Common mechanism unites membrane poration by amyloid and antimicrobial peptides

Abstract

Poration of bacterial membranes by antimicrobial peptides such as magainin 2 is a significant activity performed by innate immune systems. Pore formation by soluble forms of amyloid proteins such as islet amyloid polypeptide (IAPP) is implicated in cell death in amyloidoses. Similarities in structure and poration activity of these two systems suggest a commonality of mechanism. Here, we investigate and compare the mechanisms by which these peptides induce membrane leakage and bacterial cell death through the measurement of liposome leakage kinetics and bacterial growth inhibition. For both systems, leakage occurs through the nucleation-dependent formation of stable membrane pores. Remarkably, we observe IAPP and magainin 2 to be fully cross-cooperative in the induction of leakage and inhibition of bacterial growth. The effects are dramatic, with mixtures of these peptides showing activities >100-fold greater than simple sums of the activities of individual peptides. Direct protein–protein interactions cannot be the origin of cooperativity, as IAPP and its enantiomer D-IAPP are equally cross-cooperative. We conclude that IAPP and magainin 2 induce membrane leakage and cytotoxicity through a shared, cross-cooperative, tension-induced poration mechanism.

Membrane permeation is a central event in many cytotoxic processes. Antimicrobial peptides, such as magainin 2, protegrin-1, and alamethicin, function by permeabilizing the membranes of pathogenic bacteria (1). Toxic amyloid oligomers, such as those made by islet amyloid polypeptide (IAPP), amyloid-β (Aβ), and α-synuclein, have also been shown to form membrane pores that are capable of contributing to cell death in amyloid disease (2, 3).

Magainin 2 (Fig. 1) is 23-residue broad-spectrum antimicrobial peptide first identified from Xenopus laevis (4). It has been hypothesized to induce membrane leakage through either a toroidal (5) or chaotic (6) pore mechanism, depending on the specific experimental conditions. IAPP (Fig. 1) is a 37-residue peptide that forms amyloid fibrils and contributes to progressive beta-cell death in type II diabetics (7) and failures of islet transplants in type I diabetics (8). Prefibrillar states of IAPP have been shown to be cytotoxic by a number of groups who have suggested detergent, carpeting, and pore mechanisms (2).

Fig. 1.

Sequence and structure of magainin 2 and IAPP. (A) Primary structure of magainin 2, rat IAPP, and hIAPP. (B and C) Helical wheel representations of the α-helical segments of (B) magainin 2 and (C) rat IAPP. The green circles represent hydrophobic residues, yellow circles represent polar residues, and blue circles represent basic residues. The arrows show the hydrophobic moment. Helical wheel diagrams were produced using MPEx (47).

Numerous physicochemical characteristics link these peptides and their membrane binding behavior. Magainin 2 and IAPP are both short, cationic peptides that are natively unstructured in solution (9, 10). Both are characterized by a disorder-to-order transition upon exposure to lipid bilayers, with both peptides adopting an amphipathic α-helical conformation (11, 12). Initial peptide binding is parallel to the membrane surface, with the hydrophilic face of the peptides exposed to solvent and the head group region of the bilayer, whereas the hydrophobic regions interact with the phospholipid acyl chains. Membrane permeation by these two peptides has also been shown to include stochastic initiation of leakage, a subsequent evolution of leakage, and the presence of equilibrium pores (13, 14). Given these similarities, as well as those of many other amyloid and antimicrobial peptides, it has been hypothesized that these two classes of peptide may act against their separate targets via a similar mechanism (15).

The toxic oligomer hypothesis of amyloid disease postulates that oligomeric protein assemblies are responsible for toxic gains-of-function by normally nontoxic proteins (16). This view has been supported by optical and atomic force microscopies (17), immunochemical methods (18), and most recently, crystallography (19). Oligomerization has also been suggested to be a basis for induction of membrane leakage by antimicrobials (1). In contrast, it has also been found that peptide binding to bilayers can result in the induction of surface tension (20). One possible response of a bilayer to protein-induced surface tension is the formation and stabilization of a membrane pore.

In the current study, we aim to determine whether magainin 2 and IAPP induce membrane leakage and cell death through a common mechanism. This pairwise comparison is designed to elucidate mechanistic specifics of membrane permeation by both peptides. One of our premises is that induction of membrane surface tension, and subsequent poration, need not be dependent on specific protein–protein interactions (20). In contrast, toxic complexes of proteins require specific protein–protein contacts often achieved by prolonged incubation of protein at high concentration. Cooperative interactions between IAPP and magainin 2 are therefore examined using in vitro and in vivo assays of membrane leakage and cytotoxicity to distinguish between these two broad classes of cytotoxic mechanism.

Results

To focus on membrane conformations, we used magainin 2 from frog and IAPP from rodent (rIAPP). rIAPP adopts helical membrane-bound conformations similar to preamyloid human IAPP (hIAPP) (21). Importantly, rIAPP retains cytotoxic potential without progressing to form fibers (22).

Our liposome leakage assay (Fig. 2A) probes several facets of leakage competence (13). Briefly, a fluorescent dextran is encapsulated within a large unilamellar liposome. The size of this dextran (70 kDa) is such that it remains trapped in the liposome lumen upon exposure to leakage-inducing peptides. Leakage competence can then be experimentally probed at any time point after the addition of peptide through the extraluminal introduction of a small (∼400-Da) water-soluble quencher p-xylene-bis-pyridinium bromide (DPX). Here, we use this assay to probe leakage competence at equilibrium (after long incubation of liposomes with peptide), upon initial mixture of peptide with liposomes, and catalyzed response (change in leakage rate upon the addition of fresh peptide to equilibrated lipoprotein states). These three regimes are assessed in combination with the measurement of membrane binding isotherms and bacterial growth inhibition.

Fig. 2.

Cross-cooperative equilibrium membrane leakage. (A) Time-dependent leakage assay. (i) Unilamellar liposomes encapsulate 70-kDa fluorescent dextrans. (ii) Protein is added and (iii) the system is either allowed to evolve for 48 h (Fig. 2 B–E) or, alternatively, can be immediately assessed (Fig. 3) for leakage competence. (iv) Leakage competence is measured by addition of the small, hydrophilic quencher DPX. Apparent leakage profiles can reflect the rate of quencher entry into the liposome lumen and/or can reflect the rate at which liposomes switch from an intact state to one in which membrane integrity is lost (13). (B) Representative equilibrium leakage profiles induced by rIAPP ([1×] = 6 µM (blue, thin) and [2×] = 12 µM (blue, bold)), magainin 2 (M2) [[1×] = 3 µM (red, thin) and [2×] = 6 µM (red, bold)], or a mixture of [1×] rIAPP and [1×] magainin 2 (orange). Liposomes and peptide were equilibrated for 48 h before measurement. The black line represents the calculated behavior of the mixture condition if the individual proteins acted completely independently. (C) As part B, except replacing magainin 2 with 4.5 µM (purple, thin) and 9 µM (purple, bold) D-rIAPP as [1×] and [2×]. (D and E) Observed average rate constants from repeats of data collected as in part B or C, respectively.

Cooperativity of Equilibrium Membrane Poration.

rIAPP and magainin 2 maintain stable pores at equilibrium. Peptides were first equilibrated with 200 µM 1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (DOPG) liposomes (in monomer units) for 48 h. Anionic DOPG was chosen as a simple model that reflects the importance of electrostatic interactions for binding and activity by both peptides (23, 24). Leakage kinetics was then assessed by adding DPX extraluminally. At 6 µM rIAPP (Fig. 2B, blue, thin) and 3 µM magainin 2 (red, thin), similar leakage rate constants are observed: 3.0 × 10⁻⁵ and 6.3 × 10⁻⁵ s⁻¹, respectively. Such behavior is consistent with formation of a stable state, such as a pore, rather than transient mechanisms of leakage such as carpeting, which would result in liposomes resealing at equilibrium.

Equilibrated pore formation by rIAPP and magainin 2 is strongly cooperative. For both peptides, assays at the doubled concentrations of 12 µM rIAPP (Fig. 2B, blue, bold) or 6 µM magainin 2 (red, bold) show a >100-fold increase in leakage rate: 6.7 × 10⁻³ s⁻¹ for rIAPP, and >4 × 10⁻² s⁻¹ for magainin 2. This dramatic sensitivity to concentration suggests strongly cooperative behavior with a high apparent reaction order (13). At these concentrations, observed rates are wholly dominated by leakage with little contribution from liposome dissolution or aggregation effects. This is evident for rIAPP (13) and for magainin 2. For example, 48 h after exposure to rIAPP we observe a leakage profile upon DPX addition (Fig. 2B, blue, bold), consistent with the dextrans being initially fully protected from the added quencher. Thus, the liposomes are still intact. We also show by size-exclusion chromatography that >75% of encapsulated fluorescent dextrans are retained within the liposome lumen after exposure to magainin 2 (Fig. S1).

The similarity in magnitude of the concentration dependence between the peptides is suggestive of a common mechanism. To more easily compare the different peptides, we will refer instead to protein concentrations in units of leakage activity. In these units, [1×] is 6 µM rIAPP or 3 µM magainin 2, whereas [2×] is 12 µM rIAPP or 6 µM magainin 2, respectively.

rIAPP and magainin 2 are cross-cooperative with respect to equilibrium leakage. Rather than doubling the concentration of rIAPP or magainin 2, we instead mixed [1×] of rIAPP with [1×] magainin 2 and equilibrated the mixtures with liposomes. If the two peptides act independently, measurement of leakage at 48 h should result in a simple sum of their individual [1×] rate constants (Fig. 2B, black), to ∼9 × 10⁻⁵ s⁻¹. Remarkably, the leakage rate observed upon mixing [1×] rIAPP and [1×] magainin 2 is instead >100-fold faster (2.5 × 10⁻² s⁻¹; Fig. 2B, orange). The cooperativity and high apparent reaction order observed when doubling the individual peptide concentration is wholly recaptured when mixing equivalent activities of the two peptides (Fig. 2D). Thus, rIAPP and magainin 2 are fully cross-cooperative at inducing equilibrium membrane leakage.

Cooperativity of Initial Pore Formation.

Newly initiated leakage induced by rIAPP and magainin 2 is cross-cooperative. Leakage by 4 µM rIAPP or 1.25 µM magainin 2 measured when liposomes are first exposed to protein yields rate constants of ∼2.9 × 10⁻⁵ s⁻¹ (Fig. 3A, blue thin, red thin). The ratio of rIAPP to magainin 2 required to achieve equivalent leakage is comparable to that seen in the equilibrium measurements. That the relative activity per unit protein of these peptides fortuitously scales between the two measurement classes is also suggestive of a common mechanism. Doubling the [1×] concentration of either peptide individually increases the rate constant approximately eightfold (Fig. 3A, blue bold, red bold, ∼2.4 × 10⁻⁴ s⁻¹). As with the equilibrium study, we envisage that a mixture of noninteracting proteins, [1×] each of magainin 2 and rIAPP, should result in a simple sum of rates (Fig. 3A, black). Remarkably, we observe that the mixture yields rates equivalent to that of doubling the concentration of either individual peptide (Fig. 3A, orange). Such a response is far in excess of a linear sum of peptide activities (Fig. 3B). Thus, rIAPP and magainin 2 are fully cross-cooperative with respect to processes responsible for the initial formation of leakage competence.

Fig. 3.

Cross-cooperative initial membrane leakage. (A) Representative leakage profiles induced by rIAPP [[1×] = 4 µM (blue, thin) and [2×] = 8 µM (blue, bold)], magainin 2 [[1×] = 1.25 µM (red, thin) and [2×] = 2.5 µM (red, bold)], or a mixture of [1×] rIAPP and [1×] magainin 2 (orange). Measurement of leakage competence was made immediately after liposomes and peptide were mixed. The black line represents the calculated behavior of the mixture condition if the individual proteins acted completely independently. (B) Observed average rate constants from repeats of data collected as in part A. (C) As part B, except replacing magainin 2 with either 3 or 6 µM D-rIAPP. (D) As part B, except the rIAPP concentrations were 3.5 and 7 µM, and magainin 2 was replaced with either 1.4 or 2.8 µM hIAPP. Representative kinetics for C and D are in Fig. S2.

Equilibrated rIAPP pores can catalyze the conversion of magainin 2 to leakage competence. Autocatalysis is one of the characteristics of a nucleation-dependent process. We previously reported that equilibrated rIAPP pores can catalyze freshly added rIAPP directly to the pore state, bypassing the >24 h required for equilibration (13). Here, we recapitulate this by equilibrating 7 µM rIAPP with liposomes. Addition of 2 µM fresh peptide immediately leads to the rate constant increasing by 2.1 ± 0.5 × 10⁻³ s⁻¹ over a matched equilibrium sample in which additional peptide was not added. Next, we assayed the combination of 7 µM equilibrated rIAPP with freshly added magainin 2. Dramatically, as little as 0.7 µM fresh addition of magainin 2 jumps the leakage rate constant 2.7 ± 0.5 × 10⁻³ s⁻¹ above that observed when there is no added magainin 2 (Fig. 4A). To compare different peptides, we chose as a convenient reference the leakage rate constant of liposomes freshly mixed with 7 µM rIAPP. Addition of 0.7 µM magainin 2 or 2 µM rIAPP to this preparation results in only ∼1/10 and ∼1/5 the rate increase observed using equilibrated 7 µM rIAPP (Fig. 4B). Clearly, liposomes equilibrated with rIAPP can catalyze the conversion of both additional rIAPP and magainin 2 directly to the leakage competent state.

Fig. 4.

Cross-cooperative catalysis of pore formation. (A) Representative equilibrium leakage from a 7 µM rIAPP sample (blue), or a sample taken from the same equilibrium stock to which 0.7 µM fresh magainin 2 was added immediately before measurement of leakage (orange). Magenta trace shows the expected kinetic profile if magainin 2 and rIAPP are noninteracting. (B) Average change in leakage rate constants when the indicated amounts of magainin 2, rIAPP, D-rIAPP, or hIAPP are added to liposomes that were freshly mixed (0 h, green) or equilibrated (48 h, orange) with 7 µM rIAPP. Change in rate constant is determined relative to a matched 7 µM rIAPP sample that received no additional peptide.

Cooperativity of Enantiomers and Amyloidogenic Variants.

Magainin 2 and rIAPP have dissimilar primary structures; however, both are amphipathic and α-helical on membranes and therefore have the potential to form heteromeric complexes. We instead prepared the D-enantiomer of rIAPP (D-rIAPP). D-rIAPP behaves like L-rIAPP in that it is capable of forming and maintaining equilibrium leakage states. Moreover, D-rIAPP is cross-cooperative with L-rIAPP (Fig. 2 C and E) in equilibrium poration. A mixture of [1×] of D-rIAPP (4.5 µM) with [1×] of L-rIAPP (6 µM) gives a ∼100-fold increase in leakage rate, wholly equivalent to simply doubling the concentration of either individual peptide. Full cross-cooperativity is also observed in the initial formation of leakage competence (Fig. 3C) and the cross-catalysis of pores (Fig. 4B). There is thus no apparent discrimination between peptides of different primary structure or chirality in the cooperative formation or maintenance of membrane pores.

rIAPP is cross-cooperative with the human sequence variant. The rat isoform of IAPP was chosen for these experiments so that long-term leakage measurements could be performed without the significant loss of peptide to the formation of amyloid fibers. To evaluate whether a similar leakage mechanism occurs with hIAPP, we performed measurements of cross-cooperativity of rIAPP and hIAPP in both initial formation (Fig. 3D) and cross-catalysis (Fig. 4B) of leakage states (equilibrium leakage measurements were inaccessible due to the formation of amyloid fibrils by hIAPP). Although possessing similar primary structures (Fig. 1A), hIAPP has greater activity than rIAPP, requiring only ∼0.4 the concentration to induce a similar leakage rate. Fibril formation by the amyloidogenic hIAPP removes peptide from solution, stochastically slowing the leakage rate and resulting in kinetics that frequently asymptote at less than full leakage (Fig. S2B). As with magainin 2 and D-rIAPP, hIAPP is also cross-cooperative with rIAPP in both the initiation and catalysis of pore formation, demonstrating that the behavior observed here applies to both amyloidogenic and nonamyloidogenic isoforms of IAPP.

Membrane Binding of rIAPP.

Direct assessment of equilibrium membrane binding reveals no cooperativity. The adoption of α-helical structure upon binding bilayers (Fig. S3) permits a binding isotherm to be measured by changes in far-UV CD. Measurements taken promptly after mixture of peptide with liposomes show a difference between high versus low peptide concentrations. For example, at 5 µM rIAPP (Fig. 5, Inset, blue), binding can be described by a simple partition between aqueous and lipid-bound states (partition coefficient = 1.4 × 10⁵; Fig. 5, Inset, line, and Table S1). In contrast, at 23 µM rIAPP a significant departure is apparent with considerably more helical structure/binding observed at low lipid concentrations (Fig. 5, Inset, red). This is consistent with our previous reported data (21) and is strongly suggestive of protein–protein interactions. At equilibrium, however, a very different picture emerges. Membrane binding behavior is well described by a simple partition coefficient at all measured concentrations (2.7 × 10⁵; Fig. 5, line, and Table S1). Whereas equilibrium-induced leakage can scale by orders of magnitude over this concentration range, the direct binding isotherm suggests only linear scaling of the total bound protein.

Fig. 5.

Membrane binding of rIAPP as measured by change in mean residue ellipticity (MRE) at 222 nm. rIAPP was mixed with DOPG liposomes and measured either after 48-h incubation, or (Inset) immediately upon mixing. The lines represent a fit to a simple partitioning model.

Antimicrobial Cross-Cooperativity.

rIAPP displays antimicrobial activity that is cross-cooperative with magainin 2 and D-rIAPP. Magainin 2 is a well established bacterial toxin (4). Toxicity by IAPP in mammalian cells is correlated with its mitochondrial localization (22). We therefore chose the Gram-negative bacterium Paracoccus denitrificans as this system has physical and biological properties that make it an often used model for eukaryotic mitochondria (25). The minimum inhibitory concentration (MIC) of antimicrobial peptides is defined as the lowest concentration that leads to an OD₆₀₀ of <5% that of a toxin-free control after 24 h growth. The MICs for rIAPP, magainin 2, and D-rIAPP are 31 ± 5, 1.5 ± 0.4, and 14 ± 2 µM, respectively (Fig. 6A). Growth was then assessed in well format with mixtures of rIAPP, D-rIAPP, and magainin 2 scaled relative to their individual MICs (Fig. 6 B and D). For example, growth in one well was assayed at 0.2 MIC rIAPP (6 µM) combined with 0.5 MIC magainin 2 (0.75 µM). These renormalized mixtures of peptides demonstrate complete cross-cooperativity. Indeed, the inhibition isobole traces a nearly linear path at 1.0 MIC, such that any combination of peptide concentrations where the individual MICs sum to 1 inhibits growth to <5%. Kinetics of growth were measured for a protein-free control, 0.6 MIC of each individual peptide, and mixtures of 0.6 MIC rIAPP with either 0.6 MIC magainin 2 (Fig. 6C) or D-rIAPP (Fig. 6E). Growth rates [determined by colony-forming units (CFU)] are negligibly inhibited at 0.6 MIC of the individual peptides. A mixture of any two of those peptides would therefore not be expected to significantly inhibit growth if they acted completely independently [Bliss independence (26); Fig. 6 C and E, line]. Instead, a >2,000-fold inhibition is observed at 24 h for mixtures of 0.6 MIC rIAPP and 0.6 MIC of either magainin 2 or D-rIAPP. Each mixture showed <400 CFU/µL at 24 h, compared with an average 1.3 × 10⁶ CFU/µL observed for protein-free controls. Clearly, bacterial growth can be cooperatively inhibited by mixtures of rIAPP, magainin 2, and D-rIAPP.

Fig. 6.

Cross-cooperative growth inhibition of P. denitrificans. (A) Relative OD₆₀₀ of inocula of P. denitrificans grown for 24 h with varying amounts of the indicated peptides. OD₆₀₀ values are renormalized to a protein-free control. MIC values (growth <5% that of controls, dashed line) cited in text represent the average of at least four such experiments. (B and D) Growth inhibition of bacterial inocula after 24 h upon exposure to mixtures of rIAPP and either (B) magainin 2 or (D) D-rIAPP. Each point represents an assayed mixture, the content of which can be read off the axes. Axes are renormalized relative to the observed MIC of each individual protein. Points are shown in gray if growth observed is less than 5% that of controls; in black if greater than 5%. Gray background represents the expected growth region for idealized cross-cooperative behavior. (C and E) Time courses of bacterial growth in colony-forming units per microliter. Growth was assayed in the presence of 0.6 MIC of rIAPP (diamonds), magainin 2 (C, triangles), or D-rIAPP (E, squares). A protein-free control is shown as circles. Mixtures of 0.6 MIC of rIAPP with either 0.6 MIC magainin 2 (C) or D-rIAPP (E) are shown as stars. The open stars indicate 0–100 CFU/µL. The black line represents calculated complete Bliss independence of the mixture condition based on the response of the individual peptides.

Discussion

Initiation and maintenance of lipid bilayer pores by IAPP and magainin 2 are not dependent on direct protein–protein interactions. This is the only plausible conclusion given that mixtures of dissimilar sequences (rIAPP and magainin 2) and mirror enantiomers (l-rIAPP and D-rIAPP) are synergistic. Cooperative membrane leakage with the same high apparent reaction order is reproduced regardless of whether we add more of the same protein, or one of alternate sequence or chirality. As a thought experiment, consider the assembly of homotetrameric hemoglobin. Our observations are akin to an assertion that the rate of assembly would be unchanged even if one-half of the hemoglobin were substituted with the homotetrameric protein alcohol dehydrogenase. One possible approach to reconciling this observable is to look beyond the membrane-bound proteins and treat the bilayer as an active participant in its own poration.

Our starting point for this class of model is to adopt the assumption that thermal fluctuations can spontaneously lead to the formation of defects in any phospholipid bilayer (27). In the absence of protein, these rifts are energetically unfavorable, form rarely, and reseal more rapidly than the rate at which small molecules can traverse the bilayer. Leakage competence, however, can result from an alteration in this energy landscape brought upon by protein binding.

Amphipathic antimicrobial peptides such as magainin 2 can bind and insert into membranes laterally, increasing surface pressure within the membrane (28). The bound peptides reside in the intermediate region between the head group and acyl chains of the bilayer (29). This binding mode disproportionally expands the head group region relative to the acyl region of the membrane. One result of this is a thinning of the acyl chain region (30), which has been suggested to result in the formation of an internal surface tension within the bilayer due to the nonideal packing of the acyl chains (31). Formation of a pore could release this tension, as it creates additional membrane surface without a concomitant thinning of the acyl region. Thus, upon amphipathic peptide binding, the energetic penalty of membrane distortion that occurs upon poration is offset by the energetically favorable release of tension. Membrane tension would therefore lead to larger, more frequent, and/or longer lived bilayer defects (20).

This model was initially developed for antimicrobial peptides but is fully transferable to any membrane active peptide. Indeed, IAPP shares all of the aforementioned characteristics including lateral binding and induction of surface pressure (32), positioning in the intermediate region between the lipid head groups and acyl chains (11, 32), and thinning of the acyl chain region (33) leading to the induction of tension. We assert that amyloid peptides such as IAPP may thus act through the same mechanism.

Curvature of a bilayer in on itself to form a toroidal pore is energetically unfavorable due to such a structure being incompatible with the innate curvature of most biological phospholipids. This penalty could be attenuated by peptide binding. IAPP and magainin 2 have each been reported to induce membrane curvature (34–36). It is therefore reasonable to hypothesize that these peptides can stabilize a stochastically formed rift by binding to the pore. Importantly, direct protein–protein interactions may be present, but are not required. As a result, it does not matter if the tension and curvature are derived from a single type of peptide or a mixture of peptides of different primary structures or chiralities. The activity of individual peptides would be expected to sum on the membrane, affecting both the stochastic initiation and equilibrium behavior of pores.

This proposed model is sufficient to describe all of the data presented in this work. The initial formation of a pore is catalyzed by peptide-induced membrane tension that lowers the activation energy of spontaneous poration to a regime more accessible by thermal fluctuations (Fig. 3). These pores are then stabilized by both tension release due to poration as well as peptides binding to the pore. These factors determine the observed equilibrium leakage rates (Fig. 2) and binding isotherms (Fig. 5), resulting in stable poration at high protein concentrations and continuous cycles of transient poration at lower concentrations, as has been seen with both IAPP (13) and magainin 2 (6). Such pores could either be well-ordered toroidal pores, or could take the form of a more disordered chaotic pore (37, 38), and may vary based on the specific peptide and membrane characteristics.

The protein-stabilized pores proposed here, would behave like a dynamically sized oligomer despite the absence of direct protein–protein interactions. We suggest that such a preformed pore would also have the capacity to respond instantaneously to the increased tension and curvature brought about by fresh addition of protein (Fig. 4) without requiring spontaneous formation of a new pore. The fact that complete, activity-renormalized cross-cooperativity of bacterial growth inhibition is observed between IAPP and magainin 2 (Fig. 6) suggests that the poration mechanisms observed in the artificial liposome systems are directly applicable to bacterial cytotoxic activity. Notably, the observation of slight toxicity of 0.6 MIC rIAPP and yet complete toxicity at the combination of 0.6 MIC of each of two dissimilar proteins may be a biological manifestation of this nucleation-dependent phenomenon. Given the similarities of eukaryotic mitochondria with Gram-negative bacteria (25), it is reasonable to suggest that the antimicrobial activity observed here may also be relevant to toxicity targeted at an organism’s own mitochondria (22, 39, 40).

The toxic oligomer hypothesis (16) for amyloid may require a reinterpretation. Oligomers in some diseases may instead serve as effective delivery vehicles for toxic monomers. Oligomers may also be the active membrane binding state (although this does not appear to be a requirement for IAPP). Lastly, oligomers may simply need to be redefined as lipoprotein complexes. The latter may account for the frequent observation of lipid–amyloid oligomer associations, such as in Aβ, α-synuclein, and IAPP (2, 17). In amyloid pathology, cytotoxicity will likely be based on an interplay of stochastic nucleation of both poration and fibrillation. Fibril formation directly on cellular membranes has been shown to cause local membrane damage (33, 41), yet may also have an ensemble cytoprotective effect as it sequesters toxic monomer from solution, slowing pore nucleation and leakage on surrounding membranes (13) (Fig. S2B).

The model we assert here and previously (13) is consistent with ones developed independently for magainin 2 (6, 14) as well as the venom melittin (42). Several amyloid proteins have also now been shown to act as antimicrobial agents (43). The current research, along with the recent discovery that IAPP has cell-penetrating characteristics and targets intracellularly to mitochondrial membranes (22), supports the idea that cytotoxicity in amyloid disease is a direct consequence of this activity being directed against the organism’s own membranes. Conversely, other work has shown that the antimicrobial peptides protegrin-1 (44) and LL-37 (45) are capable of forming amyloid-like fibrils, suggesting the possibility of a relationship between a peptide’s amphipathic, porating character and its amyloidogenicity. Furthermore, the characteristics of membrane leakage induced by IAPP (13) and magainin 2 (14) bear similarities to those of the proapoptotic protein Bax (46). The model put forward here may therefore extend to this class of membrane peptides critical to organismal homeostasis. From an evolutionary perspective, the mechanism described here allows enhanced flexibility of antimicrobial peptide development, as specific interactions do not have to be maintained to retain cooperativity between peptides. This allows a wide, cross-cooperative arsenal of peptides with different binding characteristics to be presented, further enhancing the broad-spectrum character of this defense mechanism. Further insights in the mechanism or inhibition of poration caused by either antimicrobial or amyloid peptides will thus have implications spanning both fields, such as the design of novel antimicrobial peptides and therapeutics that can limit the proapoptotic effects of amyloid proteins.

Materials and Methods

Materials.

Proteins were synthesized using standard tert-butyloxycarbonyl (hIAPP, rIAPP) or fluorenylmethyloxycarbonyl (D-rIAPP, magainin 2) methods and purified by HPLC. For details, see SI Materials and Methods.

Leakage Kinetics.

Leakage measurements were performed as described in the main text. Unless otherwise stated, 200 µM DOPG liposomes (in monomer units) were used. For additional details, see SI Materials and Methods.

Bacterial Growth Assays.

Growth of Paracoccus denitrificans was performed in 96-well format in Mueller Hinton broth with differing amounts of peptide. Colony-forming units were determined by serial dilution onto LB-agar, and counting colonies after 48 h growth at 30 °C. For additional details, see main text and SI Materials and Methods.

Analysis.

Unless otherwise stated, all averages represent data from at least three independent runs. Displayed confidence intervals represent the SEM. All fitting was performed with MATLAB (MathWorks).