Why AMPs Target Bacteria but Spare Your Own Cells

How AMPs Kill Bacteria

0 insertion

Cationic antimicrobial peptides inserted readily into bacterial membrane mimics but showed zero detectable insertion into mammalian membrane models.

Glukhov et al., J Biol Chem, 2005

Glukhov et al., J Biol Chem, 2005

View as image

Antimicrobial peptides carry a net positive charge. Bacterial membranes carry a net negative charge. Human cell membranes do not. That electrochemical mismatch is the single biggest reason your body's antimicrobial peptides can kill bacteria without destroying your own cells. But the full picture involves at least four additional layers of molecular discrimination, from cholesterol shielding to lipid chain architecture, that together produce a remarkably selective killing system.

Glukhov et al. demonstrated this in 2005 by testing cationic peptides against phospholipid mixtures mimicking bacterial and mammalian membranes. Peptides readily inserted into the bacterial lipid mixtures. They showed zero detectable insertion into mammalian membrane models.^[1] Understanding exactly why that happens has occupied biophysicists for three decades, and the answers have direct implications for designing peptide-based antibiotics that are safe enough for clinical use.

The Charge Gap Between Bacterial and Human Membranes

Bacterial cell membranes are built from a fundamentally different set of phospholipids than mammalian cell membranes. This difference is electrochemical, and it controls whether an antimicrobial peptide binds or keeps moving.

Bacterial membranes are rich in phosphatidylglycerol (PG), cardiolipin (CL), and phosphatidylserine (PS). All three carry a net negative charge at physiological pH. In many Gram-positive species, PG and CL together account for 50-70% of the total membrane phospholipid content.^[4]

Human cell membranes are composed primarily of phosphatidylcholine (PC), phosphatidylethanolamine (PE), and sphingomyelin (SM). These are zwitterionic lipids. They carry both a positive and negative charge internally, producing a net-neutral surface at physiological pH.^[3]

Antimicrobial peptides carry a net positive charge of +2 to +9 from clusters of lysine and arginine residues.^[4] The electrostatic attraction between a cationic peptide and an anionic bacterial surface is the initial binding event. Taheri-Araghi and Ha modeled this interaction in 2007 and found that an optimal peptide charge exists: too little charge means weak binding, but the relationship is not simply "more charge equals more selectivity." Their physics model showed that selective asymmetrical insertion into the outer leaflet is maximized at specific charge-to-hydrophobicity ratios.^[5]

This charge-based discrimination is fast and binary. A peptide approaching a neutral mammalian membrane surface experiences no electrostatic pull. A peptide approaching a negatively charged bacterial membrane does. The encounter takes microseconds, but it determines everything that follows.

What Happens When Peptides Meet Human Membranes

Lad et al. quantified this binding selectivity in 2007 using surface pressure measurements and neutron reflectivity on model membranes. They compared how three well-characterized antimicrobial peptides (melittin, magainin II, and cecropin P1) interacted with anionic DPPG (mimicking bacteria) versus zwitterionic DPPC (mimicking mammalian cells).^[6]

Melittin adsorption to DPPG reached 2.1 mg/m² from a 0.7 μM solution. Adsorption to DPPC was 0.4 mg/m², a 5.25-fold difference. Magainin and cecropin showed even lower binding to the mammalian mimic. All three peptides penetrated the DPPC monolayer to some degree, as measured by surface pressure, but the amount of peptide actually retained at the membrane surface was minimal.^[6]

Glukhov et al. took this further by testing novel cationic peptides against full lipid mixtures (not single-lipid monolayers). They used Trp fluorescence quenching with doxyl groups positioned at varying depths along the phospholipid acyl chains to measure exactly how deep peptides penetrated. In bacterial membrane mimics, peptides inserted 2.5 to 8 Angstroms deep, oriented roughly parallel to the membrane surface. In mammalian membrane mimics containing zwitterionic lipids and cholesterol, they detected no insertion at all.^[1]

This is not a matter of degree. It is a binary outcome. The peptides either enter the bacterial membrane or they do not enter the mammalian membrane. The carpet model of AMP action described by Shai in 2002 explains why: peptides accumulate on the membrane surface until reaching a critical concentration, then disrupt the bilayer. If the initial binding step fails, the entire killing cascade never starts.^[2]

Cholesterol: The Molecular Shield Bacteria Lack

Human cell membranes contain approximately 25-30% cholesterol by mole fraction. Bacterial cell membranes contain none. This difference is ancient: cholesterol biosynthesis is a eukaryotic innovation that bacteria never acquired.^[7]

Brender, McHenry, and Ramamoorthy addressed this question directly in a 2012 review titled "Does cholesterol play a role in the bacterial selectivity of antimicrobial peptides?"^[7] The answer, based on accumulated biophysical evidence, is that cholesterol contributes to selectivity through at least three mechanisms.

First, cholesterol condenses the lipid bilayer. It fills gaps between phospholipid acyl chains, reducing membrane fluidity and making it physically harder for a peptide to wedge between lipid molecules. Second, cholesterol increases the mechanical strength of the bilayer, raising the energy barrier for the pore formation that many AMPs use to kill bacteria. Third, cholesterol may directly bind to some antimicrobial peptides, sequestering them before they can reach the phospholipid bilayer.

The practical impact: when cholesterol is added to model membranes that would otherwise be disrupted by AMPs, membrane damage drops sharply. When cholesterol is removed from normally resistant membranes, susceptibility increases. The original magainins discovered by Zasloff in 1987 were described as "nonhemolytic at their effective antimicrobial concentrations," a direct consequence of the cholesterol-containing erythrocyte membrane resisting the same peptides that destroyed bacteria.^[8]

Cholesterol does not fully explain selectivity on its own. Some AMPs, like melittin from bee venom, are cytotoxic to mammalian cells despite cholesterol's presence. The reason is that melittin has an unusually high hydrophobicity that allows it to overcome the cholesterol barrier. This points to the broader structural features that distinguish selective from non-selective peptides.^[6]

The Transmembrane Potential Gradient

Beyond membrane composition, bacteria and human cells differ in their transmembrane electrical potential. Bacterial cells maintain a resting membrane potential of approximately -130 to -150 mV (inside negative). Human cells maintain approximately -70 to -90 mV.^[4]

This voltage difference creates an additional driving force that pulls positively charged peptides toward the bacterial membrane interior. A peptide with a net charge of +6 experiences roughly twice the electrophoretic force when approaching a bacterial membrane versus a mammalian one. Taheri-Araghi and Ha's physical model incorporated this transmembrane potential and found it amplifies the charge-based selectivity described above, effectively multiplying the initial electrostatic advantage.^[5]

The transmembrane potential also plays a role after initial binding. Once a peptide has partially inserted into the outer leaflet, the electric field across the membrane helps drive the hydrophobic core deeper into the bilayer. This contributes to the formation of transmembrane pores in bacteria, while the weaker field across mammalian membranes is insufficient to drive the same insertion depth.

Beyond Headgroup Charge: Lipid Chain Architecture

Electrostatics dominate the selectivity conversation, but Shahmiri et al. reported in 2016 that the human cathelicidin LL-37 switches its mode of action based on the structure of lipid alkyl chains, not the headgroups.^[9]

In bilayers of unsaturated phospholipids (common in bacterial membranes), LL-37 formed pores. In bilayers of saturated phospholipids, it produced helical-rich fibrous peptide-lipid superstructures instead of pores. Bacterial membranes tend to have a higher proportion of unsaturated acyl chains than mammalian membranes, meaning the physical disorder of the bacterial membrane interior also contributes to susceptibility.^[9]

Böhling et al. showed the same principle operating with human beta-defensin 3 (HBD-3). Using planar asymmetric bilayers reconstructed from bacterial lipopolysaccharides and mammalian phospholipids, they found that HBD-3 could only induce lesions in membranes matching the lipid composition of sensitive bacterial strains. The correlation between lipid matrix composition and peptide activity was direct, and the lipid-specificity explained both antimicrobial potency and the absence of cytotoxicity.^[10]

Hein et al. expanded this concept in their 2022 review of defensin-lipid interactions, noting that defensins uniquely target specific membrane lipids via mechanisms distinct from other host defense peptides, which could enable the development of therapeutics with increased selectivity.^[11]

The Structural Rules That Predict Selectivity

If selectivity were purely about membrane chemistry, every cationic peptide would be selective. They are not. Melittin, protegrins, and several synthetic peptides kill bacteria and damage human cells. The structural features of the peptide itself determine whether it respects the membrane boundaries or ignores them.

Dathe and Wieprecht mapped these rules in a 1999 review of helical antimicrobial peptides. The critical parameters are hydrophobicity, net positive charge, helical stability, and amphipathicity (the degree of separation between the hydrophobic and charged faces of the helix). Their analysis showed that increasing hydrophobicity beyond a moderate threshold increases mammalian cell toxicity without improving antibacterial potency. A selectivity sweet spot exists: moderate hydrophobicity combined with strong amphipathicity and high net charge.^[3]

Randall et al. brought this question into the machine learning era in 2023 with deep mutational scanning of Protegrin-1, a potent but toxic antimicrobial peptide. They developed a method called dmSLAY (deep mutational surface localized antimicrobial display) that tested thousands of sequence variants for antibacterial activity. Their key findings: avoiding large aromatic residues and eliminating disulfide-bonded cysteine pairs while maintaining membrane-bound secondary structure improved bacterial specificity. Using these datasets, machine learning models expanded the analysis to over 5.7 million sequence variants and mapped the full mutational profiles that drive either bacterial or mammalian membrane preference.^[12]

This computational approach represents a shift from studying one peptide at a time to mapping the entire selectivity landscape. Brogden's 2005 Nature Reviews Microbiology article had noted that the existence of thousands of active AMPs with variable sequences "suggests a general mechanism for killing bacteria rather than a specific mechanism that requires preferred active structures."^[4] Randall's work shows that while the killing mechanism may be general, the selectivity rules are specific and learnable.

How Bacteria Try to Reduce This Selectivity Gap

Bacteria are not passive targets. Several species modify their membrane charge to reduce the electrostatic advantage that AMPs exploit. The best-characterized resistance mechanism is the addition of lysyl-phosphatidylglycerol to the membrane surface. By attaching a positively charged lysine to the normally anionic PG lipid, bacteria reduce the net negative charge on their outer surface, weakening the initial electrostatic attraction of cationic AMPs.^[4]

Staphylococcus aureus uses this strategy extensively through the MprF (multiple peptide resistance factor) system. Other bacteria add aminoarabinose to lipopolysaccharide or modify their teichoic acids with D-alanine, each of which partially neutralizes surface charge. These modifications reduce but do not eliminate AMP susceptibility, in part because AMPs target multiple membrane properties simultaneously. A bacterium can reduce charge but cannot add cholesterol or fundamentally restructure its lipid bilayer.^[4]

This is why AMPs that attack intracellular targets in addition to membranes may prove harder for bacteria to resist. A peptide that disrupts membranes and inhibits DNA synthesis simultaneously requires the bacterium to develop resistance to two mechanisms at once.

The Bottom Line

Antimicrobial peptide selectivity is not one mechanism but a layered system: electrostatic attraction to anionic bacterial membranes, cholesterol-mediated protection of human cells, transmembrane potential differences that favor bacterial insertion, lipid chain architecture that determines the mode of membrane disruption, and peptide structural features that can be optimized for safety. The 2005 finding that cationic peptides show zero insertion into mammalian membrane models while readily penetrating bacterial ones demonstrates that selectivity is not a matter of degree but a binary molecular outcome.

Frequently Asked Questions

Why do antimicrobial peptides kill bacteria but not human cells?

Antimicrobial peptides are positively charged molecules that are electrostatically attracted to the negatively charged surfaces of bacterial membranes. Human cell membranes are composed of neutral (zwitterionic) lipids and contain cholesterol, which stiffens the membrane and blocks peptide insertion. Glukhov et al. demonstrated in 2005 that cationic peptides showed zero detectable insertion into mammalian membrane models while readily penetrating bacterial ones.

What makes bacterial membranes different from human cell membranes?

Bacterial membranes are rich in negatively charged phospholipids like phosphatidylglycerol and cardiolipin, which together can constitute 50-70% of the total phospholipid content. Human cell membranes are primarily composed of zwitterionic (net-neutral) phospholipids like phosphatidylcholine and sphingomyelin, and contain approximately 25-30% cholesterol. Bacteria do not produce cholesterol.

Does cholesterol protect human cells from antimicrobial peptides?

Yes. Cholesterol, which constitutes about 25-30% of human cell membranes and is absent from bacterial membranes, protects human cells in three ways: it stiffens the lipid bilayer against peptide insertion, raises the energy barrier for pore formation, and may directly bind some peptides. A 2012 Frontiers in Immunology review confirmed that adding cholesterol to model membranes sharply reduces AMP-mediated damage.

Can bacteria become resistant to antimicrobial peptides?

Bacteria can reduce their susceptibility to AMPs by modifying their membrane charge. The best-known mechanism is the MprF system in Staphylococcus aureus, which adds lysyl-phosphatidylglycerol to the membrane, partially neutralizing its negative charge. These modifications reduce but do not eliminate AMP activity because AMPs target multiple membrane properties that bacteria cannot fundamentally restructure.

What determines whether an antimicrobial peptide is safe for human cells?

Peptide selectivity depends on a balance of hydrophobicity, net positive charge, helical stability, and amphipathicity. Dathe and Wieprecht showed in 1999 that excessive hydrophobicity increases toxicity to human cells without improving antibacterial activity. A 2023 study using deep mutational scanning of 5.7 million Protegrin-1 variants found that removing large aromatic residues and disulfide-bonded cysteines improved bacterial specificity.

How much more do antimicrobial peptides bind to bacterial membranes versus human membranes?

Lad et al. measured a 5.25-fold difference in binding for melittin: 2.1 mg/m² adsorption to the bacterial membrane mimic DPPG versus 0.4 mg/m² to the mammalian mimic DPPC. Magainin and cecropin showed even greater selectivity differences. Glukhov et al. found a complete binary distinction when using full lipid mixtures rather than single-lipid models: measurable insertion into bacterial mimics, zero into mammalian ones.