|Home | About | Journals | Submit | Contact Us | Français|
The mechanisms of six different antimicrobial, cytolytic, and cell-penetrating peptides, including some of their variants, are discussed and compared. The specificity of these polypeptides varies, but they all form amphipathic α-helices when bound to membranes, and there are no striking differences in their sequences. We have examined the thermodynamics and kinetics of their interaction with phospholipid vesicles, namely binding and peptide-induced dye efflux. The thermodynamics of binding calculated using the Wimley-White interfacial hydrophobicity scale are in good agreement with the values derived from experiment. The generally accepted view that binding affinity determines functional specificity is also supported by experiment in model membranes. We now propose the hypothesis that it is the thermodynamics of peptide insertion into the membrane, from a surface-bound state, that determines the mechanism.
During the past three decades, a vast number of antimicrobial peptides (1, 2) and other related cytolytic peptides (3) have been discovered and their mechanisms examined. More recently, several cell-penetrating peptides have been described, which allow for transport of large molecules, such as proteins or DNA fragments, into cells (4-8). Perhaps surprisingly, many of these antimicrobial, cytolytic, and cell-penetrating peptides fall into the same structural class: they form an amphipathic α-helix of some 14–40 residues, when bound to a membrane surface. Yet, they show remarkable specificity regarding the target membrane or organism. What has befuddled researchers for a long time is the absence of a correlation between sequence and function or mechanism. The only element that appears to separate antimicrobial from cytolytic peptides is that antimicrobials are usually cationic. This provides a simple explanation for their specificity because cationic peptides should bind better to the anionic membranes of most bacteria than to the neutral membranes of eukaryotic cells (9).
We now critically review results obtained over the past several years on a set of representative antimicrobial, cytolytic, and cell-penetrating peptides. The interactions of these peptides with model membranes were all studied with the same methods and under similar conditions. Experiments using small unilamellar vesicles (SUV)1 were common in the past but we purposely exclude them because of the strained nature of those vesicles, concentrating instead on studies that use unstrained vesicles, such as large (LUV) or giant unilamellar vesicles (GUV). The results are, therefore, directly comparable. On the basis of a quantitative analysis of the kinetics and thermodynamics of these interactions, we propose the hypothesis that the peptide sequence only specifies the mechanism indirectly, through the thermodynamics of peptide insertion into the bilayer medium from the surface-bound state. This would explain the lack of direct correlation between sequence and mechanism.
Beyond interactions with simple model lipid membranes, many questions remain regarding the function of antimicrobial and cytolytic peptides, which will not be discussed here. For example, the cell membrane is probably heterogeneous; it is likely to contain regions where peptides partition preferentially and regions from which they are excluded. We have suggested that this results in concentration of the peptides in preferred regions, increasing their efficacy (10, 11). Also, the role of membrane proteins is virtually unexplored. It is well known that there are no specific cell-surface receptors for these peptides (12, 13), but their binding and insertion into the membrane may be influenced by other proteins. Another question is the functional role of peptide oligomerization in aqueous solution. δ-Lysin, for example, forms a four-helix bundle, with the hydrophobic residues inside, shielded from water (14). One possibility is that self-association is a strategy to avoid hydrolysis by proteases, but this question is also unresolved.
We have examined the binding to membranes and the mechanism of membrane perturbation or disruption of a small group of peptides, which represent different chemical properties and biological specificities (Table 1). The peptides selected were the hemolytic peptide δ-lysin from Staphylococcus aureus, the antimicrobial peptides cecropin A from the moth Hyalophora cecropia and magainin 2 from the frog Xenopus laevis, the cytotoxic mastoparans from the wasps Vespula lewisii (mastoparan) and Vespa xanthoptera (mastoparan X), and the synthetic, cell-penetrating-peptide transportan 10 (tp10). To these we have added melittin, from bee (Apis mellifera) venom, which has been extensively studied by other authors and will also be discussed here. All these peptides form amphipathic α-helices, but they differ in length, charge, and specificity, thus spanning a broad range of properties.
Various models have been proposed for the function of these amphipathic peptides, such as the barrel-stave (15), toroidal-pore (16, 17), sinking-raft (18-20), and carpet models (21, 22). Huang and collaborators have proposed that membrane thinning, which is a consequence of peptide binding, is a critical determinant of membrane perturbation and pore formation (23). The peptides initially bind to the membrane surface and cause bilayer thinning until the peptide-to-lipid ratio (P/L) reaches a certain threshold, P/L*. Beyond this point, no more thinning occurs and peptides begin to insert into the membrane. This threshold in membrane thinning coincides closely with a change from an orientation parallel to the membrane surface to an inserted state, observed by oriented circular dichroism (23, 24). The value of P/L* depends somewhat on the peptide and the lipid, but is typically close to 1:50 (23, 24). Above P/L*, the inserted state appears to form pores that are equilibrium structures, whereas for peptide concentrations below this threshold only transient pores form (24). The conformation of the lipid bilayer around these pores has been recently investigated by X-ray diffraction (25). A Bax-derived peptide allows a continuous, curved bilayer around the pore, consistent with a toroidal pore, which has been proposed for magainin 2 and melittin (26). On the other hand, alamethicin leads to a complete interruption of the bilayer consistent with a barrel-stave pore. It should be mentioned, however, that a recent study of alamethicin in membranes proposes a model that is very different from a static barrel-stave pore; rather, the function of alamethicin is explained by the formation of transient pores that result from random association of alamethicin peptides inserted perpendicular to the membrane (27).
The current consensus is that cytolytic and antimicrobial peptides kill their target cells by membrane disruption or perturbation, and even cell-penetrating peptides must transiently perturb the membrane as they gain access to the cell interior. Instead of focusing on the different types of molecular models, we have concentrated on trying to understand peptide-induced dye efflux kinetics. Kinetic, rather than equilibrium data were used because the peptide mechanism can involve a transient pore, which may not correspond to the most populated peptide state. Moreover, the final, equilibrium state obtained by adding peptides to lipid membranes may tell us little about the mechanism of membrane disruption or perturbation. As also recently remarked by Huang and collaborators, “pore formation in cell membranes caused by water-soluble peptides typically occurs as a kinetic process” (24).
What is the rate limiting step is this process? We may subdivide it into three steps: pore formation, finding the pore by an entrapped particle, and crossing the pore. The characteristic time for crossing the pore itself is about h2/(2D), where h is the length of the pore. Crossing the pore is not rate-limiting. With h = 50 Å (bilayer thickness), this time is of the order of 100 ns, which is in agreement with recent simulations for ions through protegrin pores (28, 29). Even if a free energy barrier of about 5 kcal/mol is assumed for crossing the pore, this time only increases to 1 ms. The characteristic time of escape of a small particle from a sphere of volume V through a small hole of radius a is of the order of V/(2Da), where D is the diffusion coefficient (30, 31). Thus, the time for efflux of the contents of an LUV, with a diameter of 0.1 μm, through a pore of about 20 Å in diameter is about 1 ms. Escape times of about 10 ms were obtained from simulations (32). For a GUV, with a diameter of 10 μm, a similar calculation yields 1000 s, but experimentally efflux from a single vesicle occurs in about 10 s (33), indicating that many pores form or that they are much larger in GUVs. In conclusion, the efflux times from a single vesicle are much smaller than those observed experimentally in peptide-induced efflux from a vesicle population. This tells us two important things about the mechanism of antimicrobial peptides: a single pore is sufficient to cause rapid release of contents from a large vesicle (LUV) and, if such a pore forms, the dye efflux times measured experimentally in a vesicle population reflect mainly the time of pore formation (34).
We have measured the kinetics of peptide binding to membranes and peptide-induced dye efflux from lipid vesicles (LUVs) under similar conditions, namely at P/L of 1:50 to 1:100 or less, with peptide and lipid concentrations of 0.5–1 μM and 25–500 μM, respectively, for the chosen set of peptides, δ-lysin, tp10, mastoparans, cecropin A, and magainin 2 (18-20, 35-38). Kinetic models of possible mechanisms were tested using a global theoretical analysis (18, 19), by directly fitting the sets of differential equations that represent the kinetic models to the experimental, time-dependent data. Our experiments were performed at low P/L, under which conditions only transient pores exist according to Huang’s analysis. Therefore, there is no contradiction with existence of equilibrium pores proposed by Huang’s laboratory (24), or aggregation in the membrane, as shown recently for melittin (39), which may occur at larger P/L. However, we maintain that it is not necessary to reach the high P/L regime for these peptides to function.
The mechanisms of δ-lysin, tp10, mastoparans, cecropin A, and magainin 2 can be explained by different molecular models. However, knowing the mode of dye release, a quantitative analysis of the kinetics of peptide binding and dye efflux allowed us to divide them into two groups, according to the type of kinetic mechanism, all-or-none (35, 36) or graded (18-20, 37) (Figure 1). Melittin induces graded dye release (40), but we have not modeled its kinetics. In both types of mechanisms, the peptides initially bind to the vesicle surface and accumulate there, creating a mass imbalance across the lipid bilayer, which perturbs the membrane. Beyond that point, there are differences between the two mechanisms, which affect the kinetics. In an all-or-none mechanism (35) (Figure 1, top), partitioning of the peptides into the bilayer interior is very unfavorable, as discussed below. Pores form transiently, as a stochastic process, probably initiated by a peptide-induced defect in the membrane, as if it were under tension (24, 41-44). The presence of peptides stabilizes the pore (44), which allows the contents of the vesicle to leak out, essentially all at once. This is probably not a well-organized channel, since no significant peptide oligomerization occurs (35, 36), but rather a somewhat disordered or “chaotic” (45) toroidal pore, lined mostly by lipids, with some associated peptides, as suggested by recent molecular dynamics simulations (46-48). Previously, we argued that peptide translocation is probably limited in this mechanism (35, 36). However, while not essential for the all-or-none mechanism, peptide translocation across the membrane is not incompatible with the kinetics, even quantitatively, and it could occur to a significant extent, at least for some peptides, as proposed for magainin 2 (17). The lifetime of the pore is long enough so that the entire contents of the vesicle are released. This creates a population of empty vesicles, which increases in time in a dye efflux experiment, and influences the observed kinetics as they compete with the full vesicles for peptide binding (35). Recent experiments with GUVs have clearly demonstrated the all-or-none nature of dye release induced by magainin 2 (33, 49), in agreement with our results (36). Other GUV kinetic experiments have supported the idea that the mechanism of pore formation is stochastic and similar to tension-induced membrane disruption (24).
On the other hand, in a graded mechanism (18-20, 37) (Figure 1, bottom), the probability of a peptide transiently inserting into the hydrophobic core is larger, and peptide translocation across the bilayer may occur concomitant with membrane perturbation. In the bilayer-inserted state, which constitutes the apparent pore, the peptide “catalyzes” dye efflux from the vesicle lumen. As peptide translocation is completed, the mass balance across the bilayer is restored and the rate of efflux becomes very slow or eventually stops (18, 19, 50). That is, efflux ceases when the peptide population equilibrates across the bilayer. In principle, this is a dynamic equilibrium, where insertion could occur from both sides, but in this mechanism the rate of insertion is negligible in the absence of a mass imbalance and consequent bilayer stress. Therefore, a significant amount of dye may remain inside each vesicle at the end of a dye efflux experiment, if peptide translocation is fast compared with the rate of dye efflux. In this kinetic model, all vesicles are equivalent; they can, in fact, be modeled as one enormous vesicle. No empty vesicles ever exist, except at the very end, if complete release is achieved. A model very similar to the one we proposed (18-20) was suggested from molecular dynamics simulations (48). It has been shown that graded and all-or-none release can be obtained as two extreme cases of a model, when the efflux time is either very long or very short compared to the pore lifetime (51). A “gray zone” may exist for intermediate situations. Yet, the differences between the two mechanisms noted above are important in modeling the release kinetics and may stem from significant differences in peptide structure, as discussed below.
The two modes of dye release, graded and all-or-none, can be distinguished by a requenching experiment (32, 52, 53) in which a fluorophore, 8-aminonaphtalene-1,3,6-trisulfonic acid (ANTS), and a quencher, p-xylene-bis-pyridinium bromide (DPX), are both incorporated in the vesicle lumen. As peptides interact with the membrane, both the fluorophore and the quencher leak out at comparable rates. In graded release, the fluorescence of ANTS inside the vesicles increases because the quencher concentration decreases. Thus, a plot of the fluorescence inside against the fraction of ANTS released yields a rising curve (Figure 2, top). But in all-or-none release, the degree of quenching inside the vesicles is independent of the amount of ANTS and DPX released because only the intact vesicles contribute to the signal from inside the vesicles, and the plot yields a horizontal line (Figure 2, bottom). Cecropin A and magainin-2 cause all-or-none release and their kinetics are quantitatively described by an all-or-none kinetic mechanism (35, 36). δ-Lysin, tp10, and mastoparans cause graded release of vesicle contents and their kinetics of dye efflux are described by a graded kinetic mechanism (19, 20, 37, 54, 55). They appear to translocate across the bilayer, tp10 and mastoparans as monomers, and δ-lysin as a small oligomer. Melittin also causes graded release (40) and appears to translocate across the bilayer (56). Cecropin A and magainin-2 appear to disrupt or perturb the vesicles through a chaotic pore event, which leads to leakage of the vesicle contents, essentially all at once. Most of the differences in efficiency of cecropin A and magainin 2 toward vesicles containing varying amounts of anionic lipids appear to be due to differences in binding (35, 36). This conclusion was recently supported by experiments with GUVs, which showed that the ratio of bound peptide to lipid determined dye release kinetics induced by magainin 2 (49). According to our data, peptide oligomerization in membranes is not essential for dye release (20, 35-37), except for δ-lysin, in which case it involves dimers, trimers, or at most tetramers (18, 19).
To understand the reasons for the two different types of behaviors, we have examined the thermodynamics of binding of the peptides to the surface of a 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) membrane and their transfer to the bilayer hydrophobic interior (20, 38) (Figure 3). The thermodynamics of peptide association with membranes were obtained by measuring the kinetics of binding to LUVs of POPC or mixtures of POPC with anionic phospholipids (20, 35-38, 57). The kinetics were measured by stopped-flow fluorescence using the change in Förster resonance energy transfer (FRET) from a Trp residue on the peptide to a phospholipid labeled with the fluorophore 7-methoxycoumarin (7MC), which is incorporated in the bilayer (35-38). Occasionally, the changes in the fluorescence intensity of the Trp or a 7MC-modified lysine on the peptide were used instead of FRET (20). The measurements yield the on- and off-rate constants, kon and koff, from which the equilibrium dissociation constant is obtained as KD = koff/kon.2 The case of δ-lysin is complicated by its oligomerization in aqueous solution (14). At a concentration of 0.5 μM, which was used in the binding measurements (57), the peptide exists as a mixture of about 60% monomers and 40% dimers. This distribution is calculated from the tetramer and dimer dissociation constants that we previously estimated (19). The dimer is likely to bind more weakly than the monomer because the hydrophobic face of the amphipathic helix involved in dimerization in water is the same that binds to the membrane interface. Therefore, the experimental value of KD = 60 μM (57) is likely to be close to the monomer dissociation constant.
To include the correct mixing entropy term, KD needs to be converted to mole fraction units (58). This so-called “cratic correction” is applied by subtracting RT ln 55 from the Gibbs energy derived from KD, that is, at room temperature. We should note that the cratic correction is a controversial issue (59-61), but for consistency with the work of White and Wimley (58) we follow their recommendation and apply this correction to our values. The Gibbs energy of binding to the membrane/water interface, , can be calculated with the Wimley-White interfacial scale (62), which was extended to include transfer of several different types of amino- and carboxyterminal groups to the membrane interface (63). The binding process is a coupled binding-folding event (58) whereby a peptide, mostly disordered in aqueous solution, binds and folds to an α-helix on the membrane surface (Figure 3, dashed arrows). When an amphipathic helix forms at the interface, the Gibbs free energy of the peptide is reduced by about 0.4 kcal/mol per residue (64). Thus, is the sum of two terms, representing binding in an unfolded state and folding at the interface. The results of the calculations for this set of peptides, performed with Membrane Protein Explorer, MPEx (65), are shown in Table 2. To calculate , the degree of helicity of the peptides needs to be obtained from experiment or estimated reliably. δ-Lysin is 73% helical on a micelle surface (66, 67), but our circular dichroism (CD) measurements indicate about 90% helical content on POPC vesicles (Huskins and Almeida, unpublished observations). Mastoparan X is about 86% helical on a membrane surface (68), and mastoparan is about 70% helical, taking into account several estimates (69-71). There are no helicity measurements on tp10 and most of its variants, but we have estimated by CD that tp10W is about 60% helical on POPC/1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) 1:1 membranes (Huskins and Almeida, unpublished observations). This falls within the values of 53% for transportan (72), the parent peptide from which tp10 is derived by a deletion of the 6 N-terminal residues, and 70% for mastoparan, whose sequence constitutes the 14 C-terminal residues of tp10, out of a total of 21 residues. Magainin 2 is about 83% helical on a membrane based on NMR data (73), which we think is the most reliable estimate; other reported estimates range from about 55 to greater than 90% (74-80). Cecropin A is about 80% helical in hexafluoroisopropnanol/water solution (81, 82) and essentially maintains the same helical structure when bound to membranes (83); our CD measurements yield about 70 % helical content (Huskins and Almeida, unpublished observations). Melittin is about 70% helical on POPC or POPG membranes (64); other reported values range from about 65% to 85% on POPC/POPG mixtures (39, 84).
The agreement between the experimental () and calculated () Gibbs energies of binding is very good for tp10W, mastoparan X, magainin 2, and melittin (Table 2). For tp10W, (38) and ; for mastoparan X, (37) and ; for magainin 2 (F12W variant), (36) and ; and for melittin, (39, 85-87) and .
However, for δ-lysin and cecropin A the calculated are in poor agreement with the experimental values. For cecropin A, whereas (35), and for δ-lysin whereas (57). What is the reason for this discrepancy? We think that the disagreement is due to formation of salt bridges (hydrogen-bonded ion pairs) between positive- and negatively charged functional groups of the peptide. None of the peptides whose calculated agrees with obtained from experiment—magainin 2, melittin, mastoparan X, and tp10 variants—can possibly establish intramolecular salt bridges. On the other hand, judging by their sequences, cecropin A could establish a maximum of 2 and δ-lysin probably 3 intramolecular salt bridges, in a helical structure (Table 1). Cecropin A may contain a (i + 3)K,E salt bridge between Lys6 and Glu9 and a (i + 4)D,K salt bridge between Asp17 and Lys21. According to Marqusee and Baldwin (88), (i + 4) salt bridges stabilize the α-helix more than (i+3) salt bridges, and E,K salt bridges stabilize the α-helix more than K,E salt bridges. Therefore, we expect the (i+4)D,K salt bridge to be especially strong. δ-Lysin could make a (i+3)D,K salt bridge between Asp11 and Lys14 and a (i+4)D,K salt bridge between Asp18 and Lys22. In addition, it is possible that a salt bridge be established between one of the terminal lysines and the free terminal carboxylate. Alternative patterns involving a (i + 4)K,D salt bridge between Lys14 and Asp18 are also possible. In octanol, formation of an intramolecular salt bridge between acidic and basic groups of the peptide has been estimated to lower its Gibbs free energy by 4 kcal/mol (89). On the POPC membrane interface, this contribution must be smaller but still favorable (62). The free energies of transfer of charged residues from water to the interface are smaller than those of transfer to octanol by a factor of roughly 2 (58). We conjecture that, similarly, the free energy of salt bridge formation at the POPC/water interface will be smaller than in octanol by about the same factor, corresponding therefore to about 2 kcal/mol. This value appears reasonable because formation of a salt bridge in water, on the surface of a globular protein contributes about 1 kcal/mol to protein stability. Hence, if δ-lysin were to form 2 salt bridges when bound to the membrane surface, , in good agreement with the value obtained from experiment, . And if cecropin A were to form 2 salt bridges, , which compares to derived from experiment (Table 2).
Overall, the Wimley-White procedure provides very reasonable estimates of Gibbs energies of binding to POPC membranes. For peptides possessing cationic and anionic groups within hydrogen-bonding distance, experimental and calculated binding Gibbs energies can be reconciled if formation of salt bridges is taken into account. For the sake of completeness, let us note that a different approach to determining peptide binding thermodynamics has been developed by Seelig’s group (39, 84, 85, 90, 91) and that there are a few discrepancies in some of the free energies obtained. For instance, according to Ladokhin and White (64), the Gibbs energy is reduced by about 0.4 kcal/mol per residue upon folding to a helix at the interface, but the estimate from Seelig’s group is a factor of 2 smaller, about 0.2 kcal/mol-residue (39, 90, 91). This is somewhat compensated by a more favorable free energy of binding of the unfolded state than calculated using the Wimley-White interfacial hydrophobicity scale. A discussion of the possible origins of these differences is beyond the scope of this article, but a few comments are justified. Since the binding of cationic polypeptides to zwitterionic membranes of pure POPC or dioleoylphosphatidylcholine (DOPC) is typically weak—because of the unfavorable Gibbs energy of partitioning of the peptide cationic side chains to the bilayer/water interface—Seelig and collaborators have measured binding to mixed vesicles of POPC/POPG, containing variable amounts of the anionic lipid, and have then factored out the electrostatic effects to obtain a “bare” partition constant. Those electrostatic effects, which include attraction between the peptide and the anionic lipid and repulsion between peptides, are estimated using Gouy-Chapman theory by a combination of calculation and fitting to experimental data (39, 84, 85). What is less clear is whether and to what extent the interactions between positive charges on the peptide and the zwitterionic bilayer interface are lumped into the electrostatic component of the binding constant, not explicitly, but implicitly, through the fitting procedure. This could lead to a reduction of their unfavorable effect on partitioning, leading to an overestimate of the bare binding constants. While both the approaches of Wimley-White and Seelig provide reasonable estimates of binding Gibbs energies to the bilayer interface, mixing elements of the two is likely to yield incorrect results because compensation of small inaccuracies, which probably exist within each method, would be lost. At this point, only the Wimley-White procedure can be used to calculate binding Gibbs energies for an arbitrary peptide, and they agree well with the values derived from experiment, but we should keep in mind that there are alternative approaches and unresolved issues.
Let us now examine the thermodynamics of peptide transfer, as an α-helix, from a POPC membrane interface to the bilayer hydrophobic interior, using again the approach of White and Wimley (58). The relevant Gibbs energies of transfer for the set of peptides examined are shown in Table 2. The concept is illustrated in Figure 3 (20). A peptide is transferred from water to the membrane surface and from the surface to the bilayer interior. To complete the thermodynamic cycle, the Gibbs energy of folding () of a peptide to an α-helix in aqueous solution must be included. This value is generally not known for most peptides, but it is probably small. The number of degrees of freedom “frozen” in a coil—helix transition is about 10 per residue. This number can be understood in a simplified but intuitive manner if we consider that, in a random coil, there are three well-defined minima for the values of the ϕ and ψ dihedral angles (92). Therefore, each residue has about nine discrete conformations in a random coil, which are reduced to one in an α-helix. This is, of course, only a very rough approximation because the three minima are not equivalent and they are not discrete states. Nevertheless, several estimates of the entropy change associated with the coil—helix transition yield values that correspond to freezing about 10 degrees of freedom, ΔS ≈ −R ln 10 ≈ −4.5 cal/K/mol-residue. Schellman (93, 94) estimated the reduction in entropy upon folding into a helix to be between 4 and 5 cal/K/mol. Brandts (95) arrived at a similar value, of 5.1 cal/K/mol, and Privalov (96), on the basis of the entropy of unfolding for a set of proteins, estimated the conformational entropy change per residue to be 4.2 cal/K/mol. The Gibbs free energy change for the coil—helix transition in water () can be calculated using an all-or-none model with end effects or the zipper model (94, 97). More sophisticated approaches, such as the Zimm-Bragg (98) or Lifson-Roig (99) theories are not justified for this simple estimate. Alanine-based peptides of 14–38 residues have helix-coil transition temperatures (Tm) in aqueous buffer varying from −10 to 35°C (100). Using the values of Tm for this set of peptides and 4.5 cal/K/mol for the conformational entropy change, the all-or-none or the zipper models for the coil—helix transition yield , unfavorable for the smaller peptides and favorable for the longer ones in this set. These peptides were originally designed to establish salt bridges (88), so they might be expected to form more stable helices, but another set of alanine-based peptides that cannot make salt bridges also forms stable helices in aqueous solution (101). In water, most antimicrobial peptides are primarily unstructured, but the order-of-magnitude estimate of is probably valid.
The Gibbs energy of transfer of peptides from water to octanol can be obtained from the whole-residue octanol transfer scale (102). Although the bilayer interior is not similar to octanol, it turns out that the transfer of an α-helical polypeptide, from water to the bilayer hydrophobic interior can be reasonably estimated from the free energies of transfer from water to octanol, (103). The Gibbs energy for insertion into the bilayer nonpolar interior () of a helical peptide bound to the membrane interface is obtained by closing the thermodynamic cycle, (Figure 3). To a fairly good approximation, , because for folding in water is much smaller than all the other terms. Inspection of Table 2 reveals an interesting fact: is about 20 kcal/mol or less for peptides that cause graded dye release, which includes mastoparans, tp10, melittin, and δ-lysin, with ; but for peptides that cause all-or-none release, which includes magainin 2 and cecropin A, with . All uncertainties in experimental and calculated values of Gibbs energies of transfer notwithstanding, this suggests that insertion is easier for peptides that follow a graded mechanism.
At this point, we want to make clear that we do not mean that antimicrobial and cytolytic peptides work by partitioning into the bilayer hydrophobic core. What we suggest is that the Gibbs energy of transfer from the interface to the bilayer hydrophobic core, estimated by , provides a tool for predicting the behavior of the peptides. The idea is that the thermodynamics of pore formation reflect the thermodynamics of insertion, however perturbed the bilayer may be by interaction with the peptide, in an extreme case with formation of a pore.
Finally, we consider briefly a different class, that of polycationic cell-penetrating peptides, which includes penetratin (4), the HIV-1 TAT peptide (5, 6), polyarginine, and polylysine, for example. These peptides have been shown to cross membranes and even layers of non-polar solvents, provided that phosphate-containing or other anionic counter-ions be present (104-108). They do not form α-helices, bind very weakly to zwitterionic membranes, and their mechanism may be unrelated to that of the amphipathic, α-helical peptides discussed here, which include the cell-penetrating peptide tp10 and its variants. Yet, it is curious that application of the Wimley-White hydrophobicity scales to many of those polycationic peptides yields . For example, for the TAT peptide (YGRKKRRQRRR), both transfer to the interface and to octanol are very unfavorable, and , but . Similarly, for nonarginine (Acetyl-Arg9-amide), and , but .3 And for nonalysine (Acetyl-Lys9-amide), and , yielding a larger value, . Experimentally, it is found that nonarginine is more efficient than nonalysine (109) or TAT (110). At reasonable concentrations, these peptides will not bind to zwitterionic vesicles because of the very unfavorable, positive . However, if the anionic lipid content of the membrane is large enough, binding will occur. If the inserted or pore state is stabilized by a free energy similar to the surface-bound state in anionic vesicles relative to zwitterionic vesicles, which appears to be true for mastoparan X, tp10, cecropin A, and magainin 2 (see next section), a situation can be achieved where these peptides bind but is still below the threshold for translocation. Interesting suggestions regarding their mechanism come from molecular dynamics simulations (48). While the mechanisms of amphipathic (46, 47) and polycationic cell-penetrating peptides (48) appear similar at first sight, in both cases involving formation of a disordered toroidal pore that includes one or a few peptides, some subtle differences may be important. Particularly intriguing is the idea that arginine side chains of the cell-penetrating peptide reach out to the phosphate groups of the apposing leaflet of the bilayer, thus establishing hydrogen bonds and initiating the formation of a small pore (48).
Consider a lipid vesicle to which a peptide binds from solution, eventually causing efflux of the vesicle contents. The graph of the Gibbs free energy profile for a hypothetical path that leads to insertion is shown in Figure 4. In the simplest scenario, the rate-limiting step for efflux should be the insertion of the peptide into the bilayer and its consequent perturbation. That rate is dominated by the exponential factor e−ΔG‡/RT, where ΔG‡ is the activation energy barrier for peptide insertion. This process is unfavorable because of interactions between the peptide polar groups and the lipid acyl chains, and therefore the inserted peptide is a high free energy state. According to Hammond’s postulate (111), we can assume that it lies close to the transition state (‡) and should resemble it. Therefore, the Gibbs energy of transfer from the membrane interface to the bilayer interior () should be approximately equal to the Gibbs activation energy for insertion into the bilayer hydrophobic core (ΔG‡). If is small, as presumed, this argument provides an easy way to estimate , which can be calculated using the Wimley-White transfer scales. Membrane reorganization, including pore formation, will lower the actual activation energy considerably.
For cationic peptides, efflux from vesicles containing anionic lipids is much faster than from POPC vesicles. It turns out that for tp10, cecropin A, magainin 2, and mastoparan X the rates of pore formation obtained from our fits do not depend very much on anionic lipid content. However, increasing anionic lipid content clearly enhances binding (35, 36). This suggests that, as the Gibbs energy of the interface-bound state () decreases with increasing anionic lipid content, the Gibbs energy of the transition state (G‡) also decreases by a similar amount (Figure 4). In fact, the rates of pore formation obtained for cecropin A and magainin 2 depend little on anionic lipid content (35, 36), which is consistent with ΔG‡ remaining approximately constant. We have also observed this parallel change of calculated ΔG‡ and ΔGbind for tp10 variants in mixtures of POPC with anionic lipids, leading us to suggest that, in anionic membranes, part of the binding energy is used to disturb the bilayer (38).
Based on the set of peptides examined—which is, admittedly, very limited—we propose the hypothesis that the mechanism of antimicrobial, cytolytic and cell-penetrating peptides is determined by the thermodynamics of insertion into the membrane from the surface-bound state. Whether insertion actually occurs or not is another question. The interactions with the membrane are of course determined by the peptide sequence, but it is not the sequence directly that determines mechanism and specificity. This would explain why functionally determinant sequence motifs have not been found in antimicrobial peptides in spite of almost three decades of research.
We conclude with two predictions and two remarks:
Whether the formulated hypothesis and predictions prove correct remains to be determined. Work in our laboratories is currently in progress to test these ideas.
We thank Laura Huskins for performing the CD measurements.
1Abbreviations and Textual Footnotes
2As long as kon refers to binding to vesicles, the meaning of KD is clear. If one tries to relate it to lipid concentrations, the question arises regarding dividing KD by 2 to correct for binding to the outer monolayer only. Whether or not this is appropriate depends on whether the peptides translocate across the membrane. If they do, the lipid chemical potential is determined by the entire bilayer and division by 2 is not appropriate. We have observed small differences in koff obtained from association and dissociation kinetics, by a factor of about 2 on average (35, 36, 38), slower if measured from dissociation reactions. This suggests that translocation occurs to some extent, concomitant with membrane perturbation or pore formation by the peptide. Therefore, we prefer to leave KD undivided. If the experimental measurements reflected binding only to the outer monolayer of the vesicles, the term −RT ln 2 = −0.4 kcal/mol would have to be added to the Gibbs energy of binding () obtained from KD. This level of uncertainty is present in all these calculations.
3Usually terminally-modified versions of polycationic peptides have been used in the experiments, typically with a fluorescein chromophore on the N-terminus; therefore, we used N-terminal acetylated and C-terminal amidated sequences in the calculations, but free-terminal sequences would lead to the same qualitative conclusions.
†This work was supported by National Institutes of Health Grant GM072507.