Barrel-Stave, Toroidal, and Carpet Models of AMP Action
AMP Mechanisms
3 models
Three primary models explain how antimicrobial peptides breach bacterial membranes: barrel-stave, toroidal pore, and carpet. Each describes a fundamentally different physical interaction.
Brogden, Nat Rev Microbiol, 2005
Brogden, Nat Rev Microbiol, 2005
View as imageAntimicrobial peptides kill bacteria by attacking their membranes. That much has been clear since the first defensins were isolated from human neutrophils in 1985[1] and the magainins were discovered in frog skin two years later.[2] What remained unclear for decades was exactly how a short, positively charged peptide could breach a lipid bilayer. Three models emerged to explain it: the barrel-stave model, the toroidal pore model, and the carpet model. Each describes a different physical mechanism, and understanding their distinctions is central to how antimicrobial peptides kill bacteria.
Key Takeaways
- The barrel-stave model describes peptides forming discrete ion channel-like pores with hydrophobic faces contacting lipid tails, but alamethicin remains the only peptide with confirmed barrel-stave behavior (Bechinger, 1999)
- Toroidal pores involve both peptides and lipid headgroups lining the pore, with peptides inducing positive curvature strain that bends the membrane inward; LL-37, magainin 2, and melittin use this mechanism (Henzler Wildman et al., 2003)
- The carpet model requires no specific pore structure; peptides coat the membrane surface until a threshold concentration triggers membrane fragmentation into micelles (Brogden, 2005)
- All-atom molecular dynamics simulations now capture spontaneous pore formation at the 10+ microsecond timescale, revealing transient hourglass-shaped toroidal pores formed by magainin 2 and PGLa (Zan et al., 2026)
- The SMART (Soft Membranes Adapt and Respond, also Transiently) model proposed by Marquette and Bechinger in 2018 suggests these three models are not mutually exclusive but represent different points on a continuum of peptide-membrane interactions
- Melittin's mechanism remains disputed: electrochemical impedance data show that an equilibrium transmembrane pore cannot explain its activity at low peptide-to-lipid ratios (Wiedman et al., 2013)
What All Three Models Share
Before describing what separates these models, it helps to understand what they have in common. All three begin with the same initial event: a cationic peptide encounters a negatively charged bacterial membrane and binds to it through electrostatic attraction.[3]
Bacterial membranes are rich in anionic phospholipids like phosphatidylglycerol and cardiolipin. Mammalian cell membranes are predominantly zwitterionic (phosphatidylcholine and sphingomyelin) and contain cholesterol, which stiffens the bilayer and resists peptide insertion. This charge asymmetry is the basis of why AMPs target bacterial membranes but spare your own cells.
Once bound, the peptides adopt amphipathic conformations, typically alpha-helices with one hydrophobic face and one hydrophilic face.[4] At this point, the models diverge. The differences come down to three questions: Does the peptide insert perpendicularly into the membrane? Does it form specific peptide-peptide contacts? Does the lipid bilayer bend to line the resulting pore?
The Barrel-Stave Model
In the barrel-stave model, peptides insert vertically into the lipid bilayer and arrange themselves like the staves of a barrel around a central aqueous pore. The hydrophobic faces of the peptides contact the hydrocarbon tails of the surrounding lipids. The hydrophilic faces line the pore interior, creating a water-filled channel.[3]
This geometry closely resembles a transmembrane protein ion channel. It requires several specific conditions: the peptide must be long enough to span the membrane (roughly 20 residues in an alpha-helix for a typical 30-angstrom bilayer), the peptides must make direct lateral contacts with each other, and the resulting pore must be stable.
The textbook example is alamethicin, a 20-residue peptide produced by the fungus Trichoderma viride. Solid-state NMR studies confirmed that alamethicin adopts a transmembrane orientation in lipid bilayers, in contrast to the surface-parallel alignment observed for most cationic AMPs.[5] Alamethicin is unusual among AMPs. It is hydrophobic rather than cationic, it contains the non-standard amino acid alpha-aminoisobutyric acid (Aib) that stabilizes its helical structure, and its mechanism depends on voltage-dependent conductance states, similar to a gated ion channel.[6]
This matters because alamethicin may be the only peptide that genuinely forms barrel-stave pores. Bechinger's 1999 NMR analysis found that cationic AMPs like cecropins and magainins align parallel to the membrane surface, not perpendicular to it.[5] That orientation is incompatible with barrel-stave geometry. The barrel-stave model remains important conceptually, but its confirmed biological relevance is narrow.
The Toroidal Pore Model
The toroidal pore model addresses a key limitation of the barrel-stave framework: most AMPs lie flat on the membrane surface rather than spanning it vertically. In the toroidal model, peptides do not line the pore alone. Instead, they induce the lipid bilayer itself to curve inward, so that both peptides and lipid headgroups line the pore wall. The result is a continuous bend from the outer leaflet through the pore and into the inner leaflet, with a toroidal (doughnut-shaped) geometry.[3]
Matsuzaki and colleagues provided early evidence for this model using magainin 2. They observed that magainin-induced membrane permeabilization was accompanied by rapid phospholipid flip-flop, the movement of lipids from one leaflet to the other. This flip-flop would not occur through a barrel-stave pore where lipids remain undisturbed, but it is a natural consequence of a toroidal pore where the lipid headgroups form part of the pore structure.[7]
The strongest evidence for toroidal pores in LL-37, the only human cathelicidin, came from Henzler Wildman and colleagues in 2003. Using 15N and 31P solid-state NMR, they showed that LL-37 maintains a surface-parallel orientation at all tested concentrations and temperatures, ruling out barrel-stave behavior. LL-37 did not fragment membranes into micelles, ruling out a detergent-like mechanism. But LL-37 did increase the lamellar-to-inverted hexagonal phase transition temperature of lipid systems, demonstrating that it induces positive curvature strain. These combined observations supported a toroidal pore mechanism.[8] For more on LL-37's versatile biology, see LL-37: the only human cathelicidin and why it's so versatile.
Toroidal Pores Are Transient
A critical insight from recent computational work is that toroidal pores are not static structures. Zan and colleagues published all-atom molecular dynamics simulations in 2026 that captured spontaneous pore formation by PGLa and magainin 2 in bacterial membrane mimics at the 10+ microsecond timescale.[9] Their simulations revealed an hourglass-shaped pore geometry, narrower at the center than at the membrane surfaces, with peptides assembling as double stacks supporting the pore from both leaflets.
These pores were not permanent channels. They formed, fluctuated, and closed within the simulation window. This transient nature helps explain decades of conflicting experimental results: depending on when and how you measure, the same peptide can appear to form stable pores or no pores at all.
The Carpet Model
The carpet model describes a different mechanism entirely. Peptides accumulate on the membrane surface in a "carpet-like" arrangement, lying flat and covering the outer leaflet. No insertion into the hydrophobic core occurs at low concentrations. Once peptide concentration on the surface reaches a critical threshold, the accumulated charge and surface tension destabilize the bilayer, causing it to fragment into micelles or mixed peptide-lipid aggregates.[3]
The carpet model does not require specific peptide-peptide interactions or a defined pore structure. It is essentially a detergent-like disruption driven by the amphipathic nature of the peptide accumulating at the membrane interface. The key distinction from the other two models: the membrane is destroyed, not perforated.
Evidence for the carpet mechanism comes from several observations. Certain AMPs cause complete solubilization of lipid vesicles above a threshold concentration, converting large liposomes into small mixed micelles.[4] This all-or-none behavior, where low concentrations produce no effect and high concentrations produce total disruption, is consistent with a cooperative surface-coverage threshold rather than the graded leakage expected from discrete pore formation.
Cecropin-magainin hybrid peptides provided direct experimental support. Kang and colleagues showed in 1998 that these hybrids released entrapped molecules from phospholipid vesicles, with activity correlating to peptide hydrophobicity and the extent of alpha-helical structure formed upon membrane contact.[10] The release of very large molecules (55 kDa fluorescein-labeled IgG) from vesicles suggested wholesale membrane disruption rather than small discrete pores.
Are These Models Mutually Exclusive?
Increasingly, the answer is no. The same peptide may operate through different mechanisms depending on its concentration, the lipid composition of the target membrane, temperature, and ionic conditions.
Melittin, the principal component of bee venom, illustrates this ambiguity. It has been classified as a toroidal pore-former based on early biophysical evidence. But Wiedman and colleagues used electrochemical impedance spectroscopy in 2013 to show that melittin's activity at physiologically relevant peptide-to-lipid ratios (1:5000 to 1:100) could not be explained by an equilibrium transmembrane pore model.[11] Instead, the data suggested transient bilayer permeabilization without stable pore formation.
Pandidan and Mechler's 2019 nano-viscosimetry analysis found that melittin uses different mechanisms depending on membrane composition. In bacterial-mimetic charged lipid mixtures, the viscoelastic fingerprints pointed to a surface-acting (carpet-like) mechanism. In mammalian-mimetic neutral membranes, melittin appeared to penetrate the bilayer already at low concentrations, consistent with insertion-based mechanisms.[12]
The SMART Model
Marquette and Bechinger proposed the SMART model (Soft Membranes Adapt and Respond, also Transiently) in 2018 to accommodate this complexity.[6] Rather than treating the three classical models as distinct categories, SMART frames them as different manifestations of a continuous set of peptide-membrane interactions. A single peptide may cause transient membrane openings at low concentration (consistent with toroidal pores), but at higher concentrations cause complete membrane disintegration (consistent with the carpet model). The key variable is the peptide-to-lipid ratio, not a fixed intrinsic mechanism.
This framework also helps explain synergistic effects between AMPs. When PGLa and magainin 2 are combined at equimolar ratios, their antimicrobial activity exceeds what either peptide achieves alone. Marquette and Bechinger's isothermal titration calorimetry and dynamic light scattering data suggest that the peptide combination mediates intermembrane interactions and liposome agglutination, a mechanism that fits none of the classical models neatly but is consistent with the SMART continuum.[6]
Beyond Membrane Models: Intracellular Targets
The membrane disruption models dominated AMP research for decades, but multiple studies now show that many antimicrobial peptides also attack intracellular targets after crossing the membrane. Brogden's 2005 review in Nature Reviews Microbiology catalogued AMPs that inhibit cell wall synthesis, DNA replication, protein synthesis, and enzymatic activity.[3]
The tri-hybrid peptide LHP7, constructed from fragments of lactoferricin, HP, and plectasin, demonstrated this dual mechanism directly. Xi and colleagues showed in 2014 that LHP7 both disrupted the S. aureus membrane (confirmed by membrane integrity assays and electron microscopy) and bound genomic and plasmid DNA at mass ratios of 2.5 to 10, inserting into the DNA groove and causing cell cycle arrest.[13]
This does not invalidate the membrane models. It suggests that membrane disruption is often the first step in a multi-hit killing mechanism. The peptide breaches the membrane, leaking ions and small molecules; then translocated peptides attack internal targets, ensuring the bacterium cannot recover. For peptides like lactoferricin B, which form blebs and holes on bacterial surfaces without completely lysing cells at MIC concentrations, the intracellular component may be essential to bactericidal activity.[14]
Which Model Applies to Which Peptide?
No universal rule assigns a peptide to one model. But some patterns emerge from decades of structural and biophysical research:
| Feature | Barrel-Stave | Toroidal Pore | Carpet |
|---|---|---|---|
| Peptide orientation | Perpendicular (transmembrane) | Surface-parallel, then inserted | Surface-parallel |
| Lipid involvement in pore | None (lipids undisturbed) | Headgroups line the pore | N/A (no discrete pore) |
| Lipid flip-flop | No | Yes | N/A |
| Peptide-peptide contacts | Required (lateral stave contacts) | Not required | Not required |
| Threshold behavior | Conductance states | Concentration-dependent | Sharp concentration threshold |
| Confirmed examples | Alamethicin | LL-37, magainin 2, melittin (at some ratios) | Cecropin-derived hybrids, dermaseptins |
| Evidence quality | Strong (single peptide) | Strong (multiple peptides, NMR + MD) | Moderate (hard to distinguish from toroidal at low concentrations) |
The defensins, a large family first characterized by Ganz and colleagues,[1] present their own mechanistic complexity. Unlike the alpha-helical AMPs discussed above, defensins are beta-sheet peptides stabilized by disulfide bonds. Reduced cryptdin-4, for example, shows potent bactericidal activity against commensal bacteria that the oxidized form cannot kill, and this activity correlates with hydrophobicity and membrane insertion rather than fitting neatly into any of the three classical models.[15] Read more about defensins as your body's first line of defense.
Experimental Methods That Distinguish the Models
The reason these models remain debated is partly methodological. Different techniques capture different aspects of peptide-membrane interactions:
Solid-state NMR reveals peptide orientation (parallel vs. perpendicular to the bilayer) and lipid packing disruption. This is how Bechinger confirmed that most cationic AMPs lie parallel to the surface,[5] and how Henzler Wildman identified LL-37 as a toroidal pore-former.[8]
Vesicle leakage assays measure whether peptides allow entrapped fluorescent dyes to escape liposomes. The size of the released molecule provides information about pore size. Kang and colleagues used this approach with cecropin-magainin hybrids.[10]
Electrochemical impedance spectroscopy (EIS) measures real-time changes in membrane resistance and capacitance, allowing discrimination between stable pores and transient perturbations. Wiedman and colleagues used EIS to challenge the equilibrium pore model for melittin.[11]
All-atom molecular dynamics (MD) simulations can now capture pore formation directly, though the computational cost is substantial. Simulations exceeding 10 microseconds are required to observe spontaneous pore assembly, as demonstrated by Zan et al. for PGLa and magainin 2.[9]
Quartz crystal microbalance (QCM) fingerprinting tracks mass and viscoelastic changes on supported bilayers, distinguishing surface binding from insertion and membrane removal. Pandidan and Mechler used this to show that melittin acts differently on charged vs. neutral membranes.[12]
Each technique captures a snapshot of the interaction, and conflicting results between methods have fueled decades of debate. The emerging consensus favors a continuum of behaviors rather than rigid categories.
The Bottom Line
The barrel-stave, toroidal, and carpet models were developed to explain how antimicrobial peptides destroy bacterial membranes, but the evidence increasingly suggests these are idealized endpoints on a spectrum rather than discrete mechanisms. Alamethicin remains the only confirmed barrel-stave peptide. Toroidal pores, confirmed for LL-37 and magainin 2 through NMR and MD simulations, appear transient rather than stable. The carpet model describes the high-concentration limit where membrane destruction replaces perforation. The SMART framework accommodates the reality that a single peptide's mechanism shifts with concentration, lipid composition, and environmental conditions.