also depends on, hydrophobic interactions with the membrane. The intrinsic propensity of a peptide for assuming a helical conformation should thus affect its capacity to permeabilize biological membranes. We have reported previously that it particularly affects host cell cytotoxicity [15] and it is confirmed here with the highly cytotoxic P3, which has a stable helical structure already in the bulk phase. At the other extreme, previously reported diastereomeric peptides, with a low helix-forming propensity, maintained a reasonable antimicrobial activity associated with a remarkably low cytotoxicity [32,40].

SPR studies on model membranes, comparing P3 with P1, which has quite similar physicochemical properties but is unstructured in the bulk phase (Table 2), indicated that the initial interaction of P3 with the anionic PE/PG membrane (bacterial model) is much faster than P1, while subsequent insertion into the membrane is comparable. An explanation may be that being prestructured and probably aggregated in solution, P3 displays an increased charge/density ratio and is thus subject to a stronger electrostatic interaction with the anionic membrane surface (step 1). Membrane insertion leads to structuring of P1 and probably results in the disaggregation of P3 [41] so that they behave similarly in the second step. Conversely, the initial interaction with a model eukaryotic membrane (PC/cholesterol), which is electrically neutral, does not markedly differ for the two peptides, irrespective of their structural characteristics, whereas there is a considerable difference in the subsequent insertion step. The preformed P3 helical system binds to PC/cholesterol over 10-fold better than P1, probably because it inserts deeper into the inner leaflet. In contrast, there is only a 2-fold difference in the binding of P3 and P1 to negatively charged PE/PG membranes, indicating that the inner leaflet contributes only slightly to their binding to PE/PG bilayers. Furthermore, as the presence of an oblique hydrophobicity gradient [42] favours haemolytic activity but not bacterial inactivation, it is possible that it also affects the second step (insertion) in eukaryotic membranes. Overall, the tendency to prestructure/aggregate in solution leads to a shift in the mode of action that correlates with an increased toxicity towards host cells, and may provide some insights into the mechanisms underlying it.

The peptides P1 or P7, like magainin, show a relatively moderate cytotoxicity towards animal cells. They permeabilize lymphocyte membranes leading to a necrotic effect, characterized by increased PI uptake and decreased FS in flow cytometric experiments (Figure 5). This correlates with the propensity for helical structuring (see also [12,15]). For P3, a markedly increased PI uptake is instead associated with a shift in the SS parameter that indicates increased granularity, corresponds to a highly pocked surface in SEM (Figure 5) and is consistent with a different type of membrane damage. An increased log SS has also been described for melittin acting on the HMy2 human lymphoblastoid cell line [45], and was interpreted in terms of the formation of lipid vesicles on the cell membrane or intracellularly, which is compatible with our observations. P3 is more cytotoxic than melittin, which shows an intermediate effect on lymphocytes at equivalent concentration, with both types of damaged cell populations being present. In fact, P3 switches from one mode to the other in a concentration-dependent manner, which could indicate a two-stage mechanism, with one type of permeabilization leading to the other, but only for aggregation-prone peptides at higher concentrations. The aggregation-prone LL-37 also behaves in a quite similar manner (unpublished work). In this respect, it is possible that this type of peptide might access alternative mechanisms for membrane penetration, such as the ‘sinking raft’ model proposed by Pokorny and Almeida [46].

Finally, our results indicate that the physical interaction of the peptides with the membrane, as proposed by the various mechanisms that can be included in the carpet model, is the principal factor leading to cellular inactivation. However, this does not exclude that other effects involving the bacterial cell wall and membrane, such as activation of autolytic enzymes in bacteria [48], or of endogenous phospholipases in host cells [49] by the AMPs, or interactions of the AMPs with internal targets once the membrane is breached, could also have important accessory roles. Nevertheless, the considerations presented in the this study are not only useful to understand better the behaviour of natural α-helical AMPs and toxins, but also furnish a framework for the design of synthetic AMPs with optimized activity/selectivity profiles for possible biomedical applications.

This work was supported by grants from the Italian Ministry of Universities and Scientific Research (PRIN 2003), and from a Friuli-Venezia Giulia regional grant. I.Z. and U.P. have been supported by a grant from the V Framework EU PANAD project QLK2-CT-2000-00411. H.-G.S. acknowledges financial support for antimicrobial peptide research from the Deutsche Forschungsgemeinschaft. The free-of-charge availability of the flow cytometry facility of the Fondazione Callerio ONLUS is also gratefully acknowledged.